Novel inorganic crystal structures predicted using autonomous simulation agents.

Area of Science:

• Materials Science
• Computational Chemistry
• Solid-State Physics

Background:

• Discovering novel crystal structures is crucial for advancing materials science.
• Computational methods, particularly density functional theory (DFT), are increasingly used for materials discovery.
• Autonomous workflows can accelerate the exploration of chemical space for new materials.

Purpose of the Study:

• To report a large dataset of computationally discovered crystal structures.
• To provide DFT-computed formation energies and phase stability data.
• To facilitate benchmarking and development of materials discovery workflows.

Main Methods:

• Utilized the Computational Autonomy for Materials Discovery (CAMD) workflow.
• Employed an autonomous, density functional theory (DFT)-based, active-learning approach.
• Generated and optimized 96,640 crystal structures.

Main Results:

• The dataset comprises 96,640 DFT-computed crystal structures.
• 894 structures are within 1 meV/atom and 26,826 structures are within 200 meV/atom of the convex hull, indicating high stability.
• Includes DFT-optimized pymatgen crystal structure objects, formation energies, and phase stability data.

Conclusions:

• The CAMD dataset offers a valuable resource for materials science research.
• It can be used to benchmark active-learning and generative models for structure prediction.
• The data can seed experimental discovery and aid in developing structure-property relationship models.