We are an interdisciplinary computational & AI4Science research group in the Department of Materials Design and Innovation at the University at Buffalo, working across materials physics, surface chemistry, machine learning, and high-performance computing. We aim to train students and postdoctoral researchers who can work fluently across these traditional disciplinary boundaries and develop new approaches at the intersection of materials science and AI.
The complexity of materials cannot be reduced to composition or a single structure. It emerges from electronic structure and chemical bonding, competing atomic arrangements, chemical and structural disorder, evolving surfaces and interfaces, and the thermodynamic, kinetic, and environmental conditions that determine which states are actually realized. Our group develops materials-centric AI and atomistic modeling to connect these levels of complexity and uncover the physical principles that govern how materials behave and function.
We combine physics-based simulation and machine learning with experimental information, working closely with experimental collaborators to connect atomic-scale predictions with measurable materials behaviors. Rather than treating AI simply as a faster surrogate for conventional calculations, we seek models that incorporate physical knowledge, quantify uncertainty, and reveal interpretable mechanisms and transferable design principles grounded in experimental observations.
A recurring challenge is that any functionally relevant material is often not a single static, perfectly ordered structure. Local chemical environments, disorder, reconstruction, and operating conditions can give rise to many competing configurations whose populations and transformations collectively determine material behaviors. We therefore develop disorder-aware and disorder-native AI that asks not only how accurately a model can predict, but also whether it represents the material in a physically meaningful way and captures the ensembles that connect atomistic simulations with experimental reality.
Across these directions, prediction is not our endpoint. We use AI to extract transferable physical insights, connect atomistic descriptions with thermodynamics and kinetics, and increasingly support uncertainty-aware scientific decisions about what to calculate, measure, or explore next. In this way, our research seeks to move from prediction to understanding and from understanding to action, while keeping AI physically interpretable and experimentally grounded.
We gratefully acknowledge funding and computing support from:
