Fermilab-led team builds AI tools to sharpen proton-scale physics for DUNE
FemtoMind is a multi-laboratory project headquartered at Fermilab. It is developing agentic AI tools for proton-scale physics calculations. It began in 2025, when researchers from Fermilab, Brookhaven National Laboratory and Thomas Jefferson National Accelerator Facility answered a call for proposals for the American Science Cloud. That cloud is part of the Department of Energy's Genesis Mission, an effort to build AI-driven scientific infrastructure across the national laboratories.
The work rests on lattice QCD, which places quark and gluon fields on a four-dimensional grid and uses high-performance computing to simulate protons, nuclei and other particles. The target is DUNE. Fermilab's top priority is sending a high-intensity neutrino beam about 800 miles to liquid-argon detectors at the Sanford Underground Research Facility in South Dakota by 2031. Over roughly the last five years, lattice simulations of a single neutrino striking a single proton have shown tensions with inferences from 1970s and 1980s bubble chamber experiments. The cause is unknown. Those old experiments used deuterium, with one proton and one neutron, so FemtoMind is now simulating neutrinos hitting both.
The early practical gain is in code. One collaborator used an AI agent to write and optimize assembly code, making the simulations three times faster over a couple of weeks. Michael Wagman says a human would have needed months. The team says hallucination is not a concern, because reference code verifies correct answers and the AI is used only to make code more efficient.
Why it matters for abundance: knowledge about how neutrinos interact with nuclei sets how much DUNE can extract from its data, and cheaper, faster simulation shortens the path to that understanding. The team is also exploring ways to share its tools with the wider lattice QCD community, which would spread the benefit beyond one project.
Caveats are real. No physics result from FemtoMind has been reported; the project is moving from initial exploration to practical tools. The threefold speedup is one collaborator's account relayed by Wagman, not independently documented. The report is a Fermilab press piece, so it is self-reported, and the team says its infrastructure ideas are changing quickly.
Source: Fermilab
Many Minded