Abstract
Discovering novel infrared functional materials (IRFMs) hold tremendous significance for laser industry. Incorporating artificial intelligence into material discovery has been recognized as a pivotal trend driving advancements in materials science. In this work, an IRFM predictor based on machine learning (ML) is developed for the pre-selection of the most promising candidates, in which interpretable analyses reveal the prior domain knowledge of IRFMs. Under the guidance of this IRFM predictor, a series of selenoborates, ABa(3)(BSe(3))(2)X (A = Rb, Cs; X = Cl, Br, I) are successfully predicted and synthesized. Comprehensive characterizations together with first-principles analyses reveal that these materials exhibit preferred properties of wide bandgaps (2.92 - 3.04 eV), moderate birefringence (0.145 - 0.170 at 1064 nm), high laser-induced damage thresholds (LIDTs) (4 - 6 Ý AGS) and large second harmonic generation (SHG) responses (0.9 - 1 × AGS). Structure-property relationship analyses indicate that the [BSe(3)] unit can be regarded as a potential gene for exploring novel IRFMs. This work may open an avenue for exploring high-performance materials.