bsd-developer/matrix-math-publish

1

stars

3

commits

Lean

primary language

Aug 25, 2026

updated

README

A machine-checked bound on the matrix multiplication exponent

This repository contains a fully machine-checked proof that

ω ≤ 10935605172023554189 / 2^62 = 2.371281376990…

where ω is the matrix multiplication exponent. The entire argument is a certificate: an exact rational witness (artifacts/) whose validity is verified independently by an exact Rust checker (crates/) and closed as a Lean 4 theorem (lean/). No floating-point computation is trusted anywhere in the claim.

Bound (ω ≤)YearTeamMachine-checked
2.3713392025Alman et al.no
2.3712813772026this workyes
2.3711772026Dupont et al. (Google DeepMind)no

The bound improves the best published level-three result and, to our knowledge, is the first bound on ω in this range whose statement and verification are closed inside a proof assistant. The stronger 2.371177 is a level-four floating-point claim without machine-checked verification; we make no claim against it.

Pre-Print: https://zenodo.org/records/22101463

Runbook

Prerequisites: rustup (the pinned toolchain in rust-toolchain.toml is picked up automatically), elan (Lean 4, pinned by lean/lean-toolchain), just, and Python 3.12.

just doctor
just build
just verify artifacts/omega-55148017.certificate.json
just prove  artifacts/omega-55148017.certificate.json
  • doctor checks the pinned toolchain versions.
  • build compiles the Rust checker and the Lean development.
  • verify runs the independent exact Rust verification of the certificate; it reports the certificate digest (55148017090a8883ab18bbd1316196fadc32b2f5f41cbf751d838d5c334f895f) and the certified value.
  • prove generates and checks the certificate-specific Lean theorem under profile CN and prints the axiom report.

The specification the checkers implement is in docs/specs/; the rounding and definitional-ω decisions are recorded in docs/adr/. The accompanying paper is in paper/.

Contributors

bsd-developer

3 commits

bsd-developer/matrix-math-publish

1

stars

3

commits

Lean

primary language

Aug 25, 2026

updated

README

A machine-checked bound on the matrix multiplication exponent

This repository contains a fully machine-checked proof that

ω ≤ 10935605172023554189 / 2^62 = 2.371281376990…

where ω is the matrix multiplication exponent. The entire argument is a certificate: an exact rational witness (artifacts/) whose validity is verified independently by an exact Rust checker (crates/) and closed as a Lean 4 theorem (lean/). No floating-point computation is trusted anywhere in the claim.

Bound (ω ≤)YearTeamMachine-checked
2.3713392025Alman et al.no
2.3712813772026this workyes
2.3711772026Dupont et al. (Google DeepMind)no

The bound improves the best published level-three result and, to our knowledge, is the first bound on ω in this range whose statement and verification are closed inside a proof assistant. The stronger 2.371177 is a level-four floating-point claim without machine-checked verification; we make no claim against it.

Pre-Print: https://zenodo.org/records/22101463

Runbook

Prerequisites: rustup (the pinned toolchain in rust-toolchain.toml is picked up automatically), elan (Lean 4, pinned by lean/lean-toolchain), just, and Python 3.12.

just doctor
just build
just verify artifacts/omega-55148017.certificate.json
just prove  artifacts/omega-55148017.certificate.json
  • doctor checks the pinned toolchain versions.
  • build compiles the Rust checker and the Lean development.
  • verify runs the independent exact Rust verification of the certificate; it reports the certificate digest (55148017090a8883ab18bbd1316196fadc32b2f5f41cbf751d838d5c334f895f) and the certified value.
  • prove generates and checks the certificate-specific Lean theorem under profile CN and prints the axiom report.

The specification the checkers implement is in docs/specs/; the rounding and definitional-ω decisions are recorded in docs/adr/. The accompanying paper is in paper/.

Contributors

bsd-developer

3 commits

Languages

Lean

97.3%

Rust

2.7%