Abstract
Ab initio molecular dynamics provides a first-principles framework for aqueous systems but is limited in accuracy by self-interaction error (SIE) in exchange-correlation functionals. While SIE corrections are known to reduce the over-structured hydrogen bonding in bulk water, their impact on ion hydration remains unclear. Here, we employ machine learning molecular dynamics to systematically compare the Perdew–Burke–Ernzerhof functional with Tkatchenko–Scheffler–van der Waals dispersion correction and the strongly constrained and appropriately normed (SCAN) functional, thereby investigating the effects of different treatments of exchange-correlation (including SIE) on the hydration structure and electronic properties of Na+ and Cl− in aqueous solutions. We find that the SCAN functional yields a more ordered and compact first hydration shell of Na+, while leaving the orientational distribution of water molecules largely unchanged. In contrast, for Cl−, the Cl–water correlation of the first hydration shell remains nearly unchanged, whereas the orientational distribution of surrounding water molecules is significantly modified. Charge density analysis reveals that these differences originate from ion-dependent electronic responses: the charge density is more localized around Na+ under the SCAN description, strengthening Na+–water interactions, whereas Cl− is comparatively less affected. These results reveal that ion-specific hydration responses are governed by electronic structure effects, particularly charge-density differences associated with reduced SIE, and highlight the importance of accurate treatment of electronic structure in aqueous simulations.