A multi-grid method for variational inequalities in contact problems (Q1313250)

The quadratic programming problem (1) \(1/2(Ax,x)- (b,x)\to \min\), \(x_ j\geq 0\), \(j\in J\); \(A: \mathbb{R}^ n\to\mathbb{R}^ n\); \(b,x\in\mathbb{R}^ n\) is considered. Here \(A\) is a positive definite symmetric matrix, \(J\subset\{1,2,\dots,n\}\). The problem (1) is derived as the result of discretization of a corresponding variational inequality using various grid methods. A multigrid method for solving (1) similar to those studied by \textit{J. Mandel} [Appl. Math. Optimization 11, 77-95 (1984; Zbl 0539.65046)], \textit{A. Brandt} and \textit{C. W. Cryer} [SIAM J. Sci. Stat. Comput. 4, 655-684 (1983; Zbl 0542.65060)] is constructed. The main distinction of the proposed method consists in the procedure of forming initial guesses to solve problems on auxiliary levels. The set of feasible increments of a current solution for the auxiliary problem is broader in general than in known algorithms of multigrid type. The convergence of the proposed method is proved and it is shown that the objective function is monotonically decreasing when the iterations are carried out. The results of numerical tests for an elasticity contact problem are presented.