Theoretical analysis of zirconium oxynitride/water interface using neural network potential

Abstract

Zr oxides and oxynitrides are promising candidates to replace precious metal cathodes in polymer electrolyte fuel cells. Oxygen reduction reaction activity in this class of materials has been correlated with the amount of oxygen vacancies, but a microscopic understanding of this correlation is still lacking. To address this, we simulate a defective Zr${}_7$O${}_8$N${}_4$/H${}_2$O interface model and compare it with a pristine ZrO${}_2$/H${}_2$O interface model. First, ab initio replica exchange Monte Carlo sampling was performed to determine defect segregation at the surface in the oxynitride slab model, then molecular dynamics accelerated by neural network potentials was used to perform 1000 of 500 ps-long simulations to attain sufficient statistical accuracy of the solid/liquid interface structure. The presence of oxygen vacancies on the surface was found to clearly modify the local adsorption structure: water molecules were found to adsorb preferentially on Zr atoms surrounding oxygen vacancies, but not on the oxygen vacancies themselves. The fact that oxygen vacancy sites are free from poisoning by water molecules may explain the activity enhancement in defective systems. The layering of water molecules was also modified considerably, which should influence the proton and O${}_2$ transport near the interfaces which is another parameter that determines the overall activity.

arXiv comment64 pages, 30 figures

References in corpus (4)

• Efficient and Accurate Machine-Learning Interpolation of Atomic Energies in Compositions with Many Species
• Insights into lithium manganese oxide-water interfaces using machine learning potentials
• Room temperature structure and energetics of water-hydroxyl layers on Pt(111)
• Facilitating {\it ab initio} configurational sampling of multicomponent solids using an on-lattice neural network model and active learning

Cited by in corpus (3)

• Atomistic Simulations of Oxide-Water Interfaces using Machine Learning Potentials
• Free energy profiles for chemical reactions in solution from high-dimensional neural network potentials: The case of the Strecker synthesis
• Benchmarking of Oxygen Adsorption using TPD Spectroscopy for Accurate DFT Prediction of ORR on Anatase Titanium Dioxide (101)