Accurate Surface and Finite-Temperature Bulk Properties of Lithium Metal at Large Scales Using Machine Learning...

Area of Science:

• Materials Science
• Computational Chemistry
• Electrochemistry

Background:

• Lithium metal properties are critical for designing advanced lithium-ion and lithium-metal batteries.
• Experimental characterization of lithium metal is challenging due to its reactivity, low melting point, and microscopic scale.
• Existing computational methods like empirical potentials lack consistent accuracy, while ab initio calculations are computationally expensive.

Purpose of the Study:

• To develop a highly accurate machine learning interaction potential for lithium metal.
• To enable reliable large-scale, long-time simulations of lithium metal properties.
• To investigate properties and phenomena relevant to battery performance and stability.

Main Methods:

• Training a machine learning interaction potential using density functional theory (DFT) data.
• Validating the potential against experimental and ab initio results across diverse simulations.
• Employing the trained potential for large-scale and long-time simulations.

Main Results:

• The machine learning potential achieves state-of-the-art accuracy in reproducing known lithium metal properties.
• Accurate prediction of thermodynamic properties, phonon spectra, and temperature-dependent elastic constants.
• Successful prediction of surface properties and identification of a Bell-Evans-Polanyi relation for high Miller index facets.

Conclusions:

• The developed machine learning potential offers a computationally efficient and accurate tool for studying lithium metal.
• This potential facilitates the exploration of lithium metal behavior at scales and conditions inaccessible to DFT.
• The findings provide insights into lithium metal's mechanical behavior and surface dynamics, relevant for battery applications like dendrite suppression.