New THOR AI Framework Solves Century Old Statistical Physics Problem Hundreds of Times Faster Than Supercomputers

A new AI framework called THOR uses tensor networks to calculate atom behavior 400 times faster than traditional supercomputer simulations.

By: AXL Media

Published: Mar 16, 2026, 4:25 AM EDT

Source: Information for this report was sourced from The University of New Mexico

A Paradigm Shift in Computational Materials Science

A collaborative team from The University of New Mexico and Los Alamos National Laboratory has introduced a transformative computational framework titled Tensors for High dimensional Object Representation, or THOR. This AI system addresses one of the most enduring challenges in statistical physics by efficiently calculating configurational integrals, which are essential for predicting how materials behave under various mechanical and thermodynamic conditions. By integrating tensor network algorithms with machine learning, THOR allows scientists to bypass traditional, slow moving simulations, providing a direct mathematical solution to problems that have historically required weeks of processing time on the world’s most powerful supercomputers.

Overcoming the Curse of Dimensionality

For nearly a century, the primary obstacle in materials physics has been the "curse of dimensionality," a phenomenon where calculations become exponentially more complex as more variables are added. Traditional methods, such as molecular dynamics and Monte Carlo simulations, attempt to estimate atom movements by modeling trillions of individual interactions over time. According to Professor Dimiter Petsev of UNM, solving these integrals directly was previously considered impossible because the dimensions involved are so vast that classical techniques would theoretically take longer than the age of the universe to finish. THOR AI changes this reality by using tensor train cross interpolation to break these massive datasets into smaller, manageable, and connected mathematical pieces.

Harnessing Symmetry for Unprecedented Speed

The efficiency of the THOR framework is further enhanced by its ability to detect and utilize the inherent crystal symmetries within a material’s structure. By identifying these repeating patterns, the AI dramatically reduces the total amount of computation needed to reach a precise answer. This specialized approach ensures that calculations which once demanded thousands of hours of supercomputer labor are now finalized in mere seconds. Senior AI scientist Boian Alexandrov notes that this level of speed does not come at the cost of accuracy, providing a new benchmark for statistical mechanics that can be applied to complex scenarios like extreme pressures or sudden phase transitions in metals.