The question is number (38.2) in page 302 in Numerical Linear Algebra (Trefethen & Bau) We have a real symmetric matrix with eigenvalues: 1, 1.01, 1.02,..., 8.98, 8.99, 9.00 and 10, 12, 16, 24.
How many steps of the conjugate gradient iteration must you take to reduce the initial error by a factor of 10^6 ??
---------------------------------------------------------------------------
The answer is easy: The condition number of the matrix is
k = lamda max / lamda min = 24/1 = 24.
So, 1/10^6 = ||e_n|| / || e_0|| < = 2(sqrt(k)-1/sqrt(k)+1)^n .
Solving the equation: We get n>= 35.03 . So, we need at least 36 steps of the CG iteration.
-------------
BUT
another question arises:
How does the distribution of the eigenvalues affect the rate
of convergence ??
In other words, what if the eigenvalues were: 1,2,3,...,24 for example. what does that afect the convergence ?
The whole calculations are based on the fact that the condition number k = lamda max / lamda min.
It seems to me that the only thing that would affect that are only the maximum and the minimum eigenvalues, NOT all eigenvalues ?
which implies that the rate of convergence will not change if we have a different kind of distribution of the eigenvalues.
Is this right ? Any ideas ??
Thank you