Unified duality theory for constrained extremum problems. II: Special duality schemes (Q398643): Difference between revisions

The authors consider the problem \(f(x) \rightarrow \min\), s.t. \(g_i(x)\geq 0, i=1,2,\dots,m,\; x \in X\), where \(X\) is a metric space. The paper strongly refers to Part I [ibid. 161, No. 3, 738--762 (2014; Zbl 1307.90198)] which the reader should study first, since not all used notations are defined in Part II; e.g., \(\mathcal E\) is used in (2), but is defined in Part I as \(\mathcal E: = \mathcal K -\mathcal H\). After citing some definitions and main results from Part I, they show that in 3.1 the Lagrange-type duality w.r.t. their Lagrange function \(L_w(x,\pi)=w(1,0;\pi)f(x)-w(f(\bar x)-f(x),g(x);\pi)\) and in 3.2 the Wolfe, Mond-Weir dualities w.r.t to the usual Lagrangian meet their duality scheme. In 3.2, they forgot to specify \(X\) as a locally convex topological linear space with dual \(X^*\). In Section 4, they investigate special types of separation functions \(w(u,v;\pi)\) which are all regular (Def. 2.2) and satisfy the main assumptions \(\mathcal A, \mathcal B, \widetilde{{\mathcal C}}\). Property \(\mathcal A\) generates the primal objective by the supremum of the Lagrangian over the feasible multipliers \(\pi\in\Pi\). Property \(\mathcal B\) assumes the increase of \(w\) in \(v\) with respect to the usual order cone in \(\mathbb R^m\). Property \(\widetilde{{\mathcal C}}\) is some upper estimation of \(w\) by the minimum of the scaled coordinates \(u,\pi_1v_1,\dots,\pi_mv_m\). The Case 4.1 \(w(u,v;1,\gamma)=u+\underline{w}(v,\gamma)\) with adequate above assumptions includes the usual Lagrangian with \(\underline{w}(v,\gamma)=<\gamma,v>\). Unfortunately, the example ii) \(\underline{w}(v,\gamma)=-\Delta_{\mathbb R^m_+}(<\gamma,v>)\) reduces again to i) since the oriented distance function satisfies \(\Delta_{\mathbb R^m_+}(y)=-y\). Case 4.2 studies augmented Lagrangians by \(w(u,v;1,\lambda)=u+\sup_{z\in v-\mathbb R^m_+}(<\lambda,z>-r\sigma(z))\) with an arbitrary penalty term \(\sigma(v)\), and 4.3 some convolution function \(w(u,v;1,\lambda)=c(u,\lambda_1v_1,\dots,\lambda_mv_m)\) with some adequate properties which imply above main assumptions. May be some of the duality investigations given firstly by \textit{E.-A. Weiss} [Arch. Math. 20, 538--545 (1969; Zbl 0191.48702)], followed by \textit{M. Schäuble} and \textit{K.-H. Seidler} [Wiss. Z. Tech. Hochschule Ilmenau 18, No. 3, 69--101 (1972; Zbl 0247.90054)], by \textit{R. Deumlich} and \textit{K.-H. Elster} [Math. Operationsforsch. Stat., Ser. Optimization 14, 125--149 (1983; Zbl 0524.90081); ibid. 15, 499--515 (1984; Zbl 0556.90072); Optimization 16, 780--803 (1985; Zbl 0605.90104)], by \textit{S. Dolecki} [Lect. Notes Econ. Math. Syst. 382, 384--405 (1992; Zbl 0774.90069)] and last but not least by \textit{I. Singer} [Duality for nonconvex approximation and optimization. New York, NY: Springer (2006; Zbl 1119.90002)] can be included in the suggested duality considerations.