GPT-6 Astra disproves Erdős’s sum-distinct lower-bound conjecture

For every ε>0, arbitrarily large n admit an n-element sum-distinct set of positive integers with maximum element N≤ε·2^n. The formal statement excludes N=0 and gives no effective n as a function of ε.

Sources for this record

What was already known

Bohman’s cited construction reached N≤0.22002·2^n. Dubroff, Fox and Xu proved a lower bound of order 2^n/sqrt(n); Erdős asked for a uniform positive multiple of 2^n.

AI involvement AI-autonomous

GPT-6 Astra found the construction and wrote Lean code in a network-isolated run without human mathematical steering. Adamczewski packaged the repository; Bloom later explained the lattice view.

Benchmark and repository document no human guidance during the model proof search.

Importance Notable

The construction disproves the proposed uniform exponential lower bound.

Limitations

Bloom reinterpreted the model’s rational-matrix proof as a lattice argument, but its detailed translation is not fully documented as a human check.

Technical verification Formal (Lean 4 with Mathlib and Comparator)

Source-reported Lean build and Comparator match to the trusted challenge in a pristine environment; no broader informal theorem is silently inferred.

Formal checking

compiled proof, trusted statement comparison, and standard-axiom check

Checker
Lean 4 kernel and FrontierMath Erdős Comparator
Scope
Problem #1 compared theorem
Basis
Reported by the linked checker
Evidence source

Checking a formal statement does not establish that it matches the original problem.

Human understanding Partial

Bloom explains the lattice route but the detailed passage from the determinant construction to sum-distinct sets is not fully digested in a human proof.

Correction and publication history