Machine Learning Prediction on Birefringence of Nonlinear Optical Crystals and Polymorphs with Different...

Area of Science:

• Materials Science
• Crystallography
• Computational Chemistry

Background:

• Nonlinear optical (NLO) crystals are crucial for laser technologies, with birefringence being a key property.
• Optimizing birefringence through material modification is essential for advanced laser applications.
• Machine learning (ML) has shown promise in predicting NLO material properties, but identifying polymorphs with distinct birefringence remains challenging.

Purpose of the Study:

• To develop a novel descriptor, hp-wACSFs, for predicting the birefringence of inorganic crystals.
• To build and evaluate ML classifiers for identifying birefringence-active NLO crystals and polymorphs with different birefringence.
• To perform virtual screening for discovering new polymorphs with tunable birefringence.

Main Methods:

• Development of hp-wACSFs, a new descriptor based on atom-centered symmetric functions.
• Construction and application of ML classifiers for two prediction tasks: birefringence activity and differential birefringence in polymorphs.
• Implementation of virtual screening using the developed ML models.

Main Results:

• The hp-wACSFs descriptor achieved performance comparable to previous methods for predicting general birefringence.
• The ML models successfully identified polymorphs with different birefringence activities, reaching an accuracy of 0.8 for this challenging task.
• Virtual screening identified potential polymorphs with distinct birefringence properties.

Conclusions:

• hp-wACSFs is an effective descriptor for predicting inorganic crystal birefringence.
• The developed ML approach advances the identification of polymorphs with varied birefringence, crucial for NLO applications.
• This work facilitates the discovery of novel NLO materials with tailored optical properties.