Convex two-level optimization (Q1109682): Difference between revisions

The paper deals with two-level optimization problems. A function F(x,y) must be minimized with respect to x and subject to some constraints on x. The vector y, being a function of x, is the result of another minimization; minimize f(x,y), subject to some (combined) constraints on x and y. The problem can be viewed as a Stackelberg problem in game theory: the leader must choose x. After x has become known, the follower chooses y. In the current paper more than one follower is allowed; they do, however, not interact with each other. Each follower minimizes a function \(f(x,y_ p)\), subject to some constraints on x and \(y_ p\) only; \(p=1,...,P\), where P is the number of followers. Two propositions and three corollaries give some properties of the solution (-spaces) under some convexity conditions. It is well-known that, even for quadratic F and f, and affine constraints, the solution space may admit many local minima. See for instance \textit{A. Bagchi}, ``Stackelberg differential games in economic models'' (1984; Zbl 0543.90095). An algorithm is presented which will find the global optimum for strictly convex F, quadratic f and affine constraints. Numerical experience is reported with respect to a range of sample problems (not given explicitly). The algorithm presented, compared to two other methods known in the literature, yields favorable results.