On a construction of a partially non-anticipative multiselector and its applications to dynamic optimization problems
Let the sets of functions $Z$ and $Ω$ be given on the time interval $T$, let there also be a multifunction (m/f) $α$ acting from $Ω$ to $Z$ and a finite set of moments $Δ$ from $T$. The work deals with two questions: the first one is the connection between the possibility of stepwise construction (specified by $Δ$) of a value $z$ of $α(ω)$ for an unknown step-by-step implemented argument $ω\inΩ$ and the existence of a multiselector $β$ of the m/f $α$ with a non-anticipatory property of special kind defined by $Δ$; and the second question is how to build the above $β$ for a given pair $(α,Δ)$. The consideration of these questions is motivated by the presence of similar step-by-step procedures in the differential game theory, for example, in the alternating integral method, in pursuit-evasion problems posed with use of counter-strategies, and in the method of guide control. It is shown that the step-by-step construction of the value $z\inα(ω)$ can be carried out for any in steps implemented argument $ω$ if and only if the multiselector $β$ is non-empty-valued. In this case, the desired value $z$ can be selected from $β(ω)$ in step-by-step procedure for any unknown in advance argument $ω$. The key point of the work is the procedure for calculation the multiselector $β$, for which a constructive and finite-step description is given. Illustrative examples are considered that include, in particular, problems of a guaranteed result optimization under functional constraints on control and/or disturbance implementations.