AI solved a century-old physics mystery in seconds

Scientists from the University of New Mexico and Los Alamos National Laboratory have presented a revolutionary computational method capable of tackling one of the most formidable problems in statistical physics — the calculation of configurational integrals. The system, named THOR AI (Tensors for High-dimensional Object Representation), has already demonstrated impressive results. The work has been published in the journal Physical Review Materials (PRM).

Configurational integrals play a key role in describing particle interactions and enable the prediction of thermodynamic and mechanical properties of materials. Without them, it is impossible to study phase transitions, the behavior of substances under extreme pressure, or to design new materials. Yet until now, the direct calculation of these integrals has remained a virtually unachievable task.

"The configurational integral, which describes particle interactions, is extremely difficult to compute, especially in materials science problems involving high pressures or phase transitions," explained the project leader, artificial intelligence specialist at Los Alamos National Laboratory, Boian Alexandrov.

The root of the problem is the so-called "curse of dimensionality." As the number of variables grows, computational complexity increases exponentially, stumping even the most powerful supercomputers. For decades, scientists have been forced to resort to approximate methods — molecular dynamics or the Monte Carlo method — which simulate atomic motion but consume enormous computational resources.

The new approach fundamentally changes the rules of the game, enabling such calculations to be performed directly. The THOR AI algorithm relies on tensor network methods — a mathematical technique that represents cumbersome multidimensional data as a set of interconnected yet significantly simpler elements.

The THOR AI system breaks down the most complex problems into a chain of compact computations using the tensor interpolation method. In addition, the algorithm automatically recognizes symmetries in the crystal structure of materials, further reducing the volume of required calculations.

The result is impressive: computations that previously took thousands of hours are now completed in mere seconds — without any loss of accuracy.

The developers tested the system on several materials — copper, crystalline argon under high pressure, and complex phase transitions of tin. In all cases, THOR AI produced results that matched the data from traditional labor-intensive simulations, yet completed the task more than 400 times faster.

Equally important is the fact that THOR AI easily integrates with modern machine learning models that describe atomic interactions. This opens up possibilities for analyzing material behavior under a wide variety of physical conditions.

According to the researchers, the new technology has the potential to significantly accelerate materials development and deepen the understanding of fundamental processes in physics, chemistry, and materials science.