Convexification and concavification methods for some global optimization problems (Q1883970)

This paper is devoted to the equivalent transformation of non-convex programming problems into some more structured classes like concave minimization problems, reverse convex problems and D.C. programming problems. The paper is divided into two parts, one devoted to monotone problems and another one dealing with non-convex objectives. The transformation schemes in the first part are defined by \(\Phi_{p,h}(y)=h(\frac{1}{p}\ln y)\) and \(\Psi_{p,h}(y)=h(-\frac{1}{p}\ln y)\). It is then proved that, when \(h\) has partial derivatives with a positive lower bound, then \(\Phi_{p,h}(.)\) is concave and \(\Psi_{p,h}(.)\) is convex for sufficiently large values of \(p\). Similar statements hold for negative lower bounds of the partial derivatives of \(h\). Using these transformations the authors prove that optimization problems with monotone data (i.e. satisfying some of the bounds for the partial derivatives) can be equivalently stated as D.C. programming problems. In the second part minimization problems with equality and box constraints are considered. For the first result it is supposed that the equality constraints \(\{g_i\}_{i=1}^{m}\) are convex or concave with suitable nonzero bounds for the minimal or maximal eigenvalues of the Hessian matrix. The transformation schemes used are \(\Upsilon_p(x)=g_0(x) + p^{2} \sum_{i \in I} g_i(x) - p^{2} \sum_{j \in J} g_j(x)\) and \(\Psi_p(x)=g_0(x) - p^{2} \sum_{i \in I} g_i(x) + p^{2} \sum_{j \in J} g_j(x)\), where \(g_0\) states for the objective functions, and \(I\) and \(J\) denote the index sets of convex and concave equality constraints respectively. It is then proved that \(\Upsilon_p(x)\) is convex and \(\Psi_p(x)\) concave for sufficiently large \(p\). A combination of this result and the transformation schemes in the first part is then used to propose similar transformation schemes for the case of strictly monotone constraints and non-monotone objective function. Finally it is proved that under suitable additional assumptions the problem considered in the second part can be converted into an equivalent concave minimization, or reverse programming problem or canonical D.C. programming problem.