Machine Learning for Atomistic Simulations of Materials

Atomistic simulations play a central role in modern materials science, providing microscopic insight into the structure, dynamics, and properties of materials. First-principles methods such as density-functional theory have been enormously successful, but their high computational cost limits their applicability to small system sizes and short time scales. Over the past decade, Machine Learning (ML) approaches have allowed researchers to overcome these limitations by enabling accurate, transferable, and efficient models trained on first-principles data.

The rapid development of ML-based requires students to learn a new toolset for their research tasks. While these techniques allow simulations at unprecedented scales and accuracy, their successful application requires a solid understanding of the underlying physics and chemistry principles. In particular, questions of model construction, data efficiency, uncertainty, interpretability, and transferability are critical for ensuring predictive and physically meaningful simulations.

This summer school focuses on the theoretical foundations and practical implementation of ML for atomistic simulations of materials. Lectures will cover the construction of ML interatomic potentials, representations of atomic environments, learning algorithms, and validation strategies, as well as applications to diverse materials classes and phenomena. The program emphasizes the integration of ML with established atomistic simulation methods, highlighting how data-driven models can complement and extend first-principles approaches. The school aims to provide participants with a coherent framework to critically assess and apply ML techniques in atomistic modeling. Beyond technical training, the school seeks to foster discussion, collaboration, and long-term interaction within a rapidly growing research community that joins physics, chemistry, materials science, and data science.

Further information