The development of new solid-state electrolytes is a key step
in improving the performance and safety of battery technology.
Although the use of first-principle methods has proved invaluable
in better understanding the process at play in these materials,
these methods remains extremely costly and limit the ability to
model the diffusion phenomena as this one is often happening
over large time-scales. To solve this issue and unlock
larger time-scale and supercells, the use of force-fields has
proven to be a effective solution. In particular, polarizable
force-fields have been shown to be effective at reproducing
accurate diffusion results. To this effect, a methodology is
proposed here for the training of such polarizable force-fields
using a Self-Adaptive Differential Evolution algorithm. The constant
optimization of the shell positions is avoided by using its optimal
position with respect to the error on cores. Furthermore, the
generation of synthetic training sets is proposed through the use
of Monte-Carlo dynamics and random thermal displacements. The potential
of force-field modeling is then demonstrated by investigating the
effect of tungsten doping on garnet type electrolytes. This investigation
shows the importance of averaging over dopant distributions and
highlights the complex interplay between the various effects resulting
of the insertion of doping species. These various effects are isolated
through the use of two distinct doping models, an implicit model where
the extra positive charge is introduced as a background charge and an
explicit one where the dopant is explicitly introduced. Finally
the computation of the electrochemical stability of solid-state
electrolytes is introduced. The different methods used to compute it are
discussed and their results for relevant Li- and Na-based
solid-state electrolytes are compared.