A Quantitative Electrostatic Potential Descriptor Enables Deep Learning-Accelerated Discovery of High-Performance...

Area of Science:

• Materials Science
• Electrochemistry
• Computational Chemistry

Background:

• High-energy lithium-ion batteries (LIBs) require advanced electrolytes for improved performance.
• Current electrolyte design lacks quantitative descriptors for solvation power, hindering rapid screening.
• Deep learning (DL) offers potential for accelerated screening but requires suitable molecular descriptors.

Purpose of the Study:

• To introduce a quantitative molecular descriptor for solvation power in LIB electrolytes.
• To enable deep learning-accelerated screening of novel electrolyte candidates.
• To identify promising electrolyte components for high-voltage LIBs.

Main Methods:

• Introduced the electrostatic potential ratio (ESP_ratio) as a descriptor for electron-donating/accepting balance.
• Utilized unsupervised clustering to define a solvation modulation zone (0.9 < ESP_ratio < 2.4).
• Combined ESP_ratio with self-supervised DL models for hierarchical screening of ~10^6 molecules.

Main Results:

• Identified a key solvation modulation zone using the ESP_ratio descriptor.
• Successfully screened a large chemical space, prioritizing candidates from unexplored areas.
• Experimental validation confirmed the efficacy of ESP_ratio-guided screening for high-voltage LIBs.

Conclusions:

• The ESP_ratio is an effective descriptor for rational electrolyte design in LIBs.
• The developed workflow significantly accelerates the discovery of novel electrolyte materials.
• This approach enables the identification of advanced electrolyte components for next-generation batteries.