Concentration-Dependent Speciation in ZnCl2 Electrolyte Solutions: Transferability of Machine-Learning Interatomic Potentials and Speciation Sensitivity of EXAFS
Predicting speciation in concentrated electrolytes requires accurate ion interactions and adequate sampling of slowly interconverting coordination environments. We combine Zn K-edge extended X-ray absorption fine structure (EXAFS) measurements with long-timescale molecular dynamics (MD) using machine-learning interatomic potentials to study aqueous ZnCl2 from dilute conditions to 30 m. We train revPBE-based DeepMD models with D3 corrections included or excluded for Zn-containing pairs and compare them with DPMD-SCAN, classical MD (CMD), and MACE-MATPES-r2SCAN. EXAFS calculations are benchmarked against hydrated and chloride-coordinated Zn reference complexes, then applied to MD ensembles alongside Zn-Cl coordination free-energy analysis. Increasing concentration reduces octahedral coordination, replaces water with chloride, and promotes chloride-bridged association. All three DeepMD models yield similar EXAFS at 3 m, but DPMD-SCAN shows greater transferability across concentrations, reproducing measured spectral trends, key features of mixed ZnCl2/LiCl spectra, and octahedral fractions consistent with other studies. DFT treatment strongly affects coordination free energies and aggregation; excluding Zn-related D3 corrections does not systematically improve EXAFS agreement. CMD shows larger spectral discrepancies and less consistent coordination trends. MACE-MATPES-r2SCAN improves upon CMD and yields spectra comparable to DPMD-SCAN, which provides more consistent agreement and greater computational efficiency. Neighbor-resolved analysis shows that distance disorder and scattering-path interference can produce similar EXAFS signatures for distinct coordination ensembles, highlighting the need for rigorous sampling and ensemble-averaged spectra to distinguish isolated from chloride-bridged Zn species.