1: \begin{abstract}

2: We propose a CJ-FEAST GSVDsolver to compute a partial

3: generalized singular value decomposition (GSVD) of a large matrix pair

4: $(A,B)$ with the generalized singular values in a given interval.

5: The solver is a highly nontrivial extension of the FEAST

6: eigensolver for the (generalized) eigenvalue problem and CJ-FEAST SVDsolver

7: for the SVD problem.

8: For a partial GSVD problem, given three left and right searching

9: subspaces, we propose a general

10: projection method that works on $(A,B)$ {\em directly}, and computes

11: approximations to the desired GSVD components. For the concerning GSVD

12: problem, we exploit the Chebyshev--Jackson (CJ) series to

13: construct an approximate spectral projector of the generalized eigenvalue

14: problem of the matrix pair $(A^TA,B^TB)$ associated

15: with the generalized singular values of interest, and use

16: subspace iteration on it to generate a  right subspace. Premultiplying

17: it with $A$ and $B$ constructs two left subspaces. Applying the

18: general projection method to the subspaces constructed leads to the CJ-FEAST

19: GSVDsolver. We derive accuracy estimates for the approximate spectral

20: projector and its eigenvalues, and establish a number of convergence

21: results on the underlying subspaces

22: and the approximate GSVD components obtained by

23: the CJ-FEAST GSVDsolver. We propose general-purpose choice strategies for the

24: series degree and subspace dimension.

25: Numerical experiments illustrate the efficiency of the CJ-FEAST GSVDsolver.

26: \end{abstract}