The approximation threshold for metric k-median

OpenAI

For every fixed ε>0, a deterministic polynomial-time algorithm on finite rational metric k-median instances with specified candidate facilities opens at most k facilities and attains factor 1+2/e+ε. Together with known P≠NP hardness, this gives infimum approximation factor 1+2/e.

Sources for this record

What was already known

The best general candidate-facility approximation was about 2+ε, while a Max-k-Coverage reduction gave hardness approaching 1+2/e.

AI involvement AI-led

The manuscript names OpenAI as author and the release attributes these manuscripts to an internal model. The specific amount of human direction is not described.

OpenAI authorship and repository production statement attribute the claimed algorithm and proof to its internal model.

Importance Major

Claims to close the algorithm-versus-hardness gap for general candidate-facility metric k-median.

Limitations

The theorem gives every fixed ε>0; it does not claim an algorithm attaining exactly 1+2/e. Facility-restricted hardness is a separate question. The cited Lean documentation maps solution theorems to this claim; the comparator challenge files contain placeholders.

Technical verification Formal (Lean 4)

The repository reports a Lean formalization of the deterministic algorithm and the P≠NP infimum threshold. The published solution modules have claim-matching theorems main_finite_work and threshold; comparator challenge files contain intentional sorry placeholders, so the solution files and stated comparator mapping are the checking evidence.

Formal checking

Lean theorem proofs in OAI.Combinatorics.KMedian.Main.main_finite_work and Threshold.threshold mapped to published comparator challenges

Checker
Lean 4 kernel, as reported by OpenAI’s formalization catalogue
Scope
Deterministic polynomial-time (1+2/e+ε) metric k-median algorithm with at most k facilities, and conditional P≠NP infimum threshold
Basis
Reported by the linked checker
Evidence source

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

Human understanding Few

No named human explanation of the central directed comparison and derandomization is documented; Few is the evidence default.

Correction and publication history

2026-10-18 · publication

Claims to close the algorithm-versus-hardness gap for general candidate-facility metric k-median.

Read this published revision