Universal materials model of deep-learning density functional theory Hamiltonian

Area of Science:

• Computational Materials Science
• Artificial Intelligence in Materials Research
• Quantum Mechanics and Condensed Matter Physics

Background:

• Developing large-scale materials models is crucial for AI-driven materials research.
• Existing methods face challenges in accurately modeling complex structure-property relationships.

Purpose of the Study:

• To propose a feasible pathway for creating universal materials models.
• To enable accurate computational modeling of diverse materials using AI.

Main Methods:

• Developed a universal materials model using deep-learning density functional theory Hamiltonian (DeepH).
• Constructed a large materials database and improved the DeepH method.
• Demonstrated fine-tuning of universal models for specific applications.

Main Results:

• Achieved a universal DeepH model capable of handling diverse compositions and structures.
• Obtained remarkable accuracy in predicting material properties.
• Showcased successful fine-tuning for enhanced specific material models.

Conclusions:

• Demonstrated the feasibility of a universal materials model based on DeepH.
• Laid the groundwork for developing large-scale AI-driven materials models.
• Opened new opportunities for AI-accelerated materials discovery.