TBNAI
IneqAgent: Agentic Inequality Proving in Lean with Symbolic Computation
Publication: cs.AI updates on arXiv.org
Author: Gaolei He, Jinhui Li, Ruobing Zuo, Xia Zeng, Zhengfeng Yang
Source published: 16/10/2026, 04:00:00 UTC
Briefing created: 16/10/2026, 04:58:27 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 16/10/2026, 04:03:24 UTC.
Briefings
IneqAgent is an agentic framework combining LLM reasoning with symbolic computation to prove inequalities in Lean4. It uses Monte Carlo Tree Search for lemma construction and integrates backends like sum-of-squares and Bernstein. On four benchmarks, it solved 175 of 186 problems (94.09%), outperforming other methods in success rate and inference cost efficiency. The system also produced Lean-verified strengthenings of Dittert's conjecture in dimensions four and five, and certified the Holens--Djokovic (4,4) inequality.
Key points
- IneqAgent solves 175 of 186 benchmarks (94.09%) using DeepSeek-V4-Flash, the highest success rate among evaluated methods.
- The framework achieves higher success rates at low LLM inference cost thresholds compared to Direct Agent using the same backbone.
Context
Formal inequality proving in Lean4 requires exact computation, often involving intermediate lemmas and goal reformulations. IneqAgent introduces an agentic framework combining LLM reasoning with symbolic computation to automate this process.
Limitations
Evidence is limited to four benchmarks; results are not generalizable beyond these specific cases. The study is preprint (arXiv:2610.20681v1) and does not include peer review or experimental validation on other systems.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 16/10/2026, 04:58:27 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeih5opirmu6c4rc3psegh25oxm3mjisfj44353xntt35mfsxkonofa