CHEAT is a set of Python modules for regression of adsorption energies and modeling catalysis on high-entropy alloys. This modeling procedure is described in detail here:

"Ab Initio to activity: Machine learning assisted optimization of high-entropy alloy catalytic activity."
DOI: https://doi.org/10.26434/chemrxiv-2022-vvrrf-v2

If this repository is utilized please cite:
Clausen, C. M., Nielsen, M. L. S., Pedersen, J. K., & Rossmeisl, J. (2022). Ab Initio to activity: Machine learning assisted optimization of high-entropy alloy catalytic activity.

It is the hope of the authors that this repository will be used, copied and modified by groups interested in doing computational studies on high-entropy alloys.

• ase
• gpaw
• scikit-learn
• torch
• torch-geometric
• torch-sparse
• torch-scatter
• iteround

The data acquisition module utilizes SLURM for computational workload management but this can be omitted.

All modules contain further explanation and instructions within each subdirectory. Data have been provides so that each module contains a working example.

The data module assists setting up DFT calculations. Optimized geometries are stored in ASE databases and can subsequently be joined into a single database to construct regression features.

The features modules will reduce optimized geometries to features suitable for regression of adsorption energies. Currently two types of feature schemes are available: a zone-reduced schemed based on equivalent atomic positions relative to the adsorption site and a graph-based feature scheme.

The regression modules trains the corrensponding regression model, Piecewise Linear regression (PWL) or Graph Convolutional Neural Network (GCN), depending on the chosen feature scheme and benchmarks adsorption energy prediction accuracy.

The surface module simulates a high-entropy alloy surface of a given size, predicts the available adsorption energies and simulates adsorbate coverage including competitive co-adsorption of *O and *OH. Based on established theory a catalytic activity can be estimated.

The search module apply the above step in a Bayesian optimization procedure to maximize the catalytic activity within the given composition space.

All DFT calculations required to reproduce the results of the paper is available here