Why classical models fail for deep eutectic solvents: high-throughput evidence of cooperative packing and network-controlled viscosity.

Deep eutectic solvents (DES) offer a sustainable frontier in solvent design, yet their vast combinatorial space makes experimental trial-and-error inefficient. Here, we developed an ultra-fast Python-based pipeline using RDKit to screen nearly 1000 DES mixtures. To evaluate classical predictive models, we benchmarked the Hoftyzer-Van Krevelen (HVK) method and Eyring's hole theory against a chemically diverse dataset of 15 experimentally characterized DES systems. Our findings reveal that classical models fail systematically because they neglect crucial physical interactions. The 2D additive HVK model constantly underestimates liquid density (MAPE = 21.01%) as it cannot account for the volume shrinkage caused by tight hydrogen-bond packing. Furthermore, hole theory majorly under-predicts dynamic viscosity (MAPE = 120.39%), often estimating values below 10 mPa s for highly viscous DESs (>1000 mPa s). This occurs because it assumes that flow is a single-molecule jump, ignoring the energy required to break the cooperative supramolecular network. We conclude that while classical models serve excellently as ultra-fast "first-tier negative filters", screening 1000 systems in ∼0.05 seconds to eliminate unviable candidates, accurate DES property prediction strictly requires explicit 3D quantum mechanical methods, such as COSMO-RS.