High Performance Computing 1
Full Multigrid
•To get a sense of the work, note there is one V cycle for each call to MGV in the inner loop of FMG. On a serial machine, this means that, at level i, it costs O(4^i) computations. Thus the total serial cost is again O( 4^m ) = O( # unknowns )
•The serial complexity is optimal, in that a constant amount of work per unknown is performed. Analysis of a parallel version is considered later.