1: \begin{abstract}
2: We consider the approximation of the smallest eigenvalue of a large parameter-dependent
3: Hermitian matrix over a continuum compact domain.
4: Our approach is based on approximating the smallest
5: eigenvalue by the one obtained by projecting the large matrix onto a suitable small
6: subspace, a practice widely employed in the literature. The projection subspaces are constructed
7: iteratively (to reduce the error of the approximation
8: where it is large) with the addition of the eigenvectors of the parameter-dependent matrix at the parameter
9: values where a surrogate error is maximal. The surrogate error is the gap between the approximation
10: and a lower bound for the smallest
11: eigenvalue proposed in [Sirkovic and Kressner, SIAM J. Matrix Anal. Appl., 37(2), 2016].
12: Unlike the classical approaches, such as the successive constraint method, that maximize such surrogate errors
13: over a discrete and finite set, we maximize the surrogate error over the continuum of all permissible
14: parameter values globally.
15: We put particular attention to the lower bound, which enables us to formally prove
16: the global convergence of our framework both in finite-dimensional and infinite-dimensional settings.
17: In the second part, we focus on the approximation of the smallest singular value of a large parameter-dependent
18: matrix, in case it is non-Hermitian, and propose another subspace framework to construct a
19: small parameter-dependent non-Hermitian matrix whose smallest singular value approximates
20: the original large-scale smallest singular value.
21: We perform numerical experiments on synthetic examples, as well as on real examples arising from parametric PDEs.
22: The numerical experiments show that the proposed techniques are able to drastically reduce the
23: size of the large parameter-dependent matrix, while ensuring an approximation error for the smallest
24: eigenvalue/singular value below the prescribed tolerance.
25: \end{abstract}
26: