arxivcs.LGcs.AIcs.CE2026-07-11
Reinforcement Learning with Verifiable Physics: Post-training LLMs with Continuous Rewards
Pengfei Cai, Utkarsh Utkarsh, Alan Edelman, Christopher Vincent Rackauckas, Rafael Gomez-Bombarelli
Partial differential equations (PDEs) are foundational to modeling in science and engineering, but constructing reliable numerical solvers remains labor-intensive, demanding expert knowledge of discretization schemes, stability conditions, and boundary treatments. Recent work has…