Direction-Based Graph Representation to Accelerate Stable Catalyst Discovery

Abstract

To realize a renewable and sustainable energy cycle, there has been a lot of effort put into discovering catalysts with desired properties from a large chemical space. To achieve this goal, several screening strategies have been proposed, most of which require validation of thermodynamic stability and synthesizability of candidate materials via computationally intensive quantum chemistry or solid-state physics calculations. This problem can be overcome by reducing the number of calculations through machine learning methods, which predict target properties using unrelaxed crystal structures as inputs. However, numerical input representations of most of the previous models are based on either too specific (e.g., atomic coordinates) or too ambiguous (e.g., stoichiometry) information, practically inapplicable to energy prediction of unrelaxed initial structures. In this work, we develop a direction-based crystal graph convolutional neural network (D-CGCNN) with the highest accuracy toward formation energy predictions of the relaxed structures using the initial structures as inputs. By comparing with other approaches, we revealed correlations between crystal graph similarities and model performances, elucidating the origin of the improved accuracy of our model. We applied this model to the ongoing high-throughput virtual screening project, where the model discovered 1,725 stable materials from 15,318 unrelaxed structures by performing 3,966 structure optimizations (∼25%).

A chemically inspired convolutional neural network using electronic structure representation

“…, deletion, substitution, and insertion) required for two graphs to be identical – was defined to measure the quantitative similarity between two graph structures. 13 Thus, the higher structural similarity between IS and RS corresponds to less structural difference between them (ESI Fig. A8†).…”

“…However, despite its progress, there is still room for improvement in the accuracy and efficiency of the model. 13 To accelerate materials discovery, it is thus necessary to develop accurate ML models for direct IS2RE with appropriate representations, which are neither too structure-sensitive nor too insensitive. 13…”

“…13 To accelerate materials discovery, it is thus necessary to develop accurate ML models for direct IS2RE with appropriate representations, which are neither too structure-sensitive nor too insensitive. 13…”

“…Further, indirect IS2RE requires nearly 200 iterations of energy and force predictions for numerical geometry optimizations, 6 and the predicted formation energy signicantly depends on "pseudo"-optimized structures, making IS2RE inefficient. 13,14 Direct IS2RE is also available with properly developed input representations such as direction, which is moderately sensitive to structures. 14 Meanwhile, a model which uses stoichiometry-based inputs, such as Roost (Representation Learning from Stoichiometry), achieved reasonable prediction accuracy, [15][16][17] but it was not able to distinguish polymorphs of materials.…”

“…However, despite its progress, there is still room for improvement in the accuracy and efficiency of the model. 13 To accelerate materials discovery, it is thus necessary to develop accurate ML models for direct IS2RE with appropriate representations, which are neither too structure-sensitive nor too insensitive. 13 Density of states (DOS) has been utilized in several ML models to predict adsorption energy due to its excellent descriptions of electronic interactions based on d-band theory.…”