Erdős Problem #377
Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: open (not automatically the live result).
Subject areas: number theory, binomial coefficients.
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
Let \(B_n = \binom{2n}{n}\) be the central binomial coefficient. For each positive integer \(n\), form the sum
The question is whether there is one absolute constant \(C>0\) such that \(f(n)\leq C\) for every \(n\). The primes dividing \(B_n\) are deliberately omitted: the question asks whether the reciprocal mass of the primes up to \(n\) that escape the central binomial coefficient stays bounded as \(n\) grows. This is the exact scope of original issue #1647, not a question about the size of \(B_n\) itself or about its largest prime factor.
The quantifiers matter. The same constant must work for all positive integers, with no exceptional set and no “sufficiently large” qualification. A statement that \(f(n)=O(\log\log n)\), or that \(f(n)\) is close to a constant for almost all \(n\), is informative but does not settle the uniform boundedness question.
A small example
Here is an exact finite check at \(n=10\):
The primes at most \(10\) are \(2,3,5,7\). Direct integer divisibility gives \(2\mid184756\), while \(3\), \(5\), and \(7\) do not divide it. Therefore
This was verified with integer arithmetic by computing comb(20, 10), testing each of the four primes by remainder, and summing with exact rational fractions; the checks returned binom(2*10,10) = 184756, the excluded set [2], the included set [3, 5, 7], f(10) = 71/105, and exact checks: PASS. The example makes the definition concrete and is reproducible without floating-point decisions. It proves nothing about a bound valid for every \(n\): a finite table can neither certify an infinite quantifier nor exclude a later value of \(f(n)\).
What is known
The primary source is P. Erdős, R. L. Graham, I. Z. Ruzsa, and E. G. Straus, “On the Prime Factors of \(\binom{2n}{n}\),” Mathematics of Computation 29 (1975), 83–92, the scanned paper. Its introduction on printed page 83 explicitly says that the authors could not decide whether this function is unbounded. That is the historical statement of the open problem, not evidence that the requested constant exists.
The same paper proves substantial average information. Its Theorem 2 (printed page 86) gives
and Theorem 3 (printed page 87) gives the corresponding second-moment limit with value \(\gamma_0^2\). Their corollary then yields \(f(n)=\gamma_0+o(1)\) for almost all integers, in the density sense. These are external published results, but their “almost all” quantifier leaves room for exceptional integers.
Later on printed page 90, in the discussion surrounding equation (7), the authors record an upper bound of the form \(f(n)\leq c\log\log n\) for all sufficiently large \(n\), with some \(c<1\). This improves the trivial Mertens-scale comparison but is still not a constant bound. The live problem page also records that a positive answer would force the reciprocal mass of the primes dividing \(B_n\) to be \((1-o(1))\log\log n\); that implication is useful context, not a proof of either direction.
GCL's current state
The frozen catalogue evidence is teorth/erdosproblems commit 536408716724a2402bdfa7c1d764bf83d64a8c89, with the relevant data/problems.yaml Git blob 4e60263a80af6207aca494d2e1388e44fa372889 . At that revision, problem 377 is open, last updated 2025-08-31; its catalogue formal status is unformalized, while statement-formalized metadata is yes. The live page, accessed 2026-10-13, presents the same question and still labels it open; it is live evidence rather than a pinned revision.
The pinned formal provider is google-deepmind/formal-conjectures commit 85f863718beeec7b58a3a1926ee92e3472bc2020, file FormalConjectures/ErdosProblems/377.lean, blob 83c32ad91760cb76e9854bc8e456810636bb4ba5. Its main theorem has the real quantifier \(\exists C>0\) followed by \(\forall n\colon\mathbb N\) and the same prime-reciprocal sum, expressed using Finset.Icc 1 n and Nat.centralBinom. The file also formalizes the mean, second moment, almost-everywhere, and logarithmic upper-bound variants, but every theorem body is by sorry. This is a theorem source with explicit proof placeholders, not a checked proof or GCL certification. V2 ledger policy alone decides accepted coverage.
What remains
The smallest missing mathematical ingredient is a uniform argument controlling the exceptional integers left open by the average theorems. A bounded successor task should first fix an input cutoff \(N\) and compute every \(f(n)\) for \(1\leq n\leq N\) exactly. It can avoid enormous binomial integers by using the valuation identity
so that \(p\) is omitted precisely when this valuation is positive. A verifier should receive \(N\), the prime list, each valuation or carry certificate, and the exact rational maximum and its witnesses. Acceptance is a replay that every listed prime and divisibility decision is correct; falsification is any missed prime, incorrect valuation, or new \(n\leq N\) exceeding the reported maximum. Stop at the declared cutoff or resource limit. Even a successful finite certificate is evidence for a range, not a proof of the global conjecture.
Sources and claim boundaries
This article uses the original issue, the pinned catalogue record, the pinned formal file, the live problem page, and the 1975 primary paper. The primary paper supports the dated average and upper-bound results cited above. The catalogue and live page support status metadata. The Lean file supports only the existence and shape of formal statements, with sorry retained. The \(n=10\) calculation is an independently reproducible finite check, not a theorem about all \(n\).
No claim here asserts a solved status, a new proof, a formal replay, mathematical certification, publication, priority, or an award. The original SOURCE RESULT/1 citation is authenticated original RESULT/1; no future comment URL is fabricated. Accepted coverage remains a V2-ledger decision.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1647 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
- GCL triage: PROTECTED_READY_BACKLOG.
- Formal-statement coverage: A provider statement was included in the pinned snapshot. Presence does not mean the statement or proof has been checked.
- External mathematical status: consult current problem page and its dated bibliography.