AutoDFT: Multi-Agent Framework Automates DFT Calculations

AutoDFT is a multi-agent framework that operates in a closed loop, integrating LLM reasoning throughout the entire density functional theory (DFT) lifecycle. In contrast to current LLM-based agents that focus solely on the initial planning phase, AutoDFT employs a strategic planner to create a foundational outline of objectives, a step planner for detailed action items, and ongoing adjustments based on interim outcomes. This methodology minimizes human involvement when convergence is delayed, unforeseen physics arise, or intermediate findings alter the problem. The goal of this framework is to enhance the robustness and general applicability of DFT calculations beyond predetermined situations.