Physicist Nathaniel Craig says GPT-6 Astra helped turn his strategy for a candidate lattice definition of the Standard Model into a construction and detailed proof. He described the collaboration in a November 15 guest post on OpenAI’s new science blog. Source
The post explains work proposed in Craig’s preprint, A gauge-invariant measure for lattice chiral fermions, submitted to arXiv on October 6, 2026. OpenAI introduces Craig as a visiting scientist at the company and leader of the particle theory group at the University of California, Santa Barbara. Source Source
What the proposal tries to do
The Standard Model describes elementary particles and three fundamental forces: the strong force, the weak force and electromagnetism. Craig’s goal is a nonperturbative definition, meaning one that doesn’t rely on treating interactions as small corrections. A lattice represents spacetime as a grid, offering a way to define the theory at a particular grid spacing and, potentially, calculate its predictions on computers. Source
One difficulty is the weak force’s different treatment of left- and right-handed fermions, a class of particles that includes quarks and electrons. As Craig explains, straightforward lattice rules can introduce unwanted partners of the opposite handedness. Bonus particles are not a helpful extra. Source
Earlier methods called overlap fermions address that doubling problem, but leave a further difficulty: setting a consistent mathematical phase for the fermions. Changing the force fields in steps that return them to their starting configuration can leave the fermion states with a different phase. The phase rule must compensate for that shift around every allowed loop, Craig explains. Source
Craig’s proposed construction compares allowed fermion states on successively finer grids, combining those comparisons into a phase rule for the original grid. He says the construction works at finite lattice spacing under a bound on field strengths at the grid scale. Those conditions are part of the claimed result, not optional fine print. Source Source
Who did what
Craig says he supplied the initial strategy but hadn’t managed to prove that it worked across all the required cases. He credits Astra with suggesting how to carry force fields onto finer grids, showing that comparisons across those grids converge to a definite answer, and taking the lead in developing the detailed proof. That division of labor is his account, published by the model’s developer—not an independently audited measure of Astra’s research performance. Source
Craig names lattice field theorists Maarten Golterman, Andreas Kronfeld and Yigal Shamir as researchers who looked over the proof and asked probing questions. He says their involvement helped build his confidence in the result; the post does not supply direct assessments from those physicists. Source
What remains unresolved
Craig says he is still considering whether the construction satisfies reflection positivity at finite lattice spacing. He describes this as a condition that helps connect lattice calculations to a quantum theory in which probabilities are conserved. Source
The construction also leaves open the question of defining the interacting theory in continuous spacetime. Craig says expert discussion is needed to determine whether the prescription has all the desired properties. Source
He calls the method “far more an answer in principle than an answer in practice.” Substantial work remains before it becomes useful for realistic lattice simulations, he says. Source
Comments