Increasing Efficiency in Electrocatalysis Simulations Using the MapSy Library
Faculty Mentor Information
Dr. Oliviero Andreussi, Boise State University; and Dr. Jakob Filser, Boise State University
Presentation Date
7-16-2026
Abstract
Computational simulations are playing an increasingly important role in the discovery and design of catalysts by enabling the rapid screening of candidate materials before experimental synthesis. This work utilizes MapSy, a newly developed Python library that automates the characterization of catalytic surfaces using symmetry-function descriptors, reducing the computational cost of predicting adsorption properties while making large-scale screening of disordered and compositionally flexible materials feasible. We systematically investigated the effect of key feature-generation parameters on the accuracy and efficiency of machine learning predictions using hydrogen adsorption on bulk metallic glass cobalt phosphide (CoP), a promising catalyst for hydrogen evolution reactions (HER), as a benchmark system. Models were trained on randomly sampled adsorption sites and evaluated by comparing the full-surface prediction against corresponding reference calculations. Ultimately, it was found that omitting structural or higher-order symmetry functions had little impact on predictive accuracy. Likewise, increasing feature complexity through larger cutoff radii or additional principal components provided minimal improvement despite a major increase in the number of features required. These results demonstrate that computationally inexpensive feature sets can achieve accurate predictions, providing guidelines for selecting efficient default settings in MapSy. Future work will encompass creating a robust set of optimal feature generation settings, extending it to a broader range of catalytic materials.
Increasing Efficiency in Electrocatalysis Simulations Using the MapSy Library
Computational simulations are playing an increasingly important role in the discovery and design of catalysts by enabling the rapid screening of candidate materials before experimental synthesis. This work utilizes MapSy, a newly developed Python library that automates the characterization of catalytic surfaces using symmetry-function descriptors, reducing the computational cost of predicting adsorption properties while making large-scale screening of disordered and compositionally flexible materials feasible. We systematically investigated the effect of key feature-generation parameters on the accuracy and efficiency of machine learning predictions using hydrogen adsorption on bulk metallic glass cobalt phosphide (CoP), a promising catalyst for hydrogen evolution reactions (HER), as a benchmark system. Models were trained on randomly sampled adsorption sites and evaluated by comparing the full-surface prediction against corresponding reference calculations. Ultimately, it was found that omitting structural or higher-order symmetry functions had little impact on predictive accuracy. Likewise, increasing feature complexity through larger cutoff radii or additional principal components provided minimal improvement despite a major increase in the number of features required. These results demonstrate that computationally inexpensive feature sets can achieve accurate predictions, providing guidelines for selecting efficient default settings in MapSy. Future work will encompass creating a robust set of optimal feature generation settings, extending it to a broader range of catalytic materials.