Lexicographic geometric programming (Q1179204): Difference between revisions

For convex lexicographic programming problems of the form \(\text{lex }\min\{F(x)\mid G(x)\leq 0\}\), with \(F: R^ n\to R^ s\) and \(G: R^ n\to R^ m\), Kuhn-Tucker type optimality conditions involving a matrix of Lagrange multipliers are obtained. For these conditions to be necessary the optimal solution \(x_ *\) is required to be properly optimal; this means that there exists no \(n\in R^ n\) such that \(\nabla F(x_ *)y\) \(\text{lex}<0\) and \(\nabla G_ a(x_ *)y\leq 0\) \((G_ a\) is the subvector of \(G\) corresponding to constraints active at \(x_ *)\). This condition is rather strong; for example, it does not hold when \(F(x)={x^ 2\choose x}\), \(G(x)=x\), \((x\in R)\) and \(x_ *=0\) (the only lexicographically minimal point in this example). On the other hand, proper minimality does not imply minimality in the nonconvex case. In fact, even without convexity, the Kuhn-Tucker conditions are equivalent to proper optimality; under convexity, they imply optimality. Based on these conditions, a lexicographical maximization problem is introduced as a dual to the initial problem and a duality theorem relating them is proved. This theory is then applied to lexicographical geometric programming, of which a numerical example is presented.