Hydration Structure of Na+ and K+ Ions in Solution Predicted by Data-Driven Many-Body Potentials

Abstract

The hydration structure of Na+ and K+ ions in solution is systematically investigated using a hierarchy of molecular models that progressively include more accurate representations of many-body interactions. We found that a conventional empirical pairwise additive force field that is commonly used in biomolecular simulations is unable to reproduce the extended X-ray absorption fine structure (EXAFS) spectra for both ions. In contrast, progressive inclusion of many-body effects rigorously derived from the many-body expansion of the energy allows the MB-nrg potential energy functions (PEFs) to achieve nearly quantitative agreement with the experimental EXAFS spectra, thus enabling the development of a molecular-level picture of the hydration structure of both Na+ and K+ in solution. Since the MB-nrg PEFs have already been shown to accurately describe isomeric equilibria and vibrational spectra of small ion–water clusters in the gas phase, the present study demonstrates that the MB-nrg PEFs effectively represent the long-sought-after models able to correctly predict the properties of ionic aqueous systems from the gas to the liquid phase, which has so far remained elusive.

Understanding the Anomalous Diffusion of Water in Aqueous Electrolytes Using Machine Learned Potentials

“…Another difference between DPMD and NPFF based PMFs is their well depths; those obtained from classical NPFF MD being much deeper than the corresponding ones from DPMD, indicating that water molecules in the first hydration shells in the NPFF representation are overbound. This phenomenon of overbound hydration shell by NPFFs is consistent with previous literature reports 12,37,38,44,45 and has been primarily attributed to two factors: (i) the lack of explicit polarizability in their description of ion−water interactions and (ii) the functional form of the Lennard-Jones term. 12,45 Charge scaling based NPFFs can sometimes alleviate this issue by scaling down the ion charges (and in turn the ion−water interactions) by a factor like 0.75.…”

“…iRDFs (and related incremental probability density functions iPDFs) have been employed earlier to study the hydration structure in aqueous electrolyte solutions. 10,37,38,40,41 Figure 3 shows the iPDFs of Cs and Na ions from DPMD and NPFF simulations. The iPDFs for most of the O neighbors show Gaussian-like behavior with large deviations observed near the hydration shell radius (see SI Section S4.1.2).…”

Balance between physical interpretability and energetic predictability in widely used dispersion-corrected density functionals

“…K and P = 1 atm. 137,138 These simulations employed MB-nrg many-body potentials that were trained on CCSD(T)/CBS 2-body and 3-body energies for ion-water dimers and trimers [137][138][139][140] The cluster configurations extracted from the NPT simulations were further optimized using the MB-nrg framework implemented in the MBX software. 141 Contrary to the water clusters, dispersion-corrected revPBE exhibits significantly larger error in interaction energies compared to the DLPNO-CCSD(T)/def2-QZVPPD reference values, especially when compared to the results obtained with the dispersion-corrected PBE and PBE0 functionals as shown in Table 1 and Figure Mundy and coworkers carried out AIMD simulations of alkali-metal ions in water and compared the performance of revPBE-D3, a GGA functional, and SCAN, a meta-GGA functional.…”