
120 High Performance Programming for Soft Computing
Therefore, after setting , the search direction s can be represented
as
(5.2.33)
where
D
is the length of the parallel component . The value
D
is found
multiplying both sides of Eq. (5.2.33) by
. (5.2.34)
Solving Eq. (5.2.34) for
D
ij
. (5.2.35)
Therefore, the main drawback of the conjugate Gram-Schmidt method
is that all the old search vectors must be stored to construct each new one,
affecting directly the computational cost of the algorithm (Jorge Nocedal
2006).
5.2.4 Linear conjugate gradient method
To overcome the problem of the computational cost of the Gram-Schmidt
conjugation but still ...