没有合适的资源?快使用搜索试试~ 我知道了~
具有随机强迫的二维稳态Navier-Stokes系统的侵入和非侵入多项式混沌近似_Intrusive and Non-Intru
1.该资源内容由用户上传,如若侵权请联系客服进行举报
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
版权申诉
0 下载量 138 浏览量
2022-01-22
17:10:20
上传
评论
收藏 262KB PDF 举报
温馨提示
试读
19页
具有随机强迫的二维稳态Navier-Stokes系统的侵入和非侵入多项式混沌近似_Intrusive and Non-Intrusive Polynomial Chaos Approximations for a Two-Dimensional Steady State Navier-Stokes System with Random Forcing.pdf
资源推荐
资源详情
资源评论
arXiv:2107.11720v1 [math.NA] 25 Jul 2021
INTRUSIVE AND NON-INTRUSIVE POLYNOMIAL CHAOS
APPROXIMATIONS FOR A TWO-DIMENSIONAL STEADY STATE
NAVIER-STOKES SYSTEM WI T H RANDOM FORCING
S. V. LOTOTSKY, R. MIKULEVICIUS, AND B. L. ROZOVSKY
Abstract. While convergence of polynomial chaos approximation for linear equa-
tions is relatively well understood, a lot less is known for non-linear equations.
The paper investigates this convergence for a particular equa tion with quadratic
nonlinearity.
July 27, 2021
1. Introduction
There are two main ways to study an equation with random input. One way is to use
deterministic tools for each particular realization of randomness; in what follows, we
call it path-wise approach. An alternative, which we call random field approach,
is to consider the random input as an additional independent variable in the equation,
along with space and/or time.
Questions such as existence/uniqueness/regularity of the solution are often addressed
with a combination of the two approaches; cf. [18, 19] for ordinary differential equa-
tions and [17, 24, 25] for equations with partial derivatives.
The difference between t he two approaches becomes noticeable in numerical compu-
tations; see, for example, [13, 35]. Path-wise approach leads to repeated numerical
solutions of the underlying equation for various realizations of the random input; a
typical example is Monte Carlo simulations. In computational terms, this approach
is non-intrusive, because no new numerical procedures are required to solve the
equation compared to the deterministic case.
The random field approach reduces the problem to a fixed system of deterministic
equations via a stochastic Galerkin approximation; in many cases, the result is a
generalized polynomial chaos (gPC) expansion [35, Chapter 6]. In computational
terms, this appro ach is intrusive, because the resulting system is more complicated
than the original equation and requires different numerical procedures to obtain a
solution.
2020 Mathematics Subject Classification. 35Q30, 35R60, 60H35, 65N15 , 65N30, 65N35, 76D05,
76M22.
Key words and phrases. Gauss Quadrature, Generalized Polynomial Chaos, Stochastic Galerkin
Approximation.
SVL: Research suppor ted by ARO Grants W911N-16-1-0103 and FA9550-21-1-0015.
RM and BLR: Research supported by ARO Grant W911N-16-1-0103.
1
2 S. V. LOTOTSKY, R. MIKULEVICIUS, AND B. L. ROZOVSKY
The stochastic collocation method [35, Chapter 7], with sampling at pre-determined
realizations of the rando m input, somewhat bridges the gap between pure r andom
sampling (Monte Carlo) and complete elimination of randomness (g PC). In com-
putational terms, the method is non-intrusive [3, 31]. In this paper, we consider
the disc rete projection, o r pseudo-spectral version of the method, when t he sampled
solution is used to approximate the coefficients in the chaos expansion via Gauss
quadrature.
For many, although apparently not all [8], equations, various empirical studies
[15, 28, 32, etc.] suggest that the stochastic Galerkin approximation method, with
a fixed computational cost, can be a much more efficient way to study statistical
properties of the solution tha n Monte Carlo or stochastic collocation methods. In the
case of nonlinear equations, this exp erimental success has yet to be fully justified the-
oretically; for linear equations, the picture is rather clear; see, for example, [9, 20, 21]
as well as [22, Chapter 5] a nd [30, Section 8.3].
Accordingly, our objective in this paper is to carry out a comparative theoretical
analysis of an intrusive and a non-intrusive approximations for a particular nonlinear
equation. Specifically, we consider the stationary Navier-Stokes system in a smooth
bounded planar domain with zero boundary conditions and with randomness in the
external force, and we establish a priori error bounds for both approximations.
The paper is org anized as follows. Section 2 describes the model and introduces the
necessary function spaces. Section 3 introduces the stochastic Galerkin approximation
and gives the proof of convergence. Section 4 investigates a non-intrusive pseudo-
spectral approximation. Section 5 puts the results in a broader context.
Throughout the paper, G is a bounded domain in R
2
with area |G| and sufficiently
regular (e.g. locally Lipschtiz) boundary ∂G. We use the following convention with
the notations of various function spaces and their elements: if X denotes a space of
scalar fields f on G, then X denotes the corresponding space of vector fields f, and
X denotes the collection of X-valued random elements f .
2. The Setting
Let (Ω, F, P) be a probability space such that the L
2
(Ω) is a separable Hilbert space
and has a complete orthogonal basis {P
n
, n ≥ 0}: with
c(n) = EP
2
n
, (2.1)
every ζ ∈ L
2
(Ω) can be represented as
ζ =
X
k≥0
E
ζP
k
)
c(k)
P
k
.
In what follows, we always assume that the basis {P
n
, n ≥ 0} has the following
property: for every m, n ≥ 0, there a r e finitely many real numb ers A
m,n;l
, l ≥ 0, such
that
P
m
P
n
=
X
l≥0
A
m,n;l
P
l
; (2.2)
STATIONARY NSE 3
in that case
A
m,n;l
=
E
P
m
P
n
P
l
c(l)
.
Property (2.2) holds when each P
n
is a polynomial or a tensor product of polynomials.
Denote by P
N
the orthogonal proj ection in L
2
(Ω) on the subspace spanned by
{P
k
, k = 0, . . . , N }.
Consider a steady-state Navier-Stokes system with random forcing in a bo unded do-
main G ⊂ R
2
with sufficiently regular boundary ∂G:
ν∆u (x) =
u · ∇)u + ∇p (x) + f (x), x ∈ G, (2.3)
div u (x) = 0, x ∈ G, u|
∂G
= 0.
In equation (2.3),
• ν > 0 is the kinematic viscosity coefficient, x = (x
1
, x
2
), ∆ =
∂
2
∂x
2
1
+
∂
2
∂x
2
2
is the
Laplace operator , and ν is constant;
• u (x) = (u
1
(x) , u
2
(x)) is the (unknown) velocity and
div u = ∇ · u =
∂u
1
∂x
1
+
∂u
2
∂x
2
,
u · ∇)u
i
= u
1
∂u
i
∂x
1
+ u
2
∂u
i
∂x
2
, i = 1, 2; (2.4)
• p = p (x) is the (unknown scalar) pressure and
∇p)
i
=
∂p
∂x
i
, i = 1, 2;
• f is the random forcing.
Two standard references for the deterministic counterpart of (2.3) a re [11, Chapter
IX] and [33, Chapter II].
We will use the following function spaces:
• C
∞
0
(G), the collection o f infinitely differentiable real-valued functions on G
with compact support in G;
• D(G) = {ϕ = (ϕ
1
, ϕ
2
), ϕ
i
∈ C
∞
0
(G) , i = 1 , 2 : div ϕ = 0};
• L
r
(G), 1 ≤ r < +∞, the collection of measurable functions g on G such that
|g|
L
r
=
Z
G
|g(x)|
r
dx
1/r
< ∞;
for g, f ∈ L
2
(G), we write
(f, g)
0
=
Z
G
f(x)g(x) dx;
• L
r
(G), the collection of vector fields g = (g
1
, g
2
) on G such that g
1
, g
2
∈
L
r
(G), and endowed with norm
|g|
L
r
=
g
1
r
L
r
+
g
2
r
L
r
1/r
;
for g, f ∈ L
2
(G), we write
(f, g)
0
=
Z
G
f
1
(x)g
1
(x) + f
2
(x)g
2
(x)
dx;
4 S. V. LOTOTSKY, R. MIKULEVICIUS, AND B. L. ROZOVSKY
• L
2
(G) = L
2
Ω; L
2
(G)
, that is, the collection of L
2
(G)-valued random ele-
ments
g(ω, x) =
g
1
(ω, x) , g
2
(ω, x)
such that
|g|
L
2
=
E |g|
2
L
2
1/2
< ∞;
• H
1,2
0
(G), the completion of C
∞
0
(G) with respect to the no rm
|g|
1,2
=
Z
G
|∇g(x)|
2
dx
1/2
=
Z
G
∂g(x)
∂x
1
2
+
∂g(x)
∂x
2
2
!
dx
!
1/2
;
note that | · |
1,2
is indeed a norm on C
∞
0
(G) because, by a version of t he
Poincar´e inequality, if g ∈ C
∞
0
(G) and |G| is the Lebesgue measure (area) of
G, then (cf. [11, Exercise II.5.4])
|g|
2
L
2
≤
|G|
2
|g|
2
1,2
; (2.5)
• H
1,2
0
(G), the collection of vector fields g = (g
1
, g
2
) on G such that g
1
, g
2
∈
H
1,2
0
(G), and endowed with norm
|g|
1,2
=
g
1
2
1,2
+
g
2
2
1,2
1/2
;
for f, g ∈ H
1,2
0
(G), we write
∇f, ∇g
0
=
2
X
i,j=1
Z
G
∂f
i
(x)
∂x
j
∂g
i
(x)
∂x
j
dx, (2.6)
so that
|g|
2
1,2
=
∇g, ∇g
;
• H
1,2
0
(G) = L
2
Ω; H
1,2
0
(G)
;
•
b
H
1,2
0
(G), the completion of D(G) with respect to t he norm | · |
1,2
;
•
b
H
1,2
0
(G) = L
2
Ω;
b
H
1,2
0
(G)
;
• H
−1,2
0
(G), the completion of L
2
(G) with respect to the no rm
|g|
−1,2
= sup
Z
G
g(x)ϕ(x) dx, ϕ ∈ H
1,2
0
(G) , |ϕ|
1,2
≤ 1,
;
• H
−1,2
0
(G), the collection of vector fields g = (g
1
, g
2
) such that g
1
, g
2
∈
H
−1,2
0
(G) , and endowed with norm
|g|
−1,2
=
g
1
2
−1,2
+
g
2
2
1,2
1/2
;
• H
−1,2
0
(G) = L
2
Ω; H
−1,2
0
(G)
.
The (Ba nach space) dual of H
1,2
0
(G) is isomorphic to H
−1,2
0
(G): see [11, Theorem
II.3.5]. We denote the corr esponding duality by hf, gi
1
, f ∈ H
−1,2
0
(g) , g ∈ H
1,2
0
(G) .
Similarly, the dual of H
1,2
0
(G) is isomorphic to H
−1,2
0
(G) and the duality is denoted
by hf , gi
1
, f ∈ H
−1,2
0
(G) , g ∈ H
1,2
0
(G) .
剩余18页未读,继续阅读
资源评论
易小侠
- 粉丝: 6468
- 资源: 9万+
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