Metric inequality, subdifferential calculus and applications (Q5945284)

The aim of the paper is the presentation of calculus rules for the limiting Fréchet subdifferential of not necessarily Lipschitz functions on Asplund spaces and the application of these results to the derivation of necessary optimality conditions for nonsmooth constrained optimization problems. The main tool of the approach is the metric inequality, i.e., a relation between the distance \(d_{S_1\cap\cdots\cap S_n}(x)\) and the sum of distances \(d_{S_1}(x)+\cdots+ d_{S_n}(x)\) for sets \(S_1,\dots, S_n\subset X\) and points \(x\not\in S_1\cap\cdots\cap S_n\). In the first part of the paper, the authors establish a condition which ensures the metric inequality in Asplund spaces. Moreover, it is pointed out that this condition characterizes even the Aspund property. In the second part, a chain rule for the limiting Fréchet subdifferential of the composite function \(g(f_1,\dots, f_n): X\to\mathbb{R}\) is given where \(f_1,\dots, f_n: X\to \mathbb{R}\) and \(g: \mathbb{R}^n\to\mathbb{R}\) are (extended real-valued) lower-semicontinuous functions. Specifying the function \(g\), the authors present as corollaries special calculus rules for the maximum, the sum, the product and the quotient of functions. In the final part of the paper, the authors apply these results to formulate necessary Lagrange multiplier rules for the minimizers of non-Lipschitz optimization problems with equality, inequality and set constraints.