Heat Equation with a Radiation Boundary Condition

u
t
(x, t) =
2
u
xx
(x, t), 0 < x < , t > 0 (1)
u(0, t) = 0, u
x
(1, t) +hu(1, t) = 0
u(x, 0) = (x)
1. Separate Variables Look for simple solutions in the form
u(x, t) = X(x)T(t).
Substituting into (1) and dividing both sides by X(x)T(t) gives

T(t)
T(t)
=
2
X

(x)
X(x)
Since the left side is independent of x and the right side is independent of t, it follows
that the expression must be a constant:

T(t)

2
T(t)
=
X

(x)
X(x)
= .
(Here

T means the derivative of T with respect to t and X

means means the derivative

of X with respect to x.) We seek to nd all possible constants and the corresponding
nonzero functions X and T.
We obtain
X

X = 0,

T
2
T = 0.
The solution of the second equation is
T(t) = Ce

2
t
(2)
where C is an arbitrary constant. Furthermore, the boundary conditions give
X(0)T(t) = 0, X

(1) +hX(1)T(t) = 0 for all t.

Since T(t) is not identically zero we obtain the desired eigenvalue problem
X

(x) X(x) = 0, X(0) = 0, X

(1) +hX(1)T(t) = 0. (3)

2. Find Eigenvalues and Eignevectors The next main step is to nd the eigenvalues
and eigenfunctions from (3). There are, in general, three cases:
(a) If = 0 then X(x) = ax +b so applying the boundary conditions we get
0 = X(0) = b, 0 = X

(1) +hX(1) = a(1 +h) a = 0 ( unless h = 1).

We conclude that
0
= 0 is note an eigenvalue unless h = 1.
1
(b) If =
2
> 0 then
X(x) = a cosh(x) +b sinh(x)
and
X

(x) = asinh(x) +bcosh(x).

Applying the boundary conditions we have
0 = X(0) = a a = 0
and
0 = X

(1) +hX(1) = b(cosh() +hsinh()) for b = 0 tanh() =

h
.
This case is a bit more complicated depending on whether h is positive or negative.
h positive corresponds to heat owing out of the rod so there are no positive
eigenvalues. There are also no positive eigenvalues for 1 < h < 0. But for
h < 1 we see that there is a single positive eigenvalue.
Consider the following alternative argument: If X

(x) = X(x) then multiplying

by X we have X(x)X

(x) = X(x)
2
. Integrate this expression from x = 0 to x =
, apply integration by parts on the right and use X(0) = 0 and X

(1) = hX(1).
We have

_
1
0
X(x)
2
dx =
_
1
0
X(x)X

(x) dx =
_
1
0
X

(x)
2
dx hX(1)
2

0
.
We conclude that
=
_

0
X

(x)
2
dx +hX(1)
2
_

0
X(x)
2
dx
and we see that is negative unless h is a large negative number. This calculation
does not, however, give any real idea large needs to be. From here on we consider
only the case h > 0.
(c) Finally, consider =
2
so that
X(x) = a cos(x) +b sin(x)
and
X

(x) = asin(x) +bcos(x).

2
Applying the boundary conditions we have
0 = X(0) = a a = 0 0 = X

(1) +hX(1) = b(cos() +hsin())

From this we conclude
tan() =

h
.
By graphing the functions on the right and left on the same axis it is easy to see
that there are innitely many values
n
with

2
<
1
< ,
3
2
<
2
< 2 ,
and in general
(2n 1)
2
<
n
< n, and
n
n

(2n 1)
2
.
Thus we have eignevalues and eigenfunctions

n
=
2
n
, X
n
(x) = sin(
n
x), n = 1, 2, 3, . (4)
From (2) we also have the associated functions T
n
(t) = e

2
nt
.
3. Write as a Formal Sum From the above considerations we can conclude that for
any integer N and constants {a
n
}
N
n=0
u
n
(x, t) =
N

n=1
b
n
T
n
(t)X
n
(x) =
N

n=1
b
n
e
nt
sin (
n
x) .
satises the dierential equation in (1) and the boundary conditions.
4. Use Fourier Series to Find Coecients The only problem remaining is to somehow
pick the constants b
n
so that the initial condition u(x, 0) = (x) is satised. To do
this we need a theory which is more general than Fourier series. This theory is called
3
Sturm-Liouville theory and we will discuss it a little bit later. The main thing is that
it guarantees that, just as with Fourier sereis, we look for u as an innite sum
u(x, t) =

n=1
b
n
e

2
nt
sin (
n
x)
and we seek {b
n
} satisfying
(x) = u(x, 0) =

n=1
b
n
sin (
n
x) .
We claim (see proof at the end of the notes)
_
1
0
X
n
(x)X
m
(x) dx =
_
1
0
sin(
n
x) sin(
m
x) dx = 0 n = m (5)
and
_
1
0
X
2
n
(x) dx =
_
1
0
sin
2
(
n
x) dx =
(1 +hcos
2
(
n
))
2

1

n
= 0 (6)
so

n
=
2
(1 +h
1
cos
2
(
n
))
n
2.
Just as we did in our formal study of Fourier series, to nd b
n
multiply both sides of
the formal series by X
n
(x) and integrate from 0 to 1:
_
1
0
(x) sin(
n
x) dx =

k=1
b
k
_
1
0
sin(
k
x) sin(
n
x) dx
= b
n
_
1
0
sin
2
(
n
x) dx =
b
n

n
Thus we nd
b
n
=
n
_
1
0
(x) sin (
n
x) dx. (7)
As an explicit example for the initial condition consider (x) = x. In this case (7) becomes
b
n
=
n
_
1
0
sin(
n
x) dx =
n
_
1
0
x
cos(
n
x)

n
dx
=
n
_
x
cos(
n
x)

1
0

_
1
0
cos(
n
x)

n
dx
_
4
=
n
_
cos(
n
)

n
+
_
1
0
cos(
n
x)

n
dx
_
=
n
_
cos(
n
)

n
+
sin(
n
)

2
n
_
=

n
(h + 1) sin(
n
)

2
n
where on the last step we have used
cos(
n
) =
hsin(
n
)

n
which follows from
tan() =

h
.
So nally we arrive at the solution
u(x, t) =

k=1
b
n
e

2
nt
sin(
n
x). (8)
Proof of Orthogonality and Derivation
n
First we obtain the desired formula for
n
.

1
n
=
_
1
0
sin
2
(
n
x) dx =
1
2
_
1
0
(1 cos(2
n
x)) dx
=
1
2
_
x
sin(2
n
x)
2
n
_

1
0
=
1
2
_
1
sin(2
n
)
2
n
_
=
1
2
_
1
sin(
n
) cos(
n
)

n
_
=
1
2
_
1 +h
1
cos
2
(
n
)

where on the last step we have used

sin(
n
)

n
=
cos(
n
)
h
which follows from
tan() =

h
.
5
Now we show orthogonality, i.e.,
_
1
0
sin(
n
x) sin(
m
x) dx = 0 for n = m.
Recall that sin(
j
x) = X
j
and that
X

j
=
j
X
j
, X
j
(0) = 0, X

j
(1) = hX
j
(1), j = n, m
so that

n
_
1
0
X
n
(x)X
m
(x) dx =
_
1
0
X

n
(x)X
m
(x) dx
=
_
1
0
X

n
(x)X

m
(x) dx +X

n
(x)X
m
(x)

1
0
=
_
1
0
X
n
(x)X

m
(x) dx + [X

n
(x)X
m
(x) X
n
(x)X

m
(x)]

1
0
=
m
_
1
0
X
n
(x)X
m
(x) dx +h[X
n
(1)X
m
(1) X
n
(1)X
m
(1)]
=
m
_
1
0
X
n
(x)X
m
(x) dx
Therefore we can conclude
(
n

m
)
_
1
0
X
n
(x)X
m
(x) dx = 0
and since
n
=
m
for n = m we have
_
1
0
X
n
(x)X
m
(x) dx = 0.
6