COSMOSol: efficient solvent screening for polymer additives with open-source COSMO-SAC.

Recycling polymers with solvents requires removing additives efficiently-yet for most of the hundreds of additives used industrially, the efficiency of various solvents to dissolve them is unknown, particularly due to the scarcity of experimental solubility data. This work proposes a systematic computational approach using the quantum mechanics-aided open-source COSMO-SAC model to screen solvents and requires only the fusion data of the additive as external input, or even no empirical input. To validate the method, we compiled solubility measurements for five representative additives across diverse solvent classes. Quantitatively, predictions typically fall within 0.5 log units of measured values (in mole fraction) and, more importantly, qualitatively correctly identify both the best and worst solvent types for each additive. Furthermore, we examined how different modeling choices affect predictions: for instance, values of fusion properties, molecular conformation and σ-profile of the additive, and dispersive interactions. When fusion data are unavailable, the infinite-dilution activity coefficient provides rankings comparable to those obtained from calculated equilibrium solubilities, particularly for poorly soluble additives. Minor limitations of the approach were observed in cases where solvents are structurally very similar-here, fine distinctions between them can be affected by interference from other effects or model simplifications. However, for choosing between different solvent classes-the decision that matters most in practice-the rankings prove robust. We additionally benchmarked COSMO-SAC against HANNA, a machine-learning activity-coefficient model: the two yield comparable solvent rankings. To facilitate reproducibility and extension of this work, we provide COSMOSol, a Python toolset containing all workflows and data. The method offers a practical, first-principles route to solvent screening-not only for polymer additives, but potentially for other compound classes where experimental data are lacking.