We study block conjugate gradient methods in the context of continuation me
thods for bifurcation problems. By exploiting symmetry in certain semilinea
r elliptic differential equations, we can decompose the problems into small
ones and reduce computational cost. On the other hand, the associated cent
ered difference discretization matrices on the subdomains are nonsymmetric.
We symmetrize them by using simple similarity transformations and discuss
some basic properties concerning the discretization matrices. These propert
ies allow the discrete pure mode solution paths branching from a multiple b
ifurcation point [0, lambda(m,n)] of the centered difference analogue of th
e original problem to be represented by the solution path branching from th
e first simple bifurcation point (0, mu(1,1)) of the counterpart of the red
uced problem. Thus, the structure of a multiple bifurcation is preserved in
discretization, while its treatment is reduced to those for simple bifurca
tion of problems on subdomains. In particular, we can adapt the continuatio
n-lanczos algorithm proposed in [1] to trace simple solution paths. Sample
numerical results are reported. (C) 1999 Elsevier Science Ltd. All rights r
eserved.