Research

GPT-6 Astra Helps Physicist Propose Lattice Definition of Standard Model

Physicist Nathaniel Craig says GPT-6 Astra took the lead on a detailed proof behind an October preprint proposing a candidate lattice definition of the Standard Model, a problem open for over 50 years.

A step toward defining the Standard Model on a lattice
A step toward defining the Standard Model on a latticeTumisu / pixabay
By Marcus Bennett7 min read

Why it matters

  • Physicist Nathaniel Craig published an October 2026 preprint proposing a candidate nonperturbative lattice definition of the Standard Model, a problem open for more than 50 years.
  • Craig states GPT-6 Astra took the lead in developing the detailed proof and suggested the key construction for carrying force fields onto finer grids.
  • Lattice experts Maarten Golterman, Andreas Kronfeld, and Yigil Shamir reviewed Astra's proof.
  • The phase-alignment problem for the full Standard Model had been open for more than 25 years.
  • Open questions remain, including reflection positivity at finite lattice spacing and usefulness for realistic simulations.

Physicist Nathaniel Craig has published an October 2026 preprint, prepared with the aid of GPT-6 Astra, that proposes a candidate nonperturbative lattice definition of the Standard Model — a foundational problem that has stood open for more than fifty years. Craig leads the particle theory group at the University of California, Santa Barbara, is a permanent member of its Kavli Institute for Theoretical Physics, and is currently a visiting scientist at OpenAI. The preprint, described in the first guest post on OpenAI's science blog, marks one of the most concrete claims to date that a frontier AI model contributed substantively to a proof in theoretical physics.

Why the Standard Model lacks a definition

The Standard Model describes the known elementary particles and three of the four fundamental forces: the strong force, the weak force, and electromagnetism. Its success rests on agreement between theory and experiment across eighteen orders of magnitude, from the low energies of hydrogen's 21-centimeter radio emission to collisions at the Large Hadron Collider.

Yet the theory's predictive machinery relies on perturbation theory — the technique of imagining interactions are switched off, solving exactly, then adding interactions back as a series of small corrections. Craig notes this approach has let physicists predict the strength of the electron's magnetism to within a few parts in a trillion.

Much of the theory's richness lies outside that regime. The proton is made of three quarks, but its mass comes primarily from the motion and interactions of quarks and gluons (the strong force carriers) inside it. Turn off the strong force and the proton falls apart entirely. For the full Standard Model in continuous spacetime, Craig writes, physicists do not have a nonperturbative definition, or even an understanding of whether the interacting theory is mathematically well defined.

What a lattice changes

Lattice field theory — discretizing spacetime into a grid of points — offers a path. The grid spacing defines the shortest distance the theory can resolve, and a finite grid permits a rigorous nonperturbative definition at a given spacing. It also enables direct computer calculation. Over recent decades the approach has delivered a quantitative, first-principles understanding of the proton's mass and major progress on the strong force.

The obstacle to extending it to the full Standard Model is chirality. A chiral theory looks different under reflection in space, like a single left-handed glove whose mirror image is a right-handed glove. The strong force and electromagnetism treat left-handed and right-handed fermions (particles such as quarks and electrons) identically. The weak force does not. It treats left- and right-handed fermions differently, which makes the Standard Model a chiral theory.

Putting chiral theories on a lattice is, in Craig's words, profoundly challenging. Attempt to place a left-handed fermion on the lattice, and the most natural rules produce a right-handed partner you never asked for. This "doubling" means a straightforward lattice formulation of the Standard Model gives the wrong particle content for the weak force — one reason a complete nonperturbative definition has been missing since the theory took shape.

What is the mirror analogy for the doubling problem?

Craig explains the problem through an analogy with light. A laser producing left-circular light (electric field rotating counterclockwise around the direction of travel), placed in a box with ordinary mirrors, does not stay purely left-circular: each mirror reflection flips the polarization, so the box ends up containing both hands of light. That mirrors the appearance of doublers on the lattice — you supplied one hand, and the lattice supplied the other.

A physical fix exists. Chiral mirrors, built from films whose molecules line up in spiraling layers, reflect left-handed light back left-handed and let right-handed light pass through. The phenomenon occurs even in nature — some scarab beetles carry a similar spiral in their shells, and chiral mirrors have existed for more than a decade.

The lattice analog has a distinguished lineage. Decades ago Paul Ginsparg and Nobel laureate Ken Wilson worked out lattice rules analogous to the chiral-mirror condition. Rajamani Narayanan and Herbert Neuberger then constructed overlap fermions — the necessary lattice definition of fundamental particles. Martin Lüscher showed how to distinguish left-handed from right-handed overlap fermions on the lattice, and Yoshio Kikukawa with collaborators explored how this could describe weak interactions in the Standard Model. One piece still remained missing.

The missing piece: setting the phase

In the optical analogy, even a box with chiral mirrors needs phase alignment. Left-circular light bouncing back and forth meets fresh light from the laser; if their electric fields are in step they reinforce, and if they are half a cycle apart they cancel. Rotating one chiral mirror about the beam compensates for the phase accumulated over a lap, producing a strong beam.

On the lattice, the analogous phase depends on what the Standard Model's force fields are doing at every grid point. Worse, carrying the allowed fermion states around a loop of force-field configurations — returning the fields to where they started — can leave a different phase. Lüscher worked out the phase prescription for a simpler class of theories, such as a chiral version of electromagnetism, but the alignment for the full Standard Model stayed open for more than 25 years.

Craig's idea was to represent the force fields on a succession of finer grids. Comparing allowed fermion states on two successive grids yields one part of the phase correction for a small change in the forces; summing those parts gives the phase rule on the original grid. The proof, he writes, shows the rule gives the right correction around every allowed loop, that the correction for a local change depends mainly on nearby fields, and that different mathematical descriptions of the same fields give the same answer. The result supplies the measure for the Standard Model at finite lattice spacing, under a bound on field strengths at the scale of the grid.

What role did GPT-6 Astra actually play?

Craig is explicit about the division of labor. "Astra helped turn that strategy into a construction and took the lead in developing the detailed proof," he writes. He adds that GPT-6 Astra "was essential to turning the refinement idea into a working construction": it suggested the right way to carry a given set of force fields onto finer grids, showed the comparisons on finer grids would converge to a definite answer, and led the detailed proof.

Human experts vetted the result. Lattice field theorists Maarten Golterman, Andreas Kronfeld, and Yigal Shamir reviewed Astra's proof and asked what Craig calls probing questions that built confidence in the result. "Their experience with the problem and the questions they asked were invaluable as the construction took shape," he writes.

What remains unresolved?

Craig frames the construction as a beginning, not an end. Scientific dialogue among experts is required to decide whether the prescription has all the desired properties, and he describes the method as "far more an answer in principle than an answer in practice" — substantial work lies ahead before it can support realistic lattice simulations.

His immediate open question is whether the construction satisfies reflection positivity at finite lattice spacing, a condition connecting lattice calculations to a quantum theory in which probabilities are conserved. The construction also leaves open the definition of the interacting theory in continuous spacetime. There, Craig points to OpenAI's recent proof of the existence of Yang-Mills theory — of which the strong and weak forces are examples — as the missing puzzle piece, one that enabled a proof of the existence of the simplest unified Yang-Mills theory including the Standard Model's gauge fields and fermions, obtained by starting with overlap fermions on the lattice and taking the continuum limit. "I'm optimistic these developments will lead to new progress in our understanding of the Standard Model," he writes.

The stakes for AI-assisted science are straightforward: if peer review confirms the construction, a frontier model will have taken the lead role in a proof addressing a problem open since the Standard Model's formulation more than fifty years ago.

Source: OpenAI News

Share this article:

More from Marcus Bennett

Marcus Bennett

Show full bio

Senior reporter covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

616 articles

Related articles

  1. GPT-5.2 derives new gluon amplitude formula in physics first
  2. OpenAI Preprint Shows Graviton Amplitudes Once Assumed Zero
  3. OpenAI Claims GPT-5 Helped Close an Erdős Problem and Speed Immunology Breakthroughs
  4. Frontier AI Models Now Solve 35 Percent of Hard Theory Problems
  5. A Quantum Physicist Is Using OpenAI's o1 to Tackle Physics' Biggest Questions

« Previous article