-
Navier-Stokes方程是描述流体运动规律的一类典型的非线性方程,其研究对人们认识和控制湍流格外重要,并被广泛应用于天气、海流等方面.近几十年来,用有限元方法数值求解Navier-Stokes方程吸引了许多研究者[1-3].而后处理的思想最初是在文献[4]中基于Fourier谱方法提出的,文献[5]将该方法应用于混合有限元方法提出了后处理的混合有限元方法.该方法主要分为两步:第一步,在粗网格上求解一个非线性系统,得到最后时刻T的有限元解uH;第二步,在最后时刻T,对第一步所得解进行后处理,主要通过在细网格上(或用高阶元)求解一个线性的Stokes问题[6],或一个线性Ossen类型的问题[7],或者一个Newton类型的问题[8].理论分析和数值实验表明:在选择合适的网格尺寸的条件下,后处理的混合有限元方法与标准的有限元方法相比收敛精度提高了一阶.由于完全非线性问题要在粗网格上求解,而对流占优的流体具有不稳定性,所以用后处理的方法模拟大雷诺数流问题具有一定的挑战性.本文中,我们把前面提到的3种后处理的有限元方法和亚格子稳定化方法[9]相结合,提出了3种稳定化的后处理混合有限元方法.
A Subgrid Stabilizing Postprocessed Mixed Finite Element Method for the Navier-Stokes Equations
-
摘要: 提出并考察了3种基于亚格子模型的后处理混合有限元方法,其主要思想是:第一步在粗网格上求解带有亚格子模型稳定项的Navier-Stokes方程,得到最后时刻T的有限元解uH;第二步在最后时刻T,对第一步所得解uH进行后处理,主要通过在细网格上(或用高阶元)分别求解带有亚格子模型稳定项的Stokes问题、Newton问题或者Ossen问题.实验结果表明:在选取适当的稳定化参数和网格尺寸的条件下,3种稳定化的后处理有限元方法提高了稳定化的混合有限元解的精确度,并且收敛阶较标准的有限元方法明显提高了一阶.从计算时间看,除ν = 1以外,在其它情况下稳定化的Newton型后处理花费的时间相对较多,而稳定化的Ossen型后处理花费的时间相对较少.从精确度来看,Newton型后处理和Ossen型后处理方法所得速度的H1-范误差和压力的L2-范误差比Stokes型后处理方法更有效.
-
关键词:
- Navier-Stokes方程 /
- 后处理 /
- 亚格子稳定化 /
- 有限元方法
Abstract: In this paper, we mainly study three postprocessed mixed finite element methods for the incompressible Navier-Stokes equations, which are based on a subgrid model. These methods consist of two steps. The first step is to solve a subgrid stabilized nonlinear Navier-Stokes problem on a coarse grid to obtain an approximate solution uH at time T. The second step is to postprocess uH on a finer grid (or by high-order finite elements), by solving a stabilized Stokes problem, a stabilized Newton-Type problem, or a stabilized Ossen problem. The numerical results show that under the conditions of selecting appropriate stabilizing parameters and grid sizes, the postprocessed finite element method can improve the precision of the mixed finite-element solution, and the order of convergence is obviously improved by one unit compared with the standard subgrid stabilized method. From the point of the computational time, in addition to ν = 1, the stabilized Newton-type postprocessed method takes a relatively more time than the others, while the stabilized Ossen-type postprocessed method takes the least time among the three methods. And from the point of precision of the computed solutions, the Newton and Oseen-type postprocessed methods are better than the Stokes-type postprocessed method.-
Key words:
- Navier-Stokes equations /
- postprocessing /
- subgrid stabilization /
- finite element method .
-
表 1 稳定化的混合有限元解和后处理解的L2范误差
h Δt ${{\left\| \mathit{\boldsymbol{u}}\left(T \right)-{{\mathit{\boldsymbol{u}}}_{N}}^{h} \right\|}_{0}}$ 收敛率1 ${{\left\| \mathit{\boldsymbol{u}}\left(T \right)-{{{\mathit{\boldsymbol{\tilde{u}}}}}_{N}}^{h} \right\|}_{0}}$ 收敛率2 1/3 1/1 000 0.002 623 32 - 0.000 546 192 - 1/4 1/2 000 0.001 144 46 2.883 42 0.000 204 008 3.423 26 1/5 1/4 000 0.000 582 591 3.025 86 8.988 45 e-005 3.673 12 1/6 1/8 000 0.000 332 497 3.076 18 4.551 18 e-005 3.732 71 表 2 稳定化的混合有限元解和后处理解的H1范误差
h Δt ${{\left\| \mathit{\boldsymbol{u}}\left(T \right)-{{\mathit{\boldsymbol{u}}}_{N}}^{h} \right\|}_{0}}$ 收敛率1 ${{\left\| \mathit{\boldsymbol{u}}\left(T \right)-{{{\mathit{\boldsymbol{\tilde{u}}}}}_{N}}^{h} \right\|}_{0}}$ 收敛率2 1/3 1/1 000 0.056 121 6 - 0.011 215 6 - 1/4 1/2 000 0.033 580 3 1.785 24 0.005 077 48 2.754 74 1/5 1/4 000 0.022 172 3 1.860 21 0.002 728 83 2.782 69 1/6 1/8 000 0.015 690 6 1.896 55 0.001 665 23 2.709 01 表 3 稳定化的混合有限元压力解和后处理压力解的误差
h Δt ${{\left\| \mathit{\boldsymbol{P}}\left(T \right)-{{\mathit{\boldsymbol{P}}}_{h}} \right\|}_{0}}$ 收敛率1 ${{\left\| \mathit{\boldsymbol{P}}\left(T \right)-{{{\mathit{\boldsymbol{\tilde{P}}}}}_{h}} \right\|}_{0}}$ 收敛率2 1/3 1/1 000 0.293 433 - 0.012 140 7 - 1/4 1/2 000 0.168 947 1.918 99 0.004 673 56 3.318 38 1/5 1/4 000 0.113 955 1.764 72 0.002 215 05 3.346 04 1/6 1/8 000 0.087 428 8 1.453 35 0.001 225 75 3.245 49 表 4 3种算法所得逼近解的L2范误差
ν 算法1 算法2 算法3 ${{\left\| \mathit{\boldsymbol{u}}\left(T \right)-{{{\mathit{\boldsymbol{\tilde{u}}}}}_{N}}^{h} \right\|}_{0}}$ 计算时间/ s ${{\left\| \mathit{\boldsymbol{u}}\left(T \right)-{{{\mathit{\boldsymbol{\tilde{u}}}}}_{N}}^{h} \right\|}_{0}}$ 计算时间/ s ${{\left\| \mathit{\boldsymbol{u}}\left(T \right)-{{{\mathit{\boldsymbol{\tilde{u}}}}}_{N}}^{h} \right\|}_{0}}$ 计算时间/ s 1 4.170 23 e-008 646.368 3.281 82 e-008 645.385 3.281 8 e-008 642.455 0.1 4.122 22 e-007 6 446.090 3.271 9 e-007 6 475.770 3.271 86 e-007 6 438.950 0.01 4.042 53 e-006 12 240.900 3.263 82 e-006 12 300.300 3.264 11 e-006 12 229.000 0.001 3.439 92 e-005 18 037.300 3.160 34 e-005 18 158.800 3.181 14 e-005 18 019.600 表 5 3种算法所得逼近解的H1范误差
ν 算法1 算法2 算法3 ${{\left\| \mathit{\boldsymbol{u}}\left(T \right)-{{{\mathit{\boldsymbol{\tilde{u}}}}}_{N}}^{h} \right\|}_{1}}$ 计算时间/ s ${{\left\| \mathit{\boldsymbol{u}}\left(T \right)-{{{\mathit{\boldsymbol{\tilde{u}}}}}_{N}}^{h} \right\|}_{1}}$ 计算时间/ s ${{\left\| \mathit{\boldsymbol{u}}\left(T \right)-{{{\mathit{\boldsymbol{\tilde{u}}}}}_{N}}^{h} \right\|}_{1}}$ 计算时间/ s 1 3.181 36 e-005 646.368 3.191 44 e-006 645.385 3.191 44 e-006 642.455 0.1 0.000 317 569 6 446.090 3.190 04 e-005 6 475.770 3.190 04 e-005 6 438.950 0.01 0.003 120 13 12 240.900 0.000 318 435 12 300.300 0.000 318 434 12 229.000 0.001 0.026 555 6 18 037.300 0.003 128 28 18 158.800 0.003 128 08 18 019.600 表 6 3种算法所得逼近解的压力误差
ν 算法1 算法2 算法3 ${{\left\| \mathit{\boldsymbol{P}}\left(T \right)-{{{\mathit{\boldsymbol{\tilde{P}}}}}_{h}} \right\|}_{0}}$ 计算时间/ s ${{\left\| \mathit{\boldsymbol{P}}\left(T \right)-{{{\mathit{\boldsymbol{\tilde{P}}}}}_{h}} \right\|}_{0}}$ 计算时间/ s ${{\left\| \mathit{\boldsymbol{P}}\left(T \right)-{{{\mathit{\boldsymbol{\tilde{P}}}}}_{h}} \right\|}_{0}}$ 计算时间/ s 1 2.533 81 e-005 646.368 2.591 14 e-006 645.385 2.591 99 e-006 642.455 0.1 2.533 85 e-005 6 446.090 2.592 76 e-006 6 475.770 2.588 36 e-006 6 438.950 0.01 2.533 86 e-005 12 240.900 2.643 26 e-006 12 300.300 2.609 82 e-006 12 229.000 0.001 2.533 92 e-005 18 037.300 3.884 7 e-006 18 158.800 3.060 6 e-006 18 019.600 -
[1] TEMAN R. Navier-Stokes Equations:Theory and Numerical Analysis[M]. Amsterdam:North-Holland Publishing, 1977. [2] GIRAULT V, RAVIART P A. Finite Element Methods for Navier-Stokes Equations:Theory and Algorithms[M]. Berlin:Springer-Verlag, 1986. [3] GLOWINSKI R. Finite Element Methods for Incompressible Viscous Flow[M]//Handbook of Numerical Analysis. Amsterdam:Elsevier, 2003. [4] ARCHILLA B G, NOVO J, TITI E S. Postprocessing the Galerkin Method:A Novel Approach to Approximate Inertial Manifolds[J]. SIAM Journal on Numerical Analysis, 1998, 35(3):941-972. doi: 10.1137/S0036142995296096 [5] doi: https://epubs.siam.org/doi/10.1137/S0036142998335893 GARC B. Postprocessing the Galerkin Method:The Finite-Element Case[J]. SIAM Journal on Numerical Analysis, 2000, 37(2):470-499. [6] AYUSO B, ARCHILLA B G, NOVO J. The Postprocessed Mixed Finite-Element Methodfor the Navier-Stokes Equations[J]. SIAM Journal on Numerical Analysis, 2005, 43(3):1091-1111. doi: 10.1137/040602821 [7] FRUTOS J D, ARCHILLA B G, NOVO J. Static Two-Grid Mixed Finite-Element Approximations to the Navier-Stokes Equations[J]. Journal of Scientific Computing, 2012, 52(3):619-637. doi: 10.1007/s10915-011-9562-7 [8] DURANGO F, NOVO J. Two-Grid Mixed Finite-Element Approximations to the Navier-StokesEquations Based on a Newton-Type Step[J]. Journal of Scientific Computing, 2018, 74(1):456-473. doi: 10.1007/s10915-017-0447-2 [9] doi: https://onlinelibrary.wiley.com/doi/abs/10.1002/num.20120#citedBy SHANG Y Q. A Two-Level Subgrid Stabilized Oseen IterativeMethod for the Steady Navier-Stokes Equations[J]. Journal of Computational Physics, 2013,233(1):210-226. [10] doi: http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=10.1515/jnum-2012-0013 HECHT F. New Development in Freefem++[J]. Journal of Numerical Mathematics, 2012, 20(3-4):159-344. [11] doi: http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=d8521578e4067f0b252c5e62ecab0883 SHANG Y Q. Error Analysis of a Fully Discrete Finite Element Variational Multiscale Method for Time-Dependent Incompressible Navier-Stokes Equations[J]. Numerical Methods for Partial Differential Equations, 2013, 29(6):2025-2046.