The recent random batch Ewald algorithm, originating
from a stochastic
approximation, performs 1 order of magnitude faster than the mainstream
algorithms such as the particle–particle particle-mesh method
for handling of long-range electrostatics in large-scale simulations.
However, this algorithm fails to fully capture the long-range electrostatic
correlations. Here, we demonstrate that, when incorporating a known
screening condition in the stochastic approximation, the algorithm
could be simply amended without the loss of any efficiency.