Anatomy of the Shape Hessian

Abstract

The computation of the Shape Gradient with respect to domain perturbations plays a central role in the theory and numerical solution of Shape Optimization problems. In 1907 J. Hadamard introduced a method which has been and still is widely used to obtain many useful results for applications. The mathematical limitation of his method rests in the fact that the deformations of the domain are a function of the smoothness of the normal to the boundary (hence the smoothness of the boundary). New developments by the Nice School (J. Cea and J. P. Zolesio) using arbitrary velocity fields of deformation relaxed the condition that the deformation be carried by the normal to the boundary. Finally the use of «Shape Lagrangians» by Delfour and Zolésio made it possible to obtain Shape Gradients by a simple constructive method which does not require the derivative of the state with respect to the domain. In this paper we apply this last method to semi convex cost functions. This extension makes it possible to compute the «Shape Hessian» or «Shape directional second derivative». We give several expressions for the «Shape Hessian» and a set of equations characterizing its kernel. © 1991 Fondazione Annali di Matematica Pura ed Applicata.