Successive Over Relaxation Method
Successive Over Relaxation Method
For this product to be less that one in absolute value, it is necessary that
|1 − ω| < 1, i.e., 0 < ω < 2. We restate this as a proposition.
Proposition 1. A necessary condition for the SOR method to converge for
any starting vector and right hand side is that 0 < ω < 2.
We shall provide a convergence theorem for the SOR method. Note that
the reduction matrix GSOR is generally not symmetric even when A is sym-
metric. Accordingly, even if A has real entries, any analysis of GSOR will
have to involve complex numbers as, in general, GSOR will have complex
eigenvalues. Accordingly, we shall provide an analysis for complex Hermit-
ian matrices A.
Definition 1. The conjugate transpose N ∗ of a general n × m matrix N
with complex entries is the m × n matrix with entries
(N ∗ )i,j = N̄j,i .
Here the bar denotes complex conjugate. An n × n matrix A with complex
entries is called Hermitian (or conjugate symmetric) if A∗ = A.
Proof. The reduction matrix for SOR is GSOR = (ω −1 D+L)−1 ((ω −1 −1)D−
U ) and we set Q = (ω −1 D + L). We shall show that ρ(GSOR ) < 1.
We note that
(5.2) (I −GSOR ) = (ω −1 D+L)−1 ((ω −1 D+L)−((ω −1 −1)D−U )) = Q−1 A.
Let x ∈ Cn be an eigenvector of GSOR with eigenvalue λ and set y =
(I − GSOR )x = (1 − λ)x. Then, by (??),
(5.3) Qy = Ax
and
(Q − A)y = (Q − A)Q−1 Ax = (A − AQ−1 A)x
(5.4)
= A(I − Q−1 A)x = AGSOR x = λAx.
Taking the inner product of (??) with y in the second place and (??) with
y in the first place gives
(Qy, y) = (Ax, y), and
(y, (Q − A)y) = (y, λAx).
These equations can be rewritten (using the fact that D is real and symmet-
ric)
ω −1 (Dy, y) + (Ly, y) = (1 − λ̄)(Ax, x), and
(ω −1 − 1)(Dy, y) − (y, U y) = (1 − λ)λ̄(x, Ax).
Now since A is Hermitian, (Ly, y) = (y, U y) so adding the two equations
above gives
(5.5) (2ω −1 − 1)(Dy, y) = (1 − |λ|2 )(Ax, x).
1A complex n × n matrix A is positive definite if (Ax, x) > 0 for all non zero x ∈ Cn .
4