Machine learning force fields for inorganic crystalline materials: principles, advances, and emerging applications

Area of Science:

• Computational Materials Science
• Artificial Intelligence in Chemistry

Background:

• Classical force fields offer efficiency but lack accuracy for complex systems.
• First-principles methods provide high accuracy but are computationally expensive.
• Machine learning force fields (MLFFs) bridge this gap for inorganic crystalline materials.

Purpose of the Study:

• To systematically review the research progress and fundamental principles of MLFFs.
• To introduce the technical characteristics of representative MLFF models and benchmarking platforms.
• To examine MLFF advantages and challenges in inorganic crystalline materials research.

Main Methods:

• Systematic literature review of MLFF research.
• Categorization of MLFF models and their technical characteristics.
• Analysis of MLFF applications and limitations.

Main Results:

• MLFFs demonstrate significant advantages in structural prediction, physical properties, defect analysis, and phase transitions.
• Key challenges include computational efficiency, simulation scale, accuracy, generalization, data needs, interpretability, and physical constraints.
• Representative MLFF models and benchmarking platforms are introduced.

Conclusions:

• MLFFs offer powerful capabilities for atomic-level studies in inorganic crystalline materials.
• Addressing current challenges is crucial for advancing MLFF research and application.
• This review provides a reference for future MLFF development and utilization.