This is because the length of the central region in Z is unknown a priori, and having vacuum on the outside and rigid bodies representing the electrodes (or their extensions, at least) allows the algorithm to optimize the length of internal part of the central region in Z.