Inside every battery sloshes a liquid electrolyte, and how ions jostle within it determines how much energy the battery delivers and how long it lasts. Previously, peeking into that crowd meant either running long experiments or launching ultra-precise calculations that devour weeks of computer time. Now there's a third path: a model that learns from millions of examples and produces the picture almost instantly.

The authors took two such models. One was trained on inorganic crystals: it knows the world of solids. The other was trained on the OMol25 set, which contains tens of millions of molecular configurations, including those found in electrolytes. It turned out that the 'molecular' model predicts density and X-ray structure factors noticeably more accurately than the 'crystal' one.

The model then showed how solvation changes with composition: which cation, which anion, how much salt, and the shape of the solvent molecule. For example, in glyme electrolytes, even a slight change in solvent topology noticeably rearranges ionic correlations. Such sensitivity to details previously eluded coarse models.

Now engineers can sift through electrolyte compositions on a computer like assembling Lego pieces, and only the best candidates need to be tested in the lab. This speeds up the search for safe, high-capacity sodium batteries, and sodium is far more abundant than lithium.