Electronic Configurations of 3d Transition-Metal Compounds Using Local Structure and Neural Networks

Abstract

Machine learning (ML) methods extract statistical relationships between inputs and results. When the inputs are solid-state crystal structures, structure−property relationships can be obtained. In this work, we investigate whether a simple neural network is able to learn the 3d orbital occupations for the transition-metal (TM) centers in crystalline inorganic solid-state compounds using only the local structure around the transition-metal centers described by rotationally invariant fingerprints based on spherical harmonics and one-hot elemental encoding. A multilayer neural network trained on density functional theory (DFT) results of about 1800 samples was developed and showed good performance in predicting the TM orbital occupations (for both spin channels). We study in detail how the local structure affects the predictions of the local properties and how they provide physical insights for the design of a future machine learning model for materials chemistry. The proposed ML method is illustrated in practical application by predicting local magnetic moments of the transition-metal atoms in a full set of inorganic structures with large unit cells. Although less accurate compared to the experimental data, the ML results compared well with the DFT results, suggesting the feasibility of electronic property prediction based only on structure input.

Analysis of the oxygen evolution activity of layered double hydroxides (LDHs) using machine learning guidance

“…To establish an ML vector, we chose 0, 1 (binary representation/binary notation) rather than atomic numbers as inputs to represent elements. In fact, this encoding method has been widely applied in feature construction for ML models in chemistry, 29,30 and can offer a high accuracy. 31 Besides, this method can offer more rationality for discrete features which are not comparable since using only 0 and 1 can minimize bias for different features, which is its most significant advantage.…”

Machine Learning Prediction of the Experimental Transition Temperature of Fe(II) Spin-Crossover Complexes

“…Other approaches to predict spin state ordering have used interpretable linear models 106 or neural networks trained using only the local structure around the TM centers. 107 While ML has been previously leveraged to predict SCO behavior through predictions of spin splitting energies or structural differences between spin states, in this study we use ML models to make predictions of transition temperatures of SCO complexes from the previously curated SCO-95 data set. 40 We train interpretable ML models, that is, random forests (RFs) on a full feature set as well as RF or kernel ridge regression (KRR) models on a feature set selected by RF-ranked recursive feature addition (RF-RFA) on a set of 76 SCO complexes for which experimental T 1/2 values are available.…”

“…Besides predicting SCO behavior through predictions of spin splitting energies, a study from our group demonstrated the utility of ANNs trained on hybrid DFT structures by data-mining the literature for SCO complex structures and correctly assigning almost all spin states based on bond lengths in a set of 46 Fe­(II) SCO complexes. Other approaches to predict spin state ordering have used interpretable linear models or neural networks trained using only the local structure around the TM centers …”