Ripple framework uses AI to formally verify chemical reaction networks
AI agents found and fixed gaps in published proofs, proving Apéry's constant computable by CRNs.
Ripple, introduced by Ho-Lin Chen and Xiang Huang, is an open-source Lean 4 framework that formalizes the mathematics of computing real numbers with chemical reaction networks (CRNs). Built almost entirely by AI agents using publicly available models, the framework covers the full spectrum of models: from the GPAC/CRN continuum and CRN-computable reals, through the large-population-protocol (LPP) compilation pipeline, to a continuous-time Markov chain (CTMC) layer bridged to deterministic mean-field limits via three machine-checked versions of Kurtz's theorem. It also formalizes two major Turing-completeness results: the Bournez-Graça-Pouly GPAC construction and the Soloveichik-Cook-Winfree-Bruck stochastic-CRN universality theorem. The entire development is verified to depend on exactly three Mathlib foundational axioms, with no unproven assumptions (no "sorry").
Remarkably, Ripple exposed genuine, fixable gaps in previously published proofs, including the approximate-majority convergence argument and the LPP main theorem. The framework also produced new results: a fully machine-checked construction of Apéry's constant ζ(3) as a CRN-computable number via its holonomic generating function, and a reduction of Ramanujan's modular 1/π series to a sharp open problem. The AI-driven formalization ensures the entire workflow is reproducible, setting a new benchmark for combining automated reasoning with biological and molecular computing models.
- Ripple formalizes the entire GPAC/CRN continuum, compilation pipeline, and CTMC layer with three verified Kurtz theorems in Lean 4.
- AI agents exposed and fixed two genuine gaps in published proofs: approximate-majority convergence and LPP main theorem.
- First machine-checked construction of Apéry's constant ζ(3) as a CRN-computable number, plus a Ramanujan series open problem.
Why It Matters
Demonstrates AI's potential to rigorously verify unconventional computing models and uncover proof errors, advancing molecular programming.