Structural search for stable Mg-Ca alloys accelerated with a neural network interatomic model.

Area of Science:

• Computational Materials Science
• Alloy Discovery
• Solid-State Chemistry

Background:

• The Mg-Ca binary system is incompletely understood, with only one known stable phase (C14-Mg2Ca) under standard conditions.
• Traditional methods for alloy discovery are limited in exploring the vast potential energy surface.
• Predicting new stable phases requires advanced computational techniques capable of handling complex structures and conditions.

Purpose of the Study:

• To systematically screen the Mg-Ca binary system for novel (meta)stable alloy phases.
• To explore the potential energy surface efficiently beyond traditional ab initio methods.
• To identify candidate Mg-Ca materials synthesizable under non-standard conditions (elevated temperature/pressure).

Main Methods:

• Integration of neural network formalism with metaheuristic structural global search algorithms.
• Efficient exploration of the potential energy surface for Mg-Ca alloys.
• Complementary use of special quasirandom structures (SQS) for various stoichiometries.

Main Results:

• Discovery of several new candidate Mg-Ca phases beyond the known C14-Mg2Ca.
• Identification of phases (C15, C36 Laves, oS36) stabilized by vibrational entropy, particularly those with magnesium kagome layers.
• Prediction of two new thermodynamically stable phases at 1:1 and 3:1 Mg:Ca stoichiometries under high pressures (up to 10 GPa).

Conclusions:

• The combination of neural networks and metaheuristic searches effectively expands the known Mg-Ca phase diagram.
• Predicted phases with magnesium kagome layers are potentially synthesizable at high temperatures due to favorable vibrational entropy.
• New Mg-Ca phases are predicted to be stable under high pressures, suggesting feasibility in multi-anvil experiments.