A regularized domain decomposition method with Lagrange multiplier (Q877781): Difference between revisions

A new regularized method, in which the regularization term acts on the kernel of the underlying operator is proposed. This design can avoid the generation of a great roundoff error when the regularization parameters are very small. For the regularized method, the interface equation of the multiplier can be built directly, but the condition number of the interface operator depends on the regularization parameters. In order to solve the interface equation in an efficient manner a cheap preconditioned iterative method is developed, in which one needs to solve a small coarse problem and some local problems only one time at each iteration step (here each local solver is defined on the common space between two neighbouring subdomains). This iterative method can be viewed as a preconditioned conjugate gradient (PCG) method on a subspace with a block-diagonal preconditioner, and possesses a PCG-type convergence speed, which is determined by the condition of the preconditioned system. Main result: The author shows that the proposed condition number grows only as the logarithm of the dimension of the local problem associated with an individual substructure, and is independent of the regularization parameters. The optimal error estimate of the regularized approximation, a preconditioned iterative algorithm for solving the interface equation and the convergence of the iterative algorithm are established. Precise proofs of a series of technical lemmas (sufficiently detailed) are presented. Finally, the author reports some numerical results to confirm the theoretical results.