Analysis of parametric models. Linear methods and approximations (Q2305542)

This is a high lever discussion of parametric models that have many applications such as optimal control, uncertainty quantification, model reduction, etc. The main purpose of the paper is to provide a survey and a unifying approach. In the abstract setting of the paper, the problem takes the form of a function \(r(p)\) in some Hilbert space \(\mathcal{U}\) depending on parameters \(p\) (e.g., \(r(p)\) is describing the state) and it is the unique solution of some equation \(F(r(p),p)=0\). The Hilbert space is essential because taking an inner product in \(\mathcal{U}\) allows to define a linear map \(R:u\mapsto \langle r(p),u\rangle \in\mathbb{R}\). Then a reproducing kernel or an orthonormal basis for \(\mathcal{U}\) can be defined. The operator \(C=R^*R\) is the correlation, which has a spectral decomposition, and can have several differential factorizations, giving rise to different (known) methods and techniques. The paper analyses the equivalence and relations between all these approaches and links these to known, seemingly different, methods and applications.