
188
R.D. Fierro and P.C. Hansen
Table 2: The average amount of work
Wd(X)
in the algorithms for estimating the largest
singular value and corresponding vector of the matrix X. Here, p and I denote the average
number of power and Lanczos iterations.
POWER METHOD
Wd(L) 2p,~ 2
Wd(A) 4pmn
LANCZOS METHOD
2In 2
+
14in
+
10/3
4lmn + 81m + 6In +
1013
sponding factor for the convergence of the eigenvalues, cf. [5, w and [11, p. 62 and
(12-4-1)]. Hence, the comparison of the POWER and LANCZOS methods from [5, w for
computing eigenvalues essentially carries over to the computation of singular vectors. As