Convexity conditions and the Legendre-fenchel transform for the product of finitely many positive definite quadratic forms (Q607786)

Optimization problems having a product of convex functions as an objective or a constraint have been extensively investigated in the field of global optimization. These problems may find applications in such areas as microeconomics, geometric design, finance, VLSI chip design, and system reliability. However, they are not easy to solve because a product of convex functions is not convex in general. Hence the fundamental question is: when is the product of finitely many convex functions convex? Answering this question could identify a subclass of optimization problems that can be computationally tractable. Another fundamental issue associated with the product of convex functions is: what is the Legendre-Fenchel transform of the product of finitely many convex functions? The aim of this paper is to address these questions in the case of finitely many positive definite quadratic forms. The contribution is twofold: a general sufficient convexity condition for the product of quadratic forms is established and an explicit expression of its Legendre-Fenchel transform is derived. First, the convexity result claims that if the condition number of ``scaled matrices'' are not too large (bounded above by a constant which depends on the number of quadratic forms), then the product function is convex. Secondly, if the product function is convex, then its Legendre-Fenchel transform can be explicitly expressed as a nonnegative function which is positively homogeneous of degree \(\frac{2m}{2m-1}\), where \(m\) is the number of quadratic functions. The analysis in the paper shows that the computation of the Legendre-Fenchel transform can be implemented via solving a system of smooth equations with a special structure.