A new proof of a theorem on optimal control of switched systems (Q878460)

This paper presents a proof of the following result. Given a control system of the form \[ \dot{x}=f_{v(t)}(t,x(x),u(t)), \tag{1} \] referred to as a two-switched system, where \(x\in R^n\), \(v(t)\in\{0,1\}\) denotes the switching control, \(u(t)\in \Omega\subset R^m\), \(\Omega\) compact and convex, is the ``usual'' control, \(v(t),u(t)\) measurable, and \(f_0,f_1\) are of the class \(C^1\). We associate with the system (1) an embedding \[ \dot{x}=(1-\alpha(t))f_0(t,x(t),u_0(t))+\alpha(t)f_1(t,x(t),u_1(t)), \tag{2} \] with \(\alpha(t)\in [0,1]\) and \(u_0(t),u_1(t)\in\Omega\). Suppose that both the systems (1) and are initialized at the same \(x_0\). Let \(\psi(t)\) stand for the trajectory of the system corresponding to a given triple \((u_0(t),u_1(t),\alpha(t))\), defined on a time interval \([t_0,t_f].\) Then the main result of the paper (Theorem 3) establishes that trajectories of (1) are dense in the set of trajectories of (2). More specifically, it has been shown that for every \(\varepsilon>0\) there exist controls \(v_{\varepsilon}(t)\) and \(u_{\varepsilon}(t)\) such that the corresponding trajectory \(\phi_{\varepsilon}(t)\) of the original system (1) satisfies the inequality \(| | \phi_{\varepsilon}(t)-\psi(t)| | <\varepsilon\) for every \(t\in [t_0,t_f]\). This density result has been originally obtained in the paper by \textit{S. C. Bengea} and \textit{R. A. DeCarlo}, ''Optimal control of switching systems'', Automatica, 41, 11--27 (2005; Zbl 1088.49018)]. A contribution of the reviewed paper lies in providing a simpler, alternative proof.