Physicist Nathaniel Craig has worked with the GPT-6 Astra artificial intelligence model on a candidate lattice definition of the Standard Model of particle physics. The construction, described in an October preprint, could address a foundational question that has remained open for more than 50 years, but the result still requires further expert scrutiny.
Craig is a visiting scientist at OpenAI, leads the particle theory group at the University of California, Santa Barbara, and is a permanent member of its Kavli Institute for Theoretical Physics. He says the underlying idea was his, but GPT-6 Astra was instrumental in developing it and took the lead on the detailed proof.
What is missing from the Standard Model?
The Standard Model describes the known elementary particles and three of the four fundamental forces: the strong force, the weak force and electromagnetism. Its predictions agree with experiments across eighteen orders of magnitude, from the low energies of hydrogen’s 21-centimeter radio emission to the high energies of collisions at the Large Hadron Collider.
Calculations often use perturbation theory: a problem is first solved without interactions, which are then added back as small corrections. The method can, for example, predict the strength of the electron’s magnetism to within a few parts in a trillion. It does not work for every phenomenon, however. The proton’s mass comes mainly from the motion and interactions of its constituent quarks and the gluons that carry the strong force. If that force were switched off, the proton would fall apart.
There is as yet no nonperturbative definition of the full Standard Model in continuous spacetime that includes interactions from the outset; nor is it settled whether the interacting theory is mathematically well defined. Lattice field theory offers a possible alternative by replacing spacetime with a grid of points. The grid spacing sets the shortest distance that can be resolved and permits a rigorous nonperturbative definition at that spacing, as well as direct computer calculations. The method has already proved successful for the strong force, including in determining the proton’s mass from first principles.
Chirality and the doubling problem
Putting the full model on a lattice is difficult because it is chiral: the weak force treats left- and right-handed fermions, including quarks and electrons, differently. The strong force and electromagnetism, by contrast, treat the two types alike. The most straightforward lattice rules produce an unwanted right-handed partner alongside a left-handed fermion, giving the wrong particle content for the weak force.
Craig illustrates this with the polarization of light. If left-circularly polarized light is sent between two ordinary mirrors, reflection reverses its direction of travel and changes it to right-circular polarization. Both polarizations then appear in the box, much as an unwanted fermion partner appears on the lattice. In the analogy, the handedness of light refers to photon helicity, rather than fermion chirality.
Special layered chiral mirrors can instead preserve the left-handed polarization of reflected light. A similar conceptual solution emerged in lattice theory: Paul Ginsparg and Ken Wilson developed the relevant lattice rules, while Rajamani Narayanan and Herbert Neuberger constructed what are known as overlap fermions. Martin Lüscher showed how to distinguish their left- and right-handed forms, and Yoshio Kikukawa and collaborators explored how the approach could describe weak interactions in the Standard Model.
The missing phase rule
Removing the unwanted fermions is not enough on its own: a phase must also be specified, and it depends on the force fields at every point on the lattice. Lüscher worked out a prescription for simpler theories, such as a chiral version of electromagnetism, but the question remained open for the full Standard Model for more than 25 years.
Craig again uses an optical analogy. Light between chiral mirrors builds up properly only if the phase of the wave returning from a round trip is aligned with light arriving from the source. The alignment can be adjusted by rotating one mirror relative to the other. The phase on the lattice is different in nature, but requires a comparable correction: if force fields are changed in steps and then returned to their starting configuration, the phase of the fermion states may end up different from its initial value. The correction must work for every allowed loop.
Craig’s idea was to represent the same force fields on a succession of finer grids. Comparing fermion states on two successive grids gives one part of the phase correction; adding those parts yields the rule for the original lattice. According to the proof, the correction works for every allowed loop, depends mainly on nearby fields when one point is changed, and gives the same answer for different mathematical descriptions of the same fields. The construction supplies the required fermion measure at finite lattice spacing, subject to a bound on field strengths at the scale of the grid.
The AI’s role and remaining questions
Craig says several rounds of work and discussions with outside experts were needed to turn the idea into a proof. GPT-6 Astra suggested how to transfer the force fields to finer grids, showed that the comparisons converge to a definite answer, and took the lead in developing the detailed proof. Lattice field theorists Maarten Golterman, Andreas Kronfeld and Yigal Shamir reviewed the proof and helped test the result through their questions.
For now, Craig says, the proposal is more an answer in principle than one ready for practical use. Expert discussion must determine whether it has all the desired properties, and substantial work remains before realistic lattice simulations are possible. One particular question is whether it satisfies reflection positivity at finite lattice spacing, a condition that helps connect lattice calculations to a quantum theory in which probabilities are conserved.
The construction also does not by itself settle the definition of the interacting theory in continuous spacetime. Craig says OpenAI’s recent proof concerning the existence of Yang–Mills theory supplies a missing element: according to the account, OpenAI used overlap fermions and the continuum limit to prove the existence of the simplest unified Yang–Mills theory containing the Standard Model’s gauge fields and fermions.



