Approximation of strongly relaxed minimizers with ordinary delayed controls (Q1895791)

The paper deals with optimal control for a system of ordinary differential equations with several delayed controls. For a delay \(\theta> 0\), such system looks like \(dx/dt= f(t, x(t), u(t), u(t- \theta))\) for a.a. \(t\in (0, 1)\), \(x(0)= \xi_0\), and \(u(t)\in \Omega\) for a.a. \(t\in (-\theta, 1)\). The relaxed system employs a relaxed control (i.e., a Young measure) \(\nu: (0,1)\to \text{rpm}(\Omega\times \Omega): t\mapsto \nu_t\), where \(\text{rpm}(\Omega\times \Omega)\) stands for Radon probability measures on the compact set \(\Omega\times \Omega\). The relaxed (i.e., continuously extended) equation then takes the form \(dx/dt= \int_{\Omega\times \Omega} f(t, x(t), r_0, r_1)\nu_t(dr_0 dr_1)\) for \(t\in (0, T)\). However, only relaxed controls \(\nu\) which satisfy the condition \[ \int^1_\theta dt \int_{\Omega\times \Omega} \varphi(t, r_1)\nu_t(dr_0 dr_1)= \int^1_0 dt \int_{\Omega\times \Omega} \varphi(t, r_0) \nu_{t- \theta}(dr_0 dr_1) \] for any \(\varphi\in L^1(\theta, 1; C(\Omega))\) are admissible, i.e., weakly\(^*\) attainable by a sequence of original delayed controls. This essential result is proved in the paper by an explicit construction of such sequence.