next up previous

2.4.3 Conjugate Gradient     continued...

For example, when the bound on convergence measures the size of with respect to the -norm,

we have [32,45]

Clearly rapid convergence is expected when , as for SD. Further estimates of convergence bounds can only be derived when certain properties about the eigenvalue distribution are known (e.g., m large eigenvalues and n-m small eigenvalues clustered in a region ) [32].