Machine learning models for predicting the dielectric constants of oxides based on high-throughput first-principles calculations

Abstract

Prediction models of both the electronic and ionic contributions to the static dielectric constants have been constructed using data from density functional perturbation theory calculations of approximately 1200 metal oxides via supervised machine learning. We developed two types of random forest regression models for oxides with the ground-state crystal structures: one model requires only compositional information and the other model also uses structural information. Although the training data included various atomic frameworks, the prediction models performed well even when only compositional information was used as feature descriptors. In prediction of the electronic contributions to the dielectric constants, the accuracies of the regression models with and without structural information were comparable, while the structural descriptors more clearly improved the prediction accuracy for the ionic contributions. We also analyzed the feature importance for prediction of the dielectric constants. The mean atomic mass and mass density were determined to be significant features in prediction of the electronic contributions without and with structural information, respectively. The standard deviation of the principal quantum number and mean neighbor distance variation were found to be important for the respective prediction models of the ionic contributions. The correlations between the dielectric constants and these features are discussed, along with the underlying physical mechanisms.

Structure–Composition–Property Relationships in Antiperovskite Nitrides: Guiding a Rational Alloy Design

“…This is attributed to the linear relationship between band gap and tolerance factor in Figure 4a, because a larger band gap generally leads to a smaller ɛele. 34 On the other hand, we find that the ɛele is not sensitive to the X-site element. Based on the calculated ɛele, we estimate the exciton binding energies Eb of the six studied anti-perovskites X3NA (see Table 1).…”

Prediction of dielectric constants of ABO3-type perovskites using machine learning and first-principles calculations

“…These two observations are consistent with those of previous studies. 2,21,27,48 However, such behavior is not clearly observed in ABO 3 perovskites. Many perovskites satisfy the inverse relationship for the electronic contribution, but several structures exhibit large values for both properties.…”

“…20 With the development of artificial intelligence, material scientists have begun to implement machine-learning (ML) algorithms to search for materials with ideal properties among vast possible chemical spaces using high-throughput screening methods. [21][22][23][24] For example, Srivastava et al suggested an outline for the development of perovskite solar cells through ML methods. 25 Such a method has also been employed to find promising dielectric materials.…”

Interpretable and Explainable Machine Learning for Materials Science and Chemistry

“…Tree-based models are interpretable because the order and threshold of decisions executed by the tree to reach an answer can be observed, even when trees are used in ensembles (such as random forests or boosted trees) the degree to which a given parameter splits the data can be obtained and is linked to how important that parameter is for the final prediction; thus providing an interpretation of feature importance. This kind of approach has been been used for example to show which features affect the band gap of a material, or the dielectric response [14,15]. Support vector machine methods are also interpretable using similar feature importance inspection.…”

“…There are a range of "what-if" analysis approaches that work by examining how the value of the model output changes when one or more of the input values are modified. Partial dependence plots (PDPs) examine how changing a given feature affects the output, ignoring the effects of all other features [24], for example we could look at the effect of the mean atomic mass of a material on the dielectric response, marginalising all other factors using a model such as that presented in reference [14]. One drawback of marginalising all other features is that confounding relationships are missed and can mask effects, for example imagine increased mean atomic mass increased the dielectric response in a dense material, but decreased it in a porous material, these factors would cancel in PDP.…”