Gramians for a New Class of Nonlinear Control Systems Using Koopman and a Novel Generalized SVD

Abstract: Model reduction with error bounds in nonlinear systems with non-affine control inputs remains an active field of research. In this work we present a construction for Controllability and Observability Gramians in a class of non-affine control input systems satisfying certain induced norm properties. We do so using a combination of representational forms, including a novel function decomposition that resembles linear Singular Value Decomposition (SVD), in tandem with an additional unconventional decomposition of the dynamics, and Koopman operator theory. The resulting representation allows one to place error bounds on the $H_{\infty}$ norm on a reduced-order representation of the system computed using finite-dimensional nonlinear Controllability and Observability Gramians.