An Agentic AI Framework for Automated DFT Workflows

Density functional theory (DFT) has become a cornerstone of modern materials research, yet its practical use remains challenging for newcomers due to the steep learning curve of input preparation, parameter convergence, and methodological validation. Setting up accurate and reproducible simulations often requires substantial experience, resulting in inefficient workflows and significant time loss when errors occur.

To address these challenges, we have developed an AI agentic framework integrated with the RESCU DFT solver that automates every stage of a first-principles workflow. The system consists of multiple large language model (LLM)-driven agents—each specialized in a particular task, including parameter selection, structure research, input generation, and automation scripting. Acting under a supervisory node, these agents transform high-level user queries into complete, validated RESCU input files and runnable workflows, supported by literature-aware reasoning and data-driven parameterization.

The framework has been successfully applied to complex semiconductor band-alignment studies in strained Si/Ge heterostructures, demonstrating its ability to autonomously produce accurate inputs, optimize parameters, and enable large-scale automated data generation. This approach substantially reduces the barrier to entry for DFT research, accelerates project development, and enhances reproducibility across computational materials science.