Correlação da solubilidade do pazopanibe em misturas de cossolventes (acetato de etila + etanol) e (acetato de etila + 2-propanol) em diferentes temperaturas usando diferentes modelos matemáticos

Objectives: To evaluate the predictive capacity and relevance of six mathematical models for correlating Pazopanib solubility in ethyl acetate + ethanol and ethyl acetate + 2-propanol cosolvent mixtures. Methods: A Python-based computational framework using nonlinear optimization algorithms (Levenberg-Marquardt) was employed to fit van't Hoff, Buchowski-Ksiazczak (lh), Apelblat, van't Hoff-Yaws, Wilson, and NRTL models to experimental data ranging from 288.15 K to 328.15 K. Results: Although three-parameter models (Apelblat and van't Hoff-Yaws) yielded nearly perfect statistical fits (R2 »0.9999), they suffered from severe multicollinearity and coefficients lacking physical significance. Two-parameter semi-empirical models, particularly van't Hoff and lh, demonstrated greater stability and thermodynamic consistency. Activity coefficient-based models (Wilson and NRTL) exhibited severe numerical instability and an order-of-magnitude increase in residual errors. Conclusions: For industrial design and solubility prediction, the van't Hoff and Buchowski-Ksiazczak models are technically superior due to their parsimony and statistical robustness, whereas activity-based models prove unreliable for this specific system.