1: \begin{abstract}

2: This paper describes the software package Cucheb, a GPU implementation of the

3: filtered Lanczos procedure for the solution of large sparse symmetric

4: eigenvalue problems. The filtered Lanczos procedure uses a carefully chosen

5: polynomial spectral transformation to accelerate convergence of the Lanczos

6: method when computing eigenvalues within a desired interval. This method has

7: proven particularly effective for eigenvalue problems that arise in electronic

8: structure calculations and density functional theory. We compare our

9: implementation against an equivalent CPU implementation and show that using the

10: GPU can reduce the computation time by more than a factor of 10.

11: \end{abstract}

12: