Machine learning search for stable binary Sn alloys with Na, Ca, Cu, Pd, and Ag.

Area of Science:

• Materials Science
• Computational Chemistry
• Solid State Physics

Background:

• Metal-Sn (M-Sn) binaries are known for diverse properties relevant to energy storage, electronics, and superconductivity.
• Exploring the vast configuration space of M-Sn systems is computationally intensive.
• Previous studies have hinted at the potential of various M-Sn compounds.

Purpose of the Study:

• To conduct a large-scale screening for new synthesizable materials within five M-Sn binary systems (M = Na, Ca, Cu, Pd, Ag).
• To identify thermodynamically stable intermetallics under varying temperature and pressure conditions.
• To demonstrate the efficacy of machine learning potentials in accelerating materials discovery.

Main Methods:

• Development and application of the MAISE-NET framework for constructing neural network interatomic potentials.
• Utilizing MAISE-NET to accelerate *ab initio* global structure searches.
• Screening over two million candidate phases computationally.

Main Results:

• Discovery of 29 thermodynamically stable intermetallics across the studied M-Sn systems.
• Prediction of novel ambient-pressure materials, including hP6-NaSn2 and tI36-PdSn2.
• Identification of high-temperature, tin-rich ground states in Na-Sn, Cu-Sn, and Ag-Sn systems.
• Re-examination of known Cu-Sn phases, explaining their stabilization through entropy.

Conclusions:

• The study successfully identified numerous new stable M-Sn intermetallics, significantly expanding the known phase space.
• Machine learning potentials integrated with *ab initio* methods offer a powerful and efficient strategy for materials discovery.
• The findings provide a foundation for further experimental synthesis and application development of these novel materials.