Genetic Algorithm Driven Force Field Parameterization for Molten Alkali-Metal Carbonate and Hydroxide Salts

Abstract

Molten alkali-metal carbonates and hydroxides play important roles in the molten carbonate fuel cell and in Earth’s geochemistry. Molecular simulations allow us to study these systems at extreme conditions without the need for difficult experimentation. Using a genetic algorithm to fit ab intio molecular dynamics-computed densities and radial distribution functions, as well as experimental enthalpies of formation, we derive new classical force fields able to accurately predict liquid chemical potentials. These fitting properties were chosen to ensure accurate liquid phase structure and energetics. Although the predicted dynamics is slow when compared to experiments, in general the trends in dynamic properties across different systems still hold true. In addition, these newly parametrized force fields can be extended to the molten carbonate–hydroxide mixtures by using standard combining rules.

First-Principles Modeling of Transport Mechanisms in Carbonate–Hydroxide Electrolytes

“…The details of the simulations and models used are reported in another work. 21 From these simulations, we computed the diffusion coefficients of hydroxide (D OH − ) and carbonate (D CO 3 2− ) ions, and their values are 0.129 and 0.054 Å 2 /ps, respectively. Although estimating these diffusion coefficients from the AIMD trajectory should be possible in principle, the simulation time of <100 ps is not long enough to obtain reliable values.…”

“…Therefore, we should expect a higher magnitude of total conductivity than the one computed by using classical force fields. 21 We believe these simulation results will facilitate the fundamental understanding and development of electrolytes for various applications such as capacitors, batteries, and fuel cells.…”

“…19,20 Details of the force field parameters and protocols are described elsewhere. 9,21,22 The equilibrated supercell from the classical MD trajectory was geometry optimized within the density functional theory (DFT) framework at the same level of theory as described below. The gradients on the wave functions and on the nuclear positions were optimized with convergence criteria of 10 −6 and 10 −3 au, respectively.…”

Predicting chemical reaction equilibria in molten carbonate fuel cells via molecular simulations

“…and chemical potentials. In an accompanying work, we have parameterized new force fields for alkali carbonate and hydroxide electrolytes that accurately reproduce both thermodynamic and transport properties 28 . Specifically, these models predict the experimental chemical potentials of the pure carbonate and hydroxide salts to within ∼1 kT .…”

Sintering Rate and Mechanism of Supported Pt Nanoparticles by Multiscale Simulation

“…In view of the characteristics of the data and computational complexity, the parameter fitting using traditional optimization algorithms (such as the Levenberg–Marquardt method) is time-comsuming. , To obtain the appropriate force field parameters efficiently and accurately, this paper uses genetic algorithms (GAs) to implement the fitting of force field parameters. Genetic algorithm is an artificial intelligence global probability search methodology that uses the process of evolution and natural selection theory and is under the umbrella of evolutionary computing algorithm. This method can be effectively used for the optimization and nonlinear fitting of the force field parameters of the current system.…”