Nonconvex global optimization problems in control theory (Q1585399): Difference between revisions

The authors give a review of their recent results related to a special class of nonconvex global optimization problems admitting effective solution. The approach developed by the authors consists of proving the basic relations of the convex duality theory for nonconvex optimization problems and of applying of the rule for solution of the problem which is based on these relations. The problem under consideration is \(G_0 (z) \to \inf\) in domain \(z \in Z, G(z) \leq 0_Y\). Here \(Z\) is an affine subspace of a linear space, the functions \(G_0: Z \to R, G: Z \to Y\) are given, and \(Y\) is a finite-dimensional real linear space ordered by a nonempty convex cone. The rule consists of carrying out the following operations. (i) For \(\tau^* \geq 0\), solve the problem \(S(\tau^*,z) \to \inf\) in the domain \(z \in Z\) restricting ourselves to the search for the value \(S_0(\tau^*)=\inf_{z \in Z} S(\tau^*,z)\) with \(S(\tau^*,z)=G_0 (z)+\tau^* G(z)\). (ii) Find a certain solution \(\tau_0^*\) to the dual problem \(S_0(\tau^*) \to \max\) in the domain \(\tau^* \geq 0\), where the maximum is necessarily attained. (iii) Find all the solutions \(z^0\) to the problem (i) with \(\tau=\tau_0^*\), and then discard those solutions \(z^0\) that do not satisfy at least one of the following relations \(G(z^0)\leq 0, \tau_0^* G(z^0)=0\). The resultant set \(\{ z^0 \}\) of remaining points coincides with the set of all the solutions to the initial problem. The method (i)-(iii) is of interest, since many practically significant classes of nonconvex problems that admit its application were found. The applicability of the method is equivalent to the validity of the Lagrange duality relation \(\max_{\tau^* \geq 0} \inf_{z \in Z} S(\tau^*,z)= \inf_{z \in D} G_0 (z)\), where \(D=\{ z \in Z: G(z) \leq 0 \}\). The paper gives a review of the main results obtained by the authors concerning an extension of the domain of applicability of the method (i)-(iii). Applications of the abstract theory to special optimal control problems are discussed in detail.