Interpretable Deep Learning Model for Analyzing the Relationship between the Electronic Structure and Chemisorption Property

Abstract

The use of machine learning (ML) is exploding in materials science as a result of its high predictive performance of material properties. Tremendous trainable parameters are required to build an outperforming predictive model, which makes it impossible to retrace how the model predicts well. However, it is necessary to develop a ML model that can extract human-understandable knowledge while maintaining performance for a universal application to materials science. In this study, we developed a deep learning model that can interpret the correlation between surface electronic density of states (DOSs) of materials and their chemisorption property using the attention mechanism that provides which part of DOS is important to predict adsorption energies. The developed model constructs the well-known d-band center theory without any prior knowledge. This work shows that human-interpretable knowledge can be extracted from complex ML models.

High-Throughput Screening of Sulfur Reduction Reaction Catalysts Utilizing Electronic Fingerprint Similarity

“…12 DOS can also be implemented to accurately predict catalytic properties, such as adsorption energies. 10,20,21 This screening process, which does not involve the optimization of intermediate structures, is particularly suitable for complex catalytic reactions with numerous transforming intermediates, such as the sulfur reduction reaction (SRR). 22 Li−S batteries, with their extremely high theoretical capacity density of 2567 Wh/kg, are one of the most promising nextgeneration energy storage devices.…”

“…Even if the materials have different structures, they can still be grouped into a cluster based on their DOS fingerprint similarity . DOS can also be implemented to accurately predict catalytic properties, such as adsorption energies. ,, …”

Inverse design for materials discovery from the multidimensional electronic density of states

“…In this regard, the DOS pattern itself or its derivative, although much more complex than the d-band center, has served as an improved and complementary descriptor in catalyst design. [27][28][29][30][31] Fung et al 29 developed an ML model designed to accurately predict adsorption energies from DOS patterns. Similarly, Hong et al 30 predicted adsorption energies from DOS patterns and interpreted the correlation between DOSs and chemisorption properties.…”

“…[27][28][29][30][31] Fung et al 29 developed an ML model designed to accurately predict adsorption energies from DOS patterns. Similarly, Hong et al 30 predicted adsorption energies from DOS patterns and interpreted the correlation between DOSs and chemisorption properties. Knøsgaard et al 31 successfully estimated quasiparticle band structures from DOS ngerprints using standard DFT calculations.…”

Formulating electronic descriptors to rationally design graphene-supported single-atom catalysts for oxygen electrocatalysis

“…Therefore, several approaches have been proposed to reduce the usage of precious PGMs while maintaining high catalytic performance. [5][6][7][8][9][10][11][12] Recently, graphenesupported single-atom catalysts (G-SACs) have been demonstrated to be one of the best of these strategies because G-SACs not only decrease the amount of loaded precious metal but also improve the mass activity by maximizing the atomic efficiency with high product selectivity. 2,3,13,14 However, G-SACs always face stability issues owing to the high chemical potential of single atoms, particularly agglomeration and dissolution.…”