A dynamic routing-guided interpretable framework for salt-solvent chemistry.

Area of Science:

• Electrochemistry
• Materials Science
• Computational Chemistry

Background:

• Salt-solvent chemistry is crucial for electrochemical systems, influencing properties like ionic conductivity.
• Designing optimal electrolytes is challenging due to vast chemical spaces and limited data.
• Existing models struggle with sparse, imbalanced data and complex structure-behavior relationships.

Purpose of the Study:

• To develop a robust framework (SCAN) for modeling and interpreting salt-solvent chemistry.
• To accurately predict ionic conductivity in non-aqueous electrolytes.
• To provide chemical insights into factors governing conductivity.

Main Methods:

• Developed SCAN, a dynamic routing-guided framework for salt-solvent chemistry.
• Applied SCAN to non-aqueous electrolytes, handling long-tailed data distributions.
• Integrated gradient-decoupling, symbolic regression, and quantum chemistry calculations for interpretation.

Main Results:

• Achieved a benchmark conductivity prediction error of 0.372 mS cm⁻¹, a 65.3% reduction over baselines.
• Created a conductivity atlas for over 11.5 million salt-solvent systems.
• Validated SCAN with an 81.08% success rate for top-predicted candidates, identifying electrolytes with >20 mS cm⁻¹ conductivity.

Conclusions:

• SCAN effectively models complex salt-solvent interactions and predicts ionic conductivity with high accuracy.
• The framework facilitates the discovery of novel electrolytes for electrochemical applications.
• SCAN offers valuable chemical insights into molecular flexibility and ion-solvent interactions influencing conductivity.