Large Language Models for Material Property Predictions: elastic constant tensor prediction and materials design

Abstract

Efficient and accurate prediction of material properties is critical for advancing materials design and applications. Leveraging the rapid progress of large language models (LLMs), we introduce ElaTBot, a domain-specific LLM for predicting elastic constant tensors and enabling materials discovery as a case study. The proposed ElaTBot LLM enables simultaneous prediction of elastic constant tensors, bulk modulus at finite temperatures, and the generation of new materials with targeted properties. Integrating general LLMs (GPT-4o) and Retrieval Augmented Generation (RAG) further enhances its predictive capabilities. A specialized variant, ElaTBot-DFT, designed for 0 K elastic constant tensor prediction, reduces the prediction errors by 33.1% compared with domain-specific, material science LLM (Darwin) trained on the same dataset. This natural language-based approach highlights the broader potential of LLMs for material property predictions and inverse design. Their multitask capabilities lay the foundation for multimodal materials design, enabling more integrated and versatile exploration of material systems.