Optimal Control with State Constraints: Capture Basin is Closed under a Convexity Assumption.

39 Views Asked by At

I'm currently reading the paper "Reachability and minimal times for state constrained nonlinear problems without any controllability assumption" from Bokanowski, Forcadel, and Zidani.

We consider a controlled system \begin{align} \label{system} &\dot{y}(s)=f(y(s),\alpha(s))\quad \text{for a.e. }s\in[0,T], &y(0)=x, \end{align} where $y(s)\in\mathbb{R}^d$ represents the position at time $s\in[0,T]$ with initial position $x\in\mathbb{R}^d$. $\alpha:\,[0,T]\rightarrow\mathcal{A}$ is a measurable function which maps the time $s$ into the compact and non-empty set of admissible controls $\mathcal{A}\subset\mathbb{R}^m$. The right hand side $f:\,\mathbb{R}^d\times\mathcal{A}\rightarrow\mathbb{R}^d$ is a continuous function which models the dynamics. We have to make the following two assumption on $f$:

Assumption1: The function $f$ fulfills the following lipschitz condition and is bounded: There exists an $L_f\geq 0$ such that for every $x,\tilde{x}\in\mathbb{R}^d$ and $a\in\mathcal{A}$, \begin{align*} &\vert f(x,a)-f(\tilde{x},a)\vert\leq L_f\vert x-\tilde{x}\vert\quad\text{and}\\ &\vert f(x,a)\vert\leq L_f. \end{align*}

Assumption2: For every $x\in\mathbb{R}^d$, the set $f(x,\mathcal{A})\subset\mathbb{R}^d$ is convex.

I understand that Assumption 1 implies that the system has a unique (absolutely continuous) solution on $[0,T]$. Now, we define the Capture Basin.

Definition (Capture Basin): Let $\mathcal{K}\subset\mathbb{R}^d$ be a non-empty and closed set of state contrains and let the target $\mathcal{C}\subset\mathcal{K}$ be non-empty and closed, too. Then, for a given time $t\in[0,T]$, the Capture Basin is defined by \begin{align*} CapB(t):= \left\{x\in\mathbb{R}^d\mid\exists\alpha :[0,T]\rightarrow\mathcal{A} \text{ measureable},\,y_x^\alpha(t)\in\mathcal{C}\,\text{and}\,y_x^\alpha(\theta)\in\mathcal{K}\,\forall\theta\in[0,t]\right\}, \end{align*} where $y_x^\alpha$ denotes the solution of our System depending on the intial position $x$ and control $\alpha$.

In the paper it is said that Assumption 2 implies that $CapB(t)$ is closed (Remark 1). This is what I don't understand. Does anybody have an idea?