1: \begin{abstract}

2: A detailed new upgrade of the FEAST eigensolver targeting  

3: non-Hermitian eigenvalue problems is presented and thoroughly discussed.

4: It aims at broadening the class of eigenproblems that can be addressed within the 

5: framework of the FEAST algorithm.

6: The algorithm is ideally suited for computing

7: selected interior eigenvalues and their associated right/left bi-orthogonal eigenvectors, 

8: located within a subset of the complex plane. 

9: It combines subspace iteration with efficient contour integration 

10: techniques that approximate the left and right spectral projectors. 

11: We discuss the various algorithmic choices that have been 

12: made to improve the stability and usability of the new non-Hermitian eigensolver.

13: The latter

14:  retains the convergence property and multi-level parallelism of 

15: Hermitian FEAST, making it a valuable new software tool for the scientific community.

16: \end{abstract}