Research

Xie Lab develops intelligent representations of matter that connect physical law, multiscale simulation, and science-grounded AI to steer the discovery and making of useful materials.

Research

Modeling materials as they actually operate

Research centers on intelligent representations of matter: models should accelerate discovery while preserving mechanism, symmetry, thermodynamics, uncertainty, and physical interpretability.

AI for materials

Machine learning force fields, graph neural networks, language-model workflows, Bayesian optimization, and materials informatics for accelerated materials design.

Theory and methods

Improved first-principles based methods for systems with strong vibronic coupling, anharmonicity, disorder, defects, and finite-temperature effects.

Mechanisms and phenomena

Quantum and statistical mechanics for charge, mass, and energy transport; excited states; spontaneous symmetry breaking; and multiscale structure-property relations.

Materials discovery

Rational design, high-throughput screening, and machine learning to identify materials for energy, optical, optoelectronic, ferroic, and environmental uses.

Research philosophy

Representation as scientific infrastructure

Scientific progress is limited not only by data or compute, but by how matter is represented. The lab builds representations that remain faithful to physical law and useful for AI-driven design.

Mechanism before black boxes

Models are evaluated not only by prediction quality, but also by whether they reveal useful mechanisms, invariances, and design rules.

Structure-preserving intelligence

Symmetry, geometry, thermodynamics, kinetics, and uncertainty are treated as design constraints, not afterthoughts.

Knowledge that guides discovery

Scientific understanding should help validate mechanisms, navigate design spaces, and guide the discovery and making of useful materials.

Conceptual framework

From physical structure to intelligent action

The group connects materials physics and AI through representations that encode structure, symmetry, thermodynamics, kinetics, uncertainty, and design feasibility.

Nature

Atoms, materials systems, experiments, simulations, and data.

Physical structure

Symmetry, interactions, energy landscapes, disorder, and kinetics.

Representation

Latent spaces, graph features, constraints, uncertainty, and design variables.

Intelligence

Equivariant learning, generative models, Bayesian optimization, and active learning.

Transformation

Materials design, mechanistic validation, and useful discovery.