Inner approximation of state-constrained optimal control problems (Q2768035)

Let the standard optimal control problem \(P(S, F)(0,x_0)\) be given by NEWLINE\[NEWLINE\begin{aligned} \ell(x(T))\to \min\quad\text{s.t.}\quad \dot x(t) &\in F(x(t)),\;t\in [0,T]\quad\text{a.e.},\\ x(0) &= x_0,\\ x(t) &\in S,\;t\in [0,T]\text{ a.e.},\end{aligned}NEWLINE\]NEWLINE where \(\ell: \mathbb{R}^n\to\mathbb{R}\) is a real-valued function, \(F:\mathbb{R}^n\Rightarrow \mathbb{R}^n\) is a set-valued function, \(C\subset \mathbb{R}^n\) is a nonempty set and \(x_0\in S\).NEWLINENEWLINENEWLINEFor \(r> 0\) the authors present an inner approximation of the problem replacing the set \(S\) by the smaller set \(S_r= \{x: d_{\text{cl}(\mathbb{R}^n\setminus S)}(x)\geq r\}\), the time interval \([0,T]\) by the smaller interval \([\tau,T]\) (where \(\tau\in [0,T]\)), the initial condition \(x(t)= x_0\) by the condition \(x(\tau)= \alpha\) (where \(\alpha\in S_r\)) and the set-valued mapping \(F(x)\) by the mapping \(F_r(x)\) which is a special convex combination of \(F(x)\) and a suitable velocity vector \(v(x)\subset T^C_S(x)\) (the Clarke tangent cone to \(S\)) such that \(F_r(x)= F(x)\) for \(x\in S\) and \(F_r(x)= v(x)\) if \(x\) lies on the boundary of \(S\).NEWLINENEWLINENEWLINEHence the new problem \(P(S_r,F_r)(\tau,\alpha)\) has the form NEWLINE\[NEWLINE\begin{aligned} \ell(x(T))\to\min\quad \text{s.t.}\quad \dot x(t) &\in F_r(x(t)),\;t\in [\tau, T]\text{ a.e.},\\ x(\tau) &=\alpha,\\ x(t) &\in S_r,\;t\in [\tau, T]\text{ a.e.}\end{aligned}NEWLINE\]NEWLINE Denoting the optimal value of this problem by \(V(S_r, F_r)(\tau,\alpha)\) then -- assuming some standard hypotheses -- it is shown thatNEWLINENEWLINENEWLINE1. \(V(S, F)\) is Lipschitz with respect to \((\tau, \alpha)\in [0, T]\times S\),NEWLINENEWLINENEWLINE2. there exists \(c_0> 0\) such that for any \(\tau> 0\) (sufficiently small) and any \((\tau,\alpha)\in [0,T]\times \text{int }S\) it is NEWLINE\[NEWLINE|V(S_r, F)(\tau, \alpha)- V(S, F)(\tau, \alpha)|\leq c_0 r,NEWLINE\]NEWLINE 3. for any \(\varepsilon> 0\), \(r> 0\) (sufficiently small) and \((\tau,\alpha)\in [0,T]\times \text{int }S\) it is NEWLINE\[NEWLINE|V(S, F_r)(\tau, \alpha)- V(S, F)(\tau,\alpha)|\leq \varepsilon.NEWLINE\]NEWLINENEWLINENEWLINEFor the entire collection see [Zbl 0968.00020].