Erdős Problem #729 — Large factorial ratios with bounded large prime factors

At least 50 years old · documented by

For every real constant C>0C>0, does there exist an integer K≥3K\ge3 such that there are infinitely many triples (a,b,n)∈N3(a,b,n)\in\mathbb N^3 satisfying a>0a>0, b>0b>0, n>0n>0,

a+b>n+Clog⁡n,a+b>n+C\log n,

and the denominator of the rational number

n!a!b!\frac{n!}{a!b!}

has no prime divisor greater than KK? Equivalently, for every prime p>Kp>K, the pp-adic valuation of that denominator is zero.

References

Progress summary

Refreshed
Claimed solved

A positive solution has been announced and formalized by AI, but no independently verified proof is yet available.

The problem asks whether, for every constant C>0C>0, infinitely many triples satisfy a+b>n+Clog⁡na+b>n+C\log n while the denominator of n!/(a!b!)n!/(a!b!) has only primes bounded in terms of CC. Erdős proved in 1968 that a!b!∣n!a!b!\mid n! forces a+b≤n+O(log⁡n)a+b\le n+O(\log n); the problem asks whether this bound can be exceeded by every fixed logarithmic amount.

Known results

  • Erdős, 1968: a!b!∣n!a!b!\mid n! implies a+b≤n+O(log⁡n)a+b\le n+O(\log n).
  • Carl Pomerance subsequently noted that a related logarithmic-gap result follows by modifying his earlier argument, with a note covering k≤.7log⁡mk\le .7\log m.

January 2026 claimed solution

Barreto and Leeham announced an affirmative solution, obtained from an argument for Problem #728 and subsequently autoformalized by Aristotle from a GPT-5.2 Pro proof. The associated writeup states the required logarithmic-gap factorial-divisibility result, but the available repository extract still displays by sorry; independent mathematical verification is therefore lacking.

Current status (as of June 2026): A positive AI-generated solution and claimed formalization exist, but the problem remains unverified pending an independently checkable completed proof.

Sources

Solutions 0

No solutions have been posted yet.