scientific article; zbMATH DE number 4026217

From MaRDI portal
Publication:3767923





[For the entire collection see Zbl 0619.00014.] The usual distributed control system \(Ay=Bv\) involves a linear partial differential operator A, a linear operator B, and a control function v. These ingredients generate the state y, a function whose form depends upon the choice of v. In the presence of uncertainty, the control system may be modelled by \(Ay=Bv+\beta g\), where \(\beta\) is linear and g belongs to some space of possible perturbations. The resulting state \(y=y(v,g)\) depends upon the control v and the noise g, and a control \(\hat u\) is called Pareto optimal for the quadratic objective J(v,g) if no competing control u obeys both (a) J(u,g)\(\leq J(\hat u,g)\) for all g and (b) there exists \(g_ 0:\) \(J(u,g_ 0)<J(\hat u,g_ 0)\). This paper describes necessary and sufficient conditions for Pareto optimality (relative to a given control \(u_ 0)\) in rather general examples of elliptic and parabolic type, and illustrates a general method for deriving such conditions. The Lagrange multipliers which arise from the state constraint implicit in the Pareto objective must be in generalized spaces; Example 3.2 shows that the dual space of \(H^{5/2}(\partial \Omega)\) is an unavoidable choice for some simple problems.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3767923)