Teaching thermodynamic laws to AI unlocks a polymer modeling challenge

For more than half a century, materials scientists have struggled with how to simulate the complexity of polymer materials. An individual chain can comprise tens of thousands of atoms, a melt or composite contains billions, and the properties engineers actually care about, such as how an adhesive grips a surface, how a self-assembling block copolymer locks into a nanostructure, or how a biopolymer film stretches without tearing, emerge only over length and time scales that forcible atomistic simulation cannot reach.