NSPI: Neuro-Symbolic Framework for Polynomial Inequality Proving
A new neuro-symbolic framework called NSPI combines large language models with symbolic computation to automate polynomial inequality proving. The LLM generates approximate sum-of-squares decompositions as conjectures, which are then refined via symbolic computation into exact certificates. This approach addresses scalability challenges in automated mathematical reasoning, where purely symbolic methods struggle with high-degree or multi-variable inequalities, while LLM-guided methods have been limited to small-variable competition problems. The work is published on arXiv under ID 2605.15445.
Key facts
- NSPI is a neuro-symbolic framework for polynomial inequality proving.
- It combines LLMs with symbolic computation.
- The LLM proposes approximate sum-of-squares decompositions.
- Symbolic computation refines these into exact certificates.
- Purely symbolic methods scale poorly with variable count or degree.
- LLM-guided methods have made progress on small-variable competition inequalities.
- The paper is on arXiv with ID 2605.15445.
- The approach targets scalability challenges in automated reasoning.
Entities
Institutions
- arXiv