Nonsmooth Hessenberg differential-algebraic equations (Q1995798)

In this paper, the authors study parametric Hessenberg differential-algebraic equations (DAEs) of size \(\nu\) with nonsmooth right-hand side functions. The authors establish a comprehensive well-posedness theory that extends classical theory including existence, uniqueness, maximal continuation and Lipschitzian/lexicographic dependence on parameters. Hessenberg DAEs are a useful modeling tool for various dynamic problems in science and engineering such as constrained mechanical systems, robotics control problems, electrical circuits and incompressible fluid dynamics. The classical theory needs assumptions on sufficient smoothness of right-hand side functions. However, the relaxation of smoothness assumptions is necessary for modeling a broad range of hybrid and discontinuous systems. The existing nonsmooth DAE theory is limited to semi-explicit DAEs of ``index one'' that behave locally like nonsmooth ODEs. This work is concerned with nonsmooth Hessenberg DAEs of arbitrarily high differentiation index in generalized sense which can be verified with matrix-theoretic conditions involving, for example, Clarke-Jacobian projections. Existence and maximal continuation of regular solutions are established. Furthermore, regular solutions is shown to be lexicographically smooth with respect to parameters. As a consequence, a sensitivity analysis is presented by an auxiliary, nonsmooth DAE system which admits lexicographic derivatives as its unique solution. Thanks to the recent progress in nonsmooth automatic/algorithmic differentiation, the parametric sensitivity functions can be evaluated and supplied to dynamic optimization methods. As an application, (open-loop) optimal control theory that is numerically implementable is given in details. Applying these results to smooth Hessenberg DAEs, \(C^1\) parametric dependence of regular solutions is formalized, forward sensitivity analysis is obtained and (open-loop) optimal control is developed for smooth Hessenberg DAEs in a rigorous manner.