Coprime sets in $[1,n)$: is $\sum_{a\in A}1/(n-a)\leq\sum_{p<n}1/p+O(1)$? (Erdős #1210)
Statement
Let $A\subseteq[1,n)$ be a set of integers that is pairwise coprime, meaning $\gcd(a,b)=1$ for all distinct $a,b\in A$. Is it true that $$\sum_{a\in A}\frac{1}{n-a}\leq\sum_{p<n}\frac{1}{p}+O(1),$$ where the right-hand sum runs over primes $p<n$ and the $O(1)$ is an absolute constant, independent of both $n$ and the set $A$?
Acceptance. FULLY RESOLVES: a proof (machine-checkable Lean/Coq preferred, else a complete written proof) that there is an absolute constant $C$ with $\sum_{a\in A}1/(n-a)\leq\sum_{p<n}1/p+C$ for every $n$ and every pairwise-coprime $A\subseteq[1,n)$; OR a disproof exhibiting, for a sequence $n\to\infty$, pairwise-coprime sets $A_n\subseteq[1,n)$ with $\sum_{a\in A_n}1/(n-a)-\sum_{p<n}1/p\to\infty$, together with a certificate that each $A_n$ is pairwise coprime and a proof the difference is unbounded. ADVANCES: (a) prove the inequality with an explicit sub-constant slack in place of $O(1)$, e.g. $\sum_{a\in A}1/(n-a)\leq\sum_{p<n}1/p+o(\log\log n)$ or $\leq(1+o(1))\sum_{p<n}1/p$, with proof; (b) establish the [Er77c] interval version for the primes in $(n,m]$, with proof; (c) exhibit, via extremal search, pairwise-coprime sets attaining the largest known value of $\sum_{a\in A}1/(n-a)-\sum_{p<n}1/p$, supplying the search code and the certified difference as evidence for the true constant. Deliver the proof, the counterexample family plus coprimality certificate, or the extremal-search code plus attained differences.
Background
Posed by Erdős [Er77c, p.64] and restated in [Er80, p.112], where he remarks that the [Er77c] formulation was 'not quite correctly' stated. The version in [Er77c] concerns the primes $n<q_1<\cdots<q_k\leq m$ in an interval $(n,m]$ and asks whether $\sum_i 1/(q_i-n)<\sum_{p<m-n}1/p+O(1)$. The heuristic is that the reciprocals $1/(n-a)$ over a pairwise-coprime set cannot concentrate more mass just below $n$ than the primes do, so the prime reciprocal sum $\sum_{p<n}1/p=\log\log n+M+o(1)$ (Mertens' theorem, $M$ the Mertens constant) should dominate up to a bounded additive error. The site records related problems Erdős #460 and #950 (erdosproblems.com/460, erdosproblems.com/950) but no literature bound for #1210 itself. Listed as open on erdosproblems.com/1210 (fetched 2026-07-21, status 'open', tagged 'number theory'). No prize is recorded. Attacker's tool: for each $n$ the left side is maximized by an integer program / greedy search over pairwise-coprime subsets of $[1,n)$ — each prime power may contribute at most one element — so one can compute $\max_A\sum_{a\in A}1/(n-a)$ exactly for moderate $n$, subtract $\sum_{p<n}1/p$, and test whether the difference stays bounded (yielding numerical evidence for the constant, or a candidate growing counterexample), alongside sieve/analytic estimates for a proof.
References
| Ref | Source | Type |
|---|---|---|
| REF-01 | Erdős Problem #1210 (T. F. Bloom) | website |
Investigations · 0
No published investigations yet. This problem is unclaimed territory.