The Fem For Elliptic Problems

N.
Zabaras 1
THE FEM FOR ELLIPTIC PROBLEMS
Abstract formulation: A unied treatment
The variational (V
h
) and minimization (M
h
) problems
Error estimates
The energy norm
Applications to various elliptic boundary value problems
Supplemental mathematical background
1. Sobolev spaces H
k
(), k = 1, 2, . . .
2. The Poincare inequality
Lecture 2: A unied abstract treatment of the FEM for elliptic problems Materials Process Design and Control Laboratory
N. Zabaras 2
ABSTRACT FORMULATION OF THE FEM FOR ELLIPTIC
PROBLEMS
Let V be a Hilbert space with scalar product (., .)
V
and corresponding
norm | |
V
.
a(., .) is a bilinear form on V V and L a linear form on V such that:
1. a(., .) is symmetric
2. a(., .) is continuous, i.e., there is constant > 0 such that
[a(v, w)[ |v|
V
|w|
V
, v, w V
3. a(., .) is V -elliptic, i.e., there is a constant > 0 such that
a(v, v) |v|
2
V
, v V
4. L is continuous, i.e., there is a constant > 0 such that
[L(v)[ |v|
V
, v V
N. Zabaras 3
The abstract minimization (M) and variational (V ) problems
The minimization problem (M):
Find u V such that F(u) = min
vV
F(v),
where
F(v) =
1
2
a(v, v) L(v)
Variational problem (V ):
Find u V such that a(u, v) = L(v) v V .
Theorem
The problems (M) and (V ) are equivalent. There
exists a unique solution u V of these problems
and the following stability estimate holds:
|u|
V

Stability estimate
N. Zabaras 4
Proof of the theorem
1. The existence of a solution follows from the Lax-Milgram theorem.
2. If u V is a solution of (M), then u is a solution of (V ).
Let v V and 1 be arbitrary. Then (u + v) V .
Since u is a minimum, F(u) F(u + v) 1
Let g() F(u + v), 1. Then g(0) g(), 1.
Hence g
/
(0) = 0 (if g
/
(0) exists). Using the symmetry of a(., .):
g() =
1
2
a(u + v, u + v) L(u + v)
=
1
2
a(u, u) +

2
a(u, v) +

2
a(v, u) +

2
2
a(v, v) L(u) L(v)
=
1
2
a(u, u) L(u) + a(u, v) L(v) +

2
2
a (v, v)
g
/
(0) = 0 g
/
(0) = a(u, v) L(v) = 0 u is a solution of (V ).
N. Zabaras 5
3. If u V is a solution of (V ), then u is a solution of (M).
Let v V and 1 be arbitrary. Then (u + v) V with
a(u, v) = L(v), v V .
Using the earlier given expression for g() and the V -elliptic condition
for a(., .), we can write the following:
F(u + v) F(u) = (a(u, v) L(v)) +

2
2
a(v, v)
=

2
2
a(v, v)

2
2
|v|
V
0
where we used a(u, v) = L(v).
It follows that u is a solution of (M).
N. Zabaras 6
4. Stability condition
To prove the stability condition, we use the continuity of L and ellip-
ticity of a(., .):
|u|
2
V
a(u, u) = L(u) |u|
V
For |u|
V
,= 0, we then conclude:
|u|
V

N. Zabaras 7
5. Uniqueness of solution of (M) and (V )
Let u
1
and u
2
be two solutions of (V ), then:
a(u
i
, v) = L(v) v V, i = 1, 2
By subtraction, we see that: a(u
1
u
2
, v) = 0, v V
Applying the stability condition for the case of u = u
1
u
2
,
L = 0 (i.e. = 0), we obtain:
|u
1
u
2
|
V
0 u
1
= u
2
Important note
Even without the symmetry condition a(u, v) = a(v, u), there
exists a unique u V such that
a(u, v) = L(v) v V (V )
and the stability estimate |u|
V

holds. However, in this case,

we cannot dene an associated minimization problem (M).
N. Zabaras 8
Discrete problems (M
h
) and (V
h
)
Let V
h
be a nite-dimensional subspace of V of dimension M. Let
1
, . . . ,
M
be a basis for V
h
so that
i
V
h
.
Any v V
h
has the unique representation
v =
M
i=1
i
, where
i
1
We can now formulate the following discrete analogues of the prob-
lems (M) and (V ):
Problem (M
h
): Find u
h
V
h
such that
F(u
h
) F(v) v V
h
Discrete optimization problem
Problem (V
h
): Find u
h
V
h
such that
a(u
h
,
j
) = L(
j
), j = 1, . . . , M. Discrete variational problem
N. Zabaras 9
Matrix form of (V
h
)
Let u
h
=
M
i=1
i
,
i
1. (V
h
) takes the matrix form A = b:
M
i=1
a(
i
,
j
)
i
= L(
j
), i, j = 1, . . . , M, with A
ji
= a(
i
,
j
), b
j
= L(
j
)
a(v, v) = a
_
_
_
M
i=1
i
,
M
j=1
j
_
_
_
=
M
i,j=1
i
a (
i
,
j
)
j
= A
L(v) = L
_
_
_
M
j=1
j
_
_
_
=
M
j=1
j
L(
j
) = b , v V
h
,
where is the dot product in 1
M
.
Problem (M
h
):
1
2
A b = min
1
M
_
1
2
A b
_
Properties of matrix A:
A = a(v, v) |v|
2
V
> 0 (for v ,= 0) A > 0, for ,= 0
A is positive denite.
a(
i
,
j
) = a(
j
,
i
) A is symmetric.
There is a unique solution of the system A = b.
N. Zabaras 10
Stability condition for the solution of problems (V
h
) and (M
h
)
There exists a unique solution u
h
V
h
to the equivalent problems
(V
h
) and (M
h
). Further, the following stability estimate holds:
|u
h
|
V

The stability estimate follows by choosing v = u

h
in
a(u
h
, v
h
) = L(v
h
) and using the V -ellipticity condition for
a(., .) and the L-continuity condition, i.e.
|u
h
|
2
V
a(u
h
, u
h
) = L(u
h
) |u
h
|
V
From the above equation for |u
h
|
V
,= 0, the stability condition
is derived.
The above stability estimate can be viewed as the theoretical
basis for the success of the FE method
N. Zabaras 11
Error estimate
Theorem
Let u V be the solution of (V ) and u
h
V
h
that of (V
h
) where V
h
V .
Then:
|u u
h
|
V

|u v|
V
v V
h
Error estimate
Recall that:
appears in the a(., .)-continuity condition:
[a(v, w)[ |v|
V
|w|
V
, v, w V
appears in the ellipticity condition for a(., .):
a(v, v) |v|
2
V
, v V
N. Zabaras 12
Proof of the error estimate
We can easily show that: a(u u
h
, w) = 0 w V
h
.
For an arbitrary v V
h
, dene w = u
h
v. Using the V ellipticity and
continuity conditions of a(., .) and that a(u u
h
, w) = 0, we obtain:
|u u
h
|
2
V
ellipticity
..
a(u u
h
, u u
h
) +
= 0
..
a(u u
h
, w)
= a(u u
h
, u u
h
+ w) = a(u u
h
, u v)
continuity
..
|u u
h
|
V
|u v|
V
Division by |u u
h
|
V
,= 0 provides the required error estimate.
Can choose v =
h
u where
h
u V
h
is a suitable interpolant of u. We
would then like to calculate the interpolation error |u
h
u|
V
.
The FEM error estimate can now take the following form:
|u u
h
|
V

|u
h
u|
V
Error estimate
N. Zabaras 13
The energy norm: |v|
a
=
_
a(v, v), v V
Dene the energy norm |.|
a
as follows: |v|
2
a
= a(v, v), v V .
|.|
a
is equivalent to |.|
V
:
c|v|
V
|v|
a
C|v|
V
, v V , with c =
and C =
The corresponding inner product is dened as: (v, w)

a
= a(v, w).
We showed that u
h
is the projection of u onto V
h
, i.e.:
(u u
h
, w)
a
= 0 w V
h
Using (uu
h
, w)
a
= 0, w V
h
and the Cauchy inequality, we obtain:
|u u
h
|
2
a
= (u u
h
, u u
h
)
a
= (u u
h
, u u
h
)
a
+ (u u
h
, w)
a
= (u u
h
, u u
h
+ w)
a
= (u u
h
, u v)
a
|u u
h
|
a
|u v|
a
|u u
h
|
a
|u v|
a
, v V
h
i.e. u
h
is a best approximation of u in the energy norm.
N. Zabaras 14
Example 1: u + u = f in ,
u
n
= 0 on and f L
2
()
V = H
1
() and the following variational problem is obtained:
a(v, w) =
_
[v w + vw] dx
L(v) =
_
fvdx
_
_
a(., .) is a symmetric bilinear form on V V .
L is a linear form.
a(v, v) = |v|
2
H
1
()
a(., .) is V elliptic and using Cauchys in-
equality we obtain:
a(v, w) a(v, v)
1
2
a(w, w)
1
2
= |v|
H
1
()
|w|
H
1
()
a is continuous
The continuity of L is shown using the Cauchy inequality in L
2
:
[L(v)[
fvdx
|f|
L
2
()
|v|
L
2
()
|f|
L
2
()
|v|
H
1
()
N. Zabaras 15
Example 2: u
//
= f on I = (0, 1), u(0) = u(1) = 0 and f L
2
(I)
V = H
1
0
(I) and the following variational problem is obtained:
a(v, w) =
_
I
v
/
w
/
dx, L(v) =
_
I
fvdx
a(., .) is obviously symmetric and bilinear and L is linear. The continuity
of L is derived as in Example 1.
The continuity of a(., .) is shown as follows:
[a(v, w)[ |v
/
|
L
2
(I)
|w
/
|
L
2
(I)
|v|
H
1
0
(I)
|w|
H
1
0
(I)
The V elliptic condition for a(., .) can be shown using the fact that
_
I
v
2
dx
_
I
(v
/
)
2
dx v H
1
0
(I)
a(v, v) =
_
I
(v
/
)
2
dx
1
2
_
_
I
v
2
dx +
_
I
(v
/
)
2
dx
_
=
1
2
|v|
H
1
0
(I)
, v H
1
0
(I).
If V
h
consists of piecewise linear functions on I and u is smooth enough:
|u u
h
|
H
1
()
Ch
N. Zabaras 16
Proof of
_
I
v
2
dx
_
I
(v
/
)
2
dx, v H
1
0
(I)
Using v(0) = 0, we can write: v(x) = v(0) +
x
_
0
v
/
(y)dy =
x
_
0
v
/
(y)dy
Using Cauchys inequality:
[v(x)[ = [
x
_
0
v
/
(y)dy[
1
_
0
[v
/
(y)[dy
(
1
_
0
1
2
dy)
1
2
(
1
_
0
(v
/
(y))
2
dy)
1
2
= (
1
_
0
(v
/
(y))
2
dy)
1
2
Taking the square and then integrating the last equation from 0 to 1:
1
_
0
v
2
(x)dx
1
_
0
(v
/
(y))
2
dy
The above proof needs the boundary condition v(0) = 0 (e.g. take v = 1
to see that the inequality does not stand!).
In order to control the norm of the function v by the norm of the derivative
v
/
we need a xed point to start from.
N. Zabaras 17
Example 3: u = f in 1
2
, u = 0 on , with f L
2
()
Here, V = H
1
0
(), with a(v, w) =
_
v wdx, L(v) =
_
fvdx
a(., .) is symmetric & bilinear and L is linear. For the continuity of a:
[a(v, w)[ |[v[|
L
2
()
|[w[|
L
2
()
|v|
H
1
0
()
|w|
H
1
0
()
The V ellipticity can be shown using Poincares inequality:
_
v
2
dx c
_
[v[
2
dx v H
1
0
()
where c is a constant. Indeed, we can write:
a(v, v)
_
[v[
2
dx
1
c + 1
__
_
v
2
+ [v[
2
_
dx
_

1
c + 1
|v|
2
H
1
()
For piecewise continuous functions in and u suciently smooth:
|u u
h
|
H
1
()
Ch
N. Zabaras 18
Example 4:
d
4
u
dx
4
= f, x I = (0, 1),
u(0) = u
/
(0) = u(1) = u
/
(1) = 0 with f L
2
(I)
Introduce: H
2
(I) = v L
2
(I) : v
/
, v
//
L
2
(I). Then V = H
2
0
(I),
where:
H
2
0
(I) =
_
v H
2
(I) : v(0) = v
/
(0) = v(1) = v
/
(1) = 0
_
Find u V such that a(u, v) = L(v) v V , where
a(v, w) =
_
I
v
//
w
//
dx, L(v) =
_
I
fvdx
The symmetry of a(., .) and continuity of a(., .) and L are easy to show.
Using twice the Poincare inequality and v(0) = v
/
(0) = 0, we can write:
_
I
v
2
dx
_
I
(v
/
)
2
dx
_
I
(v
//
)
2
dx, v H
2
0
(I)
|v|
2
H
2
(I)
=
_
I
v
2
+ (v
/
)
2
+ (v
//
)
2
dx 3
_
I
(v
//
)
2
dx 3a(v, v)
(ellipticity condition for a with = 1/3)
N. Zabaras 19
Sobolev spaces H
k
(), k = 1, 2, . . .
Let D
v =

[[
v
x
1
1
x
2
2
, [[ =
1
+
2
,
i
non-negative natural number
We now dene for k = 1, 2, . . . ,
H
k
() = v L
2
() : D
v L
2
(), [[ k Sobolev space
with norm
|v|
H
k
()
=
_
_
_
_
[[k
_
[D
v[
2
dx
_
_
_
_
1
2
Sobolev space norm
The space H
k
() consists of all functions v on that, together with
the partial derivatives D
v of order [[ at most k, belong to L

2
().
The space H
k
() is a Hilbert space.
N. Zabaras 20
Example 5: The biharmonic problem
u = f in , u =
u
n
= 0 on
Physical models: Clamped plate under transverse load,
Stokes equations in uid mechanics, etc.
Here V = H
2
0
() =
_
v H
2
() : v =
v
n
= 0 on
_
.
Use:
_
(u) vdx =
_
u
n
vds
_
u
v
n
ds+
_
uvdx =
_
uvdx.
(V ): Find u V such that: a(u, v) = L(v), with:
a(u, v) =
_
uvdx and L(v) =

_
fvdx
Easy to show the symmetry of a(., .) and continuity of a(., .) and L.
To show the V -ellipticity of a(., .) need to rst show the Poincare inequality:
|v|
2
H
2
()
c
_
(v)
2
dx, v H
2
0
(), for some constant c
(see derivation at the end of the lecture).
N. Zabaras 21
Example 6: A convection-diusion problem
u +
1
u
x
1
+
2
u
x
2
+ u = f in , u = 0 on , , constants with > 0
Convection in the direction = (
1
,
2
). Assume [[/ is small. We next
simplify by taking = 1.
v V = H
1
0
() and (V ): Find u V : a(u, v) = L(v) v V , where:
a(v, w) =
_
v w +
_
1
v
x
1
+
2
v
x
2
+ v
_
wdx, L(v) =
_
fvdx
Note that:
_
1
v
x
1
+
2
v
x
2
v dx =
_
v
2
(
1
n
1
+
2
n
2
ds
_
1
v
x
1
+
2
v
x
2
v dx
_
1
v
x
1
+
2
v
x
2
v dx = 0.
Thus: a(v, v) =
_
_
[v[
2
+ v
2
_
dx = |v|
2
H
1
()
a(., .) is V -elliptic.
The stiness matrix A is not symmetric. As noted earlier (p. 7), there still
exists a unique solution u
h
V
h
to (V
h
): a(u
h
, v) = L(v) v V
h
.
The following error estimate is obtained ( = 1):
|u u
h
|
H
1
()
|u v|
H
1
()
v V
h
N. Zabaras 22
Example 7: A heat conduction problem:
q
i
= k
i
(x)
u
x
i
in (no-sum on i)
with div q = f in , u = 0 on
1
, q n = g on
2
With the k
i
non-constant, this is an example of a partial dif-
ferential equation with variable coecients.
Here V =
_
v H
1
() : v = 0 on
1
_
with
a(u, v) = L(v), v V , where:
a(v, w) =
_
i=1
k
i
(x)
v
x
i
w
x
i
dx,
L(v) =
_
fvdx +
_
2
gvds.
The general conditions are satised under the following hy-
pothesis: There are positive constants c and C such that
c k
i
(x) C, x , i = 1, 2, 3,
f L
2
(), g L
2
(
2
) and the area of
1
is positive.
N. Zabaras 23
The Poincare inequality:
_
v
2
dx c()
_
[v[
2
dx, v H
1
0
()
Let be an open bounded region. The region

= + can be enclosed
in a square region K: K = (x
1
, x
2
) : x
1
[0, a], x
2
[0, a].
v(x) is extended onto K be setting v = 0 on K . Hence, v is
continuously dierentiable in K and vanishes on the boundary of K.
Using: v(x
1
, x
2
) =
x
1
_
0
v
(, x
2
)d and the Cauchy inequality, we obtain:
[v(x
1
, x
2
)[
2
= [
x
1
_
0
v
(, x
2
)d[
2
((
x
1
_
0
1
2
d)
1/2
(
x
1
_
0
[
v
[
2
d)
1/2
)
2
x
1
x
1
_
0
[
v
[
2
d a
a
_
0
[
v
[
2
d
N. Zabaras 24
The Poincare inequality
Integrating with respect to x
1
and x
2
over K, we obtain:
a
_
0
a
_
0
[v(x
1
, x
2
)[
2
dx
1
dx
2
a
a
_
0
a
_
0
(
a
_
0
[
v(, x
2
)
[
2
d)dx
1
dx
2
= a
2
a
_
0
a
_
0
[
v
[
2
ddx
2
= a
2
_
K
[
v
x
1
[
2
dx
1
dx
2
Similarly we obtain:
a
_
0
a
_
0
[v(x
1
, x
2
)[
2
dx
1
dx
2
a
2
_
K
[
v
x
2
[
2
dx
1
dx
2
.
From the above two inequalities and using v = 0 on K , we obtain:
_
[v(x
1
, x
2
)[
2
dx
1
dx
2

a
2
2
_
([
v
x
1
[
2
+ [
v
x
2
[
2
)dx
1
dx
2
We conclude that:
_
v
2
dx c()
_
[v[
2
dx, v H
1
0
(), where
c() > 0 is a constant that depends on (c dictates the value of a only).
N. Zabaras 25
The Poincare inequality: |v|
2
H
2
()
C()
_
(v)
2
dx, v H
2
0
()
From v = 0 on
v
s
on . With
v
n
= 0 on
v
x
1
=
v
x
2
= 0 on .
Using these boundary conditions and integrating by parts rst in x
1
and
then in x
2
, one can easily show that:
_
2
v
x
2
1
2
v
x
2
2
dx
1
dx
2
=
_
(

2
v
x
1
x
2
)
2
dx
1
dx
2
, v H
2
0
() (a)
Using the earlier Poincare inequality for functions in H
1
0
(), we obtain:
_
(
v
x
1
)
2
dx
1
dx
2
c()
_
2
v
x
2
1
)
2
+ (

2
v
x
1
x
2
)
2
dx
1
dx
2
(b)
_
(
v
x
2
)
2
dx
1
dx
2
c()
_
2
v
x
2
2
)
2
+ (

2
v
x
1
x
2
)
2
dx
1
dx
2
(c)
_
v
2
dx
1
dx
2
c()
_
(
v
x
1
)
2
+ (
v
x
2
)
2
dx
1
dx
2
(d)
From (b)+(c) and (a):
_
[v[
2
dx
1
dx
2
c()
_
(v)
2
dx
1
dx
2
(e)
From (d) and (e):
_
v
2
dx
1
dx
2
c
2
()
_
(v)
2
dx
1
dx
2
(f)
|v|
2
H
2
()
=
_
v
2
+ (v)
2
+ (
2
v
x
2
1
)
2
+ (
2
v
x
2
2
)
2
+ 2(

2
v
x
1
x
2
)
2
dx
1
dx
2

|v|
2
H
2
()
=
_
v
2
+(v)
2
+(v)
2
dx
1
dx
2
(c
2
+ c + 1)
. .
C
_
(v)
2
dx
1
dx
2
N. Zabaras 26
The semi-norm [v[
1
: [v[
1
=
_
[[
v
x
1
[
2
+ [
v
x
1
[
2
]dx
1
dx
2
c
2
|v|
1
[v[
1
c
1
|v|
1
, v H
1
0
()
Based on the rst Poincare inequality, we can write that:
_
v
2
dx
1
dx
2
C
2
()
_
([
v
x
1
[
2
+ [
v
x
2
[
2
)dx
1
dx
2
Dene c
2
2
=
1
2
min1,
1
C
2
.
We thus can conclude:
c
2
2
_
[v
2
dx + [
v
x
1
[
2
+ [
v
x
1
[
2
]dx
1
dx
2

_
([
v
x
1
[
2
+ [
v
x
2
[
2
)dx
1
dx
2
It is also clear that (e.g. for c
1
= 1):
[v[
1
c
1
|v|
1
, v H
1
()
We can now conclude that:
c
2
|v|
1
[v[
1
c
1
|v|
1
, v H
1
0
()
For v H
1
0
(), the seminorm [v[
1
is thus equivalent to the norm |v|
1
.

The Fem For Elliptic Problems

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

The Fem For Elliptic Problems

Hochgeladen von

Copyright:

Verfügbare Formate

N.

holds. However, in this case,

The stability estimate follows by choosing v = u

The corresponding inner product is dened as: (v, w)

v of order [[ at most k, belong to L

uvdx and L(v) =

Das könnte Ihnen auch gefallen