Researchers from Radical AI have published a paper describing TorchSim, a next-generation open-source atomistic simulation engine built entirely in PyTorch. By rewriting the core
primitives of atomistic simulation within a modern deep-learning framework, TorchSim delivers orders-of-magnitude acceleration for popular machine learning interatomic potentials
(MLIPs).
Designed for the new era of AI-enabled materials science, TorchSim unifies molecular dynamics, energy minimization, and gradient-based learning under one differentiable and GPU-native
platform. The result is speed, flexibility, and ease of integration with emerging machine learning atomistic models.
TorchSim is freely available as an open-source project, advancing the mission to democratize high-performance simulation for researchers worldwide.
Computational simulation/modeling
Not applicable
TorchSim: An efficient atomistic simulation engine in PyTorch
25-Oct-2025