Pattern Learning Electronic Density of States.

Area of Science:

• Condensed matter physics
• Materials science
• Computational materials science

Background:

• Electronic density of states (DOS) is crucial for understanding metal properties.
• First-principles density-functional theory (DFT) calculations are standard for DOS but computationally expensive.

Purpose of the Study:

• To develop a fast machine learning (ML) method for predicting DOS patterns.
• To enable prediction for both bulk and surface structures in multi-component alloys.

Main Methods:

• Utilized principal component analysis for pattern learning.
• Employed four key features: d-orbital occupation ratio, coordination number, mixing factor, and inverse miller indices.
• Developed an ML model independent of the number of electrons, unlike DFT's O(N^3) scaling.

Main Results:

• Achieved 91-98% pattern similarity with DFT calculations.
• Demonstrated a method significantly faster than traditional DFT.
• Showcased applicability to both bulk and surface alloy structures.

Conclusions:

• The developed ML method offers a fast and accurate alternative for electronic structure calculations.
• This approach overcomes the traditional accuracy-speed trade-off in electronic structure prediction.
• Enables efficient exploration of material properties for multi-component alloys.