A class of algebraic multi-p methods for solving the p-version of the
finite element equations is first presented and discussed. It is then
shown how these multi-p methods can be used as preconditioners for the
conjugate gradient (CG) method. In particular, it is shown that given
any preconditioner M-p to CG, a multi-p preconditioner B-p based on M
-p can be constructed, which leads to a smaller condition number (and
hence faster convergence). Numerical experiments on representative pro
blems indicate that the condition numbers after multi-p preconditionin
gs are, in fact, independent of p. The numerical results also show gre
ater efficiency for the preconditioned conjugate gradient (PCG) method
with the multi-p preconditioners in terms of number of iterations and
CPU time when compared with two sophisticated linear equation solvers
: (1) a direct frontal solver specially designed for the p-version of
the finite element analysis; (2) a highly tuned PCG code in ITPACKSC,
http://www.netlib.org.itpack/, 1994. Preliminary comparisons of the nu
mber of iterations are also made with ROCKITS [Solvers International,
Inc., Boulder, CO, 1994], a new commercial code used for the p-version
.