Leveraging Large Language Models for Suitability Assessment of Electrolytes for Rechargeable Batteries

Chemical mixtures (e.g., drugs, cosmetics, electrolytes for rechargable batteries) play a significant role in chemical and material sciences. Electrolyte formulations are characterized by multiple factors, such as components (salt and solvents), their respective fractions, and temperature, which collectively influence target properties like ionic conductivity. The evaluation of an electrolyte's suitability or usability under specified conditions, like distinct temperature ranges and ionic conductivity thresholds, poses a tremendous challenge for individuals lacking expert knowledge in this domain. This underscores the need for specialized insight when assessing the viability of electrolytes in specific applications.

Large language models (LLMs) have emerged as a promising tool for domain-specific scientific discovery. These models have shown remarkable progress in exhibiting expert-level language understanding and generation across specialized areas, thereby enhancing productivity and enabling deeper insights. Despite these advancements, directly applying LLMs to complex tasks, such as reasoning about electrolyte properties and their suitability for rechargeable batteries, continues to be challenging.

To address this issue and assist non-expert comprehension, a unique dataset was curated consisting of question-answer (QA) pairs centered around compounds (salt and solvents), their respective mol fractions, the temperature, and the ionic conductivity of the mixture. The data samples were extracted from the paper by M. Zohari, V. Sharma et al. [1] and filtered based on suitability constraints as described in the publication. This curated dataset was then used to develop a fine-tuned LLM to classify and explicate electrolyte suitability.

The results of our fine-tuned model indicate improved performance compared to the base model, Granite3.2-2B-instruct, when assessing the suitability of never-seen electrolytes. Research efforts are ongoing to refine the LLM's reasoning capabilities and provide concise explanations on the individual components that contribute to an electrolyte's overall properties. This work presents a fine-tuned LLM designed to support the interpretation of complex electrolyte systems and assess their suitability for specific application scenarios.

[1] Chemical Foundation Model Guided Design of High Ionic Conductivity Electrolyte Formulations (M. Zohari, V. Sharma et al., 2025)