Professional Documents
Culture Documents
j k
a
j k
2
u
x
k
x
j
+
l
b
l
u
x
l
+ cu = 0 ,
with a corresponding nonhomogeneous partial differential equation,
j k
a
j k
2
u
x
k
x
j
+
l
b
l
u
x
l
+ cu = f
where f is some nonzero function. To keep our discussion reasonably brief, let us limit ourselves
to problems involving one spatial variable x and one temporal variable t .
For example, we might take our original heat ow problem on a rod of length L ,
Find the solution u = u(x, t ) to the heat equation
u
t
6
2
u
x
2
= 0 for 0 < x < L and 0 < t (22.1a)
that also satises the boundary conditions
u(0, t ) = 0 and u(L, t ) = 0 for 0 < t (22.1b)
and the initial conditions
u(x, 0) = u
0
(x) for 0 < x < L (22.1c)
where u
0
is some known function (precisely what u
0
is will not be important in
these discussions).
and, letting f denote any reasonable function of x and y , consider
Find the solution u = u(x, t ) to the heat equation
u
t
6
2
u
x
2
= f (x, t ) for 0 < x < L and 0 < t (22.2a)
4/16/2012
Nonhomogeneous Problems Chapter & Page: 222
that also satises the boundary conditions
u(0, t ) = 0 and u(L, t ) = 0 for 0 < t (22.2b)
and the initial condition
u(x, 0) = u
0
(x) for 0 < x < L (22.2c)
where u
0
is some known function (precisely what u
0
is will not be important in
these discussions).
The only change we are making is the addition of the f in the partial differential equation. This
f (x, t ) describes the generation of heat or cold at position x and time t due to, say, radioactivity
of the material making up the rod, laser heating of spots of the rod, tiny furnaces or refrigerators
installed in the rod, or magic.
One thing we are not changing (in this section, at least) are the boundary conditions. We want
the same Sturm-Liouville appropriate boundary conditions as in the corresponding homogeneous
problem. The reason is that we will be using the set of eigenfunctions
{
1
(x),
2
(x),
3
(x), . . . }
obtained in solving the boundary-value/eigen-problemarisinginthe corresponding homogeneous
problem.
Now here are the basic ideas: For each value of t , we can treat f (x, t ) and u(x, t ) as
functions of just x . Since our set of eigenfunctions {
1
(x),
2
(x),
3
(x), . . .} is a complete
orthogonal set (with respect to some weight function w(x) on our interval of interest (a, b) ),
we can express f (x, t ) and u(x, t ) in terms of these eigenfunctions,
f (x, t ) =
k
f
k
(t )
k
(x) (22.3)
and
u(x, t ) =
k
g
k
(t )
k
(x) . (22.4)
Remember that the use of these eigenfunctions should ensure that u(x, t ) satises the given
(Sturm-Liouville appropriate) boundary conditions.
The f
k
(t )s are the (generalized) Fourier coefcients of f (x, t ) with respect to the
k
s
at time t . For each t , they can be computed using the standard formula for computing these
coefcients,
f
k
(t ) =
k
(x) | f (x, t )
2
=
_
b
a
k
[formula of g
k
(t ) and its derivatives]
k
(x) =
k
f
k
(t )
k
(x) .
The uniqueness of the generalized Fourier coefcients then assures us that, for each k , we must
have
formula of g
k
(t ) and its derivatives = f
k
(t ) .
That is, we have a differential equation for each g
k
. Solving this differential equation will give
the formula for each g
k
(t ) , a formula which will involve an arbitrary constant or two. These
constants can then be found as they were found in the homogeneous case, by setting
u
0
(x) = u(x, 0) =
k
g
k
(0)
k
(x)
and using the fact that we must then have
g
k
(0) =
k
| u
0
2
.
(This assumes u(x, 0) = u
0
(x) is the initial condition. Obvious adjustments must be made if
there are other initial conditions.)
!
Example 22.1: Let us solve the sample problem given above assuming f (x, t ) depends
only on x . That is, assume f (x) and u
0
(x) are any two known reasonable functions of
x , and lets nd the solution u = u(x, t ) to the heat equation
u
t
6
2
u
x
2
= f (x, t ) for 0 < x < L and 0 < t (22.5)
that also satises the boundary conditions
u(0, t ) = 0 and u(L, t ) = 0 for 0 < t (22.6)
and the initial condition
u(x, 0) = u
0
(x) for 0 < x < L . (22.7)
In solving the corresponding homogeneous problem (with f 0 ) we got the complete
set of eigenfunctions
{
1
(x),
2
(x),
3
(x), . . . }
given by
k
(x) = sin
_
k
L
x
_
for k = 1, 2, 3, . . . .
The corresponding eigenfunction expansion of the supposedly known function f is
f (x, t ) =
k=1
f
k
(t ) sin
_
k
L
x
_
version: 4/16/2012
Nonhomogeneous Problems Chapter & Page: 224
where
f
k
(t ) =
_
sin
_
k
L
x
_
f (x, t )
_
_
_
_sin
_
k
L
x
__
_
_
2
= =
2
L
_
L
0
f (x, t ) sin
_
k
L
x
_
dx .
The corresponding eigenfunction expansion for u(x, t ) for each t > 0 is
u(x, t ) =
k=1
g
k
(t ) sin
_
k
L
x
_
where the g
k
s are functions to be determined. Now observe what happens if we replace u
with
u
k
(x, t ) = g
k
(t )
k
(x) = g
k
(t ) sin
_
k
L
x
_
in the the left side of our partial differential equation:
u
k
t
6
2
u
k
x
2
=
t
_
g
k
(t ) sin
_
k
L
x
__
6
2
x
2
_
g
k
(t ) sin
_
k
L
x
__
=
dg
k
dt
sin
_
k
L
x
_
6g
k
(t )
_
_
k
L
_
2
sin
_
k
L
x
_
_
=
_
dg
k
dt
+ 6
_
k
L
_
2
g
k
(t )
_
sin
_
k
L
x
_
.
So we do get that
some formula of g
k
(t )
k
(x)
as desired. And thus, using the eigenfunction expansions for u and f , we have
u
t
6
2
u
x
2
= f (x, t )
t
_
k=1
u
k
(x, t )
_
6
2
x
2
_
k=1
u
k
(x, t )
_
=
k=1
f
k
(t ) sin
_
k
L
x
_
k=1
_
u
k
t
6
2
u
k
x
2
_
=
k=1
f
k
(t ) sin
_
k
L
x
_
k=1
_
dg
k
dt
+ 6
_
k
L
_
2
g
k
(t )
_
sin
_
k
L
x
_
=
k=1
f
k
(t ) sin
_
k
L
x
_
.
Hence, for each k , we must have
dg
k
dt
+ 6
_
k
L
_
2
g
k
(t ) = f
k
(t ) .
Recalling what the corresponding eigenvalues were, we see that we can also write this as
dg
k
dt
+ 6
k
g
k
(t ) = f
k
(t ) where
k
=
_
k
L
_
2
. (22.8)
version: 4/16/2012
Nonhomogeneous Problems Chapter & Page: 225
The above is a rst-order linear ordinary differential equation that can be solved using the
integrating factor e
6
k
t
.
1
Its solution is
g
k
(t ) = e
6
k
t
_
f (t )e
6
k
t
dt
which we can rewrite using a denite integral as
g
k
(t ) = e
6
k
t
_
t
s=0
f (s)e
6
k
s
ds + c
k
e
6
k
t
(22.9)
where c
k
is an arbitrary constant. For brevity, lets further rewrite this as
g
k
(t ) = g
0,k
(t ) + c
k
e
6
k
t
with g
0,k
(t ) = e
6
k
t
_
t
s=0
f (s)e
6
k
s
ds ,
and observe that
g
0,k
(0) = e
6
k
0
_
0
s=0
f (s)e
6
k
s
ds = 0 ,
which, in turn, gives us
g
k
(0) = g
0,k
(0) + c
k
e
6
k
0
= c
k
.
Thus,
u(x, t ) =
k=1
g
k
(t ) sin
_
k
L
x
_
=
k=1
_
g
0,k
(t ) + c
k
e
6
k
t
_
sin
_
k
L
x
_
. (22.10a)
with
g
0,k
(t ) = e
6
k
t
_
t
s=0
f (s)e
6
k
s
ds (22.10b)
Finally, we apply the initial condition:
u
0
(x) = u(x, 0) =
k=1
_
g
0,k
(0) + c
k
e
6
k
0
_
sin
_
k
L
x
_
=
k=1
[0 +c
k
] sin
_
k
L
x
_
=
k=1
c
k
sin
_
k
L
x
_
.
Clearly, the c
k
s here are the Fourier sine series coefcents for u
0
,
c
k
=
2
L
_
L
0
u
0
(x) sin
_
k
L
x
_
dx .
1
You may want to dash back to Appendix A (A Brief Review of Elementary Ordinary Differential Equations) to
review how to solve these equations.
version: 4/16/2012
Nonhomogeneous Problems Chapter & Page: 226
Combining this with formula (22.10) for u(x, t ) then yields our nal solution:
u(x, t ) =
k=1
_
g
0,k
(t ) + c
k
e
6
k
t
_
sin
_
k
L
x
_
(22.11a)
where
k
=
_
k
L
_
2
, (22.11b)
g
0,k
(t ) = e
6
k
t
_
t
s=0
f (s)e
6
k
s
ds (22.11c)
and
c
k
=
2
L
_
L
0
u
0
(x) sin
_
k
L
x
_
dx . (22.11d)
The nal set of formulas for the generic solution given above used a particular form for the
solution to the ordinary differential equations given in (22.8). Rather than memorize the above,
just remember the general process, and dont worry about using any particular form for solving
those ordinary differential equations.
!
k=1
f
k
sin
_
k
3
x
_
and
0 = u
0
(x) =
k=1
u
0,k
sin
_
k
3
x
_
,
are easily found:
f
k
=
2
3
_
3
0
1 sin
_
k
3
x
_
dx = =
2
_
1 (1)
k
_
k
and
u
0,k
= 0 .
version: 4/16/2012
Nonhomogeneous Problems Chapter & Page: 227
The corresponding eigenfunction expansion for u(x, t ) for each t > 0 is
u(x, t ) =
k=1
u
k
(x, t ) with u
k
(x, t ) = g
k
(t ) sin
_
k
3
x
_
where the g
k
s are functions to be determined. As before, we have
u
k
t
6
2
u
k
x
2
=
t
_
g
k
(t ) sin
_
k
3
x
__
6
2
x
2
_
g
k
(t ) sin
_
k
3
x
__
=
dg
k
dt
sin
_
k
3
x
_
6g
k
(t )
_
_
k
3
_
2
sin
_
k
3
x
_
_
=
_
dg
k
dt
+ 6
_
k
3
_
2
g
k
(t )
_
sin
_
k
3
x
_
.
From this, the fact that
k
=
_
k
3
_
2
and our sine series for f , we then get
u
t
6
2
u
x
2
= 1
t
_
k=1
u
k
(x, t )
_
6
2
x
2
_
k=1
u
k
(x, t )
_
=
k=1
f
k
sin
_
k
3
x
_
k=1
_
u
k
t
6
2
u
k
x
2
_
=
k=1
f
k
sin
_
k
3
x
_
k=1
_
dg
k
dt
+ 6
k
g
k
(t )
_
sin
_
k
3
x
_
=
k=1
f
k
sin
_
k
3
x
_
.
where
f
k
=
2
_
1 (1)
k
_
k
.
Consequently, each g
k
(t ) must satisfy
dg
k
dt
+ 6
k
g
k
(t ) = f
k
.
Accidentally forgetting that we had already solved a more general version of this, we re-solve
this simple, rst-order linear differential equation, obtaining (this time)
g
k
(t ) = e
6
k
t
_
f
k
e
6
k
t
dt
= e
6
k
t
_
f
k
6
k
e
6
k
t
+c
k
_
=
f
k
6
k
+ c
k
e
6
k
t
Replacing f
k
and
k
in the rst term with their values, this reduces to
g
k
(t ) =
3
_
1 (1)
k
_
(k)
3
+ c
k
e
6
k
t
.
version: 4/16/2012
Nonhomogeneous Problems Chapter & Page: 228
Thus,
u(x, t ) =
k=1
g
k
(t ) sin
_
k
L
x
_
=
k=1
_
3
_
1 (1)
k
_
(k)
3
+ c
k
e
6
k
t
_
sin
_
k
3
x
_
.
Finally, we apply our initial condition:
0 = u(x, 0) =
k=1
_
3
_
1 (1)
k
_
(k)
3
+ c
k
e
6
k
0
_
sin
_
k
3
x
_
=
k=1
_
3
_
1 (1)
k
_
(k)
3
+ c
k
_
sin
_
k
3
x
_
.
Since each term must be zero, we must have
c
k
=
3
_
1 (1)
k
_
(k)
3
for k = 1, 2, 3, . . . .
Plugging this back into our last formula for u(x, t ) and simplifying slightly, we get
u(x, t ) =
k=1
3
_
1 (1)
k
_
3
k
3
_
1 e
6
k
t
_
sin
_
k
3
x
_
with
k
=
_
k
3
_
2
. (22.12)
Its may be worth noting that the exponential terms shrink to zero rapidly as t increases,
and that the k
3
terms also shrink quickly to zero as k increases. So, for large t ,
u(x, t )
k=1
3
_
1 (1)
k
_
3
k
3
[1 0] sin
_
k
3
x
_
=
3
_
1 (1)
1
_
3
1
3
sin
_
1
3
x
_
+
3
_
1 (1)
2
_
3
2
3
sin
_
2
3
x
_
+
3
_
1 (1)
3
_
3
3
3
sin
_
3
3
x
_
+
3
sin
_
3
x
_
.
?
Exercise 22.2: Let denote some real value, and consider the problem of nding the
solution u = u(x, t ) to the heat equation
u
t
6
2
u
x
2
= e
t
f (x) for 0 < x < 3 and 0 < t
that also satises the boundary conditions
u(0, t ) = 0 and u(3, t ) = 0 for 0 < t
and the initial condition
u(x, 0) = u
0
(x) for 0 < x < 3 .
Assume f (x) and u
0
(x) are reasonable known functions on (0, L) .
a: Find the formula (analogous to formula (22.11)) for the solution u(x, t ) .
b: What happens to u(x, t ) as t when > 0 ?
c: What happens to u(x, t ) as t when < 0 ?
?
Exercise 22.3: Consider the problem of nding the solution u = u(x, t ) to the heat
equation
u
t
6
2
u
x
2
= f (x) for 0 < x < 3 and 0 < t
that also satises the boundary conditions
u
x
(0, t ) = 0 and u
x
(3, t ) = 0 for 0 < t
and the initial condition
u(x, 0) = u
0
(x) for 0 < x < 3 .
Assume f (x) and u
0
(x) are reasonable known functions on (0, L) .
a: Find the formula (analogous to formula (22.11)) for the solution u(x, t ) .
b: What happens to u(x, t ) as t ?
version: 4/16/2012