The 2-coordinate descent method for solving double-sided simplex constrained minimization problems (Q463005): Difference between revisions

The author considers the following minimization problem: \[ f(x) ~\longmapsto~ \min \] subject to \[ \sum_{j=1}^n x_j = K,\, l_j \leq x_j \leq u_j,~ j = 1, \dots, n, \] where \(f:\mathbb R^n ~\longmapsto~ \mathbb R\) is a continuously differentiable function with a Lipschitz continuous gradient, and \(l_j, ~ u_j\) may be real or infinite. Several variants of a two-coordinate descent method are proposed to solve the problem. Convergence to stationary points for general non-convex functions is established. In case of a convex objective function, lower finite bounds \(l_j\) and no upper bounds of the variables, the author proves a sublinear rate of convergence. The last section of the paper is devoted to solving illustrative numerical examples showing the effectiveness of the proposed method.