these operations are costly and do not scale very well;
also are you assuming you would have the same size of the field on all MPI ranks? How would boundary conditions be handled?
The operation you refer to is not "adding", but
gathering
https://openfoam-parallelisation-course.github.io/workshop/chapters/02-collective-comms/
I suggest you
algorithmically similfy your data before exchanging over MPI. For example, if you only need the global "max" value for your field, each proc can find the local max and those local maxima can be reduced with a maxOp...