Copositivity and constrained fractional quadratic problems (Q403649)

The paper deals with the problem of minimizing a fractional problem involving the ratio of two quadratic functions over a polytope. Following the ideas presented in [\textit{I. M. Bomze} et al., J. Glob. Optim. 18, No. 4, 301--320 (2000; Zbl 0970.90057)] for finding a global quadratic nonconvex program over the standar simplex, an exact completely positive formulation for the constrained fractional quadratic problem (CFQP) is first introduced. The completely positive condition is relaxed, and a convex semidefinite lower bounding problem is obtained. The authors prove that dual attainability is impossible for this formulation, and they propose a second dual formulation, based on a more general cone, for which this property is verified. Applications of the CFQP, and in particular of the standard fractional quadratic problem on the correction of linear systems and symmetric eigenvalue complementarity problems are discussed. Preliminary computational experience with a set of randomly generated CFQPs is reported which illustrates the quality of the lower-bounds as compared with those given by a more traditional approach, such as BARON [\textit{N. V. Sahinidis} and \textit{M. Tawarmalani}, The GAMS Solver Manual. GAMS Development Corporation. Washington (2004)]. The authors also compare their approach with the performance of Gloptipoly 3, a general-purpose semidefinite programming-based method to optimize rational functions over a semi-algebraic set.