Computational study reveals accurate infinite-dilution activity coefficient predictions across diverse chemical systems, indicating improved solvent screening.
Key Points
To develop an accurate, physics-guided 3D molecular framework for predicting temperature-dependent infinite-dilution activity coefficients used in solvent screening and separation processes.
Trained and evaluated the Physics-Guided Solute-Solvent Interaction (PGSSI) framework using a benchmark dataset of 39,840 samples encompassing 695 solvents, 559 solutes, and 13,222 unique solute–solvent pairs.
Assessed predictive accuracy across a temperature range of 249.84 to 555.60 K using a fixed test set, with targeted evaluations on isomeric pairs, hydrogen-bonded systems, and aqueous mixtures.
PGSSI achieved a mean absolute error (MAE) of 0.2043 and an R² of 0.9536 on the fixed test set across the full temperature range.
The framework demonstrated 3D-aware predictive gains on isomeric pairs and captured temperature-dependent thermodynamic trends effectively across both hydrogen-bond-rich and water-containing systems.