Differentiability of the optimal time vector and positional control of a linear subcritical system (Q5926580)

The authors study the differentiability conditions for the optimal time function of the linear subcritical system \(\;\dot x = A(t)x+b(t)u\), \(|u|\leq 1,\) with \(A:\mathbf R\to \text{End}(\mathbf R^n)\), \(b:\mathbf R\to \mathbf R^n\;\) belonging to the class \(C^k\), \(k\geq 0.\) The optimal time function \((t,x)\to \tau_n (t,x)\) is defined by \(\tau_n(t_0,x_0)=\min_{u(.)\in \mathcal U} \{\vartheta \geq 0: x(t_0+\vartheta,t_0,x_0,u(.))=0\}\), where \(\mathcal U\) is the set of measurable functions ranging in \([-1,1]\) and \(x(t,t_0,x_0,u(.))\) is the solution of the considered system with the control \(u=u(t)\) and the initial condition \(x(t_0)=x_0\). The authors have shown that regardless of the smoothness of \(A\) and \(b\), in an arbitrary neighborhood of the origin in the \((t,x)\)-space, there exist (nonzero) points at which the optimal time function looses its smoothness. Precisely, there exists an \(n\)-dimensional weakly invariant manifold \(\mathcal N^n\) that contains the origin and the optimal time function is not differentiable at any point \((t,x)\in \mathcal N^n\). Furthermore, at all points \((t,x)\in D=\{(t,x):\;t\in \mathbf R\), \(\tau_n(t,x)<\sigma (t)\}\) not lying on \(\mathcal N^n\), the optimal time function has one more derivative than the function \(t\to (A(t),b(t)).\) Using a sound definition of a time-optimal positional control, its existence is verified.