Autonomous intelligent agents for accelerated materials discovery.

Area of Science:

• Computational materials science
• Artificial intelligence in chemistry
• Autonomous experimentation

Background:

• Discovering new materials is crucial for technological advancement but faces challenges in vast, high-dimensional search spaces.
• Traditional methods are often slow and costly, limiting the exploration of potential material candidates.

Purpose of the Study:

• To develop and demonstrate an end-to-end computational system for autonomous materials discovery.
• To optimize cost-effectiveness in materials research by employing a sequential, agent-based experimental design approach.

Main Methods:

• The system utilizes an agent-based framework that leverages past knowledge, surrogate models, physical principles, and heuristic rules.
• It incorporates exploration-exploitation strategies for efficient decision-making on subsequent experiments.
• Autonomous agents controlled external computational experiments, including cloud-based density functional theory (DFT) simulations.

Main Results:

• Simulations were performed to design and test agents for materials discovery campaigns focused on relative stability.
• The autonomous platform was deployed in 16 real-world discovery campaigns across various binary and ternary chemistries (metal oxides, phosphides, sulfides, alloys).
• A total of 383 new stable or nearly stable materials were discovered autonomously.

Conclusions:

• The presented autonomous system significantly accelerates the discovery of novel materials.
• The agent-based, sequential approach offers a cost-effective and efficient strategy for navigating complex materials search spaces.
• This work demonstrates the successful application of autonomous platforms in real-world materials discovery, reducing the need for direct researcher intervention.