Speaker
Description
Widely used machine-learning (ML) approaches in materials science and catalysis are designed to accurately describe, in average, a wide range of materials. Nonetheless, only a handful of compounds might show the desired properties to be suitable for a given application. Thus, global ML models may overlook these statistically exceptional materials of interest. Here, we discuss how the subgroup-discovery (SGD) approach can be utilized to identify descriptions focused on exceptional materials [1]. We present a systematic analysis of the Pareto front of optimal SGD solutions with respect to generality and exceptionality, two conflicting objectives in SGD [2]. The concepts are illustrated by the identification of “rules” describing single-atom alloys (SAAs) capable of providing a strong activation of adsorbed CO$_2$, the first step towards the catalytic conversion of the molecule towards chemicals and fuels [3].
[1] B. R. Goldsmith, M. Boley, J. Vreeken, M. Scheffler, and L. M. Ghiringhelli, Uncovering structure-property relationships of materials by subgroup discovery, New. J. Phys. 19, 013031, 2017; https://doi.org/10.1088/1367-2630/aa57c2
[2] L. Foppa and M. Scheffler, Coherent Collections of Rules Describing Exceptional Materials Identified with a Multi-Objective Optimization of Subgroups, arXiv:2403.18437, 2024; https://doi.org/10.48550/arXiv.2403.18437
[3] H. I. Rivera-Arrieta and L. Foppa, Rules Describing CO$_2$ Activation on Single-Atom Alloys from DFT-meta-GGA Calculations and Artificial Intelligence, ChemRxiv, 2024; https://doi.org/10.26434/chemrxiv-2024-1dr10
