Bridging Hydrocarbon Thermodynamics and Electron-Density Isosurfaces with Explainable Machine Learning.

Area of Science:

• Physical Chemistry
• Computational Chemistry
• Chemical Engineering

Background:

• Accurate thermodynamic property prediction is crucial for hydrocarbon process design and modeling.
• Experimental data is often scarce or expensive, and existing prediction models require updates.
• Quantitative Structure-Property Relationship (QSPR) models offer a computational alternative for property estimation.

Purpose of the Study:

• To develop compact, physically interpretable QSPR models for predicting key thermodynamic properties of hydrocarbons.
• To establish a link between molecular-scale electronic structure and macroscopic thermodynamic behavior.
• To enable rapid and transparent property estimation for process design and thermodynamic modeling.

Main Methods:

• Generated molecular geometries and wave functions using Density Functional Theory (DFT).
• Quantitatively analyzed molecular surfaces defined on the 0.001 au electron-density isosurface to obtain descriptors.
• Employed variance-inflation-factor (VIF) pruning, LASSO, and SISSO to develop sparse linear models with 1-2 composite descriptors.
• Validated descriptor set robustness across different DFT functionals, basis sets, and solvent models.

Main Results:

• Achieved high predictive accuracy (test-set R² > 0.95) for thermodynamic properties.
• Developed concise, closed-form QSPR models with clear physical interpretability.
• Identified key descriptors related to density, size, shape, and electrostatics that rationalize property trends.
• Demonstrated the models' ability to predict volatility and critical behavior.

Conclusions:

• The developed QSPR models provide a practical bridge between first-principles calculations and macroscopic thermodynamic properties.
• These models enable rapid property estimation with transparent structure-property relationships.
• The approach facilitates improved process design and thermodynamic modeling for hydrocarbons.