简介:Weintroducemultilevelaugmentationmethodsforsolvingoperatorequationsbasedondirectsumdecompositionsoftherangespaceoftheoperatorandthesolutionspaceoftheoperatorequationandamatrixsplittingscheme.Weestablishageneralsettingfortheanalysisofthesemethods,showingthatthemethodsyieldapproximatesolutionsofthesameconvergenceorderasthebestapproximationfromthesubspace.Theseaugmentationmethodsallowustodevelopfast,accurateandstablenonconventionalnumericalalgorithmsforsolvingoperatorequations.Inparticular,forsecondkindequations,specialsplittingtechniquesareproposedtodevelopsuchalgorithms.Thesealgorithmsarethenappliedtosolvethelinearsystemsresultingfrommatrixcompressionschemesusingwavelet-likefunctionsforsolvingFredholmintegralequationsofthesecondkind.Forthisspecialcase,acompleteanalysisforcomputationalcomplexityandconvergenceorderispresented.Numericalexamplesareincludedtodemonstratetheefficiencyandaccuracyofthemethods.IntheseexamplesweusetheproposedaugmentationmethodtosolvelargescalelinearsystemsresultingfromtherecentlydevelopedwaveletGalerkinmethodsandfastcollocationmethodsappliedtointegralequationsofthesecondkind.Ournumericalresultsconfirmthatthisaugmentationmethodisparticularlyefficientforsolvinglargescalelinearsystemsinducedfromwaveletcompressionschemes.