Crystal Structure Prediction of Binary Alloys via Deep Potential

Area of Science:

• Materials Science
• Computational Physics
• Condensed Matter Physics

Background:

• Crystal structure prediction is a long-standing challenge in physics and materials science.
• Accurate energy calculations and efficient global search algorithms are crucial for predicting stable crystal structures.
• Machine learning potentials offer a promising approach, combining Density Functional Theory (DFT) accuracy with empirical force field speed.

Purpose of the Study:

• To predict stable crystal structures for intermetallic compounds in the aluminum-magnesium system using a machine learning model.
• To identify novel meta-stable phases with low formation energies.
• To develop a robust screening criterion for efficient crystal structure prediction and DFT refinement.

Main Methods:

• Utilized a pre-developed Deep Potential model for interatomic energy calculations.
• Employed global search algorithms to explore the configurational space of the aluminum-magnesium system.
• Applied a novel screening criterion to select candidate structures for subsequent DFT refinement.

Main Results:

• Identified six meta-stable phases in the aluminum-magnesium system with negative or near-zero formation energies.
• Discovered Mg12Al8 exhibiting excellent ductility and Mg5Al27 with a high Young's modulus.
• Demonstrated that the proposed screening criterion significantly reduces the computational cost for constructing accurate convex hulls.

Conclusions:

• Machine learning potentials, like Deep Potential, are effective tools for accelerating crystal structure prediction.
• The identified aluminum-magnesium phases warrant further experimental investigation.
• The developed screening criterion offers a computationally efficient pathway for discovering new materials with desired properties.