ML-Accelerated Sampling for Realistic Catalytic Interfaces

Heenen Group

Heterogeneous catalysis is governed by interfaces, yet the structures that actually operate under reaction conditions remain poorly understood. In our group, we develop ML-accelerated sampling strategies to close the gap between idealized computational models and the complexity of real catalytic surfaces, with the goal of achieving (semi)quantitative agreement with experiment.

Catalytic interfaces are inherently dynamic since surface composition, adsorbate coverage, and oxidation state evolve continuously under operando conditions. Capturing this behavior in atomistic simulations demands both high accuracy and extensive conformational sampling — a combination that conventional methods cannot provide. We address this through advanced sampling algorithms, including enhanced sampling and rare-event methods, that are tightly coupled to machine-learned interatomic potentials (MLIPs). This combination enables us to resolve slow structural transformations, reconstruct free energy landscapes, and follow surface evolution on time and length scales relevant to experiment.

A core part of our methodological work focuses on the sampling algorithms, with MLIP training strategies developed in direct response to their demands. Reliable MLIPs for reactive, multi-element interfaces require active learning workflows in which uncertainty quantification guides the iterative refinement of the potential alongside the sampling process itself. This coupling is not incidental. Accurate free energy estimates and thermodynamically consistent ensemble averages demand that both the potential and the sampling protocol are developed in concert, ensuring that efficiency gains never come at the cost of physical correctness.

We apply these tools to systems where interface complexity is the central scientific problem, including operando surface evolution, interface reactivity, stability, and deactivation under reaction conditions relevant to electrocatalyst surfaces and film growth during chemical vapor deposition.

Selected Publications

Selected recent publications are listed below (a full list can be found under Publications):

Operando Characterization and Molecular Simulations Reveal the Growth Kinetics of Graphene on Liquid Copper During Chemical Vapor Deposition, V. Rein. et al., ACS Nano (2024)

wfl Python toolkit for creating machine learning interatomic potentials and related atomistic simulation workflows, E. Gelžinytė, et al., J. Chem. Phys. (2023)

Graphene at Liquid Copper Catalysts: Atomic-Scale Agreement of Experimental and First-Principles Adsorption Height, H. Gao, et al., Adv. Sci. (2022)

The mechanism for acetate formation in electrochemical CO(2) reduction on Cu: selectivity with potential, pH, and nanostructuring, H. H. Heenen, et al., Energy Environ. Sci. (2022)

On the Role of Long-Range Electrostatics in Machine-Learned Interatomic Potentials for Complex Battery Materials, C. G. Staacke, H. H. Heenen, C. Scheurer, G. Csányi, K. Reuter, and J. T. Margraf, ACS Appl. Energy Mater. (2021)

Catalytic polysulfide conversion and physiochemical confinement for lithium-sulfur batteries, Z. Sun*, S. Vijay*, H. H. Heenen*, et al., Adv. Energy Mater. (2020)

Solvation at metal/water interfaces: An ab initio molecular dynamics benchmark of common computational approaches, H. H. Heenen, J. A. Gauthier, H. H. Kristoffersen, T. Ludwig, and K. Chan, J. Chem. Phys. (2020)

Multi-ion Conduction in Li₃OCl Glass Electrolytes, H. H. Heenen, J. Voss, C. Scheurer, K. Reuter, and A. C. Luntz, J. Phys. Chem. Lett. (2019)

Implications of Occupational Disorder on Ion Mobility in Li₄Ti₅O₁₂ Battery Materials, H. H. Heenen, C. Scheurer, and K. Reuter, Nano Lett. (2017)