About an optimal visiting problem (Q434359)

The authors study an `optimal visiting problem' which is described by a controlled dynamical system in \(\mathbb R^{n},\) i.e., \(y^{\prime }(t)=f(y(t) ,\alpha (t))\) for \(t>0\), \(y(0) =x\) where \(\alpha :[0,+\infty [ \rightarrow A\) is the measurable control and \(f\) is assumed sufficiently regular, together with a number of closed disjoint sets \(\mathcal{T}_{j}\subset\mathbb R^{n},j=1,\dots,m\) which represent \(m\) target sets which have to be reached (not necessarily in a particular order). For every control \(\alpha \) define the visiting time from \(x\) by: \[ t_{x}(\alpha) =\inf \left\{ t\geq 0\mid \exists t_{1},\dots,t_{m}\in \left[ 0,t\right] ,y(t_{j}) \in \mathcal{T} _{j}~\forall j=1,\dots,m\right\} \] The problem then consists of minimizing, over the choice of admissible controls, the necessary time for a trajectory to touch each target set resulting in the value function or ``minimum visiting'' function \(T(x) =\inf_{\alpha }t_{x}(\alpha) .\) The traveling salesman problem can be viewed as a problem of this type. The objective is to characterize \(T(x) \) via study of a Hamilton-Jacobi equation. This requires the optimality principle of dynamic programming which, however, does not hold when there are more than one target in the above problem. The authors take an approach of introducing some ``external variables'' \(w_{1}(.) ,\dots, w_{m}(.) \), one for each target, such that \(w_{j}(t) \geq 0\) and \(=0\) if and only if \(y(\tau) \in \mathcal{T}_{j}\) for some \(\tau \in [0,t].\) The evolutions of these added variables are represented by a ``hysteresis law'' representing a particular kind of memory satisfying \(w_{j}(t) =0\) implies \(w_{j}(s) =0\) for \(s\geq t\). Under a suitable ``semigroup property'' they are able to completely recover the optimality principle. They define the function \(\Psi :\mathbb{R}^{n}\times \mathbb{R}^{m}\) by \((x,w_{1},\dots, w_{m}) \mapsto w_{1}+\dots+w_{m}\). Denoting by \(w^{0}\) the initial state of \(w=(w_{1},\dots, w_{m})\) they consider the optimal control problem given by \[ V(t,x,w^{0}) =\inf_{\alpha }\Psi (y(t),w(t)) \] For any \(w^{0}>>0\), the optimal visiting function satisfies \[ T(x) =\inf \left\{ t\geq 0\right\} \mid V(t,x,w^{0})=0 \] So, if the value function \(V\) is known one can reconstruct \(T\). They then study \(V\). They prove its continuity in a suitable subspace of \(\mathbb{R}^{n}\times\mathbb{R}^{m}\) and then they derive the discontinuous Hamilton-Jacobi equation satisfied by the value function in the viscosity sense. To prove its uniqueness they transform the discontinuous equation to a continuous equation with suitable boundary conditions of the Neumann type. The paper concludes with some pointers about other formulations and approximations of their problem.