Solvation strategies for free-energy calculations in a halogen-bonded complex: implicit, explicit, and machine learning approaches.

In pursuit of an efficient solvation approach for the halogen bonded complex between molecular iodine and tetramethylthiourea that reliably reproduces experimental trends, we investigated a range of solvent models, from implicit representations to periodic metadynamics simulations alongside micro-solvation and ONIOM-based methods as robust alternatives. Implicit solvent models fail to describe halogen-bonded complexes in high-polar solvents but provide surprisingly accurate estimates of binding free energies in all low to moderately polar solvents. For accurate and reliable modeling, especially in polar media, explicit solvent representations are essential. Periodic metadynamics simulations typically provide enhanced accuracy in calculating free energy differences, particularly for systems with complex solvation behavior. However, they are computationally demanding and restricted to generalized gradient approximation functionals (GGA). To overcome this limitation and improve accuracy, we employed the machine learning perturbation theory technique, enabling the estimation of free energies at levels of theory beyond the GGA.