Two-level hierarchically preconditioned conjugate gradient methods for solving linear elasticity finite element equations (Q914330)

Some preconditioned conjugate gradient methods for systems of linear algebraic equations obtained by the finite element method approximating second order elliptic equations in 2-D and 3-D domains are considered. The system is supposed to be generated by one or by some finite elements from the set containing the most usable triangular and rectangular elements with coordinate functions, which form the basis in the space P, where \(P^ 2\supset P\supset P_ 2\), \(P_ 2\) is the space of polynomials of second order and \(P^ 2\) is the space of polynomials of second order in each variable. The distortion of finite elements by some linear coordinate transformation is allowed. Preconditioners which are suggested to use are based on a hierarchical splitting of the finite element space. A priori preconditions of \textit{O. Axelsson} and \textit{I. Gustafsson} [Math. Comput. 40, 219-242 (1983; Zbl 0511.65079)] are replaced by those obtained from them via some fixed point iteration. For such iterations it is suggested to use multigrid iterations, incomplete factorization of the a priori preconditioner or some spectrally equivalent matrices. Relaying on an analysis of the constants in the strengthened Cauchy inequalities the authors investigate convergence factors of the conjugate gradient method using the considered preconditioners. Results of numerical experiments presented in the article confirm the efficiency of these preconditioners in comparison for instance with direct methods.