AI and the World's Hard Problems · Theoretical computer science · arXiv · 2026-10-22
ORSMT: Unleashing the Potential of Satisfiability Modulo Theories for Semantic Verification and Repair in LLM-Based Optimization
ORSMT tackles semantic verification in LLM-based optimization modeling, where checking whether the written mathematics, solver code and returned report describe the same problem requires semantic comparison. It provides a typed semantic interface and exact SMT relation queries, proving or refuting feasible-region and objective relations, localizing refutations, and reconstructing one artifact from another. Across 3,034 problems and six generator sizes it improves the model-code match rate in all 72 combinations, and protected selection raises the aggregate rate from 43.2% to 88.2% and the correct-and-matched share from 25.2% to 40.2%.