Denote the resulting system by A(uk)uk+1 = fk , and note that A is positive definite i,j
and diagonally dominant. If SFP is intended to be used as a stand-alone algorithm, then at each kth iteration, the system can be solved using, for instance, PCG or linear MG. However, in the context of a nonlinear MG with SFP used as smoother, we found that partially solving the system with a few Gauss–Seidel or SOR iterations works better; we name this SFP algorithm, solved up to some accuracy by any of the above linear solvers, the SFP2 method. Algorithm 4 describes the particular case when lexicographic Gauss–Seidel (GSLEX) is used as the inner solver.