Scribd
Upload
304 views4 pages

Natural Boundary Conditions

In contrast to essential boundary conditions, which are built into the solution space, natural boundary conditions are built into the weak form.

Uploaded by

sobaseki1
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
304 views4 pages

Natural Boundary Conditions

In contrast to essential boundary conditions, which are built into the solution space, natural boundary conditions are built into the weak form.

Uploaded by

sobaseki1
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Natural Boundary Conditions

In contrast to essential boundary conditions, which are built into the solution space,
natural boundary conditions are built into the weak form.
Example 1. Consider the 1-D Poisson equation with a Neumann boundary condition on
the left and a homogeneous Dirichlet (essential) boundary condition on the right,

d
2
u
dx
2
= f(x), 0 < x < 1, (1)
du
dx
(0) =
0
,
u(1) = 0.
If we multiply the ODE by v(x) and integrate by parts, we obtain
_
1
0
f(x)v(x) dx =
_
1
0
d
2
u
dx
2
v(x) dx =
_
1
0
du
dx
dv
dx
dx
du
dx
(1)v(1) +
du
dx
(0)
. .

0
v(0).
Suppose we impose the same right (essential) boundary condition on v as the solution u
satises, so v(1) = 0. Then we obtain
_
1
0
f(x)v(x) dx =
_
1
0
du
dx
dv
dx
dx +
0
v(0). (2)
If we dene
V
0
= {v H
1
[0, 1] : v(1) = 0}
a(u, v) =
_
1
0
du
dx
dv
dx
dx
L(v) =
_
1
0
f(x)v(x) dx
0
v(0),
then we obtain from (2) the weak form of the above Poisson equation: Find u V
0
such
that
a(u, v) = L(v) v V
0
.
The weak solution is also the minimizer over V
0
of the energy functional
J(v) =
1
2
a(v, v) L(v).
Remark. As before, one can show that a is bilinear, symmetric, coersive, and bounded
on V
0
. Hence, the abstract theory of boundary value problems can be applied to show the
existence, uniqueness, and continuous dependence of a weak solution in the Hilbert space
V
0
H
1
[0, 1].
1
Remark. If we impose on the weak solution u some additional smoothness, then we can
show that u solves the Poisson eqn (1). If u C
2
[0, 1], we can apply integration by parts to
(2) to obtain
_
1
0
f(x)v(x) dx =
_
1
0
d
2
u
dx
2
v(x) dx +
du
dx
(1)v(1)
du
dx
(0)v(0) +
0
v(0) (3)
for all v V
0
. But if (3) holds for all v V
0
, it must hold for all v H
1
0
[0, 1], i.e., it holds
if we impose the additional restriction v(0) = 0 on top of the essential boundary condition
v(1) = 0. Then
_
1
0
_

d
2
u
dx
2
f(x)
_
v(x) dx = 0 v H
1
0
[0, 1],
which implies that

d
2
u
dx
2
f(x) = 0 x [0, 1],
so the ODE in (1) holds. If we substitute this into (3) we obtain
du
dx
(1)v(1) +
_

0

du
dx
(0)
_
v(0) = 0 v V
0
.
The condition v V
0
forces v(1) = 0, so that
_

0

du
dx
(0)
_
v(0) = 0 v V
0
.
But v V
0
allows v(0) to vary. By picking v(0) = 1, we enforce the left boundary condition,

0

du
dx
(0) = 0.
Example 2. Consider the 1-D steady-state diusion equation with radiation boundary
condition on the left and a homogeneous Dirichlet boundary condition on the right,

d
dx
_
(x)
du
dx
_
= f(x), 0 < x < 1, (4)
(0)
du
dx
(0) = (u(0)
0
),
u(1) = 0,
where is a positive parameter. In the context of heat transfer, the left boundary condition
means that the heat ux across the boundary is proportional to the dierence between the
boundary temperature u(0) and some ambient temperature
0
. If we multiply the ODE by
v(x) and integrate by parts and apply the boundary conditions for u and assume v(1) = 0
(essential BC), we obtain
_
1
0
f(x)v(x) dx =
_
1
0
(x)
du
dx
dv
dx
dx (1)
du
dx
(1) v(1)
. .
0
+(0)
du
dx
(0)
. .
(u(0)
0
)
v(0)
2
and hence
_
1
0
(x)
du
dx
dv
dx
+ u(0)v(0) =
_
1
0
f(x)v(x) dx +
0
v(0).
The left hand side denes the bilinear form a(u, v) and the right hand side gives the linear
functional L(v). The weak solution lies in the function space V
0
= {v H
1
[0, 1] : v(1) = 0}.
Clearly a is symmetric and it is easy to show that a is bounded with respect to the Sobolev
H
1
norm. Coersivity can be established as long as the ux parameter is nonnegative. The
corresponding energy functional is
J(v) =
1
2
a(v, v) L(v) =
_
1
0
_
_
1
2
_
dv
dx
_
2
f(x)
_
_
v(x) dx +

2
v(0)
2

0
v(0).
Example 3. Natural Boundary Conditions in Higher Dimensions. Consider the
steady-state diusion equation with mixed boundary conditions
div ((x)u) = f(x), x (5)
u(x) = 0, x
1
(essential BC)
(x)u n(x) = (x), x
0
(natural BC)
where n denotes the outward unit normal to the boundary and =
1
+
0
. By this we
mean that
1
,
0
form a partition of , so that =
1

0
and
1

0
= . If we multiply
the PDE by v(x), apply Greens identity, apply the boundary conditions for u, and assume
that v satises the (essential) homogeneous Dirichlet boundary conditions on
1
, we obtain
_

(x)u v dx =
_

f(x)v(x) dx +
_

(x)u n(x) v(x) dS (6)


=
_

f(x)v(x) dx +
_

1
(x)u n(x) v(x)
. .
0
dS +
_

0
(x)v(x) dS.
This gives us the weak form: Find u V
0
such that
a(u, v) = L(v) v V
0
where
V
0
= {v H
1
() : v(x) = 0 x
1
}.
a(u, v) =
_

(x)u v dx
L(v) =
_

f(x)v(x) dx +
_

0
(x)v(x) dS
As in the 1-D case, one can show that a is bilinear, symmetric, bounded, and coercive, so
a unique weak solution u exists and minimizes J(v) = a(v, v)/2 L(v) over v V
0
. If
u C
2
(), then one can show that the weak solution satises the PDE and the boundary
conditions.
3
Exercises
1. Derive the weak form for the ODE BVP

d
dx
_
(x)
du
dx
_
= f(x), 0 < x < 1,
u(0) = g
0
,
du
dx
(1) =
1
,
i.e., give the appropriate space V
0
in which the weak solution lies, give the bilinear
functional a(u, v), and give the bounded linear functional L(v) associated with this
BVP. Note that you will need to apply a weak version of the superposition principle
to handle the left boundary condition.
2. Implement the nite element method for the BVP in problem 1 with continuous piece-
wise linear hat basis functions to obtain an approximate solution with
0
= 2,
g
1
= 2e
1
, = 1 + x
2
, and f(x) = 2(1 + x)
2
e
x
. The true solution is u(x) = 2e
x
.
Note that the left boundary condition is Dirichlet, but not homogeneous. Hand in the
usual material, i.e., a description of your implementation, code listings, plots, and a
summary of convergence results as you vary the grid spacing h.
4

You might also like