Google AI expands mathematical solutions for nuclear fusion

A team of researchers used Google's artificial intelligence to extend a set of mathematical solutions related to the magnetic fields of experimental fusion reactors. The result concerns the calculations used to model how magnetic fields contain fuel heated to extreme temperatures, not a breakthrough that produces electricity or makes fusion commercially viable.

In magnetic confinement fusion reactors, very powerful magnetic fields hold an electrically charged, extremely hot gas together, preventing it from escaping before atomic nuclei can fuse and release energy. To design configurations capable of maintaining this confinement, physicists need accurate mathematical models of the magnetic fields, which feed into the computer simulations used to study plasma behaviour.

In 1967, physicist Harold Grad proposed a conjecture about the mathematical conditions needed for the equilibrium of three-dimensional magnetic fields, an idea that influenced decades of research into plasma configurations. In September 2026, researcher Matt Landreman of the University of Maryland published exact solutions that contradict that conjecture, with help from OpenAI's GPT-6 Astra Pro model. Those solutions mattered because they provided concrete mathematical examples that researchers could use to test software modelling magnetic confinement, instead of relying solely on approximations. However, all of Landreman's solutions shared a particular geometric symmetry, leaving no equivalent exact examples for configurations lacking that property.

AI agents at work in the Antigravity environment

A new study by Opal Issan, Jonathan Citrin, Matt Landreman and Brendan Tracey extended the existing solutions by introducing a parameter that breaks this symmetry. The new mathematical configurations give researchers additional reference cases for checking programs that calculate magnetic equilibrium in complex three-dimensional arrangements.

To reach this result, the team used Google's Gemini 4 Argon model through the Antigravity environment, where multiple AI agents explored Landreman's earlier results and looked for ways to extend them. The agents found ways to remove the symmetry restriction from Landreman's solutions, producing new configurations that researchers then verified. According to the team, promising mathematical candidates emerged within hours, a sign that AI can speed up research involving extensive calculations and interconnected mathematical conditions.

The initial discovery was only part of the process: researchers still had to check the derivations and establish that the proposed solutions were mathematically consistent. The group also used computational checks and AI assistance when preparing parts of the manuscript, while keeping responsibility for the published work. The immediate outcome is a more extensive set of exact mathematical solutions for researchers studying magnetic confinement, not evidence that artificial intelligence has solved nuclear fusion or removed its engineering obstacles.