Physics-informed machine learning for inorganic scintillator discovery.

Area of Science:

• Materials Science
• Computational Chemistry
• Solid-State Physics

Background:

• Inorganic scintillators activated with lanthanide dopants are crucial for various applications.
• A key requirement for scintillation is the precise positioning of activator energy levels (4f and 5d1) within the host bandgap.

Purpose of the Study:

• To develop a machine learning (ML) based strategy for high-throughput screening of novel inorganic scintillators.
• To rapidly and reliably estimate the relative positions of activator energy levels within diverse host materials.

Main Methods:

• Utilized physics-based chemical trends and experimental data to train an ML model.
• Integrated first-principles calculations of band edges with the ML model.
• Applied the strategy to perovskite oxides and elpasolite halides.

Main Results:

• The ML model accurately captures chemical trends in host-dependent energy levels.
• Successfully screened promising scintillator candidates in a high-throughput manner.
• Demonstrated the model's effectiveness using perovskite oxides and elpasolite halides.

Conclusions:

• The developed ML approach provides a practical tool for initial screening of potential inorganic scintillators.
• This method systematically down-selects candidate materials for further investigation.
• Accelerates the discovery of novel materials for scintillator applications.