Optical materials discovery and design with federated databases and machine learning.

Area of Science:

• Materials Science
• Computational Chemistry
• Solid State Physics

Background:

• Vast numbers of hypothetical inorganic materials are generated using computational methods like density-functional theory.
• These materials are increasingly accessible through open databases and standardized APIs, such as the OPTIMADE API.
• The OPTIMADE federation provides access to over 30 million crystal structures, many novel and identified by machine learning.

Purpose of the Study:

• To develop an efficient workflow for screening large datasets of hypothetical inorganic materials.
• To identify next-generation optical materials, specifically those with high refractive indices.
• To leverage automated calculations, federated data curation, and machine learning for materials discovery.

Main Methods:

• Utilized the OPTIMADE API to access a federated database of over 30 million crystal structures.
• Applied MODNet, a neural network model, for property prediction within an active learning framework.
• Implemented a high-throughput computation strategy combined with active learning for targeted screening.

Main Results:

• Successfully isolated specific structures and chemistries with potential for high-refractive-index optical applications.
• Demonstrated an efficient method for non-exhaustively screening a dynamic and large materials space.
• Identified promising candidates for further theoretical calculations and experimental validation.

Conclusions:

• The presented workflow effectively utilizes automated calculations, federated datasets, and machine learning for accelerated materials discovery.
• The approach is adaptable and can be periodically re-assessed as new data and databases become available.
• This work facilitates the discovery of novel optical materials by combining computational power with data federation.