Research

Our department develops theoretical and computational methods to understand, predict, and design materials and molecular systems. We combine quantum-mechanical theory, physics-informed machine learning, and high-performance computing to investigate electronic structure, chemical space, and the dynamics of matter.

Selected published results are summarized as research highlights.

Research Areas

Scalable Machine Learning for Electronic-Structure Theory

The electronic structure of matter determines many fundamental properties of materials and molecules, but accurate quantum-mechanical simulations become computationally demanding for large systems. We develop scalable, physics-informed machine learning methods that extend the reach of first-principles calculations, particularly density functional theory (DFT). Our Materials Learning Algorithms (MALA) framework enables efficient predictions of electronic-structure properties for complex atomistic systems.

HighlightsMALA: scalable ML for electronic structureBridging the gap in electronic structureElectronic structure at any length scaleMachine learning for DFT: a review

Representation Learning for Chemical Space and Materials

Chemical space encompasses an enormous diversity of molecules and materials with distinct structures and properties. We develop machine learning approaches that identify meaningful representations of molecular and material systems, capturing underlying physical and chemical relationships. Our goal is to discover transferable patterns across chemical space, improve the prediction of material properties, and enable the systematic exploration and design of new compounds and functional materials.

Quantum Dynamics and Electronic Transport

Understanding how electrons evolve and interact with their environment is essential for describing the behavior of materials and nanoscale devices. We develop theoretical and computational methods for quantum dynamics and electronic transport, combining first-principles approaches with advanced machine learning techniques. Our research includes physics-informed methods for solving and inverting quantum-mechanical equations and neural-operator approaches for modeling electron dynamics.

HighlightsInverting the Kohn–Sham equations

Magnetic Materials and Spin-Lattice Dynamics

We combine first-principles calculations and machine-learned interatomic potentials to investigate the coupled dynamics of atomic structures and magnetic moments. Our approaches enable large-scale molecular-spin dynamics simulations to study magnetostructural phase transitions, spin-lattice interactions, and the stability of magnetic materials. Understanding these processes provides insights into functional magnetic materials and emerging technologies such as spintronics and ultrafast magnetic information storage.

HighlightsIron in Earth’s coreMagneto-elastic spin-lattice dynamics

THz-Driven Materials and Molecular Systems

We are developing theoretical and computational approaches to investigate the response of materials and molecular systems to terahertz (THz) electromagnetic fields. Our aim is to understand how THz excitation influences the coupled dynamics of electrons and atoms and to explore whether tailored electromagnetic fields can selectively control material properties and molecular transformations.