Is there way to solve eigenproblems of the form

Ax = lambda Bx

and return a matrix of eigenvectors and corresponding eigenvalues in decreasing or increasing order or values from eigenvalues, like it does in ARPACK++?

Edit: Or at least is there a way to solve Ax= lambda x in such a way that you solve for only for the first few eigenvalues and eigenvectors that are the smallest or the biggest? The purpose of this is because I am trying to do modal analysis but I only really need the first 10 or 20 eigenvalues and eigenvectors from the 100s that exist.

if A and B are symmetric and one is positive definite, then you can use the following class:
http://eigen.tuxfamily.org/dox/classEig ... olver.html

Otherwise you can solve for B^-1 A.

The algorithm we use does not allow to compute only the biggest eigenvalues. If your matrix sizes are about 100, then I doubt such an algorithm would be faster anyway.