Physicists want to understand the universe at its smallest scales. One of their tools is lattice QCD, a mathematical method that places quarks and gluons - the building blocks of protons and nuclei - on a four-dimensional grid, so that supercomputers can simulate how these particles behave. A project called FemtoMind, headquartered at Fermilab, is now adding artificial intelligence to this work, according to Fermilab. The project is part of the American Science Cloud under the Department of Energy's Genesis Mission. It is developing agentic AI tools to speed up calculations and find new patterns in proton-scale physics. The project's name comes from the femtoscale, the size range of particles like the proton.
Preparing for DUNE
A major motivation is the Deep Underground Neutrino Experiment, or DUNE. Fermilab aims to deliver a high-intensity beam of neutrinos - nearly massless particles that very rarely interact with matter - to DUNE's detectors in South Dakota by 2031. To interpret DUNE's data, scientists must understand neutrino interactions with atomic nuclei very precisely. But in recent years, lattice QCD simulations of a neutrino hitting a single proton have disagreed with previous results. FemtoMind is therefore simulating neutrinos hitting a proton and a neutron together, matching the deuterium targets those older experiments used, so researchers can compare theory and experiment directly and trace the source of the tensions. One collaborator has already used an AI agent to write and optimize code, making the simulations three times faster in a few weeks.
The team stresses that scientists remain in charge: reference code checks that the AI's answers are correct, guarding against fabricated results. The researchers hope the tools will free up time for new ideas, lower the barrier to entry for computational physics, and help make use of new supercomputers coming online.
This matters because DUNE's discoveries will depend on theory that is ready before the beam turns on. By making simulations faster and more reliable, FemtoMind strengthens the foundation for neutrino science and offers a model for AI-assisted research in other fields.