An approach for pushing large language models beyond computational math toward rigorous, proof-based mathematical analysis. It contributes DEMI-MathAnalysis, a curated corpus of proof problems in real analysis (sequences and limits, infinite series, convex functions) drawn from classic problem books, plus a step-by-step guiding framework that structures the model's reasoning. Fine-tuning open models (e.g. Llama 3.2, Qwen2) on this dataset together with the framework yields markedly more logical, complete, and rigorous proofs, targeting the gap left by benchmarks that focus on computation and avoid formal proof.
Mathematical problem-solving is a key field in artificial intelligence (AI) and a critical benchmark for evaluating the capabilities of large language models (LLMs). While extensive research has focu…