No label defined (Q3972031)

The paper deals with control systems of the form (1) \(\dot x=f(t,x,u)\), \(u\in\Omega(t)\subset R^ m\), \((t,x)\in R\times R^ n\), and the corresponding differential inclusion (2) \(\dot x\in F(t,x)=\{y\in R^ n;\;y=f(t,x,u),\;u\in\Omega(t)\}\). The paper points out the non- equivalence between the first order necessary optimality conditions based on different formulations of Pontryagin's Maximum Principle (PMP), and then explains how these notions of extremality corresponding to different PMP's are related. Furthermore, the paper proposes a new definition of extremality which is better than others in the sense that it provides stronger necessary conditions for optimality. Since the new definition involves all possible smooth control representations of the corresponding new differential inclusion, it seems very hard to check effectively if a solution is extremal in the new sense. So the main part of the paper is devoted to prove that it is enough to deal only with a particular class of easily identifiable representations. The principal conclusions are the following: System \((f,\Omega)\) is called ``tame'', if the following condition \[ \limsup_{\displaystyle{{(x,y)\to(x_ 0,y_ 0)} \atop {y\in F(t,x),| y-y_ 0|\leq M| x-x_ 0|}}}\hbox {dist}(0,\{u\in\Omega(t):\;y=f(t,x,u)\})<+\infty \tag{3} \] is satisfied for \(M>0\) and \(y_ 0\in F(t,x_ 0)\), a.e. in \(R\). The differential inclusion (2) is called ``nondegenerate'', if the convex hull of the set \(F(t,x)\) has nonempty interior for all \(x\in R^ n\), a.e. in \(R\). If \((f,\Omega)\) is a regular tame control representation of (1) of the nongenerate differential inclusion (2), then the set of all extremals of (1) is just the set of all extremals of (2). This implies that a (smooth) control representation \((f,\Omega)\) of the differential inclusion (2) is the best.