Researchers from The University of New Mexico and Los Alamos National Laboratory have developed a groundbreaking computational framework that solves a century-old challenge in statistical physics. The Tensors for High-dimensional Object Representation (THOR) AI framework leverages tensor network algorithms to efficiently compute the configurational integral—a fundamental quantity in statistical mechanics that describes the thermodynamic and mechanical behavior of materials. The configurational integral, which captures the interactions among particles in a system, has long been considered intractable due to its extreme dimensionality. In materials science, this integral can involve thousands of dimensions, making traditional numerical integration methods impractical. Even the most powerful supercomputers would require computational times exceeding the age of the universe to solve such problems directly using classical approaches. Until now, scientists have relied on indirect methods like molecular dynamics and Monte Carlo simulations, which approximate the integral by simulating atomic motion over long time scales. While useful, these techniques are computationally intensive, often taking weeks to run, and still face inherent limitations in accuracy and scalability. The THOR framework overcomes these challenges by representing the high-dimensional integrand as a chain of interconnected, lower-dimensional components through a technique known as tensor train cross interpolation. A customized version of this method identifies key crystal symmetries, dramatically reducing computational complexity. As a result, the configurational integral can now be computed in seconds—without sacrificing accuracy. The framework integrates seamlessly with machine learning potentials, which model interatomic interactions with high precision. This combination enables accurate and scalable simulations across a wide range of physical conditions, including extreme pressures and phase transitions. Applied to real-world materials such as copper, argon under high pressure, and tin undergoing a solid-solid phase transition, THOR AI delivers results that match the best existing simulations from Los Alamos—but more than 400 times faster. This leap in performance marks a paradigm shift, replacing centuries-old approximations with first-principles calculations. “Solving the configurational integral directly was once thought impossible,” said Boian Alexandrov, a senior AI scientist at Los Alamos and project lead. “With THOR, we’re not just improving efficiency—we’re redefining what’s possible in materials modeling.” Dimiter Petsev, professor at UNM’s Department of Chemical and Biological Engineering and longtime collaborator with Alexandrov, recognized the potential of the method when first introduced. “This isn’t just a faster algorithm—it’s a new standard for accuracy and efficiency,” he said. “It allows us to benchmark and improve all other approaches.” Duc Truong, a Los Alamos scientist and lead author of the study published in Physical Review Materials, emphasized the broader impact: “THOR AI opens the door to accelerated discovery in materials science, physics, and chemistry. We’re no longer limited by computational bottlenecks.” The THOR Project is publicly available on GitHub, enabling researchers worldwide to access and build upon this transformative tool.