AI solves a century-old problem in seconds: Materials physics calculations accelerated hundreds of times

An international team from the University of New Mexico and Los Alamos National Laboratory has developed an AI system that dramatically accelerates one of the most difficult calculations in statistical physics—configuration integrals. These integrals describe how particles interact and are essential for predicting material properties, from phase transitions to behavior under extreme pressure. The work was published in Physical Review Materials.

A problem unsolved for decades

Configuration integrals mathematically describe all possible arrangements of particles in a system. They underpin much of thermodynamics and condensed matter physics. However, computing them directly is nearly impossible: as the number of particles grows, complexity increases exponentially—this is the well-known “curse of dimensionality.”

For decades, scientists relied on approximate methods such as molecular dynamics, Monte Carlo simulations, and variational approaches. All of these require enormous computational resources and often provide only approximate results.

Project lead Boyan Alexandrov explained that configuration integrals describing particle interactions are extremely difficult to compute, especially in materials science problems involving high pressures or phase transitions.

How THOR AI works

The new system, called THOR AI (Tensors for High-dimensional Object Representation), uses tensor networks—a powerful mathematical framework that represents massive multidimensional data as compact, interconnected structures.

THOR breaks a complex integral into a sequence of much simpler tasks, automatically detects symmetries in crystal lattices, and applies tensor interpolation. As a result, calculations that previously took thousands of hours on supercomputers can now be completed in seconds—without loss of accuracy.

The team tested the algorithm on several classic problems:

• copper under normal and high pressures
• crystalline argon in extreme conditions
• phase transitions in tin

In all cases, the results matched those of traditional methods, while computation time was reduced by more than 400 times.

What this changes

THOR AI can be integrated with modern machine learning models (such as neural network potentials) that describe atomic interactions. This opens the door to rapid simulation of materials under virtually any conditions—from extreme pressures inside planets to high temperatures in nuclear reactors.

The development could significantly accelerate the creation of new materials, including:

• superconductors
• aerospace alloys
• next-generation batteries
• materials for fusion energy

The authors note that it is now possible, for the first time, to directly and quickly compute what was previously considered nearly impossible.

In brief

Scientists from the University of New Mexico and Los Alamos National Laboratory developed THOR AI, a system that solves configuration integrals—key calculations in materials physics—hundreds of times faster than traditional methods.

Using tensor networks, the system detects symmetries and completes computations in seconds instead of thousands of hours, while maintaining full accuracy.

This breakthrough could significantly accelerate the development of new materials. The study was published in Physical Review Materials.