Implementace deflated verzí sdružených gradientů

Abstract

The conjugate gradient algorithm is one of the most popular methods for the solution of large systems of linear equations with symmetric positive semi-definite matrix. One of the schemes accelerating the convergence of conjugate gradients is deflation which effectively hides parts of the matrix spectrum that slows down the convergence. This master's thesis deals with efficient parallel implementation of the deflated conjugate gradient method with various modifications. Detailed theoretical considerations and the crucial choice of the deflation space are also discussed. The implementation is showcased on a wide range of benchmarks