1: \begin{abstract}

2: The FEAST method for solving large sparse eigenproblems is equivalent to subspace iteration with an approximate spectral projector and implicit orthogonalization.

3: This relation allows to characterize the convergence of this method in terms of the error of a certain rational approximant to an indicator function.

4: We propose improved rational approximants leading to

5: FEAST variants with faster convergence, in particular,

6: when using rational approximants based on the work of Zolotarev.

7: Numerical experiments demonstrate the possible computational savings especially for pencils whose eigenvalues are not

8:  well separated and when the dimension of the search space is only slightly larger than the number of wanted eigenvalues.

9: %This allows for the minimization of communication in the FEAST method as well as

10: %for load balancing.

11: %This allows for increase in robustness and capabilities to achieve load balancing in parallel applications.

12: The new approach improves both convergence robustness and

13:  load balancing when FEAST runs on multiple search intervals in parallel.

14: \end{abstract}

15: