1: \begin{abstract}

2: \noindent

3: We consider the minimization or maximization of the $J$th largest eigenvalue of 

4: an analytic and Hermitian matrix-valued function, and build on  

5: Mengi \emph{et al.} (2014, \emph{SIAM J. Matrix Anal. Appl.}, 35, 699-724).  

6: This work addresses the setting when the matrix-valued function involved is 

7: very large. We describe subspace procedures that convert the original problem into a small-scale 

8: one by means of orthogonal projections and restrictions to certain subspaces,

9: and that gradually expand these subspaces based on the optimal solutions of small-scale 

10: problems. Global convergence and superlinear rate-of-convergence results with respect to the dimensions of 

11: the subspaces are presented in the infinite dimensional setting, where  the matrix-valued function is 

12: replaced by a compact operator depending on parameters. In practice, it suffices to solve eigenvalue 

13: optimization problems involving matrices with sizes on the scale of tens, instead of the original 

14: problem involving matrices with sizes on the scale of thousands. 

15: \\

16: 

17: 

18: \noindent

19: \textbf{Key words.}

20: Eigenvalue optimization, large-scale, orthogonal projection, eigenvalue perturbation theory, parameter dependent compact operator, matrix-valued function

21: \\

22: 

23: \noindent

24: \textbf{AMS subject classifications.} 65F15, 90C26, 47B37, 47B07

25: 

26: \end{abstract}

27: