Predicting the Flory–Huggins Parameter for Polymer–Solvents From Quantum Chemical Descriptors

ABSTRACT This study develops a predictive model for the Flory–Huggins interaction parameter ( χ ) using quantum chemical descriptors and accounting for temperature effects. A dataset of 2474 χ values across 19 polymers and 88 solvents was utilized. An optimized convolutional neural network (CNN) model, incorporating 30 selected features, demonstrated superior performance, achieving a mean absolute error (MAE) of 0.140, a root mean squared error (RMSE) of 0.177, and a coefficient of determination ( R 2 ) of 0.973 on the test set. These results represent a significant improvement over existing quantitative structure–property relationship (QSPR) models (particularly for those with test set sizes exceeding 150 samples, which typically report RMSE > 0.25 and R 2 < 0.94). Mechanism analysis revealed that χ is primarily governed by charge‐related properties (e.g., the most positive hydrogen charge in the polymer, H_p, and the most negative atomic charge in the solvent, N_s), polarity properties (polymer and solvent dipole moments, μ_p and μ_s), and solvent thermal energy (E_s). These factors collectively regulate the balance between specific molecular interactions and disorder effects. This study delivers not only a state‐of‐the‐art predictive tool but also physical insight, establishing a new paradigm for the rational design of advanced polymer materials.