MatPC: Prompting Large Language Model, Crystal Structure Prediction, and First-Principles for Semantic-Driven...

Area of Science:

• Materials Science
• Computational Chemistry
• Artificial Intelligence

Background:

• Traditional material design is often slow and relies on intuition.
• Discovering novel materials with specific properties, like photovoltaics, requires efficient computational methods.

Purpose of the Study:

• To develop an AI-guided framework for semantic-driven material design.
• To accelerate the identification of novel photovoltaic materials using large language models (LLMs).

Main Methods:

• Integration of LLMs with first-principles methods and crystal structure prediction (MatPC).
• Utilizing prompt-engineered LLMs for semantic embeddings to identify material candidates.
• A computational workflow combining LLMs, similarity scoring, dimensional reduction, formula screening, crystal structure prediction (hybrid GA-GNN), and DFT validation.

Main Results:

• An unconventional Bi2WO6 polymorph was identified as a promising photovoltaic material.
• Detailed analysis of the electronic and optical properties of the identified material via first-principles calculations.
• Demonstration of an efficient material discovery pipeline leveraging LLMs.

Conclusions:

• The developed AI framework significantly accelerates the material design process.
• LLMs are effective tools for semantic-driven material discovery.
• The identified Bi2WO6 polymorph shows potential for photovoltaic applications.