Convex envelope of bivariate cubic functions over rectangular regions (Q2301177)

Convex envelope of a given non-convex function \(f\) defined on some region \(X \subset R^n \) is the best underestimator of \(f\). Generating set \(G(f, X)\) of \(f\) over \(X\) is the smallest subset of \(X\), which can replace \(X\) in the convex envelope of \(f\) over \(X\). When \(X\) is a polytope, then the vertex set of \(X\) is equal to the generating set \(G(f,X)\). Non-polyhedral convex envelopes are proper subsets of \(G(f,X)\) and under appropriate assumptions about \(f\) the generating set is a subset of the border of \(X\). There are known cases in the literature, in which \(G(f,X)\) is not a subset of the border of the set \(X\) for the case of univariate functions, (see e.g. [\textit{C. E. Gounaris} and \textit{C. A. Floudas}, J. Glob. Optim. 42, No. 1, 51--67 (2008; Zbl 1173.90503)]). According to author's knowledge, there are no such results for bivariate functions. The new result of the present paper consists in deriving the convex envelope over a rectangular region for a class of bivariate cubic functions. A possibility of extending the methodology used in the paper to a class of more general bivariate functions is briefly mentioned in the conclusions of the paper.