Â鶹ÒùÔº

Machine learning framework accelerates theoretical design of nonlinear optical materials

A research team from the Xinjiang Technical Institute of Â鶹ÒùÔºics and Chemistry of the Chinese Academy of Sciences has made strides in the theoretical design of nonlinear optical (NLO) materials by leveraging machine learning techniques. The team introduced a new strategy to explore uncharted chemical spaces, enabling the quantitative prediction of second harmonic generation (SHG) coefficients for complex NLO systems spanning infrared to deep ultraviolet wavelengths.

The development of high-performance NLO materials has been hindered by the vastness of chemical space and the absence of efficient theoretical prediction tools. To address this challenge, the researchers integrated machine learning with crystal structure generation methods, creating a predictive model that utilizes compositional and structural descriptors to guide the synthesis of new NLO materials.

Central to their approach is a machine learning model trained to predict the maximum SHG coefficients of NLO materials. By analyzing the relationships between chemical composition, structural features, and material properties, the model offers a systematic pathway for understanding structure-property correlations.

To further streamline the discovery process, the researchers implemented a rapid crystal structure generation technique, establishing an efficient workflow for exploring unknown chemical spaces. This framework allows researchers to input a crystal structure file and quickly obtain the predicted SHG coefficient, significantly reducing the time required for material screening.

Moreover, the researchers applied this workflow to infrared NLO materials, identifying seven promising compounds with strong SHG responses. One of these compounds, CsIn₅Se₈, was experimentally synthesized and characterized, exhibiting an SHG coefficient (d24) exceeding that of the benchmark material AgGaS₂. This experimental validation underscores the effectiveness of the proposed theoretical design approach.

The study, in Small, represents an advancement in the quantitative prediction of SHG coefficients using machine learning.