Table of Contents
Physicist Nathaniel Craig says GPT-6 Astra helped develop a candidate lattice definition of the Standard Model, taking the lead on a detailed proof needed to turn his initial strategy into a mathematical construction. The work addresses a longstanding difficulty in describing the theory's particles and forces without treating their interactions as small corrections.
Craig, who leads the particle theory group at the University of California, Santa Barbara, and is a visiting scientist at OpenAI, published the proposal in an October preprint. He has now detailed the collaboration in the first guest post on OpenAI's science blog, which the company introduced on X. The new blog will feature research results and practical accounts of how scientists use and evaluate AI models.
The result concerns the mathematical foundations needed to study the full Standard Model on a spacetime grid. Craig calls it a candidate definition, with further expert examination and substantial work needed before it becomes useful for realistic computer simulations.
The Phase Problem Behind the Construction
A lattice replaces continuous spacetime with discrete points. The distance between those points sets the shortest scale the model can resolve, giving physicists a controlled framework for defining interactions and calculating their consequences numerically.
That approach has been particularly successful for the strong force. Extending it to the full Standard Model is harder because the weak force distinguishes between left- and right-handed fermions, the class of particles that includes quarks and electrons. This property, called chirality, must survive the transition from continuous equations to a grid.
Straightforward lattice formulations introduce unwanted particle partners, a problem known as fermion doubling. Earlier work established ways to preserve the required chiral structure: Paul Ginsparg and Kenneth Wilson developed a modified lattice symmetry relation, and Rajamani Narayanan and Herbert Neuberger developed overlap fermions. Martin Lüscher subsequently established a framework for constructing chiral gauge theories on a lattice, while Yoshio Kikukawa and collaborators explored applications to the Standard Model's weak interactions.
Craig's construction targets a remaining ingredient: a consistent phase for the fermion measure, the mathematical prescription used when summing over fermionic contributions to the theory. Phase determines how quantum contributions reinforce or cancel one another.
The difficulty appears when the force fields change through a sequence of configurations and eventually return to their starting point. Following the allowed fermion states around that loop can leave an accumulated phase even though the fields have returned to their original configuration. A valid measure needs a prescription that compensates consistently around every allowed loop.
Lüscher solved this problem for a simpler class of theories, including a chiral version of electromagnetism. Craig says the corresponding prescription for the full Standard Model remained unresolved for more than 25 years.
Turning Grid Refinement Into a Proof
Craig's strategy compares the same force fields represented on progressively finer lattices. Each comparison between neighboring grid scales supplies part of the required phase correction. Adding those contributions produces a phase prescription for the original lattice.
Making that idea work requires more than showing that a few comparisons give sensible answers. The sequence must converge as the grids become finer, and the resulting rule must satisfy the consistency conditions needed to reconstruct the fermion measure.
The preprint presents a construction for a class of four-dimensional chiral gauge theories that includes the Standard Model. It uses overlap fermions and restricts the allowed force fields through an admissibility condition, a bound on field strengths at the lattice scale.
Within that setting, the construction establishes an exponentially local current: the influence of distant fields on a local change falls off rapidly with distance. It also supplies the global consistency properties needed for a smooth, gauge-invariant fermion measure. Gauge invariance means that changing the mathematical description of the same physical fields does not change the answer.
Craig credits Astra with proposing how to transfer the force fields onto finer grids, establishing convergence, and leading development of the detailed proof. His own starting point was the refinement strategy, which he had not been able to prove worked across all the required cases.
The work involved several rounds of interaction with the model and discussions with outside specialists. Craig says lattice field theorists Maarten Golterman, Andreas Kronfeld, and Yigal Shamir examined the proof and asked questions that helped him assess the construction.
From a Lattice Definition to Physical Calculations
The motivation for a nonperturbative definition is practical as well as mathematical. Perturbation theory starts from a solvable system without interactions, then adds those interactions as successive corrections. It can produce extremely precise predictions when the corrections become progressively smaller.
But some phenomena depend on strong interactions from the outset. Most of the proton's mass comes from the motion and interactions of its quarks and gluons, rather than simply adding the quarks' masses. Turning off the strong force eliminates the bound proton itself. Lattice calculations have made it possible to calculate its mass from first principles without relying on that expansion.
A corresponding framework for the full Standard Model must also accommodate the chiral weak force. Craig's proposed measure addresses that foundational requirement at finite lattice spacing, rather than immediately supplying a computationally efficient simulation method.
He describes the result as "far more an answer in principle than an answer in practice." One specific question he is investigating is whether the construction satisfies reflection positivity at finite spacing. That condition helps connect the lattice formulation to a quantum theory with conserved probabilities.
There is also a separate problem in taking the grid spacing toward zero to recover continuous spacetime. Craig connects this to OpenAI's work on the existence of Yang–Mills theories and a unified completion of the Standard Model. That program concerns constructing interacting quantum field theories in the continuum, including a unified theory containing the Standard Model's gauge fields and fermions.