В МФТИ ускорили процесс поиска новых материалов с определенными оптическими характеристиками.
The findings have been published in the journal Science China Materials. Materials with a high birefringence index (Δn), particularly in the deep ultraviolet range, are actively utilized in the development of polarization and various other devices. Traditional experimental methods for discovering such materials are time-consuming and often complex. Consequently, researchers have created a straightforward and clear method for selecting materials with a high Δn, leveraging quantum-mechanical models and modern machine learning technologies.
When training a neural network, it is crucial to accurately select the features that describe the training dataset. The optical properties of a material depend on both its crystalline structure and chemical composition. Thus, the model architecture must identify the relationships between chemical composition, crystalline structure, and the target property of the material.
One such architecture that integrates the aforementioned properties is the graph neural network. In this study, the ALIGNN graph neural network model was used as the primary tool. Researchers trained it to predict the birefringence value of materials based on their crystalline structures. To assess the accuracy of predictions, physicists introduced the D-optimality criterion. This criterion indicates how reliable the model's predictions are on new data and evaluates how well the characteristics of new materials "fit" within the dataset used for training the model.
“The proposed approach, which includes gathering a training sample, developing machine learning models, and screening databases, has enabled us to identify new materials with specified optical properties. In the future, this approach could be expanded to other classes of materials,” says Ivan Kruglov, head of the Materials Computer Design Laboratory at MIPT.
“Previously, we had not encountered studies that employed the D-optimality criterion not only to assess the reliability of machine learning model predictions but also to evaluate the 'novelty' of the data. Additionally, the developed approach combines machine learning methods with traditional quantum-mechanical calculations, ensuring accuracy and good interpretability of the results. We continue to develop models using new approaches, and possibly their quality will improve even further in the near future!” explains Lyudmila Bereznikova, a researcher at the Materials Computer Design Laboratory at MIPT.
Utilizing the trained graph neural network, after screening the database Materials Project, scientists discovered two new materials with high birefringence values in the deep ultraviolet range: NaYCO3F2 and SClO2F. The identified materials have Δn = 0.202 and Δn = 0.101 at a wavelength of 1064 nm. Analysis revealed that the high refractive index in NaYCO3F2 is due to the arrangement of planar structural elements [CO3]. In the case of SClO2F, the anisotropic S–Cl bonds contribute significantly to the Δn value.
This work demonstrates for the first time the possibility of simultaneously using a graph neural network in conjunction with the D-optimality criterion for accurate Δn prediction and the search for new optical materials. The developed model is presumed to be adaptable for searching materials with other physical properties, such as hardness or thermal conductivity.
The study involved researchers from MIPT, the Xinjiang Technical Institute of Physics and Chemistry of the Chinese Academy of Sciences, and the XPANCEO New Technology Research Center (UAE).