1: \begin{abstract}

2: \noindent

3: This work concerns the minimization of the pseudospectral abscissa of a matrix-valued function

4: dependent on parameters analytically. The problem is motivated by robust stability and 

5: transient behavior considerations for a linear control system that has optimization

6: parameters. We describe a subspace procedure to cope with the setting when the matrix-valued

7: function is of large size. The proposed subspace procedure solves a sequence of reduced problems

8: obtained by restricting the matrix-valued function to small subspaces, whose dimensions increase

9: gradually. It possesses desirable features such as the global convergence of the minimal values

10: of the reduced problems to the minimal value of the original problem, and a superlinear convergence

11: exhibited by the decay in the errors of the minimizers of the reduced problems. 

12: In mathematical terms, the problem we consider is a large-scale

13: nonconvex minimax eigenvalue optimization problem such that the eigenvalue function

14: appears in the constraint of the inner maximization problem. Devising and analyzing 

15: a subspace framework for the minimax eigenvalue optimization problem at hand with the eigenvalue

16: function in the constraint require special treatment that makes use of a Lagrangian and dual variables.

17: There are notable advantages in minimizing the pseudospectral abscissa over maximizing the

18: distance to instability or minimizing the $\mathcal{H}_\infty$ norm; the optimized pseudospectral

19: abscissa provides quantitative information about the worst-case transient growth, and the initial

20: guesses for the parameter values to optimize the pseudospectral abscissa can be arbitrary, unlike the case 

21: to optimize the distance to instability and $\mathcal{H}_\infty$ norm that would normally require initial 

22: guesses yielding asymptotically stable systems.

23: 

24: \end{abstract}

25: