In recent years, (LinK1-n)2CO3 binary melts have been investigated as the promising electrolyte matrixes of the fuel cell. Many researchers moved to the investigation of local structure and properties of the melts. In this paper, machine learning interatomic potentials were developed based on first-principle datasets in order to research the local structure and properties of molten (LinK1-n)2CO3 (n = 0.4,0.5,0.6) binary salts. The trained machine learning potentials, which enable similar accuracy relative to DFT and high efficiency, can yield precise descriptions of microstructure and macro properties of Li2CO3-K2CO3 binary melts. The information of the microstructure evolution was analyzed through partial radial distribution functions, coordination numbers and bond angle distribution functions. It is observed that lithium cation exhibits more stable coordination and destroys the C-O ion pairs. Further, comparing the inter-ionic partial radius distribution function diagrams of binary melts and those of pure Li2CO3 and K2CO3 melts, the changing regularities of interatomic strength and arrangement were obtained. Macro properties including density, self-diffusion coefficients, thermal conductivity and viscosity at target temperature range were calculated by the potentials and MD simulations. The temperature effect and concentration effect in properties were explored. This work exhibits a thorough understanding of the local structure and properties of Li2CO3-K2CO3 melts and reveals the accuracy of machine learning potentials on lithium-potassium melts for the first time.
Read full abstract