First calculate sums in a given processor, summing in Processor P over all nodal points j in P. Then one must form
a Global ``combine''.
Typically, this same sum must be sent to all processors.