Application of Spectral Element Method Based on Convolution Filtering in Long-Term Wavefield Modeling

Ren JunSheng,Zhang Huai,Zhou YuanZe,Zhang Zhen,Shi YaoLin

DOI: https://doi.org/10.6038/cjg2022q0160

2024-01-01

Abstract:With the continuous development of high-performance computing, seismic wavefield numerical modeling has increasingly become an essential means to understand complex underground media deeply. As a hybrid method, the spectral element method (SEM) considers the traditional advantages of the finite element. The method improves the numerical accuracy and stability, and is one of the most commonly used method in the numerical modeling of the seismic wavefield. With the popularization and development of ultra-large-scale parallel computing, the grid scale of seismic wavefield numerical modeling has gradually increased. The grid size has become smaller and smaller. The design of numerical scheme for the stability of SEM in long- term seismic wave modeling has been a continuous research hotspot. Unlike the traditional structure preservation algorithm, we use the time dispersion transformation method to compensate for the error caused by time-discrete and introduce a filter to break the limitation of Courant-Friedrichs-Lewy (CFL) conditions on time sampling intervals. To overcome the computational problem caused by the non-uniform spatial sampling points of the original filter method in the SEM, this paper replaces the traditional FFT ( Fast Fourier Transform) filtering with spatial convolutional filtering, which is also suitable for the simulation of complex terrain. Finally, this method can also be used in the structure preservation algorithm (such as the 4th-order symplectic Nystrom method). It is found that the time sampling step of the 4th-order symplectic Nystrom method can be further increased. The new method can be effectively used in the simulation of the Earth's free oscillations and other work that requires long-term numerical simulations.