Semantic Search over 9 Million Mathematical Theorems
This addresses the difficulty for mathematicians and theorem-proving agents in finding precise theorems, though it is incremental as it applies semantic search to a new domain.
The paper tackled the problem of searching for specific mathematical theorems in large corpora by introducing semantic theorem retrieval over 9.2 million theorem statements, achieving substantial improvements in retrieval quality compared to existing baselines on queries from professional mathematicians.
Searching for mathematical results remains difficult: most existing tools retrieve entire papers, while mathematicians and theorem-proving agents often seek a specific theorem, lemma, or proposition that answers a query. While semantic search has seen rapid progress, its behavior on large, highly technical corpora such as research-level mathematical theorems remains poorly understood. In this work, we introduce and study semantic theorem retrieval at scale over a unified corpus of $9.2$ million theorem statements extracted from arXiv and seven other sources, representing the largest publicly available corpus of human-authored, research-level theorems. We represent each theorem with a short natural-language description as a retrieval representation and systematically analyze how representation context, language model choice, embedding model, and prompting strategy affect retrieval quality. On a curated evaluation set of theorem-search queries written by professional mathematicians, our approach substantially improves both theorem-level and paper-level retrieval compared to existing baselines, demonstrating that semantic theorem search is feasible and effective at web scale. The theorem search tool is available at \href{https://huggingface.co/spaces/uw-math-ai/theorem-search}{this link}, and the dataset is available at \href{https://huggingface.co/datasets/uw-math-ai/TheoremSearch}{this link}.