Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated


Statement and provenance

If a1<a2<⋯a_1<a_2<\cdots are positive integers and ai+1−ai=O(1)a_{i+1}-a_i=O(1) as i→∞i\to\infty, then some finite set of indices II satisfies ∑i∈I1/ai=1\sum_{i\in I}1/a_i=1.

This is the consequence of Bloom's Theorem 2 identified by the cached Problem 299 page, accessed 2026-09-05. It is not a separately numbered result in Bloom's paper. It disproves the existence asked for in Problem 299.

Rewritten derivation

Choose integers i0≥1i_0\ge1 and H≥1H\ge1 so that ai+1−ai≤Ha_{i+1}-a_i\le H for every i≥i0i\ge i_0. Induction gives ai0+j≤ai0+Hja_{i_0+j}\le a_{i_0}+Hj for all j≥0j\ge0. Consequently the set A={ai:i≥1}A=\{a_i:i\ge1\} satisfies, for N≥ai0N\ge a_{i_0},

∣A∩[1,N]∣≥1+⌊N−ai0H⌋.|A\cap[1,N]| \ge1+\left\lfloor\frac{N-a_{i_0}}H\right\rfloor.

Indeed, all indices i0+ji_0+j through the indicated floor have ai0+j≤Na_{i_0+j}\le N, and strict increase makes their values distinct. Divide by NN and take the lower limit to obtain d‾(A)≥1/H>0\underline d(A)\ge1/H>0. Hence d‾(A)>0\overline d(A)>0 as well. Theorem 2 supplies a finite S⊆AS\subseteq A of reciprocal sum one. Each element of SS has a unique index because the sequence is strictly increasing; taking those indices gives the required II.

Existing formalization

The Google DeepMind declaration for Problem 299 uses a strictly increasing positive sequence and an eventual big-O gap bound. As inspected on 2026-09-05, its body is sorry, with an external formal-proof tag linking the Bloom–Mehta Lean 3 density proof. That link supports the density theorem; no separate completed Lean declaration for the bounded-gap reduction was identified in the inspected source. No formal proof was built here.

Dependencies

Theorem 2 only; its full proof is kept on its own page and is not repeated here.

Bears on