Unveiling the principle descriptor for predicting the electron inelastic mean free path based on a machine learning...

Area of Science:

• Materials Science
• Computational Physics
• Surface Science

Background:

• The TPP-2M formula is widely used for estimating electron inelastic mean free paths (IMFPs) in solids.
• However, TPP-2M exhibits limitations in accuracy for certain materials due to its reliance on traditional least-squares analysis.

Purpose of the Study:

• To develop a novel machine learning framework for more accurate IMFP prediction.
• To overcome the limitations of existing empirical formulas like TPP-2M.
• To uncover new physical descriptors influencing electron scattering.

Main Methods:

• Implementation of a machine learning framework enabling selection from numerous combined terms (descriptors).
• Development of a new empirical formula based on machine learning insights.
• Analysis of newly identified descriptors to understand their physical significance.

Main Results:

• The proposed machine learning framework achieves higher average accuracy and stability in IMFP estimation.
• New descriptors were identified, providing deeper physical insights into electron inelastic scattering.
• A comprehensive understanding of electron IMFPs, encompassing single and collective electron behaviors, was obtained.

Conclusions:

• Machine learning provides a robust and efficient method for predicting electron IMFPs.
• The framework has significant potential for data-driven regression problems and discovering empirical formulas.
• This approach can reveal deeper connections between experimental data and fundamental material parameters.