An Agentic Framework for Autonomous Metamaterial Modeling and Inverse Design
Darui Lu, Jordan M. Malof, Willie J. Padilla
Abstract
The evolution from large language models to agentic systems has created a new Frontier of scientific discovery, enabling the automation of complex research tasks that have traditionally required human expertise. We developed and demonstrated such a framework specifically for the inverse design of photonic metamaterials. When queried with a desired optical spectrum, the Agent autonomously proposes and develops a forward deep learning model, accesses external tools via APIs for tasks like optimization, utilizes memory, and generates a final design via a deep inverse method. We demonstrate the framework’s effectiveness, highlighting its ability to reason, plan, and adapt its strategy autonomously and in real-time, mirroring the processes of a human researcher. Notably, the Agentic Framework possesses internal reflection and decision flexibility, allowing exploration of a large design space and the production of highly varied output. Our results suggest that autonomous agents have the potential to accelerate research in photonics and broader domains of scientific computing while reducing the expertise requirements.