On Error Correction Methods for Acceleration of Convergence

摘要

This paper provides a unified theory of the error correction methods for acceleration of convergence in linear solvers used in computational electromagnetism, which include the multigrid and deflation methods, explicit and implicit error correction methods as well as the AV method for eddy current analysis. It is shown that in all these methods the error correction with projection matrices play a crucial role, which are constructed so that their null spaces include the slowly converging numerical errors.