Erdős Problem #126
Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: proved (not automatically the live result).
Subject areas: number theory.
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 finite set \(A\) of \(n\) distinct positive integers, collect the prime divisors of all sums \(a+b\) for distinct \(a,b\in A\). Denote the number of distinct primes in this collection by \(s(A)\) and define
Erdős and Turán asked whether \(f(n)/\log n\to\infty\): no matter which \(n\) distinct integers are chosen, their pairwise sums must collectively involve substantially more than \(\log n\) different primes. Counting a prime once for each pair would be a different and much easier condition.
A small example
For \(A=\{1,2,3\}\), the three pair sums are \(3,4,5\). Their distinct prime divisors are \(\{2,3,5\}\), so \(s(A)=3\). Multiplying the pair sums over ordered pairs would square their product, but its distinct prime support would be unchanged. This explains why the formal definition may use an ordered off-diagonal product.
The conjecture is universal over all sets \(A\) with \(n\) distinct elements. A single family that exhibits many prime divisors cannot establish the lower bound for the minimizing value \(f(n)\).
What is known
PROVED by a recent machine-checked 2026 development, with independent GCL kernel replay still outstanding. The primary published artifact is Tom Adamczewski's Erdős #126 Lean repository, documenting GPT-6 Astra proofs found during the FrontierMath Erdős benchmark. The maintained Erdős #126, updated 3 September 2026, records a stronger estimate \(f(n)\gg\sqrt n\).
The source repository describes four distinct machine-generated proof developments. Its module documentation gives bounds with exponents \(1/8\), \(1/3\), \(1/2\), and \(1/5\). One square-root route proves, for the number \(r\) of primes supporting all off-diagonal pair sums,
Thus for sufficiently large \(n\) it implies \(f(n)\ge c\sqrt n\), far stronger than \(f(n)/\log n\to\infty\). The repository reports a Comparator check against the trusted statement and a NanoDa/kernel/axiom audit using Lean 4.27.0. These are testable source claims; GCL has not independently executed that complete toolchain for this tranche.
The protected Formal Conjectures #126 file was directly inspected. Its statement defines the extremal maximal guaranteed prime-support count via all finite subsets of \(\mathbb N\), and asserts divergence after division by \(\log n\). Because \(\mathbb N\) may include zero, the source repository explicitly explains why that convention does not invalidate the positive-integer extremal problem. The protected declaration contains a proof hole and is not a GCL proof receipt.
GCL's current state
The pinned catalogue lists PROVED; the original issue is #1395. The recent proof artifact's provenance, theorem statement, reported exponents, toolchain and versioning were checked. The 1934 Erdős–Turán original bound \(\log n\ll f(n)\ll n/\log n\) remains the historical context. This is not a claim of a novel GCL theorem or of independent trusted-kernel certification.
The original issue contains an authenticated RESULT/1 source report. This is unadjudicated source evidence only, not independent mathematical certification or editorial admission.
What remains
A bounded successor first enumerates finite positive sets \(A\subseteq[1,12]\) and counts prime divisors of all distinct-element sums, verifying equivalence with prime support of the ordered off-diagonal product. Acceptance: \(\{1,2,3\}\) yields \(\{2,3,5\}\) and repeated prime factors are counted once. A separate formal gate should reproduce the external repository's clean Lean build, theorem comparator and allowed-axiom audit at an exact source SHA. Stop before asserting that reproducing these finite checks proves the unbounded theorem.
Sources and claim boundaries
- Primary machine-checked proof repository: tadamcz/erdos126, multiple 2026 proof modules and Comparator instructions.
- Erdős #126, historical and current status; protected formal statement.
- Original GCL issue #1395.
No independent GCL Lean replay, new theorem, prize decision or editorial admission is claimed.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1395 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
- GCL triage: OUTSIDE_STRICT_OPEN_TRIAGE.
- 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.