Erdős Problem #384
Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: proved (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
Erdős Problem 384 asks how small a prime divisor must occur in a nontrivial binomial coefficient. In the usual notation, take integers with \(1<k<n-1\) and inspect \(\binom{n}{k}\). The maintained problem page currently displays the question as asking for a prime \(p<n/2\), apart from \(\binom{7}{3}=5\cdot7\).
The primary paper gives a nearby, and mathematically coherent, normalization. Ecklund begins with \(n\geq2k\) and proves that \(\binom{n}{k}\) has a prime divisor satisfying
with the single exception \(\binom{7}{3}\). For \(k\geq2\), the maximum is \(n/2\), so this is a non-strict bound. Passing from \(k\leq n/2\) to every \(1<k<n-1\) by \(\binom{n}{k}=\binom{n}{n-k}\) introduces the symmetric exception \(\binom{7}{4}=35\) as well. Thus the source-level theorem is: outside \((n,k)=(7,3)\) and \((7,4)\), some prime divisor \(p\) of \(\binom{n}{k}\) satisfies \(2p\leq n\).
That distinction is not cosmetic. The displayed strict inequality fails already at \((n,k)=(4,2)\), where \(\binom42=6\) has prime divisors \(2\) and \(3\), neither satisfying \(2p<4\). The exact issue-specific source audit is GitHub issue #1654, assignment ERDOS-CATALOGUE-0384-S01.
A small example
Here is an exact finite check of the non-strict theorem's shape. Take \(n=10\) and \(k=4\). Direct integer arithmetic gives
The bound is \(n/2=5\), and the witness \(p=2\) satisfies \(2p=4\leq10\). A reproducible check, using no floating-point arithmetic, is:
import math
n, k = 10, 4
b = math.comb(n, k)
primes = [p for p in range(2, b + 1)
if all(p % d for d in range(2, p) if d*d <= p) and b % p == 0]
print(b, primes, [p for p in primes if 2*p <= n])
The executed output was 210 [2, 3, 5, 7] [2, 3, 5]. This verifies one concrete instance and makes the endpoint convention visible. It does not prove the universal theorem, reproduce Ecklund's estimates, or turn a finite search into a resolution of any stronger conjecture. The same exact calculation at \((n,k)=(4,2)\) returns \(6\), prime divisors \([2,3]\), and no strict witnesses with \(2p<4\); this is the small obstruction to the strict formulation.
What is known
The primary resolution is E. F. Ecklund Jr., “On prime divisors of the binomial coefficient,” Pacific Journal of Mathematics 29 (1969), 267–270, PDF. The theorem is printed on p. 267. The paper proves the result by estimates for prime-counting functions, a Sylvester–Schur-type argument, and a finite check. On pp. 269–270 it says that the remaining ranges were checked for \(4\leq k\leq60\) with \(2k\leq n\leq k^2\) and for \(61\leq k\leq105\) with \(2k\leq n\leq4k\), using an IBM 1620 and manual cross-checking. This description is evidence about the published proof, not a modern replay.
The maintained Erdős Problems page, accessed 2026-10-13, labels the problem proved, cites [ErGr80, p. 73], and attributes the proof to Ecklund. It also records stronger conjectural directions, including the proposed \(p<n/k\) conclusion when \(n>k^2\); those stronger claims are not established by the theorem audited here. The page's displayed strict statement and sole exception do not match Ecklund's non-strict theorem plus symmetry, so the editorial record preserves the discrepancy rather than silently repairing the live wording.
The primary paper is available and unambiguous about its theorem and page. No claim is made here about an independent proof, a prize, or a new result. The finite calculation above is only an interface test.
GCL's current state
The frozen catalogue evidence is teorth/erdosproblems at commit 536408716724a2402bdfa7c1d764bf83d64a8c89, specifically data/problems.yaml, entry number: "384", Git blob 4e60263a80af6207aca494d2e1388e44fa372889. Its derived fields say informal_status: proved, formal_status: Lean, status: proved (Lean), and formalized: yes, with updates through 2026-09-22. These are catalogue metadata; V2 ledger alone decides accepted coverage.
The exact upstream FormalConjectures file is present at commit 817d8a74e6bcd505ac8407388cc12ac3cfda3d0d, file blob d6d8b2b80d6706fea07403d1b10e5ab4386f1663. It states the non-strict theorem with both symmetric exceptions and separately records the strict formulation as false, using \((4,2)\) as the counterexample. Both theorem bodies end in sorry; therefore this is a formal statement and reported interface, not a completed or replayed proof. The GCL-admitted FC-GDM-85F86371 snapshot is public metadata only and was not treated as a protected theorem source.
The original source RESULT/1 citation is authenticated original RESULT/1. This local article does not post an issue comment or confer acceptance, certification, or publication status.
What remains
The mathematical resolution of the non-strict Problem 384 statement is already supplied by Ecklund. The bounded successor task should therefore be semantic and formal rather than a fresh search for a proof. First, fix the exact target as natural numbers \(n,k,p\), hypotheses \(1<k<n-1\), exclusions \((n,k)\ne(7,3)\) and \((n,k)\ne(7,4)\), primality of \(p\), divisibility \(p\mid\binom{n}{k}\), and \(2p\leq n\). Then prove the elementary symmetry bridge between the lower-half source theorem and the unrestricted statement.
A separate falsification lemma should state that the strict target with \(2p<n\) is false at \((4,2)\). A later formal proof may import the mathematical content of Ecklund as an admitted external theorem, but it must label that dependency and must not present sorry as a proof. Acceptance would require an exact statement comparison, a source citation to Ecklund p. 267, and a successful pinned Lean replay when such a replay is authorized. Failure of the comparison, or discovery that the target still says strict inequality, is a stopping condition and should return the task to source normalization.
Sources and claim boundaries
The primary PDF used in this audit has SHA-256 86152d9c7e5d8c62cb47576144fc518b9a81ff4b186b03446c9996ffde64ee27. The frozen catalogue commit and blob are recorded above; the formal file commit and blob are recorded above. The live page is a maintained external source and may change after the access date. The original issue remains the assignment authority.
Established external evidence is Ecklund's published theorem and its stated finite-check ranges. The exact finite examples were independently verified in this worktree. The catalogue and live-page labels are status evidence, not mathematical certification. The FormalConjectures file is a theorem source with sorry, not a checked Lean proof. GCL has supplied no mathematical certification here, and no claim of such certification is made. Any later accepted coverage must follow the protected V2 ledger and retain the source distinction between the coherent non-strict theorem and the false strict wording.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1654 — 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.