Decoding Temperature-Dependent Conformer Populations and Dipole Moments in Liquid Ethylene Glycol with Classical and Machine-Learned Potentials.

Delineating the conformer populations in neat liquid ethylene glycol (EG) has been challenging as experiments lack sufficient resolution, while simulations suffer from low accuracy and/or high computational demand. Ab initio MD (AIMD) simulations of liquid EG at ambient conditions have reported that about 20% of the molecules have their central (OCCO) dihedral in the trans conformation. Yet, adequate conformational sampling is an issue with AIMD simulations, given the limited duration of the trajectory. We describe the development of deep neural network-based potentials (DP) fitted to density functional theory data and confirm the trans population to be 22% at ambient conditions. The DP models provide well-converged structural properties and conformer populations over the entire temperature range of liquid EG. An increase in the population of trans conformers was observed with increasing temperature and is rationalized on two grounds: the increasingly similar coordination environments of trans and gauche conformers, and a decrease in the polarity of the liquid with temperature. Despite the latter, the trans population saturates to a value of 24% at 400 K as rising intermolecular distances favor the gauche conformation with intramolecular hydrogen bonding. Wannier Centroid analysis carried out using a Deep Wannier model reveals a mean molecular dipole moment in liquid EG of 3.3 D at ambient conditions, which decreases to a value of 3.0 D at 463 K.