The paper deals with fast solution of large saddle-point systems arising in wavelet-Galerkin discretizations of separable elliptic PDEs. The periodized orthonormal compactly supported wavelets of the tensor product type together with the fictitious domain method are used. A special structure of matrices makes possible to use the fast Fourier transform that determines the complexity of the algorithm. Numerical experiments confirm theoretical results.
@article{702785,
title = {Complexity and memory requirements of an algorithm for solving saddle-point linear systems with singular blocks},
booktitle = {Programs and Algorithms of Numerical Mathematics},
series = {GDML\_Books},
publisher = {Institute of Mathematics AS CR},
address = {Prague},
year = {2004},
pages = {131-135},
url = {http://dml.mathdoc.fr/item/702785}
}
Kučera, Radek. Complexity and memory requirements of an algorithm for solving saddle-point linear systems with singular blocks, dans Programs and Algorithms of Numerical Mathematics, GDML_Books, (2004), pp. 131-135. http://gdmltest.u-ga.fr/item/702785/