Section 8.

1 Partial Dierential Equations in Physics and Engineering 167

Solutions to Exercises 8.1
1. u
xx
+u
xy
= 2u is a second order, linear, and homogeneous partial dierential equation. u
x
(0, y) =
0 is linear and homogeneous.
5. u
t
u
x
+ u
xt
= 2u is second order and nonlinear because of the term u
t
u
x
. u(0, t) + u
x
(0, t) = 0 is
linear and homogeneous.
9. (a) Let u(x, y) = e
ax
e
by
. Then
u
x
= ae
ax
e
by
u
y
= be
ax
e
by
u
xx
= a
2
e
ax
e
by
u
yy
= b
2
e
ax
e
by
u
xy
= abe
ax
e
by
.
So
Au
xx
+ 2Bu
xy
+ Cu
yy
+ Du
x
+ Eu
y
+Fu = 0
Aa
2
e
ax
e
by
+ 2Babe
ax
e
by
+ Cb
2
e
ax
e
by
+Dae
ax
e
by
+Ebe
ax
e
by
+ Fe
ax
e
by
= 0
e
ax
e
by

Aa
2
+ 2Bab + Cb
2
+ Da + Eb +F

= 0
Aa
2
+ 2Bab + Cb
2
+ Da + Eb +F = 0,
because e
ax
e
by
= 0 for all x and y.
(b) By (a), in order to solve
u
xx
+ 2u
xy
+ u
yy
+ 2 u
x
+ 2 u
y
+u = 0,
we can try u(x, y) = e
ax
e
by
, where a and b are solutions of
a
2
+ 2ab +b
2
+ 2a + 2b + 1 = 0.
But
a
2
+ 2ab +b
2
+ 2a + 2b + 1 = (a + b + 1)
2
.
So a + b + 1 = 0. Clearly, this equation admits innitely many pairs of solutions (a, b). Here are
four possible solutions of the partial dierential equation:
a = 1, b = 2 u(x, y) = e
x
e
2y
a = 0, b = 1 u(x, y) = e
y
a = 1/2, b = 1/2 u(x, y) = e
x/2
e
y/2
a = 3/2, b = 1/2 u(x, y) = e
3x/2
e
y/2
13. We follow the outlined solution in Exercise 12. We have
A(u) = ln(u), (x) = e
x
, A(u(x(t)), t)) = A((x(0))) = ln(e
x(0)
) = x(0).
So the characteristic lines are
x = tx(0) + x(0) x(0) = L(xt) =
x
t + 1
.
Partial Dierential Equations in Rectangular Coordinates
So u(x, t) = f(L(x, t)) = f

x
t+1

. The condition u(x, 0) = e

x
implies that f(x) = e
x
and so
u(x, t) = e
x
t+1
.
Check: u
t
= e
x
t+1
x
(t+1)
2
, u
x
= e
x
t+1
1
t+1
,
u
t
+ ln(u)u
x
= e
x
t+1
x
(t + 1)
2
+
x
t + 1
e
x
t+1
1
t + 1
= 0.
17. We have
A(u) = u
2
, (x) =

x, A(u(x(t)), t)) = A((x(0))) = x(0).

So the characteristic lines are
x = tx(0) + x(0) x(0)(t + 1) x = 0.
Solving for x(0), we nd
x(0) =
x
t + 1
,
and so
u(x, t) = f

x
t + 1

.
Now
u(x, 0) = f(x) =

x.
So
u(x, t) =

x
t + 1
.
Section 8.2 Wave Equation, the Method of Separation of Variables 169
Solutions to Exercises 8.2
1. The solution is
u(x, t) =

n=1
sin
nx
L
_
b
n
cos c
nt
L
+b

n
sinc
nt
L
_
,
where b
n
are the Fourier sine coecients of f and b

n
are
L
cn
times the Fourier coecients
of g. In this exercise, b

n
= 0, since g = 0, b
1
= 0.05; and b
n
= 0 for all n > 1, because f is
already given by its Fourier sine series (period 2). So u(x, t) = 0.05 sinx cos t.
5. (a) The solution is
u(x, t) =

n=1
sin(nx) (b
n
cos(4nt) +b

n
sin(4nt)) ,
where b
n
is the nth sine Fourier coecient of f and b

n
is L/(cn) times the Fourier coecient
of g, where L = 1 and c = 4. Since g = 0, we have b

n
= 0 for all n. As for the Fourier
coecients of f, we can get them by using Exercise 17, Section 2.4, with p = 1, h = 1, and
a = 1/2. We get
b
n
=
8

2
sin
n
2
n
2
.
Thus
u(x, t) =
8

n=1
sin
n
2
n
2
sin(nx) cos(4nt)
=
8

k=0
(1)
k
(2k + 1)
2
sin((2k + 1)x) cos(4(2k + 1)t).
(b) Here is the initial shape of the string. Note the new Mathematica command that we
used to dene piecewise a function. (Previously, we used the If command.)
Clear f
f x_ : 2 x ; 0 x 1 2
f x_ : 2 1 x ; 1 2 x 1
Plot f x , x, 0, 1
Because the period of cos(4(2k + 1)t) is 1/2, the motion is periodic in t with period
1/2. This is illustrated by the following graphs. We use two dierent ways to plot the
graphs: The rst uses simple Mathematica commands; the second one is more involved and
is intended to display the graphs in a convenient array.
170 Chapter 8 Partial Dierential Equations in Rectangular Coordinates
Clear partsum
partsum x_, t_ :
8 Pi^2 Sum Sin 1 ^k 2 k 1 Pi x Cos 4 2 k 1 Pi t 2 k 1 ^2, k, 0, 10
Plot Evaluate partsum x, 0 , f x , x, 0, 1
Here is the motion in an array.
Section 8.2 Wave Equation, the Method of Separation of Variables 171
tt Table
Plot Evaluate partsum x, t , x, 0, 1 , PlotRange 0, 1 , 1, 1 ,
Ticks .5 , 1, .5, .5, 1 , DisplayFunction Identity , t, 0, 1, 1 20 ;
Show GraphicsArray Partition tt, 4
-1
-0.5
0.5
1
-1
-0.5
0.5
1
-1
-0.5
0.5
1
-1
-0.5
0.5
1
-1
-0.5
0.5
1
-1
-0.5
0.5
1
-1
-0.5
0.5
1
-1
-0.5
0.5
1
-1
-0.5
0.5
1
-1
-0.5
0.5
1
-1
-0.5
0.5
1
-1
-0.5
0.5
1
-1
-0.5
0.5
1
-1
-0.5
0.5
1
-1
-0.5
0.5
1
-1
-0.5
0.5
1
-1
-0.5
0.5
1
-1
-0.5
0.5
1
-1
-0.5
0.5
1
-1
-0.5
0.5
1
9. The solution is
u(x, t) =

n=1
sin(nx) (b
n
cos(nt) +b

n
sin(nt)) ,
where b

1
=
1

and all other b

n
= 0. The Fourier coecients of f are
b
n
= 2
_
1
0
x(1 x) sin(nx) dx.
To evaluate this integral, we will use integration by parts to derive rst the formula: for
a = 0,
_
x sin(ax) dx =
x cos(a x)
a
+
sin(a x)
a
2
+ C,
and
_
x
2
sin(ax) dx =
2 cos(a x)
a
3

x
2
cos(a x)
a
+
2 x sin(a x)
a
2
+C;
172 Chapter 8 Partial Dierential Equations in Rectangular Coordinates
thus
_
x(1 x) sin(ax) dx
=
2 cos(a x)
a
3

x cos(a x)
a
+
x
2
cos(a x)
a
+
sin(a x)
a
2

2 x sin(a x)
a
2
+ C.
Applying the formula with a = n, we get
_
1
0
x(1 x) sin(nx) dx
=
2 cos(n x)
(n)
3

x cos(n x)
n
+
x
2
cos(n x)
n
+
sin(n x)
(n)
2

2 x sin(n x)
(n)
2

1
0
=
2 ((1)
n
1)
(n)
3

(1)
n
n
+
(1)
n
n
=
2 ((1)
n
1)
(n)
3
=
_
4
(n)
3
if n is odd,
0 if n is even.
Thus
b
n
=
_
8
(n)
3
if n is odd,
0 if n is even,
and so
u(x, t) =
8

k=0
sin((2k + 1)x) cos((2k + 1)t)
(2k + 1)
3
+
1

sin(x) sin(t).
13. To solve

2
u
t
2
+
u
t
=

2
u
x
2
,
u(0, t) = u(, t) = 0,
u(x, 0) = sinx,
u
t
(x, 0) = 0,
we follow the method of the previous exercise. We have c = 1, k = .5, L = , f(x) = sinx,
and g(x) = 0. Thus the real number
Lk
c
= .5 is not an integer and we have n >
kL

for all
n. So only Case III from the solution of Exercise 12 needs to be considered. Thus
u(x, t) =

n=1
e
.5t
sin nx
_
a
n
cos
n
t +b
n
sin
n
t,
where

n
=
_
(.5n)
2
1.
Setting t = 0, we obtain
sinx =

n=1
a
n
sinnx.
Section 8.2 Wave Equation, the Method of Separation of Variables 173
Hence a
1
= 1 and a
n
= 0 for all n > 1. Now since
b
n
=
ka
n

n
+
2

n
L
_
L
0
g(x) sin
n
L
x dx, n = 1, 2, . . . ,
it follows that b
n
= 0 for all n > 1 and and the solution takes the form
u(x, t) = e
.5t
sinx
_
cos
1
t +b
1
sin
1
t
_
,
where
1
=
_
(.5)
2
1 =

.75 =

3
2
and
b
1
=
ka
1

1
=
1

3
.
So
u(x, t) = e
.5t
sin x
_
cos(

3
2
t) +
1

3
sin(

3
2
t)
_
.
17. (a) That G is even follows from its Fourier series representation that we derive in a
moment. That G is 2L-periodic follows from the fact that g is 2L-periodic and its integral
over one period is 0, because it is odd (see Section 7.1, Exercise 15).
Since G is an antiderivative of g

, to obtain its Fourier series, we apply Exercise 33,

Section 7.3, and get
G(x) = A
0

n=1
b
n
(g)
n
cos
n
L
x,
where b
n
(g) is the nth Fourier sine coecient of g

,
b
n
(g) =
2
L
_
L
0
g(x) sin
n
L
x dx
and
A
0
=
L

n=1
b
n
(g)
n
.
In terms of b

n
, we have
L

b
n
(g)
n
=
2
n
_
L
0
g(x) sin
n
L
x dx = cb

n
,
and so
G(x) =
L

n=1
b
n
(g)
n

L

n=1
b
n
(g)
n
cos
n
L
x
=

n=1
cb

n
_
1 cos(
n
L
x)
_
.
174 Chapter 8 Partial Dierential Equations in Rectangular Coordinates
Since the Fourier series of G contains only cosine terms, it follows that G is even.
(b) From (a), it follows that
G(x + ct) G(x ct) =

n=1
cb

n
__
1 cos(
n
L
(x + ct))
_

_
1 cos(
n
L
(x ct))
__
=

n=1
cb

n
_
cos(
n
L
(x + ct)) cos(
n
L
(x ct))
_
(c) Continuing from (b) and using the notation in the text, we obtain
1
2c
_
x+ct
xct
g

(s) ds =
1
2c
[G(x + ct) G(x ct)]
=

n=1
b

n
1
2
_
cos(
n
L
(x + ct)) cos(
n
L
(x ct))
_
=

n=1
b

n
sin(
n
L
x) sin(
n
L
ct)
=

n=1
b

n
sin(
n
L
x) sin(
n
t).
(d) To derive dAlemberts solution, proceed as follows:
u(x, t) =

n=1
b
n
sin(
n
L
x) cos(
n
t) +

n=1
b

n
sin(
n
L
x) sin(
n
t)
=
1
2
_
f

(x ct) +f

(x +ct)
_
+
1
2c
[G(x + ct) G(x ct)] ,
where in the last equality we used Exercise 16 and part (c).
Section 8.3 The One Dimensional Heat Equation 175
Solutions to Exercises 8.3
1. Multiply the solution in Example 1 by
78
100
to obtain
u(x, t) =
312

k=0
e
(2k+1)
2
t
2k + 1
sin(2k + 1)x.
5. We have
u(x, t) =

n=1
b
n
e
(n)
2
t
sin(nx),
where
b
n
= 2
_
1
0
x sin(nx) dx = 2
_

x cos(n x)
n
+
sin(n x)
n
2

2
_

1
0
= 2
cos n
n
= 2
(1)
n+1
n
.
So
u(x, t) =
2

n=1
(1)
n+1
e
(n)
2
t
sin(nx)
n
.
9. (a) The steady-state solution is a linear function that passes through the points (0, 0)
and (1, 100). Thus, u(x) = 100x.
(b) The steady-state solution is a linear function that passes through the points (0, 100)
and (1, 100). Thus, u(x) = 100. This is also obvious: If you keep both ends of the insulated
bar at 100 degrees, the steady-state temperature will be 100 degrees.
13. We have u
1
(x) =
50

x +100. We use (13) and the formula from Exercise 10, and get
(recall the Fourier coecients of f from Exercise 3)
u(x, t) =
50

x + 100
+

n=1
_
132

sin(n

2
)
n
2
100
_
2 (1)
n
n
__
e
n
2
t
sin nx.
17. Fix t
0
> 0 and consider the solution at time t = t
0
:
u(x, t
0
) =

n=1
b
n
sin
n
L
xe

2
n
t
0
.
We will show that this series converges uniformly for all x (not just 0 x L) by appeal-
ing to the Weierstrass M-test. For this purpose, it suces to establish the following two
inequalities:
(a)

b
n
sin
n
L
xe

2
n
t
0

M
n
for all x; and
176 Chapter 8 Partial Dierential Equations in Rectangular Coordinates
(b)

n=1
M
n
< .
To establish (a), note that
|b
n
| =

2
L
_
L
0
f(x) sin
n
L
x dx

2
L
_
L
0

f(x) sin
n
L
x

dx
(The absolute value of the integral is
the integral of the absolute value.)

2
L
_
L
0
|f(x)| dx = A (because | sinu| 1 for all u).
Note that A is a nite number because f is bounded, so its absolute value is bounded and
hence its integral is nite on [0, L]. We have

b
n
sin
n
L
xe

2
n
t
0

Ae

2
n
t
0
= Ae

c
2

2
t
0
L
2
n
2
A
_
e

c
2

2
t
0
L
2
_
n
= Ar
n
,
where r = e

c
2

2
t
0
L
2
< 1. Take M
n
= Ar
n
. Then a holds and

M
n
is convergent because it
is a geometric series with ratio r < 1.
Section 8.4 Heat Conduction in Bars: Varying the Boundary Conditions 177
Solutions to Exercises 8.4
1. Since the bar is insulated and the temperature inside is constant, there is no exchange
of heat, and so the temperature remains constant for all t > 0. Thus u(x, t) = 100 for all
t > 0. This is also a consequence of (2), since in this case all the a
n
s are 0 except a
0
= 100.
5. The solution is given by (4) with c = L = 1, and where the coecients c
n
are the
Fourier cosine coecients of the function f(x) = cos x. This function is clearly its own
cosine series with a
n
= 0 for all n = 1 and a
1
= 1. Thus
u(x, t) = e

2
1
t
cos x = e

2
t
cos x.
9. This is a straightforward application of Exercise 7. For Exercise 1 the average is 100.
For Exercise 2 the average is a
0
= 0.
13. The solution is given by (8), where c
n
is given by (11). We have
_
1
0
sin
2

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

1
0
=
1
2
_
1
1
2
n
sin(2
n
)
_
.
Since
n
is a solution of tan = , we have sin
n
=
n
cos
n
, so
sin2
n
= 2 sin
n
cos
n
= 2
n
cos
2

n
,
and hence
_
1
0
sin
2

n
x dx =
1
2
_
1 + cos
2

n
_
.
Also,
_
1/2
0
sin
n
x dx =
1

n
_
1 cos

n
2
_
.
Applying (11), we nd
c
n
=
_
1/2
0
100 sin
n
x dx
_
_
1
0
sin
2

n
x dx
=
100

n
_
1 cos

n
2
_
_
1
2
_
1 + cos
2

n
_
=
200
_
1 cos
n
2
_

n
(1 + cos
2

n
)
.
Thus the solution is
u(x, t) =

n=1
200
_
1 cos
n
2
_

n
(1 + cos
2

n
)
e

2
n
t
sin
n
x.
178 Chapter 8 Partial Dierential Equations in Rectangular Coordinates
17. Part (a) is straightforward as in Example 2. We omit the details that lead to the
separated equations:
T

kT = 0,
X

kX = 0, X

(0) = X(0), X

(1) = X(1),
where k is a separation constant.
(b) If k = 0 then
X

= 0 X = ax + b,
X

(0) = X(0) a = b
X

(1) = X(1) a = (a + b) 2a = b;
a = b = 0.
So k = 0 leads to trivial solutions.
(c) If k =
2
> 0, then
X

2
X = 0 X = c
1
cosh x + c
2
sinhx;
X

(0) = X(0) c
2
= c
1
X

(1) = X(1) c
1
sinh +c
2
cosh = c
1
cosh c
2
sinh
c
1
sinh c
1
cosh = c
1
cosh c
2
sinh
c
1
sinh = c
2
sinh
c
1
sinh =
c
1

sinh .
Since = 0, sinh = 0. Take c
1
= 0 and divide by sinh and get
c
1
=
c
1

2
= 1 k = 1.
So X = c
1
cosh x +c
2
sinh x. But c
1
= c
2
, so
X = c
1
cosh x +c
2
sinh x = c
1
coshx c
1
sinhx = c
1
e
x
.
Solving the equation for T, we nd T(t) = e
t
; thus we have the product solution
c
0
e
x
e
t
,
where, for convenience, we have used c
0
as an arbitrary constant.
(d) If k =
2
< 0, then
X

+
2
X = 0 X = c
1
cos x +c
2
sin x;
X

(0) = X(0) c
2
= c
1
X

(1) = X(1) c
1
sin +c
2
cos = c
1
cos c
2
sin
c
1
sin c
1
cos = c
1
cos c
2
sin
c
1
sin = c
2
sin
c
1
sin =
c
1

sin .
Section 8.4 Heat Conduction in Bars: Varying the Boundary Conditions 179
Since = 0, take c
1
= 0 (otherwise you will get a trivial solution) and divide by c
1
and get

2
sin = sin sin = 0 = n,
where n is an integer. So X = c
1
cos nx +c
2
sin nx. But c
1
= c
2
, so
X = c
1
_
n cos nx sin nx
_
.
Call X
n
= n cos nx sinnx.
(e) To establish the orthogonality of the X
n
s, treat the case k = 1 separately. For k =
2
,
we refer to the boundary value problem
X

+
2
n
X = 0, X(0) = X

(0), X(1) = X

(1),
that is satised by the X
n
s, where
n
= n. We establish orthogonality using a trick from
Sturm-Liouville theory (Chapter 6, Section 6.2). Since
X

m
=
2
m
X
m
and X

n
=
2
n
X
n
,
multiplying the rst equation by X
n
and the second by X
m
and then subtracting the
resulting equations, we obtain
X
n
X

m
=
2
m
X
m
X
n
and X
m
X

n
=
2
n
X
n
X
m
X
n
X

m
X
m
X

n
= (
2
n

2
m
)X
m
X
n
_
X
n
X

m
X
m
X

n
_

= (
2
n

2
m
)X
m
X
n
where the last equation follows by simply checking the validity of the identity X
n
X

m

X
m
X

n
=
_
X
n
X

m
X
m
X

n
_

. So
(
2
n

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

m
(x) X
m
(x)X

n
(x)
_

dx
= X
n
(x)X

m
(x) X
m
(x)X

n
(x)

1
0
,
because the integral of the derivative of a function is the function itself. Now we use the
boundary conditions to conclude that
X
n
(x)X

m
(x) X
m
(x)X

n
(x)

1
0
= X
n
(1)X

m
(1) X
m
(1)X

n
(1) X
n
(0)X

m
(0) +X
m
(0)X

n
(0)
= X
n
(1)X
m
(1) + X
m
(1)X
n
(1) + X
n
(0)X
m
(0) X
m
(0)X
n
(0)
= 0.
Thus the functions are orthogonal. We still have to verify the orthogonality when one of
the X
n
s is equal to e
x
. This can be done by modifying the argument that we just gave.
(f) Superposing the product solutions, we nd that
u(x, t) = c
0
e
x
e
t
+

n=1
c
n
T
n
(t)X
n
(x).
180 Chapter 8 Partial Dierential Equations in Rectangular Coordinates
Using the initial condition, it follows that
u(x, 0) = f(x) = c
0
e
x
+

n=1
c
n
X
n
(x).
The coecients in this series expansion are determined by using the orthogonality of the
X
n
s in the usual way. Let us determine c
0
. Multiplying both sides by e
x
and integrating
term by term, it follows from the orthogonality of the X
n
that
_
1
0
f(x)e
x
dx = c
0
_
1
0
e
2x
dx +

n=1
c
n
=0
..
_
1
0
X
n
(x)e
x
dx .
Hence
_
1
0
f(x)e
x
dx = c
0
_
1
0
e
2x
dx = c
0
1 e
2
2
.
Thus
c
0
=
2e
2
e
2
1
_
1
0
f(x)e
x
dx.
In a similar way, we prove that
c
n
=
1

n
_
1
0
f(x)X
n
(x) dx
where

n
=
_
1
0
X
2
n
(x) dx .
This integral can be evaluated as we did in Exercise 15 or by straightforward computations,
using the explicit formula for the X
n
s, as follows:
_
1
0
X
2
n
(x) dx =
_
1
0
_
n cos nx sinnx
_
2
dx
=
_
1
0
_
n
2

2
cos
2
nx + sin
2
nx 2n cos(nx) sin(nx)
_
dx
=
=(n
2

2
)/2
..
_
1
0
n
2

2
cos
2
nx dx+
=1/2
..
_
1
0
sin
2
nx dx
2n
=0
..
_
1
0
cos(nx) sin(nx) dx
=
n
2

2
+ 1
2
.
Section 8.5 The Two Dimensional Wave and Heat Equations 181
Solutions to Exercises 8.5
5. We proceed as in Exercise 3. We have
u(x, y, t) =

n=1

m=1
(B
mn
cos
mn
t + B

mn
sin
mn
t) sinmx sinny,
where
mn
=

m
2
+n
2
, B
mn
= 0, and
B

mn
=
4

m
2
+n
2
_
1
0
_
1
0
sin mx sin ny dx dy
=
4

m
2
+n
2
_
1
0
sin mx dx
_
1
0
sin ny dy
=
_
16

m
2
+n
2
(mn)
2
if m and n are both odd,
0 otherwise.
Thus
u(x, y, t) =

k=0

l=0
16 sin((2k + 1)x) sin((2l + 1)y)
_
(2k + 1)
2
+ (2l + 1)
2
(2k + 1)(2l + 1)
2
sin
_
(2k + 1)
2
+ (2l + 1)
2
t
182 Chapter 8 Partial Dierential Equations in Rectangular Coordinates
Solutions to Exercises 8.6
1. The solution is given by
u(x, y) =

n=1
B
n
sin(nx) sinh(ny),
where
B
n
=
2
sinh(2n)
_
1
0
x sin(nx) dx
=
2
sinh(2n)
_

x cos(n x)
n
+
sin(n x)
n
2

2
_

1
0
=
2
sinh(2n)
(1)
n
n
=
2
sinh(2n)
(1)
n+1
n
.
Thus,
u(x, y) =
2

n=1
(1)
n+1
nsinh(2n)
sin(nx) sinh(ny).
5. Start by decomposing the problem into four subproblems as described by Figure 3. Let
u
j
(x, y) denote the solution to problem j (j = 1, 2, 3, 4). Each u
j
consists of only one term
of the series solutions, because of the orthogonality of the sine functions. For example, to
compute u
1
, we have
u
1
(x, y) =

n=1
A
n
sinnx sinh[n(1 y)],
where
A
n
=
2
sinhn
_
1
0
sin 7x sinnx dx.
Since the integral is 0 unless n = 7 and, when n = 7,
A
7
=
2
sinh 7
_
1
0
sin
2
7x dx =
1
sinh7
.
Thus
u
1
(x, y) =
1
sinh 7
sin7x sinh[7(1 y)].
Section 8.6 Laplaces Equation in Rectangular Coordinates 183
In a similar way, appealing to the formulas in the text, we nd
u
2
(x, y) =
1
sinh
sinx sinh(y)
u
3
(x, y) =
1
sinh 3
sinh[3(1 x)] sin(3y)
u
4
(x, y) =
1
sinh 6
sinh6x sin(6y);
u(x, y) = u
1
(x, y) + u
2
(x, y) +u
3
(x, y) +u
4
(x, y)
=
1
sinh 7
sin7x sinh[7(1 y)] +
1
sinh
sin(x) sinh(y)
+
1
sinh 3
sinh[3(1 x)] sin(3y) +
1
sinh6
sinh(6x) sin(6y)
13. Consider the function
f(z) = iAcos[(z z
0
)],
where A and and the given real constants and z
0
= +i , where and are the given
real constants. Clearly, f is entire (analytic for all z). Appealing to (15), Section 1.6, we
nd that
f(z) = iA{cos[ Re ((z z
0
))] cosh[ Im((z z
0
))]
sin[ Re ((z z
0
))] sinh[ Im((z z
0
))]}
= iA{cos[(x )] cosh[(y )] sin[(x )] sinh[(y )]}
= Asin[(x )] sinh[(y )] + iAcos[(x )] cosh[(y )].
Thus the real part of f is
(x, y) = Asin[(x )] sinh[(y )]
and its imaginary part is
(x, y) = Acos[(x )] cosh[(y )].
This completes the solution, since the real and imaginary parts are harmonic functions and
the imaginary part is a harmonic conjugate of the real part.
184 Chapter 8 Partial Dierential Equations in Rectangular Coordinates
Solutions to Exercises 8.7
1. We apply (2), with a = b = 1:
u(x, y) =

n=1

m=1
E
mn
sin mx sinny,
where
E
mn
=
4

2
(m
2
+n
2
)
_
1
0
_
1
0
x sinmx sinny dx dy
=
4

2
(m
2
+n
2
)
_
1
0
x sinmx dx
=
1(1)
n
n
..
_
1
0
sin ny dy
=
4

4
(m
2
+n
2
)
1 (1)
n
n
_

x cos(m x)
m
+
sin(mx)
m
2

1
0
=
4

4
(m
2
+n
2
)
1 (1)
n
n
(1)
m
m
.
Thus
u(x, y) =
8

k=0

m=1
(1)
m
(m
2
+ (2k + 1)
2
)m(2k + 1)
sinmx sin((2k + 1)y).
5. We will use an eigenfunction expansion based on the eigenfunctions (x, y) = sin mx sinny,
where (x, y) =
2
_
m
2
+n
2
_
sinmx sinny. So plug
u(x, y) =

n=1

m=1
E
mn
sin mx sinny
into the equation u = 3u 1, proceed formally, and get
(

n=1

m=1
E
mn
sinmx sinny) = 3

n=1

m=1
E
mn
sinmx sinny 1

n=1

m=1
E
mn
(sinmx sinny) = 3

n=1

m=1
E
mn
sinmx sinny 1

n=1

m=1
E
mn

2
_
m
2
+ n
2
_
sinmx sinny
= 3

n=1

m=1
E
mn
sinmx sinny 1

n=1

m=1
_
3 +
2
_
m
2
+ n
2
__
E
mn
sin mx sinny = 1.
Thinking of this as the double sine series expansion of the function identically 1, it follows
that
_
3+
2
_
m
2
+n
2
__
E
mn
are double Fourier sine coecients, given by (see (8), Section 3.7)
_
3 +
2
_
m
2
+ n
2
__
E
mn
= 4
_
1
0
_
1
0
sinmx sinny dx dy
= 4
1 (1)
m
m
1 (1)
n
n
=
_
0 if either m or n is even,
16

2
mn
if both m and n are even.
.
Section 8.7 Poissons Equation: The Method of Eigenfunction Expansions 185
Thus
E
mn
=
_
0 if either m or n is even,
16

2
mn
_
3+
2
_
m
2
+n
2
__
if both m and n are even,
and so
u(x, y) =
16

k=0

l=0
sin((2k + 1)x) sin((2l + 1)y)
(2k + 1) (2l + 1)
_
3 +
2
_
(2k + 1)
2
+ (2l + 1)
2
__.
186 Chapter 8 Partial Dierential Equations in Rectangular Coordinates
Solutions to Exercises 8.8
1. We use a combination of solutions from (2) and (3) and try a solution of the form
u(x, y) =

n=1
sin mx
_
A
m
cosh m(1 y) +B
m
sinh my

.
(If you have tried a dierent form of the solution, you can still do the problem, but your
answer may look dierent from the one derived here. The reason for our choice is to simplify
the computations that follow.) The boundary conditions on the vertical sides are clearly
satised. We now determine A
m
and B
m
so as to satisfy the conditions on the other sides.
Starting with u(1, 0) = 100, we nd that
100 =

m=1
A
m
cosh msinmx.
Thus A
m
cosh m is the sine Fourier coecient of the function f(x) = 100. Hence
A
m
cosh m =
2

_

0
100 sinmx dx A
m
=
200
mcoshm
[1 (1)
m
] .
Using the boundary condition u
y
(x, 1) = 0, we nd
0 =

m=1
sin mx
_
A
m
(m) sinh[m(1 y)] + mB
m
coshmy

y=1
.
Thus
0 =

m=1
mB
m
sinmx coshm.
By the uniqueness of Fourier series, we conclude that mB
m
cosh m = 0 for all m. Since
mcoshm = 0, we conclude that B
m
= 0 and hence
u(x, y) =
200

m=1
[1 (1)
m
]
mcoshm
sinmx cosh[m(1 y)]
=
400

k=0
sin[(2k + 1)x]
(2k + 1) cosh(2k + 1)
cosh[(2k + 1)(1 y)].
5. We combine solutions of dierent types from Exercise 4 and try a solution of the form
u(x, y) = A
0
+B
0
y +

m=1
cos
m
a
x
_
A
m
cosh[
m
a
(b y)] +B
m
sinh[
m
a
y]

.
Using the boundary conditions on the horizontal sides, starting with u
y
(x, b) = 0, we nd
that
0 = B
0
+

m=1
m
a
B
m
cos
m
a
x cosh
_
m
a
b

.
Section 8.8 Neumann and Robin Conditions 187
Thus B
0
= 0 and B
m
= 0 for all m 1 and so
A
0
+

m=1
A
m
cos
m
a
x cosh[
m
a
(b y)].
Now, using u(x, 0) = g(x), we nd
g(x) = A
0
+

m=1
A
m
cosh[
m
a
b] cos
m
a
x.
Recognizing this as a cosine series, we conclude that
A
0
=
1
a
_
a
0
g(x) dx
and
A
m
cosh[
m
a
b] =
2
a
_
a
0
g(x) cos
m
a
x dx;
equivalently, for m 1,
A
m
=
2
a cosh[
m
a
b]
_
a
0
g(x) cos
m
a
x dx.
9. We follow the solution in Example 3. We have
u(x, y) = u
1
(x, y) +u
2
(x, y),
where
u
1
(x, y) =

m=1
B
m
sin mx sinhmy,
with
B
m
=
2
mcosh(m)
_

0
sin mx dx =
2
m
2
cosh(m)
(1 (1)
m
);
and
u
2
(x, y) =

m=1
A
m
sinmx cosh[m( y)],
with
A
m
=
2
cosh(m)
_

0
sinmx dx =
2
mcosh(m)
(1 (1)
m
).
Hence
u(x, y) =
2

m=1
(1 (1)
m
)
mcosh(m)
sinmx
_
sinhmy
m
+ cosh[m( y)]
_
=
4

k=0
sin(2k + 1)x
(2k + 1) cosh[(2k + 1)]
_
sinh[(2k + 1)y]
(2k + 1)
+ cosh[(2k + 1)( y)]
_
.