https://ai-smith.net Michael Brown (slapglif) documents his exploits in formal theorem proving, Lean 4 verification, and unsolved-adjacent mathematics. 11 theorems, zero sorries. en-us Tue, 23 Jun 2026 00:00:00 GMT CryptSmith kurosama112@gmail.com (Michael Brown) https://ai-smith.net/posts.html#11-theorems-zero-sorries https://ai-smith.net/posts.html#11-theorems-zero-sorries Sun, 22 Jun 2026 00:00:00 GMT How I formalized 11 open-problem-adjacent theorems — from Wieferich primes to Goldbach's conjecture — using native_decide and mathlib4. Every proof compiles. Every bound is exhaustive. Lean 4 Proof https://ai-smith.net/posts.html#prime-sieve-pitfall https://ai-smith.net/posts.html#prime-sieve-pitfall Sun, 22 Jun 2026 00:00:00 GMT When generating Finset literal sets for prime gap theorems, your sieve MUST extend to bound+gap, not just bound. Missing this cost me two build failures on Cousin and Sexy primes. Proof https://ai-smith.net/posts.html#e7-fine-structure-constant https://ai-smith.net/posts.html#e7-fine-structure-constant Sun, 22 Jun 2026 00:00:00 GMT Deriving α ≈ 1/137 from the exceptional Lie algebra E7. Not numerology — 62 experiments, >4σ significance, and a growing body of evidence that the fine-structure constant has geometric roots in exceptional groups. Theory https://ai-smith.net/posts.html#itt-mathlib-lottery-ticket https://ai-smith.net/posts.html#itt-mathlib-lottery-ticket Fri, 20 Jun 2026 00:00:00 GMT Intrinsic Ticket Theory — formalizing the Lottery Ticket Hypothesis in Lean 4 with mathlib. Proving that sparse subnetworks exist within randomly-initialized neural networks that can match the full network's performance. Lean 4 ML https://ai-smith.net/posts.html#native-decide-brute-force https://ai-smith.net/posts.html#native-decide-brute-force Thu, 18 Jun 2026 00:00:00 GMT Why native_decide is the unsung hero of formal verification. When your conjecture is decidable and finite, don't reach for induction — reach for native_decide. It compiles your proof to native code and runs it. Proof Lean 4 https://ai-smith.net/posts.html#unsolved-adjacent-strategy https://ai-smith.net/posts.html#unsolved-adjacent-strategy Sun, 15 Jun 2026 00:00:00 GMT I don't solve open problems — I prove their finite prefixes. This is the 'unsolved-adjacent' strategy: take an open conjecture, bound it, and formally verify the bound. Each theorem is a stepping stone toward the real thing. Theory