I've been studying many iterative methods, like Jacobi, fixed-point, Newton's and the conjugate gradient methods.
Currently, I'm studying the CG method, but it's not the first time where the eigenvectors (and eigenvalues) of the usual matrix involved are important in determining how the iterative method behaves, like what's the convergence of the method.
In particular, I was reading the book "A first course in numerical methods" (by Greif, Ascher) and at a certain point (pp. 186-7) there's:
If the eigenvalues of A are located in only a few narrow clusters, then the CG method requires only a few iterations to converge. Convergence is slower, though, if the eigenvalues are widely spread.
which makes me wonder why, why if the eigenvalues of $A$ are located in an only a few narrow clusters the CG converges fast?
This is not the first time that I see similar observations where the eigenvalues and eigenvectors of the matrix involved in the iterative method somehow determine the behavior or performance of the iterative method.
My questions, apart from the specific case above, are:
What are the relations between the eigenvectors and eigenvalues of the matrices involved in an iterative method and the behaviour of the iterative method (if this can be generalized)?
What are the general behaviors of iterative methods that we can predict given the eigenvectors or eigenvalues?
If this can't be generalized, either I could ask specific questions, or you could like list a few examples where eigenvectors and eigenvalues influence different iterative methods. Of course, an exhaustive list would be nice for everyone.
Note: I know what are eigenvectors and eigenvalues, even though when I think about them I can't really make sense of why they are useful. You could argue that they are useful like in the example above, but I can't see the relation very well and clearly.