Removing the stability limit of the explicit finite-difference scheme with eigenvalue perturbation

被引:27
作者
Gao, Yingjie [1 ,2 ]
Zhang, Jinhai [1 ,2 ]
Yao, Zhenxing [1 ,2 ]
机构
[1] Chinese Acad Sci, Inst Geol & Geophys, Key Lab Earth & Planetary Phys, Beijing 100029, Peoples R China
[2] Chinese Acad Sci, Inst Earth Sci, Beijing, Peoples R China
基金
中国博士后科学基金; 中国国家自然科学基金;
关键词
STABLE FDTD METHOD; TIME DISPERSION;
D O I
10.1190/GEO2018-0447.1
中图分类号
P3 [地球物理学]; P59 [地球化学];
学科分类号
0708 ; 070902 ;
摘要
The explicit finite-difference scheme is popular for solving the wave equation in the field of seismic exploration due to its simplicity in numerical implementation. However, its maximum time step is strictly restricted by the Courant-Friedrichs-Lewy (CFL) stability limit, which leads to a heavy computational burden in the presence of small-scale structures and high-velocity targets. We remove the CFL stability limit of the explicit finite-difference scheme using the eigenvalue perturbation, which allows us to use a much larger time step beyond the CFL stability limit. For a given time step that is within the CFL stability limit, the eigenvalues of the update matrix would be distributed along the unit circle; otherwise, some eigenvalues would be distributed outside of the unit circle, which introduces unstable phenomena. The eigenvalue perturbation can normalize the unstable eigenvalues and guarantee the stability of the update matrix by using an arbitrary time step. The update matrix can be preprocessed before the numerical simulation, thus retaining the computational efficiency well. We further incorporate the forward time-dispersion transform (FTDT) and the inverse time-dispersion transform (ITDT) to reduce the time-dispersion error caused by using an unusually large time step. Our numerical experiments indicate that the combination of the eigenvalue perturbation, the FTDT method, and the ITDT method can simulate highly accurate waveforms when applying a time step beyond the CFL stability limit. The time step can be extended even toward the Nyquist limit. This means that we could save many iteration steps without suffering from time-dispersion error and stability problems.
引用
收藏
页码:A93 / A98
页数:6
相关论文
共 18 条
[1]   Time step n-tupling for wave equations [J].
Amundsen, Lasse ;
Pedersen, Orjan .
GEOPHYSICS, 2017, 82 (06) :T249-T254
[2]  
[Anonymous], 2014, THESIS
[3]  
Chu C., 2009, 79 ANN INT M SEG, P2662, DOI [10.1190/1.3255400., DOI 10.1190/1.3255400]
[4]   Partial differential equations of mathematical physics [J].
Courant, R ;
Friedrichs, K ;
Lewy, H .
MATHEMATISCHE ANNALEN, 1928, 100 :32-74
[5]  
Etgen J.T., 1994, 64 ANN INT M, P1266
[6]  
Etgen J.T., 2009, SEG Expanded Abstracts, V28, P2552, DOI DOI 10.1190/1.3255375
[7]   Computational methods for large-scale 3D acoustic finite-difference modeling: A tutorial [J].
Etgen, John T. ;
O'Brien, Michael J. .
GEOPHYSICS, 2007, 72 (05) :SM223-SM230
[8]   Alternative Method for Making Explicit FDTD Unconditionally Stable [J].
Gaffar, Md. ;
Jiao, Dan .
IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, 2015, 63 (12) :4215-4224
[9]   An Explicit and Unconditionally Stable FDTD Method for Electromagnetic Analysis [J].
Gaffar, Md. ;
Jiao, Dan .
IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, 2014, 62 (11) :2538-2550
[10]   Third-order symplectic integration method with inverse time dispersion transform for long-term simulation [J].
Gao, Yingjie ;
Zhang, Jinhai ;
Yao, Zhenxing .
JOURNAL OF COMPUTATIONAL PHYSICS, 2016, 314 :436-449