arXiv cs.AIMonday · May 25, 2026FREE

RMA: an Agentic System for Research-Level Mathematical Problems

Research Math Agents (RMA), presented in arXiv:2605.22875, is an agentic framework designed for automated reasoning on research-level mathematical problems. Unlike prior work focused on competition math or formal theorem proving, RMA targets problems requiring long-horizon reasoning, literature grounding, and iterative proof refinement. The framework decomposes proof solving into specialized modules for problem analysis, literature search, fair comparison, knowledge-bank construction, and proof verification, coordinated by initializer, proposer, and verifier agents through shared structured memory. These agents operate in a multi-role, multi-round workflow, collaboratively generating, refining, and verifying candidate proofs. RMA was evaluated on the First Proof benchmark, consisting of ten research-level problems contributed by expert mathematicians. Through comprehensive expert evaluation, RMA solved eight out of ten problems, outperforming strong baselines including GPT-5.2R and Aletheia.