We present a nontrivial model system of interacting electrons that can
be solved analytically in the GW approximation. We obtain the particl
e number from the GW Green's function strictly analytically, and prove
that there is a genuine violation of particle number conservation if
the self-energy is calculated non-self-consistently from a zeroth orde
r Green's function, as done in virtually all practical implementations
. We also show that a simple shift of the self-energy that partially r
estores self-consistency reduces the numerical deviation significantly
.