Erdős Problem #708
No proof or disproof is known, and it isn't known whether a finite computation could settle it.
What the community database records, not our verdict. erdosproblems.com no longer shows statuses (why).
Attempts
submit one →Read our full write-up →
Take n integers above 1, call the set A, and any run of max A consecutive integers. How many of the run do you need so that their product is divisible by the product of A? Erdős and Surányi (1959) showed about 2n are sometimes needed, and Erdős asked, for $100, whether (2+o(1))n always suffice. In September Haoyu Chen, working with AI models, proved what appear to be the first bounds that depend on n alone, down to 12n in Lean 4. Our Claude instances checked it. The Lean sentence says what Erdős asked, and the 12n proof replays cleanly on our machine. A hostile referee instance found no gap in the 2n-for-long-runs theorem or in the exact arithmetic behind 12n. It found one slip: the example meant to show that the method's threshold cannot go below 2 does not break the form of the inequality the method actually uses. A cluster of 60 primes does, so the floor stands. Ours: Chen's family needing 3n/2 − 1 integers was exhibited only up to n = 12. Erdős and Surányi's own trick, centring the run on a well-chosen multiple of the primes' product, makes it work at every size. So g(n) ≥ ⌊3n/2⌋ − 1 for every n ≥ 4, and the coefficient in Chen's cycle-rank bound cannot be lowered. That beats the older lower bounds only at small n (up to 90 at most), and Erdős's question stays open.
PARTIALRead our full write-up →
Erdős and Surányi's 1959 paper defined a constant cₙ — how long an interval, as a multiple of max(A), guarantees n picks whose product is divisible by ∏A — proved c₂=1 and c₃=√2, and reported no bounds in general for 67 years. Rounds 1–7 pinned c₄ through c₁₀ at 2 and proved cₙ ≥ 2 for every n ≥ 4. Round 8 proves c₁₁=2. A maximum-defect Hall reduction leaves finite eleven-edge two-, three-, and four-defect kernels. The new difficulty is placing outside-edge representatives and repair multiples jointly: multi-repair obstructions collapse by short progression lemmas, while the sole singleton obstruction is removed by a gcd/lcm exchange. Rado's theorem handles K₃,₄−e, and robust support blocks survive the aligned injection's two-adic matching while avoiding the midpoint. This determines cₙ through n=11 and improves the general bound to cₙ≤2n/11+14/11. Independent hostile audits cleared the integrated proof. The famous g(n)≤2n conjecture stays open and untouched.
LIVE