Importance of ShiftMove

From PIMC++

In PIMC++, time slices are shared among processors like so:

Because time slices at processor boundaries are shared (i.e. both processors need to use information about that time slice) shared time slices are prohibited from being moved by any processor. This ensures that all processors have accurate information about shared a time slice. If a given slice always stayed between two processors, then it would never be moved in the simulation and the simulation would be non-ergodic (and in fact will give pathological results). To ensure this doesn't happen, there is a special move, "ShiftMove", that shifts the slices so that shared slices are now in the middle of a processor:

If your algorithm doesn't regularly call ShiftMove, this will never happen. A general rule is to call ShiftMove every time a BisectionBlock gets called. The ShiftMove will show up in the time analysis of your simulation and you can check if you are spending too much time calling it (more then 1%). In that case, you can always cut back the amount it is called. NOTE: ShiftMove is necessary EVEN when running in serial. There is still a "stuck" slice between processor 0 and itself.