Leveraging Language Model, Crystal Structure Prediction and First-Principles Calculation for Material Design.

Area of Science:

• Condensed matter physics
• Materials science
• Computational chemistry

Background:

• Large language models (LLMs) show promise for materials discovery.
• Integrating LLMs with ab initio methods like crystal structure prediction (CSP) and density functional theory (DFT) is crucial for advanced materials design.
• Current methods need enhancement for efficient navigation of complex material spaces.

Purpose of the Study:

• To develop an integrated computational framework for inverse material design.
• To combine LLM-guided screening with GA and GNN-based CSP methods.
• To identify novel photovoltaic materials with tailored optoelectronic properties.

Main Methods:

• Developed an integrated framework combining LLMs, genetic algorithms (GA), and graph neural network (GNN)-based CSP.
• Utilized transformer-based vector similarity analysis and unsupervised clustering.
• Performed first-principles calculations (DFT) to validate material properties.

Main Results:

• Successfully identified a previously overlooked oxide material with significant photovoltaic potential.
• Demonstrated that the material possesses a direct band gap.
• Calculated high theoretical efficiencies suitable for photovoltaic applications.

Conclusions:

• The LLM + CSP + DFT framework efficiently navigates material space for inverse design.
• This hierarchical approach can discover nonintuitive functional materials.
• The identified oxide material shows promise for future photovoltaic applications.