High-Throughput Exploration of Ti–V–Nb–Mo Carbide MXenes Using Neural Network Potentials and Their Evaluation as Catalysts for Hydrogen Evolution Reaction

Abstract

Realization of a sustainable hydrogen economy in the future requires the development of efficient and cost-effective catalysts for its production at scale. MXenes (M_n+1X_n) are a class of 2D materials with ‘n’ layers of carbon or nitrogen (X) interleaved by ‘n+1’ layers of transition metal (M) and have emerged as promising materials for various applications including catalysts for hydrogen evolution reaction (HER). Their properties are intimately related to both their composition and their atomic structure. Recently, high entropy MXenes were synthesized, opening a vast compositional space of potentially stable and functionally superior materials. Detailed atomistic modeling enables us to systematically explore this extensive design space, which is otherwise infeasible in experiments. We have developed a Neural Network Potential (NNP) to model (Ti_xV_yNb_zMo_p)_n+1C_n MXenes (x+y+z+p = 1; n = 1,2,3) by training against Density Functional Theory (DFT) data in an active learning fashion. We then used the developed NNP to perform hybrid Monte Carlo-Molecular Dynamics (MC-MD) simulations to identify thermodynamically stable compositions and investigate the relative arrangement of transition metal atoms within and across layers. Thermodynamic stability increased with Mo content and its presence on the surface layer. We further investigated the catalytic performance of stable MXenes for the HER and observed that the center of the oxygen p-band (ε_p) correlated well with the energy of adsorption of a hydrogen atom ΔG(*H). Subsurface metal atoms significantly influenced the ΔG(*H) values at the surface via both ligand and strain effects. Our work expands the space of potentially stable MXene compositions, providing targets for synthesis and their evaluation in various applications.