Synthesis-Aware Materials Redesign via Large Language Models.

Area of Science:

• Materials Science
• Computational Chemistry
• Crystallography

Background:

• Predicting the synthetic feasibility of inorganic crystal structures is crucial for materials discovery.
• Existing methods primarily focus on prediction rather than redesigning unsynthesizable materials.

Purpose of the Study:

• To develop a novel framework using large language models (LLMs) to transform synthetically infeasible inorganic crystal structures into synthetically feasible ones.
• To provide actionable solutions for redesigning materials, bridging the gap between computational design and experimental synthesis.

Main Methods:

• Integration of an invertible structural representation with large language models.
• Iterative fine-tuning strategy to modify unsynthesizable materials.
• Utilizing a "learn-and-regenerate" approach inherent in LLMs.

Main Results:

• Demonstrated LLMs can effectively modify various inorganic materials to enhance synthesizability.
• Successfully redesigned unsynthesizable structures into viable candidates.
• Indirect experimental validation: 34 of the top 100 redesigned structures were found in existing literature.

Conclusions:

• The proposed framework effectively addresses the critical gap between materials design and synthesis.
• LLMs offer a powerful tool for generating experimentally realizable inorganic compounds.
• This approach accelerates the discovery of novel, synthesizable materials.