The traditional methods for numerically solving PDE include local methods(suchas finite difference or finite element method) and global methods(such as spectrummethod). Recently, Wei and Hoffman et.al. have proposed a new method called Dis-crete Singular Convolution method, which both has the high accuracy of global meth-ods and ?exibility of local methods. Different DSC kernels and relative methods havebeen successfully used in many realms such as hydrodynamics and beam analysis.DSCkernel can be regard as a kind of derivative operator but it also has strong tie to wavelettheory.However, DSC algorithm also encounters difficulties such as the oscillation ofthe numerical solution when used in solving non-linear PDE, especially in the hydro-dynamics.Therefore, looking for some robust and reliable methods which can controlthose oscillations by wavelet theory is an important and compelling task.In this thesis, a new DSC kernel with the properties of interpolation,symmetryand compact support is constructed based on the Daubechies wavelet function. More-over, The oscillatory errors produced in the process of DSC method have been ana-lyzed by the convolution properties of distributions and by the context of the theoryof wavelet.Ultimately, a new scheme called DSC-Wavelet algorithm is proposed. Thisnon-oscillation scheme can simply, self-adaptively and accurately solve all kinds ofnon-linear PDEs and efficiently reduce the Gibbs oscillation. At the end of this the-sis,Burger's equation, one kind of Convection-Dominated Diffusion Equations withextremely large values of Reynolds number has been used to test the effect of the newscheme.
|