Farkas' lemma for separable sublinear inequalities without qualifications (Q732779): Difference between revisions

In 1902 Farkas established that the following conditions are equivalent whenever the involved functions are linear: {\parindent=6mm\begin{itemize}\item[(i)] \(x\in\mathbb R^n\), \(f_1(x)\leq 0,\dots,f_m(x)\leq 0 \Rightarrow f_0(x)\geq0\); \item[(ii)] \((\exists\lambda_i\geq 0) (\forall x\in\mathbb R^n)\;f_0(x) + \lambda_1 f_1(x) +\dots+ \lambda_m f_m(x) \geq0\), where \(f_i : \mathbb R^n \to\mathbb R\). \end{itemize}} In this paper, it is proved that (i) and (ii) are equivalent whenever \(f_0\) is a sublinear (convex and positively homogeneous) function and each \(f_i, i=1,\dots, m,\) is a separable sublinear function. The separable sublinear functions form a broad class of nonlinear functions. Then, as a corollary, a polar cone characterization of a level set is obtained in which it is described by separable sublinear functions. Also a version of the Farkas lemma for max functions, which arise in robust optimization, is given. Then, the sublinear programming problem with separable sublinear constructions is considered and an optimality condition for such problems is presented. It is necessary to remind that this type of problems covers robust linear programming problems which commonly arise in engineering applications. But the authors did not mention the advantages of this optimality characterization, especially by some comparisons. In the end of the paper, the Lagrangian dual of the above problems are formulated and more than the weak duality, it is shown that the strong duality always holds.