1: \begin{abstract}

2: This paper is concerned with the derivation of necessary  conditions for the optimal shape of a design problem governed by a non-smooth PDE. The main particularity thereof is the lack of differentiability of the nonlinearity  in the state equation, which, at the same time, is solved on an unknown domain. We follow the functional variational approach introduced in \cite{pen} where the set of admissible shapes is parametrized by a large class of continuous mappings. 

3: %This methodology allows for both boundary and topological variations. It has the advantage that one can  rewrite the shape optimization problem as a control problem in a function space.

4:  It has been recently established \cite{p1}  that each parametrization associated to an  optimal shape  is the limit of a sequence of global optima of minimization problems with convex admissible set consisting of functions. Though non-smooth, these problems allow for the derivation of an   optimality system equivalent with the first order necessary optimality condition \cite{p2}. In the present manuscript we let  the approximation parameter vanish therein. 

5: % One of the main difficulties will appear when passing to the limit in the gradient equation. In this context, we make use of the co-area formula and  we study the convergence behaviour of curvilinear integrals associated to optima of the approximating minimization problems. 

6:  The  final necessary conditions for the non-smooth shape optimization problem consist of  an adjoint equation, a limit gradient equation that features a measure concentrated on the boundary of the optimal shape and, because of the non-smoothness, an inclusion  that involves its Clarke subdifferential.\end{abstract}

7: