Predicting Properties of Cyclohexene with Electronic Structure Methods through Adaptive Force Matching

Electronic structure calculations are powerful tools for determining molecular energies and structures; however, their computational cost limits their application to large systems and hinders the calculation of finite-temperature properties that require ensemble averages. We demonstrated that the Adaptive Force Matching (AFM) method can be used to bridge this gap and predict macroscopic properties of a molecule solely on the basis of electronic structure information. As a proof-of-concept, we focused our work on cyclohexene. By fitting AFM models to reference forces computed using the MP2 and B3LYP-D3 methods, we successfully predicted 16 distinct properties. For all properties with available experimental values for validation, the AFM models achieve very good agreement. For other properties, predictions from AFM models fill gaps in the missing experimental data. While models developed by fitting neat and hydrated phases separately predict the solubility of cyclohexene in outstanding agreement with experimental reference values, a mixed phase model is required for certain properties, such as interfacial tension between cyclohexene and water. The results demonstrate the feasibility and accuracy of using AFM-based molecular dynamics for predicting a wide range of macroscopic properties directly from electronic structure information, opening new avenues for computational materials science and accelerating the discovery and design of novel chemicals.