A validated machine-learning interatomic potential framework for understanding reactive hematite-water interfaces toward next-generation functional materials

Reactive oxide–water interfaces play a fundamental role in photocatalysis, electrochemical energy conversion, corrosion, and environmental remediation. However, atomistic understanding of the structural and electronic processes governing these interfaces remains limited because conventional ab initio molecular dynamics (AIMD) is computationally prohibitive for the long timescales required to capture rare interfacial events. This study presents a validated machine-learning interatomic potential (MLIP) framework that combines first-principles Density Functional Theory (DFT), uncertainty-driven active learning, and on-the-fly machine learning to investigate the reactive hematite–water interface as a model oxide system.

High-fidelity reference configurations were generated using dispersion-corrected DFT and iteratively refined through active-learning molecular dynamics, enabling efficient exploration of complex interfacial configurations while retaining first-principles accuracy. Validation using force-error analysis, Bayesian uncertainty monitoring, long-time energy stability, and pair-correlation-function comparisons confirms that the trained potential faithfully reproduces the structural, energetic, and chemical characteristics of the underlying DFT model.

Long-timescale simulations reveal atomistic mechanisms governing molecular adsorption, water dissociation, proton transfer, surface hydroxylation, hydrogen-bond rearrangement, and hydration-layer restructuring at the hematite surface. Electronic-structure analysis of representative machine-learning-generated configurations further demonstrates adsorption-induced Fe–O/O–H hybridisation and interfacial electronic reconstruction, providing new insight into how water modifies the local electronic environment and reactive behaviour of hematite.

These findings establish a computationally efficient and transferable strategy for investigating complex oxide–water interfaces while advancing fundamental understanding of interfacial processes central to catalytic, electrochemical, and environmental materials. More broadly, the framework provides a scalable approach for accelerating the design of next-generation functional materials for sustainable energy, photocatalysis, and advanced interface engineering.

Dr Mary Taiwo Ajide is a Chemical Engineer and Founder of AIM²E® Lab, where she develops AI-enabled computational frameworks for predictive materials modelling. Her research integrates Density Functional Theory (DFT), machine-learning interatomic potentials (MLIPs), molecular dynamics, and multiscale modelling to investigate reactive interfaces, energy materials, and interfacial phenomena. She is also the Founder of the AIM²E® Network, promoting interdisciplinary collaboration and scientific engagement. Her work focuses on advancing computational materials science through machine learning and first-principles simulations for sustainable materials engineering, catalysis, electrochemical systems, and next-generation functional materials.