Comment on "Advancing material property prediction: using physics-informed machine learning models for viscosity".

Area of Science:

• Physical Chemistry
• Materials Science
• Computational Chemistry

Background:

• Data-driven machine learning (ML) faces challenges in property prediction with limited data.
• Integrating physics-based models with ML can enhance predictive performance.
• Chew et al. utilized molecular dynamics (MD) descriptors in quantitative structure-property relationships (QSPR) for predicting liquid viscosity.

Discussion:

• This comment explores the physical basis for heat of vaporization being a key descriptor for viscosity.
• It references Eyring's rate theory to explain the relationship between molecular interactions and fluid flow.
• An alternative thermodynamic-based modeling approach is proposed, requiring less input data.

Key Insights:

• Heat of vaporization is identified as a crucial descriptor for predicting temperature-dependent viscosity.
• Physics-informed ML, incorporating MD simulations, offers improved accuracy over purely data-driven methods.
• Eyring's rate theory provides a theoretical framework for understanding viscosity-descriptor relationships.

Outlook:

• Developing thermodynamic-based models can reduce data requirements for property prediction.
• Further integration of physical principles into ML holds promise for advancing materials modeling.
• This work encourages exploring alternative architectures for more efficient and accurate predictive models.