Theory: Explainable AI for Feynman Integral Reductions

by Tongzhi Yang (U of Zurich)

Monday 28 Apr 2025, 15:00 → 16:00 Europe/Berlin

SR2

Feynman integrals are central to precision collider physics, with their evaluation often limited by the complexity of integration-by-parts (IBP) reductions. In this talk, we present a new method to optimize IBP reductions using explainable artificial intelligence (AI) techniques. By leveraging the FunSearch algorithm—which combines large language models with genetic algorithms—we develop and identify efficient priority functions that drastically reduce the number of required seeding integrals and boost computational efficiency. Once found, these optimized functions can be applied universally to Feynman integrals at any loop order or multiplicity, without further AI input. As AI is used only for optimization, the IBP results remain exact. We validate the approach on various multi-loop integrals, including challenging five-loop self-energy integrals, showing significant speedups—over 3000× in some cases—while maintaining precision and broad applicability.