1: \begin{abstract}

2:   We propose new iterative methods for computing nontrivial extremal

3:   generalized singular values and vectors. The first method is a

4:   generalized Davidson-type algorithm and the second method employs a

5:   multidirectional subspace expansion technique.  Essential to the

6:   latter is a fast truncation step designed to remove a low quality

7:   search direction and to ensure moderate growth of the search space.

8:   Both methods rely on thick restarts and may be combined with two

9:   different deflation approaches.  We argue that the methods have

10:   monotonic and (asymptotic) linear convergence, derive and discuss

11:   locally optimal expansion vectors, and explain why the fast truncation

12:   step ideally removes search directions orthogonal to the desired

13:   generalized singular vector.  Furthermore, we identify the relation

14:   between our generalized Davidson-type algorithm and the

15:   Jacobi--Davidson algorithm for the generalized singular value

16:   decomposition.  Finally, we generalize several known convergence

17:   results for the Hermitian eigenvalue problem to the Hermitian positive

18:   definite generalized eigenvalue problem.  Numerical experiments

19:   indicate that both methods are competitive.

20: \end{abstract}

21: