Many scientific and societal challenges in chemical transformation, energy science, and materials discovery depend on understanding how complex materials and reactive interfaces form, function, and evolve under realistic conditions. The relevant states are often not fixed or fully known: they depend on local chemistry, configurational complexity, synthesis and processing history, and operating conditions.
Our group develops physics-grounded AI and atomistic/agentic models to make these complex states predictive and actionable. We combine first-principles modeling, statistical mechanics and kinetics, machine learning, and experiment–model integration to determine what is physically accessible, what controls chemical function, what experiments reveal, and how uncertainty should guide the next scientific action.
Thurst A: Reactive Catalytic Interfaces
How do local chemical environments and evolving interfacial structures determine catalytic function under operating conditions?
Catalytic interfaces are rarely static. Local coordination, bonding, electronic structure, adsorbates, and environmental conditions can create chemically distinct active sites, while reconstruction and degradation can change which sites are actually present during operation.
We develop physics-grounded AI together with first-principles, atomistic, thermodynamic, and kinetic modeling to connect local chemistry with active-site populations, reactivity, and stability. Rather than describing catalysts only through idealized structures or composition-averaged descriptors, we seek predictive principles that explain both what individual environments can do chemically and which environments become relevant under realistic conditions.
Thurst B: Disordered Energy Materials
How do chemical disorder, configurational complexity, and synthesis determine which material states form and how they function?
Many energy materials cannot be represented adequately by a single ideal crystal structure. Chemical ordering, local correlations, defects, and configurational ensembles can reshape bonding, stability, transport, and reactivity, while synthesis and operating conditions determine which of these states are actually accessible.
We develop disorder-aware and disorder-native AI that connects atomistic representations with statistical mechanics, thermodynamics, kinetics, and experimental information. Our goal is to understand when configurational complexity must be represented explicitly, predict physically accessible ordered and disordered states, and establish links among composition, synthesis, material state, and chemical function.
Thurst C: Closed-Loop Agentic Discovery
How can experimental evidence and uncertainty be transformed into mechanistic understanding and better scientific decisions?
Complex chemical and material states are often observed only indirectly through incomplete and multimodal measurements. This makes scientific discovery not only a prediction problem, but also an inference and decision problem: what states are consistent with the evidence, what remains uncertain, and what calculation or experiment would be most informative next?
We develop physics-grounded digital twins and reasoning-enabled agentic AI that connect atomistic models, experimental observables, mechanistic knowledge, and uncertainty-aware decision-making. Working closely with experimental collaborators, we seek closed-loop workflows in which simulations and measurements iteratively constrain one another, allowing AI to support not only prediction but also interpretation, hypothesis refinement, and selection of the next scientific action.
Across these directions, our goal is not simply to make prediction faster. We seek AI and computational frameworks that connect physical mechanisms, accessible chemical and material states, experimental evidence, and uncertainty—turning prediction into understanding and understanding into informed scientific action.