When UC Santa Barbara physicist Nathaniel Craig wanted to put the Standard Model of particle physics on a grid, he already had a strategy. He would take the same force fields and look at them on a series of ever finer grids. He still needed to show that the procedure would work across all the cases the theory had to cover. In a Nov. 15 account published by OpenAI, Craig, a visiting scientist at the company, says GPT-6 Astra suggested how to carry the fields from one grid to the next and led the work on the proof.

The account concerns a preprint Craig submitted to arXiv on Oct. 6. The paper carries his name and credits the model extensively in its acknowledgments. The new account gives readers a way to see how he divided the work between a physicist with a plan, a model that turned it into a proof, and other researchers who pressed him on the details.

Why use a grid in the first place?

Rows of black cabinets on the Mira supercomputer floor
Mira at Argonne National Laboratory in 2013. The machine later supported U.S. lattice calculations of the strong force.Photo: Argonne National Laboratory / U.S. Department of Energy via Wikimedia Commons, CC BY 2.0

The Standard Model describes elementary particles and three forces: electromagnetism, the strong force and the weak force. It has survived demanding experimental tests. A different problem arises when physicists try to give the full theory a mathematical definition that works without assuming its interactions are small enough to treat as corrections.

Putting space and time on a grid gives physicists a shortest distance to work with. It also makes some problems suitable for computer simulation. A 2016 project described by Argonne National Laboratory used that approach to study the strong force. With access to computers such as Mira, the researchers could run calculations on very fine grids at the physical mass of the pion, a particle made from a quark and an antiquark. Argonne reports that the project produced 4,800 new samples of the fields that carry the strong force for use in calculations.

The weak force presents a different problem. It treats particles differently depending on their handedness, a property that can distinguish a particle from its mirror counterpart. Simple grid rules can introduce an extra particle of the opposite handedness. Earlier work gave researchers a way around that doubling and worked out a way to assign consistent weights in simpler theories. Extending that last step to a class that includes the Standard Model remained difficult.

Where Astra entered the proof

Kohn Hall at UC Santa Barbara in evening light
Kohn Hall, home of UC Santa Barbara’s Kavli Institute for Theoretical Physics, in 2006. Craig is a member of the institute.Photo: Jonathan Pfeiffer via Wikimedia Commons, CC BY 2.0

Craig proposed comparing the allowed particle states on successive grids. He says Astra found a way to represent the same force fields on each new grid and showed that the comparison converges to a definite answer as the grid gets finer. The resulting rule gives physicists a consistent way to assign weights to configurations at a fixed grid spacing, under restrictions on the fields at that scale.

The preprint's acknowledgments say Craig combined sections he wrote with extensive revisions to the model's proof. He and colleagues proposed tests, including a check in the proof assistant Lean, that Astra carried out. The arXiv entry includes a separate verification report. Craig says researchers Maarten Golterman, Andreas Kronfeld and Yigal Shamir examined the proof and asked questions. The paper and its supplements let other researchers inspect the claims and the checks the author reports. A Lean check applies to the mathematical statements encoded for it. Researchers still have to ask whether those statements capture what the physics requires.

A candidate theory, before a practical simulation

There is already one public response in the research literature. In a Nov. 3 preprint, physicist Edward Witten cited Craig's construction and described a different route that builds on some of the same ideas. Witten identifies a barrier to using standard statistical sampling for his variant. Craig also lists realistic simulations as future work for his own approach. His paper constructs a candidate at finite grid spacing. It leaves open further mathematical checks, including a condition that would help connect the construction to a quantum theory with conserved probabilities. The October paper also leaves open what happens when its grid spacing goes to zero. A successful calculation on a particular grid and a mathematical definition for continuous space and time are separate milestones.