Abstract
<jats:p>In the context of global warming, batteries play a pivotal role in facilitating the transition from fossil fuels to renewable energy sources. Solid-state batteries, characterized by their solid state electrolytes (SSEs), offer substantial advantages over conventional liquid electrolytes, including higher energy density, enhanced safety, along with reduced toxicity and flammability. Despite these benefits, SSEs often struggle to achieve ionic conductivities comparable to their liquid counterparts. Consequently, establishing dependable and accurate methods for predicting ionic conductivity (IC) in promising materials remains of critical importance. The prediction of IC for SSEs at room temperature continues to be a significant challenge even with modern computing capabilities. Reliable predictions would require extremely long simulation trajectories (typically exceeding 50 ns), which are computationally infeasible for ab initio EMD simulations [1, 2]. One common workaround in EMD is to perform simulations at elevated temperatures and extrapolate the results to room temperature [3]. However, this strategy often fails due to possible phase transitions or nonphysical alterations of the potential energy surface at higher temperatures [2]. In this work, a detailed investigation is presented on the application of Nonequilibrium Molecular Dynamics (NEMD) [4, 5] simulations to accelerate ion-dynamics in solid electrolytes. When the applied field is sufficiently small, ionic conductivity can be evaluated from the proportional relationship between the applied field and the system’s current response [2, 4–6]. This study demonstrates that combining NEMD with high external fields and machine-learned interatomic potentials (MLIAPs) provides a powerful framework for high-throughput screening of ionic conductivity trends across different materials. Furthermore, the effects of time averaging against ensemble averaging on screening accuracy are investigated along with convergence behavior. Various metrics are introduced to evaluate their performance in terms of correctness, convergence speed, and statistical confidence. The results show that the proposed high-field CCD-NEMD approach can reproduce experimental</jats:p>