O09_05

Leveraging LLMs for Quantum Chemistry: A Comparative Study of Input File Generation for Gaussian, DFTB+, and ORCA

Gergely JUHASZ *¹, Johannes Mario MEISSNER², Ilya KULYATIN²

¹School of Science, Chemistry, Tokyo Institue of Technology
²ResearchCopilot
( * E-mail: juhasz@chem.titech.ac.jp)

Computational tools are now essential in chemical research and development, especially for predicting the properties of new compounds, screening processes, and studying the relationship between structure and properties. However, using these tools effectively can be challenging because it requires a deep understanding of how to choose the right methods, set up models, and adjust parameters. This knowledge often comes from extensive study of the literature, established benchmarks, and experiences shared among colleagues.

In this presentation, we explore how large language models (LLMs) can assist in planning quantum chemistry simulations and generating input files. Specifically, we tested recent versions of OpenAI’s GPT, Claude, and open-source LLMs to generate input files for popular quantum chemistry software packages such as Gaussian, DFTB+, and ORCA. We compare these LLMs to see how well they help select computational methods and create accurate, grammatically correct input files. Our study examines the strengths and weaknesses of these models, helping to identify where they can be useful and where expert input is still necessary.

The potential impact of these tools goes beyond just making processes faster. LLMs could make quantum chemistry more accessible to researchers who don’t have a strong background in computational methods, allowing them to conduct advanced research. Additionally, LLMs could help make it easier to reproduce calculations from published studies, addressing a common problem in the field where replicating results can be difficult due to the complexity of setting up input files. By integrating LLMs into computational chemistry workflows, we could create a more inclusive research environment where more scientists can use advanced computational techniques and ensure their work is reproducible.