CHAPTER 1

Partial Differential Equations on Bounded

Domains
1. Separation of Variables
Exercise 1. The solution is
u(x, t) =

an en t sin nx

n=1

where
2
an =

Thus
u(x, t) =

/2

sin nxdx =

2
((1)n cos(n/2))
n

2 t
2
2 9t
2 25t
e sin x e4t sin 2x +
e
sin 3x +
e
sin 5x +

3
5

Exercise 2. The solution is given by formula (4.14) in the text, where the coefficients are given by (4.15) and (4.16). Since G(x) = 0 we have cn = 0. Then
Z
Z
2
2 /2
x sin nx dx +
( x) sin nx dx
dn =
0
/2
R
Using the antiderivative formula x sin nx dx = (1/n2 ) sin nx (x/n) cos nx we
integrate to get
n
4
sin
dn =
n2
2
Exercise 3. Substituting u(x, y) = (x)(y) we obtain the Sturm-Liouville problem
= , x (0, l); (0) = (l) = 0
and the differential equation

= 0

The SLP has eigenvalues and eigenfunctions

n = n2 2 /l2 ,

n (x) = sin(nx/l)

and the solution to the equation is

n (y) = an cosh(ny/l) + bn sinh(ny/l)
1

1. PARTIAL DIFFERENTIAL EQUATIONS ON BOUNDED DOMAINS

Therefore
u(x, y) =

(an cosh(ny/l) + bn sinh(ny/l)) sin(nx/l)

n=1

Now we apply the boundary conditions:

u(x, 0) = F (x) =

an sin(nx/l)

n=1

and
u(x, 1) = G(x) =

(an cosh(n/l) + bn sinh(n/l)) sin(nx/l)

n=1

Thus

an =
and

2
l

F (x) sin(nx/l)dx

an cosh(n/l) + bn sinh(n/l) =
which gives the coefficients an and bn .

2
l

G(x) sin(nx/l)dx

Exercise 4. Substituting u = y(x)g(t) into the PDE and boundary conditions

gives the SLP
y = y, y(0) = y(1) = 0
and, for g, the equation
g + kg + c2 g = 0
The SLP has eigenvalues and eigenfunctions
n = n2 2 ,

yn (x) = sin nx, n = 1, 2, . . .

The g equation is a linear equation with constant coefficients; the characteristic

equations is
m2 + km + c2 = 0
which has roots
p
1
m = (k k 2 4c2 n2 2 )
2
By assumption k < 2c, and therefore the roots are complex for all n. Thus the
solution to the equation is (see the Appendix in the text on ordinary differential
equations)
gn (t) = ekt (an cos(mn t) + bn sin(mn t))
where
1p 2 2 2
mn =
4c n k 2 )
2
Then we form the linear combination

X
u(x, t) =
ekt (an cos(mn t) + bn sin(mn t)) sin(nx)
n=1

Now apply the initial conditions. We have

X
u(x, 0) = f (x) =
an sin(nx)
n=1

2. FLUX AND RADIATION CONDITIONS

and thus
an =

f (x) sin nx dx

The initial condition ut = 0 at t = 0 yields

ut (x, 0) = 0 =

n=1

Therefore

(bn mn kan ) sin(nx)

bn mn kan = 0

or

k
kan
=
bn =
mn
mn

f (x) sin nx dx

2. Flux and Radiation Conditions

Exercise 1. This problem models the transverse vibrations of a string of length l
when the left end is fixed (attached) and the right end experience no force; however,
the right end can move vertically. Initially the string is displaced by f (x) and it is
not given an initial velocity.
Substituting u = y(x)g(t) into the PDE and boundary conditions gives the SLP
y = y,

y(0) = y (l) = 0

and, for g, the equation

g + c2 g = 0
The SLP has eigenvalues and eigenfunctions
n = ((2n + 1)/l)2 ,

yn (x) = sin((2n + 1)x/l), n = 0, 1, 2, . . .

and the equation for g has general solution

gn (t) = an sin((2n + 1)ct/l) + bn cos((2n + 1)ct/l)
Then we form

X
u(x, t) =
(an sin((2n + 1)ct/l) + bn cos((2n + 1)ct/l)) sin((2n + 1)x/l)
n=0

Applying the initial conditions,

u(x, t) = f (x) =

bn sin((2n + 1)x/l)

n=0

which yields
bn =
And

1
|| sin((2n + 1)x/l)||2
ut (x, 0) = 0 =

n=0

f (x) sin((2n + 1)x/l)

an cn sin((2n + 1)x/l)

1. PARTIAL DIFFERENTIAL EQUATIONS ON BOUNDED DOMAINS

which gives an = 0. Therefore the solution is

u(x, t) =

bn cos((2n + 1)ct/l)) sin((2n + 1)x/l)

n=0

Exercise 2. This problem models the steady state temperatures in a rectangular

plate that is insulated on both sides, whose temperature is zero on the top, and
whose temperature is f (x) along the bottom. Letting u = g(y)(x) and substituting
into the PDE and boundary conditions gives the Sturm-Liouville problem
= ,

(0) = (a) = 0

and the differential equation

g g = 0

The eigenvalues and eigenfunctions are 0 = 0, (x) = 1 and

n = n2 2 /a2 ,

n (x) = cos(nx/a),

n = 1, 2, 3, . . .

The solution to the g equation is, corresponding to the zero eigenvalue, g0 (y) =
c0 y + d0 , and corresponding to the positive eigenvalues,
gn (y) = cn sinh(ny/a) + dn cosh(ny/a)
Thus we form the linear combination

X
(cn sinh(ny/a) + dn cosh(ny/a)) cos(nx/a)
u(x, y) = c0 y + d0 +
n=1

Now apply the boundary conditions on y to compute the coefficients:

u(x, 0) = f (x) = d0 +

dn cos(nx/a)

n=1

which gives
d0 =
Next

1
a

f (x)dx, dn =

u(x, b) = 0 = c0 b + d0 +

2
a

f (x) cos(nx/a)dx

(cn sinh(nb/a) + dn cosh(nb/a)) cos(nx/a)

n=1

Therefore

c0 = d0 /b,

cn =

cosh(nb/a)
dn
sinh(nb/a)

Exercise 3. The flux at x = 0 is (0, t) ux (0, t) = a0 u(0, t) > 0, so there

is heat flow into the bar and therefore adsorption. At x = 1 we have (1, t) =
ux (1, t) = a1 u(1, t) > 0, and therefore heat is flowing out of the bar, which is
radiation. The right side of the inequality a0 + a1 > a0 a1 is positive, so the
positive constant a1 , which measures radiation, must greatly exceed the negative
constant a0 , which measures adsorption.
In this problem substituting u = y(x)g(t) leads to the Sturm-Liouville problem
y = y,

y (0) a0 y(0) = 0, y (1) + a1 y(1) = 0

2. FLUX AND RADIATION CONDITIONS

and the differential equation

g = g
There are no nonpositive eigenvalues. If we take = k 2 > 0 then the solutions are
y(x) = a cos kx + b sin kx
Applying the two boundary conditions leads to the nonlinear equation
tan k =

(a0 + a1 )k
k 2 a0 a1

To determine the roots k, and thus the eigenvalues = k 2 , we can graph both sides
of this equation to observe that there are infinitely many intersections occuring at
kn , and thus there are infinitely many eigenvalues n = kn2 . The eigenfunctions are
a0
sin kn x
yn (x) = cos kn x +
kn
So the solution has the form

X
2
a0
cn en t (cos kn x +
u(x, t) =
sin kn x)
kn
n=1
The cn are then the Fourier coefficients

cn = (f, yn )/||yn ||2

If a0 = 1/4 and a1 = 4 then

3.75k
k2 + 1
From a graphing calculator, the first four roots are approximately k1 = 1.08, k2 =
3.85, k3 = 6.81, k4 = 9.82.
tan k =

Exercise 4. Letting c(x, t) = y(x)g(t) leads to the periodic boundary value problem
y = y, y(0) = y(2l), y (0) = y (2l)
and the differential equation
g = Dg
which has solution
g(t) = eDt
The eigenvalues and eigenfunctions are found exactly as in the solution of Exercise
4, Section 3.4, with l replaced by 2l. They are
0 = 0, n = n2 2 /l2 , n = 1, 2, . . .
and
y0 (x) = 1,
Thus we form

yn (x) = an cos(nx/l) + bn sin(nx/l), n = 1, 2, . . .

c(x, t) = a0 /2 +

en

2 Dt/l2

(an cos(nx/l) + bn sin(nx/l)

n=1

Now the initial condition gives

c(x, 0) = f (x) = a0 /2 +

n=1

(an cos(nx/l) + bn sin(nx/l)

1. PARTIAL DIFFERENTIAL EQUATIONS ON BOUNDED DOMAINS

which is the Fourier series for f . Thus the coefficients are given by
Z
1 2
lf (x) cos(nx/l)dx, n = 0, 1, 2, /ldots
an =
l 0
and
Z
1 2
bn =
lf (x) sin(nx/l)dx, n = 1, 2, /ldots
l 0
3. Laplaces Equation
Exercise 2. Substituting u(r, ) = g()y(r) into the PDE and boundary conditions
gives the Sturm-Liouville problem
g = g,

g(0) = g(/2) = 0

and the differential equation

r2 y + ry + y = 0
This SLP has been solved many times in the text and in the problems. The eigenvalues and eigenfunctions are
n = 4n2 ,

gn () = sin(2n),

n = 1, 2, . . .

The y equation is a Cauchy-Euler equation and has bounded solution

yn (r) = r2n
Form
u(r, ) =

bn r2n sin(2n)

n=1

Then the boundary condition at r = R gives

X
u(R, ) = f () =
bn R2n sin(2n)
n=1

Hence the coefficients are

bn =

1
R2n

/2

f () sin(2n)d

Exercise 3. Substituting u(r, ) = g()y(r) into the PDE and boundary conditions
gives the Sturm-Liouville problem
g = g,

g(0) = g (/2) = 0

and the differential equation

r2 y + ry + y = 0
The eigenvalues and eigenfunctions are
n = (2n + 1)2 ,

gn () = sin((2n + 1)),

n = 0, 1, 2, . . .

The y equation is a Cauchy-Euler equation and has bounded solution

yn (r) = r2n+1

3. LAPLACES EQUATION

Form
u(r, ) =

bn r2n+1 sin((2n + 1))

n=1

Then the boundary condition at r = R gives

u(R, ) = f () =

bn R2n+1 sin((2n + 1))

n=1

Hence the coefficients are

bn =

1
R2n+1

/2

f () sin((2n + 1))d

Exercise 4. Substituting r = 0 into Poissons integral formula (4.34) of the text

we instantly get the temperature at the origin as
Z 2
1
f ()d
u(0, 0) =
2 0
The right side is the average of the function f () over the interval [0, 2].
Exercise 5. Let w = u + v where u satisfies the Neumann problem and v satisfies
the boundary condition n v = 0. Then
E(w)

= E(u + v)
Z
Z
1
(hu hv)dA
(u u + 2u v + v v)dV
=
2
Z
Z
Z
1
= E(u) +
u v dV +
v vdV
hv dA
2

Z
Z
Z
Z
1
v vdV
hv dA
= E(u) +
vu n dA
vu dV +
2

Z
Z
Z
1
vh dA +
= E(u) +
hv dA
v vdV
2

Z
1
v vdV
= E(u) +
2

So E(u) E(w).
Exercise 6. Multiply both sides of the PDE by u and integrate over . We obtain
Z
Z
u2 dV
uu dV = c

Now use Greens first identity to obtain

Z
Z
Z
uu ndA
u u dV = c
u2 dV

or

au2 dA

u u dV = c

u2 dV

1. PARTIAL DIFFERENTIAL EQUATIONS ON BOUNDED DOMAINS

The left side is negative and the right side is positive. Then both must be zero, or
Z
u2 dV = 0

Hence u = 0 in .
The uniqueness is standard. Let u and v be two solutions to the boundary
value problem
u cu = f, x

n u + au = g, x

w cw = 0, x

n w + aw = 0, x

Then the difference w = u v satisfies the homogeneous problem

By the first part of the problem we know w = 0 and therefore u = v.

Exercise 7. The solution is given by equation (4.31) in the text. Here
f () = 4 + 3 sin
The right side is its Fourier series, so the Fourier coefficients are given by
a0
= 4, Rb1 = 3
2
with all the other Fourier coefficients identically zero. So the solution is
3r
sin
u(r, ) = 4 +
R
Exercise 8. Multiplying the equation u = 0 by u, integrating over , and then
using Greens identity gives
Z
Z
Z
uu dV =
uu ndA
u udV = 0

Thus

which implies

u udV = 0
u = 0

Thus u =constant.

4. Cooling of a Sphere
Exercise 1. The problem is
2
y y = y,

y(0) bounded, y() = 0

Making the transformation Y = y we get Y = Y . If = k 2 < 0 then

Y = a sinh k + b cosh k

or
y = 1 (a sinh k + b cosh k)
For boundedness at = 0 we set b = 0. Then y() = 0) forces sinh k = 0. Thus
k = 0. Consequently, there are no negative eigenvalues.

4. COOLING OF A SPHERE

Figure 1. Temperature at the center of the sphere in Exercise 2.

Exercise 2. From the formula developed in the text the temperature at = 0 is
u(0, t) = 74

2
1
(1)n+1 en kt
n
n=1

where k = 5.58 inches-squared per hour. A graph of an approximation using fifty

terms is shown in the figure.
Exercise 3. The boundary value problem is
2
ut = k(u + u )

u (R, t) = hu(R, t), t > 0

u(, 0) = f (),

0R

Assume u = y()g(t). Then the PDE and boundary conditions separate into the
boundary value problem
y + (2/)y + y = 0,

y (R) = hy(R), y bounded

and the differential equation

g = kg

The latter has solution g(t) = exp(kt). One can show that the eigenvalues are
positive. So let = p2 and make the substitution Y = y, as in the text, to obtain
Y + p2 Y = 0
This has solution
Y () = a cos p + b sin p

10

1. PARTIAL DIFFERENTIAL EQUATIONS ON BOUNDED DOMAINS

But Y (0) = 0 forces a = 0 (because y is bounded). Then the other boundary

condition forces p to satisfy the nonlinear equation
Rp
1 Rh
If we graph both sides of this equation against p we note that there are infinitely
many intersections, giving infinitely many roots pn , n = 1, 2, . . ., and therefore
infinitely many eigenvalues n = p2n . The corresponding eigenfunctions are
p
yn () = 1 sin( n )
tan Rp =

Thus we have

u(, t) =

n=1

p
cn en kt 1 sin( n )

The cn are found from the initial condition. We have

X
p
cn 1 sin( n )
u(, 0) = f () =
n=1

Thus

cn =

f () sin( n )d
RR 2
sin ( n )d
0

RR
0

Exercise 4. Representing the Laplacian in spherical coordinates, the boundary

value problem for u = u(, ), where (0, 1) and (0, ), is
2
1
u = u + u + 2
(sin u ) = 0

sin
u(1, ) = f (), 0

Observe, by symmetry of the boundary condition, u cannot depend on the angle .

Now assume u = R()Y (). The PDE separates into two equations,
2 R + 2R R = 0
and

1
(sin Y ) = Y
sin
We transform the Y equation by changing the independent variable to x = cos .
Then we get, using the chain rule,
d
1 d
=
sin d
dx
So the Y equation becomes


d
dy

(1 x2 )
= y
dx
dx

1<x<1

By the given facts, this equation has bounded, orthogonal, solutions yn (x) = Pn (x)
on [1, 1] when = n = n(n + 1), n = 0, 1, 2, . . .. Here Pn (x) are the Legendre
polynomials.
Now, the Requation then becomes
2 R + 2R n(n + 1)R = 0

4. COOLING OF A SPHERE

11

This is a Cauchy-Euler equation (see the Appendix in the text) with characteristic
equation
m(m 1) + m n(n + 1) = 0

The roots are m = n, (n + 1). Thus

Rn () = an n

are the bounded solutions (the other root gives the solution n1 , which is unbounded at zero). Therefore we form
u(, ) =

an n Pn (cos )

n=0

or, equivalently

u(, x) =

an n Pn (x)

n=0

Applying the boundary condition gives the coefficients. We have

u(1, x) = f (arccos x) =

an Pn (x)

n=0

By orthogonality we get

or

1
an =
||Pn ||2

f (arccos x)Pn (x)dx

Z
1
f ()Pn (cos ) sin d
||Pn ||2 0
By direct differentiation we get
an =

P0 (x) = 1, P1 (x) = x, P2 (x) =

1
5
3
(3x2 1), P3 (x) = x3 x
2
2
2

Also the norms are given by

2
2
, ||P2 ||2 =
3
5
When f () = sin the first few Fourier coefficients are given by

a0 = , a1 = a3 = 0, a2 = 0.49
4
Therefore a two-term approximation is given by
||P0 ||2 = 2, ||P1 ||2 =

u(, )

0.49 2

(3 cos2 1)
4
2

Exercise 5. To determine the temperature of the earth we must derive the temperature formula for any radius R (the calculation in the text uses R = ). The
method is exactly the same, but now the eigenvalues are n = n2 2 /R2 and the
eigenfunctions are yn = 1 sin(n/R), for n = 1, 2, . . .. Then the temperature is
u(, t) =

2RT0 X (1)n+1 lamn kt 1

e
sin(n/R)
n=1
n

12

1. PARTIAL DIFFERENTIAL EQUATIONS ON BOUNDED DOMAINS

Now we compute the geothermal gradient at the surface, which is u (, t) at = R.

We obtain

2T0 X n2 2 kt/R2
e
u (R, t) =
R n=1

If G is the value of the geothermal gradient at the current time t = tc , then

RG X n2 2 ktc /R2
=
e
2T0
n=1
We must solve for tc . Notice that the sum has the form

X
2
ean
n=1

where a = ktc /R . We can make an approximation by noting that the sum

represents a Riemann sum approximation to the integral
Z
2
1p
/a
eax dx =
2
0

So we use this value to approximate the sum, i.e.,

X
2 2
2
R
en ktc /R
2 ktc
n=1

Solving for tc gives

T02
G2 k
Substituting the numbers in from Exercise 5 in Section 2.4 gives tc = 5.15(10)8
years. This is the same approximation we found earlier.
tc =

5. Diffusion in a Disk
Exercise 1. The differential equation is (ry ) = ry. Multiply both sides by y
and integrate over [0, R] to get
Z R
Z R

(ry ) ydr =
ry 2 dr
0

Integrating the left hand side by parts gives

Z R
Z
2
ryy |R
r(y
)
dr
=

0
0

ry 2 dr

But, since y and y are assumed to be bounded, the boundary term vanishes. The
remaining integrals are nonnegative and so 0.
Exercise 2. Let y, and w, be two eigenpairs. Then (ry ) = ry and
(rw ) = rw. Multiply the first of these equations by w and the second by
y, and then subtract and integrate to get
Z R
Z R
[(ry ) w + (rw ) y]dr = ( )
ruwdr
0

5. DIFFUSION IN A DISK

13

Figure 2. The Bessel functions J0 (zn r).

Now integrate both terms in the first integral on the left hand side by parts to get
(ry w + rw y) |R
0 +

(ry w rw y )dr = ( )

ruwdr
0

The left side of the equation is zero and so y and w are orthogonal with respect to
the weight function r.
Exercise 3. We have
u(r, t) =

cn e0.25n t J0 (zn r)

n=1

where
cn =

R1
0

5r4 (1 r)J0 (zn r)dr

R1
J (z r)2 rdr
0 0 n

We have z1 = 2.405, z2 = 5.520, z3 = 8.654. Use a computer algebra program to

calculate a three-term approximation.
Exercise 4. The eigenfunctions J0 (zn r), n = 1, 2, 3, 4 are sketched in the figure.
For larger n the number of oscillations increases.
Exercise 5. The Maple worksheet follows.

14

1. PARTIAL DIFFERENTIAL EQUATIONS ON BOUNDED DOMAINS

6. Sources on Bounded Domains

Exercise 1. Use Duhamels principle to solve the problem
utt c2 uxx = f (x, t), 0 < x < , t > 0
u(0, t) = u(, t) = 0, t > 0
u(x, 0) = ut (x, 0) = 0,

0<x<1

Consider the problem for w = w(x, t, ), where is a parameter:

wtt c2 wxx = 0, 0 < x < , t > 0
w(0, t, ) = w(, t, ) = 0, t > 0
w(x, 0, ) = 0, wt (x, 0, ) = 0,

0<x<1

This problem was solved in Section 4.1 (see (4.14)(4.14)). The solution is
w(x, t, ) =

cn ( ) sin nct sin nx

n=1

where

Z
2
f (x, ) sin nx dx
nc 0
So the solution to the original problem is
Z t
w(x, t , )d
u(x, t) =
cn ( ) =

Exercise 2. If f = f (x), and does not depend on t, then the solution can be
written

 Z t
Z
2X
n2 k(t )
u(x, t) =
e
d sin nx
f (r) sin nr dr (
n=1 0
0

But a straightforward integration gives

Z t
2
2
1
en k(t ) d =
(1 en kt )
2
kn
0
Therefore

Z

2X 1
n2 kt
u(x, t) =
f (r) sin nr dr sin nx
(1 e
)
n=1 kn2
0
Taking the limit as t gives

U (x) lim u(x, t) = u(x, t) =


Z

2X 1
f
(r)
sin
nr
dr
sin nx
n=1 kn2
0

Now consider the steady state problem

kv = x( x),

v(0) = v() = 0

This can be solved directly by integrating twice and using the boundary conditions
to determine the constants of integration. One obtains
1
(2x2 x4 3 x)
v(x) =
12k

6. SOURCES ON BOUNDED DOMAINS

15

To observe that the solution v(x) is the same as the limiting solution U (x) we
expand the right side of the vequation in its Fourier sine series on [0, ]. Then
kv =
where

cn sin nx

n=1

Z
2
r( r) sin nrdr
0
Integrating the differential equation twice gives
cn =

kv(x) + kv (0)x =
Evaluating at x = gives

n=1

kv (0) =

cn n2 (sin nx x)

cn n2

n=1

Whence
kv(x) =

cn n2 sin nx

n=1

or


Z

2 X 1
v(x) =
f (r) sin nr dr sin nx
k n=1 n2
0

Hence v(x) = U (x).

Exercise 3. Following the hint in the text we have

u(x, y) =

gn (y) sin nx

n=1

where we find

gn n2 gn = fn (y),

gn (0) = gn (1) = 0

where fn (y) are the Fourier coefficients of f (x, y). From the variation of parameters
formula
Z
2 y
sinh(n( y))fn ()d
gn (y) = aeny + beny
n 0
Now gn (0) = 0 implies b = a. So we can write
Z
2 y
sinh(n( y))fn ()d
gn (y) = 2a sinh ny
n 0
which gives, using gn (1) = 0,
a=

1
n sinh n

sinh(n( 1))fn ()d

Thus the gn (y) are given by

Z
Z
2 sinh ny 1
2 y
gn (y) =
sinh(n( 1))fn ()d
sinh(n( y))fn ()d
n sinh n 0
n 0

16

1. PARTIAL DIFFERENTIAL EQUATIONS ON BOUNDED DOMAINS

Exercise 4. The problem is

ut = u + f (r, t)

0 r < R, t > 0

u(R, t) = 0, t > 0
u(r, 0) = 0, 0 < r < R
For w = w(r, t, ) we consider the problem
wt = w

0 r < R, t > 0

w(R, t ) = 0, t > 0
w(r, 0, ) = f (r, ), 0 < r < R

This is the model for heat flow in a disk of radius R; the solution is given by
equation (4.53) in Section 4.5 of the text. It is
X
w(r, t, ) =
cn ( )en kt J0 (zn r/R)
where zn are the zeros of the Bessel function J0 , n = zn2 /R2 and
Z R
1
f (r, )J0 (zn r/R)rdr
cn ( ) =
||J0 (zn r/R)||2 0

Then

u(r, t) =

w(r, t , )d

7. Parameter Identification Problems

Exercise 1. We have

p1 = p0 er1 t , p2 = p0 er2 t
Dividing the two equations and then taking natural logarithms gives
1
r1 r2 = (ln p1 ln p2 )
t
By the mean value theorem
d ln x
1
| ln a ln b|
=
|x=c =
|a b|
dx
c

for some c between a and b. Thus

1
1
|p1 p2 |
|r1 r2 | = | ln p1 ln p2 |
t
M t
Exercise 2. We have (x)utt = uxx . Putting u = Y (x)g(t) gives, upon separating
variables,
y = (x)y, y(0) = y(1) = 0
We have
yf = (x)f yf , yf (0) = yf (1) = 0
Integrating from x = 0 to x = s gives
Z s
yf (s) + yf (0) = f
(x)yf (x)dx
0

7. PARAMETER IDENTIFICATION PROBLEMS

Now integrate from s = 0 to s = 1 to get

Z 1Z
yf (0) = f
Z

(x)yf (x)dxds

= f

17

(1 x)(x)yf (x)dx

The last step follows by interchanging the order of integration. If (x) =r ho0 is a
constant, then
yf (0)
0 =
R1
f 0 (1 x)yf (x)dx

Exercise 3. From Exercise 2 in Section 4.6 we have the solution


Z

2X 1
n2 kt
f
(r)
sin
nr
dr
sin nx
(1

e
)
u(x, t) =
n=1 kn2
0
Therefore

U (t) = u(/2, t) =

2X
1
n2 kt
sin nr sin(n/2) 2 (1 e
) f (r)dr
n=1
kn

We want to recover f (x) if we know U (t). This problem is not stable, as the
following example shows. Let

2
u(x, t) = m3/2 (1 em t ) sin mx, f (x) = m sin mx

This pair satisfies the model. If m is sufficiently large, thenu(x, t) is uniformly

small; yet f (0) is large. So a small error in measuring U (t) will result in a large
change in f (x).
Exercise 4. The problem is
ut = uxx , x, t > 0
u(x, 0) = 0, x > 0
u(0, t) = f (t), t > 0
Taking Laplace transforms and solving gives

U (x, s) = F (s)ex

Here we have discarded the unbounded part of the solution. So, by convolution,
Z t
2
x
ex /4(t ) d
u(x, t) =
f ( ) p
3
4(t

)
0
Hence, evaluating at x = 1,

U (t) =

f ( ) p

1
4(t )3

e1/4(t ) d

which is an integral equation for f (t). Suppose f (t) = f0 is constant and U (5) = 10.
Then

Z 5 1/4(5 )

20
e
p
d = 2 erf c (1/ 5)
=
3
f0
4(5 )
0
where erf c = 1 erf . Thus f0 = 59.9 degrees.

18

1. PARTIAL DIFFERENTIAL EQUATIONS ON BOUNDED DOMAINS

Exercise 5. The problem

ut = Duxx vux ,

x R, t > 0;

u(x, 0) = eax , x R

can be solved by Fourier transforms to get

2
a
e(xvt) /((a+4Dt)
u(x, t) =
a + 4Dt
Thus, choosing a = v = 1 we get
2
1
U (t) = u(1, t) =
e(1t) /((1+4Dt)
1 + 4Dt

8. Finite Difference Methods

Exercise 1. The Cauchy-Euler algorithm for this problem is
Yn+1 = Yn + h(2nhYn + Yn2 ),

n = 0, 1, 2, . . . ; Y0 = 1

where h is the step size. With h = 0.1 the values of Yn at the points tn = nh, n =
0, . . . , 10 are:
1, 1.1, 1.199, 1.295, 1.385, 1.466, 1.534, 1.585, 1.615, 1.617, 1.587
Exercise 2. A time snapshot of the solution surface at t = 1 is shown in the
accompanying figure. The Maple program in Figure 4.8 of the text was run with
h = 0.1 and k = 0.1, which gives r = k/h2 = 10. This violates the stability
condition, and one can observe the highly oscillatory behavior of the numerical
scheme.
Exercise 3. The Maple worksheet and surface plot is given in the two figures.
Exercise 4. The Maple worksheet and surface plot is given in the two figures.
Exercise 5. The Maple worksheet and surface plot for the first problem in Exercise
5 is given in the accompanying figures.

8. FINITE DIFFERENCE METHODS

Figure 3. Time t = 1 profile of the numerical solution in Exercise

2 when h/ k = 10.

19

20

1. PARTIAL DIFFERENTIAL EQUATIONS ON BOUNDED DOMAINS

Figure 4. Maple program to solve Exercise 3.

8. FINITE DIFFERENCE METHODS

Figure 5. Solution surface in Exercise 3.

21

22

1. PARTIAL DIFFERENTIAL EQUATIONS ON BOUNDED DOMAINS

Figure 6. Maple program to solve Exercise 4.

8. FINITE DIFFERENCE METHODS

Figure 7. Solution surface in Exercise 4.

23

24

1. PARTIAL DIFFERENTIAL EQUATIONS ON BOUNDED DOMAINS

Figure 8. Maple program to solve Exercise 5.

8. FINITE DIFFERENCE METHODS

Figure 9. Solution surface in Exercise 5.

25

CHAPTER 2

Appendix

1. The equation y + 2y = ex is first order, linear. Multiply by the integrating

factor e2x and the equation becomes (ye2x ) = ex Integrate both sides and multiply
by e2x to get
y(x) = Ce2x + ex
2. Here, y = 3y. Separate variables and integrate to obtain
y(x) = Ce3x

3. The equation y + 8y = 0 is second-order, linear, with constant

coefficients. The
characteristic equation is m2 + 8 = 0 which has roots m = 8. Then

y(x) = A cos 8x + B sin 8x

4. The equation y xy = x2 y 2 is a Bernoulli equation. Make the substitution
w = 1/y and the equation turns into a linear equationw + xw = x2 . The
integrating factor is exp(x2 /2). Multiplying by the integrating factor gives
2

(wex

/2

) = x2 ex

/2

Integrating both sides from 0 to x gives

2

wex

/2

w(0) =

Then
2

w(x) = 1/y(x) = ex

/2

r2 er

/2

dr

0
2

w(0) ex

/2

r2 er

/2

dr

5. The equation x2 y 3xy +4y = 0 is a Cauchy-Euler equation. The characteristic

equation is m(m 1) 3m + 4 = 0 which has roots m = 2, 2. Thus
y(x) = ax2 + bx2 ln x

6. The equation y + x(y )2 = 0 does not have y appearing explicitly. So let v = y

to obtain
v + xv 2 = 0
We separate variables to get
1
1
= x2 + A
v
2
27

28

2. APPENDIX

Then
y(x) =

x
2
dx
arctan
+B
+B =
+A
2A
2A

x2 /2

7. The equation y + y + y = 0 is linear, second-order, with constant coefficients.

The characteristic equation is m2 + m + 1 = 0, which has roots m = 1/2 3i/2.

Thus

!
3x
3x
x/2
a cos
y(x) = e
+ b sin
2
2
8. In the equation yy (y )3 = 0 the independent variable does not appear
dv
. Then we get
explicitly. So let v = y which gives y = v dy
yv

dv
= v3
dy

Separating variables and solving gives v = 1/(C + ln y). Then

(C + ln y)dy = dx
Then
Cy + y ln y y = x + B
9. The equation 2x2 y +3xy y = 0 is a Cauchy-Euler equation with characteristic
equation 2m(m 1) + 3m 1 = 0. The roots are m = 1, 1/2. Then

b
y(x) = a x +
x

10. The equation y 3y 4y = 2 sin x is a linear, nonhomogeneous equation. The

homogeneous solution is yh (x) = ae4t +bet . Guess a particular solution of the form
yp (x) = A sin x + B cos x. Substitute into the equation to find A = 5/8, B = 3/8.
Then
5
3
y(x) = ae4t + bet + sin x cos x
8
8
11. The homogeneous solution of y + 4y = x sin x is yh (x) = a cos 2x + b sin 2x. A
particular solution can be found by undetermined coefficients or using the variation
of parameters formula from this appendix. We choose the latter. We have
Z x
cos 2s sin 2x cos 2x sin 2s
yp (x) =
s sin 2s ds
2
0
The calculation is left as an exercise. The solution is the sum of yh and yp .
12. The equation y 2xy = 1 is linear with integrating factor exp(x2 ). Multiplying by the integrating factor leads to
2

(yex ) = ex

2. APPENDIX

29

Integrating from 0 to x then gives

x2

y(x) = y(0)e

ex

r 2

dr

13. The second order linear equation y + 5y + 6y = 0 has characteristic equation

m2 + 5m + 6 = 0 with roots 3, 2. Hence
y(x) = ae3x + be2x

14. Separate variables to obtain

(1 + 3y 3 )dy = x2 dx
Integrating gives the implicit solution
1
3
y + y4 = + C
4
3