Weekly

News about Lean, the theorem prover and programming language.

Week of Oct 15, 2026

Edition 30: 21 items published for Sep 24 to Oct 15, 2026.

Lean core

Lean 4.35.0 ships vcgen, lake check, and bundled independent kernels

Lean 4.35.0, published on 2026-10-12, adds vcgen, a faster and more general successor of mvcgen, plus lake check and lake comparator, which check a project with Lean's kernel and optionally several independent kernels now included in the toolchain.

Checked against: v4.35.0 release noteslean4 v4.35.0

AI and formalization

OpenAI publishes 719 model-produced math manuscripts, about 42% Lean-formalized

OpenAI's openai/math repository holds manuscripts and proof artifacts produced by an internal OpenAI model, organized into 372 families, with 300 of 719 top-line results formalized in Lean. An October 7, 2026 update withdrew three manuscripts after a sign error and revised 14 others.

Checked against: openai/math

Papers and events

Renaissance Philanthropy and XTX Markets announce $17.1 million in AI for Math grants

The AI for Math Fund selected 22 applications representing 30 organizations from more than 470 submissions, bringing its total commitments to $35.1 million. Winners include TorchLean, a Lean-native framework for verified machine learning, and a project turning a Lean 4 visual proofwriting prototype into open-source infrastructure for reading AI-generated proofs.

Image: renaissancephilanthropy.org

Checked against: renaissancephilanthropy.org

AI and formalization

AI proof-search agent resolves nine open Erdős problems, verified in Lean

A paper in Science by George Tsoukalas, Swarat Chaudhuri and coauthors describes an agent for formal proof search that autonomously resolved nine of 353 open Erdős problems and proved 44 of 492 On-Line Encyclopedia of Integer Sequences conjectures. A basic agent alternating LLM-based generation with Lean-based verification replicated the Erdős successes.

Checked against: science.org

AI and formalization

Machine-checked Lean 4 proof of the general Courtade–Kumar inequality released

A repository by @dpwoodru contains a complete, machine-checked Lean 4 proof of the general Courtade–Kumar inequality (the most informative Boolean function conjecture), formalizing the paper by Z. Chen, A. Gohari, A. Javanmard, H. Lin, V. Mirrokni, C. Nair and D. P. Woodruff. The whole proof spans 45,500 modules.

Checked against: dpwoodru/general-courtade-kumar-lean

AI and formalization

Google's Gemini 4 Argon writes a Lean-verified quantum arcsin implementation

Google's launch post for Gemini 4 Argon (dated 2026-10-13) reports an arcsin implementation, a building block in recent quantum simulation algorithms, written by Argon and formally verified in Lean. It cuts spacetime cost by more than 100x compared to a published quantum CORDIC baseline.

Image: blog.google

Checked against: blog.google

Verification

Trail of Bits proves Miden core library arithmetic correct with AI-generated Lean proofs

During an audit of the Miden VM, a zero-knowledge VM with its own assembly language, Trail of Bits had agents build a Lean model of the VM executor, producing 95 correctness proofs covering all binary arithmetic components of the core library. The work identified two subtle bugs that the existing unit test suite did not catch.

Image: blog.trailofbits.com

Checked against: blog.trailofbits.com

Verification

Emina Torlak presents SymCert, a Lean framework for verified Cedar policy analyses

SymCert, implemented in Lean, builds verified SMT-based analyses of Cedar policies, an open-source authorization language. In the FMCAD 2026 paper, five analyses used in a Cedar analysis service at Amazon Web Services were modeled and proved in one work day, and three extensions for new Cedar features took 4 to 8 days each.

Checked against: paper (pdf)

Verification

Flapjack ports the verified Pancake compiler from HOL4 to Lean

Flapjack is bringing the Pancake compiler, originally verified in HOL4, and its proofs to Lean, with a path compiling a fragment of Pancake to RISC-V where all stages except parsing and AST conversion have a correctness theorem. A MIPS32 backend targets the Ziren zkVM, but its output has been run only on Ziren's Lean model.

Checked against: flapjack-lang.dev

Lean core

Lean reference manual documents vcgen, the successor of mvcgen in Lean 4.35

A pull request by Sebastian Graf documents the vcgen tactic and its Std.WP framework, adds the tutorial "Verifying Imperative Programs Using vcgen", and moves the mvcgen chapter and tutorial to their own files. The Kraken project already uses vcgen to prove that FFmpeg's FMA3 dot-product loop, over two arrays that may alias, terminates without faulting.

Checked against: reference-manual#927zrlk/kraken

Lean core

Joachim Breitner and Lean FRO describe the proof behind con-leche checker

con-leche, a checker for Lean 4, comes with a proof in Lean that every environment it accepts has a model in set theory, so no accepted environment holds a proof of False. A new document gives a pen-and-paper account of that argument on a simplified fragment of the checker.

Checked against: leanprover.github.io

Lean core

Sebastian Ullrich removes Lean's 255-object-field limit per constructor

A merged pull request by Sebastian Ullrich removes the limit of 255 object fields per constructor, and thus per structure. When a constructor has more, the compiler keeps the first 254 in the constructor object and stores the rest in a chain of spill objects, with one pointer followed per spill object on access.

Checked against: lean4#15516

Lean core

Julia Markus Himmel reworks order instances across Lean core types

Julia Markus Himmel redefined the order on String as the lexicographic order of its UTF-8 encoding, made Min Float follow the IEEE-754-2019 minimum operation, and added missing order instances for BitVec, List, Array, Vector and ByteArray. Sofia Rodrigues fixed leaks and refactored the libuv bindings.

Checked against: lean4#15581lean4#15257lean4#15274lean4#15304lean4#15329lean4#15175lean4#15545

Lean core

Kim Morrison proposes supply-chain audits for Mathlib dependency updates

A Mathlib pull request by Kim Morrison adds a supply-chain audit step to the hourly update_dependencies.yml workflow, posting diffs, high-risk files touched and suspicious added lines for each changed dependency. The diffs come from paranoid-lake-update, which uses bare clones and git only, so nothing from upstream is executed.

Checked against: mathlib4#43911