Quantifying Chemical Structure and Machine-Learned Atomic Energies in Amorphous and Liquid Silicon.

Area of Science:

• Materials Science
• Computational Chemistry
• Condensed Matter Physics

Background:

• Understanding amorphous materials requires advanced simulation techniques.
• Current methods struggle to fully elucidate intricate atomic structures.

Purpose of the Study:

• To apply machine-learning techniques for quantitative chemical insight into amorphous silicon (a-Si) atomic structures.
• To associate coordination defects with stability regions in a-Si.
• To analyze the transition in local energies during the vitrification of liquid silicon.

Main Methods:

• Combining quantitative descriptions of nearest- and next-nearest-neighbor structures.
• Integrating a quantitative description of local stability.
• Analyzing ensembles of a-Si networks with tailored ordering via varying quench rates (down to 10¹⁰ K s^-1).

Main Results:

• Machine learning provides new quantitative chemical insights into a-Si.
• Coordination defects in a-Si are linked to distinct stability regions.
• A clear transition in local energies during liquid silicon vitrification was observed.

Conclusions:

• The machine-learning approach offers a straightforward and inexpensive method for analyzing amorphous and liquid states.
• This technique is expected to significantly advance the quantitative understanding of disordered matter.
• The findings have broad implications for materials science and condensed matter physics.