AI-developed proof improves the permanent-approximation base

For every nonnegative rational n×n matrix, Anari’s formal development constructs one rational-output algorithm and an absolute constant c<√2 with a two-sided cⁿ approximation guarantee for the permanent. It also proves deterministic polynomial time in the ordinary binary input size.

Sources for this record

What was already known

Anari and Rezaei’s Bethe analysis gave a deterministic (√2)ⁿ approximation for nonnegative matrices. Earlier Bethe inequalities reconstructed in this project are inputs, not the new bound.

AI involvement AI-centered

Nima Anari supplied the high-level plan and guided the project. The disclosure credits ChatGPT 5.6 Sol Pro with the proof and Lean development, and reports Anari’s critical review of the result.

ChatGPT developed the retained proof and Lean formalization; Anari supplied the high-level attack and guidance, so the mathematical work is joint with a documented central model role.

Importance Notable

Improves the stated exponential approximation base and supplies a binary polynomial-time implementation.

Limitations

The accompanying manuscript was still in preparation at the reviewed code version. The formal theorem covers the stated algorithm and complexity, while manuscript priority claims are outside the proof. No named human explanation of the full new proof mechanism is documented.

Technical verification Formal (Lean 4.30.0/Mathlib 4.30.0 with Comparator, Lean kernel replay and axiom audit + human)

The cited repository reports a closed Lean theorem with no project-specific axioms or omitted proof obligations, including the finite-precision optimization routine and ordinary binary polynomial-time claim. Its metadata reports fresh Lean builds, an axiom audit, Comparator and kernel replay, plus Anari's critical author review. These are source-reported checks of the selected theorem, not checks reproduced in this review; manuscript priority claims are outside Lean.

Independent checking

Full Lean build, theorem dependency and axiom audit, Comparator statement comparison and kernel replay

Checker
Lean kernel and Comparator in the project-reported pinned build
Scope
BeyondBethe.theoremOne and the matching Palomar.beyond_bethe_permanent declaration: rational c^n approximation with c < sqrt(2) and ordinary binary polynomial time
Basis
Reported by the linked checker

Author verification and repeated critical mathematical passes over the listed proof components

Checker
Nima Anari, author-reviewer
Scope
Theorem 1, including its approximation statement, constants, boundary cases, reconstructed inputs, numerical algorithm and binary complexity claim
Basis
Reported by the linked checker
Evidence source

Human understanding Few

Anari's plan and verification show involvement and checking, and the repository maps proof components to Lean, but the reviewed sources do not provide a named human explanation of the full new mechanism. Few is the evidence default, not a finding that Anari understands little.

Correction and publication history

2026-10-18 · publication

Improves the stated exponential approximation base and supplies a binary polynomial-time implementation.

Read this published revision