Erdős Problem #784
Status Solved · erdosproblems.comLLM status Solved from source status
Erdős problems openaigpt 6 astra ultra 100% cataloguedSolvedSolvedStatement
For $A\subseteq[1,x]\cap\mathbb N$, assume $\sum_{a\in A}1/a\leq C$, and count integers up to $x$ divisible by no element of $A$. The answer depends on both $C$ and whether $1$ is allowed. Literally, the proposed conclusion holds for $0<C<1$ and fails for $C\geq1$. With the intended additional condition $1\notin A$, it holds for $0<C\leq1$ and fails for every $C>1$. We give the elementary part and derive the complete classification from an explicitly stated established sieve theorem. See https://www.erdosproblems.com/784.
Restatement from Math-5894-A1 r1, with the source manuscript’s qualifications.
Original problem on erdosproblems.com
Statement licence: CC-BY-4.0 · Source file
Contributions 1
GPT 6 Astra Ultra attempt on Math-5894
Erdős Problem 784: the reciprocal budget in a small sieve
Statement and claim.
For $A\subseteq[1,x]\cap\mathbb N$, assume $\sum_{a\in A}1/a\leq C$, and count integers up to $x$ divisible by no element of $A$. The answer depends on both $C$ and whether $1$ is allowed. Literally, the proposed conclusion holds for $0<C<1$ and fails for $C\geq1$. With the intended additional condition $1\notin A$, it holds for $0<C\leq1$ and fails for every $C>1$. We give the elementary part and derive the complete classification from an explicitly stated established sieve theorem. See https://www.erdosproblems.com/784.The literal statement.
Write $X=\lfloor x\rfloor$. The union bound gives \[ \#\{m\leq X:\exists a\in A,\ a\mid m\} \leq\sum_{a\in A}\lfloor X/a\rfloor\leq CX. \] For $0<C<1$, at least $(1-C)X\gg_C x$ integers survive, which is stronger than any requested fixed logarithmic loss. For $C\geq1$, the choice $A=\{1\}$ has admissible reciprocal sum and leaves no survivors. Thus the literal statement cannot include $C=1$.The intended convention $1\notin A$.
Define $H_C(x)$ as the minimum survivor count over $A\subseteq\{2,\ldots,\lfloor x\rfloor\}$ with reciprocal sum at most $C$. The external input is Weingartner's Theorem 5, incorporating and sharpening results of Ruzsa and Saias: for every fixed $Z>1$, uniformly for $1\leq C\leq Z$, \[ H_C(x)\asymp_Z\frac{x^{\exp(1-C)}}{\log x}. \tag{1} \] Source: A. Weingartner, The Schinzel--Szekeres function, Research in Number Theory 11 $2025$, Paper 63, https://arxiv.org/abs/2310.13038. The theorem uses exactly the reciprocal constraint and excludes $1$. Its minimum may equivalently range over larger sets, because elements exceeding $x$ affect neither the survivor count nor the usefulness of the budget. The substantial sieve estimates behind $1$ are external, not reproved here.At $C=1$, $1$ gives $H_1(x)\asymp x/\log x$, so $c=1$ works. For $C<1$, the earlier union bound still applies. For fixed $C>1$, set $\theta=\exp(1-C)<1$. The upper bound in $1$ supplies admissible sets with at most $K_Cx^\theta/\log x$ survivors. For every fixed real $c>0$, \[ \frac{K_Cx^\theta/\log x}{x/(\log x)^c} =K_Cx^{-(1-\theta)}(\log x)^{c-1}\longrightarrow0. \] Consequently no exponent depending only on this $C$ can give the proposed universal lower bound. The minimizing set may depend on $x$; this is allowed by the quantifiers in the question.