Math-5571 · mathematics · Mathematics

Math-5571

Erdős Problem #310

Status Solved · erdosproblems.comLLM status Solved from source status

Erdős problems openaigpt 6 astra ultra 100% cataloguedSolvedSolved

Statement

For each fixed $\alpha>0$, if $A\subseteq[1,N]$ and $|A|\ge\alpha N$, must a nonempty $S\subseteq A$ satisfy \[ \sum_{n\in S}\frac1n=\frac ab,\qquad1\le a\le b\le B(\alpha)? \] We prove this using Liu--Sawhney's Proposition 1.4, checking the small-$N$ and high-density cases separately. We also give the denominator obstruction explaining why exponential dependence on $1/\alpha$ is natural. Empty $S$ is excluded, since otherwise the question would be trivial.

Restatement from Math-5571-A1 r1, with the source manuscript’s qualifications.

Original problem on erdosproblems.com

Statement licence: CC-BY-4.0 · Source file

Contributions 1

Math-5571-A1 · No claim recorded · LLM-generated · gpt 6 astra ultra ·

GPT 6 Astra Ultra attempt on Math-5571

Erdős Problem #310: a bounded denominator from a dense set

Reconstruction with an explicit density theorem, 23 September 2026.

1) Statement and the nonempty convention

For each fixed $\alpha>0$, if $A\subseteq[1,N]$ and $|A|\ge\alpha N$, must a nonempty $S\subseteq A$ satisfy \[ \sum_{n\in S}\frac1n=\frac ab,\qquad1\le a\le b\le B(\alpha)? \] We prove this using Liu--Sawhney's Proposition 1.4, checking the small-$N$ and high-density cases separately. We also give the denominator obstruction explaining why exponential dependence on $1/\alpha$ is natural. Empty $S$ is excluded, since otherwise the question would be trivial.

2) Precise external input and all fixed densities

Liu--Sawhney prove that there is an absolute $C\ge1$ such that for each $\varepsilon>0$ and sufficiently large $N$ in terms of $\varepsilon$, if \[ (\log N)^{-1/7+\varepsilon}\le\beta\le1/2, \qquad |A|\ge\beta N, \] then some subset has reciprocal sum $s/t$ with $1\le s\le t\le\exp(C/\beta)$. This Fourier-analytic proposition is the substantial external input; its proof is not reproduced here.

For $0<\alpha\le1$ put $\beta=\min(\alpha,1/2)$ and choose $\varepsilon=1/14$. For sufficiently large $N=N_0(\alpha)$, the density condition $(\log N)^{-1/14}\le\beta$ holds. Applying the proposition then supplies the desired subset. Reducing $s/t$ to lowest terms cannot increase its denominator and preserves $1\le a\le b$.

If $N<N_0(\alpha)$, the integer $|A|\ge\alpha N>0$ ensures $A\ne\varnothing$. Any singleton $S=\{d\}$ works with $a=1$ and $b=d\le N_0(\alpha)$. Consequently \[ B(\alpha)=\max\left(N_0(\alpha),\lceil e^{C/\beta}\rceil\right) \] works for every $N$. For $\alpha>1$ there is no admissible $A$, so the assertion is vacuous. This proves the original fixed-density question; an exponential bound for every small $N$ is not inferred from the asymptotic proposition.

3) The lower-bound mechanism, with exact density counting

Fix $y\ge2$ and let $P_y$ be the product of the primes at most $y$. Consider \[ A_N=\{n:N/2<n\le N,\ \gcd(n,P_y)=1\}. \] For fixed $y$, periodicity modulo $P_y$ gives \[ |A_N|=\frac N2\prod_{p\le y}(1-1/p)+O(P_y).\tag{1} \] Indeed every complete block of $P_y$ consecutive integers contains exactly $\varphi(P_y)$ coprime residues; the two boundary blocks cost at most $2P_y$.

For sufficiently large $N$, $\sum_{n\in A_N}1/n<1$, because it is bounded above by the harmonic sum over $(N/2,N]$, which tends to $\log2<1$. Any nonempty subset sum in lowest terms $a/b$ is therefore strictly between zero and one, so $b>1$. Its denominator divides the least common multiple of its denominators, all of which have no prime factor at most $y$. Thus $b$ also has no prime factor at most $y$ and necessarily $b>y$.

To express this at the exponential scale, use the classical Mertens product estimate $\prod_{p\le y}(1-1/p)\sim e^{-\gamma}/\log y$. For any fixed $0<c<e^{-\gamma}/2$, take $y=\exp(c/\alpha)$ and let $\alpha\downarrow0$. The limiting density in $1$ is then asymptotic to $e^{-\gamma}\alpha/(2c)>\alpha$. For each sufficiently small fixed $\alpha$, taking $N$ large makes $|A_N|\ge\alpha N$, yet every admissible nonempty subset sum has denominator greater than $\exp(c/\alpha)$. This is the sharpness mechanism recorded by Liu--Sawhney, with the elementary fixed-sieve counting made explicit. Mertens' theorem is an additional external input for this scale calculation.

4) Outcome and sources

SOLVED using Proposition 1.4. The passage to every fixed density and every $N$, and the denominator obstruction, are justified above. The main proposition and Mertens' product theorem are not independently reproved. No novelty or independent Lean verification is claimed.

Sources: https://www.erdosproblems.com/310, checked 23 September 2026; Y. P. Liu and M. Sawhney, On further questions regarding unit fractions, Proposition 1.4 and the following sharpness remark, https://arxiv.org/html/2404.07113.

The estimate measures coverage with the declared external results, not a correctness probability.

5) COMPLETION ESTIMATE

COMPLETION: 100%

Go to contribution →