How good is DFT for ions in water? Machine-learning assisted benchmarks of the structure of aqueous electrolytes Niamh O’Neill , Angelos Michaelides, Christoph Schran, Yusuf Hamied Department of Chemistry, University of Cambridge, UK Performing atomistic simulations of electrolyte solutions is a complex task. They require an accurate yet computationally costly electronic structure model to faithfully capture the delicate balance of interactions, as exemplified by NaCl in water – the prototypical electrolyte. Fortunately, recent simultaneous advances in machine learning potentials (MLPs) and electronic structure methods now enable this challenge to be addressed. First, developments in electronic structure codes have made high accuracy wavefunction based methods feasible for systems previously confined to density functional theory (DFT). In tandem, MLPs can now reliably act as a surrogate electronic structure model at a fraction of the computational cost. Leveraging these two developments in the field of atomistic simulations we obtain wavefunction based models with the random phase approximation (RPA) and Moller-Plessett Second Perturbation Theory (MP2) for Na Cl ions in water. The RPA model strongly agrees with experiment for structural properties. We also obtain a reference quality potential of mean force at RPA level of theory, allowing comparisons with DFT and classical force-field models. These models are primed for application to computationally intensive properties such as transport coefficients and conductivity, that are highly sensitive to their chosen reference electronic structure method.
P31
© The Author(s), 2023
Made with FlippingBook Learn more on our blog