Automated COHP Calculations for the Prediction of Stability and Catalytic Activity in Mixed Transition Metal Oxides

The materials discovery of effective catalysts for the oxygen reduction and oxygen evolution reactions (ORR/OER) is crucial for the advancement of hydrogen fuel cells and electrolyzers [1]. For this, it is essential to identify affordable and Earth-abundant alternatives to precious metal catalysts that provide both stability and high catalytic activity. Recent studies have shown that the electronic structure of binary bulk transition metal-oxides (TMOs), represented by the crystal orbital Hamiltonian populations (COHP) of the metal-oxygen bond, serves as an accurate descriptor for O and OH adsorption [2]. This further demonstrates that leveraging these bulk descriptors enables efficient catalyst screening without the need for expensive surface simulations [3].<br/>In this talk, we will discuss an extended model of bulk ternary mixed transition metal oxides (A_xB_yO_z) and the importance of bond strength between metal d states and oxygen 2p states for predicting their stability and catalytic activity. We will also cover high-throughput COHP [4] calculations using Density Functional Theory (DFT) codes like VASP through the lobster package to analyze these interactions. To handle these high throughput simulations, we developed a python library called AutoCatLab. This library is implemented on finite state machine workflow which consists of several configurable steps including Job scheduling, monitoring and analysing results. In the current configuration, AutoCatLab is configured to compute COHP, handles basic and modified DFT calculations, and analyse the calculation results. I will describe working model of AutoCatLab which automatically deals with subtle aspects of these calculations, including VASP input generation, Job scheduling, output data management, and calculation convergence. This library also further analyse the results through an ASE parser, storing into ASE based SQL database [5]. Additionally, this also supports both CPU and GPU computation with the rescheduling mechanism for failed calculations.<br/>Lastly, our prediction models, such as Gaussian process regression (GPR) and Crystal graph convolutional neural network (CGCNN), will also be discussed in predicting the bulk stability of mixed ternary oxides.<br/><br/>[1]. Seh, Z. W.; Kibsgaard, J.; Dickens, C. F.; Chorkendorff, I.; Nørskov, J. K.; Jaramillo, T. F. Combining theory and experiment in electrocatalysis: Insights into materials design. <i>Science. 355, No. eaad 4998</i> (2017).<br/>[2]. Comer, B. M., Li, J., Abild-Pedersen, F., Bajdich, M., Winther, K. T. (2022), Unraveling Electronic Trends in O* and OH* Surface Adsorption in the MO2 Transition-Metal Oxide Series., <i>JPCC. 126 (18)</i> (2022).<br/>[3]. Comer, B. M., Bothra, .N, Lunger, J., Abild-Pedersen, F., Bajdich, M., Winther, K. T., Generalized Prediction of O and OH Adsorption on Transition Metal Oxide Surfaces from Bulk Descriptors <i>ACS Catal. 14, 5286−5296</i> (2024).<br/>[4] Dronskowski, R. and Blöchl, P.E., Crystal orbital Hamilton populations (COHP): energy resolved visualization of chemical bonding in solids based on density-functional calculations. <i>J. Phys. Chem. 97(33), pp.8617-8624</i> (1993).<br/>[5] The Atomic Simulation Environment—A Python library for working with atoms, <i>J. Phys.: Condens. Matter Vol. 29 273002,</i> (2017).