A progressive learning method for predicting the band gap of ABO₃ perovskites using an instrumental variable

Abstract

A progressive learning method with an instrumental variable and bond-valence vector sums was used to improve the bandgap prediction precision.

Machine learning for perovskite materials design and discovery

“…In addition to using the classification models to screen promising candidates with appropriate E g , the ML regression models also have excellent predictive performance. Li et al 45 constructed a ML model with ABO 3 perovskite formation energy (E f ) as the target. Then, the E f predicted by the model was used as the instrumental variable to build a progressive learning model to predict the E g of the perovskite materials.…”

Use of metamodels for rapid discovery of narrow bandgap oxide photocatalysts

“…The largest performance differences are seen between linear and nonlinear models. Least absolute shrinkage and selection operator (LASSO), kernel ridge regression (KRR), artificial neural networks, support vector regression (SVR), random forest (RF), extra tree (EXT), and different types of gradient boosting regression (GBR) have been used to predict the bandgap of diverse compounds ( Li et al., 2019 , 2020a ; Lu et al., 2018 ). Although these models could recapitulate the properties of photocatalysts well, two problems remain.…”

“…Simple, basic atomic and physicochemical features were calculated for the dataset ( Lu et al., 2019 ). Each component of the photocatalysts (A, A’, B, B’) was described by 14 features obtained from the periodic table, materials handbooks, and material databases ( Table S1 ) ( Li et al., 2019 , 2020a ; Lu et al., 2019 ; Rajan et al., 2018 ). To refine the number of features to the most relevant subset the full descriptor set of 57 features (56 features related to A, A’, B and B’, while n x is the amount of O) was subjected to feature selection.…”

Recent progress in the data-driven discovery of novel photovoltaic materials

“…Atomic descriptors are publicly accessible in the Mendeleev package [100] , Villars database [56] , and RD-Kit [64] , while structural descriptors are extracted from quantum-optimized crystal structures. Li et al [101] employed the Python Materials Genomics (pymatgen) package [102] to obtain the atomic information and crystal structural parameters for 1593 ABO 3 perovskites sourced from the Materials Project [52] . The atomic descriptors included the atom number, atom mass, Pauling electronegativity, melting point, and electron numbers in valence orbitals, while the structural features included the octahedral distortion and bond lengths and angles.…”

“…For example, Wen et al employed embedded methods by combining RFE and tree-based models to select the features, which led to up to nine features remaining [141] . Li et al adopted the same method to filter the optimal instrumental features for the bandgap of ABO 3 perovskites, in which the model could reach a stable 𝑅 2 value of 0.94 in cross validation with the selected 24 features [101] .…”