Machine Learning-Based Investigation of Atomic Packing Effects: Chemical Pressures at the Extremes of Intermetallic...

Area of Science:

• Materials Science
• Computational Chemistry
• Solid State Physics

Background:

• Intermetallic phases exhibit complex atomic arrangements driven by unknown forces, hindering new material design.
• Density Functional Theory (DFT)-chemical pressure (CP) analysis visualizes atomic packing tensions but is computationally intensive.

Purpose of the Study:

• To develop a machine learning (ML)-based implementation of the chemical pressure (CP) approach.
• To overcome the computational limitations of traditional DFT-CP methods for analyzing intermetallic complexity.

Main Methods:

• Developed an ML-CP model trained on DFT-CP data from the Intermetallic Reactivity Database.
• Validated the ML-CP approach by comparing it with DFT-CP for various intermetallic systems.
• Applied ML-CP to analyze the complex structure of Mg2Al3.

Main Results:

• The ML-CP model accurately reproduces DFT-CP results, significantly reducing computational cost.
• Analysis of Mg2Al3 revealed that its complex structure originates from simple assembly rules for Frank-Kasper polyhedra.
• The ML-CP model is readily deployable via web interface or command-line tool.

Conclusions:

• The ML-CP approach provides an efficient and accessible tool for exploring intermetallic complexity.
• This method facilitates the understanding of driving forces in intermetallic formation, aiding in the design of new materials.
• The ML-CP model opens new avenues for investigating a wide range of intermetallic systems.