Machine Learning Band Gaps of Doped-TiO₂ Photocatalysts from Structural and Morphological Parameters

Abstract

Titanium dioxide (TiO2) photocatalysts in the form of thin films are of great interest due to their tunable optical band gaps, E g’s, which are promising candidates for applications of visible-light photocatalytic activities. Previous studies have shown that processing conditions, dopant types and concentrations, and different combinations of the two have great impacts on structural, microscopic, and optical properties of TiO2 thin films. The lattice parameters and surface area are strongly correlated with E g values, which are conventionally simulated and studied through first-principle models, but these models require significant computational resources, particularly in complex situations involving codoping and various surface areas. In this study, we develop the Gaussian process regression model for predictions of anatase TiO2 photocatalysts’ energy band gaps based on the lattice parameters and surface area. We explore 60 doped-TiO2 anatase photocatalysts with E g’s between 2.280 and 3.250 eV. Our model demonstrates a high correlation coefficient of 99.99% between predicted E g’s and their experimental values and high prediction accuracy as reflected through the prediction root-mean-square error and mean absolute error being 0.0012 and 0.0010% of the average experimental E g, respectively. This modeling method is simple and straightforward and does not require a lot of parameters, which are advantages for applications and computations.

Machine Learning Lattice Constants for Cubic Perovskite Compounds

“…As one of the machine learning techniques, the GPR has been utilized in various materials systems to predict important physical parameters in diverse application fields of. [46][47][48][49][50][51][52][53][54][55][56][57][58][59][60] This model could serve as a guideline for cubic perovskite design, both oxides and halides, and could be used as part of machine learning to aid understandings of relationships between ionic radii and lattice constants.…”

Machine learning the lattice constant of cubic pyrochlore compounds

“…The introduction of the methodology follows previous studies 36‐59 . GPRs are nonparametric probabilistic models.…”

“…The model is highly stable and accurate that contributes to efficient and low‐cost lattice constant estimations and understandings of which based on ionic radii and electronegativities of both ternary and mixed pyrochlores. As a machine learning technique, 34,35 the GPR model has been used various areas of materials science and engineering to acquire important physicochemical parameters through rapid and robust predictions 36‐59 . This model could serve as a guideline for pyrochlore lattice design and might be adopted for lattice mismatch estimations in thin film configurations.…”

Machine learning lattice parameters of monoclinic double perovskites

“…As one of the computational intelligence techniques, the GPR model has been utilized in other materials systems to predict significant physical parameters in different fields of applications. [35][36][37][38][39][40][41][42][43][44][45][46][47][48][49][50][51][52] This model can serve as a guideline for monoclinic double perovskite design and can be used as part of machine learning to aid the understanding of relationships between ion sizes and lattice constants. be the mean function, and consider the GPR model y = b(x) T β + l(x), where l(x)$GP(0, k(x, x 0 )) and b(x) R p .…”

“…Detailed kernel and basis function specifications can be found in previous studies. [35][36][37][38][39][40][41][42][43][44][45][46][47][49][50][51][52] For model parameter estimations, cross validation and Bayesian optimizations are used. For the former, 10 randomized folds are utilized (see Tables 1-3), and for the latter, the expected improvement per second plus (EIPSP) algorithm is adopted.…”