没有合适的资源?快使用搜索试试~ 我知道了~
Discontinuous Galerkin Method: Chi-Wang Shu
需积分: 18 34 下载量 18 浏览量
2009-08-18
09:58:24
上传
评论 2
收藏 189KB PDF 举报
温馨提示
试读
36页
间断Galerkin方法所求解守恒律方程的有效方法,欢迎大家前来下载
资源详情
资源评论
资源推荐
Superconvergence of Discontinuous Galerkin and Local Discontinuous Galerkin
Schemes for Linear Hyperbolic and Convection Diffusion Equations in One
Space Dimension
Yingda Cheng
1
Department of Mathematics and ICES, University of Texas, Austin, TX 78712
and
Chi-Wang Shu
2
Division of Applied Mathematics, Brown University, Providence, RI 02912
Abstract
In this paper, we study the superconvergence property for the discontinuous Galerkin
(DG) and the local discontinuous Galerkin (LDG) methods, for solving one-dimensional
time dependent linear conservation laws and convection-diffusion equations. We prove su-
perconvergence towards a particular projection of the exact solution when the upwind flux
is used for conservation laws and when the alternating flux is used for convection-diffusion
equations. The order of superconvergence for both cases is proved to be k+
3
2
when piecewise
P
k
polynomials with k ≥ 1 are used. The proof is valid for arbitrary non-uniform regular
meshes and for piecewise P
k
polynomials with arbitrary k ≥ 1, improving upon the results
in [8, 9] in which the proof based on Fourier analysis was given only for uniform meshes and
piecewise P
1
polynomials.
Keywords: discontinuous Galerkin method; local discontinuous Galerkin method; su-
perconvergence; upwind flux; projection; error estimates.
1
E-mail: ycheng@math.utexas.edu
2
E-mail: shu@dam.brown.edu. Research supported by NSF grant DMS-0809086 and DOE grant DE-
FG02-08ER25863.
1
1 Introduction
In this paper, we consider one-dimensional linear hyperbolic conservation laws
u
t
+ cu
x
= 0, (1.1)
and convection-diffusion equations
u
t
+ cu
x
= bu
xx
, (1.2)
where c, b are constants and b > 0. We study the the superconvergence of the discontinuous
Galerkin (DG) solutions and the local DG (LDG) solutions towards a particular projection
of the exact solution. Superconvergence requires upwind fluxes for the DG scheme and
alternating fluxes for the LDG scheme. This superconvergence also implies a good control
on the time evolution of the errors.
The DG method discussed here is a class of finite element methods using completely
discontinuous piecewise polynomial space for the numerical solution and the test functions.
It is originally devised to solve hyperbolic conservation laws containing only first order spatial
derivatives, e.g. [14, 13, 12, 11, 15, 17]. It has the advantage of flexibility for arbitrarily
unstructured meshes, with a compact stencil, and with the ability to easily accommodate
arbitrary h-p adaptivity. The DG method was later generalized to the LDG method by
Cockburn and Shu to solve the convection-diffusion equation [16]. Their work was motivated
by the successful numerical experiments of Bassi and Rebay [5] for the compressible Navier-
Stokes equations.
For ordinary differential equations, Adjerid et al. [1, 4] proved the DG solution is super-
convergent at Radau points. In [8], we proved superconvergence of the DG solution towards
a particular projection of the exact solution in the case of piecewise linear polynomials on
uniform meshes for the linear conservation law (1.1) and considered its impact on the time
growth of the errors. We also demonstrated numerically that the conclusions hold true for
very general cases, including higher order DG methods, nonlinear equations, systems, and
2
two dimensions. For convection-diffusion equations, in [7], Celiker and Cockburn studied the
steady state solution of (1.2), and proved that for a large class of DG methods, the numerical
fluxes (traces) are superconvergent, also see [6] for related discussions on elliptic problems.
In [3, 2], Adjerid et al. showed for convection or diffusion dominant time dependent equa-
tions, the LDG solution will be superconvergent at Radau points. In [9], we discussed the
superconvergence property of the LDG scheme for convection-diffusion equations. We proved
the superconvergence result for the heat equation in the case of piecewise linear solutions
on uniform meshes, and gave numerical tests to demonstrate the validity of the result for
higher order schemes and nonlinear equations.
The proof in [8, 9] uses Fourier analysis and works only for piecewise linear approximation
space and uniform meshes. In this paper, we use a different framework to prove the super-
convergence results and do not rely on Fourier analysis. The proof now works for arbitrary
non-uniform regular meshes and schemes of any order.
Even though the proof in this paper is given for the simple scalar equations (1.1) and
(1.2), the same superconvergence results can be easily proved for one-dimensional linear
systems along the same lines. The generalization to two space dimensions is more involved,
see [8] for some discussion.
This paper is organized as follows: in Section 2, we consider the superconvergence of
the DG method for the linear conservation law (1.1). We prove our main superconvergence
result in Theorem 2.2. In Section 3, we prove the superconvergence of the LDG method for
the linear convection-diffusion equation (1.2), and discuss the effect of fluxes on superconver-
gence. Finally, conclusions and plans for future work are provided in Section 4. The proofs
for some of the technical lemmas are collected in the Appendix.
3
2 Conservation laws
In this section, we consider, without loss of generality, the linear conservation law (1.1) with
c = 1:
u
t
+ u
x
= 0
u(x, 0) = u
0
(x)
u(0, t) = u(2π, t)
. (2.1)
Here, u
0
(x) is a smooth 2π-periodic function. We consider only periodic boundary conditions
in this paper for simplicity. Since we do not use Fourier analysis, the assumption of periodic
boundary condition is not essential.
The usual notation of the DG method is adopted. If we want to solve this equation on
the interval I = [a, b], first we divide it into N cells as follows
a = x
1
2
< x
3
2
< . . . < x
N+
1
2
= b. (2.2)
We denote
I
j
= (x
j−
1
2
, x
j+
1
2
), x
j
=
1
2
x
j−
1
2
+ x
j+
1
2
, (2.3)
as the cells and cell centers respectively. h
j
= x
j+
1
2
− x
j−
1
2
denotes length of each cell. We
denote h = max
j
h
j
as length of the largest cell.
Define V
k
h
= {υ : υ|
I
j
∈ P
k
(I
j
), j = 1, ··· , N} to be the approximation space, where
P
k
(I
j
) denotes all polynomials of degree at most k on I
j
. The DG scheme using the upwind
flux will become: find u
h
∈ V
k
h
, such that
Z
I
j
(u
h
)
t
v
h
dx −
Z
I
j
u
h
(v
h
)
x
dx + u
−
h
v
−
h
|
j+
1
2
− u
−
h
v
+
h
|
j−
1
2
= 0 (2.4)
holds for any v
h
∈ V
k
h
. Here and below (v
h
)
−
j+
1
2
= v
h
(x
−
j+
1
2
) denotes the left limit of the
function v
h
at the discontinuity point x
j+
1
2
. Likewise for v
+
h
.
In addition, if k ≥ 1, we can define P
−
h
u to be a projection of u into V
k
h
, such that
Z
I
j
P
−
h
u v
h
dx =
Z
I
j
u v
h
dx (2.5)
for any v
h
∈ P
k−1
on I
j
, where k is the polynomial degree of the DG solution, and
(P
−
h
u)
−
= u
−
at x
j+1/2
. (2.6)
4
Notice that this special projection is used in the error estimates of the DG methods to derive
optimal L
2
error bounds in the literature, e.g. in [18]. We are going to show that indeed
the numerical solution is closer to this special projection of the exact solution than to the
exact solution itself, extending the results in [8]. Let us denote e = u − u
h
to be the error
between the exact solution and numerical solution, ε = u − P
−
h
u to be the projection error,
and ¯e = P
−
h
u − u
h
to be the error between the numerical solution and the projection of the
exact solution.
We introduce two functionals which are essential to our estimates. We prove in Lemma
2.1 that they are related to the L
2
norm of a function on I
j
.
B
−
j
(M) =
Z
I
j
M(x)
x − x
j−1/2
h
j
d
dx
M(x)
x − x
j
h
j
dx,
B
+
j
(M) =
Z
I
j
M(x)
x − x
j+1/2
h
j
d
dx
M(x)
x − x
j
h
j
dx.
Lemma 2.1. For any function M(x) ∈ C
1
on I
j
,
B
−
j
(M) =
1
4h
j
Z
I
j
M
2
(x)dx +
M
2
(x
j+1/2
)
4
, (2.7)
B
+
j
(M) = −
1
4h
j
Z
I
j
M
2
(x)dx −
M
2
(x
j−1/2
)
4
. (2.8)
The proof of this lemma is given in the Appendix.
Theorem 2.2. Let u be the exact solution of the equation (2.1). If k ≥ 1, define u
h
to be
the DG solution of (2.4) with the initial condition u
h
(·, 0) = P
−
h
u
0
. We have the following
error estimate:
||¯e(·, t)||
L
2
≤ C
1
(t + 1) h
k+3/2
, (2.9)
and
||e(·, t)||
L
2
≤ C
1
t h
k+3/2
+ C
2
h
k+1
, (2.10)
where C
1
= C
1
(||u||
k+3
), C
2
= C
2
(||u||
k+3
).
5
剩余35页未读,继续阅读
Phaeton45
- 粉丝: 2
- 资源: 5
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功
评论0