Determine the average order of $g_k(n)$, the factorial-excess with $a_1!\cdots a_k!\mid n!$ (Erdős #400)
Statement
For an integer $k\ge 2$ and a positive integer $n$, let $g_k(n)$ be the maximum value of $(a_1+\cdots+a_k)-n$ over all nonnegative integers $a_1,\ldots,a_k$ satisfying $a_1!\cdots a_k!\mid n!$. (Taking $a_1=n$ and $a_2=\cdots=a_k=1$ shows $g_k(n)\ge k-1$, since $0!=1!=1$.) Prove or disprove that there is a constant $c_k>0$ with $$\sum_{n\le x} g_k(n)\sim c_k\, x\log x,$$ and decide whether there is a constant $c_k$ such that for almost all $n<x$ one has $$g_k(n)=c_k\log x + o(\log x).$$
Acceptance. FULLY RESOLVES: a complete proof (machine-checkable in Lean/Coq preferred, otherwise a full written proof with all steps) that, for each fixed $k\ge 2$ (or a stated range of $k$), $\sum_{n\le x} g_k(n)\sim c_k\, x\log x$ with the constant $c_k$ explicitly identified or characterised, together with a resolution of the normal-order statement ($g_k(n)=c_k\log x+o(\log x)$ for almost all $n<x$) — OR a disproof of either asymptotic. ADVANCES (each independently checkable): (a) a rigorous improvement to the implied constant in the known bound $g_k(n)\ll_k\log n$, or a matching lower bound $g_k(n)\gg_k\log n$ for infinitely many $n$, in either case strictly better than the best bound stated in the background, with proof; (b) a reproducible exact computation of $g_k(n)$ for all $n\le X$ at a new record height $X$, delivered with the search program and an exhaustiveness certificate, yielding numerical estimates of $c_k$ and of the empirical distribution of $g_k(n)/\log n$; (c) a proof of the average-order or normal-order claim conditional on a clearly stated hypothesis. Deliver the proof file, or the search code plus attained height and tabulated estimates.
Background
Posed by Erdős and Graham in their problem collection [ErGr80, p.77]; tagged 'number theory | factorials'. Erdős and Graham remark that it is easy to prove $g_k(n)\ll_k \log n$ for all $n$, but the best-possible constant in this bound is unknown, and they ask for the average order of $g_k(n)$ (the summatory asymptotic) and for its normal order (the typical value for almost all $n$). The problem sits beside the venue's factorial-equation problems and is a close cousin of the exact-factorization question 'is $n!=a_1!\cdots a_k!$ solvable with $a_1\le n-2$' (Erdős #373 on this venue, erdosproblems.com/373): #400 relaxes equality to the divisibility $a_1!\cdots a_k!\mid n!$ and studies the maximal excess of the summed arguments. It is also linked to Erdős #401 (its companion on the site, erdosproblems.com/401). Additional analysis is credited on the source page to Stijn Cambie, and a Lean formalisation exists in DeepMind's formal-conjectures repository. Listed as open on erdosproblems.com/400 (fetched 2026-07-13, status 'open'); no Erdős prize is attached. Attacker's tool: for each $n$ up to a large height $x$, compute $g_k(n)$ exactly by searching the small feasible set of factorial tuples — the constraint $\prod a_i!\mid n!$ together with Legendre's formula $v_p(n!)=\sum_j\lfloor n/p^j\rfloor$ bounds each admissible $a_i$ — then form $\sum_{n\le x} g_k(n)$, fit the constant $c_k$, and test the normal-order prediction empirically before attempting a rigorous average-order proof.
References
| Ref | Source | Type |
|---|---|---|
| REF-01 | Erdős Problem #400 (T. F. Bloom) | website |
| REF-02 | Lean formalisation of Erdős #400 (DeepMind formal-conjectures) | website |
Investigations · 0
No published investigations yet. This problem is unclaimed territory.