LLM2Cert synthesizes certificates checked by SMT solvers and Lean
Original title: LLM2Cert: LLM Synthesis of Certificates with SMT/Lean Verification
The method verifies 81 of 105 DynaCertBench instances, 17.1 points above one-shot proposals without feedback.
Researchers present LLM2Cert, which proposes Lyapunov and barrier certificates and checks each proof obligation with an SMT solver or a Lean proof-repair agent. They also release DynaCertBench, 105 instances spanning stability, region of attraction, forward invariance, and safety, including higher-dimensional and control-constrained systems. The paper reports 81 verified instances and 22 certificates that All-SMT leaves unresolved. Code and data are on GitHub.
Unconfirmed: Verification rates are the authors' reported results on their own benchmark.
- DynaCertBench contains 105 instances, including 80 systems curated from the literature.
- LLM2Cert verifies 81 of 105 instances, reported as 77.1 percent.
- Feedback lifts results by 17.1 percentage points over one-shot proposals.
- Of 81 fixed certificates, 22 remain unresolved by All-SMT.
- Code and data are at github.com/FormalCert-AI/llm2cert.
- Products
- LLM2CertDynaCertBench
- Organizations
- FormalCert-AI
From the original article
1 min read at huggingface.co
Lyapunov functions and barrier certificates give state-space guarantees of stability and safety, but synthesizing and verifying them---especially for nonlinear, high-dimensional systems---remains an open challenge.
By Zhipeng Cao, Peiyang Song, Nikolaus Vertovec · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.