Convexity properties of the minimum time function (Q1894803)

The paper deals with the minimum time optimal control problem governed by the system \[ \begin{aligned} y'(t) & = f \bigl( y(t), u(t) \bigr) \tag{*} \\ y(0) & = x \in \mathbb{R}^n, \quad u : [0, + \infty) \to U, \end{aligned} \] where \(U \subset \mathbb{R}^m\) is a compact set. Given a target \(K\) and denoting by \(\Omega \subset \mathbb{R}^n\) the set of points which can be driven in a finite time on \(K\), the minimum time function \(\tau : \Omega \to [0, + \infty)\) is defined as \[ \tau (x) : = \inf \biggl\{ T : \bigl( y(t), u(t) \bigr) \text{ solution of } (*),\;y(T) \in K \biggr\}. \] Under a Petrov type controllability assumption, it is proved that the semiconcavity of the distance function from the target (which can be considered as a mild regularity assumption on \(K)\) implies the semiconcavity of \(\tau\). For linear control systems a semiconvexity property can also be established. These conditions can be used to study the structure (Hausdorff dimension estimates, propagation of singularities) of the set of nondifferentiability points of \(\tau\).