Math 462: HW3 Solutions

Due on August 1, 2014

Jacky Chong

Jacky Chong
Remark: We are working in the context of Riemann Integrals.

Problem 1
2.4.1 Solve the diffusion equation with the initial condition
(x) = 1

for

|x| < l

and

(x) = 0

for

|x| > l.

Write your answer in terms of Erf(x).

Solution: Consider the heat kernel
St (x) := S(x, t) =
and the initial condition

(
(x) =

2
1
ex /4kt
4kt

if |x| < l

if |x| 0

then the solution to the diffusion equation is given by

u(x, t) = St (x)
the convolution of S(x, t) and (x) in the spatial variable. Observe
Z +
2
1
e(xy) /4kt (y) dy
4kt
Z l
2
1

e(xy) /4kt dy
4kt l
Z (x+l)/4kt
2
1

ep dp
(xl)/4kt
Z (x+l)/4kt
Z (xl)/4kt
2
2
1
1

ep dp
ep dp
0
0




x+l
xl
1
1
Erf
Erf
.
2
2
4kt
4kt

St (x) =
=
=
=
=

Problem 2
2.4.16 Solve the diffusion equation with constant dissipation:
ut kuxx + bu = 0

for

<x<

with u(x, 0) = (x),

where b > 0 is a constant.

Solution: Consider the following diffusion equation
(

vt kvxx = 0

for t > 0

and < x <

v(x, 0) = (x)

Page 2 of 8

Jacky Chong

Problem 2

then we know that the solution is given by

+

1
v(x, t) = St (x) =
4kt

e(xy)

/4kt

(y) dy.

However, observe that

ebt
u(x, t) := ebt v(x, t) =
4kt

e(xy)

/4kt

(y) dy

solves the following problem

(
ut kuxx + bu = 0

and < x <

for t > 0

v(x, 0) = (x)
since

ut kuxx + bu = bebt v + ebt vt kebt vxx + bebt v

= ebt (vt kvxx ) = 0

and
u(x, 0) = v(x, 0) = (x).

Problem 3
2.4.18 Solve the heat equation with convection:
ut kuxx + V ux = 0

for

< x < with u(x, 0) = (x),

where V is a constant.
Solution: Consider the following diffusion equation
(

vt kvzz = 0

and < z <

for t > 0

v(z, 0) = (y)
then we know that the solution is given by
+

1
4kt

1
u(x, t) := v(x V t, t) =
4kt

v(z, t) = St (z) =

e(zy)

/4kt

(y) dy.

Define

e(xV ty)

/4kt

(y) dy

and observe that

ut = V vz + vt

and uxx = vzz .

Then it follows
ut kuxx + V ux = V vz + vt kvzz + V vz = vt kvzz = 0
and
u(x, 0) = v(x, 0) = (x)

Page 3 of 8

Jacky Chong

Problem 3

Problem 4
2.5.3 Let u satisfy the diffusion equation ut = 12 uxx . Let
1 2
v(x, t) = ex /2t u
t

x 1
,
t t


.

Show that v satisfies the backward diffusion equation vt = 21 vxx for t > 0.
Solution: Suppose u(z, ) satisfies the diffusion equation u 12 uzz = 0. Define
1 2
v(x, t) = ex /2t u
t
Then it follows



1 x2 /2t
x 1
e
,

u
t t
2t3/2


2
x
x 1
,
5/2 ex /2t uz

t t
t

vt (x, t) =

and
vx (x, t) =
and

x
t3/2

x2 /2t


u

x 1
,
t t


+

x 1
,
t t


.



x2 x2 /2t
x 1
e
,
u
t t
2t5/2


1 x2 /2t
x 1
e
u
,
,

t t
t5/2
1

t3/2

x2 /2t


uz

x 1
,
t t





x2 x2 /2t
x 1
x 1
1 x2 /2t
vxx (x, t) = 5/2 e
u
,
u
,
+ 3/2 e
t t
t t
t
t




1 x2 /2t
2x x2 /2t
x 1
x 1
,
+ 5/2 e
,
.
+ 5/2 e
uz
uzz
t t
t t
t
t

Hence it follows

1
vt + vxx = 0.
2

Problem 5
3.4.3 Solve utt = c2 uxx + cos x, u(x, 0) = sin x, ut (x, 0) = 1 + x.
Solution: Using the formula, we get that
u(x, t) =

1
1
[sin(x + ct) + sin(x ct)] +
2
2c

x+ct

(1 + s) ds +
xct
Z x+ct

1
2c

Z tZ

x+c(t )

cos s dsd
0

xc(t )

1
1
[sin(x + ct) + sin(x ct)] +
(1 + s) ds
2
2c xct
Z t
1
+
sin(x + c(t )) sin(x c(t )) d
2c 0
1
1
= [sin(x + ct) + sin(x ct)] + t(1 + x) 2 [cos(x + ct) + cos(x ct) 2 cos x]
2
2c
1
= sin x cos ct + t(1 + x) + 2 [cos x cos x cos ct]
c
1
= sin x cos ct + t(1 + x) + 2 cos x(1 cos ct)
c

Page 4 of 8

Jacky Chong

Problem 5

Problem 6
3.4.7 Let A be a positive-definite n n matrix. Let
S(t) =

X
(1)m A2m t2m+1
.
(2m + 1)!
m=0

(a) Show that this series of matrices converges uniformly for bounded t and its sum S(t) solves the problem
S 00 (t) + A2 S(t) = 0, S(0) = 0, S 0 (0) = I, where I is the identity matrix. Therefore, it makes sense to
denote S(t) as A1 sin tA and to denote its derivative S 0 (t) as cos(tA).
(b) Show that the solution to (13) is (14).
Solution:
(a) Let A = (aij ) be a n n matrix, then
v
uX
n
u n X
k A kF = t
|aij |2
i=1 j=1

the Frobenius norm of A. Suppose t < M < , then observe

l
l
2m
X (1)m A2m t2m+1
X
k A kF M 2m+1


.




(2m + 1)!
(2m + 1)!
m=k

Since the series

m=k

2m
X
k A kF M 2m+1
(2m + 1)!
m=0

converges by the ratio-test, then we know that for any  > 0 there exists N N such that
l
2m
X
k A kF M 2m+1
<
(2m + 1)!

m=k

provided k, l > N . In particular, for all k, l > N we see that

l

l
2m
X (1)m A2m t2m+1
X
k A kF M 2m+1


<




(2m + 1)!
(2m + 1)!
m=k

m=k

which means the partial sum

Sn (t) =

n
X
(1)m A2m t2m+1
(2m + 1)!
m=0

is uniformly Cauchy with respect to the Frobenius norm. Hence it follows Sn (t) S(t) uniformly.
By the uniform convergence of Sn (t) to S(t), then by Appendix A.2 we have that
S 0 (t) =
since

X
(1)m A2m t2m
2m!
m=0

X
X
(1)m A2m t2m
(1)m A2m t2m

2m!
2m!
m=0
m=0

Page 5 of 8

Jacky Chong

Problem 6

uniformly by an argument similar to the one given above. Now, observe S(0) = 0 and S 0 (0) = I. Finally,
S 00 (t) + A2 S(t) =
=

X
X
(1)m A2m+2 t2m+1
(1)m A2m t2m1
+
(2m 1)!
(2m + 1)!
m=0
m=1

X
X
(1)m+1 A2m+2 t2m+1
(1)m A2m+2 t2m+1
+
(2m + 1)!
(2m + 1)!
m=0
m=0

X
X
(1)m A2m+2 t2m+1
(1)m A2m+2 t2m+1
+
=0
(2m + 1)!
(2m + 1)!
m=0
m=0

(b) Consider
u(t) = cos(tA) + A1 sin(tA) +

A1 sin((t s)A)f (s) ds

then observe

Z t
d
S 0 (t s)f (s) ds
u(t) = S 00 (t) + S 0 (t) + S(0)f (t) +
dt
0
Z t
00
0
0
= S (t) + S (t) +
S (t s)f (s) ds
0

and

d2
u(t) =S 000 (t) + S 00 (t) + S 0 (0)f (t) +
dt2

S 00 (t s)f (s) ds
0
Z t
2
= A cos(tA) A sin(tA) + f (t)
A sin((t s)A)f (s) ds.
0

Hence it follows

d2
u(t) + A2 u(t) = f (t).
dt2

Problem 7
2
2
#1 If you know that the Fourier Transform of f (x) := ex is f() = e what is the Fourier Transform
2
2
of ex /2 ? What is the Fourier transform of e(x1) /2 ?

Solution: Since we know that

x2 () =
e\

ex e2ix dx = e ,

then observe
x2 /2 () =
e\

ex

/2 2ix

dx

2 #
x
=
exp
e2ix dx
2

h
i
Z + y2

= 2
e
exp 2iy( 2) dy
Z

=
=

"

y 2 ( 2)
2 e\
2e2

2 2

Page 6 of 8

Jacky Chong

Problem 7

Next, consider
+

\2 /2 () =
e(x1)

e(x1)

Z +

ey

/2 2ix

dx

/2 2i(y+1)

dy

=e
=

2i

ey

/2 2iy

dy

2e2

2 2

2i

Problem 8
#2 Show that for t > 0
Z

1 x
f
dx =
t
t

f (x) dx.

Solution: U-substitution.

Problem 9
#3 Compute the Fourier Transform of the Pi function defined as follows: (x) = 1 if |x| < 1 and it is zero
otherwise. (Its graph looks like a ). What is the Fourier Transform of (x/t).
Solution: Using the definition, we have that
+

(x)e2ix dx

[
(x)()
=

Z 1

e2ix dx

=
=

1
2i

e2i
sin 2
=
.
2i

Suppose t > 0, then we see that

+

(x/t)e2ix dx

\
(x/t)()
=

Z
=t

(y)e2iyt dy

[
= t(y)(t)
sin 2t
sin 2t
=t
=
.
t

Page 7 of 8

Jacky Chong

Problem 9

Suppose t < 0, then we see that

Z

(x/t)e2ix dx

\
(x/t)()
=

=t

(y)e2iyt dy

(y)e2iyt dy

= t

[
= t(y)(t)
sin 2t
sin 2t
= t
=
.
t

Hence it follows

sin 2|t|
\
(x/t)()
=
.

Problem 10
#4 Use FT to compute the convolution
+

e(xy) ey dy.

Solution: Let f (x) = ex , then we see that

Z

e(xy) ey dy = f f (x).

Applying the multiplicative property of convolutions under Fourier Transform, we get that
f[
f () = f()f().
Observe
Z

ex e2ix dx

"
2 #

Z +
x
e2ix dx
=
exp

Z +



2
=
ey exp 2iy( ) dy

x2 () =
ed

y2

2 2
= e\
( ) = e ,
then it follows
2 2
f()f() = e2 =

2 2

2e2 =
2

\
ex2 /2 ().
2

Taking the inversion, we get that

r
f f (x) =

x2 /2
e
.
2

Page 8 of 8