Machine learning (ML)-assisted first-principles DFT study of mechanical properties, electron–hole recombination, and enhanced light absorption in transition metal trioxides for photocatalytic applications

Developing efficient transition metal trioxides (TMOs) is difficult due to the critical impact of electron–hole recombination and optical absorption on photocatalytic activity. In this study, we develop a machine learning (ML)-assisted DFT framework to accelerate the screening of TMOs by predicting key properties such as bandgap energy, electron–hole recombination rates, and light-absorption efficiency. A comprehensive dataset of over 50 TMO materials is generated, and descriptors including unit cell volume, density, formation energy, and structural parameters are analyzed. ML models such as supervised regression models are trained to predict electronic and photocatalytic properties. Principal component analysis (PCA) reveals that bandgap energy, unit cell volume, and density are the most influential factors controlling electron–hole recombination, surface area, and light absorption. A machine learning-driven random forest regression model accurately predicts the energy bandgap of transition metal trioxides (WO3, VO3, and CrO3), achieving high performance with training R2 values of 0.937–0.959 and strong generalization with testing R2 values of 0.906–0.970. The random forest regression (RFR) model exhibits satisfactory performance during testing (R2 = 0.936), (R2 = 0.964), and (R2 = 0.894) enabling the rapid identification of top-performing TMO candidates. DFT calculated electronic band structures reveal a clear evolution from metallic to semiconducting behavior, with ReO3 exhibiting a zero-band gap (Eg), while VO3 (0.99 eV), CrO3 (1.41 eV), and WO3 (1.55 eV) display suitable band gaps (Eg) for photoinduced charge excitation. Optical response analysis shows strong light absorption on the order of 105–106 cm−1 indicating efficient photon harvesting across the ultraviolet and visible regions. WO3 combines a favorable band gap with high optical conductivity (~ 23.64 Ω−1 cm−1), strong absorption (8.0 × 105 cm−1), and comparatively low reflectivity (~ 0.35), making it particularly effective for visible-light-driven photocatalysis. This work demonstrates the effectiveness of combining ML with first-principles calculations to efficiently screen and design TMOs, providing a powerful strategy for the rational development of high-performance photocatalysts. © The Author(s), under exclusive licence to Società Italiana di Fisica and Springer-Verlag GmbH Germany, part of Springer Nature 2026.