A coordinate gradient descent method for nonsmooth separable minimization (Q959979)

The authors presented a block coordinate gradient descent method for minimizing the sum of a smooth function and a convex separable function. The method may be viewed as a hybrid of gradient-projection and coordinate descent methods, or as a block coordinate version of descent methods in [\textit{J. V. Burke}, Math. Program. 33, 260--279 (1985; Zbl 0581.90084)] and [\textit{M. Fukushima} and \textit{H. Mine}, Int. J. Syst. Sci. 12, 989--1000 (1981; Zbl 0467.65028)]. The global convergence is proved and, under a local Lipschitzian error bound assumption, linear convergence of this method. The local Lipschitzian error bound holds under assumptions analogous to those for constrained smooth optimization, e.g. the convex function is polyhedral and the smooth function is (nonconvex) quadratic or is the composition of a strongly convex function with a linear mapping. The numerical results are presented to verify the practical efficiency of the method.