Sie sind auf Seite 1von 396

Partial Dierential Equations: Graduate Level Problems

and Solutions
Igor Yanovsky

Partial Dierential Equations

Igor Yanovsky, 2005

Disclaimer: This handbook is intended to assist graduate students with qualifying


examination preparation. Please be aware, however, that the handbook might contain,
and almost certainly contains, typos as well as incorrect or inaccurate solutions. I can
not be made responsible for any inaccuracies contained in this handbook.

Partial Dierential Equations

Igor Yanovsky, 2005

Contents
1 Trigonometric Identities

2 Simple Eigenvalue Problem

3 Separation of Variables:
Quick Guide

4 Eigenvalues of the Laplacian:

Quick Guide

5 First-Order Equations
5.1 Quasilinear Equations . . . . . . . . . . . . . . .
5.2 Weak Solutions for Quasilinear Equations . . . .
5.2.1 Conservation Laws and Jump Conditions
5.2.2 Fans and Rarefaction Waves . . . . . . . .
5.3 General Nonlinear Equations . . . . . . . . . . .
5.3.1 Two Spatial Dimensions . . . . . . . . . .
5.3.2 Three Spatial Dimensions . . . . . . . . .

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

10
10
12
12
12
13
13
13

6 Second-Order Equations
14
6.1 Classication by Characteristics . . . . . . . . . . . . . . . . . . . . . . . 14
6.2 Canonical Forms and General Solutions . . . . . . . . . . . . . . . . . . 14
6.3 Well-Posedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
7 Wave Equation
7.1 The Initial Value Problem . . . . . . . . . .
7.2 Weak Solutions . . . . . . . . . . . . . . . .
7.3 Initial/Boundary Value Problem . . . . . .
7.4 Duhamels Principle . . . . . . . . . . . . .
7.5 The Nonhomogeneous Equation . . . . . . .
7.6 Higher Dimensions . . . . . . . . . . . . . .
7.6.1 Spherical Means . . . . . . . . . . .
7.6.2 Application to the Cauchy Problem
7.6.3 Three-Dimensional Wave Equation .
7.6.4 Two-Dimensional Wave Equation . .
7.6.5 Huygens Principle . . . . . . . . . .
7.7 Energy Methods . . . . . . . . . . . . . . .
7.8 Contraction Mapping Principle . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

8 Laplace Equation
8.1 Greens Formulas . . . . . . . . . . . . . . . . . . . . .
8.2 Polar Coordinates . . . . . . . . . . . . . . . . . . . .
8.3 Polar Laplacian in R2 for Radial Functions . . . . . .
8.4 Spherical Laplacian in R3 and Rn for Radial Functions
8.5 Cylindrical Laplacian in R3 for Radial Functions . . .
8.6 Mean Value Theorem . . . . . . . . . . . . . . . . . . .
8.7 Maximum Principle . . . . . . . . . . . . . . . . . . .
8.8 The Fundamental Solution . . . . . . . . . . . . . . . .
8.9 Representation Theorem . . . . . . . . . . . . . . . . .
8.10 Greens Function and the Poisson Kernel . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

23
23
24
24
24
24
26
26
26
27
28
28
29
30

.
.
.
.
.
.
.
.
.
.

31
31
32
32
32
33
33
33
34
37
42

Partial Dierential Equations

Igor Yanovsky, 2005

8.11 Properties of Harmonic Functions . . . . . . . . . . . . . . . . . . . . . .


8.12 Eigenvalues of the Laplacian . . . . . . . . . . . . . . . . . . . . . . . . .

44
44

9 Heat Equation
45
9.1 The Pure Initial Value Problem . . . . . . . . . . . . . . . . . . . . . . . 45
9.1.1 Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . 45
9.1.2 Multi-Index Notation . . . . . . . . . . . . . . . . . . . . . . . . 45
9.1.3 Solution of the Pure Initial Value Problem . . . . . . . . . . . . . 49
9.1.4 Nonhomogeneous Equation . . . . . . . . . . . . . . . . . . . . . 50
9.1.5 Nonhomogeneous Equation with Nonhomogeneous Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
9.1.6 The Fundamental Solution . . . . . . . . . . . . . . . . . . . . . 50
10 Schr
odinger Equation

52

11 Problems: Quasilinear Equations

54

12 Problems: Shocks

75

13 Problems: General Nonlinear Equations


86
13.1 Two Spatial Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
13.2 Three Spatial Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . 93
14 Problems: First-Order Systems

102

15 Problems: Gas Dynamics Systems


15.1 Perturbation . . . . . . . . . . . .
15.2 Stationary Solutions . . . . . . . .
15.3 Periodic Solutions . . . . . . . . .
15.4 Energy Estimates . . . . . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

127
127
128
130
136

16 Problems: Wave Equation


16.1 The Initial Value Problem . . . .
16.2 Initial/Boundary Value Problem
16.3 Similarity Solutions . . . . . . . .
16.4 Traveling Wave Solutions . . . .
16.5 Dispersion . . . . . . . . . . . . .
16.6 Energy Methods . . . . . . . . .
16.7 Wave Equation in 2D and 3D . .

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

139
139
141
155
156
171
174
187

.
.
.
.
.
.
.
.

196
196
205
216
221
223
232
242
249

.
.
.
.
.
.
.

17 Problems: Laplace Equation


17.1 Greens Function and the Poisson Kernel . . .
17.2 The Fundamental Solution . . . . . . . . . . .
17.3 Radial Variables . . . . . . . . . . . . . . . .
17.4 Weak Solutions . . . . . . . . . . . . . . . . .
17.5 Uniqueness . . . . . . . . . . . . . . . . . . .
17.6 Self-Adjoint Operators . . . . . . . . . . . . .
17.7 Spherical Means . . . . . . . . . . . . . . . .
17.8 Harmonic Extensions, Subharmonic Functions

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

Partial Dierential Equations

Igor Yanovsky, 2005

18 Problems: Heat Equation


255
18.1 Heat Equation with Lower Order Terms . . . . . . . . . . . . . . . . . . 263
18.1.1 Heat Equation Energy Estimates . . . . . . . . . . . . . . . . . . 264
19 Contraction Mapping and Uniqueness - Wave

271

20 Contraction Mapping and Uniqueness - Heat

273

21 Problems: Maximum Principle - Laplace and Heat


279
21.1 Heat Equation - Maximum Principle and Uniqueness . . . . . . . . . . . 279
21.2 Laplace Equation - Maximum Principle . . . . . . . . . . . . . . . . . . 281
22 Problems: Separation of Variables - Laplace Equation

282

23 Problems: Separation of Variables - Poisson Equation

302

24 Problems: Separation of Variables - Wave Equation

305

25 Problems: Separation of Variables - Heat Equation

309

26 Problems: Eigenvalues of the Laplacian - Laplace

323

27 Problems: Eigenvalues of the Laplacian - Poisson

333

28 Problems: Eigenvalues of the Laplacian - Wave

338

29 Problems: Eigenvalues of the Laplacian - Heat


346
29.1 Heat Equation with Periodic Boundary Conditions in 2D
(with extra terms) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360
30 Problems: Fourier Transform

365

31 Laplace Transform

385

32 Linear Functional Analysis


32.1 Norms . . . . . . . . . . . .
32.2 Banach and Hilbert Spaces
32.3 Cauchy-Schwarz Inequality
32.4 H
older Inequality . . . . . .
32.5 Minkowski Inequality . . . .
32.6 Sobolev Spaces . . . . . . .

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

393
393
393
393
393
394
394

Partial Dierential Equations

Trigonometric Identities
cos(a + b) = cos a cos b sin a sin b
cos(a b) = cos a cos b + sin a sin b
sin(a + b) = sin a cos b + cos a sin b
sin(a b) = sin a cos b cos a sin b

cos a cos b =
sin a cos b =
sin a sin b =

cos(a + b) + cos(a b)
2
sin(a + b) + sin(a b)
2
cos(a b) cos(a + b)
2

cos 2t = cos2 t sin2 t

Igor Yanovsky, 2005


0
mx
nx
cos
dx =
cos
L
L
L
L

 L
0
mx
nx
sin
dx =
sin
L
L
L
L
 L
mx
nx
cos
dx = 0
sin
L
L
L

L
0

nx
mx
cos
cos
dx =
L
L

L
0

mx
nx
sin
dx =
sin
L
L

sin 2t = 2 sin t cos t



1 + cos t
2
1 cos t
2

1
cos2 t =
2
1
sin2 t =
2

einx eimx dx =
0


0

cot t + 1 = csc t

sin x =

eix + eix
2
eix eix
2i

cosh x =
sinh x =

ex + ex
2
ex ex
2

d
cosh x = sinh(x)
dx
d
sinh x = cosh(x)
dx
cosh2 x sinh2 x = 1


du
+ u2

du

a2 u2
a2

1
u
tan1 + C
a
a
u
= sin1 + C
a
=


0

n = m
n=m
n = m
n=m

L
2

n = m
n=m

L
2

n = m
n=m


0

0
L

n = m
n=m

0
L

n = 0
n=0

x sin x cos x

2
2

x sin x cos x
cos2 x dx = +
2
2

tan2 x dx = tan x x

cos2 x
sin x cos x dx =
2

cos x =

einx dx =

1 + tan2 t = sec2 t
2

sin2 x dx =

ln(xy) = ln(x) + ln(y)


x
ln = ln(x) ln(y)
y
ln xr = r lnx

ln x dx = x ln x x

x ln x dx =



z2

ez dz =

x2
x2
ln x
2
4

e 2 dz =

Partial Dierential Equations


A=

a b
c d


,

1
=
det(A)

Igor Yanovsky, 2005

d b
c a

Partial Dierential Equations

Igor Yanovsky, 2005

Simple Eigenvalue Problem


X  + X = 0
Boundary conditions

Eigenvalues n
 n 2
 L1 2

X(0) = X(L) = 0

Eigenfunctions Xn

(n 2 )
L


(n 12 ) 2
L
 n 2
L 
 2n
2
L

X(0) = X  (L) = 0
X (0) = X(L) = 0
X (0) = X  (L) = 0
X(0) = X(L), X  (0) = X  (L)
X(L) = X(L), X  (L) = X  (L)

 n 2
L

sin n
L x

(n 1 )
sin L2 x
(n 1 )
cos L2 x
cos n
L x
2n
sin L x
cos 2n
L x
n
sin L x
cos n
L x

n = 1, 2, . . .
n = 1, 2, . . .
n = 1, 2, . . .
n = 0, 1, 2, . . .
n = 1, 2, . . .
n = 0, 1, 2, . . .
n = 1, 2, . . .
n = 0, 1, 2, . . .

X  X = 0
Boundary conditions
X(0) = X(L) = 0, X  (0) = X  (L) = 0
X  (0) = X (L) = 0, X (0) = X (L) = 0

Eigenvalues n
 n 4
L 
 n
4
L

Eigenfunctions Xn
sin n
L x
cos n
L x

n = 1, 2, . . .
n = 0, 1, 2, . . .

Partial Dierential Equations

Igor Yanovsky, 2005

Separation of Variables:
Quick Guide
u = 0.

Laplace Equation:
X (x)

Eigenvalues of the Laplacian: Quick Guide

Laplace Equation:

 (y)

Y
= .
X(x)
Y (y)
X  + X = 0.
=

uxx + uyy + u = 0.

X  Y 
+
+ = 0. ( = 2 + 2 )
X
Y
Y  + 2 Y = 0.
X  + 2 X = 0,

X (t)
Y  ()
=
= .
X(t)
Y ()
Y  () + Y () = 0.
uxx + uyy + k2 u = 0.
utt uxx = 0.

Wave Equation:

T  (t)
X (x)
=
= .
X(x)
T (t)
X  + X = 0.

Y 
X 
=
+ k 2 = c2 .
X
Y
X  + c2 X = 0,

Y  + (k2 c2 )Y = 0.

utt + 3ut + u = uxx .


T
X 
T 
+3 +1 =
= .
T
T
X
X  + X = 0.
X 
T 
+1 =
= .
T
X
X  + X = 0.
utt + ut = c2 uxx + uxxt,

( > 0)

X 
= ,
X

T
T  X 
1 T 
+
=
1
+
.
c2 T
c2 T
c2 T X
4th Order: utt = k uxxxx.
1 T 
X 
=
= .
X
k T
X  X = 0.

Heat Equation:
T
T

= k

ut = kuxx .
X 
= .
X

X = 0.
k
ut = uxxxx .

X  +
4th Order:

uxx + uyy + k2 u = 0.
X 
Y 
=
+ k 2 = c2 .
Y
X
Y  + c2 Y = 0,

utt uxx + u = 0.

X 
T
=
= .
T
X
X  X = 0.

X  + (k2 c2 )X = 0.

Partial Dierential Equations

Igor Yanovsky, 2005

10

First-Order Equations

5.1

Quasilinear Equations

Consider the Cauchy problem for the quasilinear equation in two variables
a(x, y, u)ux + b(x, y, u)uy = c(x, y, u),
with parameterized by (f (s), g(s), h(s)). The characteristic equations are
dy
= b(x, y, z),
dt

dx
= a(x, y, z),
dt

dz
= c(x, y, z),
dt

with initial conditions


x(s, 0) = f (s),

y(s, 0) = g(s),

z(s, 0) = h(s).

dy
In a quasilinear case, the characteristic equations for dx
dt and dt need not decouple from
the dz
dt equation; this means that we must take the z values into account even to nd
the projected characteristic curves in the xy-plane. In particular, this allows for the
possibility that the projected characteristics may cross each other.
The condition for solving for s and t in terms of x and y requires that the Jacobian
matrix be nonsingular:


xs ys
= xs yt ys xt = 0.
J
xt yt

In particular, at t = 0 we obtain the condition


f  (s) b(f (s), g(s), h(s)) g  (s) a(f (s), g(s), h(s)) = 0.
Burgers Equation. Solve the Cauchy problem

ut + uux = 0,
u(x, 0) = h(x).

(5.1)

The characteristic equations are


dy
dz
dx
= z,
= 1,
= 0,
dt
dt
dt
and may be parametrized by (s, 0, h(s)).
x = h(s)t + s, y = t, z = h(s).
u(x, y) = h(x uy)

(5.2)

The characteristic projection in the xt-plane1 passing through the point (s, 0) is the
line
x = h(s)t + s
along which u has the constant value u = h(s). Two characteristics x = h(s1 )t + s1
and x = h(s2 )t + s2 intersect at a point (x, t) with
t=
1

s2 s1
.
h(s2 ) h(s1 )

y and t are interchanged here

Partial Dierential Equations

Igor Yanovsky, 2005

11

From (5.2), we have


ux = h (s)(1 ux t)

ux =

h (s)
1 + h (s)t

Hence for h (s) < 0, ux becomes innite at the positive time


t=

1
.
h (s)

The smallest t for which this happens corresponds to the value s = s0 at which h (s)
has a minimum (i.e.h (s) has a maximum). At time T = 1/h (s0 ) the solution u
experiences a gradient catastrophe.

Partial Dierential Equations

5.2
5.2.1

Igor Yanovsky, 2005

12

Weak Solutions for Quasilinear Equations


Conservation Laws and Jump Conditions

Consider shocks for an equation


ut + f (u)x = 0,

(5.3)

where f is a smooth function of u. If we integrate (5.3) with respect to x for a x b,


we obtain

d b
u(x, t) dx + f (u(b, t)) f (u(a, t)) = 0.
(5.4)
dt a
This is an example of a conservation law. Notice that (5.4) implies (5.3) if u is C 1 , but
(5.4) makes sense for more general u.
Consider a solution of (5.4) that, for xed t, has a jump discontinuity at x = (t).
We assume that u, ux , and ut are continuous up to . Also, we assume that (t) is C 1
in t.
Taking a < (t) < b in (5.4), we obtain

 b

d

u dx +
u dx + f (u(b, t)) f (u(a, t))
dt a


 b
ut (x, t) dx +
ut(x, t) dx
=  (t)ul ((t), t)  (t)ur ((t), t) +
a

+ f (u(b, t)) f (u(a, t)) = 0,


where ul and ur denote the limiting values of u from the left and right sides of the shock.
Letting a (t) and b (t), we get the Rankine-Hugoniot jump condition:
 (t)(ul ur ) + f (ur ) f (ul ) = 0,
 (t) =
5.2.2

f (ur ) f (ul )
.
ur ul

Fans and Rarefaction Waves

For Burgers equation

1
u2 = 0,
ut +
2
x

x x


x x
=
u

= .

we have f  (u) = u, f  u
t
t
t
t
For a rarefaction fan emanating from (s, 0) on xt-plane, we have:

xs


ul ,
t f (ul ) = ul ,
u(x, t) = xs
ul xs
t ,
t ur ,

xs

ur ,
t f (ur ) = ur .

Partial Dierential Equations

5.3

Igor Yanovsky, 2005

13

General Nonlinear Equations

5.3.1

Two Spatial Dimensions

Write a general nonlinear equation F (x, y, u, ux, uy ) = 0 as


F (x, y, z, p, q) = 0.
is parameterized by


: f (s) , g(s) , h(s) , (s) , (s)
    
x(s,0) y(s,0) z(s,0) p(s,0) q(s,0)

We need to complete to a strip. Find (s) and (s), the initial conditions for p(s, t)
and q(s, t), respectively:

F (f (s), g(s), h(s), (s), (s)) = 0

h (s) = (s)f  (s) + (s)g (s)

The characteristic equations are


dy
dx
= Fp
= Fq
dt
dt
dz
= pFp + qFq
dt
dq
dp
= Fx Fz p
= Fy Fz q
dt
dt
We need to have the Jacobian condition. That is, in order to solve the Cauchy problem
in a neighborhood of , the following condition must be satised:
f  (s) Fq [f, g, h, , ](s) g  (s) Fp [f, g, h, , ](s) = 0.
5.3.2

Three Spatial Dimensions

Write a general nonlinear equation F (x1 , x2 , x3 , u, ux1 , ux2 , ux3 ) = 0 as


F (x1 , x2 , x3, z, p1 , p2, p3 ) = 0.
is parameterized by


: f1 (s1 , s2 ), f2 (s1 , s2 ), f3 (s1 , s2 ), h(s1 , s2 ), 1 (s1 , s2 ), 2 (s1 , s2 ), 3 (s1 , s2 )
                    
x1 (s1 ,s2 ,0) x2 (s1 ,s2 ,0) x3 (s1 ,s2 ,0) z(s1 ,s2 ,0)

p1 (s1 ,s2 ,0)

p2 (s1 ,s2 ,0)

p3 (s1 ,s2 ,0)

We need to complete to a strip. Find 1 (s1 , s2 ), 2 (s1 , s2 ), and 3 (s1 , s2 ), the initial
conditions for p1 (s1 , s2 , t), p2 (s1 , s2 , t), and p3 (s1 , s2 , t), respectively:



F f1 (s1 , s2 ), f2 (s1 , s2 ), f3 (s1 , s2 ), h(s1 , s2 ), 1 , 2 , 3 = 0


f1
f2
f3
h
= 1
+ 2
+ 3

s1
s1
s1
s1
h
f1
f2
f3

= 1
+ 2
+ 3
s2
s2
s2
s2
The characteristic equations are
dx2
dx3
dx1
= Fp1
= Fp2
= Fp3
dt
dt
dt
dz
= p1 Fp1 + p2 Fp2 + p3 Fp3
dt
dp1
dp2
dp3
= Fx1 p1 Fz
= Fx2 p2 Fz
= Fx3 p3 Fz
dt
dt
dt

Partial Dierential Equations

6
6.1

Igor Yanovsky, 2005

14

Second-Order Equations
Classification by Characteristics

Consider the second-order equation in which the derivatives of second-order all occur
linearly, with coecients only depending on the independent variables:
a(x, y)uxx + b(x, y)uxy + c(x, y)uyy = d(x, y, u, ux, uy ).

(6.1)

The characteristic equation is

b b2 4ac
dy
=
.
dx
2a
b2 4ac > 0 two characteristics, and (6.1) is called hyperbolic;
b2 4ac = 0 one characteristic, and (6.1) is called parabolic;
b2 4ac < 0 no characteristics, and (6.1) is called elliptic.
These denitions are all taken at a point x0 R2 ; unless a, b, and c are all constant,
the type may change with the point x0 .

6.2

Canonical Forms and General Solutions

uxx uyy = 0 is hyperbolic (one-dimensional wave equation).


uxx uy = 0 is parabolic (one-dimensional heat equation).
uxx + uyy = 0 is elliptic (two-dimensional Laplace equation).
By the introduction of new coordinates and in place of x and y, the equation
(6.1) may be transformed so that its principal part takes the form , , or .
If (6.1) is hyperbolic, parabolic, or elliptic, there exists a change of variables (x, y) and
(x, y) under which (6.1) becomes, respectively,
, u, u, u )
u = d(,
, u, u, u ),
u = d(,

x, y, u, ux , uy),
uxx uyy = d(

, u, u, u ).
u + u = d(,
Example 1. Reduce to canonical form and nd the general solution:
uxx + 5uxy + 6uyy = 0.
Proof. a = 1, b = 5, c = 6 b2 4ac = 1 > 0
characteristics.
The characteristics are found by solving

51
3
dy
=
=
dx
2
2
to nd y = 3x + c1 and y = 2x + c2 .

(6.2)

hyperbolic two

Partial Dierential Equations

Igor Yanovsky, 2005

15

(x, y) = 3x y, (x, y) = 2x y.

Let

x = 3,

x = 2,

y = 1,

y = 1.

u = u((x, y), (x, y));


ux = u x + u x = 3u + 2u ,
uy = u y + u y = u u ,
uxx = (3u + 2u )x = 3(u x + u x) + 2(u x + u x) = 9u + 12u + 4u ,
uxy = (3u + 2u )y = 3(u y + u y ) + 2(uy + u y ) = 3u 5u 2u ,
uyy = (u + u )y = (u y + u y + u y + u y ) = u + 2u + u .
Inserting these expressions into (6.2) and simplifying, we obtain
u = 0,

which is the Canonical form,

u = f (),
u = F () + G(),
u(x, y) = F (3x y) + G(2x y),

General solution.

Example 2. Reduce to canonical form and nd the general solution:


y 2 uxx 2yuxy + uyy = ux + 6y.

(6.3)

Proof. a = y 2 , b = 2y, c = 1 b2 4ac = 0 parabolic one characteristic.


The characteristics are found by solving
2y
1
dy
=
=
2
dx
2y
y
y2
+ c = x.
to nd
2
2

Let = y2 + x. We must choose a second constant function (x, y) so that is not


parallel to . Choose (x, y) = y.
x = 1,

x = 0,

y = y,

y = 1.

u = u((x, y), (x, y));


ux = u x + u x = u ,
uy = u y + u y = yu + u ,
uxx = (u )x = u x + u x = u ,
uxy = (u )y = u y + u y = yu + u ,
uyy

= (yu + u )y = u + y(u y + u y ) + (u y + u y )
= u + y 2 u + 2yu + u .

Partial Dierential Equations

Igor Yanovsky, 2005

Inserting these expressions into (6.3) and simplifying, we obtain


u = 6y,
u = 6,

which is the Canonical form,

u = 3 + f (),
u = 3 + f () + g(),

y2

y2
+x +g
+x ,
u(x, y) = y 3 + y f
2
2

General solution.

16

Partial Dierential Equations

Igor Yanovsky, 2005

17

Problem (F03, #4). Find the characteristics of the partial dierential equation
xuxx + (x y)uxy yuyy = 0,

x > 0, y > 0,

(6.4)

and then show that it can be transformed into the canonical form
( 2 + 4)u + u = 0
whence and are suitably chosen canonical coordinates. Use this to obtain the general
solution in the form

g( ) d 
u(, ) = f () +
1
( 2 + 4  ) 2
where f and g are arbitrary functions of and .
Proof. a = x, b = x y, c = y b2 4ac = (x y)2 + 4xy > 0 for x > 0,
y > 0 hyperbolic two characteristics.
The characteristics are found by solving



2x
dy
b b2 4ac
x y (x y)2 + 4xy
x y (x + y)
2x = 1
=
=
=
=
2y
dx
2a
2x
2x
= xy
2x

dx
dy
= ,
y
x
ln y = ln x1 + c2 ,
c2
y= .
x

y = x + c1 ,

Let = x y and = xy
x = 1,

x = y,

y = 1,

y = x.

u = u((x, y), (x, y));


ux = u x + u x = u + yu ,
uy = u y + u y = u + xu ,
uxx = (u + yu )x = u x + u x + y(ux + u x ) = u + 2yu + y 2 u ,
uxy = (u + yu )y = u y + u y + u + y(uy + u y ) = u + xu + u yu + xyu ,
uyy = (u + xu )y = u y u y + x(u y + u y ) = u 2xu + x2 u ,
Inserting these expressions into (6.4), we obtain
x(u + 2yu + y 2 u ) + (x y)(u + xu + u yu + xyu ) y(u 2xu + x2 u ) = 0,
(x2 + 2xy + y 2 )u + (x y)u = 0,


(x y)2 + 4xy u + (x y)u = 0,
(2 + 4)u + u = 0,

which is the Canonical form.

Partial Dierential Equations

Igor Yanovsky, 2005

We need to integrate twice to get the general solution:


(2 + 4)(u) + u = 0,



(u )
d,
d =
u
2 + 4
1
ln u = ln (2 + 4) + g(),
2
1
ln u = ln (2 + 4) 2 + g(),
g()
u =
1,
(2 + 4) 2

g() d
u(, ) = f () +
1 ,
(2 + 4) 2

General solution.

18

Partial Dierential Equations

6.3

Igor Yanovsky, 2005

19

Well-Posedness

Problem (S99, #2). In R2 consider the unit square dened by 0 x, y 1.


Consider
a)

ux + uyy = 0;

b)

uxx + uyy = 0;

c)

uxx uyy = 0.

Prescribe data for each problem separately on the boundary of so that each of these
problems is well-posed. Justify your answers.
Proof. The initial / boundary value problem for the HEAT EQUATION is wellposed:

x , t > 0,
ut = u
u(x, 0) = g(x)
x ,

u(x, t) = 0
x , t > 0.
Existence - by eigenfunction expansion.
Uniqueness and continuous dependence on the data by maximum principle.
The method of eigenfunction expansion and maximum
principle give well-posedness for more general problems:

x , t > 0,

ut = u + f (x, t)
u(x, 0) = g(x)
x ,

u(x, t) = h(x, t)
x , t > 0.
It is also possible to replace the Dirichlet boundary condition u(x, t) = h(x, t) by a
Neumann or Robin condition, provided we replace n , n by the eigenvalues and eigenfunctions for the appropriate boundary value problem.
a) Relabel the variables (x t, y x).
We have the BACKWARDS HEAT EQUATION:
ut + uxx = 0.
Need to dene initial conditions u(x, 1) = g(x), and
either Dirichlet, Neumann, or Robin boundary conditions.
b) The solution to the LAPLACE EQUATION

u = 0
in ,
u=g
on
exists if g is continuous on , by Perrons method. Maximum principle gives uniqueness.
To show the continuous dependence on the data, assume


u2 = 0
in ,
in ,
u1 = 0
on ;
on .
u1 = g 1
u2 = g 2

Partial Dierential Equations

Igor Yanovsky, 2005

20

Then (u1 u2 ) = 0 in . Maximum principle gives


max(u1 u2 ) = max(g1 g2 ).

Thus,

max |u1 u2 | = max |g1 g2 |.

Thus, |u1 u2 | is bounded by |g1 g2 |, i.e. continuous dependence on data.


Perrons method gives existence of the solution to the POISSON EQUATION

u = f
in ,
u
on
n = h


for
 f C () and h C (), satisfying the compatibility condition h dS =
f dx. It is unique up to an additive constant.
c) Relabel the variables (y t).
The solution to the WAVE EQUATION
utt uxx = 0,
is of the form u(x, y) = F (x + t) + G(x t).
The existence of the solution to the initial/boundary value problem

0 < x < 1, t > 0


utt uxx = 0
0<x<1
u(x, 0) = g(x), ut(x, 0) = h(x)

u(0, t) = (t),
u(1, t) = (t)
t 0.
is given by the method of separation of variables
(expansion in eigenfunctions)
and by the parallelogram rule.
Uniqueness is given by the energy method.
Need initial conditions u(x, 0), ut (x, 0).
Prescribe u or ux for each of the two boundaries.

Partial Dierential Equations

Igor Yanovsky, 2005

21

Problem (F95, #7). Let a, b be real numbers. The PDE


uy + auxx + buyy = 0
is to be solved in the box = [0, 1]2.
Find data, given on an appropriate part of , that will make this a well-posed problem.
Cover all cases according to the possible values of a and b. Justify your statements.
Proof.
ab < 0 two sets of characteristics hyperbolic.
Relabeling the variables (y t), we have
a
1
utt + uxx = ut .
b
b
The solution of the
equation
is ofthe form
 a
u(x, t) = F (x + b t) + G(x ab t).
Existence of the solution to the initial/boundary
value problem is given by the method of separation
of variables (expansion in eigenfunctions)
and by the parallelogram rule.
Uniqueness is given by the energy method.
Need initial conditions u(x, 0), ut (x, 0).
Prescribe u or ux for each of the two boundaries.
ab > 0 no characteristics elliptic.
The solution to the Laplace equation with boundary conditions u = g on exists
if g is continuous on , by Perrons method.
To show uniqueness, we use maximum principle. Assume there are two solutions u1
and u2 with with u1 = g(x), u2 = g(x) on . By maximum principle
max(u1 u2 ) = max(g(x) g(x)) = 0.

Thus, u1 = u2 .

ab = 0 one set of characteristics parabolic.


a = b = 0. We have uy = 0, a rst-order ODE.
u must be specied on y = 0, i.e. x -axis.
a = 0, b = 0. We have uy + buyy = 0, a second-order ODE.
u and uy must be specied on y = 0, i.e. x -axis.
a > 0, b = 0. We have a Backwards Heat Equation.
ut = auxx .
Need to dene initial conditions u(x, 1) = g(x), and
either Dirichlet, Neumann, or Robin boundary conditions.

Partial Dierential Equations

Igor Yanovsky, 2005

a < 0, b = 0. We have a Heat Equation.


ut = auxx .
The initial / boundary value problem for the heat equation is well-posed:

x , t > 0,

ut = u
u(x, 0) = g(x)
x ,

u(x, t) = 0
x , t > 0.
Existence - by eigenfunction expansion.
Uniqueness and continuous dependence on the data by maximum principle.

22

Partial Dierential Equations

Igor Yanovsky, 2005

23

Wave Equation

The one-dimensional wave equation is


utt c2 uxx = 0.

(7.1)

The characteristic equation with a = c2 , b = 0, c = 1 would be

b b2 4ac
1
4c2
dt
=
=
= ,
2
dx
2a
2c
c
and thus
1
t = x + c1
c
= x + ct

and

1
x + c2 ,
c
= x ct,

t=

which transforms (7.1) to


u = 0.

(7.2)

The general solution of (7.2) is u(, ) = F ()+G(), where F and G are C 1 functions.
Returning to the variables x, t we nd that
u(x, t) = F (x + ct) + G(x ct)

(7.3)

solves (7.1). Moreover, u is C 2 provided that F and G are C 2 .


If F 0, then u has constant values along the lines x ct = const, so may be described
as a wave moving in the positive x-direction with speed dx/dt = c; if G 0, then u is
a wave moving in the negative x-direction with speed c.

7.1

The Initial Value Problem

For an initial value problem, consider the Cauchy problem



utt c2 uxx = 0,
u(x, 0) = g(x), ut(x, 0) = h(x).

(7.4)

Using (7.3) and (7.4), we nd that F and G satisfy


F (x) + G(x) = g(x),

cF  (x) cG (x) = h(x).

(7.5)

x
If we integrate the second equation in (7.5), we get cF (x) cG(x) = 0 h() d + C.
Combining this with the rst equation in (7.5), we can solve for F and G to nd

x
1
h() d + C1
F (x) = 12 g(x) + 2c
0x
1
1
G(x) = 2 g(x) 2c 0 h() d C1 ,
Using these expressions in (7.3), we obtain dAlemberts Formula for the solution
of the initial value problem (7.4):
1
1
u(x, t) = (g(x + ct) + g(x ct)) +
2
2c

x+ct

h() d.
xct

If g C 2 and h C 1 , then dAlemberts Formula denes a C 2 solution of (7.4).

Partial Dierential Equations

7.2

Igor Yanovsky, 2005

24

Weak Solutions

Equation (7.3) denes a weak solution of (7.1) when F and G are not C 2 functions.
Consider the parallelogram with sides that are
segments of characteristics. Since
u(x, t) = F (x + ct) + G(x ct), we have
u(A) + u(C) =
= F (k1 ) + G(k3) + F (k2 ) + G(k4 )
= u(B) + u(D),
which is the parallelogram rule.

7.3

Initial/Boundary Value Problem

0 < x < L, t > 0


utt c uxx = 0
0<x<L
u(x, 0) = g(x), ut (x, 0) = h(x)

u(0, t) = (t), u(L, t) = (t)


t 0.

(7.6)

Use separation of variables to obtain an expansion in eigenfunctions. Find u(x, t) in


the form

u(x, t) =

nx
nx
a0 (t) 
+
+ bn (t) sin
.
an (t) cos
2
L
L
n=1

7.4

Duhamels Principle

utt c uxx = f (x, t)


u(x, 0) = 0

ut (x, 0) = 0.



an + n an = fn (t)
an (0) = 0


an (0) = 0

7.5

Utt c Uxx = 0

U (x, 0, s) = 0

Ut (x, 0, s) = f (x, s)



an + n an = 0

an (0, s) = 0


an (0, s) = fn (s)

u(x, t) =

U (x, ts, s) ds.


0


an (t) =

an (t s, s) ds.

The Nonhomogeneous Equation

Consider the nonhomogeneous wave equation with homogeneous initial conditions:



utt c2 uxx = f (x, t),
(7.7)
u(x, 0) = 0, ut(x, 0) = 0.
Duhamels Principle provides the solution of (7.7):
1
u(x, t) =
2c

 t

x+c(ts)


f (, s) d ds.

xc(ts)

If f (x, t) is C 1 in x and C 0 in t, then Duhamels Principle provides a C 2 solution of


(7.7).

Partial Dierential Equations

Igor Yanovsky, 2005

We can solve (7.7) with nonhomogeneous initial conditions,



utt c2 uxx = f (x, t),
u(x, 0) = g(x), ut(x, 0) = h(x),

25

(7.8)

by adding together dAlemberts formula and Duhamels principle gives the solution:
u(x, t) =

1
1
(g(x + ct) + g(x ct)) +
2
2c

x+ct

h() d +
xct

1
2c

 t
0

x+c(ts)

xc(ts)


f (, s) d ds.

Partial Dierential Equations

7.6
7.6.1

Igor Yanovsky, 2005

26

Higher Dimensions
Spherical Means

For a continuous function u(x) on Rn , its spherical mean or average on a sphere of


radius r and center x is

1
u(x + r)dS ,
Mu (x, r) =
n ||=1
where n is the area of the unit sphere S n1 = { Rn : || = 1} and dS is surface
measure. Since u is continuous in x, Mu (x, r) is continuous in x and r, so
Mu (x, 0) = u(x).
Using the chain rule, we nd

n

1

Mu (x, r) =
uxi (x + r) i dS = 
r
n ||=1
i=1

To compute the RHS, we apply the divergence theorem in = { Rn : || < 1},


which has boundary = S n1 and exterior unit normal n() = . The integrand is
V n where V () = r 1 u(x + r) = x u(x + r). Computing the divergence of V ,
we obtain
div V () = r

n


uxi xi (x + r) = rx u(x + r),

so,

i=1

 =
=
=
=



r
1
rx u(x + r) d =
x
u(x + r) d
(  = r)
n ||<1
n
||<1

r 1
x
u(x +  ) d 
(spherical coordinates)
n r n
| |<r
 r

1
n1
x

u(x + ) dS d
n r n1
0
||=1
 r
 r
1
1
n1

M
(x,
)
d
=

n1 Mu (x, ) d.
n x
u
n1 x
n r n1
r
0
0

If we multiply by r n1 , dierentiate with respect to r, and then divide by r n1 ,


we obtain the Darboux equation:



n1
2
Mu (x, r) = x Mu (x, r).
+
r 2
r r

Note that for a radial function u = u(r), we have Mu = u, so the equation provides the
Laplacian of u in spherical coordinates.
7.6.2

Application to the Cauchy Problem

We want to solve the equation


utt = c2 u
u(x, 0) = g(x),

x Rn , t > 0,
ut (x, 0) = h(x)

(7.9)

x Rn .

We use Poissons method of spherical means to reduce this problem to a partial dierential equation in the two variables r and t.

Partial Dierential Equations

Igor Yanovsky, 2005

27

Suppose that u(x, t) solves (7.9). We can view t as a parameter and take the spherical
mean to obtain Mu (x, r, t), which satises


1
1
2
M
(x,
r,
t)
=
u
(x
+
r,
t)dS
=
c2 u(x + r, t)dS = c2 Mu (x, r, t).
u
tt

t2
n ||=1
n ||=1
Invoking the Darboux equation, we obtain the Euler-Poisson-Darboux equation:
2
Mu (x, r, t) = c2
t2

2
n1
+
r 2
r r


Mu (x, r, t).

The initial conditions are obtained by taking the spherical means:


Mu (x, r, 0) = Mg (x, r),

Mu
(x, r, 0) = Mh (x, r).
t

If we nd Mu (x, r, t), we can then recover u(x, t) by:


u(x, t) = lim Mu (x, r, t).
r0

7.6.3

Three-Dimensional Wave Equation

When n = 3, we can write the Euler-Poisson-Darboux equation as




2

2
(x,
r,
t)
=
c
(x,
r,
t)
.
rM
rM
u
u
t2
r 2

For each xed x, consider V x (r, t) = rMu (x, r, t) as a solution of the one-dimensional
wave equation in r, t > 0:
2
2 x
2
V
(r,
t)
=
c
V x (r, t),
t2
r 2
V x (r, 0) = rMg (x, r) Gx(r),

(IC)

= rMh (x, r) H (r),

(IC)

Vtx (r, 0)
x

V (0, t) = lim rMu (x, r, t) = 0 u(x, t) = 0.


r0
x

(BC)

Gx (0) = H (0) = 0.
We may extend Gx and H x as odd functions of r and use dAlemberts formula for
V x (r, t):


1 x
1 r+ct x
G (r + ct) + Gx (r ct) +
H () d.
V x (r, t) =
2
2c rct
Since Gx and H x are odd functions, we have for r < ct:
 r+ct

x
x
x
and
H () d =
G (r ct) = G (ct r)
rct

ct+r

H x () d.

ctr

After some more manipulations, we nd that the solution of (7.9) is given by the
Kirchhos formula:
 


1
t
t
u(x, t) =
g(x + ct)dS +
h(x + ct)dS .
4 t
4 ||=1
||=1
If g C 3 (R3 ) and h C 2 (R3 ), then Kirchhos formula denes a C 2 -solution of (7.9).
2

It is seen by expanding the equation below.

Partial Dierential Equations


7.6.4

Igor Yanovsky, 2005

28

Two-Dimensional Wave Equation

This problem is solved by Hadamards method of descent, namely, view (7.9) as a special
case of a three-dimensional problem with initial conditions independent of x3 .
We need to convert surface integrals in R3 to domain integrals in R2 .


 
 
t
1
g(x1 + ct1 , x2 + ct2 )d1 d2
h(x1 + ct1 , x2 + ct2 )d1 d2


+
2t
2
u(x1 , x2 , t) =
4 t
4
1 12 22
1 12 22
12 +22 <1
12 +22 <1
If g C 3 (R2 ) and h C 2 (R2 ), then this equation denes a C 2 -solution of (7.9).
7.6.5

Huygens Principle

Notice that u(x, t) depends only on the Cauchy data g, h on the surface of the hypersphere {x + ct : || = 1} in Rn , n = 2k + 1; in other words we have sharp signals.
If we use the method of descent to obtain the solution for n = 2k, the hypersurface
integrals become domain integrals. This means that there are no sharp signals.
The fact that sharp signals exist only for odd dimensions n 3 is known as Huygens
principle.
3

For x Rn :



1

f (x + t)dS = n1
f (x + y)dy
t
t
||=1
|y|t




f (x + y)dy
|y|t

= tn1




f (x + t)dS
||=1

Partial Dierential Equations

7.7

Igor Yanovsky, 2005

29

Energy Methods

Suppose u C 2 (Rn (0, )) solves



utt = c2 u
u(x, 0) = g(x), ut(x, 0) = h(x)

x Rn , t > 0,
x Rn ,

(7.10)

where g and h have compact support.


Dene energy for a function u(x, t) at time t by

1
(u2 + c2 |u|2 ) dx.
E(t) =
2 Rn t
If we dierentiate this energy function, we obtain
 
 
n
n


 2



d 1
dE
=
u2xi dx =
uxi uxi t dx
ut + c2
ut utt + c2
dt
dt 2 Rn
Rn
i=1
i=1


n
n



ut utt dx + c2
uxi ut

c2
uxi xi ut dx
=
n
Rn


=

Rn

i=1

Rn

i=1

ut (utt c2 u) dx = 0,

or
dE
dt

 
 
n
n


 2



d 1
2
2
=
uxi dx =
uxi uxi t dx
ut + c
ut utt + c2
dt 2 Rn
Rn
i=1
i=1



ut utt + c2 u ut dx
=
n



R
u
2
ds
ut utt dx + c
ut
ut u dx
=
n
n
Rn
Rn
R
ut (utt c2 u) dx = 0.
=
Rn

Hence, E(t) is constant, or E(t) E(0).


In particular, if u1 and u2 are two solutions of (7.10), then w = u1 u2 has zero Cauchy
data and hence Ew (0) = 0. By discussion above, Ew (t) 0, which implies w(x, t)
const. But w(x, 0) = 0 then implies w(x, t) 0, so the solution is unique.

Partial Dierential Equations

7.8

Igor Yanovsky, 2005

30

Contraction Mapping Principle

Suppose X is a complete metric space with distance function represented by d(, ).


A mapping T : X X is a strict contraction if there exists 0 < < 1 such that
d(T x, T y) d(x, y)

x, y X.

An obvious example on X = Rn is T x = x, which shrinks all of Rn , leaving 0 xed.


The Contraction Mapping Principle. If X is a complete metric space and T :
X X is a strict contraction, then T has a unique xed point.
The process of replacing a dierential equation by an integral equation occurs in
time-evolution partial dierential equations.
The Contraction Mapping Principle is used to establish the local existence and uniqueness of solutions to various nonlinear equations.

Partial Dierential Equations

Igor Yanovsky, 2005

31

Laplace Equation

Consider the Laplace equation


u = 0

in Rn

(8.1)

and the Poisson equation


u = f

in Rn .

(8.2)

Solutions of (8.1) are called harmonic functions in .


Cauchy problems for (8.1) and (8.2) are not well posed. We use separation of variables
for some special domains to nd boundary conditions that are appropriate for (8.1),
(8.2).
Dirichlet problem:

u(x) = g(x),
u(x)
= h(x),
n
u
+ u = ,
n

Neumann problem:
Robin problem:

8.1

x
x
x

Greens Formulas


u
ds
u v dx =
v
v u dx



 u
v 
u
ds = (vu uv) dx
v
n
n



u
ds =
u dx
n




u
ds
|u|2 dx =
u
u u dx

(8.3)

(v = 1 in (8.3))
(u = v in (8.3))

n = (n1 , n2 ) R2

ux vx dxdy =
vux n1 ds
vuxx dxdy




uxk v dx =
uvnk ds
uvxk dx

u 2 v dx =


u

u  v v  u dx =

v
ds
n


n = (n1 , . . ., nn ) Rn .


v

u
ds +
n

 v
u 
v
ds +
u
n
n

uv dx.

u

u 
v
v
ds.
n
n

Partial Dierential Equations

8.2

Igor Yanovsky, 2005

32

Polar Coordinates

Polar Coordinates. Let f : Rn R be continuous. Then






f dx =
f dS dr
Rn

Br (x0 )

Rn .

In particular
for each x0




d
f dx =
f dS
dr Br (x0 )
Br (x0 )
for each r > 0.
u = u(x(r, ), y(r, ))
x(r, ) = r cos
y(r, ) = r sin
ur = ux xr + uy yr = ux cos + uy sin ,
= ux x + uy y = ux r sin + uy r cos ,

urr = (ux cos + uy sin )r = (uxxxr + uxy yr ) cos + (uyx xr + uyy yr ) sin
= uxx cos2 + 2uxy cos sin + uyy sin2 ,
= (ux r sin + uy r cos )

= (uxx x uxy y )r sin ux r cos + (uyx x + uyy y )r cos uy r sin


= (uxx r sin uxy r cos )r sin ux r cos + (uyx r sin + uyy r cos )r cos uy r sin
= r 2 (uxx sin2 2uxy cos sin + uyy cos2 ) r(ux cos + uy sin ).
urr +

1
u
r 2

= uxx cos2 + 2uxy cos sin + uyy sin2 + uxx sin2 2uxy cos sin + uyy cos2 1r (ux cos + uy sin )
= uxx + uyy 1r ur .

1
1
uxx + uyy = urr + ur + 2 u .
r
r
1 2
2
2
1
2
+
+
=
+
.
x2 y 2
r 2 r r r 2 2

8.3

Polar Laplacian in R2 for Radial Functions


1 
rur r =
u =
r

8.4


2
1
u.
+
r 2 r r

Spherical Laplacian in R3 and Rn for Radial Functions



u =

In R3 :


2
n1
u.
+
r 2
r r


1
1 
ru rr =
u = 2 r 2 ur r =
r
r
4


2
2
u.
+
r 2 r r

These formulas are taken from S. Farlow, p. 411.

Partial Dierential Equations

8.5

33

Cylindrical Laplacian in R3 for Radial Functions


1 
u =
rur r =
r

8.6

Igor Yanovsky, 2005


2
1
+
u.
r 2 r r

Mean Value Theorem

Gauss Mean Value Theorem. If u C 2 () is harmonic in , let and pick


r > 0 so that Br () = {x : |x | r} . Then

1
u( + rx) dSx,
u() = Mu (, r)
n |x|=1
where n is the measure of the (n 1)-dimensional sphere in Rn .

8.7

Maximum Principle

Maximum Principle. If u C 2 () satises u 0 in , then either u is a constant,


or
u() < sup u(x)
x

for all .
Proof. We may assume A = supx u(x) , so by continuity of u we know that
{x : u(x) = A} is relatively closed in . But since

n
u( + rx) dx,
u()
n |x|1
if u() = A at an interior point , then u(x) = A for all x in a ball about , so
{x : u(x) = A} is open. The connectedness of implies u() < A or u() A for
all .
The maximum principle shows that u C 2 () with u 0 can attain an interior
maximum only if u is constant. In particular, if is compact, and u C 2 () C()
satises u 0 in , we have the weak maximum principle:
max u(x) = max u(x).
x

Partial Dierential Equations

8.8

Igor Yanovsky, 2005

34

The Fundamental Solution

A fundamental solution K(x) for the Laplace operator is a distribution satisfying


K(x) = (x)

(8.4)

where is the delta distribution supported at x = 0. In order to solve (8.4), we should


rst observe that  is symmetric in the variables x1 , . . ., xn , and (x) is also radially
symmetric (i.e., its value only depends on r = |x|). Thus, we try to solve (8.4) with a
radially symmetric function K(x). Since (x) = 0 for x = 0, we see that (8.4) requires
K to be harmonic for r > 0. For the radially symmetric function K, Laplace equation
becomes (K = K(r)):
n 1 K
2K
= 0.
+
r 2
r r

(8.5)

The general solution to (8.5) is



if n = 2
c1 + c2 log r
K(r) =
2n
if n 3.
c1 + c2 r

(8.6)

After we determine c2 , we nd the fundamental solution for the Laplace operator:



1
if n = 2
2 log r
K(x) =
1
2n
r
if n 3.
(2n)n
We can derive, (8.6) for any given n. For intance, when n = 3, we have:
2
K  + K  = 0.
r

Let
K =
K =
K  =

1
w(r),
r
1 
1
w 2 w,
r
r
1 
2
2
w 2 w  + 3 w.
r
r
r

Plugging these into , we obtain:


1 
w = 0,
r
w  = 0.

or

Thus,
w = c1 r + c2 ,
c2
1
w(r) = c1 + .
K =
r
r

See the similar problem, F99, #2, where the fundamental solution for ( I) is
found in the process.

Partial Dierential Equations

Igor Yanovsky, 2005

35

Find the Fundamental Solution of the Laplace Operator for n = 3


We found that starting with the Laplacian in R3 for a radially symmetric function K,
2
K  + K  = 0,
r
and letting K =
K = c1 +

1
r w(r),

we obtained the equation:

w = c1 r + c2 , which implied:

c2
.
r

We now nd the constant c2 that ensures that for v C0 (R3 ), we have



K(|x|) v(x) dx = v(0).
R3

Suppose v(x) 0 for |x| R and let = BR (0); for small  > 0 let
 = B (0).
K(|x|) is harmonic (K(|x|) = 0) in  . Consider Greens identity ( =
B (0)):





K(|x|)
K(|x|)
v
v
v
dS +
v
dS.
K(|x|)v dx =
K(|x|)
K(|x|)
n
n
n
n


B(0)



=0, since v0 f or xR



K(|x|)v dx =
K(|x|)v dx.


lim

0

Since K(r) = c1 +


c2
is integrable at x = 0.
r

On B (0), K(|x|) = K(). Thus, 5






 v 



 

v
c2 

 




dS
=
K()
dS

K(|x|)
+

c
42 max v  0, as  0.

1


n

B(0)
B(0) n


K(|x|)
dS =
v(x)
n
B(0)

c2
v(x) dS
2
B(0) 


c2
c2
v(0)
dS
+
[v(x) v(0)] dS
=
2
2

B(0)
B(0) 


c2
= 2 v(0) 42 + 4c2 max v(x) v(0)

xB(0)



0, (v is continuous)

= 4c2 v(0) v(0).


1
, we obtain
Thus, taking 4c2 = 1, i.e. c2 = 4


K(|x|)v dx = lim
K(|x|)v dx = v(0),
0

that is K(r) =
5

1
4r

is the fundamental solution of .

In R3 , for |x| = ,
K(|x|)

K() = c1 +

K(|x|)
n

c2
.


K()
c2
= 2,
r


(since n points inwards.)

n points toward 0 on the sphere |x| =  (i.e., n = x/|x|).

Partial Dierential Equations

Igor Yanovsky, 2005

36

Show that the Fundamental Solution of the Laplace Operator is given by.

1
if n = 2
2 log r
(8.7)
K(x) =
1
2n
if n 3.
(2n)n r
Proof. For v C0 (Rn ), we want to show

K(|x|) v(x) dx = v(0).
Rn

Suppose v(x) 0 for |x| R and let = BR (0); for small  > 0 let
 = B (0).
K(|x|) is harmonic (K(|x|) = 0) in  . Consider Greens identity ( =
B (0)):





K(|x|)
K(|x|)
v
v
v
dS +
v
dS.
K(|x|)v dx =
K(|x|)
K(|x|)
n
n
n
n


B(0)



=0, since v0 f or xR


lim

K(|x|)v dx =

0

K(|x|)v dx.


Since K(r) is integrable at x = 0.

On B (0), K(|x|) = K(). Thus, 6





 v 





 

v
 
 = K()

dS
K(|x|)
  dS K()n n1 max v  0, as  0.


n
B(0)
B(0) n

v(x)
B (0)

K(|x|)
dS =
n

1
v(x) dS
n1


n
B (0)


1
1

v(0) dS +

[v(x) v(0)] dS
=
n1
n1
B (0) n 
B(0) n 


1
n1
v(x) v(0)
v(0)

max
=
n
n n1
xB(0)




0, (v is continuous)

= v(0).
Thus,


K(|x|)v dx = lim

0

Note that for |x| = ,


K(|x|)

1
2

K() =

K()

=
r

K(|x|)v dx = v(0).


log 

1
2n
(2n)n 

K(|x|)
n

1
2
1
n n1

if n = 2
if n 3.
if n = 2
if n 3,

n points toward 0 on the sphere |x| =  (i.e., n = x/|x|).

1
,
n n1

(since n points inwards.)

Partial Dierential Equations

8.9

Igor Yanovsky, 2005

37

Representation Theorem

Representation Theorem, n = 3.
Let be bounded domain in R3 and let n be the unit exterior normal to . Let
u C 2 (). Then the value of u at any point x is given by the formula

 


1
1
u(y)
1 u(y)
1
u(y)
dS
dy. (8.8)
u(x) =
4 |x y| n
n |x y|
4 |x y|
Proof. Consider the Greens identity:


 w
u 
u
w
dS,
(uw wu) dy =
n
n

where w is the harmonic function


1
,
w(y) =
|x y|
which is singular at x . In order to be able to apply Greens identity, we consider
a new domain  :
 = B (x).
Since u, w C2 ( ), Greens identity can be applied. Since w is harmonic (w = 0)
in  and since  = B (x), we have

 

1
1 u(y)
u(y)

dy =

dS
(8.9)
u(y)

n |x y| |x y| n
 |x y|




1
1 u(y)

dS.
(8.10)
u(y)
+
n |x y| |x y| n
B(x)
We will show that formula (8.8) is obtained by letting  0.

 



1
u(y)
u(y)
dy =
dy.
Since
is integrable at x = y.
lim
0
|x y|
 |x y|
|x y|
The rst integral on the right of (8.10) does not depend on . Hence, the limit as  0
of the second integral on the right of (8.10) exists, and in order to obtain (8.8), need



1
1 u(y)

dS = 4u(x).
u(y)
lim
0 B(x)
n |x y| |x y| n

B(x)





1
1 u(y)
1 u(y)

dS =
dS
u(y)
u(y)
2
n |x y| |x y| n
 n
B(x) 




1 u(y)
1
1
dS
u(x) dS +
[u(y) u(x)]
=
2
2
 n
B(x) 
B(x) 



1 u(y)
1
dS.
[u(y) u(x)]
= 4u(x) +
2
 n
B (x) 

Partial Dierential Equations

Igor Yanovsky, 2005

7
The last integral tends to 0 as  0:

 



1 u(y)
1

dS 
[u(y) u(x)]

2
 n
B(x) 

38


 1
 u(y) 




dS
u(y) u(x) +


n
B(x)
B (x)








4 max u(y) u(x) + 4 max u(y) .
yB(x)
y

 



1
2

0, (u continuous in )

Note that for points y on B (x),


1
1
=
|x y|


and

1
1
= 2.
n |x y|


0, (|u| is f inite)

Partial Dierential Equations

Igor Yanovsky, 2005

39

Representation Theorem, n = 2.
Let be bounded domain in R2 and let n be the unit exterior normal to . Let
u C 2 (). Then the value of u at any point x is given by the formula


 
u(y)
1

1
log |x y| log |x y|
dS.
(8.11)
u(y) log |x y| dy +
u(y)
u(x) =
2
2
n
n
Proof. Consider the Greens identity:


 w
u 
(uw wu) dy =
u
w
dS,
n
n

where w is the harmonic function


w(y) = log |x y|,
which is singular at x . In order to be able to apply Greens identity, we consider
a new domain  :
 = B (x).
Since u, w C2 ( ), Greens identity can be applied. Since w is harmonic (w = 0)
in  and since  = B (x), we have

u(y) log |x y| dy
(8.12)

 


u(y)

log |x y| log |x y|
dS
u(y)
=
n
n




u(y)

log |x y| log |x y|
dS.
u(y)
+
n
n
B (x)
We will show that formula (8.11) is obtained by letting  0.

 

u(y) log |x y| dy = u(y) log |x y| dy.
lim
0


since log |x y| is integrable at x = y.

The rst integral on the right of (8.12) does not depend on . Hence, the limit as  0
of the second integral on the right of (8.12) exists, and in order to obtain (8.11), need



u(y)

log |x y| log |x y|
dS = 2u(x).
u(y)
lim
0 B(x)
n
n

B (x)





u(y)
u(y)

1
log |x y| log |x y|
dS =
u(y) log 
dS
u(y)
n
n
n
B(x) 




u(y)
1
1
u(x) dS +
[u(y) u(x)] log 
dS
=
n
B(x) 
B(x) 



u(y)
1
[u(y) u(x)] log 
dS.
= 2u(x) +
n
B(x) 

Partial Dierential Equations

Igor Yanovsky, 2005

8
The last integral tends to 0 as  0:

 



u(y)
1

[u(y) u(x)] log 
dS 

n
B(x) 

40




 u(y) 




dS
u(y) u(x) + log 

n
B(x)
B(x)








2 max u(y) u(x) + 2 log  max u(y) .
yB (x)
y

 



1


0, (u continuous in )

Note that for points y on B (x),


log |x y| = log 

and

log |x y| = .
n


0, (|u| is f inite)

Partial Dierential Equations

Igor Yanovsky, 2005

41

Representation Theorems, n > 3 can be obtained in the same way. We use the
Greens identity with
w(y) =

1
,
|x y|n2

which is a harmonic function in Rn with a singularity at x.


The fundamental solution for the Laplace operator is (r = |x|):

1
if n = 2
2 log r
K(x) =
1
2n
r
if n 3.
(2n)n
Representation Theorem. If Rn is bounded, u C 2 (), and x , then

 

u(y)
K(x y)
K(x y)
dS.
(8.13)
K(x y)u(y) dy +
u(y)
u(x) =
n
n

Proof. Consider the Greens identity:




 w
u 
w
dS,
(uw wu) dy =
u
n
n

where w is the harmonic function


w(y) = K(x y),
which is singular at y = x. In order to be able to apply Greens identity, we consider a
new domain  :
 = B (x).
Since K(x y) is
Since u, K(x y) C2 ( ), Greens identity can be applied.
harmonic (K(x y) = 0) in  and since  = B (x), we have

 

u(y)
K(x y)
K(x y)
dS (8.14)
K(x y)u(y) dy =
u(y)

n
n





u(y)
K(x y)
K(x y)
dS.
(8.15)
u(y)
+
n
n
B (x)
We will show that formula (8.13) is obtained by letting  0.

 



K(x y)u(y) dy = K(x y)u(y) dy.
since K(x y) is integrable at x = y.
lim
0

The rst integral on the right of (8.15) does not depend on . Hence, the limit as  0
of the second integral on the right of (8.15) exists, and in order to obtain (8.13), need



u(y)
K(x y)
K(x y)
dS = u(x).
u(y)
lim
0 B(x)
n
n

Partial Dierential Equations

Igor Yanovsky, 2005

42





u(y)
u(y)
K(x y)
K()
K(x y)
dS =
K()
dS
u(y)
u(y)
n
n
n
n
B (x)
B(x)




K()
u(y)
K()
dS +
[u(y) u(x)] K()
dS
u(x)
=
n
n
n
B(x)
B (x)



1
u(y)
1
dS
u(x) dS
[u(y) u(x)] dS
K()
=
n1
n1
n 
n 
n
B(x)
B (x)
B(x)


1
u(y)
1
n1
dS.
u(x)n 

[u(y) u(x)] dS
K()
=
n1
n1

n 
n
B (x)
B(x)
 n




u(x)

The last two integrals tend to 0 as  0:






1

[u(y) u(x)] dS
 n1
n

1
n n1



u(y) 
dS 
K()
n
B(x)
B(x)






max u(y) u(x)n n1 + K() max u(y)n n1 .
yB (x)
y

 


0, (|u| is f inite)

0, (u continuous in )

8.10

Greens Function and the Poisson Kernel

With a slight change in notation, the Representation Theorem has the following special
case.

Theorem. If Rn is bounded, u C 2 () C 1 () is harmonic, and , then

 
u(x)
K(x )
K(x )
dS.
(8.16)
u(x)
u() =
n
n

Let (x) be any harmonic function in , and for x, consider


G(x, ) = K(x ) + (x).
If we use the Greens identity (with u = 0 and  = 0), we get:


u 

ds.
u
0 =
n
n

Adding (8.16) and (8.17), we obtain:



 
u(x)
G(x, )
G(x, )
dS.
u(x)
u() =
n
n

(8.17)

(8.18)

Suppose that for each we can nd a function (x) that is harmonic in and
satises (x) = K(x ) for all x . Then G(x, ) = K(x ) + (x) is a
fundamental solution such that
x .

G(x, ) = 0
9

Note that for points y on B (x),



1
log 
2
K(x y) = K() =
1

K(x y)
n

2n
(2n)n


K()
=
r

1
2
1
n n1

if n = 2
if n 3.
if n = 2
if n 3,

1
,
n n1

(since n points inwards.)

Partial Dierential Equations

Igor Yanovsky, 2005

43

G is called the Greens function and is useful in satisfying Dirichlet boundary conditions.
The Greens function is dicult to construct for a general domain since it requires
solving the Dirichlet problem  = 0 in , (x) = K(x ) for x , for each
.
From (8.18) we nd 10

u() =

u(x)

G(x, )
dS.
n

Thus if we know that the Dirichlet problem has a solution u C 2 (), then we can
calculate u from the Poisson integral formula (provided of course that we can compute
G(x, )).
10
If we did not assume u = 0 in our derivation, we would have (8.13) instead of (8.16), and an
extra term in (8.17), which would give us a more general expression:

u() =

G(x, ) u dx +

u(x)

G(x, )
dS.
n

Partial Dierential Equations

8.11

Igor Yanovsky, 2005

44

Properties of Harmonic Functions

Liouvilles Theorem. A bounded harmonic function dened on all of Rn must be a


constant.

8.12

Eigenvalues of the Laplacian

Consider the equation



u + u = 0
u=0

in
on ,

(8.19)

where is a bounded domain and is a (complex) number. The values of for which
(8.19) admits a nontrivial solution u are called the eigenvalues of  in , and the
solution u is an eigenfunction associated to the eigenvalue . (The convention
u + u = 0 is chosen so that all eigenvalues will be positive.)
Properties of the Eigenvalues and Eigenfunctions for (8.19):
1. The eigenvalues of (8.19) form a countable set {n}
n=1 of positive numbers with
n as n .
2. For each eigenvalue n there is a nite number (called the multiplicity of n ) of
linearly independent eigenfunctions un .
3. The rst (or principal) eigenvalue, 1 , is simple and u1 does not change sign in .
4. Eigenfunctions corresponding to distinct eigenvalues are orthogonal.
5. The eigenfunctions may be used to expand certain functions on in an innite
series.

Partial Dierential Equations

Igor Yanovsky, 2005

45

Heat Equation

The heat equation is


for x , t > 0,

ut = ku

(9.1)

with initial and boundary conditions.

9.1
9.1.1

The Pure Initial Value Problem


Fourier Transform

 by
If u C0 (Rn ), dene its Fourier transform u

1
eix u(x) dx
for Rn .
u
() =
n
(2) 2 Rn
We can dierentiate u
:




1

u
() =
eix (ixj )u(x) dx = (ix
n
j ) u ().
j
(2) 2 Rn
Iterating this computation, we obtain


  
k
k
u
() = (ix
j ) u ().
j

(9.2)

Similarly, integrating by parts shows





1

1
u
ix u
e
(x) dx =
(eix )u(x) dx
() =
n
n
xj
xj
(2) 2 Rn
(2) 2 Rn xj

1
(ij )eix u(x) dx
=
n
(2) 2 Rn
u().
= (ij )
Iterating this computation, we obtain


ku
().
() = (ij )k u
xkj

(9.3)

Formulas (9.2) and (9.3) express the fact that Fourier transform interchanges dierentiation and multiplication by the coordinate function.
9.1.2

Multi-Index Notation

A multi-index is a vector = (1 , . . . , n ) where each i is a nonnegative integer.


The order of the multi-index is || = 1 + . . . + n . Given a multi-index , dene
Du =

||u
= x11 xnn u.
xn n

x1 1

We can generalize (9.3) in multi-index notation:




1
(1)||
ix


e
D u(x) dx =
Dx (eix )u(x) dx
D u() =
n
n
(2) 2 Rn
(2) 2 Rn

1
(i)eix u(x) dx
=
n
n
2
(2)
R

().
= (i) u
(i) = (i1 )1 (in )n .

Partial Dierential Equations

Igor Yanovsky, 2005

46

Parsevals theorem (Plancherels theorem).


, u L2 (Rn ) and
Assume u L1 (Rn ) L2 (Rn ). Then u
||
u||L2 (Rn) = ||u ||L2(Rn ) = ||u||L2(Rn ) ,


Also,

|u(x)| dx =

|
u()|2 d.

u(x) v(x) dx =

or

u
() 
v() d.

The properties (9.2) and (9.3) make it very natural to consider the fourier transform
on a subspace of L1 (Rn ) called the Schwartz class of functions, S, which consists of the
smooth functions whose derivatives of all orders decay faster than any polynomial, i.e.
S = {u C (Rn ) : for every k N and Nn , |x|k |Du(x)| is bounded on Rn }.
For u S, the Fourier transform u
 exists since u decays rapidly at .
 is bounded. (ii) If u S, then u
 S.
Lemma. (i) If u L1 (Rn ), then u
Dene the inverse Fourier transform for u L1 (Rn ):

1

eix u(x) dx
for Rn ,
u () =
n
(2) 2 Rn

1
eix u
() d
for x Rn .
u(x) =
n
2
n
(2)
R

or

Fourier Inversion Theorem (McOwen). If u S, then (


u) = u; that is,
1
u(x) =
n
(2) 2


ix

Rn

1
u
() d =
(2)n

 
R2n

ei(xy) u(y) dy d = (
u) (x).

Fourier Inversion Theorem (Evans). Assume u L2 (Rn ). Then, u = (


u) .

Partial Dierential Equations

Igor Yanovsky, 2005

47

Shift: Let u(x


a) = v(x), and determinte v():
 
y


u(x
a)() = 
v() =
=



1
1

eix v(x) dx =
ei(y+a) u(y) dy
2 R
2 R
1

eiy eia u(y) dy = eia u


().
2 R


().
u(x
a)() = eia u
Delta function:
1

1
eix (x) dx = ,
2
R
1


= eia .
(x
a)() = eia ()
2

(x)()
=

Convolution:
(f g)(x) =

(f
g)() =
=
=


since u(x) =
R


(x y) u(y) dy .

(using result from Shift)


f (x y)g(y) dy,


 
1
1
ix
e
f (x y) g(y) dy dx =
eix f (x y) g(y) dy dx
n
n
(2) 2 Rn
(2) 2 Rn Rn
Rn
  


1
i(xy)
iy
f
(x

y)
dx
e
g(y)
dy
e
n
(2) 2 Rn Rn


n
1
iz
e
f (z) dz
eiy g(y) dy = (2) 2 f()
g().
n
(2) 2 Rn
Rn
Rn

n

g().
(f
g)() = (2) 2 f() 

Gaussian: (completing the square)





2
2
2

x2 +2ix 2
2
1
1
1
ix x2
x +2ix
x2
2
2
e
e
dx =
e
dx =
e
dx e 2
() =
e
2 R
2 R
2 R


2
2
2
2
(x+i)

2
2
1
1
1
=
e 2 dx e 2 =
e 2 dy e 2 =
2e 2 = e 2 .
2 R
2 R
2

2
x2 
e 2 () = e 2 .
Multiplication by x:
1

ixu()
=
2



d
u
().
eix ixu(x) dx =
d
R

d

().
xu(x)()
= i u
d

Partial Dierential Equations

Igor Yanovsky, 2005

48

Multiplication of ux by x: (using the above result)








1
1  ix
1
ix

e
e
xu

xux (x) dx =
(i)eix x + eix u dx
xux (x)() =

2 R
2
2 R



=0


1
1
eix u dx
= i eix x u dx
2
2
R
R

 d
d

() u
() = u
() u
().
= i xu(x)()
u
() = i i u
d
d
d

u
() u
().
xu
x (x)() =
d

Table of Fourier Transforms:



ax2 
e 2 () =

f (ax)() =
eibx
f (x) =
a|x| () =
e


1
() =
2
a + x2

H(a
|x|)() =

H(x)()
=



H(x)
H(x) () =

1() =

11

11

2
1
e 2a ,
(Gaussian)
a
1 
b
f
,
a
a

1, |x| L
1 2 sin(L)
,
f
(x)() =

2
0, |x| > L,
2a
1

,
(a > 0)
2
2 a + 2

2 a||
e
,
(a > 0)
2a

2 1
sin a,


1
1

() +
,

i
2

2 1
,
(sign)

i

2().


Results with marked with  were taken from W. Strauss, where the denition of Fourier Transform
is dierent. An extra multiple of 12 was added to each of these results.

Partial Dierential Equations


9.1.3

Igor Yanovsky, 2005

49

Solution of the Pure Initial Value Problem

Consider the pure initial value problem



for t > 0, x Rn
ut = u
u(x, 0) = g(x)
for x Rn .

(9.4)

We take the Fourier transform of the heat equation in the x-variables.



1


)(,
t)
=
eix ut (x, t) dx = u
(u
(, t)
n
t
n
t
2
(2)
R
n


(ij )2 u
(, t) = ||2 u
(, t).
u(, t) =
j=1

The heat equation therefore becomes

u
(, t) = ||2 u
(, t),
t
2

which is an ordinary dierential equation in t, with the solution u


(, t) = Ce|| t.
The initial condition u
(, 0) = 
g() gives
2

u
(, t) = g() e|| t,

2
=
u(x, t) =
g() e|| t


 1
1
g

n
n
(2) 2
(2) 2 Rn



 |x|2
n
1
1
2
2
4t
eix|| t d =
n g
n g e
2
2
t
(4 ) 2
(4 ) 2
Rn


2
2
|x|
|xy|
1
1
4t
e 4t g(y) dy.
=
n g e
n
n
(4t) 2
(4t) 2 R

=
=
=
Thus,

12

1   ||2 t 
n g e
(2) 2


2
e|| t eix d

solution of the initial value problem (9.4) is




1
u(x, t) =
K(x, y, t) g(y) dy =
n
n
(4t) 2
R


Rn

|xy|2
4t

g(y) dy.

Uniqueness of solutions for the pure initial value problem fails: there are nontrivial
solutions of (9.4) with g = 0. 13 Thus, the pure initial value problem for the heat
equation is not well-posed, as it was for the wave equation. However, the nontrivial
solutions are unbounded as functions of x when t > 0 is xed; uniqueness can be
regained by adding a boundedness condition on the solution.
12

13

Identity (Evans, p. 187.) :



n
|x|2
2
2
eix|| t d = e 4t
.
t
Rn
The following function u satises ut = uxx for t > 0 with u(x, 0) = 0:
u(x, t) =


k=0

2
1
dk
x2k k e1/t .
(2k)!
dt

Partial Dierential Equations


9.1.4

Igor Yanovsky, 2005

50

Nonhomogeneous Equation

Consider the pure initial value problem with homogeneous initial condition:

for t > 0, x Rn
ut = u + f (x, t)
u(x, 0) = 0
for x Rn .

(9.5)

Duhamels principle gives the solution:


 t
u(x, t) =
0

9.1.5

Rn

y, t s) f (y, s) dy ds.
K(x

Nonhomogeneous Equation with Nonhomogeneous Initial Conditions

Combining two solutions above, we nd that the solution of the initial value problem

for t > 0, x Rn
ut = u + f (x, t)
(9.6)
u(x, 0) = g(x)
for x Rn .
is given by

u(x, t) =

9.1.6

Rn

y, t) g(y) dy +
K(x

 t
0

Rn

y, t s) f (y, s) dy ds.
K(x

The Fundamental Solution

Suppose we want to solve the Cauchy problem



x Rn , t > 0
ut = Lu
u(x, 0) = g(x)
x Rn .

(9.7)

where L is a dierential operator in Rn with constant coecients. Suppose K(x, t) is


a distribution in Rn for each value of t 0, K is C 1 in t and satises

Kt LK = 0,
(9.8)
K(x, 0) = (x).
We call K a fundamental solution for the initial value problem. The solution of
(9.7) is then given by convolution in the space variables:

K(x y, t) g(y) dy.
u(x, t) =
Rn

Partial Dierential Equations

Igor Yanovsky, 2005

51

For operators of the form t L, the fundamental solution of the initial value problem,
K(x, t) as dened in (9.8), coincides with the free space fundamental solution, which
satises


t L K(x, t) = (x, t),
provided we extend K(x, t) by zero to t < 0. For the heat equation, consider

|x|2

1
4t
e
t>0
n/2

(4t)
K(x,
t) =
0
t 0.

(9.9)

is smooth for (x, t) = (0, 0).


Notice that K
dened as in (9.9), is the fundamental solution of the free space heat
K
equation.
Proof. We need to show:

t  K(x, t) = (x, t).

(9.10)

To verify (9.10) as distributions, we must show that for any v C0 (Rn+1 ): 14





(x, t) v(x, t) dx dt v(0, 0).


K(x, t) t  v dx dt =
Rn+1

Rn+1

To do this, let us take  > 0 and dene

|x|2

1
4t
e
t>
 (x, t) =
(4t)n/2
K
0
t .
as distributions, so it suces to show that (t )K
 as distribu K
Then K
tions. Now

 


 t  v dx dt =
K(x,
t) t  v(x, t) dx dt
K
n

R

 
 

K(x, t) t v(x, t) dx dt
K(x, t) v(x, t) dx dt
=


Rn

Rn


t=   


 

+
t K(x, t) v(x, t) dx dt
K(x, t) v(x, t) dx dt
K(x, t) v(x, t) dx
=


Rn
Rn
Rn
t=


 

K(x,
) v(x, ) dx.
t  K(x, t) v(x, t) dx dt +
=


Rn

Rn

t) = 0; moreover, since limt0+ K(x,


t) = 0 (x) = (x),
But for t > , (t )K(x,

we have K(x, ) 0 (x) as  0, so the last integral tends to v(0, 0).


14

Note, for the operator L = /t, the adjoint operator is L = /t.

Partial Dierential Equations

10

Igor Yanovsky, 2005

52

Schr
odinger Equation

Problem (F96, #5). The Gauss kernel


G(t, x, y) =

1 e

(xy)2
4t

(4t) 2

is the fundamental solution of the heat equation, solving


Gt = Gxx,

G(0, x, y) = (x y).

By analogy with the heat equation, nd the fundamental solution H(t, x, y) of the
Schr
odinger equation
Ht = iHxx,

H(0, x, y) = (x y).

Show that your expression H(x) is indeed the fundamental solution for the
Schr
odinger equation. You may use the following special integral


ix2
e 4 dx = i4.

Proof. Remark: Consider the initial value problem for the Schr
odinger equation

x Rn , t > 0,
ut = iu
u(x, 0) = g(x)
x Rn .
If we formally replace t by it in the heat kernel, we obtain the Fundamental
Solution of the Schr
odinger Equation: 15
|x|2
1
4it
(x Rn , t = 0)
n e
(4it) 2

|xy|2
1
e 4it g(y) dy.
u(x, t) =
n
(4it) 2 Rn

H(x, t) =

In particular, the Schr


odinger equation is reversible in time, whereas the heat equation
is not.
Solution: We have already found the fundamental solution for the heat equation
using the Fourier transform. For the Schr
odinger equation is one dimension, we have

u
(, t) = i 2 u
(, t),
t
2

which is an ordinary dierential equation in t, with the solution u


(, t) = Cei t .
The initial condition u
(, 0) = 
g() gives
2

u
(, t) = g() ei t ,

1   i 2 t 
2
g e
=
u(x, t) =
g() ei t

2

 1

2
1
ei t eix d
= g
2
2 R


2
1
g
eixi t d
= (need some work) =
=
2
R

 |x|2
|xy|2
1
1
g e 4it =
e 4it g(y) dy.
=
4it
4it R
15

Evans, p. 188, Example 3.

Partial Dierential Equations

Igor Yanovsky, 2005

53

For the Schr


odinger equation, consider

|x|2

1
4it
e
t>0

(4it)n/2
(x,
t) =
0
t 0.

(10.1)

is smooth for (x, t) = (0, 0).


Notice that
dened as in (10.1), is the fundamental solution of the Schr

odinger equation. We need to show:


(10.2)
t i (x, t) = (x, t).
To verify (10.2) as distributions, we must show that for any v C0 (Rn+1 ): 16



(x, t) v(x, t) dx dt v(0, 0).


(x,
t) t i v dx dt =
Rn+1

Rn+1

To do this, let us take  > 0 and dene

|x|2

1
4it
e
t>
n/2
 (x, t) =
(4it)

0
t .
as distributions, so it suces to show that (t i)
 as distribu
Then
tions. Now

 


(x, t) t i v(x, t) dx dt
 t i v dx dt =

Rn


 

t) v(x, t) dx dt +
(x,
) v(x, ) dx.
t i (x,
=


Rn

Rn

t) = 0; moreover, since limt0+ (x,


t) = 0 (x) = (x),
But for t > , (t i)(x,

we have (x, ) 0 (x) as  0, so the last integral tends to v(0, 0).


16

Note, for the operator L = /t, the adjoint operator is L = /t.

Partial Dierential Equations

11

Igor Yanovsky, 2005

54

Problems: Quasilinear Equations

Problem (F90, #7). Use the method of characteristics to nd the solution of the
rst order partial dierential equation
x2 ux + xyuy = u2
which passes through the curve u = 1, x = y 2 . Determine where this solution becomes
singular.
Proof. We have a condition u(x = y 2 ) = 1.
dx
dt

= x2

dy
dt
dz
dt

= xy
= z2

x=

1
t c1 (s)

is parametrized by : (s2 , s, 1).

x(0, s) =

1
1
= s2 x =
c1 (s)
t +

1
s2

dy
s2 y
c2 (s)
s
=
y=
y(s, 0) = c2 (s) = s y =
,
dt
1 ts2
1 ts2
1 ts2
1
1
1
z(0, s) =
=1 z=
.
z=
t c3 (s)
c3 (s)
1t

Thus,
x
=s
y

y=

x
y

1 t xy2

u(x, y) =

1
1

y2
x2

t=

1
x

y2
1
.
2
x
x

x2
.
x2 + x y 2

The solution becomes singular when y 2 = x2 + x.


It can be checked that the solution satises the PDE and u(x = y 2 ) =
Problem (S91, #7). Solve the rst order PDE
fx + x2 yfy + f = 0
f (x = 0, y) = y 2
using the method of characteristics.
Proof. Rewrite the equation
ux + x2 yuy = u,
u(0, y) = y 2 .
is parameterized by : (0, s, s2).
dx
dt
dy
dt
dz
dt

s2
,
1 ts2

= 1

x = t,

= x2 y

dy
= t2 y
dt

= z

z = s2 et .
t3

Thus, x = t and s = ye 3 = ye
u(x, y) = (ye

x3
3

t3

x3
3

)2 ex = y 2 e 3 x

y = se 3 ,

, and

3 x

The solution satises both the PDE and initial conditions.

y4
y4 +y2 y2

= 1.

Partial Dierential Equations

Igor Yanovsky, 2005

55

Problem (S92, #1). Consider the Cauchy problem


ut = xux u + 1

< x < , t 0

u(x, 0) = sin x

< x <

and solve it by the method of characteristics. Discuss the properties of the solution; in
particular investigate the behavior of |ux(, t)| for t .
Proof. is parametrized by : (s, 0, sin s). We have
dx
= x x = set ,
dt
dy
= 1 y = t,
dt
1 sin s
dz
= 1z z = 1
.
dt
et
Thus, t = y, s = xey , and
sin(xey )
1
+
.
y
e
ey
It can be checked that the solution satises the PDE and the initial condition.
As t , u(x, t) 1. Also,
u(x, y) = 1

|ux(x, y)| = | cos(xey )| = 1.


ux (x, y) oscillate between 1 and 1. If x = 0, ux = 1.

Problem (W02, #6). Solve the Cauchy problem


ut + u2 ux = 0,
u(0, x) = 2 + x.
Proof. Solved

t > 0,

Partial Dierential Equations

Igor Yanovsky, 2005

56

Problem (S97, #1). Find the solution of the Burgers equation


ut + uux = x,

t0

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

< x < .

Proof. is parameterized by : (s, 0, f (s)).


dx
= z,
dt
dy
= 1 y = t,
dt
dz
= x.
dt
Note that we have a coupled system:

x = z,
z = x,
which can be written as a second order ODE:
x
+ x = 0,

x(s, 0) = s, x(s,
0) = z(0) = f (s).

Solving the equation, we get


x(s, t) = s cos t + f (s) sin t,

and thus,

z(s, t) = x(t)

= s sin t + f (s) cos t.




x = s cos y + f (s) sin y,
x cos y = s cos2 y + f (s) sin y cos y,

u sin y = s sin2 y + f (s) cos y sin y.


u = s sin y + f (s) cos y.
x cos y u sin y = s(cos2 y + sin2 y) = s.
u(x, y) = f (x cos y u sin y) cos y (x cos y u sin y) sin y.
Problem (F98, #2). Solve the partial dierential equation
uy u2 ux = 3u,

u(x, 0) = f (x)

using method of characteristics. (Hint: nd a parametric representation of the solution.)


Proof. is parameterized by : (s, 0, f (s)).
dx
dt
dy
dt
dz
dt

= z 2
= 1
= 3z

dx
= f 2 (s)e6t
dt
y = t,
z = f (s)e3t.

1
1
x = f 2 (s)e6t + f 2 (s) + s,
6
6

Partial Dierential Equations


Thus,


x = 16 f 2 (s)e6y + 16 f 2 (s) + s,
f (s) = ez3y

Igor Yanovsky, 2005

z2
z2
s = x 6y + .
6e
6

2
z 2 3y
z
e .
z = f x 6y +
6e
6

u2 3y
u2
e .
u(x, y) = f x 6y +
6e
6

x=

57

1 z 2 6y 1 z 2
z2
z2
+ s,
e
+
+
s
=

6 e6y
6 e6y
6e6y
6

Partial Dierential Equations

Igor Yanovsky, 2005

58

Problem (S99, #1) Modied Problem. a) Solve


ut +

u3
3

=0

(11.1)

for t > 0, < x < with initial data



a(1 ex ),
x<0
u(x, 0) = h(x) =
x
a(1 e ), x > 0
where a > 0 is constant. Solve until the rst appearance of discontinuous derivative
and determine that critical time.
b) Consider the equation
ut +

u3
3

= cu.

(11.2)

How large does the constant c > 0 has to be, so that a smooth solution (with no discontinuities) exists for all t > 0? Explain.
Proof. a) Characteristic form: ut + u2 ux = 0.

: (s, 0, h(s)).

dz
dy
dx
= z2,
= 1,
= 0.
dt
dt
dt
x = h(s)2 t + s, y = t, z = h(s).
u(x, y) = h(x u2 y)

(11.3)

The characteristic projection in the xt-plane17 passing through the point (s, 0) is the
line
x = h(s)2 t + s
along which u has the constant value u = h(s).
The derivative of the initial data is discontinuous, and that leads to a
rarefaction-like behavior at t = 0. However, if the question meant to ask to
determine the rst time when a shock forms, we proceed as follows.
Two characteristics x = h(s1 )2 t + s1 and x = h(s2 )2 t + s2 intersect at a point (x, t)
with
s2 s1
.
t=
h(s2 )2 h(s1 )2
From (11.3), we have
ux = h (s)(1 2uuxt)

ux =

h (s)
1 + 2h(s)h (s)t

Hence for 2h(s)h (s) < 0, ux becomes innite at the positive time
t=

1
.
2h(s)h (s)

The smallest t for which this happens corresponds to the value s = s0 at which h(s)h (s)
has a minimum (i.e.h(s)h (s) has a maximum). At time T = 1/(2h(s0 )h (s0 )) the
17

y and t are interchanged here

Partial Dierential Equations

Igor Yanovsky, 2005

solution u experiences a gradient catastrophe.


Therefore, need to nd a minimum of

2a(1 ex ) aex
f (x) = 2h(x)h (x) =
2a(1 ex ) (aex )

x<0
2a2 ex (1 2ex ),
f  (x) =
2
x
x
2a e (1 2e ), x > 0


=0

2a2 ex (1 ex ),
2a2 ex (1 ex ),

x<0
x>0


x = ln( 12 ) = ln(2), x < 0
x = ln(2),
x>0

f (ln( 12 )) = 2a2 eln( 2 ) (1 eln( 2 ) ) = 2a2 ( 12 )( 12 ) =


2
f (ln(2)) = 2a2 ( 12 )(1 12 ) = a2 ,
x>0

t=

a2
2 ,

x<0

2
1
= 2

min{2h(s)h (s)}
a

Proof. b) Characteristic form: ut + u2 ux = cu.


dx
=
dt
dy
=
dt
dz
=
dt
Solving for s

59

z 2 = h(s)2 e2ct

x=s+

y = t,

cz

z = h(s)ect

: (s, 0, h(s)).

1
h(s)2 (1 e2ct ),
2c

( h(s) = uecy ).

and t in terms of x and y, we get:


1
t = y,
s = x h(s)2 (1 e2cy ).
2c

Thus,


1
u(x, y) = h x u2 e2cy (1 e2cy ) ecy .
2c
1

cy
(1 uux e2cy (1 e2cy )),
ux = h (s)e
c

cy
h (s)ecy
h (s)e
=
.
ux =
1 + 1c h (s)ecy u (1 e2cy )
1 + 1c h (s)h(s)(1 e2cy )
Thus, c > 0 that would allow a smooth solution to exist for all t > 0 should satisfy
1
1 + h (s)h(s)(1 e2cy ) = 0.
c
We can perform further calculations taking into account the result from part (a):
min{2h(s)h (s)} =

a2
.
2

Partial Dierential Equations

Igor Yanovsky, 2005

60

Problem (S99, #1). Original Problem. a). Solve


ut +

u3x
=0
3

(11.4)

for t > 0, < x < with initial data



x<0
a(1 ex ),
u(x, 0) = h(x) =
x
a(1 e ), x > 0
where a > 0 is constant.
Proof. Rewrite the equation as
u3x
+ uy = 0,
3
p3
+ q = 0.
F (x, y, z, p, q) =
3
is parameterized by : (s, 0, h(s), (s), (s)).
We need to complete to a strip. Find (s) and (s), the initial conditions for p(s, t)
and q(s, t), respectively:
F (x, y, u, ux, uy ) =

F (f (s), g(s), h(s), (s), (s)) = 0,


(s)3
+ (s) = 0,
3
(s)3
.
(s) =
3

h (s) = (s)f  (s) + (s)g (s)




3 3s
(s) = a 3e ,
x<0
aes = (s),

3 3s
x>0
aes = (s),
(s) = a e3 ,

Therefore, now is parametrized by



3 3s
: (s, 0, a(1 es ), aes , a 3e ),
3 3s
: (s, 0, a(1 es ), aes , a e3 ),
dx
dt
dy
dt
dz
dt


2

= Fp = p =

a2 e2s
a2 e2s


x(s, t) =

x<0
x>0

x<0
x>0
a2 e2s t + c4 (s)
a2 e2s t + c5 (s)


x=

a2 e2s t + s
a2 e2s t + s

y(s, t) = t + c1 (s) y = t

3 3s
a3 e3s a 3e
= 23 a3 e3s ,
x<0
3
= pFp + qFq = p + q =
a3 e3s
2 3 3s
3
3s
+ 3 = 3a e ,
x>0
a e


2 3 3s
2 3 3s
s
x<0
3 a e t + c6 (s),
3 a e t a(1 e ),

z
=
z(s, t) =
x>0
23 a3 e3s t + c7 (s),
23 a3 e3s t a(1 es ),
=

dp
dt

Fq = 1


= Fx Fz p = 0

p(s, t) = c2 (s)

p=

aes ,
aes ,

x<0
x>0

x<0
x>0

Partial Dierential Equations

Igor Yanovsky, 2005

dq
dt

= Fy Fz q = 0

q(s, t) = c3 (s)

q=

3 e3s

3 ,
a3 e3s
,
3

61

x<0
x>0

Thus,

u(x, y) =

2 3 3s
s
3 a e y a(1 e ),
23 a3 e3s y a(1 es ),

where s is dened as

a2 e2s y + s,
x=
a2 e2s y + s,

x<0
x>0

x<0
x > 0.

b). Solve the equation


ut +

u3x
= cu.
3

(11.5)

Proof. Rewrite the equation as


u3x
+ uy + cu = 0,
3
p3
+ q + cz = 0.
F (x, y, z, p, q) =
3
is parameterized by : (s, 0, h(s), (s), (s)).
We need to complete to a strip. Find (s) and (s), the initial conditions for p(s, t)
and q(s, t), respectively:
F (x, y, u, ux, uy ) =

F (f (s), g(s), h(s), (s), (s)) = 0,


(s)3
+ (s) + ch(s) = 0,
3

(s)3
3 + ca(1 ex ), x < 0
(s)3
ch(s) =
(s) =
3
x
3
(s)
3 + ca(1 e ), x > 0
h (s) = (s)f  (s) + (s)g (s)


3 3s
(s) = a 3e + ca(1 ex ),
x<0
aes = (s),

3 3s
x>0
aes = (s),
(s) = a e3 + ca(1 ex ),

Therefore, now is parametrized by



3 3s
aes , a 3e + ca(1 ex ),
: (s, 0, a(1 es ),
3 3s
: (s, 0, a(1 es ), aes , a e3 + ca(1 ex ),

x<0
x>0

x<0
x>0

Partial Dierential Equations

Igor Yanovsky, 2005

dx
= Fp = p2
dt
dy
= Fq = 1
dt
dz
= pFp + qFq = p3 + q
dt
dp
= Fx Fz p = cp
dt
dq
= Fy Fz q = cq
dt
We can proceed solving the characteristic equations with initial conditions above.

62

Partial Dierential Equations

Igor Yanovsky, 2005

63

Problem (S95, #7). a) Solve the following equation, using characteristics,


ut + u3 ux = 0,

a(1 ex ),
u(x, 0) =
a(1 ex ),

for x < 0
for x > 0

where a > 0 is a constant. Determine the rst time when a shock forms.
Proof. a) is parameterized by : (s, 0, h(s)).
dz
dy
dx
= z3,
= 1,
= 0.
dt
dt
dt
x = h(s)3 t + s, y = t, z = h(s).
u(x, y) = h(x u3 y)

(11.6)

The characteristic projection in the xt-plane18


passing through the point (s, 0) is the line
x = h(s)3 t + s
along which u has a constant value u = h(s).
Characteristics x = h(s1 )3 t + s1 and x = h(s2 )3 t + s2 intersect at a point (x, t) with
s2 s1
.
t=
h(s2 )3 h(s1 )3
From (11.6), we have
ux = h (s)(1 3u2 ux t)

ux =

h (s)
1 + 3h(s)2 h (s)t

Hence for 3h(s)2 h (s) < 0, ux becomes innite at the positive time
t=

1
.
3h(s)2 h (s)

The smallest t for which this happens corresponds to the value s = s0 at which
h(s)2 h (s) has a minimum (i.e.h(s)2 h (s) has a maximum). At time T = 1/(3h(s0 )2 h (s0 ))
the solution u experiences a gradient catastrophe.
Therefore, need to nd a minimum of

3a2 (1 ex )2 aex
= 3a3 ex (1 ex )2 ,
x<0
f (x) = 3h(x)2 h (x) =
2
x 2 x
3 x
x 2
= 3a e (1 e ) , x > 0
3a (1 e ) ae




3a3 ex (1 ex )2 ex 2(1 ex )ex
x<0
= 3a3 ex (1 ex )(1 3ex ),


=0
f (x) =
3
x
x 2
x
x x
3 x
x
x
= 3a e (1 e )(1 + 3e ), x > 0
3a e (1 e ) + e 2(1 e )e

x = 0, x = ln 3, x < 0,

We check which ones give the minimum of f (x) :
The zeros of f (x) are
x = 0, x = ln 3, x > 0.


18

f (0) = 3a3 , f ( ln 3) = 3a3 13 (1 13 )2 = 4a9 ,


3
f (0) = 3a3 , f (ln 3) = 3a3 13 (1 13 )2 = 4a9 ,

y and t are interchanged here

x<0
x>0

Partial Dierential Equations

t=

Igor Yanovsky, 2005

1
1
1
=
= 3.
2

min{3h(s) h (s)}
min f (s)
3a

64

Partial Dierential Equations

Igor Yanovsky, 2005

65

b) Now consider
ut + u3 ux + cu = 0
with the same initial data and a positive constant c. How large does c need to be in
order to prevent shock formation?
b) Characteristic form: ut + u3 ux = cu.
dx
dt
dy
dt
dz
dt

z 3 = h(s)3 e3ct

y = t,

cz

z = h(s)ect

x=s+

: (s, 0, h(s)).
1
h(s)3 (1 e3ct ),
3c

( h(s) = uecy ).


1
z(s, t) = h x h(s)3 (1 e3ct ) ect ,
3c

1
u(x, y) = h x u3 e3cy (1 e3cy ) ecy .
3c
=


1
ux = h (s) ecy 1 u2 ux e3cy (1 e3cy ) ,
c
h (s)ecy
h (s)ecy
=
.
ux =
1 + 1c h (s)u2 e2cy (1 e3cy )
1 + 1c h (s)h(s)2 (1 e3cy )
Thus, we need
1
1 + h (s)h(s)2 (1 e3cy ) = 0.
c
We can perform further calculations taking into account the result from part (a):
min{3h(s)2 h (s)} = 3a3 .

Partial Dierential Equations

Igor Yanovsky, 2005

66

Problem (F99, #4). Consider the Cauchy problem


uy + a(x)ux = 0,
u(x, 0) = h(x).
Give an example of an (unbounded) smooth a(x) for which the solution of the Cauchy
problem is not unique.
Proof. is parameterized by : (s, 0, h(s)).
 t
 t
dx
= a(x) x(t) x(0) =
a(x)dt x =
a(x)dt + s,
dt
0
0
dy
= 1 y(s, t) = t + c1 (s) y = t,
dt
dz
= 0 z(s, t) = c2 (s)
z = h(s).
dt
Thus,
 y

a(x)dy
u(x, t) = h x
0

Problem (F97, #7). a) Solve the Cauchy problem


ut xuux = 0
u(x, 0) = f (x)

< x < , t 0,
< x < .

b) Find a class of initial data such that this problem has a global solution for all t.
Compute the critical time for the existence of a smooth solution for initial data, f ,
which is not in the above class.
Proof. a) is parameterized by : (s, 0, f (s)).
dx
dt
dy
dt
dz
dt

Check:


xz

dx
= xf (s)
dt

x = sef (s)t ,

y = t,

z = f (s).


z = f xef (s)t ,


u(x, y) = f xeuy .
=

ux = f  (s) (euy + xeuy ux y)


uy = f  (s) xeuy (uy y + u)

ux =
uy =

f (s)euy
1f  (s)xyeuy
f  (s)euy xu
1f (s)xyeuy

uy xuux =

ux f  (s)xeuy ux y = f  (s)euy
uy f  (s)xeuy uy y = f  (s)xeuy u
f  (s)euy xu
f  (s)euy

xu
= 0. 
1 f  (s)xyeuy
1 f  (s)xyeuy

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

Partial Dierential Equations

Igor Yanovsky, 2005

b) The characteristics would intersect when 1 f  (s)xyeuy = 0. Thus,


tc =

1
f  (s)xeutc

67

Partial Dierential Equations

Igor Yanovsky, 2005

68

Problem (F96, #6). Find an implicit formula for the solution u of the initial-value
problem
ut = (2x 1)tux + sin(x) t,
u(x, t = 0) = 0.
Evaluate u explicitly at the point (x = 0.5, t = 2).
Proof. Rewrite the equation as
uy + (1 2x)yux = sin(x) y.
is parameterized by : (s, 0, 0).
dx
dt
dy
dt
dz
dt

= (1 2x)y = (1 2x)t

= 1

x=

1
1
2
(2s 1)et + ,
2
2

1 2 1
s = (x )et +
,
2
2

y = t,

= sin(x) y = sin

(2s 1)et +

 t


t.
2



t dt + z(s, 0),
2
2
0
 t


2
(2s 1)et +
t dt.
sin
z(s, t) =
2
2
0 y 


2
(2s 1)ey +
y dy
sin
u(x, y) =
2
2
0 y 


2
2
(2x 1)ey ey +
y dy
sin
=
2
2
 y
0 y 



(2x 1) +
y dy =
sin
sin(x) y dy,
=
2
2
0
0
2
y
u(x, y) = y sin(x) .
2
Note: This solution does not satisfy the PDE.

z(s, t) =

sin

(2s 1)et +

Problem (S90, #8). Consider the Cauchy problem


ut = xux u,

< x < , t 0,
f (x) C .

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

Assume that f 0 for |x| 1.


Solve the equation by the method of characteristics and discuss the behavior of the
solution.
Proof. Rewrite the equation as
uy xux = u,
is parameterized by : (s, 0, f (s)).
dx
dt
dz
dt

x = set ,

z = f (s)et .

u(x, y) = f (xey )ey .

dy
=1
dt

y = t,

Partial Dierential Equations

Igor Yanovsky, 2005

The solution satises the PDE and initial conditions.


As y +, u 0. u = 0 for |xey | 1 u = 0 for |x|

1
ey .

69

Partial Dierential Equations

Igor Yanovsky, 2005

Problem (F02, #4). Consider the nonlinear hyperbolic equation


< x < .

uy + uux = 0

a) Find a smooth solution to this equation for initial condition u(x, 0) = x.


b) Describe the breakdown of smoothness for the solution if u(x, 0) = x.
Proof. a) is parameterized by : (s, 0, s).
dx
dt
dy
dt
dz
dt

= z=s

x = st + s

= 1

y = t,

= 0

z = s.

u(x, y) =

s=

x
x
=
.
t+1
y+1

x
; solution is smooth for all positive time y.
y+1

b) is parameterized by : (s, 0, s).


dx
dt
dy
dt
dz
dt

= z = s

x = st + s s =

= 1

y = t,

= 0

z = s.

u(x, y) =

x
x
=
.
1t
1y

x
; solution blows up at time y = 1.
y1

70

Partial Dierential Equations

Igor Yanovsky, 2005

71

Problem (F97, #4). Solve the initial-boundary value problem


ut + (x + 1)2 ux = x

for x > 0, t > 0

u(x, 0) = f (x)

0 < x < +

u(0, t) = g(t)

0 < t < +.

Proof. Rewrite the equation as


uy + (x + 1)2 ux = x

for x > 0, y > 0

u(x, 0) = f (x)

0 < x < +

u(0, y) = g(y)

0 < y < +.

For region I, we solve the following characteristic equations with is parameterized


19 by : (s, 0, f (s)).
dx
dt
dy
dt
dz
dt

(x + 1)2

x=

s+1
1,
(s + 1)t 1

y = t,

s+1
1,
(s + 1)t 1
z = ln|(s + 1)t 1| t + c1 (s),
=

x=

z = ln|(s + 1)t 1| t + f (s).


In region I, characteristics are of the form
s+1
1.
x=
(s + 1)y 1
Thus, region I is bounded above by the line
x
1
1,
or
y=
.
x=
y 1
x+1
xxyy
xy+y+1 ,

we have


x xy y

x xy y


+ 1 y 1 y + f
,
u(x, y) = ln
xy + y + 1
xy + y + 1



x xy y
1


.
u(x, y) = ln
y+f
xy + y + 1
xy + y + 1

Since t = y, s =

For region II, is parameterized by : (0, s, g(s)).


dx
dt
dy
dt
dz
dt

(x + 1)2

x=

1
1,
t1

y = t + s,

1
1,
t1
z = ln|t 1| t + c2 (s),
=

x=

z = ln|t 1| t + g(s).
19

Variable t as a third coordinate of u and variable t used to parametrize characteristic equations


are two dierent entities.

Partial Dierential Equations

Igor Yanovsky, 2005

x
s = y x+1
, we have
 x


x
x


u(x, y) = ln
1
+g y
.
x+1
x+1
x+1
x
, both solutions are equal if f (0) = g(0).
Note that on y = x+1

Since t =

x
x+1 ,

72

Partial Dierential Equations

Igor Yanovsky, 2005

73

Problem (S93, #3). Solve the following equation


ut + ux + yuy = sin t
for 0 t, 0 x, < y < and with
for t = 0, x 0

u=x+y
2

and

for x = 0, t 0.

u= t +y

Proof. Rewrite the equation as (x x1 , y x2 , t x3 ):


ux3 + ux1 + x2 ux2 = sin x3

for 0 x3 , 0 x1 , < x2 < ,

u(x1 , x2 , 0) = x1 + x2 ,
u(0, x2, x3 ) = x23 + x2 .
For region I, we solve the following characteristic equations with is parameterized
20
by : (s1 , s2 , 0, s1 + s2 ).
dx1
dt
dx2
dt
dx3
dt
dz
dt

x2

sin x3 = sin t

x1 = t + s1 ,
x2 = s2 et ,
x3 = t,

z = cos t + s1 + s2 + 1.

Since in region I, in x1 x3 -plane, characteristics are of the form x1 = x3 + s1 , region


I is bounded above by the line x1 = x3 . Since t = x3 , s1 = x1 x3 , s2 = x2 ex3 , we
have
u(x1 , x2 , x3 ) = cos x3 + x1 x3 + x2 ex3 + 1,
t

u(x, y, t) = cos t + x t + ye

+ 1,

or
x t.

For region II, we solve the following characteristic equations with is parameterized
by : (0, s2, s3 , s2 + s23 ).
dx1
= 1 x1 = t,
dt
dx2
= x2 x2 = s2 et ,
dt
dx3
= 1 x3 = t + s3 ,
dt
dz
= sin x3 = sin(t + s3 ) z = cos(t + s3 ) + cos s3 + s2 + s23 .
dt
Since t = x1 , s3 = x3 x1 , s2 = x2 ex3 , we have
u(x1 , x2 , x3 ) = cos x3 + cos(x3 x1 ) + x2 ex3 + (x3 x1 )2 ,
x t.
u(x, y, t) = cos t + cos(t x) + yet + (t x)2 ,

or

Note that on x = t, both solutions are u(x = t, y) = cos x + yex + 1.


20
Variable t as a third coordinate of u and variable t used to parametrize characteristic equations
are two dierent entities.

Partial Dierential Equations

Igor Yanovsky, 2005

74

Problem (W03, #5). Find a solution to


xux + (x + y)uy = 1
which satises u(1, y) = y for 0 y 1. Find a region in {x 0, y 0} where u is
uniquely determined by these conditions.
Proof. is parameterized by : (1, s, s).
dx
= x x = et .

dt
dy
= x + y y  y = et .
dt
dz
= 1 z = t + s.
dt
The homogeneous solution for the second equation is yh (s, t) = c1 (s)et. Since the
right hand side and yh are linearly dependent, our guess for the particular solution is
yp (s, t) = c2 (s)tet . Plugging in yp into the dierential equation, we get
c2 (s)tet + c2 (s)et c2 (s)tet = et

c2 (s) = 1.

Thus, yp (s, t) = tet and


y(s, t) = yh + yp = c1 (s)et + tet .
Since y(s, 0) = s = c1 (s), we get
y = set + tet .

With  and , we can solve for s and t in terms of x and y to get

y = sx + x ln x

t = ln x,
y x ln x
.
s =
x

u(x, y) = t + s = ln x +
u(x, y) =

y x ln x
.
x

y
.
x

We have found that the characteristics in the xy-plane are of the form
y = sx + x ln x,
where s is such that 0 s 1. Also, the characteristics originate from .
Thus, u is uniquely determined in the region between the graphs:
y = x ln x,
y = x + x ln x.

Partial Dierential Equations

12

Igor Yanovsky, 2005

Problems: Shocks

Example 1. Determine the exact solution to Burgers equation

1
u2 = 0,
t>0
ut +
2
x
with initial data

1
u(x, 0) = h(x) =
0

if x < 1,
if 1 < x < 1,
if x > 1.

Proof. Characteristic form: ut + uux = 0.


The characteristic projection in xt-plane passing through the point (s, 0) is the line
x = h(s)t + s.
Rankine-Hugoniot shock condition at s = 1:
shock speed:

 (t) =

F (ur ) F (ul )
=
ur ul

1 2
2 ur

0 12
12 u2l
1
= .
=
ur ul
01
2

The 1/slope of the shock curve = 1/2. Thus,


1
x = (t) = t + s,
2
and since the jump occurs at (1, 0), (0) = 1 = s. Therefore,
1
t 1.
2
Rankine-Hugoniot shock condition at s = 1:
x=

shock speed:

 (t) =

F (ur ) F (ul )
=
ur ul

1 2
2 ur

1
12 u2l
0
1
= .
= 2
ur ul
1 0
2

The 1/slope of the shock curve = 1/2. Thus,


1
x = (t) = t + s,
2
and since the jump occurs at (1, 0), (0) = 1 = s. Therefore,
1
x = t + 1.
2
At t = 2, Rankine-Hugoniot shock condition at s = 0:
shock speed:

 (t) =

F (ur ) F (ul )
=
ur ul

1 2
2 ur

1
12 u2l
1
= 2 2 = 0.
ur ul
1 1

The 1/slope of the shock curve = 0. Thus,


x = (t) = s,
and since the jump occurs at (x, t) = (0, 2), (2) = 0 = s. Therefore,
x = 0.

75

Partial Dierential Equations

For t < 2,

Igor Yanovsky, 2005

1
u(x, t) =
0

if x < 12 t 1,
if 12 t 1 < x < 12 t + 1,
if x > 12 t + 1.


and for t > 2,

u(x, t) =

1
1

if x < 0,
if x > 0.

76

Partial Dierential Equations

Igor Yanovsky, 2005

77

Example 2. Determine the exact solution to Burgers equation

1
u2 = 0,
t>0
ut +
2
x
with initial data

1
u(x, 0) = h(x) =
0

if x < 1,
if 1 < x < 1,
if x > 1.

Proof. Characteristic form: ut + uux = 0.


The characteristic projection in xt-plane passing through the point (s, 0) is the line
x = h(s)t + s.
For Burgers equation, for a rarefaction fan emanating from (s, 0) on xt-plane, we have:

xs

ul ,
t ul ,
xs
xs
u(x, t) =
ul t ur ,
t ,

xs
ur ,
t ur .

1,

x+1

t ,
u(x, t) =
0,

x1

t ,

1,

x < t 1,
t 1 < x < 1,
1 < x < 1,
1 < x < t + 1,
x > t + 1.

i.e. 1 <

x+1
t

<0

i.e.

x1
t

<1

0<

Partial Dierential Equations

Igor Yanovsky, 2005

78

Partial Dierential Equations

Igor Yanovsky, 2005

79

Example 3. Determine the exact solution to Burgers equation

1
u2 = 0,
t>0
ut +
2
x
with initial data

u(x, 0) = h(x) =

2
0

if 0 < x < 1,
if otherwise.

Proof. Characteristic form: ut + uux = 0.


The characteristic projection in xt-plane passing through the point (s, 0) is the line
x = h(s)t + s.
Shock: Rankine-Hugoniot shock condition at s = 1:
shock speed:

 (t) =

F (ur ) F (ul )
=
ur ul

1 2
2 ur

12 u2l
02
= 1.
=
ur ul
02

The 1/slope of the shock curve = 1. Thus,


x = (t) = t + s,
and since the jump occurs at (1, 0), (0) = 1 = s. Therefore,
x = t + 1.
Rarefaction: A rarefaction emanates from (0, 0) on xt-plane.

0
if x < 0,

x
if 0 < x < 2t,
t
For 0 < t < 1,
u(x, t) =

2
if 2t < x < t + 1.

0
if x > t + 1.
Rarefaction catches up to shock at t = 1.
Shock: At (x, t) = (2, 1), ul = x/t, ur = 0. Rankine-Hugoniot shock condition:
F (ur ) F (ul )
=
ur ul
x
dxs
=
,
dt
2t

x = c t,

 (t) =

1 2
2 ur

0 12 ( xt )2
12 u2l
1x
,
=
=
ur ul
0 xt
2t

and since the jump occurs at (x, t) = (2, 1), x(1) = 2 = c. Therefore, x = 2 t.

if x < 0,
0

x
For t > 1,
u(x, t) =
if 0 < x < 2 t,
t

0
if x > 2 t.

Partial Dierential Equations

Igor Yanovsky, 2005

80

Partial Dierential Equations

Igor Yanovsky, 2005

81

Example 4. Determine the exact solution to Burgers equation

1
u2 = 0,
t>0
ut +
2
x
with initial data


1+x
u(x, 0) = h(x) =
0

if x < 0,
if x > 0.

Proof. Characteristic form: ut + uux = 0.


The characteristic projection in xt-plane passing through the point (s, 0) is the line
x = h(s)t + s.
For s > 0, the characteristics are x = s.
For s < 0, the characteristics are x = (1 + s)t + s.
There are two ways to look for the solution on the left half-plane. One is to notice
that the characteristic at s = 0 is x = t and characteristic at s = 1 is x = 1 and
that characteristics between s = and s = 0 are intersecting at (x, t) = (1, 1).
Also, for a xed t, u is a linear function of x, i.e. for t = 0, u = 1 + x, allowing
a continuous change of u with x. Thus, the solution may be viewed as an implicit
rarefaction, originating at (1, 1), thus giving rise to the solution
u(x, t) =

x+1
.
t+1

Another way to nd a solution on the left half-plane is to solve for s to nd


s=

xt
.
1+t

Thus, u(x, t) = h(s) = 1 + s = 1 +

Shock: At (x, t) = (0, 0), ul =


 (t) =

F (ur ) F (ul )
=
ur ul

x+1
t+1 ,

xt
x+1
=
.
1+t
t+1

ur = 0. Rankine-Hugoniot shock condition:

2
0 12 ( x+1
12 u2l
1x+1
t+1 )
,
=
=
x+1
ur ul
2 t+1
0 t+1

1 2
2 ur

1x+1
dxs
=
,
dt
2 t+1

x = c t + 1 1,
and since the jump occurs
at (x, t) = (0, 0), x(0) = 0 = c 1, or c = 1. Therefore,
the shock curve is x = t + 1 1.


x+1
if x < t + 1 1,
t+1
u(x, t) =

0
if x > t + 1 1.

Partial Dierential Equations

Igor Yanovsky, 2005

82

Partial Dierential Equations

Igor Yanovsky, 2005

83

Example 5. Determine the exact solution to Burgers equation

1
u2 = 0,
t>0
ut +
2
x
with initial data

u0
u(x, 0) = h(x) = u0 (1 x)

if x < 0,
if 0 < x < 1,
if x 1,

where u0 > 0.
Proof. Characteristic form: ut + uux = 0.
The characteristic projection in xt-plane passing through the point (s, 0) is the line
x = h(s)t + s.
For s > 1, the characteristics are x = s.
For 0 < s < 1, the characteristics are x = u0 (1 s)t + s.
For s < 0, the characteristics are x = u0 t + s.
The characteristics emanating from (s, 0), 0 < s < 1 on xt-plane intersect at (1, u10 ).
Also, we can check that the characteristics do not intersect before t = u10 for this
problem:

1
1
= .
tc = min 
h (s)
u0
To nd solution in a triangular domain between x = u0 t and x = 1, we note that
characteristics there are x = u0 (1 s)t + s. Solving for s we get

x u0 t u0 (1 x)
x u0 t
.
Thus, u(x, t) = h(s) = u0 (1 s) = u0 1
=
.
s=
1 u0 t
1 u0 t
1 u0 t
We can also nd a solution in the triangular domain as follows. Note, that the characteristics are the straight lines
dx
= u = const.
dt
Integrating the equation above, we obtain
x = ut + c
Since all characteristics in the triangular domain meet at (1, u10 ), we have c = 1
and

u0 (1 x)
u
.
or
u=
x = ut + 1
u0
1 u0 t

if
x < u0 t,
u0
1
u0 (1x)
u(x, t) =
For 0 < t < ,
if u0 t < x < 1,
1u0 t

u0

0
if
x > 1.

u
u0 ,

Partial Dierential Equations

Igor Yanovsky, 2005

84

Shock: At (x, t) = (1, u10 ), Rankine-Hugoniot shock condition:


1 2
0 12 u20
u 1 u2
F (ur ) F (ul )
1
= 2 r 2 l =
= u0 ,
ur ul
ur ul
0 u0
2
1
(t) =
u0 t + c,
2
 
and since the jump occurs at (x, t) = (1, u10 ), x u10 = 1 = 12 + c, or c = 12 . Therefore,
the shock curve is x = u0 t+1
2 .

 (t) =

Partial Dierential Equations

1
,
For t >
u0

Igor Yanovsky, 2005


u(x, t) =

u0
0

if x <
if x >

85

u0 t+1
2 ,
u0 t+1
2 .

Problem. Show that for u = f (x/t) to be a nonconstant solution of ut + a(u)ux = 0,


f must be the inverse of the function a.
Proof. If u = f (x/t),
ut = f 

x x
2
t
t

and

ux = f 

x 1
.
t
t

Hence, ut + a(u)ux = 0 implies that


f 


x
x 1

x x
2 +a f
f
=0
t
t
t
t
t

or, assuming f  is not identically 0 to rule out the constant solution, that


x x
= .
a f
t
t
This shows the functions a and f to be inverses of each other.

Partial Dierential Equations

13

Igor Yanovsky, 2005

86

Problems: General Nonlinear Equations

13.1

Two Spatial Dimensions

Problem (S01, #3). Solve the initial value problem


x2
1 2
ux uy = ,
2
2
u(x, 0) = x.
You will nd that the solution blows up in nite time. Explain this in terms of the
characteristics for this equation.
Proof. Rewrite the equation as
F (x, y, z, p, q) =

x2
p2
q+
= 0.
2
2

is parameterized by : (s, 0, s, (s), (s)).


We need to complete to a strip. Find (s) and (s), the initial conditions for p(s, t)
and q(s, t), respectively:

F (f (s), g(s), h(s), (s), (s)) = 0,


F (s, 0, s, (s), (s)) = 0,
s2
(s)2
(s) +
= 0,
2
2
(s)2 + s2
.
(s) =
2

h (s) = (s)f  (s) + (s)g (s),


1 = (s).

(s) =

s2 + 1
.
2
2 +1

Therefore, now is parametrized by : (s, 0, s, 1, s


dx
dt
dy
dt
dz
dt
dp
dt
dq
dt
Thus, we


).

= Fp = p,
= Fq = 1

y(s, t) = t + c1 (s)

y = t,

= pFp + qFq = p2 q,
= Fx Fz p = x,
s2 + 1
.
2
found y and q in terms of s and t. Note that we have a coupled system:
= Fy Fz q = 0

q(s, t) = c2 (s)

q=

x = p,
p = x,

which can be written as two second order ODEs:


x + x = 0,

x(s, 0) = s,

x (s, 0) = p(s, 0) = 1,

p + p = 0,

p(s, 0) = 1,

p (s, 0) = x(s, 0) = s.

Partial Dierential Equations

Igor Yanovsky, 2005

87

Solving the two equations separately, we get


x(s, t) = s cos t + sin t,
p(s, t) = cos t s sin t.
From this, we get

2 s2 + 1
s2 + 1
dz
= p2 q =
= cos2 t 2s cos t sin t + s2 sin2 t
.
cos t s sin t
dt
2
2
 t
s2 + 1
dt + z(s, 0),
cos2 t 2s cos t sin t + s2 sin2 t
z(s, t) =
2
0
t
sin t cos t
s2 t s2 sin t cos t t(s2 + 1) t
z(s, t) =
+
+ s cos2 t +

+ s,
2
2
2
2
2
0
 sin t cos t
s2 sin t cos t t
+ s,
+ s cos2 t
=
2
2
0
sin t cos t
s2 sin t cos t
=
+ s cos2 t
s+s =
2
2
s2 sin t cos t
sin t cos t
+ s cos2 t
.
=
2
2
Plugging in x and y found earlier for s and t, we get
(x sin(y))2 sin(y) cos(y)
sin(y) cos(y) x sin(y)
+
cos2 (y)

2
cos(y)
cos2 (y)
2
2
(x + sin y) sin y cos y
sin y cos y x + sin y
+
cos2 y +

=
2
cos y
cos2 y
2
2
(x + sin y) sin y
sin y cos y
+ (x + sin y) cos y +
=
2
2 cos y
2
sin y cos y (x + sin y) sin y
+
.
= x cos y +
2
2 cos y

u(x, y) =

Partial Dierential Equations

Igor Yanovsky, 2005

88

Problem (S98, #3). Find the solution of


x2
u2x
=
,
2
2
u(x, 0) = h(x),

t 0, < x <

ut +

< x < ,

where h(x) is smooth function which vanishes for |x| large enough.
Proof. Rewrite the equation as
F (x, y, z, p, q) =

x2
p2
+q+
= 0.
2
2

is parameterized by : (s, 0, h(s), (s), (s)).


We need to complete to a strip. Find (s) and (s), the initial conditions for p(s, t)
and q(s, t), respectively:

F (f (s), g(s), h(s), (s), (s)) = 0,


F (s, 0, h(s), (s), (s)) = 0,
s2
(s)2
+ (s) +
= 0,
2
2
(s)2 + s2
.
(s) =
2

h (s) = (s)f  (s) + (s)g (s),


h (s) = (s).

(s) =

h (s)2 + s2
.
2


Therefore, now is parametrized by : (s, 0, s, h(s), h (s)2 +s ).


dx
dt
dy
dt
dz
dt
dp
dt
dq
dt
Thus, we


= Fp = p,
= Fq = 1

y(s, t) = t + c1 (s)

y = t,

= pFp + qFq = p2 + q,
= Fx Fz p = x,
h (s)2 + s2
.
2
found y and q in terms of s and t. Note that we have a coupled system:
= Fy Fz q = 0

q(s, t) = c2 (s)

q=

x = p,
p = x,

which can be written as a second order ODE:


x + x = 0,

x(s, 0) = s,

x (s, 0) = p(s, 0) = h (s).

Solving the equation, we get


x(s, t) = s cos t + h (s) sin t,
p(s, t) = x (s, t) = h (s) cos t s sin t.

Partial Dierential Equations

Igor Yanovsky, 2005

89

From this, we get



2 h (s)2 + s2
dz
= p2 + q = h (s) cos t s sin t
dt
2

h (s)2 + s2
= h (s)2 cos2 t 2sh (s) cos t sin t + s2 sin2 t
.
2
 t
h (s)2 + s2
dt + z(s, 0)
h (s)2 cos2 t 2sh (s) cos t sin t + s2 sin2 t
z(s, t) =
2
0
 t
h (s)2 + s2
dt + h(s).
h (s)2 cos2 t 2sh (s) cos t sin t + s2 sin2 t
=
2
0

We integrate the above expression similar to S  01#3 to get an expression for z(s, t).
Plugging in x and y found earlier for s and t, we get u(x, y).

Partial Dierential Equations

Igor Yanovsky, 2005

90

Problem (S97, #4).


Describe the method of the bicharacteristics for solving the initial value problem
2

2

u(x, y) +
u(x, y)
= 2 + y,
x
y
u(x, 0) = u0 (x) = x.

u0 (x)| < 2 and consider the solution such that u


Assume that | x
y > 0.
Apply all general computations for the particular case u0 (x) = x.

Proof. We have
u2x + u2y = 2 + y
u(x, 0) = u0 (x) = x.
Rewrite the equation as
F (x, y, z, p, q) = p2 + q 2 y 2 = 0.
is parameterized by : (s, 0, s, (s), (s)).
We need to complete to a strip. Find (s) and (s), the initial conditions for p(s, t)
and q(s, t), respectively:

F (f (s), g(s), h(s), (s), (s)) = 0,


F (s, 0, s, (s), (s)) = 0,
(s)2 + (s)2 2 = 0,
(s)2 + (s)2 = 2.

h (s) = (s)f  (s) + (s)g (s),


1 = (s).

(s) = 1.

Since we have a condition that q(s, t) > 0, we choose q(s, 0) = (s) = 1.


Therefore, now is parametrized by : (s, 0, s, 1, 1).
dx
dt
dy
dt
dz
dt

Fp = 2p

Fq = 2q

dx
=2
dt
dy
= 2t + 2
dt

x = 2t + s,

y = t2 + 2t,

dz
= 2t2 + 4t + 4,
dt
2
2
2
z = t3 + 2t2 + 4t + s = t3 + 2t2 + 4t + x 2t = t3 + 2t2 + 2t + x,
3
3
3
=

pFp + qFq = 2p2 + 2q 2 = 2y + 4

dp
= Fx Fz p = 0 p = 1,
dt
dq
= Fy Fz q = 1 q = t + 1.
dt
We solve y = t2 + 2t, a quadratic equation in t, t2 + 2t y = 0, for t in terms of y to
get:

t = 1 1 + y.



2

u(x, y) = (1 1 + y)3 + 2(1 1 + y)2 + 2(1 1 + y) + x.


3

Both u satisfy the PDE. ux = 1, uy = y + 1 u2x + u2y = y + 2 


u+ satises u+ (x, 0) = x . However, u does not satisfy IC, i.e. u (x, 0) = x 43 .

Partial Dierential Equations

Igor Yanovsky, 2005

91

Problem (S02, #6). Consider the equation


ux + ux uy = 1,
u(x, 0) = f (x).
Assuming that f is dierentiable, what conditions on f insure that the problem is
noncharacteristic? If f satises those conditions, show that the solution is
u(x, y) = f (r) y +

2y
,
f  (r)

where r must satisfy y = (f  (r))2(x r).


Finally, show that one can solve the equation for (x, y) in a suciently small neighborhood of (x0 , 0) with r(x0 , 0) = x0 .
Proof. Solved.
In order to solve the Cauchy problem in a neighborhood of , need:
f  (s) Fq [f, g, h, , ](s) g  (s) Fp [f, g, h, , ](s) = 0,

1 h (s)
= 0,
1 h (s) 0 1 +
h (s)
h (s) = 0.
Thus, h (s) = 0 ensures that the problem is noncharacteristic.
To show that one can solve y = (f  (s))2 (x s) for (x, y) in a suciently small
neighborhood of (x0 , 0) with s(x0 , 0) = x0 , let
G(x, y, s) = (f  (s))2 (x s) y = 0,
G(x0 , 0, x0) = 0,
Gr (x0 , 0, x0) = (f  (s))2 .
Hence, if f  (s) = 0, s, then Gs (x0 , 0, x0) = 0 and we can use the implicit function
theorem in a neighborhood of (x0 , 0, x0) to get
G(x, y, h(x, y)) = 0
and solve the equation in terms of x and y.

Partial Dierential Equations

Igor Yanovsky, 2005

92

Problem (S00, #1). Find the solutions of


(ux)2 + (uy )2 = 1
2

in a neighborhood of the curve y = x2 satisfying the conditions

x2

x2
=0
and
uy x,
> 0.
u x,
2
2
Leave your answer in parametric form.
Proof. Rewrite the equation as
F (x, y, z, p, q) = p2 + q 2 1 = 0.
2

is parameterized by : (s, s2 , 0, (s), (s)).


We need to complete to a strip. Find (s) and (s), the initial conditions for p(s, t)
and q(s, t), respectively:

F (f (s), g(s), h(s), (s), (s)) = 0,


 s2

F s, , 0, (s), (s) = 0,
2
2
(s) + (s)2 = 1.

h (s) = (s)f  (s) + (s)g (s),


0 = (s) + s(s),
(s) = s(s).

Thus, s2 (s)2 + (s)2 = 1

(s)2 =

s2

1
.
+1

Since, by assumption, (s) > 0, we have (s) = 12 .


s +1

 2
, s12 +1 .
Therefore, now is parametrized by : s, s2 , 0, ss
2 +1
dx
dt
dy
dt
dz
dt
dp
dt
dq
dt

2s
= Fp = 2p =
s2 + 1
2
= Fq = 2q =
2
s +1

2st
x=
+ s,
s2 + 1
2t
s2
y=
+ ,
2
s2 + 1

= pFp + qFq = 2p2 + 2q 2 = 2


= Fx Fz p = 0

= Fy Fz q = 0

Thus, in parametric form,


z(s, t) = 2t,
2st
+ s,
x(s, t) =
s2 + 1
s2
2t
+ .
y(s, t) =
2
s2 + 1

z = 2t,

s
,
s2 + 1
1
.
q=
s2 + 1
p=

Partial Dierential Equations

13.2

Igor Yanovsky, 2005

93

Three Spatial Dimensions

Problem (S96, #2). Solve the following Cauchy problem21 :


ux + u2y + u2z = 1,
u(0, y, z) = y z.
Proof. Rewrite the equation as
ux1 + u2x2 + u2x3 = 1,
u(0, x2, x3 ) = x2 x3 .
Write a general nonlinear equation
F (x1 , x2 , x3, z, p1 , p2, p3 ) = p1 + p22 + p23 1 = 0.
is parameterized by

,
s1
:
0



s2 , s1 s2 , 1 (s1 , s2 ), 2 (s1 , s2 ), 3 (s1 , s2 )



         

x1 (s1 ,s2 ,0) x2(s1 ,s2 ,0) x3(s1 ,s2 ,0) z(s1 ,s2 ,0) p1 (s1 ,s2 ,0)

p2 (s1 ,s2 ,0)

p3 (s1 ,s2 ,0)

We need to complete to a strip. Find 1 (s1 , s2 ), 2 (s1 , s2 ), and 3 (s1 , s2 ), the initial
conditions for p1 (s1 , s2 , t), p2 (s1 , s2 , t), and p3 (s1 , s2 , t), respectively:



F f1 (s1 , s2 ), f2 (s1 , s2 ), f3 (s1 , s2 ), h(s1 , s2 ), 1 , 2 , 3 = 0,




F 0, s1 , s2 , s1 s2 , 1 , 2 , 3 = 1 + 22 + 23 1 = 0,

1 + 22 + 23 = 1.

h
s1

h
s2

f1
f2
f3
+ 2
+ 3
,
s1
s1
s1
s2 = 2 .
f1
f2
f3
= 1
+ 2
+ 3
,
s2
s2
s2
s1 = 3 .
= 1

Thus, we have: 2 = s2 , 3 = s1 , 1 = s21 s22 + 1.



,
s1 ,
s2 , s1 s2 , s21 s22 + 1,
:
0



 


x1 (s1 ,s2 ,0) x2(s1 ,s2 ,0) x3(s1 ,s2 ,0) z(s1 ,s2 ,0)

21

p1 (s1 ,s2 ,0)

s2 ,


s1


p2(s1 ,s2 ,0) p3(s1 ,s2 ,0)

This problem is very similar to an already hand-written solved problem F95 #2.

Partial Dierential Equations

Igor Yanovsky, 2005

94

The characteristic equations are


dx1
dt
dx2
dt
dx3
dt
dz
dt

= Fp1 = 1

x1 = t,

= Fp2 = 2p2

= Fp3 = 2p3

dx2
= 2s2
dt
dx3
= 2s1
dt

x2 = 2s2 t + s1 ,

x3 = 2s1 t + s2 ,

= p1 Fp1 + p2 Fp2 + p3 Fp3 = p1 + 2p22 + 2p23 = s21 s22 + 1 + 2s22 + 2s21


= s21 + s22 + 1

z = (s21 + s22 + 1)t + s1 s2 ,

dp1
= Fx1 p1 Fz = 0 p1 = s21 s22 + 1,
dt
dp2
= Fx2 p2 Fz = 0 p2 = s2 ,
dt
dp3
= Fx3 p3 Fz = 0 p3 = s1 .
dt
Thus, we have

t = x1
=
t
x

s = x 2s t
x = 2s t + s
2
2
1
1
2
2

x3 = 2s1 t + s2
s2 = x3 2s1 t

z = (s2 + s2 + 1)t + s s
z = (s2 + s2 + 1)t + s s
1 2
1 2
1
2
1
2

u(x1 , x2 , x3 ) =

2x1 x2

s2 = x314x

z = (s21 + s22 + 1)t + s1 s2

x 2x x
x 2x x
x2 2x1 x3 2
x3 2x1 x2 2
2
1 3
3
1 2
+
+
1
x
+
.
1
1 4x21
1 4x21
1 4x21
1 4x21

Problem (F95, #2). Solve the following Cauchy problem


ux + uy + u3z = x + y + z,
u(x, y, 0) = xy.
Proof. Solved

t = x1

s1 = x2 2x12x3
14x

Partial Dierential Equations

Igor Yanovsky, 2005

95

Problem (S94, #1). Solve the following PDE for f (x, y, t):
ft + xfx + 3t2 fy = 0
f (x, y, 0) = x2 + y 2 .
Proof. Rewrite the equation as (x x1 , y x2 , t x3 , f u):
x1 ux1 + 3x23 ux2 + ux3 = 0,

u(x1 , x2, 0) = x21 + x22 .

F (x1 , x2 , x3, z, p1 , p2, p3 ) = x1 p1 + 3x23 p2 + p3 = 0.


is parameterized by

s2
:
s1 ,



0
, s21 + s22 , 1 (s1 , s2 ), 2 (s1 , s2 ), 3 (s1 , s2 )

           

x1 (s1 ,s2 ,0) x2(s1 ,s2 ,0) x3(s1 ,s2 ,0) z(s1 ,s2 ,0) p1 (s1 ,s2 ,0)

p2 (s1 ,s2 ,0)

p3 (s1 ,s2 ,0)

We need to complete to a strip. Find 1 (s1 , s2 ), 2 (s1 , s2 ), and 3 (s1 , s2 ), the initial
conditions for p1 (s1 , s2 , t), p2 (s1 , s2 , t), and p3 (s1 , s2 , t), respectively:



F f1 (s1 , s2 ), f2 (s1 , s2 ), f3 (s1 , s2 ), h(s1 , s2 ), 1 , 2 , 3 = 0,




F s1 , s2 , 0, s21 + s22 , 1 , 2 , 3 = s1 1 + 3 = 0,

h
s1

h
s2

3 = s1 1 .
f1
f2
f3
= 1
+ 2
+ 3
,
s1
s1
s1
2s1 = 1 .
f1
f2
f3
= 1
+ 2
+ 3
,
s2
s2
s2
2s2 = 2 .

Thus, we have: 1 = 2s1 , 2 = 2s2 , 3 = 2s21 .



s2 , 
0
, s21 + s22 , 2s1 ,
:
s1 ,



  

2s2 ,


2s21


x1 (s1 ,s2 ,0) x2(s1 ,s2 ,0) x3(s1 ,s2 ,0) z(s1 ,s2 ,0) p1(s1 ,s2 ,0) p2 (s1 ,s2 ,0) p3 (s1 ,s2 ,0)

The characteristic equations are


dx1
dt
dx2
dt
dx3
dt
dz
dt
dp1
dt
dp2
dt
dp3
dt

= Fp1 = x1

= Fp2 = 3x23
= Fp3 = 1

x1 = s1 et ,
dx2
= 3t2
dt

x2 = t3 + s2 ,

x3 = t,

= p1 Fp1 + p2 Fp2 + p3 Fp3 = p1 x1 + p2 3x23 + p3 = 0


= Fx1 p1 Fz = p1
= Fx2 p2 Fz = 0

= Fx3 p3 Fz = 6x3 p2

z = s21 + s22 ,

p1 = 2s1 et ,
p2 = 2s2 ,

dp3
= 12ts2
dt

p3 = 6t2 s2 + 2s21 .

With t = x3 , s1 = x1 ex , s2 = x2 x33 , we have


3
f (x, y, t) = x2 e2t + (y t3 )2 .
u(x1 , x2 , x3 ) = x21 e2x + (x2 x33 )2 .
The solution satises the PDE and initial condition.

Partial Dierential Equations

Igor Yanovsky, 2005

96

Problem (F93, #3). Find the solution of the following equation


ft + xfx + (x + t)fy = t3
f (x, y, 0) = xy.
Proof. Rewrite the equation as (x x1 , y x2 , t x3 , f u):
x1 ux1 + (x1 + x3 )ux2 + ux3 = x3 ,
u(x1 , x2 , 0) = x1 x2 .
Method I: Treat the equation as a QUASILINEAR equation.
is parameterized by : (s1 , s2 , 0, s1s2 ).
dx1
dt
dx2
dt
dx3
dt
dz
dt

= x1

= x1 + x3
= 1
= x33

x1 = s1 et ,
dx2
= s1 et + t
dt

x2 = s1 et +

t2
+ s2 s1 ,
2

x3 = t,
dz
= t3
dt

z=

Since t = x3 , s1 = x1 ex3 , s2 = x2 s1 et

t4
+ s1 s2 .
4

t2
2

+ s1 = x2 x1

x43
x2
+ x1 ex3 (x2 x1 3 + x1 ex3 ),
4
2
4
2
t
t
+ xet (y x + xet ).
f (x, y, t) =
4
2
The solution satises the PDE and initial condition.
u(x1 , x2 , x3 ) =

x23
2

+ x1 ex3 , we have

or

Method II: Treat the equation as a fully NONLINEAR equation.


F (x1 , x2 , x3, z, p1 , p2, p3 ) = x1 p1 + (x1 + x3 )p2 + p3 x33 = 0.
is parameterized by

s2
:
s1 ,



0
, s1 s2 , 1 (s1 , s2 ), 2 (s1 , s2 ), 3 (s1 , s2 )

         

x1 (s1 ,s2 ,0) x2(s1 ,s2 ,0) x3(s1 ,s2 ,0) z(s1 ,s2 ,0) p1 (s1 ,s2 ,0)

p2 (s1 ,s2 ,0)

p3 (s1 ,s2 ,0)

We need to complete to a strip. Find 1 (s1 , s2 ), 2 (s1 , s2 ), and 3 (s1 , s2 ), the initial
conditions for p1 (s1 , s2 , t), p2 (s1 , s2 , t), and p3 (s1 , s2 , t), respectively:



F f1 (s1 , s2 ), f2 (s1 , s2 ), f3 (s1 , s2 ), h(s1 , s2 ), 1 , 2 , 3 = 0,




F s1 , s2 , 0, s1s2 , 1 , 2 , 3 = s1 1 + s1 2 + 3 = 0,

h
s1

h
s2

3 = s1 (1 + 2 ).
f1
f2
f3
= 1
+ 2
+ 3
,
s1
s1
s1
s2 = 1 .
f1
f2
f3
= 1
+ 2
+ 3
,
s2
s2
s2
s1 = 2 .

Thus, we have: 1 = s2 , 2 = s1 , 3 = s21 s1 s2 .



s2 , 
0
, s1 s2 ,
s2 ,
:
s1 ,





s1 , s21 s1 s2

  

x1 (s1 ,s2 ,0) x2(s1 ,s2 ,0) x3(s1 ,s2 ,0) z(s1 ,s2 ,0) p1(s1 ,s2 ,0) p2 (s1 ,s2 ,0)

p3 (s1 ,s2 ,0)

Partial Dierential Equations

Igor Yanovsky, 2005

97

The characteristic equations are


dx1
dt
dx2
dt
dx3
dt
dz
dt
dp1
dt
dp2
dt
dp3
dt

= Fp1 = x1

x1 = s1 et ,

= Fp3 = 1

dx2
= s1 et + t
dt

= Fp2 = x1 + x3

x2 = s1 et +

x3 = t,

= p1 Fp1 + p2 Fp2 + p3 Fp3 = p1 x1 + p2 (x1 + x3 ) + p3 = x33 = t3


= Fx1 p1 Fz = p1 p2 = p1 s1
= Fx2 p2 Fz = 0

22

z=

t4
+ s1 s2 ,
4

p1 = 2s1 et s1 ,

p3 = t3 s1 t s21 s1 s2 .

x43
x2
3
3
+ x1 ex (x2 x1 3 + x1 ex ),
4
2
2
t
t4
+ xet (y x + xet ).
f (x, y, t) =
4
2
The solution satises the PDE and initial condition.
u(x1 , x2 , x3 ) =

p2 = s1 ,

= Fx3 p3 Fz = 3x23 p2 = 3t2 s1


3

With t = x3 , s1 = x1 ex , s2 = x2 s1 et t2 + s1 = x2 x1

22

t2
+ s2 s1 ,
2

x23
2

+ x1 ex , we have

or

Variable t in the derivatives of characteristics equations and t in the solution f (x, y, t) are dierent
entities.

Partial Dierential Equations

Igor Yanovsky, 2005

98

Problem (F92, #1). Solve the initial value problem


ut + ux + uy + u = 0

for t > 0

u(x, y, 0) = (x, y),


in which , and are real constants and is a smooth function.
Proof. Rewrite the equation as (x x1 , y x2 , t x3 )23 :
ux1 + ux2 + ux3 = u,
u(x1 , x2 , 0) = (x1 , x2 ).
is parameterized by : (s1 , s2 , 0, (s1, s2 )).
dx1
dt
dx2
dt
dx3
dt
dz
dt

x1 = t + s1 ,

x2 = t + s2 ,

= 1

x3 = t,

dz
= dt z = (s1 , s2 )et .
z



(x , x , x )  1 0 0 
1 2 3
=  0 1 0  = 1 = 0
J is invertible.
J det
(s1 , s2 , t)
 1 
= z

Since t = x3 , s1 = x1 x3 , s2 = x2 x3 , we have
u(x1 , x2 , x3 ) = (x1 x3 , x2 x3 )ex3 ,
u(x, y, t) = (x t, y t)et .

or

The solution satises the PDE and initial condition.24


23
Variable t as a third coordinate of u and variable t used to parametrize characteristic equations
are two dierent entities.
f
g
24
Chain Rule: u(x1 , x2 , x3 ) = (f (x1, x2 , x3 ), g(x1 , x2 , x3 )), then ux1 =
+
.
f x1
g x1

Partial Dierential Equations

Igor Yanovsky, 2005

99

Problem (F94, #2). Find the solution of the Cauchy problem


ut (x, y, t) + aux (x, y, t) + buy (x, y, t) + c(x, y, t)u(x, y, t) = 0
u(x, y, 0) = u0 (x, y),
where 0 < t < +, < x < +, < y < +,
a, b are constants, c(x, y, t) is a continuous function of (x, y, t), and u0 (x, y) is a continuous function of (x, y).
Proof. Rewrite the equation as (x x1 , y x2 , t x3 ):
aux1 + bux2 + ux3 = c(x1 , x2 , x3 )u,
u(x1 , x2 , 0) = u0 (x1 , x2).
is parameterized by : (s1 , s2 , 0, u0(s1 , s2 )).
dx1
dt
dx2
dt
dx3
dt
dz
dt

x1 = at + s1 ,

x2 = bt + s2 ,

x3 = t,

dz
c(x1 , x2 , x3 )z
= c(at + s1 , bt + s2 , t)z
dt
 t
c(a + s1 , b + s2 , )d + c1 (s1 , s2 ),
ln z =
=

z(s1 , s2 , t) = c2 (s1 , s2 )e

t
0

c(a+s1 ,b+s2 ,)d

t

z(s1 , s2 , t) = u0 (s1 , s2 )e 0 c(a+s1 ,b+s2 ,)d .





(x , x , x )  1 0 0 
1 2 3
=  0 1 0  = 1 = 0
J det
(s1 , s2 , t)
 a b 1 

dz
= c(at + s1 , bt + s2 , t)dt
z

z(s1 , s2 , 0) = c2 (s1 , s2 ) = u0 (s2 , s2 ),

J is invertible.

Since t = x3 , s1 = x1 ax3 , s2 = x2 bx3 , we have


u(x1 , x2 , x3 ) = u0 (x1 ax3 , x2 bx3 )e
= u0 (x1 ax3 , x2 bx3 )e

u(x, y, t) = u0 (x at, y bt)e

t
0

 x3
0

c(a+x1 ax3 ,b+x2 bx3 ,)d

c(x1 +a(x3 ),x2+b(x3 ),)d

 x3

c(x+a(t),y+b(t),)d

or

Partial Dierential Equations

Igor Yanovsky, 2005

100

Problem (F89, #4). Consider the rst order partial dierential equation
ut + ( + t)ux + etuy = 0

(13.1)

in which , and are constants.


a) For this equation, solve the initial value problem with initial data
u(x, y, t = 0) = sin(xy)

(13.2)

for all x and y and for t 0.


b) Suppose that this initial data is prescribed only for x 0 (and all y) and consider
(13.1) in the region x 0, t 0 and all y. For which values of , and is it possible
to solve the initial-boundary value problem (13.1), (13.2) with u(x = 0, y, t) given for
t 0?
For non-permissible values of , and , where can boundary values be prescribed in
order to determine a solution of (13.1) in the region (x 0, t 0, all y).
Proof. a) Rewrite the equation as (x x1 , y x2 , t x3 ):
( + x3 )ux1 + ex3 ux2 + ux3 = 0,
u(x1 , x2 , 0) = sin(x1 x2 ).
is parameterized by : (s1 , s2 , 0, sin(s1 s2 )).
dx1
dt
dx2
dt
dx3
dt
dz
dt

= ex3

dx1
= + t
dt

= + x3

dx2
= et
dt

= 1

x3 = t,

= 0

z = sin(s1 s2 ).

(x , x , x ) 
1 2 3
= 
J det
(s1 , s2 , t)

Since t = x3 , s1 = x1

x23
2

x1 =

t2
+ t + s1 ,
2

x2 = et + s2 ,


1
0 0 
0
1 0  = 1 = 0
t + et 1 

J is invertible.

x3 , s2 = x2 ex3 + , we have

x23
x3 )(x2 ex3 + )),
2
t2
t)(y et + )).
u(x, y, t) = sin((x
2
The solution satises the PDE and initial condition.
u(x1 , x2 , x3 ) = sin((x1

or

b) We need a compatibility condition between the initial and boundary values to hold
on y-axis (x = 0, t = 0):
u(x = 0, y, 0) = u(0, y, t = 0),
0 = 0.

Partial Dierential Equations

Igor Yanovsky, 2005

101

Partial Dierential Equations

14

Igor Yanovsky, 2005

102

Problems: First-Order Systems




Problem (S01, #2a). Find the solution u =

u1 (x, t)
u2 (x, t)


, (x, t) R R,

to the (strictly) hyperbolic equation




1 0
ux = 0,
ut
5 3

satisfying

u1 (x, 0)
u2 (x, 0)


=

eixa
0


, a R.

Proof. Rewrite the equation as




1 0
Ux = 0,
Ut +
5 3
  ixa 
 (1)
e
u (x, 0)
=
.
U (x, 0) =
0
u(2)(x, 0)
The eigenvalues of thematrix
 A are
2
, e2 =
eigenvectors are e1 =
5



1 0
2
=
,
=
0 3
5

1 = 1, 2 = 3 and the corresponding


0
. Thus,
1
 1


1
0
0
1
2
=
.
,
=
5
1
det
2 1

Let U = V . Then,
Ut + AUx = 0,
Vt + AVx = 0,
Vt + 1 AVx = 0,
Vt + Vx = 0.
Thus, the transformed problem is


1 0
Vx = 0,
Vt +
0 3

1
V (x, 0) = U (x, 0) =

1
2
5
2

0
1



eixa
0

1
= eixa
2

1
5


.

We have two initial value problems




(1)
(1)
(2)
(2)
vt 3vx = 0,
vt vx = 0,
v (1)(x, 0) = 12 eixa;
v (2)(x, 0) = 52 eixa ,
which we solve by characteristics to get
1
5
v (2)(x, t) = eia(x+3t).
v (1)(x, t) = eia(x+t),
2
2

 (1)  
2 0
v
=
We solve for U :
U = V =
5 1
v (2)

Thus,

U=

u(1)(x, t)
u(2)(x, t)


=

eia(x+t)
52 eia(x+t) + 52 eia(x+3t)

1 ia(x+t)
2e
5 ia(x+3t)
2e


.


.

Can check that this is the correct solution by plugging it into the original equation.

Partial Dierential Equations

Igor Yanovsky, 2005

Part (b) of the problem is solved in the Fourier Transform section.

103

Partial Dierential Equations

Igor Yanovsky, 2005

104

Problem (S96, #7). Solve the following initial-boundary


value problem in the do (1) 
u
:
main x > 0, t > 0, for the unknown vector U =
u(2)


2 3
Ux = 0.
(14.1)
Ut +
0 1


sin x
U (x, 0) =
and u(2)(0, t) = t.
0
Proof. The eigenvaluesof the
 matrix A are
 1 = 2, 2 = 1 and the corresponding
1
1
, e2 =
. Thus,
eigenvectors are e1 =
0
1






1
1 1
2 0
1 1
1
=
.
=
,
=
,
=
0 1
0 1
0 1
det
Let U = V . Then,
Ut + AUx = 0,
Vt + AVx = 0,
Vt + 1 AVx = 0,
Vt + Vx = 0.
Thus, the transformed problem is


2 0
Vx = 0,
Vt +
0 1


 

1 1
sin x
sin x
1
=
.
V (x, 0) = U (x, 0) =
0 1
0
0

(14.2)
(14.3)

Equation (14.2) gives traveling wave solutions of the form


v (1)(x, t) = F (x + 2t),

v (2)(x, t) = G(x t).

We can write U in terms of V :



 


  (1)  
1 1
F (x + 2t)
F (x + 2t) + G(x t)
1 1
v
=
=
.
U = V =
0 1
G(x t)
G(x t)
0 1
v (2)
(14.4)

Partial Dierential Equations

Igor Yanovsky, 2005

105

For region I, (14.2) and (14.3) give two initial value problems (since any point in
region I can be traced back along both characteristics to initial conditions):


(1)
(1)
(2)
(2)
vt + vx = 0,
vt 2vx = 0,
v (1)(x, 0) = sin x;
v (2)(x, 0) = 0.
which we solve by characteristics to get traveling wave solutions:
v (1)(x, t) = sin(x + 2t),

v (2)(x, t) = 0.

Thus, for region I,



U = V =

1 1
0 1



sin(x + 2t)
0


=

sin(x + 2t)
0


.

For region II, solutions of the form F (x + 2t) can be traced back to initial conditions.
Thus, v (1) is the same as in region I. Solutions of the form G(x t) can be traced back
to the boundary. Since from (14.4), u(2) = v (2), we use boundary conditions to get
u(2)(0, t) = t = G(t).
Hence, G(x t) = (x t).
Thus, for region II,

U = V =

1 1
0 1



sin(x + 2t)
(x t)


=

sin(x + 2t) (x t)
(x t)

Solutions for regions I and II satisfy (14.1).


Solution for region I satises both initial conditions.
Solution for region II satises given boundary condition.


.

Partial Dierential Equations

Igor Yanovsky, 2005

106

Problem (S02, #7). Consider the system


  

 

u
1 2
u
=
.
v
2
2
t
x v

(14.5)

Find an explicit solution for the following mixed problem for the system (14.5):




u(x, 0)
f (x)
=
for x > 0,
v(x, 0)
0
u(0, t) = 0
for t > 0.
You may assume that the function f is smooth and vanishes on a neighborhood of x = 0.
Proof. Rewrite the equation as


1 2
Ux = 0,
Ut +
2 2
 

 (1)
f (x)
u (x, 0)
=
.
U (x, 0) =
0
u(2)(x, 0)
The eigenvalues ofthe matrix A
 are
1
2
, e2 =
vectors are e1 =
2
1



3 0
1
=
,
=
0 2
2

1 = 3, 2 = 2 and the corresponding eigen. Thus,


2
1


,

1
1
=
=
det
5

1 2
2 1


.

Let U = V . Then,
Ut + AUx = 0,
Vt + AVx = 0,
Vt + 1 AVx = 0,
Vt + Vx = 0.
Thus, the transformed problem is


3 0
Vx = 0,
Vt +
0 2
1

V (x, 0) =

1
U (x, 0) =
5

(14.6)
1 2
2 1



f (x)
0

f (x)
=
5

1
2


.

(14.7)

Equation (14.6) gives traveling wave solutions of the form:


v (1)(x, t) = F (x + 3t),

v (2)(x, t) = G(x 2t).

(14.8)

We can write U in terms of V :



 


  (1)  
1 2
F (x + 3t)
F (x + 3t) 2G(x 2t)
1 2
v
=
=
.
U = V =
2 1
G(x 2t)
2F (x + 3t) + G(x 2t)
2 1
v (2)
(14.9)

Partial Dierential Equations

Igor Yanovsky, 2005

107

For region I, (14.6) and (14.7) give two initial value problems (since value at any
point in region I can be traced back along both characteristics to initial conditions):


(1)
(1)
(2)
(2)
vt 3vx = 0,
vt + 2vx = 0,
v (1)(x, 0) = 15 f (x);
v (2)(x, 0) = 25 f (x).
which we solve by characteristics to get traveling wave solutions:
1
2
v (2)(x, t) = f (x 2t).
v (1)(x, t) = f (x + 3t),
5
5
  1


 1
f (x + 3t)
f (x + 3t) + 45 f (x 2t)
1 2
5
5
=
.
Thus, for region I, U = V =
2
2
25 f (x 2t)
2 1
5 f (x + 3t) 5 f (x 2t)
For region II, solutions of the form F (x + 3t) can be traced back to initial conditions.
Thus, v (1) is the same as in region I. Solutions of the form G(x 2t) can be traced back
to the boundary. Since from (14.9),
u(1) = v (1) 2v (2),

we have

1
u(1)(x, t) = F (x + 3t) 2G(x 2t) = f (x + 3t) 2G(x 2t).
5
The boundary condition gives
1
u(1)(0, t) = 0 = f (3t) 2G(2t),
5
1
2G(2t) = f (3t),
5
1
3
G(t) = f t ,
10
2

1
3
f (x 2t) .
G(x 2t) =
10
2
 


1
1 2
5 f (x + 3t)
=
Thus, for region II, U = V =
1
3
2 1
10 f ( 2 (x 2t))
Solutions for regions I and II satisfy (14.5).
Solution for region I satises both initial conditions.
Solution for region II satises given boundary condition.

1
1
3
5 f (x + 3t) 5 f ( 2 (x 2t))
2
1
3
5 f (x + 3t) + 10 f ( 2 (x 2t))


.

Partial Dierential Equations

Igor Yanovsky, 2005

108

Problem (F94, #1; S97, #7). Solve the initial-boundary value problem
ut + 3vx = 0,
vt + ux + 2vx = 0
in the quarter plane 0 x, t < , with initial conditions
u(x, 0) = 1 (x),

v(x, 0) = 2 (x),

25

0 < x < +

and boundary condition


u(0, t) = (t),

t > 0.

Proof. Rewrite the equation as Ut + AUx = 0:




0 3
Ux = 0,
Ut +
1 2

U (x, 0) =

u(1)(x, 0)
u(2)(x, 0)


=

1 (x)
2 (x)

(14.10)


.

The eigenvalues ofthe matrix


A are 1 = 1, 2 = 3 and the corresponding eigen
3
1
, e2 =
. Thus,
vectors are e1 =
1
1






1 1 1
1
1 0
3 1
1
=
.
=
,
=
,
=
1 3
0 3
1 1
det
4
Let U = V . Then,
Ut + AUx = 0,
Vt + AVx = 0,
Vt + 1 AVx = 0,
Vt + Vx = 0.
Thus, the transformed problem is


1 0
Vx = 0,
Vt +
0 3
1

V (x, 0) =

1
U (x, 0) =
4

1 1
1 3

(14.11)


1 (x)
2 (x)

1
=
4

1 (x) + 2 (x)
1 (x) + 32 (x)


. (14.12)

Equation (14.11) gives traveling wave solutions of the form:


v (1)(x, t) = F (x + t),

v (2)(x, t) = G(x 3t).

(14.13)

We can write U in terms of V :



 


  (1)  
3 1
F (x + t)
3F (x + t) + G(x 3t)
3 1
v
=
=
.
U = V =
1 1
G(x 3t)
F (x + t) + G(x 3t)
1 1
v (2)
(14.14)
25

In S97, #7, the zero initial conditions are considered.

Partial Dierential Equations

Igor Yanovsky, 2005

109

For region I, (14.11) and (14.12) give two initial value problems (since value at
any point in region I can be traced back along characteristics to initial conditions):
 (1)
 (2)
(1)
(2)
vt vx = 0,
vt + 3vx = 0,
v (1)(x, 0) = 14 1 (x) + 14 2 (x);
v (2)(x, 0) = 14 1 (x) + 34 2 (x),
which we solve by characteristics to get traveling wave solutions:
1
1
1
3
v (2)(x, t) = 1 (x 3t) + 2 (x 3t).
v (1)(x, t) = 1 (x + t) + 2 (x + t),
4
4
4
4
Thus, for region I,


 1
3 1
4 1 (x + t) + 14 2 (x + t)
U = V =
1
3
1 1
4 1 (x 3t) + 4 2 (x 3t)


1 31 (x + t) 32 (x + t) + 1 (x 3t) + 32 (x 3t)
.
=
1 (x + t) + 2 (x + t) + 1 (x 3t) + 32 (x 3t)
4
For region II, solutions of the form F (x + t) can be traced back to initial conditions.
Thus, v (1) is the same as in region I. Solutions of the form G(x 3t) can be traced back
to the boundary. Since from (14.14),
we have
u(1) = 3v (1) + v (2),
3
3
u(1)(x, t) = 1 (x + t) 2 (x + t) + G(x 3t).
4
4
The boundary condition gives
3
3
u(1)(0, t) = (t) = 1 (t) 2 (t) + G(3t),
4
4
3
3
G(3t) = (t) 1 (t) + 2 (t),
4
4

t 3
t 3
t
1
+ 2
,
G(t) =
3
4
3
4
3

x 3t 3
x 3t 3
x 3t
1
+ 2
.
G(x 3t) =
3
4
3
4
3
Thus, for region II,



3 1
14 1 (x + t) + 14 2 (x + t)
U = V =
3
x3t
3
x3t
( x3t
1 1
3 ) 4 1 ( 3 ) + 4 2 ( 3 )

 3
3
x3t
3
x3t
3
x3t
4 1 (x + t) 4 2 (x + t) + ( 3 ) 4 1 ( 3 ) + 4 2 ( 3 )
.
=
3
x3t
3
x3t
14 1 (x + t) + 14 2 (x + t) + ( x3t
3 ) 4 1 ( 3 ) + 4 2 ( 3 )

Partial Dierential Equations

Igor Yanovsky, 2005

Solutions for regions I and II satisfy (14.10).


Solution for region I satises both initial conditions.
Solution for region II satises given boundary condition.

110

Partial Dierential Equations

Igor Yanovsky, 2005

111

Problem (F91, #1). Solve explicitly the following initial-boundary value problem for
linear 22 hyperbolic system
ut = ux + vx
vt = 3ux vx ,
where 0 < t < +, 0 < x < + with initial conditions
u(x, 0) = u0 (x),

v(x, 0) = v0 (x),

0 < x < +,

and the boundary condition


u(0, t) + bv(0, t) = (t),

0 < t < +,

where b = 13 is a constant.
What happens when b = 13 ?
Proof. Let us change the notation (u u(1), v u(2)). Rewrite the equation as


1 1
Ux = 0,
(14.15)
Ut +
3 1
!

 (1)
(1)
u0 (x)
u (x, 0)
.
=
U (x, 0) =
(2)
u(2)(x, 0)
u0 (x)
The eigenvalues ofthe matrix A
 are
1
1
, e2 =
vectors are e1 =
1
3



2 0
1
=
,
=
0 2
1

1 = 2, 2 = 2 and the corresponding eigen. Thus,


1
3


,

1 =

1
4

3 1
1 1


.

Let U = V . Then,
Ut + AUx = 0,
Vt + AVx = 0,
Vt + 1 AVx = 0,
Vt + Vx = 0.
Thus, the transformed problem is


2 0
Vx = 0,
Vt +
0 2

V (x, 0) =

1
U (x, 0) =
4

3 1
1 1

(14.16)



u(1)(x, 0)
u(2)(x, 0)

1
=
4

(1)

(2)

3u0 (x) + u0 (x)


(1)
(2)
u0 (x) u0 (x)

(14.17)
Equation (14.16) gives traveling wave solutions of the form:
v (1)(x, t) = F (x + 2t),

v (2)(x, t) = G(x 2t).

(14.18)

Partial Dierential Equations

Igor Yanovsky, 2005

112

We can write U in terms of V :



 


  (1)  
1 1
F (x + 2t)
F (x + 2t) + G(x 2t)
1 1
v
=
=
.
U = V =
1 3
G(x 2t)
F (x + 2t) 3G(x 2t)
1 3
v (2)
(14.19)
For region I, (14.16) and (14.17) give two initial value problems (since value at any
point in region I can be traced back along characteristics to initial conditions):
 (2)
 (1)
(1)
(2)
vt + 2vx = 0,
vt 2vx = 0,
(1)

(2)

v (1)(x, 0) = 34 u0 (x) + 14 u0 (x);

(1)

(2)

v (2)(x, 0) = 14 u0 (x) 14 u0 (x),

which we solve by characteristics to get traveling wave solutions:


3 (1)
1 (2)
1 (1)
1 (2)
v (2)(x, t) = u0 (x 2t) u0 (x 2t).
v (1)(x, t) = u0 (x + 2t) + u0 (x + 2t);
4
4
4
4
Thus, for region I,
!


3 (1)
1 (2)
1 1
u
(x
+
2t)
+
u
(x
+
2t)
4 0
4 0
U = V =
1 (1)
1 (2)
1 3
u
(x

2t)

4 0
4 u0 (x 2t)
!
3 (1)
1 (2)
1 (1)
1 (2)
u
(x
+
2t)
+
u
(x
+
2t)
+
u
(x

2t)

u
(x

2t)
4 0
4 0
4 0
4 0
.
=
3 (1)
1 (2)
3 (1)
3 (2)
u
(x
+
2t)
+
u
(x
+
2t)

u
(x

2t)
+
4 0
4 0
4 0
4 u0 (x 2t)
For region II, solutions of the form F (x + 2t) can be traced back to initial conditions.
Thus, v (1) is the same as in region I. Solutions of the form G(x 2t) can be traced back
to the boundary. The boundary condition gives
u(1)(0, t) + bu(2)(0, t) = (t).
Using (14.19),
v (1)(0, t) + G(2t) + bv (1)(0, t) 3bG(2t) = (t),
(1 + b)v (1)(0, t) + (1 3b)G(2t) = (t),

3
1 (2)
(1)
(1 + b) u0 (2t) + u0 (2t) + (1 3b)G(2t) = (t),
4
4


(1)
(2)
(t) (1 + b) 34 u0 (2t) + 14 u0 (2t)
,
G(2t) =

1 3b

(1)
(2)
( 2t ) (1 + b) 34 u0 (t) + 14 u0 (t)
,
G(t) =
1 3b


3 (1)
1 (2)
)

(1
+
b)
u
((x

2t))
+
u
((x

2t))
( x2t
2
4 0
4 0
.
G(x 2t) =
1 3b
Thus, for region II,



3 (1)
1 (2)
u
(x
+
2t)
+
u
(x
+
2t)
1 1
4 0
4 0
 (1)


(2)
U = V =
( x2t
)(1+b) 34 u0 ((x2t))+ 41 u0 ((x2t))
1 3
2
13b
 (1)


(2)
x2t
( 2 )(1+b) 34 u0 ((x2t))+ 14 u0 ((x2t))
3 (1)
1 (2)
u (x + 2t) + 4 u0 (x + 2t) +
 (1)13b
 .
= 4 0
(2)
3( x2t
)3(1+b) 34 u0 ((x2t))+ 14 u0 ((x2t))
1 (2)
3 (1)
2
4 u0 (x + 2t) + 4 u0 (x + 2t)
13b
The following were performed, but are arithmetically complicated:
Solutions for regions I and II satisfy (14.15).

Partial Dierential Equations

Igor Yanovsky, 2005

113

Solution for region I satises both initial conditions.


Solution for region II satises given boundary condition.
If b = 13 , u(1)(0, t) + 13 u(2) (0, t) = F (2t) + G(2t) + 13 F (2t) G(2t) = 43 F (2t) = (t).
Thus, the solutions of the form v (2) = G(x 2t) are not dened at x = 0, which leads
to ill-posedness.

Partial Dierential Equations

Igor Yanovsky, 2005

114

Problem (F96, #8). Consider the system


ut = 3ux + 2vx
vt = vx v
in the region x 0, t 0. Which of the following sets of initial and boundary data
make this a well-posed problem?
a)

u(x, 0) = 0, x 0
v(x, 0) = x2 , x 0
v(0, t) = t2 , t 0.

b)

u(x, 0) = 0, x 0
v(x, 0) = x2 , x 0
u(0, t) = t,

c)

t 0.

u(x, 0) = 0, x 0
v(x, 0) = x2 , x 0
u(0, t) = t,

t0

v(0, t) = t , t 0.
Proof. Rewrite the equation as Ut + AUx = BU . Initial conditions are same for
(a),(b),(c):




3 2
0 0
Ux =
U,
Ut +
0
1
0 1
 

 (1)
0
u (x, 0)
=
.
U (x, 0) =
x2
u(2)(x, 0)
The eigenvalues ofthe matrix 
A are
1
1
, e2 =
vectors are e1 =
0
2



3 0
1
=
,
=
0 1
0

1 = 3, 2 = 1, and the corresponding eigen. Thus,


1
2


,

1
=
2

2 1
0 1

Let U = V . Then,
Ut + AUx = BU,
Vt + AVx = BV,
Vt + 1 AVx = 1 BV,
Vt + Vx = 1 BV.
Thus, the transformed problem is




3 0
0 1
Vx =
V,
Vt +
0 1
0 1
1

V (x, 0) =

1
U (x, 0) =
2

2 1
0 1



(14.20)
0
x2

x2
=
2

1
1


.

(14.21)

Equation (14.20) gives traveling wave solutions of the form


v (1)(x, t) = F (x + 3t),

v (2)(x, t) = G(x t).

(14.22)

Partial Dierential Equations

Igor Yanovsky, 2005

115

We can write U in terms of V :



 


  (1)  
1 1
F (x + 3t)
F (x + 3t) + G(x t)
1 1
v
=
=
.
U = V =
0 2
G(x t)
2G(x t)
0 2
v (2)
(14.23)
For region I, (14.20) and (14.21) give two initial value problems (since a value at any
point in region I can be traced back along both characteristics to initial conditions):


(1)
(1)
(2)
(2)
vt + vx = v (2),
vt 3vx = v (2),
2
2
v (1)(x, 0) = x2 ;
v (2)(x, 0) = x2 ,
which we do not solve here. Thus, initial conditions for v (1) and v (2) have to be dened.
Since (14.23) denes u(1) and u(2) in terms of v (1) and v (2), we need to dene two initial
conditions for U .
For region II, solutions of the form F (x + 3t) can be traced back to initial conditions.
Thus, v (1) is the same as in region I. Solutions of the form G(x t) are traced back to
the boundary at x = 0. Since from (14.23), u(2)(x, t) = 2v (2)(x, t) = 2G(x t), i.e.
u(2) is written in term of v (2) only, u(2) requires a boundary condition to be dened on
x = 0.

Thus,
a) u(2)(0, t) = t2 , t 0. Well-posed.
b) u(1)(0, t) = t, t 0. Not well-posed.
c) u(1)(0, t) = t, u(2) (0, t) = t2 , t 0. Not well-posed.

Partial Dierential Equations

Igor Yanovsky, 2005

116

Problem (F02, #3). Consider the rst order system


ut + ux + vx = 0
vt + ux vx = 0
on the domain 0 < t < and 0 < x < 1. Which of the following sets of initialboundary data are well posed for this system? Explain your answers.
a) u(x,0) = f(x), v(x,0) = g(x);
b) u(x,0) = f(x), v(x,0) = g(x), u(0,t) = h(x), v(0,t) = k(x);
c) u(x,0) = f(x), v(x,0) = g(x), u(0,t) = h(x), v(1,t) = k(x).
Proof. Rewrite the equation as Ut +AUx = 0. Initial conditions are same for (a),(b),(c):


1 1
Ux = 0,
Ut +
1 1
 (1)
 

u (x, 0)
f (x)
U (x, 0) =
=
.
g(x)
u(2)(x, 0)

=
2,

The eigenvalues of thematrix A are

1
2


 2 and the corresponding
1
1
, e2 =
. Thus,
eigenvectors are e1 =
1 + 2
1 2







1
1
1
1
+
2
0
2
1
1

,
=
,
=
.
=
1 + 2 1 2
0 2
2 2 1 + 2 1
Let U = V . Then,
Ut + AUx = 0,
Vt + AVx = 0,
Vt + 1 AVx = 0,
Vt + Vx = 0.
Thus, the transformed problem is


2
0

Vx = 0,
Vt +
0 2
1

V (x, 0) =

1
U (x, 0) =
2 2

(14.24)





1
1 + 2
(1 + 2)f (x) + g(x)
f (x)
1
=
.
g(x)
1 + 2 1
2 2 (1 + 2)f (x) g(x)
(14.25)

Equation (14.24) gives traveling wave solutions of the form:

v (2)(x, t) = G(x + 2t).


v (1)(x, t) = F (x 2t),

(14.26)

However, we can continue and obtain the solutions. We have two initial value problems
 (2) (2)
 (1) (1)
vt 2vx = 0,
vt + 2vx = 0,
v (1)(x, 0) =

(1+ 2)
1
f (x) +
g(x);
2 2
2 2

v (2)(x, 0) =

(1+ 2)

f (x)
2 2

which we solve by characteristics to get traveling wave solutions:

(1 + 2)
1
(1)

f (x 2t) + g(x 2t),


v (x, t) =
2 2
2 2

(1
+
1
2)

f (x + 2t) g(x + 2t).


v (2)(x, t) =
2 2
2 2

g(x),
2 2

Partial Dierential Equations

Igor Yanovsky, 2005

117

We can obtain general solution U by writing U in terms of V :



U = V =

v (1)
v (2)


=

1
1
1 + 2 1 2

2 2


(1 + 2)f (x 2t) + g(x 2t)
.
(1 + 2)f (x + 2t) g(x + 2t)
(14.27)

In region I, the solution is obtained by solving two initial value problems(since a


value at any point in region I can be traced back along both characteristics to initial
conditions).

G(x + 2t) can be traced back to initial


In region II, the solutions of the form v (2) =
conditions and those of the form v (1) = F (x 2t), to left boundary. Since by (14.27),
u(1) and u(2) are written in terms of both v (1) and v (2), one initial condition and one
boundary condition at x = 0 need to be prescribed.

+
2t) can be traced back to
In region III, the solutions of the form v (2) = G(x

(1)
right boundary and those of the form v = F (x 2t), to initial condition. Since by
(14.27), u(1) and u(2) are written in terms of both v (1) and v (2), one initial condition
and one boundary condition at x = 1 need to be prescribed.
To obtain the solution for region IV, two boundary conditions, one for each boundary, should be given.
Thus,
a) No boundary conditions. Not well-posed.
b) u(1)(0, t) = h(x), u(2)(0, t) = k(x). Not well-posed.
c) u(1)(0, t) = h(x), u(2)(1, t) = k(x). Well-posed.

Partial Dierential Equations

Igor Yanovsky, 2005

118

Problem (S94, #3). Consider the system of equations


ft + gx = 0
gt + fx = 0
ht + 2hx = 0
on the set x 0, t 0, with the following initial-boundary values:
a) f , g, h prescribed on t = 0, x 0;
b) f , g, h prescribed on t = 0, x 0;
c) f + g, h prescribed on t = 0, x 0;

f, h
prescribed on x = 0, t 0.
f g, h prescribed on x = 0, t 0.
f , g, h prescribed on x = 0, t 0.

For each of these 3 sets of data, determine whether or not the system is well-posed.
Justify your conclusions.
Proof. The third equation is decoupled from the rst two and can be considered separately. Its solution can be written in the form
h(x, t) = H(x 2t),
and therefore, h must be prescribed on t = 0 and on x = 0, since the characteristics
propagate from both the x and t axis.
We rewrite the rst two equations as (f u1 , g u2 ):


0 1
Ux = 0,
Ut +
1 0

 (1)
u (x, 0)
.
U (x, 0) =
u(2)(x, 0)
The eigenvalues ofthe matrix
A are 1 = 1, 2 = 1 and the corresponding eigen
1
1
, e2 =
. Thus,
vectors are e1 =
1
1






1 1 1
1 0
1 1
1
.
=
,
=
,
=
1 1
0 1
1 1
2
Let U = V . Then,
Ut + AUx = 0,
Vt + AVx = 0,
Vt + 1 AVx = 0,
Vt + Vx = 0.
Thus, the transformed problem is


1 0
Vx = 0,
Vt +
0 1
1

V (x, 0) =

1
U (x, 0) =
2

(14.28)
1 1
1 1



u(1) (x, 0)
u(1) (x, 0)


.

(14.29)

Equation (14.28) gives traveling wave solutions of the form:


v (1)(x, t) = F (x + t),

v (2)(x, t) = G(x t).

(14.30)

Partial Dierential Equations

Igor Yanovsky, 2005

119

We can write U in terms of V :



 


  (1)  
1 1
F (x + t)
F (x + t) + G(x t)
1 1
v
=
=
.
U = V =
1 1
G(x t)
F (x + t) + G(x t)
1 1
v (2)
(14.31)
For region I, (14.28) and (14.29) give two initial value problems (since a value at any
point in region I can be traced back along both characteristics to initial conditions).
Thus, initial conditions for v (1) and v (2) have to be dened. Since (14.31) denes u(1)
and u(2) in terms of v (1) and v (2), we need to dene two initial conditions for U .
For region II, solutions of the form F (x + t) can be traced back to initial conditions.
Thus, v (1) is the same as in region I. Solutions of the form G(x t) are traced back
to the boundary at x = 0. Since from (14.31), u(2)(x, t) = v (1)(x, t) + v (2)(x, t) =
F (x + t) + G(x t), i.e. u(2) is written in terms of v (2) = G(x t), u(2) requires a
boundary condition to be dened on x = 0.
a) u(1), u(2) prescribed on t = 0;

u(1) prescribed on x = 0.

Since u(1)(x, t) = F (x + t) + G(x t),


u(2)(x, t) = F (x + t) + G(x t), i.e. both
u(1) and u(2) are written in terms of F (x + t)
and G(x t), we need to dene two initial
conditions for U (on t = 0).
A boundary condition also needs to be prescribed
on x = 0 to be able to trace back v (2) = G(x t).
Well-posed.
b) u(1), u(2) prescribed on t = 0;

u(1) u(2) prescribed on x = 0.

As in part (a), we need to dene two initial


conditions for U .
Since u(1) u(2) = 2F (x + t), its denition
on x = 0 leads to ill-posedness. On the
contrary, u(1) + u(2) = 2G(x t) should be
dened on x = 0 in order to be able to trace
back the values through characteristics.
Ill-posed.
c) u(1) + u(2) prescribed on t = 0;

u(1), u(2) prescribed on x = 0.

Since u(1) + u(2) = 2G(x t), another initial


condition should be prescribed to be able to
trace back solutions of the form v (2) = F (x + t),
without which the problem is ill-posed.
Also, two boundary conditions for both u(1)
and u(2) dene solutions of both v (1) = G(x t)
and v (2) = F (x + t) on the boundary. The former
boundary condition leads to ill-posedness.
Ill-posed.

Partial Dierential Equations

Igor Yanovsky, 2005

120

Problem (F92, #8). Consider the system


ut + ux + avx = 0
vt + bux + vx = 0
for 0 < x < 1 with boundary and initial conditions
u=v=0
u = u0 ,

for x = 0
v = v0

for t = 0.

a) For which values of a and b is this a well-posed problem?


b) For this class of a, b, state conditions on u0 and v0 so that the solution u, v will be
continuous and continuously dierentiable.
Proof. a) Let us change the notation (u u(1), v u(2)). Rewrite the equation as


1 a
Ux = 0,
(14.32)
Ut +
b 1

U (x, 0) =

U (0, t) =

u(1)(x, 0)
u(2)(x, 0)
u(1)(0, t)
u(2)(0, t)

(1)

=


u0 (x)
(2)
u0 (x)

!
,

= 0.

The eigenvalues of the matrix A are 1 = 1



0
1 ab
.
=
0
1 + ab

ab, 2 = 1 +

ab.

Let U = V , where is a matrix of eigenvectors. Then,


Ut + AUx = 0,
Vt + AVx = 0,
Vt + 1 AVx = 0,
Vt + Vx = 0.
Thus, the transformed problem is



1 ab
0
Vx = 0,
Vt +
0
1 + ab

(14.33)

V (x, 0) = 1 U (x, 0).


The equation (14.33) gives traveling wave solutions of the form:

v (2)(x, t) = G(x (1 + ab)t).


v (1)(x, t) = F (x (1 ab)t),

(14.34)

We also have U = V , i.e. both u(1) and u(2) (and their initial and boundary conditions)
are combinations of v (1) and v (2).
In order for this problem to be well-posed, both sets of characteristics should emanate
from the boundary at x = 0. Thus, the eigenvalues of the system are real (ab > 0) and
1,2 > 0 (ab < 1). Thus,
0 < ab < 1.
(1)

(2)

b) For U to be C 1 , we require the compatibility condition, u0 (0) = 0, u0 (0) =


0.

Partial Dierential Equations

Igor Yanovsky, 2005

121

Problem (F93, #2). Consider the initial-boundary value problem


ut + ux = 0
vt (1 cx2 )vx + ux = 0
on 1 x 1 and 0 t, with the following prescribed data:
u(x, 0),

v(x, 0),

u(1, t),

v(1, t).

For which values of c is this a well-posed problem?


Proof. Let us change the notation (u u(1), v u(2)).
The rst equation can be solved with u(1)(x, 0) = F (x) to get a solution in the form
u(1)(x, t) = F (x t), which requires u(1) (x, 0) and u(1) (1, t) to be dened.
With u(1) known, we can solve the second equation
(2)

ut (1 cx2 )u(2)
x + F (x t) = 0.
Solving the equation by characteristics, we obtain
the characteristics in the xt-plane are of the form
dx
= cx2 1.
dt
We need to determine c such that the prescribed
data u(2)(x, 0) and u(2)(1, t) makes the problem to
be well-posed. The boundary condition for u(2)(1, t)
requires the characteristics to propagate to the
left with t increasing. Thus, x(t) is a decreasing
function, i.e.
dx
<0

cx2 1 < 0 for 1 < x < 1


dt

c < 1.

We could also do similar analysis we have done in other problems on rst order systems involving nding eigenvalues/eigenvectors of the system and using the fact that
u(1)(x, t) is known at both boundaries (i.e. values of u(1)(1, t) can be traced back either
to initial conditions or to boundary conditions on x = 1).

Partial Dierential Equations

Igor Yanovsky, 2005

122

Problem (S91, #4). Consider the rst order system


ut + aux + bvx = 0
vt + cux + dvx = 0
for 0 < x < 1, with prescribed initial data:
u(x, 0) = u0 (x)
v(x, 0) = v0 (x).
a) Find conditions on a, b, c, d such that there is a full set of characteristics and, in
this case, nd the characteristic speeds.
b) For which values of a, b, c, d can boundary data be prescribed on x = 0 and for which
values can it be prescribed on x = 1? How many pieces of data can be prescribed on
each boundary?
Proof. a) Let us change the notation (u u(1), v u(2)). Rewrite the equation as


a b
Ux = 0,
(14.35)
Ut +
c d
!

 (1)
(1)
u0 (x)
u (x, 0)
.
=
U (x, 0) =
(2)
u(2)(x, 0)
u0 (x)
The system is hyperbolic if for each value of u(1) and u(2) the eigenvalues are real
and the matrix is diagonalizable, i.e. there is a complete set of linearly independent
eigenvectors. The eigenvalues of the matrix A are


a + d (a d)2 + 4bc
a + d (a + d)2 4(ad bc)
=
.
1,2 =
2
2
We need (a d)2 + 4bc > 0. This also makes the problem to be diagonalizable.
Let U = V , where is a matrix of eigenvectors. Then,
Ut + AUx = 0,
Vt + AVx = 0,
Vt + 1 AVx = 0,
Vt + Vx = 0.
Thus, the transformed problem is


1 0
Vx = 0,
Vt +
0 2

(14.36)

Equation (14.36) gives traveling wave solutions of the form:


v (1)(x, t) = F (x 1 t),
The characteristic speeds are

v (2)(x, t) = G(x 2 t).


dx
dt

= 1 ,

dx
dt

(14.37)

= 2 .

b) We assume (a + d)2 4(ad bc) > 0.


a + d > 0, ad bc > 0 1 , 2 > 0 2 B.C. on x = 0.
a + d > 0, ad bc < 0 1 < 0, 2 > 0 1 B.C. on x = 0, 1 B.C. on x = 1.
a + d < 0, ad bc > 0 1 , 2 < 0 2 B.C. on x = 1.

Partial Dierential Equations


a + d < 0,
a + d > 0,
a + d < 0,
a + d = 0,
x = 1.

ad bc < 0 1 < 0, 2 > 0


ad bc = 0 1 = 0, 2 > 0
ad bc = 0 1 = 0, 2 < 0
ad bc < 0

1 < 0, 2 >

Igor Yanovsky, 2005

123

1 B.C. on x = 0, 1 B.C. on x = 1.
1 B.C. on x = 0.
1 B.C. on x = 1.
0

1 B.C. on x = 0, 1 B.C. on

Partial Dierential Equations

Igor Yanovsky, 2005

124

Problem (S94, #2). Consider the dierential operator



  
u
ut + 9vx uxx
L
=
vt ux vxx
v


u(x, t)
on 0 x 2, t 0, in which the vector
consists of two functions that
v(x, t)
are periodic in x.
a) Find the eigenfunctions and eigenvalues of the operator L.
b) Use the results of (a) to solve the initial value problem
 
u
L
=0
for t 0,
v
   ix 
u
e
for t = 0.
=
0
v

Proof. a) We nd the space eigenvalues and eigenfunctions. We rewrite the system


as




0 9
1 0
Ut +
Ux +
Uxx = 0,
1 0
0 1
and nd eigenvalues




1 0
0 9
Uxx = U.
(14.38)
Ux +
0 1
1 0


  &n=
un (t)einx
u(x, t)
&n=
. Plugging this into (14.38), we get
Set U =
=
n=
inx
v(x, t)
n= vn (t)e
 
 &

 &


 &
inx
2
inx
inx
(t)e
u
(t)e
(t)e
1
0
n
u
0 9
inu
n
n
n
&
&
+
= &
,
invn (t)einx
n2 vn (t)einx
vn (t)einx
0 1
1 0


 

1
0 9
inun (t)
+
0
invn (t)
1 0
 


0
9in
un (t)
+
vn (t)
in 0
 2
n
in



n2 un (t)
=
n2 vn (t)



n2 0
un (t)
=
0 n2
vn (t)


un (t)
9in
= 0,
n2
vn (t)

0
1



(n2 )2 9n2 = 0,
which gives 1 =

n2 +3n,

2 =

n2 3n,

are corresponding eigenvectors.

are eigenvalues, and v1 =

3i
1

un (t)
vn (t)
un (t)
vn (t)


,

,




3i
, v2 =
,
1

Partial Dierential Equations




u
v

Igor Yanovsky, 2005




u
v

u
v

9vx uxx
ux vxx

125

. We have

+L
= 0, L
=
b) We want to solve
t
 
 
 
 
u
u
u
u
= L
=
, i.e.
= et . We can write the solution as
v
v
v
v t

 &


un (t)einx
&
=
an e1 t v1 einx + bn e2 t v2 einx
U (x, t) =
vn (t)einx
n=






3i
3i
(n2 +3n)t
inx
(n2 3n)t
=
an e
e + bn e
einx .
1
1
n=
 ix 






e
3i
3i
inx
inx
an
,
e + bn
e =
U (x, 0) =
0
1
1
n=

an = bn = 0, n = 1;
1
1
and a1 = b1
a1 = b1 = .
a1 + b1 =
3i
6i




1
1 4t 3i
3i
e
eix + e2t
eix
U (x, t) =
1
1
6i
6i

 1 4t
(e
+ e2t)
2
eix .
=
1 4t
2t )
(e

e
6i

26 27
26

ChiuYens and Sung-Has solutions give similar answers.


Questions about this problem:
1. Needed to nd eigenfunctions, not eigenvectors.
2. The notation of L was changed. The problem statement incorporates the derivatives wrt. t into L.
3. Why can we write the solution in this form above?
27

Partial Dierential Equations

Igor Yanovsky, 2005

126

Problem (W04, #6). Consider the rst order system


ut ux = vt + vx = 0
in the diamond shaped region 1 < x + t < 1, 1 < x t < 1.
For each of
the following boundary value problems state whether this problem is well-posed. If it is
well-posed, nd the solution.
a) u(x + t) = u0 (x + t) on x t = 1, v(x t) = v0 (x t) on x + t = 1.
b) v(x + t) = v0 (x + t) on x t = 1, u(x t) = u0 (x t) on x + t = 1.
Proof. We have
ut ux = 0,
vt + vx = 0.
u is constant along the characteristics: x + t = c1 (s).
Thus, its solution is u(x, t) = u0 (x + t).
It the initial condition is prescribed at x t = 1,
the solution can be determined in the entire region
by tracing back through the characteristics.
v is constant along the characteristics: x t = c2 (s).
Thus, its solution is v(x, t) = v0 (x t).
It the initial condition is prescribed at x + t = 1,
the solution can be determined in the entire region
by tracing forward through the characteristics.

Partial Dierential Equations

15

Igor Yanovsky, 2005

127

Problems: Gas Dynamics Systems

15.1

Perturbation

Problem (S92, #3).

28 29

Consider the gas dynamic equations

ut + uux + (F ())x = 0,
t + (u)x = 0.
Here F () is a given C -smooth function of . At t = 0, 2-periodic initial data
u(x, 0) = f (x),

(x, 0) = g(x).

a) Assume that
f (x) = U0 + f1 (x),

g(x) = R0 + g1 (x)

where U0 , R0 > 0 are constants and f1 (x), g1 (x) are small perturbations. Linearize the equations and given conditions for F such that the linearized problem is
well-posed.
b) Assume that U0 > 0 and consider the above linearized equations for 0 x 1,
t 0. Construct boundary conditions such that the initial-boundary value problem is
well-posed.
Proof. a) We write the equations in characteristic form:
ut + uux + F  ()x = 0,

t + ux + ux = 0.
Consider the special case of nearly constant initial data
u(x, 0) = u0 + u1 (x, 0),
(x, 0) = 0 + 1 (x, 0).
Then we can approximate nonlinear equations by linear equations. Assuming
u(x, t) = u0 + u1 (x, t),
(x, t) = 0 + 1 (x, t)
remain valid with u1 = O(1), 1 = O(1), we nd that

ut = u1t,

t = 1t ,

ux = u1x ,

x = 1x,

F () = F (0 + 1 (x, t)) = F  (0) + 1 F  (0 ) + O(2 ).


Plugging these into , gives



u1t + (u0 + u1 )u1x + F  (0 ) + 1 F  (0 ) + O(2 ) 1x = 0,

1t + u1x (0 + 1 ) + (u0 + u1 )1x = 0.


Dividing by gives
u1t + u0 u1x + F  (0 )1x = u1 u1x 1 1xF  (0 ) + O(2 ),
1t + u1x 0 + u0 1x = u1x 1 u1 1x.
28
29

See LeVeque, Second Edition, Birkh


auser Verlag, 1992, p. 44.
This problem has similar notation with S92, #4.

Partial Dierential Equations

Igor Yanovsky, 2005

For small , we have




u1t + u0 u1x + F  (0 )1x = 0,


1t + u1x 0 + u0 1x = 0.

This can be written as




 



u1
u0 F  (0 )
0
u1
+
=
.
0
1 t
u0
1 x
0

 u0 F  (0 )

 0
u0



 = (u0 )(u0 ) 0 F  (0) = 0,


2 2u0 + u20 0 F  (0) = 0,



1,2 = u0 0 F  (0 ), u0 > 0, 0 > 0.
For well-posedness, need 1,2 R or F  (0 ) 0.




b) We have
u0 > 0, and 1 = u0 + 0 F (0 ), 2 = u0 0 F (0 ).
If u0 > 0 F  (0 ) 1 > 0, 2 > 0 2 BC at x = 0.
F  (0 ) 1 > 0, 2 = 0 1 BC at x = 0.
If u0 = 0
If 0 < u0 < 0 F  (0 ) 1 > 0, 2 < 0 1 BC at x = 0, 1 BC at x = 1.

15.2

Stationary Solutions

Problem (S92, #4).

30

Consider

ut + uux + x = uxx ,
t + (u)x = 0
for t 0, < x < .
Give conditions for the states U+ , U , R+, R, such that the system has
stationary solutions (i.e. ut = t = 0) satisfying
  

  

u
U+
u
U
=
,
lim
=
.

lim
x+
x
R+
R

Proof. For stationary solutions, we need

u2
x + uxx = 0,
ut =
2 x
t = (u)x = 0.
Integrating the above equations, we obtain
u2
+ ux = C1 ,
2
u = C2 .

30

This problem has similar notation with S92, #3.

128

Partial Dierential Equations


Conditions  give ux = 0 at x = . Thus
U+2
U2
+ R+ =
+ R ,
2
2
U+ R+ = U R .

Igor Yanovsky, 2005

129

Partial Dierential Equations

15.3

Igor Yanovsky, 2005

130

Periodic Solutions

Problem (F94, #4). Let u(x, t) be a solution of the Cauchy problem


ut = uxxxx 2uxx,

< x < +, 0 < t < +,

u(x, 0) = (x),
where u(x, t) and (x) are C functions periodic in x with period 2;
i.e. u(x + 2, t) = u(x, t), x, t.
Prove that
||u(, t)|| Ceat ||||
'
'
2
2
2
2
where ||u(, t)|| =
0 |u(x, t)| dx, |||| =
0 |(x)| dx, C, a are some constants.
Proof. METHOD I:

u(x, t) =

Since u is 2-periodic, let

an (t)einx .

n=

Plugging this into the equation, we get

an (t)einx =

n=

an (t)

n4 an (t)einx + 2

n=
4

n2 an (t)einx ,

n=
2

= (n + 2n )an (t),
4 +2n2 )t

an (t) = an (0)e(n

Also, initial condition gives

u(x, 0) =

an (0)einx = (x),

n=




 
inx 

an (0)e  = |(x)|.

n=

||u(x, t)||22

u (x, t) dx =
0

2
0

a2n (t)

n=

 

an (t)e

 

n=

einx einx dx = 2

inx


imx

an (t)e

dx

m=

a2n (t) = 2

n=

4 +2n2 )t

a2n (0)e2(n

n=









  2(n4 +2n2 )t
  2  2t 
2
2
2






2
an (0)
e
an (0) e
e2(n 1) t
 = 2 
n=

= C2 e2t||||2.

||u(x, t)|| Cet ||||.

n=

n=



||||2

n=



= C1 , (convergent)

Partial Dierential Equations


METHOD II:

Igor Yanovsky, 2005

131

Multiply this equation by u and integrate

uut = uuxxxx 2uuxx ,


1 d 2
(u ) = uuxxxx 2uuxx ,
2 dt
 2
 2
 2
1 d
2
u dx =
uuxxxx dx
2uuxx dx,
2 dt 0
0
0
 2
2
2  2
1 d


2
2
||u||2 = uuxxx  + ux uxx 
uxx dx
2uuxx dx,
2 dt
0
0
  0    0 
1 d
||u||22 =
2 dt

=0
2

(2ab a2 + b2 )
 2
 2
 2

u2xx dx +
(u2 + u2xx ) dx =
u2 dx = ||u||2,
0

u2xx dx

=0
2

2uuxx dx

d
||u||2 2||u||2,
dt
||u||2 ||u(0)||2e2t,
||u|| ||u(0)||et. 

METHOD III: Can use Fourier transform. See ChiuYens solutions, that have both
Method II and III.

Partial Dierential Equations

Igor Yanovsky, 2005

132

Problem (S90, #4).


Let f (x) C be a 2-periodic function, i.e., f (x) = f (x + 2) and denote by
 2
2
|f (x)|2 dx
||f || =
0

the L2 -norm of f .
a) Express ||dpf /dxp||2 in terms of the Fourier coecients of f .
b) Let q > p > 0 be integers. Prove that  > 0, K = N (, p, q), constant, such that
 dq f 2
 dpf 2




 p   q  + K||f ||2.
dx
dx
c) Discuss how K depends on .
Proof. a) Let

31

f (x) =

fn einx ,

d f
dxp

fn (in)peinx ,

 2 
 2

 dpf 2



2


p inx 2

fn (in) e
dx =
|i2 |p 
fn np einx  dx
 p  =
dx
0
0

 2 



2

fn np einx  dx = 2
fn2 n2p .
=
0

n=0

b) We have

 dq f 2
 dpf 2




 p   q  + K||f ||2,
dx
dx




fn2 n2p 2
fn2 n2q + K 2
fn2 ,
2
n=0
2p

n=0
2q

K,

(1 n )
  

K,

q

2p

n=0

some q  > 0.

< 0, f or n large

Thus, the above inequality is true for n large enough. The statement follows.
31

Note:
 L


e

inx imx

dx

0
L

n = m
n=m

Partial Dierential Equations


Problem (S90, #5).

32

Igor Yanovsky, 2005

133

Consider the ame front equation




ut + uux + uxx + uxxxx = 0


with 2-periodic initial data
u(x, 0) = f (x),

f (x) = f (x + 2) C .

a) Determine the solution, if f (x) f0 = const.


b) Assume that
f (x) = 1 + g(x),

0 <  1,

|g| = 1,

g(x) = g(x + 2).

Linearize the equation. Is the Cauchy problem well-posed for the linearized equation,
i.e., do its solutions v satisfy an estimate
||v(, t)|| Ke(tt0 ) ||v(, t0)||?
c) Determine the best possible constants K, .
Proof. a) The solution to
ut + uux + uxx + uxxxx = 0,
u(x, 0) = f0 = const,
is u(x, t) = f0 = const.
b) We consider the special case of nearly constant initial data
u(x, 0) = 1 + u1 (x, 0).
Then we can approximate the nonlinear equation by a linear equation. Assuming
u(x, t) = 1 + u1 (x, t),
remain valid with u1 = O(1), from , we nd that
u1t + (1 + u1 )u1x + u1xx + u1xxxx = 0.
Dividing by gives
u1t + u1x + u1 u1x + u1xx + u1xxxx = 0.
For small , we have
u1t + u1x + u1xx + u1xxxx = 0.
Multiply this equation by u1 and integrate
u1 u1t + u1 u1x + u1 u1xx + u1 u1xxxx = 0,

d u21
u21
+
+ u1 u1xx + u1 u1xxxx = 0,
dt 2
2 x
 2

 2
2
u21 2
1 d
2
u1 dx +  +
u1 u1xx dx +
u1 u1xxxx dx = 0,
2 dt 0
0
0
20 
=0

2  2
2
2  2
1 d



2
2
||u1 ||2 + u1 u1x 
u1x dx + u1 u1xxx  u1x u1xx +
u21xx dx = 0,
2 dt
0
0
0
0
0
  
     
=0

=0

1 d
||u1 ||22 =
2 dt
32

S90 #5, #6, #7 all have similar formulations.

=0
2

u21x dx

u21xx dx.

Partial Dierential Equations

Igor Yanovsky, 2005

134

Since u1 is 2-periodic, let

u1 =

n=

u1x = i
u1xx =

an (t)einx .

Then,
u21x

inx

nan (t)e

n=

n=

n2 an (t)einx

u21xx =

n=

Thus,

n2 an (t)einx

2
.

n=

1 d
||u1 ||22 =
2 dt


2
=
nan (t)einx ,

u21x dx

u21xx dx

inx

2
2

dx
n2 an (t)einx dx
0
0



2
2
4
n an (t)2 = 2
an (t)2 (n2 + n4 ) 0.
= 2
n an (t) 2
=

nan (t)e

||u1 (, t)||2 ||u1(, 0)||2,

where K = 1, = 0.

Problem (W03, #4). Consider the PDE


ut = ux + u4 for t > 0
u = u0 for t = 0
for 0 < x < 2. Dene the set A = {u = u(x) : u
(k) = 0 if k < 0}, in which

{
u(k, t)} is the Fourier series of u in x on [0, 2].
a) If u0 A, show that u(t) A.
b) Find dierential equations for u
(0, t), u
(1, t), and u
(2, t).
Proof. a) Solving
ut = ux + u4
u(x, 0) = u0 (x)
by the method of characteristics, we get
u(x, t) =

u0 (x + t)
1

(1 3t(u0 (x + t))3 ) 3

0k = 0 if k < 0. Thus,


Since u0 A, u
u0 (x) =

u
0k ei

kx
2

k=0

Since
u
k =

1
2


0

u(x, t) ei

kx
2

dx,

Partial Dierential Equations


we have
u(x, t) =


k=0

that is, u(t) A.

u
k ei

kx
2

Igor Yanovsky, 2005

135

Partial Dierential Equations

15.4

Igor Yanovsky, 2005

136

Energy Estimates

Problem (S90, #6). Let U (x, t) C be 2-periodic in x. Consider the linear


equation
ut + U ux + uxx + uxxxx = 0,
u(x, 0) = f (x),

f (x) = f (x + 2) C .

a) Derive an energy estimate for u.


b) Prove that one can estimate all derivatives || pu/xp||.
c) Indicate how to prove existence of solutions. 33
Proof. a) Multiply the equation by u and integrate
uut + U uux + uuxx + uuxxxx = 0,
1
1 d 2
(u ) + U (u2 )x + uuxx + uuxxxx = 0,
2 dt
2


 2
 2
1 2
1 d 2 2
2
u dx +
U (u )x dx +
uuxx dx +
uuxxxx dx = 0,
2 dt 0
2 0
0
0

2  2
1 2 2 1 2
1 d

2
2
||u|| + U u 
Ux u dx + uux 
u2x dx
2 dt
2
2
0
0
0
0
  
=0

2
2  2


u2xx dx = 0,
+uuxxx  ux uxx  +
0
0
0
 2
 2
 2
1
1 d
2
2
2
||u||
Ux u dx
ux dx +
u2xx dx = 0,
2 dt
2 0
0
0
 2
 2
 2
1
1 d
||u||2 =
Uxu2 dx +
u2x dx
u2xx dx (from S90, #5)
2 dt
2 0
0
0

 2
1
1 2
2
max Ux
Ux u dx
u2 dx.

2 0
2 x
0
d
||u||2 max Ux ||u||2,

x
dt
||u(x, t)||2 ||u(x, 0)||2 e(maxx Ux )t .
This can also been done using Fourier Transform. See ChiuYens solutions where the
above method and the Fourier Transform methods are used.
33

S90 #5, #6, #7 all have similar formulations.

Partial Dierential Equations


34

Problem (S90, #7).

Igor Yanovsky, 2005

137

Consider the nonlinear equation




ut + uux + uxx + uxxxx = 0,

f (x) = f (x + 2) C .

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

a) Derive an energy estimate for u.


b) Show that there is an interval 0 t T , T depending on f ,
such that also ||u(, t)/x|| can be bounded.
Proof. a) Multiply the above equation by u and integrate
uut + u2 ux + uuxx + uuxxxx = 0,
1
1 d 2
(u ) + (u3 )x + uuxx + uuxxxx = 0,
2 dt
3


 2
 2
1 2 3
1 d 2 2
u dx +
(u )x dx +
uuxx dx +
uuxxxx dx = 0,
2 dt 0
3 0
0
0
 2
 2
1 3 2
1 d
2
2
ux dx +
u2xx dx = 0,
||u|| + u 
2 dt
3
0
0
0
  


=0
2

 2
1 d
2
2
||u|| =
ux dx
u2xx dx 0,
2 dt
0
0
||u(, t)|| ||u(, 0)||.

(from S90, #5)

b) In order to nd a bound for ||ux (, t)||, dierentiate  with respect to x:


utx + (uux )x + uxxx + uxxxxx = 0,
Multiply the above equation by ux and integrate:
ux utx + ux (uux)x + ux uxxx + ux uxxxxx = 0,

 2
 2
 2
1 d 2
2
(ux ) dx +
ux (uux )x dx +
uxuxxx dx +
ux uxxxxx dx = 0.
2 dt 0
0
0
0
We evaluate one of the integrals in the above expression using the periodicity:
 2
 2
 2
 2
 2
2
3
ux (uux)x dx =
uxx uux =
ux (ux + uuxx) =
ux +
uux uxx,
0
0
0
0
0

 2
1 2 3
uxx uux =
ux ,

2 0
0

 2
1 2 3
ux (uux )x =
ux .

2 0
0
We have

34

1 d
||ux ||2 +
2 dt


0

u3x


dx +
0

ux uxxx dx +

S90 #5, #6, #7 all have similar formulations.

ux uxxxxx dx = 0.

Partial Dierential Equations


Let w = ux , then

Igor Yanovsky, 2005

 2
 2
 2
1 d
||w||2 =
w 3 dx
wwxx dx
wwxxxx dx
2 dt
0
0
0
 2
 2
 2
 2
3
2
2
w dx +
wx dx
wxx dx
w 3 dx,
=
0
0
0
0
 2
d

||ux||2 =
u3x dx.
dt
0

138

Partial Dierential Equations

16

Igor Yanovsky, 2005

139

Problems: Wave Equation

16.1

The Initial Value Problem

Example (McOwen 3.1 #1). Solve the initial value problem:

utt c uxx = 0,
u(x, 0) = 
x3 , ut (x, 0) = 
sin x .

g(x)

h(x)

Proof. DAlemberts formula gives the solution:



1 x+ct
1
(g(x + ct) + g(x ct)) +
h() d
u(x, t) =
2
2c xct

1
1 x+ct
1
3
3
(x + ct) + (x ct) +
sin d
=
2
2
2c xct
1
1
= x3 + 2xc2 t2
cos(x + ct) +
cos(x ct) =
2c
2c
1
= x3 + 2xc2 t2 + sin x sin ct.
c
Problem (S99, #6). Solve the Cauchy problem

utt = a2 uxx + cos x,
u(x, 0) = sin x, ut (x, 0) = 1 + x.

(16.1)

Proof. We have a nonhomogeneous PDE with nonhomogeneous initial conditions:

utt c2 uxx = cos

 x,

f (x,t)

u(x, 0) = 
sin x,

ut (x, 0) = 1 
+ x .

g(x)

h(x)

The solution is given by dAlemberts formula and Duhamels principle.35


 x+ct
1
1
A
(g(x + ct) + g(x ct)) +
u (x, t) =
h() d
2
2c xct
 x+ct
1
1
(sin(x + ct) + sin(x ct)) +
(1 + ) d
=
2
2c xct
2 =x+ct
1
+
= sin x cos ct + xt + t.
= sin x cos ct +
2c
2 =xct
D

u (x, t) =
=

1
2c
1
2c

 t
0

 t

x+c(ts)

xc(ts)

1
f (, s) d ds =
2c

 t 

x+c(ts)

cos d ds
0

xc(ts)


1
sin[x + c(t s)] sin[x c(t s)] ds = 2 (cos x cos x cos ct).
c

u(x, t) = uA (x, t) + uD (x, t) = sin x cos ct + xt + t +


35

Note the relationship: x , t s.

1
(cos x cos x cos ct).
c2

Partial Dierential Equations

Igor Yanovsky, 2005

140

We can check that the solution satises equation (16.1). Can also check that uA , uD
satisfy


2 A
2 D
uD
uA
tt c uxx = 0,
tt c uxx = cos x,
uA (x, 0) = sin x, uA
uD (x, 0) = 0, uD
t (x, 0) = 1 + x;
t (x, 0) = 0.

Partial Dierential Equations

16.2

Igor Yanovsky, 2005

141

Initial/Boundary Value Problem

Problem 1. Consider the initial/boundary value problem

0 < x < L, t > 0


utt c uxx = 0
0<x<L
u(x, 0) = g(x), ut (x, 0) = h(x)

u(0, t) = 0,
u(L, t) = 0
t 0.

(16.2)

Proof. Find u(x, t) in the form

nx
nx
a0 (t) 
+
+ bn (t) sin
.
an (t) cos
2
L
L

u(x, t) =

n=1

Functions an (t) and bn (t) are determined by the boundary conditions:

a0 (t) 
+
an (t)
2

0 = u(0, t) =

an (t) = 0.

Thus,

n=1

u(x, t) =

bn (t) sin

n=1

nx
.
L

(16.3)

If we substitute (16.3) into the equation utt c2 uxx = 0, we get




nx
nx
n 2

2
+c
= 0,
or
bn (t) sin
bn (t) sin
L
L
L
n=1
n=1

nc 2
bn (t) = 0,
bn (t) +
L
whose general solution is
bn (t) = cn sin

nct
nct
+ dn cos
.
L
L

(16.4)

nct
nc
nct
Also, bn (t) = cn ( nc
L ) cos L dn ( L ) sin L .
The constants cn and dn are determined by the initial conditions:

g(x) = u(x, 0) =
h(x) = ut (x, 0) =


n=1

nx 
nx
=
,
bn (0) sin
dn sin
L
L
bn (0) sin

n=1

n=1

nx
nx
nc
=
sin
.
cn
L
L
L
n=1

By orthogonality, we may multiply by sin(mx/L) and integrate:


 L
 L

mx
mx
nx
L
dx =
sin
dx = dm ,
g(x) sin
dn sin
L
L
L
2
0
0 n=1
 L
 L

nx
mx
mx
nc
mc L
dx =
sin
sin
dx = cm
.
h(x) sin
cn
L
L
L
L
L 2
0
0
n=1

Thus,
2
dn =
L


0

nx
dx,
g(x) sin
L

2
cn =
nc

h(x) sin
0

The formulas (16.3), (16.4), and (16.5) dene the solution.

nx
dx.
L

(16.5)

Partial Dierential Equations

Igor Yanovsky, 2005

Example (McOwen 3.1 #2). Consider the initial/boundary value problem

0 < x < , t > 0

utt uxx = 0
0<x<
u(x, 0) = 1,
ut (x, 0) = 0

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

142

(16.6)

Proof. Find u(x, t) in the form

a0 (t) 
+
an (t) cos nx + bn (t) sin nx.
u(x, t) =
2
n=1

Functions an (t) and bn (t) are determined by the boundary conditions:

0 = u(0, t) =

a0 (t) 
+
an (t)
2

an (t) = 0.

Thus,

n=1

u(x, t) =

bn (t) sin nx.

(16.7)

n=1

If we substitute this into utt uxx = 0, we get

bn (t) sin nx +

n=1
bn (t)

bn (t)n2 sin nx = 0,

or

n=1

+ n2 bn (t) = 0,

whose general solution is


bn (t) = cn sin nt + dn cos nt.

(16.8)

Also, bn (t) = ncn cos nt ndn sin nt.


The constants cn and dn are determined by the initial conditions:
1 = u(x, 0) =
0 = ut (x, 0) =


n=1

bn (0) sin nx =
bn (0) sin nx =

n=1


n=1

dn sin nx,
ncn sin nx.

n=1

By orthogonality, we may multiply both equations by sin mx and integrate:




sin mx dx = dm ,
2
0

0 dx = ncn .
2
0
Thus,
2
(1 cos n) =
dn =
n

4
n ,

0,

n odd,
n even,

Using this in (16.8) and (16.7), we get



4
n odd,
n cos nt,
bn (t) =
0,
n even,

and

cn = 0.

(16.9)

Partial Dierential Equations

4  cos(2n + 1)t sin(2n + 1)x


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

Igor Yanovsky, 2005

143

Partial Dierential Equations

Igor Yanovsky, 2005

144

We can sum the series in regions bouded by characteristics. We have


u(x, t) =

4  cos(2n + 1)t sin(2n + 1)x


,

(2n + 1)

or

n=0

2  sin[(2n + 1)(x + t)] 2  sin[(2n + 1)(x t)]


u(x, t) =
+
.

(2n + 1)

(2n + 1)
n=0

(16.10)

n=0

The initial condition may be written as


1 = u(x, 0) =

4  sin(2n + 1)x

(2n + 1)

for 0 < x < .

(16.11)

n=0

We can use (16.11) to sum the series in (16.10).


1 1
+ = 1.
2 2
Since sin[(2n + 1)(x t)] = sin[(2n + 1)((x t))], and 0 < (x t) < in R2 ,
In R1 ,

u(x, t) =

1 1
= 0.
2 2
Since sin[(2n + 1)(x + t)] = sin[(2n + 1)(x + t 2)] = sin[(2n + 1)(2 (x + t))],
and 0 < 2 (x + t) < in R3 ,
in R2 ,

u(x, t) =

1 1
u(x, t) = + = 0.
2 2
Since 0 < (x t) < and 0 < 2 (x + t) < in R4 ,
in R3 ,

in R4 ,

1 1
u(x, t) = = 1.
2 2

Partial Dierential Equations

Igor Yanovsky, 2005

145

Problem 2. Consider the initial/boundary value problem

2
0 < x < L, t > 0

utt c uxx = 0
0<x<L
u(x, 0) = g(x), ut (x, 0) = h(x)

ux (L, t) = 0
t 0.
ux (0, t) = 0,

(16.12)

Proof. Find u(x, t) in the form

a0 (t) 
nx
nx
u(x, t) =
+
+ bn (t) sin
.
an (t) cos
2
L
L
n=1

Functions an (t) and bn (t) are determined by the boundary conditions:

n

nx
nx
ux (x, t) =
sin
+ bn (t)
cos
,
an (t)
L
L
L
L
n=1

0 = ux (0, t) =

bn (t)

n=1

bn (t) = 0.

Thus,

nx
a0 (t) 
an (t) cos
+
.
u(x, t) =
2
L

(16.13)

n=1

If we substitute (16.13) into the equation utt c2 uxx = 0, we get

n 2

nx
nx
a0 (t)  
2
+
+c
= 0,
an (t) cos
an (t)
cos
2
L
L
L
n=1
n=1

nc 2
and
an (t) +
an (t) = 0,
a0 (t) = 0
L
whose general solutions are
nct
nct
+ dn cos
.
and
an (t) = cn sin
a0 (t) = c0 t + d0
L
L

(16.14)

nct
nc
nct
Also, a0 (t) = c0 and an (t) = cn ( nc
L ) cos L dn ( L ) sin L .
The constants cn and dn are determined by the initial conditions:

g(x) = u(x, 0) =
h(x) = ut (x, 0) =

n=1

d0 
nx
nx
a0 (0) 
+
=
+
,
an (0) cos
dn cos
2
L
2
L
a0 (0)
2

n=1

an (0) cos

n=1

c0
nx
nx
nc
=
+
cos
.
cn
L
2
L
L
n=1

By orthogonality, we may multiply both equations by cos(mx/L), including m = 0,


and integrate:
 L
 L
mx
L
L
dx = dm ,
g(x) dx = d0 ,
g(x) cos
2
L
2
0
0
 L
 L
mx
L
mc L
dx = cm
.
h(x) dx = c0 ,
h(x) cos
2
L
L 2
0
0
Thus,
2
dn =
L


0

nx
dx,
g(x) cos
L

2
cn =
nc


0

nx
dx,
h(x) cos
L

2
c0 =
L

h(x) dx.
0

(16.15)
The formulas (16.13), (16.14), and (16.15) dene the solution.

Partial Dierential Equations

Igor Yanovsky, 2005

146

Example (McOwen 3.1 #3). Consider the initial/boundary value problem

0 < x < , t > 0


utt uxx = 0
(16.16)
u(x, 0) = x,
ut (x, 0) = 0
0<x<

t 0.
ux (0, t) = 0, ux (, t) = 0
Proof. Find u(x, t) in the form

a0 (t) 
+
an (t) cos nx + bn (t) sin nx.
u(x, t) =
2
n=1

Functions an (t) and bn (t) are determined by the boundary conditions:


ux (x, t) =

an (t)n sin nx + bn (t)n cos nx,

n=1

0 = ux (0, t) =

bn (t)n

bn (t) = 0.

Thus,

n=1

a0 (t) 
u(x, t) =
+
an (t) cos nx.
2

(16.17)

n=1

If we substitute (16.17) into the equation utt uxx = 0, we get

n=1

n=1
an (t)


a0 (t)  
+
an (t) cos nx +
an (t)n2 cos nx = 0,
2
a0 (t)

=0

and

+ n2 an (t) = 0,

whose general solutions are


a0 (t) = c0 t + d0

and

an (t) = cn sin nt + dn cos nt.

(16.18)

Also, a0 (t) = c0 and an (t) = cn n cos nt dn n sin nt.


The constants cn and dn are determined by the initial conditions:
x = u(x, 0) =
0 = ut (x, 0) =

n=1

n=1

d0 
a0 (0) 
+
+
an (0) cos nx =
dn cos nx,
2
2
a0 (0)
2

an (0) cos nx =

n=1

c0
+
2

cn n cos nx.

n=1

By orthogonality, we may multiply both equations by cos mx, including m = 0, and


integrate:



x dx = d0 ,
x cos mx dx = dm ,
2
2
 0
0

0 dx = c0 ,
0 cos mx dx = cm m .
2
2
0
0
Thus,
d0 = ,

dn =

2
(cos n 1),
n2

cn = 0.

Using this in (16.18) and (16.17), we get


a0 (t) = d0 = ,

an (t) =

2
(cos n 1) cos nt,
n2

(16.19)

Partial Dierential Equations

Igor Yanovsky, 2005

2  (cos n 1) cos nt cos nx

u(x, t) = +
.
2
n=1
n2

147

Partial Dierential Equations

Igor Yanovsky, 2005

148

We can sum the series in regions bouded by characteristics. We have

2  (cos n 1) cos nt cos nx

,
or
u(x, t) = +
2

n2
n=1

1
u(x, t) = +
2


n=1

(cos n 1) cos[n(x t)] 1  (cos n 1) cos[n(x + t)]


+
. (16.20)
n2

n2
n=1

The initial condition may be written as

2  (cos n 1) cos nx

u(x, 0) = x = +
2

n2

for 0 < x < ,

n=1

which implies

1  (cos n 1) cos nx
x
=
2
4

n2

for 0 < x < ,

(16.21)

n=1

We can use (16.21) to sum the series in (16.20).


xt x+t
+
+
= x.
2
2
4
2
4
Since cos[n(x t)] = cos[n((x t))], and 0 < (x t) < in R2 ,
In R1 ,

u(x, t) =

(x t) x + t
+
+
= t.
2
2
4
2
4
Since cos[n(x+t)] = cos[n(x+t2)] = cos[n(2 (x+t))], and 0 < 2 (x+t) <
in R3 ,
in R2 ,

u(x, t) =

x t 2 (x + t)
+
+
= t.
2
2
4
2
4
Since 0 < (x t) < and 0 < 2 (x + t) < in R4
in R3 ,

u(x, t) =

in R4 ,

u(x, t) =

(x t) 2 (x + t)
+
+
= x.
2
2
4
2
4

Partial Dierential Equations

Igor Yanovsky, 2005

149

Example (McOwen 3.1 #4). Consider the initial boundary value problem

2
for
x > 0, t > 0

utt c uxx = 0
(16.22)
for x > 0
u(x, 0) = g(x), ut (x, 0) = h(x)

u(0, t) = 0
for t 0,
where g(0) = 0 = h(0). If we extend g and h as odd functions on < x < , show
that dAlemberts formula gives the solution.
Proof. Extend g and h as odd functions on < x < :


g(x),
x0
h(x),

g(x) =
h(x)
=
g(x), x < 0
h(x),
Then, we need to solve

xx = 0
for
u
tt c2 u

u
(x, 0) = g(x), u
t(x, 0) = h(x)

x0
x < 0.

< x < , t > 0


for < x < .

(16.23)

To show that dAlemberts formula gives the solution to (16.23), we need to show that
the solution given by dAlemberts formula satises the boundary condition u(0, t) = 0.

1 x+ct
1
(
g(x + ct) + g(x ct)) +
h() d,
u
(x, t) =
2
2c xct

1 ct
1
(
g(ct) + g(ct)) +
h() d
u
(0, t) =
2
2c ct
1
1
(
g(ct) g(ct)) + (H(ct) H(ct))
=
2
2c
1
= 0 + (H(ct) H(ct)) = 0,
2c
x

h()
d; and since h is odd, then H is even.
where we used H(x) =
0

Example (McOwen 3.1 #5). Find in closed form (similar to dAlembets formula)
the solution u(x, t) of

for x, t > 0
utt c uxx = 0
(16.24)
u(x, 0) = g(x), ut (x, 0) = h(x) for x > 0

u(0, t) = (t)
for t 0,
where g, h, C 2 satisfy (0) = g(0),  (0) = h(0), and  (0) = c2 g (0). Verify that
u C 2 , even on the characteristic x = ct.
Proof. As in (McOwen 3.1 #4), we can extend g and h to be odd functions. We want
to transform the problem to have zero boundary conditions.
Consider the function:
U (x, t) = u(x, t) (t).

(16.25)

Partial Dierential Equations


Then (16.24) transforms to:

Utt c2 Uxx =  (t)

  

fU (x,t)

U (x, 0) = g(x) (0),






gU (x)

0 .

U (0, t) = 

Igor Yanovsky, 2005

Ut (x, 0) = h(x)  (0)





hU (x)

u (t)

We use dAlemberts formula and Duhamels principle on U .


After getting U , we can get u from u(x, t) = U (x, t) + (t).

150

Partial Dierential Equations

Igor Yanovsky, 2005

151

Example (Zachmanoglou, Chapter 8, Example 7.2). Find the solution of

2
for x > 0, t > 0

utt c uxx = 0
(16.26)
for x > 0
u(x, 0) = g(x), ut (x, 0) = h(x)

for t > 0.
ux (0, t) = 0
Proof. Extend g and h as even functions on < x < :


g(x),
x0
h(x),
x0

g(x) =
h(x)
=
g(x), x < 0
h(x), x < 0.
Then, we need to solve

xx = 0
for
u
tt c2 u

u
(x, 0) = g(x), u
t(x, 0) = h(x)

< x < , t > 0


for < x < .

(16.27)

To show that dAlemberts formula gives the solution to (16.27), we need to show that
the solution given by dAlemberts formula satises the boundary condition ux (0, t) = 0.

1 x+ct
1
(
g(x + ct) + g(x ct)) +
h() d.
u
(x, t) =
2
2c xct
1 
1
ct)],
(
g (x + ct) + g (x ct)) + [h(x
+ ct) h(x
u
x (x, t) =
2
2c
1 
1

(
g (ct) + g (ct)) + [h(ct) h(ct)]
= 0.
u
x (0, t) =
2
2c
Since g is even, then g  is odd.
Problem (F89, #3). 36 Let = c, constant. Find the solution of

for x > 0, t > 0


utt c uxx = 0
u(x, 0) = g(x), ut (x, 0) = h(x) for x > 0

for t > 0,
ut (0, t) = ux (0, t)

(16.28)

where g, h C 2 for x > 0 and vanish near x = 0.


Hint: Use the fact that a general solution of (16.28) can be written as the sum of two
traveling wave solutions.
Proof. DAlemberts formula is derived by plugging in the following into the above
equation and initial conditions:
u(x, t) = F (x + ct) + G(x ct).
As in (Zachmanoglou 7.2), we can extend g and h to be even functions.
36

Similar to McOwen 3.1 #5. The notation in this problem is changed to be consistent with McOwen.

Partial Dierential Equations

Igor Yanovsky, 2005

Example (McOwen 3.1 #6). Solve the initial/boundary value problem

for 0 < x < and t > 0


utt uxx = 1
u(x, 0) = 0, ut (x, 0) = 0
for 0 < x <

2
u(0, t) = 0,
u(, t) = /2 for t 0.

152

(16.29)

Proof. If we rst nd a particular solution of the nonhomogeneous equation, this reduces the problem to a boundary value problem for the homogeneous equation ( as in
(McOwen 3.1 #2) and (McOwen 3.1 #3) ).
Hint: You should use a particular solution depending on x!
First, nd a particular solution. This is similar to the method of separation of
variables. Assume
up (x, t) = X(x),
which gives
X (x) = 1,
X (x) = 1.
The solution to the above ODE is
x2
X(x) = + ax + b.
2
The boundary conditions give
up(0, t) = b = 0,
2
2
up (, t) = + a + b = ,
2
2
Thus, the particular solution is
up (x, t) =

a = b = 0.

x2
.
2

This solution satises the following:

up tt up xx = 1
2
up (x, 0) = x2 , up t (x, 0) = 0

up(, t) = 2 .
up (0, t) = 0,
Second, we nd a solution to a boundary value problem for the homogeneous equation:

utt uxx = 0
2
ut (x, 0) = 0
u(x, 0) = x2 ,

u(0, t) = 0,
u(, t) = 0.
This is solved by the method of Separation of Variables. See Separation of Variables
subsection of Problems: Separation of Variables: Wave Equation McOwen 3.1 #2.
The only dierence there is that u(x, 0) = 1.
We would nd uh (x, t). Then,
u(x, t) = uh (x, t) + up (x, t).

Partial Dierential Equations

Igor Yanovsky, 2005

153

Problem (S02, #2). a) Given a continuous function f on R which vanishes for


|x| > R, solve the initial value problem

utt uxx = f (x) cos t,
< x < , 0 t <
u(x, 0) = 0, ut (x, 0) = 0,
by rst nding a particular solution by separation of variables and then adding the
appropriate solution of the homogeneous PDE.
b) Since the particular solution is not unique, it will not be obvious that the solution
to the initial value problem that you have found in part (a) is unique. Prove that it is
unique.
Proof. a) First, nd a particular solution by separation of variables. Assume
up (x, t) = X(x) cos t,
which gives
X(x) cos t X (x) cos t = f (x) cos t,
X  + X = f (x).
The solution to the above ODE is written as X = Xh + Xp . The homogeneous solution
is
Xh (x) = a cos x + b sin x.
To nd a particular solution, note that since f is continuous, G C 2 (R), such that
G + G = f (x).
Thus,
Xp (x) = G(x).
X(x) = Xh (x) + Xp (x) = a cos x + b sin x + G(x).


up (x, t) = a cos x + b sin x + G(x) cos t.
It can be veried that this solution satises the following:

up tt up xx = f (x) cos t,
up t(x, 0) = 0.
up (x, 0) = a cos x + b sin x + G(x),
Second, we nd a solution of the homogeneous PDE:

utt uxx = 0,
0 .
u(x, 0) = a cos x b sin x G(x), ut (x, 0) = 





g(x)

h(x)

The solution is given by dAlemberts formula (with c = 1):



1 x+t
1
A
h() d
uh (x, t) = u (x, t) = (g(x + t) + g(x t)) +
2
2 xt
 

1

a cos(x + t) b sin(x + t) G(x + t) + a cos(x t) b sin(x t) G(x t)
=
2
 1

1
= a cos(x + t) + b sin(x + t) + G(x + t) a cos(x t) + b sin(x t) + G(x t) .
2
2

Partial Dierential Equations

Igor Yanovsky, 2005

154

It can be veried that the solution satises the above homogeneous PDE with the
boundary conditions. Thus, the complete solution is:
u(x, t) = uh (x, t) + up (x, t).

Alternatively, we could use Duhamels principle to nd the solution:


1
u(x, t) =
2

 t 

37

x+(ts)

f () cos s d ds.
0

x(ts)

However, this is not how it was suggested to do this problem.


b) The particular solution is not unique, since any constants a, b give the solution.
However, we show that the solution to the initial value problem is unique.
Suppose u1 and u2 are two solutions. Then w = u1 u2 satises:

wtt wxx = 0,
w(x, 0) = 0, wt (x, 0) = 0.
DAlemberts formula gives
1
1
w(x, t) = (g(x + t) + g(x t)) +
2
2

x+t

h() d = 0.
xt

Thus, the solution to the initial value problem is unique.


37

Note the relationship: x , t s.

Partial Dierential Equations

16.3

Igor Yanovsky, 2005

155

Similarity Solutions

Problem (F98, #7). Look for a similarity solution of the form


v(x, t) = t w(y = x/t ) for the dierential equation
vt = vxx + (v 2 )x.

(16.30)

a) Find the parameters and .


b) Find a dierential equation for w(y) and show that this ODE can be reduced to rst
order.
c) Find a solution for the resulting rst order ODE.
Proof. We can rewrite (16.30) as
vt = vxx + 2vvx.

(16.31)

We look for a similarity solution of the form

x
v(x, t) = t w(y),
y= .
t
1

vt = t

w + t w yt = t

w+t

vx = t w  yx = t w  t = t w  ,

x 
+1 w = t1 w t1 yw  ,
t

vxx = (t w  )x = t w  yx = t w  t = t2 w  .
Plugging in the derivatives we calculated into (16.31), we obtain
t1 w t1 yw  = t2 w  + 2(tw)(t w  ),
w yw  = t12 w  + 2t+1 ww  .
The parameters that would eliminate t from equation above are
1
1
= , = .
2
2
With these parameters, we obtain the dierential equation for w(y):
1
1
w yw  = w  + 2ww  ,
2
2
1
1
w  + 2ww  + yw  + w = 0.
2
2
We can write the ODE as
1
w  + 2ww  + (yw) = 0.
2
Integrating it with respect to y, we obtain the rst order ODE:
1
w  + w 2 + yw = c.
2

Partial Dierential Equations

16.4

Igor Yanovsky, 2005

156

Traveling Wave Solutions

Consider the Korteweg-de Vries (KdV) equation in the form


ut + 6uux + uxxx = 0,

38

< x < , t > 0.

(16.32)

We look for a traveling wave solution


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

(16.33)

We get the ODE


cf  + 6f f  + f  = 0.

(16.34)

We integrate (16.34) to get


cf + 3f 2 + f  = a,

(16.35)

where a is a constant. Multiplying this equality by f  , we obtain


cf f  + 3f 2 f  + f  f  = af  .
Integrating again, we get
(f  )2
c
= af + b.
f2 + f3 +
2
2

(16.36)

We are looking for solutions f which satisfy f (x), f  (x), f  (x) 0 as x . (In
which case the function u having the form (16.33) is called a solitary wave.) Then
(16.35) and (16.36) imply a = b = 0, so that
(f  )2
c
= 0,
f2 + f3 +
2
2

or f  = f


c 2f .

The solution of this ODE is

c
c
2
f (x) = sech [ (x x0 )],
2
2
where x0 is the constant of integration. A solution of this form is called a soliton.
38

Evans, p. 174; Strauss, p. 367.

Partial Dierential Equations

Igor Yanovsky, 2005

157

Problem (S93, #6). The generalized KdV equation is


1
u 3 u
u
= (n + 1)(n + 2)un

,
t
2
x x3
where n is a positive integer. Solitary wave solutions are sought in which u = f (),
where = x ct and
f, f  , f  0,

as || ;

c, the wave speed, is constant.


Show that
f 2 = f n+2 + cf 2 .
Hence show that solitary waves do not exist if n is even.
Show also that, when n = 1, all conditions of the problem are satised provided c > 0
and

1
c(x ct) .
u = c sech2
2
Proof. We look for a traveling wave solution
u(x, t) = f (x ct).
We get the ODE
1
cf  = (n + 1)(n + 2)f n f  f  ,
2
Integrating this equation, we get
1
cf = (n + 2)f n+1 f  + a,
2
where a is a constant. Multiplying this equality by f  , we obtain
1
cf f  = (n + 2)f n+1 f  f  f  + af  .
2
Integrating again, we get

(16.37)

1
(f  )2
cf 2
= f n+2
+ af + b.
(16.38)
2
2
2
We are looking for solutions f which satisfy f, f  , f  0 as x . Then (16.37)
and (16.38) imply a = b = 0, so that

1 n+2 (f  )2
cf 2
=
f
,

2
2
2

(f  )2 = f n+2 + cf 2 .

We show that solitary waves do not exist if n is even. We have




f  = f n+2 + cf 2 = |f | f n + c,



f  d =
|f | f n + c d,



|f | f n + c d,
f  =



|f | f n + c d.
0 =

Partial Dierential Equations

Igor Yanovsky, 2005

Thus, either |f | 0 f = 0, or
f n + c = 0. Since f 0 as x , we have c = 0 f = 0.
Thus, solitary waves do not exist if n is even.


158

Partial Dierential Equations

Igor Yanovsky, 2005

159

When n = 1, we have
(f  )2 = f 3 + cf 2 .

(16.39)

We show that all conditions of the problem are satised provided c > 0, including

1
c(x ct) ,
or
u = c sech2
2
 c 2


c
2 c

f = c sech
= c cosh
=
.
2
2
cosh2 [ 2 c ]
We have
f

= 2c cosh

 c 3
2

 2

(f )

c sinh 2 c
 ,
cosh6 2 c

f3 =
cf 2 =

 2

 c c
 c 3
 c

= c c cosh
,
sinh
sinh
2
2
2
2

c3


cosh6
c3


cosh4

,
c
2

.
c
2

Plugging these into (16.39), we obtain: 39



c3 sinh2 2 c
c3
c3

 +
 ,
=
cosh6 2 c
cosh6 2 c
cosh4 2 c


c3 + c3 cosh2 2 c
c3 sinh2 2 c


=
,
cosh6 2 c
cosh6 2 c


c3 sinh2 2 c
c3 sinh2 2 c

 .
=

cosh6 2 c
cosh6 2 c
Also, f, f  , f  0, as || , since

2

2
c
2 c

= c
=
0, as || .
f () = c sech
c
c
2
cosh2 [ 2 c ]
e[ 2 ] + e[ 2 ]
Similarly, f  , f  0, as || .
39

cosh2 x sinh2 x = 1.
cosh x =

ex + ex
,
2

sinh x =

ex ex
2

Partial Dierential Equations

Igor Yanovsky, 2005

160

Problem (S00, #5). Look for a traveling wave solution of the PDE
utt + (u2 )xx = uxxxx
of the form u(x, t) = v(x ct). In particular, you should nd an ODE for v. Under
the assumption that v goes to a constant as |x| , describe the form of the solution.
Proof. Since (u2 )x = 2uux, and (u2 )xx = 2u2x + 2uuxx, we have
utt + 2u2x + 2uuxx = uxxxx .
We look for a traveling wave solution
u(x, t) = v(x ct).
We get the ODE
c2 v  + 2(v  )2 + 2vv  = v  ,
c2 v  + 2((v )2 + vv ) = v  ,
c2 v  + 2(vv  ) = v  ,
2 



c v + 2vv = v + a,
c2 v + v 2 = v  + as + b,

(exact dierentials)
s = x ct


v  + c2 v + v 2 = a(x ct) + b.
Since v C = const as |x| , we have v  , v  0, as |x| . Thus,  implies
c2 v + v 2 = as + b.
Since |x| , but v C, we have a = 0:
v 2 + c2 v b = 0.

c2 c4 + 4b
.
v=
2

Partial Dierential Equations

Igor Yanovsky, 2005

161

Problem (S95, #2). Consider the KdV-Burgers equation


ut + uux = uxx + uxxx
in which  > 0, > 0.
a) Find an ODE for traveling wave solutions of the form
u(x, t) = (x st)
with s > 0 and
lim (y) = 0

and analyze the stationary points from this ODE.


b) Find the possible (nite) values of
+ = lim (y).
y

Proof. a) We look for a traveling wave solution


u(x, t) = (x st),

y = x st.

We get the ODE


s +  =  + ,
1
s + 2 =  +  + a.
2
Since 0 as y , then  ,  0 as y . Therefore, at y = , a = 0.
We found the following ODE,
1

s
 +  + 2 = 0.

2
In order to nd and analyze the stationary points of an ODE above, we write it as a
rst-order system.
1 = ,
2 =  .
1 =  = 2 ,
1

s

s
1
2 =  =  + 2 = 2 1 + 21 .

2



1 = 2 = 0,
1 = 2 = 0,
1 = 2 = 0,

1 2
1 2
1 = 0;
1 = 0;
2 =  2 s 1 + 2
2 = s 1 + 2
2 = 1 1 (s 12 1 ) = 0.
Stationary points: (0, 0), (2s, 0), s > 0.
1 = 2
= f (1 , 2 ),

s
1
2 = 2 1 + 21 = g(1 , 2 ).

Partial Dierential Equations

Igor Yanovsky, 2005

162

In order to classify a stationary point, need to nd eigenvalues of a linearized system


at that point.
(
) 

f
f
0
1
1
2
.
=
J(f (1 , 2 ), g(1, 2 )) =
g
g
s + 1 1 

For (1 , 2 ) = (0, 0) :

1
det(J|(0,0) I) =  s

'

2
s
4
= 2
2 .
2


>s
If 4
(0,0)
2
If 4
<s
(0,0)



 = 2 +  +

= 0.

R, < 0.
is Stable Improper Node.
C, Re( ) < 0.
is Stable Spiral Point.

For (1 , 2 ) = (2s, 0) :



1
 = 2 + 

det(J|(2s,0) I) =  s


'

2
s
4
= 2
2 + .
+ > 0, < 0.
(2s,0) is Untable Saddle Point.

= 0.

b) Since
lim (y) = 0 = lim (x st),

we may have
lim (y) = lim (x st) = 2s.

y+

That is, a particle may start o at an unstable node (2s, 0) and as t increases, approach
the stable node (0, 0).
A phase diagram with (0, 0) being a stable spiral point, is shown below.

Partial Dierential Equations

Igor Yanovsky, 2005

163

Partial Dierential Equations

Igor Yanovsky, 2005

164

Problem (F95, #8). Consider the equation


ut + f (u)x = uxx
where f is smooth and  > 0. We seek traveling wave solutions to this equation,
i.e., solutions of the form u = (x st), under the boundary conditions
u uL and ux 0 as x ,
u uR and ux 0 as x +.
Find a necessary and sucient condition on f , uL , uR and s for such traveling waves
to exist; in case this condition holds, write an equation which denes implicitly.
Proof. We look for traveling wave solutions
u(x, t) = (x st),

y = x st.

The boundary conditions become


uL and  0 as x ,
uR and  0 as x +.

Since f ((x st))x = f  () , we get the ODE


s + f  () =  ,
s + (f ()) =  ,
s + f () =  + a,
s + f ()
+ b.
 =

We use boundary conditions to determine constant b:
At x = ,
At x = +,
s=

suL + f (uL )
+b

suR + f (uR )
+b
0 =  =

0 =  =

suL f (uL )
.

suR f (uR )
b=
.

b=

f (uL ) f (uR )
.
uL uR

40
40

For the solution for the second part of the problem, refer to Chiu-Yens solutions.

Partial Dierential Equations

Igor Yanovsky, 2005

165

Problem (S02, #5; F90, #2). Fishers Equation. Consider


ut = u(1 u) + uxx ,

< x < , t > 0.

The solutions of physical interest satisfy 0 u 1, and


lim u(x, t) = 0,

lim u(x, t) = 1.

x+

One class of solutions is the set of wavefront solutions. These have the form u(x, t) =
(x + ct), c 0.
Determine the ordinary dierential equation and boundary conditions which must
satisfy (to be of physical interest). Carry out a phase plane analysis of this equation,
and show that physically interesting wavefront solutions are possible if c 2, but not if
0 c < 2.
Proof. We look for a traveling wave solution
u(x, t) = (x + ct),

s = x + ct.

We get the ODE


c = (1 ) +  ,
 c + 2 = 0,

(s) 0, as s ,
(s) 1, as s +,

0 1.

In order to nd and analyze the stationary points of an ODE above, we write it as a


rst-order system.
y1 = ,
y2 =  .
y1 =  = y2 ,
y2 =  = c + 2 = cy2 y1 + y12 .


y2 = 0,
y1 = y2 = 0,


2
y2 = cy2 y1 + y1 = 0;
y1 (y1 1) = 0.
Stationary points: (0, 0), (1, 0).
y1 = y2
y2

= cy2

= f (y1 , y2 ),
y1 + y12

= g(y1 , y2 ).

In order to classify a stationary point, need to nd eigenvalues of a linearized system


at that point.
(
) 

f
f
0
1
y1
y2
.
=
J(f (y1 , y2 ), g(y1, y2 )) =
g
g
2y1 1 c
y
y
1

Partial Dierential Equations


For (y1 , y2 ) = (0,0) :

1
det(J|(0,0) I) = 
1 c
=

Igor Yanovsky, 2005

166



 = 2 c + 1 = 0.


c2 4
.
2

If c 2 R, > 0.
(0,0) is Unstable Improper (c > 2) / Proper (c = 2) Node.
If 0 c < 2 C, Re( ) 0.
(0,0) is Unstable Spiral Node.
For (y1 , y2 ) = (1,0) :

1
det(J|(1,0) I) = 
1 c
=



 = 2 c 1 = 0.


c2 + 4
.
2

If c 0 + > 0, < 0.
(1,0) is Unstable Saddle Point.
By looking at the phase plot, a particle may start o at an unstable node (0, 0) and as
t increases, approach the unstable node (1, 0).

Partial Dierential Equations

Igor Yanovsky, 2005

167

Partial Dierential Equations

Igor Yanovsky, 2005

168

Problem (F99, #6). For the system


t + x u = 0
t u + x (u) = x2 u
look for traveling wave solutions of the form (x, t) = (y = x st), u(x, t) = u(y =
x st). In particular
a) Find a rst order ODE for u.
b) Show that this equation has solutions of the form
u(y) = u0 + u1 tanh(y + y0 ),
for some constants u0 , u1 , , y0 .
Proof. a) We rewrite the system:
t + ux = 0
ut + x u + ux = uxx
We look for traveling wave solutions
(x, t) = (x st),

u(x, t) = u(x st),

y = x st.

We get the system of ODEs



s + u = 0,
su +  u + u = u .
The rst ODE gives
1 
u,
 =
s
1
u + a,
=
s
where a is a constant, and integration was done with respect to y. The second ODE
gives

1
1
u + a u = u ,
su + u u +
s
s
2 

Integrating, we get
su + uu + au = u .
s
1
su + u2 + au = u + b.
s
u =

1 2
u + (a s)u b.
s

b) Note that the ODE above may be written in the following form:
u + Au2 + Bu = C,
which is a nonlinear rst order equation.

Partial Dierential Equations

Igor Yanovsky, 2005

169

Problem (S01, #7). Consider the following system of PDEs:


ft + fx = g 2 f 2
gt gx = f 2 g
a) Find a system of ODEs that describes traveling wave solutions of the PDE
system; i.e. for solutions of the form f (x, t) = f (x st) and g(x, t) = g(x st).
b) Analyze the stationary points and draw the phase plane for this ODE system in the
standing wave case s = 0.
Proof. a) We look for traveling wave solutions
f (x, t) = f (x st),

g(x, t) = g(x st).

We get the system of ODEs


sf  + f  = g 2 f 2 ,
sg  g  = f 2 g.
Thus,
f =
g =

g2 f 2
,
1s
f2 g
.
1 s

b) If s = 0, the system becomes



f  = g2 f 2,
g = g f 2.
Relabel the variables f y1 , g y2 .

y1 = y22 y12 = 0,
y2 = y2 y12 = 0.
Stationary points: (0, 0), (1, 1), (1, 1).

y1 = y22 y12 = (y1 , y2 ),
y2 = y2 y12 = (y1 , y2 ).
In order to classify a stationary point, need to nd eigenvalues of a linearized system
at that point.
(
) 


2y1 2y2
y1
y2
.
=
J((y1 , y2 ), (y1, y2 )) =

2y1 1
y
y
1

For (y1 , y2 ) = (0,0) :



0
det(J|(0,0) I) = 
0 1



 = (1 ) = 0.



1 = 0, 2 = 1;
(0,0) is Unstable Node.

eigenvectors: v1 =

1
0


, v2 =

0
1


.

Partial Dierential Equations

For (y1 , y2 ) = (1, 1) :


 2
2
det(J|(1,1) I) = 
2
1
3
=
2

Igor Yanovsky, 2005



 = 2 3 2 = 0.


17
.
2

< 0, + > 0.
(-1,1) is Unstable Saddle Point.
For (y1 , y2 ) = (1,1) :
 2
2
det(J|(1,1) I) = 
2
1

1
7
= i
.
2
2
Re( ) < 0.
(1,1) is Stable Spiral Point.



 = 2 + + 2 = 0.


170

Partial Dierential Equations

16.5

Igor Yanovsky, 2005

171

Dispersion

Problem (S97, #8). Consider the following equation


f (v) = v 2 v,

ut = (f (ux ))x uxxxx ,

(16.40)

with constant .
a) Linearize this equation around u = 0 and nd the principal mode solution of the
form et+ikx . For which values of are there unstable modes, i.e., modes with = 0
for real k? For these values, nd the maximally unstable mode, i.e., the value of k with
the largest positive value of .
b) Consider the steady solution of the (fully nonlinear) problem. Show that the resulting
equation can be written as a second order autonomous ODE for v = ux and draw the
corresponding phase plane.
Proof. a) We have
ut = (f (ux))x uxxxx ,
ut = (u2x ux )x uxxxx ,
ut = 2ux uxx uxx uxxxx .

However, we need to linearize (16.40) around u = 0. To do this, we need to linearize f .


f (u) = f (0) + uf  (0) +

u2 
f (0) + = 0 + u(0 1) + = u + .
2

Thus, we have
ut = uxx uxxxx .
Consider u(x, t) = et+ikx .
et+ikx = (k2 k4 )et+ikx ,
= k2 k4 .
To nd unstable nodes, we set = 0, to get
=

1
.
k2

To nd the maximally unstable mode, i.e., the value of k with the largest positive
value of , consider
(k) = k2 k4 ,
 (k) = 2k 4k3 .
To nd the extremas of , we set  = 0. Thus,the extremas are at

1
.
k1 = 0, k2,3 =
2
To nd if the extremas are maximums or minimums, we set  = 0:
 (k) = 2 12k2 = 0,
k = 0 is the minimum.
 (0) = 2 > 0

 
1
1

= 4 < 0 k =
is the maximum unstable mode.

2
2
  
1
1
=
is the largest positive value of .

2
4


Partial Dierential Equations

Igor Yanovsky, 2005

172

b) Integrating , we get
u2x ux uxxx = 0.
Let v = ux . Then,
v 2 v vxx = 0, or
v2 v
.
v  =

In order to nd and analyze the stationary points of an ODE above, we write it as a


rst-order system.
y1 = v,
y2 = v  .
y1 = v  = y2 ,
y 2 y1
v2 v
= 1
.
y2 = v  =



y1 = y2 = 0,
y2 = 0,

y12 y1

y1 (y1 1) = 0.
y2 = = 0;
Stationary points: (0, 0), (1, 0).
y1 = y2
= f (y1 , y2 ),
2
y y1
= g(y1 , y2 ).
y2 = 1

In order to classify a stationary point, need to nd eigenvalues of a linearized system


at that point.
(
) 

f
f
0
1
y1
y2
.
= 2y1 1
J(f (y1 , y2 ), g(y1, y2 )) =
g
g
0

y1
y2
'
For (y1 , y2 ) = (0, 0), = 1 .
+ > 0, < 0. (0,0) is Unstable Saddle Point.
If < 0, R, '

If > 0, = i

C, Re( ) = 0. (0,0) is Spiral Point.


'
For (y1 , y2 ) = (1, 0), = 1 .
'
If < 0, = i 1 C, Re( ) = 0. (1,0) is Spiral Point.
If > 0, R, + > 0, < 0. (1,0) is Unstable Saddle Point.

Partial Dierential Equations

Igor Yanovsky, 2005

173

Partial Dierential Equations

16.6

Igor Yanovsky, 2005

174

Energy Methods

Problem (S98, #9; S96, #5). Consider the following initial-boundary value
problem for the multi-dimensional wave equation:
in (0, ),
utt = u
u
(x, 0) = g(x)
for x ,
u(x, 0) = f (x),
t
u
u
+ a(x)
=0
on .
n
t
Here, is a bounded domain in Rn and a(x) 0. Dene the Energy integral for this
problem and use it in order to prove the uniqueness of the classical solution of the problem.
Proof.

dE
=0 =
dt

u
ut ds +
n

(utt u)ut dx =
utt ut dx




1 2
1
2
(ut ) dx +
|u| dx +
a(x)u2t ds.
=
2
t
2
t

Thus,

a(x)u2t dx


1
=
2 t

u ut dx

u2t + |u|2 dx.

Let Energy integral be



1
E(t) =
u2 + |u|2 dx.
2 t
In order to prove that the given E(t) 0 from scratch, take its derivative with respect
to t:



dE
(t) =
ut utt + u ut dx
dt



u
ds
ut utt dx +
ut
ut u dx
=
n



ut (utt u) dx
a(x)u2t dx 0.
=




=0

Thus, E(t) E(0).


To prove the uniqueness of the classical solution, suppose u1 and u2 are two solutions
of the initial boundary value problem. Let w = u1 u2 . Then, w satises
wtt = w

in (0, ),

for x ,
w(x, 0) = 0,
wt(x, 0) = 0
w
w
+ a(x)
=0
on .
n
t
We have

1
(wt(x, 0)2 + |w(x, 0)|2) dx = 0.
Ew (0) =
2

Partial Dierential Equations

Igor Yanovsky, 2005

175

Ew (t) Ew (0) = 0 Ew (t) = 0. Thus, wt = 0, wxi = 0 w(x, t) = const = 0.


Hence, u1 = u2 .

Problem (S94, #7). Consider the wave equation


1
utt = u
c2 (x)
u
u
(x)
=0
t
n

x
on R.

Assume that (x) is of one sign for all x (i.e. always positive or always negative).
For the energy

1
1
E(t) =
u2 + |u|2 dx,
2
2 c (x) t
show that the sign of
Proof. We have

dE
dt

is determined by the sign of .


1
u
u
+
u

u
t tt
t dx
2
c (x)



1
u
ut utt dx +
ds
ut
ut u dx
=
c2 (x)
n





1 2
1
utt u dx +
ut dx
ut 2
=
c (x)

(x)



=0


> 0, if (x) > 0, x ,
1 2
ut dx =
=
< 0, if (x) < 0, x .
(x)

dE
(t) =
dt

Partial Dierential Equations

Igor Yanovsky, 2005

176

Problem (F92, #2). Let Rn . Let u(x, t) be a smooth solution of the following
initial boundary value problem:
utt u + u3 = 0
u(x, t) = 0

for (x, t) [0, T ]


for (x, t) [0, T ].

a) Derive an energy equality for u. (Hint: Multiply by ut and integrate over


[0, T ].)
b) Show that if u|t=0 = ut |t=0 = 0 for x , then u 0.
Proof. a) Multiply by ut and integrate:





u
3
ut ds +
(utt u + u )ut dx =
utt ut dx
u ut dx +
u3 ut dx
0 =




=0




 2
1 d
1 
1 2
1
1 4
2
(ut ) dx +
|u| dx +
(u ) dx =
ut + |u|2 + u4 dx.
=
2 dt
2
2 t
2 t
4 t
Thus, the Energy integral is

E(t) =

1 
u2t + |u|2 + u4 dx = const = E(0).
2

b) Since u(x, 0) = 0, ut (x, 0) = 0, we have





1
ut (x, 0)2 + |u(x, 0)|2 + u(x, 0)4 dx = 0.
E(0) =
2

Since E(t) = E(0) = 0, we have





1
ut (x, t)2 + |u(x, t)|2 + u(x, t)4 dx = 0.
E(t) =
2

Thus, u 0.

Partial Dierential Equations

Igor Yanovsky, 2005

177

Problem (F04, #3). Consider a damped wave equation



(x, t) R3 R,
utt u + a(x)ut = 0,
u|t=0 = u0 ,
ut |t=0 = u1 .
Here the damping coecient a C0 (R3 ) is a non-negative function and u0 , u1
C0 (R3 ). Show that the energy of the solution u(x, t) at time t,



1
|xu|2 + |ut|2 dx
E(t) =
2 R3
is a decreasing function of t 0.
Proof. Take the derivative of E(t) with respect to t. Note that the boundary integral
is 0 by Huygens principle.



dE
(t) =
u ut + ututt dx
dt
3


R
u
ds
ut
ut u dx +
ututt dx
=
n
R3
R3
R3



=0



ut (u + utt) dx =
ut (a(x)ut) dx =
a(x)u2t dx 0.
=
R3

Thus,

dE
dt

R3

R3

0 E(t) E(0), i.e. E(t) is a decreasing function of t.

Partial Dierential Equations

Igor Yanovsky, 2005

178

Problem (W03, #8). a) Consider the damped wave equation for high-speed waves
(0 <  << 1) in a bounded region D
2 utt + ut = u

with the boundary condition u(x, t) = 0 on D. Show that the energy functional

2 u2t + |u|2 dx
E(t) =
D

is nonincreasing on solutions of the boundary value problem.


b) Consider the solution to the boundary value problem in part (a) with initial data
u (x, 0) = 0, ut (x, 0) =  f (x), where f does not depend on  and < 1. Use part
(a) to show that

|u(x, t)|2 dx 0
D

uniformly on 0 t T for any T as  0.


c) Show that the result in part (b) does not hold for = 1. To do this consider
the case where f is an eigenfunction of the Laplacian, i.e. f + f = 0 in D and
f = 0 on D, and solve for u explicitly.
Proof. a)
dE
dt


=

2 ututt dx +
D

2u ut dx


u
2 ututt dx +
2 ut ds
2uut dx
=
D
D n
D



=0, (u=0 on D)


2
|ut |2 dx 0.
= 2 ( utt u)ut dx =  = 2
D

Thus, E(t) E(0), i.e. E(t) is nonincreasing.


b) From (a), we know dE
dt 0. We also have

2 (ut (x, 0))2 + |u (x, 0)|2 dx
E (0) =

D
2 ( f (x))2 + 0 dx =
2(1) f (x)2 dx 0 as  0.
=
D

2  2
D  (ut )

+
Since E
  (0) E (t) =
Thus, D |u |2 dx 0 as  0.
c) If = 1,

|u |2 dx,

2(1) f (x)2 dx =

E (0) =
D

then E(t) 0 as  0.

f (x)2 dx.

Sincef is independent of , E(0) does not approach 0 as  0. We can not conclude


that D |u(x, t)|2 dx 0.

Partial Dierential Equations

Igor Yanovsky, 2005

179

Problem (F98, #6). Let f solve the nonlinear wave equation


ftt fxx = f (1 + f 2 )1
for x [0, 1], with f (x = 0, t) = f (x = 1, t) = 0 and with smooth initial data f (x, t) =
f0 (x).
a) Find an energy integral E(t) which is constant in time.
b) Show that |f (x, t)| < c for all x and t, in which c is a constant.
Hint: Note that
1 d
f
log(1 + f 2 ).
=
2
1+f
2 df
Proof. a) Since f (0, t) = f (1, t) = 0, t, we have ft (0, t) = ft (1, t) = 0. Let
 1


dE
= 0 =
ftt fxx + f (1 + f 2 )1 ft dx
dt
0
 1
 1
 1
f ft
fttft dx
fxx ft dx +
dx
=
2
0
0
0 1+f
 1
 1
 1
f ft
1
fttft dx [fx ft ]0 +
fx ftx dx +
dx
=
2

0
0
0 1+f
=0
 1
 1
 1
1 2
1 2
1
(ft ) dx +
(fx ) dx +
(ln(1 + f 2 )) dx
=
2
t
2
t
2
t
0
0
0


1 d 1 2
ft + fx2 + ln(1 + f 2 ) dx.
=
2 dt 0
Thus,
1
E(t) =
2


0

1


ft2 + fx2 + ln(1 + f 2 ) dx.

b) We want to show that f is bounded. For smooth f (x, 0) = f0 (x), we have




1 1
ft (x, 0)2 + fx (x, 0)2 + ln(1 + f (x, 0)2 ) dx < .
E(0) =
2 0
Since E(t) is constant in time, E(t) = E(0) < . Thus,



1 1 2
1 1
2
ln(1 + f ) dx
ft + fx2 + ln(1 + f 2 ) dx = E(t) < .
2 0
2 0
Hence, f is bounded.

Partial Dierential Equations

Igor Yanovsky, 2005

Problem (F97, #1). Consider initial-boundary value problem


utt + a2 (x, t)ut u(x, t) = 0

x Rn , 0 < t < +

u(x) = 0

x .

ut(x, 0) = g(x)

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

Prove that L2 -norm of the solution is bounded in t on (0, +).


Here is a bounded domain, and a(x, t), f (x), g(x) are smooth functions.
Proof. Multiply the equation by ut and integrate over :
ut utt + a2 u2t ut u = 0,

2 2
ut utt dx +
a ut dx
ut u dx = 0,





1 d
u
u2t dx +
a2 u2t dx
ut
u ut dx = 0,
ds +
2 dt
n






1 d
2 dt

=0, (u=0, x)

u2t dx

1 d
2 dt

u2t

a2 u2t dx

+ |u|

1 d
+
2 dt

dx =

|u|2 dx = 0,

a2 u2t dx 0.

Let Energy integral be



E(t) =


u2t + |u|2 dx.

0, i.e. E(t) E(0).








2
2
ut (x, 0) + |u(x, 0)| dx =
g(x)2 + |f (x)|2 dx < ,
E(t) E(0) =

We have

dE
dt

since f and g are smooth functions.



 2

ut + |u|2 dx <
E(t) =


|u|2 dx <



2
u dx C
|u|2 dx <

Thus, ||u||2 is bounded t.

Thus,
,
,
,

by Poincare inequality.

180

Partial Dierential Equations

Igor Yanovsky, 2005

181

Problem (S98, #4). a) Let u(x, y, z, t), < x, y, z < be a solution of the
equation

utt + ut = uxx + uyy + uzz


(16.41)
u(x, y, z, 0) = f (x, y, z),

ut (x, y, z, 0) = g(x, y, z).



Here f , g are smooth functions which vanish if x2 + y 2 + z 2 is large enough. Prove
that it is the unique solution for t 0.
b) Suppose we want to solve the same equation (16.41) in the region z 0, <
x, y < , with the additional conditions
u(x, y, 0, t) = f (x, y, t)
uz (x, y, 0, t) = g(x, y, t)
with the same f , g as before in (16.41). What goes wrong?
Proof. a) Suppose u1 and u2 are two solutions. Let w = u1 u2 . Then,

wtt + wt = w,
w(x, y, z, 0) = 0,

wt (x, y, z, 0) = 0.
Multiply the equation by wt and integrate:
wtwtt + wt2 = wtw,


2
wt wtt dx +
wt dx =
wtw dx,
3
3
R3

R
R

w
1 d
2
2
dx
w dx +
wt dx =
wt
w wt dx,
2 dt R3 t
n
R3
R3
R3



=0



1 d
1 d
wt2 dx +
wt2 dx =
|w|2 dx,
2 dt R3
2 dt R3
R3


 2

d
wt2 dx 0,
wt + |w|2 dx = 2
dt R3
3
R





E(t)

dE
dt

0,



wt(x, 0)2 + |w(x, 0)|2 dx = 0,
E(t) E(0) =
3
R
 2

wt + |w|2 dx = 0.
E(t) =
R3

Thus, wt = 0, w = 0, and w = constant. Since w(x, y, z, 0) = 0, we have w 0.


b)

Partial Dierential Equations

Igor Yanovsky, 2005

182

Problem (F94, #8). The one-dimensional, isothermal uid equations with viscosity
and capillarity in Lagrangian variables are
vt ux = 0
ut + p(v)x = uxx vxxx
in which v(= 1/) is specic volume, u is velocity, and p(v) is pressure. The coecients
and are non-negative.
Find an energy integral which is non-increasing (as t increases) if > 0 and conserved if = 0.

Hint: if = 0, E = u2 /2 P (v) dx where P  (v) = p(v).
Proof. Multiply the second equation by u and integrate over R. We use ux = vt .
Note that the boundary integrals are 0 due to nite speed of propagation.
uut + up(v)x = uuxx uvxxx ,




uut dx +
up(v)x dx = uuxx dx uvxxx dx,
R
R
R
R



2
1
(u ) dx +
up(v) ds + ux p(v) dx
2 R t
R
R



=0




2
uux dx ux dx
uvxx dx +
ux vxx dx,
=
R
R
R
R


  

=0

 =0


2
1
(u ) dx +
vtp(v) dx = u2x dx + vt vxx dx,
2 R t
R
R


R


2

1
2
(u ) dx +
P (v) dx = ux dx +
vt vx dx
vxt vx dx,
2 R t
R t
R
R
R
  
=0





2
1
(u ) dx +
P (v) dx +
(vx ) dx = u2x dx,
2 R t
2 R t
R t
R

2


2
u
d
+ P (v) + vx dx = u2x dx 0.
dt R 2
2
R

2

u
+ P (v) + vx2 dx
E(t) =
2
2
R
is nonincreasing if > 0, and conserved if = 0.

Partial Dierential Equations

Igor Yanovsky, 2005

183

Problem (S99, #5). Consider the equation


utt =

(ux )
x

(16.42)

with (z) a smooth function. This is to be solved for t > 0, 0 x 1, with


periodic boundary conditions and initial data u(x, 0) = u0 (x) and ut(x, 0) = v0 (x).
a) Multiply (16.42) by ut and get an expression of the form

d 1
F (ut , ux) = 0
dt 0
that is satised for an appropriate function F (y, z) with y = ut , z = ux ,
where u is any smooth, periodic in space solution of (16.42).
b) Under what conditions on (z) is this function, F (y, z), convex in its variables?
c) What a
` priori inequality is satised for smooth solutions when F is convex?
d) Discuss the special case (z) = a2 z 3 /3, with a > 0 and constant.
Proof. a) Multiply by ut and integrate:
ut utt = ut (ux)x ,
 1
 1
ut utt dx =
ut (ux)x dx,
0
0

 1
1
d 1 u2t
dx =
ut (ux)0

utx (ux) dx = 
dt 0 2
  
0
=0, (2-periodic)
d
Q(ux) = (ux)uxt. Thus,
Let Q (z) = (z), then dt

 1
d 1
utx(ux ) dx =
Q(ux ) dx.
 =
dt 0
0


d 1
u2t
+ Q(ux) dx = 0.
dt 0
2

b) We have
u2t
+ Q(ux ).
2
41 For F to be convex, the Hessian matrix of partial derivatives must be positive denite.
F (ut , ux) =

41

A function f is convex on a convex set S if it satises


f (x + (1 )y) f (x) + (1 )f (y)

for all 0 1 and for all x, y S.


If a one-dimensional function f has two continuous derivatives, then f is convex if and only if
f  (x) 0.
In the multi-dimensional case the Hessian matrix of second derivatives must be positive semi-denite,
that is, at every point x S
yT 2 f (x) y 0,

for all y.

The Hessian matrix is the matrix with entries


[2 f (x)]ij

2 f (x)
.
xi xj

For functions with continuous second derivatives, it will always be symmetric matrix: fxi xj = fxj xi .

Partial Dierential Equations


The Hessian matrix is

Fut ut
2
F (ut , ux ) =
Fux ut
2

y F (x) y =
T

Igor Yanovsky, 2005

Fut ux
Fux ux

y1 y2


=

1
0
0  (ux )

1
0
0  (ux)


y1
y2

184

.
= y12 +  (ux )y22

0.

need

Thus, for a Hessian matrix to be positive denite, need  (ux) 0, so that the above
inequality holds for all y.
c) We have

d 1
F (ut , ux ) dx
dt 0
 1
F (ut , ux ) dx
0
 1
F (ut , ux ) dx
0
 1
2

ut
+ Q(ux) dx
2
0

= 0,
= const,
 1
=
F (ut (x, 0), ux(x, 0)) dx,
0
 1
2

v0
+ Q(u0x ) dx.
=
2
0

d) If (z) = a2 z 3 /3, we have


u2
F (ut , ux) = t + Q(ux )
2

a2 u4x
d 1
u2t
+
dx
dt 0
2
12
 1
2
a2 u4x
ut
+
dx
2
12
0
 1
2
a2 u4x
ut
+
dx
2
12
0

a2 u4x
u2t
+
,
2
12

= 0,
= const,
 1
2
a2 u0 4x
v0
=
+
dx.
2
12
0

Partial Dierential Equations


Problem (S96, #8).

42

Igor Yanovsky, 2005

185

Let u(x, t) be the solution of the Korteweg-de Vries equation


0 x 2,

ut + uux = uxxx,

with 2-periodic boundary conditions and prescribed initial data


u(x, t = 0) = f (x).
a) Prove that the energy integral
 2
u2 (x, t) dx
I1 (u) =
0

is independent of the time t.


b) Prove that the second energy integral,
 2


1
1 2
ux (x, t) + u3 (x, t) dx
I2 (u) =
2
6
0
is also independent of the time t.
c) Assume the initial data are such that I1 (f ) + I2 (f ) < . Use (a) + (b) to prove
that the maximum norm of the solution, |u| = supx |u(x, t)|, is bounded in time.
Hint: Use the following inequalities (here, |u|p is the Lp -norm of u(x, t) at xed time
t):

(one of Sobolevs inequalities);


|u|2 (|u|22 + |ux |22 )
6
(straightforward).
|u|33 |u|22 |u|
Proof. a) Multiply the equation by u and integrate. Note that all boundary terms are
0 due to 2-periodicity.
uut + u2 ux = uuxxx ,
 2
2
uut dx +
u2 ux dx =
uuxxx dx,
0
0
0


 2
2
1 2 3
1 d 2 2

u dx +
(u )x dx = uuxx 0
ux uxx dx,
2 dt 0
3 0
0


1 3 2
1 2 2
1 d 2 2
u dx + u 0 =
(ux )x dx,
2 dt 0
3
2 0

1 2
1 d 2 2
u dx = u2x 0 = 0.
2 dt 0
2
 2
u2 dx = C.
I1 (u) =


 2

Thus, I1 (u) = 0 u (x, t) dx is independent of the time t.


Alternatively, we may dierentiate I1 (u):

 2
 2
d 2 2
dI1
(u) =
u dx =
2uut dx =
2u(uux + uxxx) dx
dt
dt 0
0
0
 2
 2
 2
 2
2
2 3
2

2u ux dx +
2uuxxx dx =
(u )x dx + 2uuxx 0
2ux uxx dx
=
3
0
0
0
0
 2
2
2 2
(u2x )x dx = u2x 0 = 0.
= u3 0
3
0
42

Also, see S92, #7.

Partial Dierential Equations

Igor Yanovsky, 2005

186

b) Note that all boundary terms are 0 due to 2-periodicity.



 2


d 2
1 2 1 3
1
dI2
(u) =
ux + u dx =
ux uxt + u2 ut dx = 
dt
dt 0
2
6
2
0
We dierentiate the original equation with respect to x:
ut = uux + uxxx
utx = (uux )x + uxxxx .

1 2 2
ux ((uux )x + uxxxx ) dx +
u (uux + uxxx ) dx
2 0
0


 2
 2
1 2 3
1 2 2
ux (uux)x dx +
ux uxxxx dx
u ux dx +
u uxxx dx
2 0
2 0
0
0
 2
 2
2
2


uxxuux dx + ux uxxx 0
uxx uxxx dx
ux uux 0 +
0
0


2 1 2
1 2
u4
1

dx + u2 uxx0
2uux uxx dx
2 0
4 x
2
2 0
 2
 2
 2
1 u4 2
uxx uux dx
uxx uxxx dx

uux uxx dx
2 4 0
0
0
0
 2
 2
 2

2 2
uxx uxxx dx = uxx 0 +
uxxx uxx dx =
uxxx uxx dx = 0,

 =
=
=

=
=
since

 2

 2

uxxuxxx dx = + 0
 2

1 2
ux (x, t) +
I2 (u) =
2
0
0

uxx uxxx dx. Thus,



1 3
u (x, t) dx = C,
6

and I2 (u) is independent of the time t.


c) From (a) and (b), we have
 2
u2 dx = ||u||22,
I1 (u) =
0
 2

1
1 2 1 3
1
ux + u dx = ||ux||22 + ||u||33.
I2 (u) =
2
6
2
6
0
Using given inequalities, we have

1
(||u||22 + ||ux||22 )
I1 (u) + 2I2 (u) ||u||33
||u||2
6
6
3

2
I1 (u) + I2 (u) + ||u||2 ||u|| I1 (u) + I2 (u) + I1 (u)||u||

6
3
18
6
3
18
= C + C1 ||u||.
||u||2 C + C1 ||u||,

||u|| C2 .
Thus, ||u|| is bounded in time.

Also see Energy Methods problems for higher order equations (3rd and
4th) in the section on Gas Dynamics.

Partial Dierential Equations

16.7

Igor Yanovsky, 2005

187

Wave Equation in 2D and 3D

Problem (F97, #8); (McOwen 3.2 #90). Solve


utt = uxx + uyy + uzz
with initial conditions
u(x, y, z, 0) = x2 + y 2 ,
  
g(x)

ut (x, y, z, 0) = 
0 .
h(x)

Proof.
We may use the Kirchhos formula:
 


t
1
t
g(x + ct) dS +
h(x + ct) dS
u(x, t) =
4 t
4 ||=1
||=1
 



1
2
2
t
=
(x1 + ct1 ) + (x2 + ct2 ) dS + 0 =
4 t
||=1
We may solve the problem by Hadamards method of descent, since initial conditions are independent of x3 . We need to convert surface integrals in R3 to domain
integrals in R2 . Specically, we need to express the surface measure on the upper half
2
in terms of the two variables 1 and 2 . To do this, consider
of the unit sphere S+
'
1 12 22
over the unit disk 12 + 22 < 1.
f (1 , 2 ) =
'
d1 d2
1 + (f1 )2 + (f2 )2 d1 d2 = 
.
dS =
1 12 22

Partial Dierential Equations

Igor Yanovsky, 2005

188

 

1
g(x1 + ct1 , x2 + ct2 ) d1 d2

2t
u(x1 , x2 , t) =
4 t
1 12 22
12 +22 <1
 

h(x1 + ct1 , x2 + ct2 ) d1d2
t

2
+
4
1 12 22
2 + 2 <1
 1 2

(x1 + t1 )2 + (x2 + t2 )2
1

d1 d2 + 0,
2t
=
4 t
1 12 22
12 +22 <1
 

x21 + 2x1 t1 + t2 12 + x22 + 2x2 t2 + t2 22
1

t
d1 d2
=
2 t
1 12 22
12 +22 <1


tx21 + 2x1 t2 1 + t3 12 + tx22 + 2x2 t2 2 + t3 22
1

d1 d2
=
2 t
1 12 22
12 +22 <1


x21 + 4x1 t1 + 3t2 12 + x22 + 4x2 t2 + 3t2 22
1

d1 d2
=
2
1 12 22
12 +22 <1


(x21 + x22 ) + 4t(x1 1 + x2 2 ) + 3t2 (12 + 22 )
1

d1 d2
=
2
1 12 22
12 +22 <1


4t
d d
x + x2 2
1 2
 1 2
1 1
+
(x1 + x22 )
d1 d2
=
2
2
2
2
2
2
2
2
1 1 2
1 12 22
1 +2 <1
1 +2 <1








2 + 22
3t2
 1
d1 d2 = 
+
2 12 +22 <1 1 12 22





 2  1
1 2
d1 d2
r dr d
1 2
2
2


(x1 + x2 )
(x1 + x2 )
=
=
2
2
1 r2
1 12 22
0
0
12 +22 <1
 2
 1 1


2 du d
1 2
(x1 + x22 )
=
2
u = 1 r 2 , du = 2r dr
1
2
0
0
u2
 2
1 2
(x + x22 )
1 d = x21 + x22 .
=
2 1
0

  1 2
2
4t 1
x1 1 + x2 2
x1 1 + x2 2
4t


d1 d2 =
d1 d2
=

2
2
2 12 +22 <1 1 1 2
2 1 122 1 12 22
= 0.



3t2 2 1 (r cos )2 + (r sin )2
12 + 22
3t2


d1 d2 =
r drd
=
2 12 +22 <1 1 12 22
2 0
1 r2
0




r3
3t2 2 1

drd
u = 1 r 2 , du = 2r dr
=
2 0
1 r2
0
 2

2
t2 2
2
3t
d =
d = 2t2 .
=
2 0 3
0


u(x1 , x2 , t) = + + = x21 + x22 + 2t2 .

We may guess what the solution is:



1
(x + t)2 + (y + t)2 + (x t)2 + (y t)2 = x2 + y 2 + 2t2 .
u(x, y, z, t) =
2

Partial Dierential Equations

Igor Yanovsky, 2005

Check:
u(x, y, z, 0) = x2 + y 2 . 
ut (x, y, z, t) = (x + t) + (y + t) (x t) (y t),
ut (x, y, z, 0) = 0. 
utt (x, y, z, t) = 4,
ux (x, y, z, t) = (x + t) + (x t),
uxx (x, y, z, t) = 2,
uy (x, y, z, t) = (y + t) + (y t),
uyy (x, y, z, t) = 2,
uzz (x, y, z, t) = 0,
utt = uxx + uyy + uzz . 

189

Partial Dierential Equations

Igor Yanovsky, 2005

190

Problem (S98, #6).


Consider the two-dimensional wave equation wtt = a2 w, with initial data which vanish for x2 +y 2 large enough. Prove that w(x, y, t) satises the decay |w(x, y, t)| C t1 .
(Note that the estimate is not uniform with respect to x, y since C may depend on x, y).
Proof. Suppose we have the following problem with initial data:
utt = a2 u

x R2 , t > 0,
ut (x, 0) = h(x)

u(x, 0) = g(x),

x R2 .

The result is the consequence of the Huygens principle and may be proved by Hadamards
method of descent: 43
 

1
g(x1 + ct1 , x2 + ct2 ) d1 d2

2t
u(x, t) =
4 t
1 12 22
12 +22 <1

 
h(x1 + ct1 , x2 + ct2 ) d1d2
t

2
+
4
1 12 22
12 +22 <1

1
th(x + ) + g(x + ) d1 d2
'
=
2 ||2 <c2 t2
c2 t2
||2
1 c2 t2

t
g(x + ) (ct, ct) d1 d2
'
+
.
2 ||2 <c2 t2
c2 t2
||2
1 c2 t2
For a given x, let T (x) be so large that T > 1 and supp(h + g) BT (x). Then for
t > 2T we have:

tM + M + 2M ct d1 d2
1
'
|u(x, t)| =
2 2
2 ||2 <c2 T 2
c2 t2
1 c T
2

c T
2

2 

c2 T 2 4

M 1
M 1
2M c


+
+
.
c2 t
3/4 c2 t
3/4 c2 T t

u(x, t) C1 /t for t > 2T .


For t 2T :

2T M + M + 4M cT d1 d2
1
'
|u(x, t)| =
2
2 ||2<c2 t2
c2 t2
1 c||
2 t2
 ct
1
r dr/c2t2
'
(2T M + M + 4M ct)2
=
2
2
0
1 cr2 t2

M (2T + 1 + 4cT )2T
M (2T + 1 + 4cT ) 1 du
M (2T + 1 + 4cT )
2
.
=
=
1/2
2
2
t
0 u
Letting C = max(C1 , M (2T + 1 + 4cT )2T ), we have |u(x, t)| C(x)/t.
For n = 3, suppose g, h C0 (R3 ). The solution is given by the Kircchos
formula. There is a constant C so that u(x, t) C/t for all x R3 and t > 0. As
McOwen suggensts in Hints for Exercises, to prove the result, we need to estimate the
43

Nicks solution follows.

Partial Dierential Equations

Igor Yanovsky, 2005

191

area of intersection of the sphere of radius ct with the support of the functions g and
h.

Partial Dierential Equations

Igor Yanovsky, 2005

192

Problem (S95, #6). Spherical waves in 3-d are waves symmetric about the origin;
i.e. u = u(r, t) where r is the distance from the origin. The wave equation
utt = c2 u
then reduces to
1
2
utt = urr + ur .
2
c
r

(16.43)

a) Find the general solutions u(r, t) by solving (16.43). Include both the incoming waves
and outgoing waves in your solutions.
b) Consider only the outgoing waves and assume the nite out-ux condition
0 < lim r 2 ur <
r0

for all t. The wavefront is dened as r = ct. How is the amplitude of the wavefront
decaying in time?
Proof. a) We want to reduce (16.43) to the 1D wave equation. Let v = ru. Then
vtt = rutt,
vr = rur + u,
vrr = rurr + 2ur .
Thus, (16.43) becomes
1
1 1
vtt =
vrr ,
2
c r
r
1
vtt = vrr ,
c2
vtt = c2 vrr ,
which has the solution
v(r, t) = f (r + ct) + g(r ct).
Thus,
1
1
1
f (r + ct) + g(r ct) .
u(r, t) = v(r, t) =
r
r
r
  
  
incoming, (c>0)

outgoing, (c>0)

b) We consider u(r, t) = 1r g(r ct):


0 < lim r 2 ur < ,
r0


1
1
0 < lim r 2 g  (r ct) 2 g(r ct) < ,
r0
r

 r
0 < lim rg (r ct) g(r ct) < ,
r0

0 < g(ct) < ,


0 < g(ct) = G(t) < ,

t
.
g(t) = G
c

Partial Dierential Equations

Igor Yanovsky, 2005

The wavefront is dened as r = ct. We have


1
r ct
1
1
g(r ct) = G
= G(0).
u(r, t) =
r
r
c
ct

1  1

|u(r, t)| =
 G(0).
t
c
The amplitude of the wavefront decays like 1t .

193

Partial Dierential Equations

Igor Yanovsky, 2005

194

Problem (S00, #8). a) Show that for a smooth function F on the line, while
u(x, t) = F (ct + |x|)/|x| may look like a solution of the wave equation utt = c2 u
in R3 , it actually is not. Do this by showing that for any smooth function (x, t) with
compact support


u(x, t)(tt ) dxdt = 4 (0, t)F (ct) dt.
R3 R

Note that, setting r = |x|, for any function w which only depends on r one has
w = r 2 (r 2 wr )r = wrr + 2r wr .
b) If F (0) = F  (0) = 0, what is the true solution to utt = u with initial conditions
u(x, 0) = F (|x|)/|x| and ut(x, 0) = F  (|x|)/|x|?
c) (Ralston Hw) Suppose u(x, t) is a solution to the wave equation utt = c2 u in
R3 R with u(x, t) = w(|x|, t) and u(x, 0) = 0. Show that
u(x, t) =

F (|x| + ct) F (|x| ct)


|x|

for a function F of one variable.


Proof. a) We have


 
u (tt ) dxdt = lim dt
u (tt ) dx
0 R
R3 R
|x|>



(utt u) dx +
= lim dt
0

|x|>

|x|=

u
u
dS .
n
n

The nal equality is derived by integrating by parts twice in t, and using Greens
theorem:


 u
v 
u
ds.
(vu uv) dx =
v
n
n

= r
, substituting u(x, t) = F (|x| + ct)/|x|
Since dS = 2 sin  d d and n
gives:

 
u (tt ) dxdt =
4F (ct) dt.
R3

Thus, u is not a weak solution to the wave equation.


b)
c) We want to show that v(|x|, t) = |x|w(|x|, t) is a solution to the wave equation in
one space dimension and hence must have the from v = F (|x| + ct) + G(|x| ct). Then
we can argue that w will be undened at x = 0 for some t unless F (ct) + G(ct) = 0
for all t.
We work in spherical coordinates. Note that w and v are independent of and . We
have:
1
1
vtt(r, t) = c2 w = c2 2 (r 2 wr )r = c2 2 (2rwr + r 2 wrr ),
r
r
rwtt = c2 rwrr + 2wr .
Thus we see that vtt = c2 vrr , and we can conclude that
v(r, t) = F (r + ct) + G(r ct)
F (r + ct) + G(r ct)
.
w(r, t) =
r

and

Partial Dierential Equations

Igor Yanovsky, 2005

limr0 w(r, t) does not exist unless F (ct) + G(ct) = 0 for all t. Hence
w(r, t) =
u(x, t) =

F (ct + r) + G(ct r)
,
and
r
F (ct + |x|) + G(ct |x|)
.
|x|

195

Partial Dierential Equations

17

Igor Yanovsky, 2005

196

Problems: Laplace Equation

A fundamental solution K(x) for the Laplace operator is a distribution satisfying

44

K(x) = (x)
The fundamental solution for the Laplace operator is

1
if n = 2
2 log |x|
K(x) =
1
2n
|x|
if n 3.
(2n)n

17.1

Greens Function and the Poisson Kernel

Greens function is a special fundamental solution satisfying



G(x, ) = (x)
for x
G(x, ) = 0
for x ,

45

(17.1)

To construct the Greens function,


consider w (x) with w (x) = 0 in and w (x) = K(x ) on ;
consider G(x, ) = K(x ) + w (x), which is a fundamental solution satisfying
(17.1).
Problem 1. Given a particular distribution solution to the set of Dirichlet problems

for x
u (x) = (x)
for x ,
u (x) = 0
how would you use this to solve

u = 0
for x
u(x) = g(x)
for x .
Proof. u (x) = G(x, ), a Greens function. G is a fundamental solution to the Laplace
operator, G(x, ) = 0, x . In this problem, it is assumed that G(x, ) is known for
. Then


G(x, )
G(x, )u dx +
u(x)
dSx
u() =
nx

for every u C 2 (). In particular, if u = 0 in and u = g on , then we obtain


the Poisson integral formula

u() =

G(x, )
g(x) dSx,
nx


We know that u(x) = Rn K(x y)f (y) dy is a distribution solution of u = f when f is integrable
and has compact support. In particular, we have

K(x y)u(y) dy
whenever u C0 (Rn ).
u(x) =
44

Rn

The above result is a consequence of:




(x y)u(y) dy = (K) u = K (u) =
K(x y) u(y) dy.
u(x) =

45

Greens function is useful in satisfying Dirichlet boundary conditions.

Partial Dierential Equations


G(x,)

Igor Yanovsky, 2005

197

where H(x, ) = nx is the Poisson kernel.


Thus if we know that the Dirichlet problem has a solution u C 2 (), then we can
calculate u from the Poisson integral formula (provided of course that we can compute
G(x, )).

Partial Dierential Equations

Igor Yanovsky, 2005

198

Dirichlet Problem on a Half-Space. Solve the n-dimensional Laplace/Poisson


equation on the half-space with Dirichlet boundary conditions.
Proof. Use the method of reection to construct Greens function. Let be an
upper half-space in Rn . If x = (x , xn ), where x Rn1 , we can see
|x | = |x |,

and hence

K(x ) = K(x ). Thus

G(x, ) = K(x ) K(x )


is the Greens function on . G(x, ) is harmonic in ,
and G(x, ) = 0 on .
To compute the Poisson kernel, we must dierentiate G(x, )
in the negative xn direction. For n 2,
xn n

K(x ) =
|x |n ,
xn
n
so that the Poisson kernel is given by

2n 

G(x, )xn =0 =
|x |n ,

xn
n

for x Rn1 .

Thus, the solution is



u() =

2n
G(x, )
g(x) dSx =
nx
n


Rn1

g(x)
dx .
|x |n

If g(x ) is bounded and continuous for x Rn1 , then u() is C and harmonic in Rn+
and extends continuously to Rn+ such that u(  ) = g(  ).

Partial Dierential Equations

Igor Yanovsky, 2005

199

Problem (F95, #3): Neumann Problem on a Half-Space.


a) Consider the Neumann problem in the upper half plane,
= {x = (x1 , x2 ) : < x1 < , x2 > 0}:
u = ux1 x1 + ux2 x2 = 0
ux2 (x1 , 0) = f (x1 )

x ,
< x1 < .

Find the corresponding Greens function and conclude that



1
ln [(x1 1 )2 + 22 ] f (x1 ) dx1
u() = u(1 , 2 ) =
2
is a solution of the problem.

b) Show that this solution is bounded in if and only if f (x1 ) dx1 = 0.
Proof. a) Notation: x = (x, y), = (x0 , y0 ). Since K(x ) =
First, we nd the Greens function. We have

1
log (x x0 )2 + (y y0 )2 .
K(x ) =
2
Let G(x, ) = K(x ) + (x).
Since the problem is Neumann, we need:

G(x, ) = (x ),
G
y ((x, 0), ) = 0.

1
2

log |x |, n = 2.


1
log (x x0 )2 + (y y0 )2 + ((x, y), ),
2
1
y y0
G
((x, y), ) =
+ y ((x, y), ),
y
2 (x x0 )2 + (y y0 )2
1
y0
G
((x, 0), ) =
+ y ((x, 0), ) = 0.
y
2 (x x0 )2 + y02
G((x, y), ) =

Let


a
log (x x0 )2 + (y + y0 )2 .
Then,
2
1
y0
y0
a
G
((x, 0), ) =
+
= 0.
2
2
y
2 (x x0 ) + y0
2 (x x0 )2 + y02
((x, y), ) =

Thus, a = 1.
G((x, y), ) =



1
1
log (x x0 )2 + (y y0 )2 +
log (x x0 )2 + (y + y0 )2 .
2
2

46

Consider Greens identity (after cutting out B () and having  0):





u
G
dS
(uG G u ) dx =
G
u

n
n


=0
=0

46

Note that for the Dirichlet problem, we would have gotten the - sign instead of + in front of

Partial Dierential Equations


Since

Igor Yanovsky, 2005

= f (x), we have

u (x ) dx =
G((x, y), ) f (x) dx,


G((x, y), ) f (x) dx.
u() =

u
n

u
(y)

For y = 0, we have
G((x, y), ) =
=
=
Thus,
u() =

1
2

'
'
1
1
2
2
log (x x0 ) + y0 +
log (x x0 )2 + y02
2
2
'
1
2 log (x x0 )2 + y02
2


1
log (x x0 )2 + y02 .
2



log (x x0 )2 + y02 f (x) dx. 

b) Show that this solution is bounded in if and only if

f (x1 ) dx1

= 0.

Consider the Greens identity:






u
u
dS =
dx =
u dxdy =
f (x) dx = 0.
y

n
Note that the Greens identity applies to bounded domains .
 2
 R
u
R d = 0.
f dx1 +
r
R
0
???

200

Partial Dierential Equations

Igor Yanovsky, 2005

201

McOwen 4.2 # 6. For n = 2, use the method of reections to nd the Greens


function for the rst quadrant = {(x, y) : x, y > 0}.
Proof. For x ,
|x (0)| |x (2)| = |x (1)| |x (3)|,
|x (0)| =

|x (1) | |x (3)|
.
|x (2)|

But (0) = , so for n = 2,


G(x, ) =

1
|x (1)| |x (3)|
1
.
log |x |
log
2
2
|x (2)|

G(x, ) = 0, x .

Problem. Use the method of images to solve


G = (x )
in the rst quadrant with G = 0 on the boundary.
Proof. To solve the problem in the rst quadrant
we take a reection to the fourth quadrant
and the two are reected to the left half.
G = (x (0)) (x (1)) (x (2)) + (x (3)).
G =
=

|x (0) | |x (3)|
1
log
2
|x (1) | |x (2)|


(x x0 )2 + (y y0 )2 (x + x0 )2 + (y + y0 )2
1

log 
.
2
(x x0 )2 + (y + y0 )2 (x + x0 )2 + (y y0 )2

Note that on the axes G = 0.

Partial Dierential Equations

Igor Yanovsky, 2005

202

Problem (S96, #3). Construct a Greens function for the following mixed DirichletNeumann problem in = {x = (x1 , x2 ) R2 : x1 > 0, x2 > 0}:
u =

2u 2u
+
= f,
x21 x22
ux2 (x1 , 0) = 0,

x1 > 0,

u(0, x2) = 0,

x2 > 0.

x ,

Proof. Notation: x = (x, y), = (x0 , y0 ). Since K(x ) =



1
log (x x0 )2 + (y y0 )2 .
K(x ) =
2
Let G(x, ) = K(x ) + (x).
At (0, y), y > 0,
'


1
G (0, y), =
log x20 + (y y0 )2 + (0, y) = 0.
2
Also,


Gy (x, y), =
=

1
2

log |x |, n = 2.

1
1
2 2(y y0 )
+ wy (x, y)
2 (x x0 )2 + (y y0 )2
y y0
1
+ wy (x, y).
2 (x x0 )2 + (y y0 )2

At (x, 0), x > 0,




y0
1
+ wy (x, 0) = 0.
Gy (x, 0), =
2 (x x0 )2 + y02
We have


a
log (x + x0 )2 + (y y0 )2
2

b
log (x x0 )2 + (y + y0 )2
+
2

c
log (x + x0 )2 + (y + y0 )2 .
+
2
boundary conditions, we have
'
1
log x20 + (y y0 )2 + (0, y)
G((0, y), ) =
2
'
'
'
'
a
b
c
1
log x20 + (y y0 )2 +
log x20 + (y y0 )2 +
log x20 + (y + y0 )2 +
log x20 + (y + y0 )2 .
2
2
2
2
a = 1, c = b. Also,
y0
1
+ wy (x, 0)
Gy ((x, 0), ) =
2 (x x0 )2 + y02
y0
y0
y0
y0
(1)
b
(b)
1

+
+
.

2
2
2
2
2
2
2 (x x0 ) + y0
2 (x + x0 ) + y0
2 (x x0 ) + y0
2 (x + x0 )2 + y02

((x, y), ) =

Using
0 =
=
Thus,
0 =
=

Thus, b = 1, and



1
1
2
2
log (x x0 ) + (y y0 ) + (x) =
log (x x0 )2 + (y y0 )2
G((x, y), ) =
2
2

Partial Dierential Equations


log

(x + x0

)2

+ (y y0

Igor Yanovsky, 2005


)2

203




2
2
2
2
+ log (x x0 ) + (y + y0 ) log (x + x0 ) + (y + y0 ) .

It can be seen that G((x, y), ) = 0 on x = 0, for example.

Partial Dierential Equations

Igor Yanovsky, 2005

204

Dirichlet Problem on a Ball. Solve the n-dimensional Laplace/Poisson equation on


the ball with Dirichlet boundary conditions.
Proof. Use the method of reection to construct Greens function.
a2
Let = {x Rn : |x| < a}. For , dene = ||
2 as its reection in ; note

/ .
a
|x |
=
|x |
||

for |x| = a.

|x | =

||
|x |.
a

From (17.2) we conclude that for x (i.e. |x| = a),

||
1

2
log a |x |
if n = 2

n2
K(x ) =
a
K(x )
if n 3.
||

(17.3)

Dene for x, :


K(x ) 1 log || |x |
2
a
G(x, ) =
 
K(x ) a n2 K(x )

if n = 2
if n 3.

||

Since is not in , the second terms on the RHS are harmonic


in x . Moreover, by (17.3) we have G(x, ) = 0 if x .
Thus, G is the Greens function for .

u() =

a2 ||2
G(x, )
g(x) dSx =
nx
an


|x|=a

(17.2)

g(x)
dSx.
|x |n

Partial Dierential Equations

17.2

Igor Yanovsky, 2005

205

The Fundamental Solution

Problem (F99, #2). Given that Ka(x y) and Kb (x y) are the kernels for
the operators ( aI)1 and ( bI)1 on L2 (Rn ), where 0 < a < b, show that
( aI)( bI) has a fundamental solution of the form c1 Ka + c2 Kb .
Use the preceding to nd a fundamental solution for 2 , when n = 3.
Proof. METHOD :

( aI)u = f

( bI)u = f

u = Ka f

fundamental solution

*af
u
=K

( aI)( bI)u

 (a + b) + abI u
1
f()
u
= 2
( + a)( 2 + b)
1
 new =
K
2
( + a)( 2 + b)
1
(Kb Ka),
Knew =
ba
1
1
, c2 =
.
c1 =
ba
ba
2

*b f
u
=K
if u L2 ,
(
bI)u = (||2 b)
u = f

u = f
(
aI)u = (||2 a)
1
f()
u
= 2
( + a)
*a = 1
K
2 + a


u = Kb f

kernel

1
f()
( 2 + b)
*b = 1
K
2 + b

u
=

= f,
= f,
 new f(),
= K
=

1
1
1
b K
 a),
2
+ 2
=
(K
ba
+b +a
ba

n = 3 is not relevant (may be used to assume Ka, Kb L2 ).


For 2 , a = 0, b = 1 above, or more explicitly
(2 )u = f,
u = f,
( 4 + 2 )
u
 =
 =
K

1
f,
+ 2)
1
1
1
1
1 K
 0.
= 2 2
= 2
+ 2 = K
4
2
( + )
( + 1)
+1
( 4

Partial Dierential Equations

Igor Yanovsky, 2005

206

METHOD :
For u C0 (Rn ) we have:

Ka(x y) ( aI) u(y) dy,
u(x) =
Rn

Kb(x y) ( bI) u(y) dy.
u(x) =
Rn

Let
u(x) = c1 ( bI) (x),
u(x) = c2 ( aI) (x),

for
for

for (x) C0 (Rn ). Then,



Ka(x y) ( aI) c1 ( bI)(y) dy,
c1 ( bI)(x) =
n
R
Kb (x y) ( bI) c2( aI)(y) dy.
c2 ( aI)(x) =
Rn

We add two equations:

(c1 + c2 )(x) (c1 b + c2 a)(x) =

Rn

(c1 Ka + c2 Kb ) ( aI) ( bI) (y) dy.

1
, we have:
If c1 = c2 and (c1 b + c2 a) = 1, that is, c1 = ab

1
(Ka Kb ) ( aI) ( bI) (y) dy,
(x) =
a

b
n
R

which means that

1
ab (Ka

Kb ) is a fundamental solution of ( aI)( bI).

2  = ( 1) = ( 0I)( 1I).


1
in R3 .
( 0I) has fundamental solution K0 = 4r
To nd K, a fundamental solution for ( 1I), we need to solve for a radially
symmetric solution of
( 1I)K = .
In spherical coordinates, in R3 , the above expression may be written as:
2

K  + K  K = 0.
r
Let
1
w(r),
K =
r
1 
1
w 2 w,
K =
r
r
1 
2
2

w 2 w  + 3 w.
K =
r
r
r
Plugging these into , we obtain:
1  1
w w = 0,
or
r
r
w  w = 0.

Partial Dierential Equations

Igor Yanovsky, 2005

207

Thus,
w = c1 er + c2 er ,
er
er
1
w(r) = c1 + c2
.
K =
r
r
r

Suppose v(x) 0 for |x| R and let = BR (0); for small  > 0 let
 = B (0).

Note: ( I)K(|x|) = 0 in  . Consider Greens identity ( = B (0)):







K(|x|)
K(|x|)
v
v
v
dS +
v
dS
K(|x|)v vK(|x|) dx =
K(|x|)
K(|x|)
n
n
n
n

B(0)





=0, since v0 f or xR

We add  K(|x|) v dx +  v K(|x|) dx to LHS to get:





v
K(|x|)
K(|x|)( I)v v ( I)K(|x|) dx =
K(|x|)
v
dS.



n
n

B(0)
= 0, in 


lim

0

K(|x|)( I)v dx =

K(|x|)( I)v dx.

Since K(r) = c1


er
er
+ c2
is integrable at x = 0.
r
r

On B (0), K(|x|) = K(). Thus, 47





 e
 v 



 

v
e 

 
2



v  0, as  0.

dS
=
K()
dS

+
c
4
K(|x|)
max

c


1
2


n


B(0)
B(0) n


K(|x|)
dS
v(x)
n
B (0)



1
c1 e + c2 e + 2 c1 e + c2 e v(x) dS

B (0) 

1



1




c1 e + c2 e
v(x) dS
+ 2 c1 e + c2 e
=


B(0)

1


1
c1 e + c2 e + 2 c1 e + c2 e
v(0) dS
=


B(0)

1


1 



c1 e + c2 e
[v(x) v(0)] dS
+ 2 c1 e + c2 e
+


B (0)

1
2 c1 e + c2 e v(0) 42

4(c1 + c2 )v(0) = v(0).
1

1
, which gives
Thus, taking c1 = c2 , we have c1 = c2 = 8


K(|x|)( I)v dx = lim
K(|x|)( I)v dx = v(0),
0

47

In R3 , for |x| = ,

K(|x|)

K() = c1

K(|x|)
n

e
e
+ c2
.



e

e


K()
e
e
1
1
= c1
2 c2
2
c1 e + c2 e + 2 c1 e + c2 e ,
=
r







since n points inwards. n points toward 0 on the sphere |x| =  (i.e., n = x/|x|).

Partial Dierential Equations

Igor Yanovsky, 2005

208


1 er
er
1
+
cosh(r) is the fundamental solution of
= 4r
that is K(r) = 8
r
r
( I).
1
(Ka Kb ) is a fundamental solution of ( aI)( bI).
By part (a), ab

1
1
(K0 K) = 4r
+
Here, the fundamental solution of ( 0I)( 1I) is 1



1
1
4r cosh(r) = 4r 1 cosh(r) .

Partial Dierential Equations

Igor Yanovsky, 2005

209

Problem (F91, #3). Prove that

1 cos k|x|
4 |x|


is a fundamental solution for ( + k2 ) in R3 where |x| = x21 + x22 + x23 ,
i.e. prove that for any smooth function f (x) with compact support

cos k|x y|
1
f (y) dy
u(x) =
4 R3 |x y|
is a solution to ( + k2 )u = f .
1
Proof. For v C0 (Rn ), we want to show that for K(|x|) = 4
we have ( + k2 )K = , i.e.

K(|x|) ( + k2 )v(x) dx = v(0).

cos k|x|
|x| ,

Rn

Suppose v(x) 0 for |x| R and let = BR (0); for small  > 0 let
 = B (0).

( + k2 )K(|x|) = 0 in  . Consider Greens identity ( = B (0)):





K(|x|)
K(|x|)
v
v
v
dS +
v
dS
K(|x|)v vK(|x|) dx =
K(|x|)
K(|x|)
n
n
n
n

B(0)





=0, since v0 f or xR

We add  k2 K(|x|) v dx  v k2 K(|x|) dx to LHS to get:





K(|x|)
v
2
2
v
dS.
K(|x|)( + k )v v ( + k )K(|x|) dx =
K(|x|)



n
n

B(0)
= 0, in 



K(|x|)( + k )v dx =

lim

0

K(|x|)( + k2 )v dx.

cos kr
is integrable at x = 0.
Since K(r) =
4r

On B (0), K(|x|) = K(). Thus, 48





 cos k 
 v 



 

v


 
2



v  0, as  0.

dS
=
K()
dS

4
K(|x|)
max






n
4
B(0)
B(0) n
48

In R3 , for |x| = ,
K(|x|)

K() =

K(|x|)
n

cos k
.
4

cos k
1

K()
1
k sin k
cos k

k sin k +
,
=
=

2
r
4


4


since n points inwards. n points toward 0 on the sphere |x| =  (i.e., n = x/|x|).

Partial Dierential Equations

Igor Yanovsky, 2005

210


cos k
K(|x|)
1

dS =
k sin k +
v(x) dS

n

B (0)
B(0) 4

1

cos k
=
k sin k +
v(x) dS
4

B (0)


cos k
cos k
1

k sin k +
k sin k +
v(0) dS
[v(x) v(0)] dS
=
4

4

B (0)
B(0)
cos k
cos k
1

k sin k +
v(0) 42
k sin k +
[v(x) v(0)] 42
=
4

4






v(x)

0, (v is continuous)

cos k v(0) v(0).


Thus,

K(|x|)( + k )v dx = lim

0

K(|x|)( + k2 )v dx = v(0),

1 cos kr
is the fundamental solution of  + k2 .
that is, K(r) = 4
r

Problem (F97, #2). Let u(x) be a solution of the Helmholtz equation


u + k2 u = 0

x R3

satisfying the radiation conditions

1
u
,
iku = O 2 , |x| = r .
u=O
r
r
r
Prove that u 0.
Hint: A fundamental solution to the Helmholtz equation is
Use the Green formula.

1 ikr
.
4r e

1 ikr
e , a fundamental solution. Thus, ( + k2 )K = .
Proof. Denote K(|x|) = 4r
Let x0 be any point and = BR (x0 ); for small  > 0 let

 = B (x0 ).
( + k2 )K(|x|) = 0 in  . Consider Greens identity ( = B (x0 )):



K
u
u
K
2
2
K
dS +
K
dS .
u ( + k )K K( + k )u dx =
u
u
n
n
n
n

B (x0 )


 



=0

u(x0 ), as 0

(It can be shown by the method previously used that the integral over B (x0 ) approaches u(x0 ) as  0.) Taking the limit when  0, we obtain


u
eik|xx0 | u
K
eik|xx0 |
K
dS =

dS
u
u
u(x0 ) =
n
n
r 4|x x0 |
4|x x0 | r

  

eik|xx0 |
eik|xx0 |  u
eik|xx0 |
ik
iku dS

u
=
r 4|x x0 |
4|x x0 |
4|x x0 | r




= O(

1
|x|2

); (can be shown)

1
2
O

4R
4R2 = 0.
= O
R
R2
R
R2
Taking the limit when R , we get u(x0 ) = 0.

Partial Dierential Equations

Igor Yanovsky, 2005

211

Problem (S02, #1). a) Find a radially symmetric solution, u, to the equation in


R2 ,
u =

1
log |x|,
2

and show that u is a fundamental solution for 2 , i.e. show



u2 dx
(0) =
R2

for any smooth which vanishes for |x| large.


b) Explain how to construct the Greens function for the following boundary value in
a bounded domain D R2 with smooth boundary D
w=0

w
=0
n
in D.

and

2 w = f

on D,

Proof. a) Rewriting the equation in polar coordinates, we have


1 
1
1
u =
rur r + 2 u =
log r.
r
r
2
For a radially symmetric solution u(r), we have u = 0. Thus,
1
1 
rur r =
log r,
r
2
 
1
r log r,
rur r =
2

r 2 log r
r2
1

,
r log r dr =
rur =
2
4
8
r
r log r

,
ur =
4
8


1
1 2
1
r log r 1 .
r log r dr
r dr =
u =
4
8
8

1 2
r log r 1 .
8

u(r) =

We want to show that u dened above is a fundamental solution of 2 for n = 2. That


is

u 2 v dx = v(0),
v C0 (Rn ).
R2

1 2
r log r is the
See the next page that shows that u dened as u(r) = 8
1 2
2
Fundamental Solution of  . (The 8 r term does not play any role.)

In particular, the solution of


2 = f (x),
if given by
(x) =


R2

u(x y) 2 (y) dy =

1
8


R2



|x y|2 log |x y| 1 f (y) dy.

Partial Dierential Equations

Igor Yanovsky, 2005

212

b) Let


1
|x |2 log |x | 1 .
8
We use the method of images to construct the Greens function.
Let G(x, ) = K(x ) + (x). We need G(x, ) = 0 and G
n (x, ) = 0 for x .
w
2
Consider w (x) with  w (x) = 0 in , w (x) = K(x ) and n (x) = K
n (x )
on .
Note, we can nd the Greens function for the upper-half plane, and then
make a conformal map onto the domain.
K(x ) =

Partial Dierential Equations

Igor Yanovsky, 2005

213

Problem (S97, #6). Show that the fundamental solution of 2 in R2 is given by


V (x1 , x2 ) =

1 2
r ln(r),
8

r = |x |,

and write the solution of


2 w = F (x1 , x2 ).
Hint: In polar coordinates,  =
Proof. Notation:

x = (x1 , x2).

1 2
r r (r r ) + r 2 2 ;

for example, V =

1
2 (1 + ln(r)).

We have

1 2
r log(r),
8
In polar coordinates: (here, V = 0)




1
1 2
1 1

1 
rVr r =
r log(r)
r
r 2r log(r) + r
=
V =
r
r
8
r r
8 r
r




1
1 1
1
2r 2 log(r) + r 2
4r + 4r log r
=
=
8 r
8 r
r
1
(1 + log r).
=
2
V (x) =

The fundamental solution V (x) for 2 is the distribution satisfying: 2 V (r) = (r).

1

1
1 1
(1 + log r) =
(1 + log r) =
r(1 + log r)r r
2 V = (V ) = 
2
2
2 r
1 1
1
1 1
r
(1)r = 0
=
=
for r = 0.
2 r r r
2 r
Thus, 2 V (r) = (r) V is the fundamental solution. 
The approach above is not rigorous. See the next page that shows that
V dened above is the Fundamental Solution of 2 .
The solution of
2 = F (x),
if given by
(x) =

1
V (x y)  (y) dy =
8
R2
2


R2

|x y|2 log |x y| F (y) dy.

Partial Dierential Equations

Igor Yanovsky, 2005

214

Show that the Fundamental Solution of 2 in R2 is given by:


K(x) =

1 2
r ln(r),
8

r = |x |,

(17.4)

Proof. For v C0 (Rn ), we want to show



K(|x|) 2v(x) dx = v(0).
Rn

Suppose v(x) 0 for |x| R and let = BR (0); for small  > 0 let
 = B (0).
K(|x|) is biharmonic (2 K(|x|) = 0) in  . Consider Greens identity ( =
B (0)):





K(|x|) 
K(|x|) 
v
v
2
v
ds +
v
ds
K(|x|)  v dx =
K(|x|)
K(|x|)
n
n
n
n




=0, since v0 f or xR

+
B(0)



K(|x|)2v dx =
K(|x|)v 2 dx.


lim

0

K(|x|) 
v
v
ds +
K(|x|)
n
n

B(0)

K(|x|)

K(|x|) 
v
v
ds.
n
n


Since K(r) is integrable at x = 0.

On B (0), K(|x|) = K(). Thus, 49





 v 








v





K()n 1 max (v)

dS
=
K()
dS

K(|x|)




n
x
B (0)
B(0) n



 1 2
=   log()n  max (v) 0, as  0.
8
x


K(|x|)
dS =
v(x)
n
B (0)

1
v(x) dS
B(0) 2


1
1
v(0) dS +
[v(x) v(0)] dS

=
B(0) 2
B(0) 2


1
v(0) 2 max v(x) v(0) = v(0). 
=
2
xB(0)




0, (v is continuous)



 v 



 1

v
 



dS  = K()
K(|x|)
  dS  (1 + log )2 max |v| 0, as  0.
n
2
x
B(0)
B (0) n






K(|x|)
dS =
v
n
B(0)

49

1 
1
 log 
 v(x) dS
4
8
B(0)
1
 
log  +  2 max |v| 0, as  0.
4
2
xB(0)


Note that for |x| = ,


K(|x|)

K(|x|)
n

1 2
 log ,
8
K()
1
1

=  log 
,
r
4
8

K() =

1
(1 + log ),
2
K
K
1
=
=
.
n
r
2
K =

Partial Dierential Equations




Igor Yanovsky, 2005




K(|x|) v dx = lim

0

K(|x|)2 v dx = v(0).

215

Partial Dierential Equations

17.3

Igor Yanovsky, 2005

216

Radial Variables

Problem (F99, #8). Let u = u(x, t) solve the following PDE in two spatial dimensions
u = 1
for r < R(t), in which r = |x| is the radial variable, with boundary condition
u=0
on r = R(t). In addition assume that R(t) satises
dR
u
= (r = R)

dt
r
with initial condition R(0) = R0 .
a) Find the solution u(x, t).
b) Find an ODE for the outer radius R(t), and solve for R(t).
Proof. a) Rewrite the equation in polar coordinates:

1
1
u = (rur )r + 2 u = 1.
r
r
For a radially symmetric solution u(r), we have u = 0. Thus,
1
(rur )r = 1,
r
(rur )r = r,
r2
rur = + c1 ,
2
r c1
ur = + ,
2
r
r2
u(r, t) = + c1 log r + c2 .
4
Since we want u to be dened for r = 0, we have c1 = 0. Thus,
r2
+ c2 .
4
Using boundary conditions, we have
u(r, t) =

u(R(t), t) =
u(r, t) =

R(t)2
+ c2 = 0
4

c2 =

R(t)2
.
4

r 2 R(t)2
+
.
4
4

b) We have
R(t)2
r2
+
,
4
4
r
,
2
u
R
(r = R) =
,
(from  )
r
2
dt
,
2
t
,
2
t
R(0) = c1 = R0 .
Thus,
c1 e 2 ,

u(r, t) =
u
r
dR
dt
dR
R

=
=
=

log R =
R(t) =

Thus,

Partial Dierential Equations


t

R(t) = R0 e 2 .

Igor Yanovsky, 2005

217

Partial Dierential Equations

Igor Yanovsky, 2005

218

Problem (F01, #3). Let u = u(x, t) solve the following PDE in three spatial dimensions
u = 0
for R1 < r < R(t), in which r = |x| is the radial variable, with boundary conditions
u(r = R(t), t) = 0,

and

u(r = R1 , t) = 1.

In addition assume that R(t) satises


dR
u
= (r = R)

dt
r
with initial condition R(0) = R0 in which R0 > R1 .
a) Find the solution u(x, t).
b) Find an ODE for the outer radius R(t).
Proof. a) Rewrite the equation in spherical coordinates (n = 3, radial functions):

2
1
2
u = 2 (r 2 ur )r = 0.
+
u =
2
r
r r
r
(r 2 ur )r = 0,
r 2 ur = c1 ,
c1
,
ur =
r2
c1
u(r, t) = + c2 .
r
Using boundary conditions, we have
c1
+ c2 = 0
u(R(t), t) =
R(t)
c1
+ c2 = 1.
u(R1 , t) =
R1
This gives
c1 =

R1 R(t)
,
R1 R(t)

u(r, t) =

c2 =

c2 =

c1
,
R(t)

R1
.
R1 R(t)

1
R1
R1 R(t)
+
.
R1 R(t) r R1 R(t)

b) We have
1
R1
R1 R(t)
+
,
R1 R(t) r R1 R(t)
R1 R(t)
1
=
2,
R1 R(t) r
u
R1 R(t)
1
R1
= (r = R) =

=
2
r
R1 R(t) R(t)
(R1 R(t)) R(t)

u(r, t) =
u
r
dR
dt

Thus, an ODE for the outer radius R(t) is



R1
dR
dt = (R(t)R1) R(t) ,
R(0) = R0 ,

R0 > R1 .

(from  )

Partial Dierential Equations

Igor Yanovsky, 2005

219

Problem (S02, #3). Steady viscous ow in a cylindrical pipe is described by the


equation

1
(u )u + p u = 0

on the domain < x1 < , x22 + x23 R2 , where u = (u1 , u2 , u3 ) = (U (x2 , x3), 0, 0)
is the velocity vector, p(x1 , x2 , x3 ) is the pressure, and and are constants.
p
is a constant c, and that U = c/.
a) Show that x
1
b) Assuming further that U is radially symmetric and U = 0 on the surface of the pipe,
determine the mass Q of uid passing through a cross-section of pipe per unit time in
terms of c, , , and R. Note that

Q =
U dx2 dx3 .
{x22 +x23 R2 }

Proof. a) Since u = (u1 , u2 , u3 ) = (U (x2 , x3 ), 0, 0), we have

u u u
1
2
3
,
,
= (U (x2 , x3 ), 0, 0) (0, 0, 0) = 0.
(u )u = (u1 , u2 , u3 )
x1 x2 x3
Thus,

1
p u = 0,

p = u,

p p p
,
,
= (u1 , u2 , u3 ),
x1 x2 x3

p p p
,
,
= (Ux2x2 + Ux3 x3 , 0, 0).
x1 x2 x3
We can make the following observations:
p
x1

= (Ux2 x2 + Ux3 x3 ),



indep. of x1

p
x2
p
x3

= 0

p = f (x1 , x3 ),

= 0

p = g(x1, x2 ).

Thus, p = h(x1 ). But


p
x1

= U,

U

p
x1

c
1 p
= .
x1

is independent of x1 . Therefore,

p
x1

= c.

Partial Dierential Equations

Igor Yanovsky, 2005

220

b) Cylindrical Laplacian in R3 for radial functions is



1
U =
rUr r ,
r

1
c
rUr r =
,
r



cr
,
rUr r =

cr 2
+ c1 ,
rUr =
2
c1
cr
+ .
Ur =
2
r
For Ur to stay bounded for r = 0, we need c1 = 0. Thus,
cr
Ur =
,
2
cr 2
+ c2 ,
U =
4
cR2
+ c2 ,
0 = U (R) =
4
c 2
cr 2 cR2

=
(r R2 ).
U =
4
4
4

Q =
=

{x22+x23 R2 }
cR4

U dx2 dx3

It is not clear why Q is negative?

c
=
4

2
0


0

c
(r R ) rdrd =
4
2


0

R4
d
4

Partial Dierential Equations

17.4

Igor Yanovsky, 2005

221

Weak Solutions

Problem (S98, #2).


A function u H02 () is a weak solution of the biharmonic equation

2
in

 u=f
u=0
on

u
on
n = 0
provided



uv dx =

f v dx

for all test functions v H02 (). Prove that for each f L2 (), there exists a unique
weak solution for this problem. Here, H02 () is the closure of all smooth
 functions in
which vanish on the boundary and with nite H 2 norm: ||u||22 = (u2xx + u2xy +
u2yy ) dxdy < .
Hint: use Lax-Milgram lemma.
Proof. Multiply the equation by v H02 () and integrate over :
2 u = f,

2
 u v dx =
f v dx,



v
u
v ds
ds +
u
uv dx =
f v dx,
n
n



= 0


uv dx =
f v dx .




  





a(u,v)

L(v)

Denote: V = H02 (). Check the following conditions:

a(, ) is continuous: > 0, s.t. |a(u, v)| ||u||V ||v||V ,

a(, ) is V-elliptic: > 0, s.t. a(v, v)

||v||2V ,

v V ;

L() is continuous: > 0, s.t. |L(v)| ||v||V ,

v V.

u, v V ;

Partial Dierential Equations


We have

Igor Yanovsky, 2005

222

50


2







|a(u, v)|2 =  uv dx
(u)2 dx
(v)2 dx ||u||2H 2() ||v||2H 2() . 


a(v, v) =

(v)2 dx ||v||H02() . 






1

1
2
2


2
2
|L(v)| =  f v dx
|f | |v| dx
f dx
v dx

= ||f ||L2() ||v||L2() ||f ||L2() ||v||H 2() . 


0
  

Thus, by Lax-Milgram theorem, there exists a weak solution u H02 ().


Also, we can prove the stability result.
||u||2H 2() a(u, u) = |L(u)| ||u||H02() ,
0

||u||H02()

Let u1 , u2 be two solutions so that


a(u1 , v) = L(v),
a(u2 , v) = L(v)
for all v V . Subtracting these two equations, we see that:
a(u1 u2 , v) = 0

v V.

Applying the stability estimate (with L 0, i.e. = 0), we conclude that


||u1 u2 ||H02() = 0, i.e. u1 = u2 .
50

Cauchy-Schwarz Inequality:

|(u, v)|

||u||||v|| in any norm, for example

|a(u, v)|

|v|dx

a(u, u) 2 a(v, v) 2 ;



1
1
|v| 1 dx = ( |v|2 dx) 2 ( 12 dx) 2 .


|uv|dx (

u2 dx) 2 (

v 2 dx) 2 ;

Poincare Inequality:

||v||H 2() C (v)2 dx.

Greens formula:




(u)2 dx =
(u2xx + u2yy + 2uxx uyy ) dxdy = (u2xx + u2yy 2uxxy uy ) dxdy = (u2xx + u2yy + 2|uxy |2 ) dxdy ||u||2H 2().

Partial Dierential Equations

17.5

Igor Yanovsky, 2005

223

Uniqueness

Problem. The solution of the Robin problem


u
+ u = ,
n

for the Laplace equation is unique when > 0 is a constant.


Proof. Let u1 and u2 be two solutions of the Robin problem. Let w = u1 u2 . Then
w = 0
in ,
w
+ w = 0
on .
n
Consider Greens formula:



u
ds
u v dx =
v
v u dx.

Setting w = u = v gives



w
2
ds
|w| dx =
w
w w dx .
n


 
=0

Boundary condition gives




2
|w| dx =
w 2 ds .


 
 
0

Thus, w 0, and u1 u2 . Hence, the solution to the Robin problem is unique.


Problem. Suppose q(x) 0 for x and consider solutions u C 2 () C 1 () of
u q(x)u = 0

in .

Establish uniqueness theorems for


a) the Dirichlet problem:
u(x) = g(x), x ;
b) the Neumann problem: u/n = h(x), x .
Proof. Let u1 and u2 be two solutions of the Dirichlet or Neumann problem.
Let w = u1 u2 . Then
w q(x)w = 0 in ,
w
=0
on .
w = 0 or
n
Consider Greens formula:



u
ds
u v dx =
v
v u dx.

Setting w = u = v gives

|w|2 dx =

=0,

ds
w
n




Dirichlet or Neumann

w w dx.

Partial Dierential Equations




Igor Yanovsky, 2005

224

|w| dx =





0

q(x)w 2 dx .



0

Thus, w 0, and u1 u2 . Hence, the solution to the Dirichlet and Neumann problems
are unique.
Problem (F02, #8; S93, #5).
Let D be a bounded domain in R3 . Show that a solution of the boundary value problem
2 u = f

in D,

u = u = 0

on D

is unique.
Proof. Method I: Maximum Principle. Let u1 , u2 be two solutions of the boundary
value problem. Dene w = u1 u2 . Then w satises
2 w = 0 in D,
w = w = 0 on D.
So w is harmonic and thus achieves min and max on D w 0.
So w is harmonic, but w 0 on D w 0. Hence, u1 = u2 .
Method II: Greens Identities. Multiply the equation by w and integrate:


w2 w = 0,
w2 w dx = 0,


(w)
ds
w
w()w dx = 0,
n





=0
w
w ds + (w)2 dx = 0.





=0

Thus, w 0. Now, multiply w = 0 by w. We get



|w|2 dx = 0.

Thus, w = 0 and w is a constant. Since w = 0 on , we have w 0.


Problem (F97, #6).
a) Let u(x) 0 be continuous in closed bounded domain Rn , u is continuous in
,
u = u2

and

u| = 0.

Prove that u 0.
b) What can you say about u(x) when the condition u(x) 0 in is dropped?

Partial Dierential Equations

Igor Yanovsky, 2005

225

Proof. a) Multiply the equation by u and integrate:


uu = u3 ,

uu dx =
u3 dx,



u
2
ds
u
|u| dx =
u3 dx,
n


 
=0

 3

u + |u|2 dx = 0.


Since u(x) 0, we have u 0.


b) If we dont know that u(x) 0, then u can not be nonnegative on the entire
domain . That is, u(x) < 0, on some (or all) parts of . If u is nonnegative on ,
then u 0.

Partial Dierential Equations

Igor Yanovsky, 2005

226

Problem (W02, #5). Consider the boundary value problem


u +

n

k=1

u
u3 = 0
xk
u=0

in ,
on ,

where is a bounded domain in Rn with smooth boundary. If the k s are constants,


and u(x) has continuous derivatives up to second order, prove that u must vanish
identically.
Proof. Multiply the equation by u and integrate:
n


u
u u4 = 0,
xk
k=1
 


n
u
uu dx +
k
u dx
u4 dx = 0,
x
k

k=1



 

n
u
u
ds |u|2 dx +
u
k
u dx
u4 dx = 0.
xk
n

k=1







=0

We will show that = 0.





u
u
2
k
u dx =
k u ds
k u
dx,
xk
xk




 =0
u
k
u dx = 0,
2
xk

 
n
u
k
u dx = 0.

xk

uu +

k=1

Thus, we have


u4 dx = 0,
|u|2 dx





2
u4 dx = 0.
|u| +

Hence, |u| = 0 and u4 = 0. Thus, u 0.

Note that
 
n
k=1

u
u dx =
xk

u u dx =

nu2 ds

 
=0

and thus,

u u dx = 0.

u u dx,

Partial Dierential Equations

Igor Yanovsky, 2005

227

Problem (W02, #9). Let D = {x R2 : x1 0, x2 0}, and assume that f is


continuous on D and vanishes for |x| > R.
a) Show that the boundary value problem
u = f

in D,
u
u(x1 , 0) =
(0, x2) = 0
x1
can have only one bounded solution.
b) Find an explicit Greens function for this boundary value problem.
Proof. a) Let u1 , u2 be two solutions of the boundary value problem. Dene w =
u1 u2 . Then w satises
w = 0

in D,
w
(0, x2 ) = 0.
w(x1 , 0) =
x1
Consider Greens formula:



u
ds
u v dx =
v
v u dx.
D
D n
D
Setting w = u = v gives



w
2
ds
|w| dx =
w
w w dx,
n
D
D
D





w
w
w
ds +
ds +
ds
|w|2 dx =
w
w
w
w w dx
n
n
n
D
Rx1
Rx2
|x|>R
D




w
w
w
ds
w(x1 , 0)
ds +
w(0, x2)
ds +
w
w w dx,
=


x1
n
Rx1    x2
Rx2
|x|>R
D

=0
=0
=0
=0

|w|2 dx = 0 |w|2 = 0 w = const.
D

Since w(x1 , 0) = 0 w 0. Thus, u1 = u2 .


b) The similar problem is solved in the appropriate section (S96, #3).
Notice whenever you are on the boundary with variable x,
|x (1) ||x (3)|
.
|x (2)|


|x (1)||x (3)|
1
log |x | log
G(x, ) =
2
|x (2)|

|x (0)| =
So,

is the Greens function.

Partial Dierential Equations

Igor Yanovsky, 2005

Problem (F98, #4). In two dimensions x = (x, y), dene the set a as
a = +
in which
+ = {|x x0 | a} {x 0}
= {|x + x0 | a} {x 0} = +
and x0 = (1, 0). Note that a consists of two components when 0 < a < 1
and a single component when a > 1. Consider the Neumann problem
2 u = f,
u/n = 0,

x a
x a

in which

+

f (x) dx = 1

f (x) dx = 1

a) Show that this problem has a solution for 1 < a, but not for 0 < a < 1.
(You do not need to construct the solution, only demonstrate solveability.)
b) Show that maxa |u| as a 1 from above. (Hint: Denote L to be
the line segment L = + , and note that its length |L| goes to 0 as a 1.)
Proof. a) We use the Greens identity. For 1 < a,



u
ds =
u dx =
f (x) dx
0 =
n
a
 a
a
f (x) dx +
f (x) dx = 1 1 = 0. 
=
+

Thus, the problem has a solution for 1 < a.


For 0 < a < 1, + and are disjoint. Consider + :



u
ds =
u dx =
f (x) dx = 1,
0 =
+ n
+
+



u
ds =
u dx =
f (x) dx = 1.
0 =
n

We get contradictions.
Thus, the solution does not exist for 0 < a < 1.

228

Partial Dierential Equations

Igor Yanovsky, 2005

229

b) Using the Greens identity, we have: (n+ is the unit normal to + )





u
u
u dx =
ds
=
ds,
+
+
n
n
+
+

L




u
u
u
u dx =
ds =
ds =
ds.

n
L n
L n



u
u dx
u dx = 2
ds,
+
+

L n



u
f (x) dx
f (x) dx = 2
ds.
+
+

L

u
ds,
2 = 2
+
L n
 
 


 u 2  u 2
u
 u 
ds
+
|L| max |u| |L| max |u|.
1 =
 +  ds
+
L
a
n+

L n
L n
L
Thus,
max |u|
a

1
.
|L|

As a 1 (L 0)

maxa |u| .

Partial Dierential Equations

Igor Yanovsky, 2005

230

Problem (F00, #1). Consider the Dirichlet problem in a bounded domain D Rn


with smooth boundary D,
u + a(x)u = f (x)
u=0

in D,
on D.

a) Assuming that |a(x)| is small enough, prove the uniqueness of the classical solution.
b) Prove the existence of the solution in the Sobolev space H 1 (D) assuming that f
L2 (D).
Note: Use Poincare inequality.
Proof. a) By Poincare Inequality, for any u C01 (D), we have ||u||22 C||u||22.
Consider two solutions of the Dirichlet problem above. Let w = u1 u2 . Then, w
satises

w + a(x)w = 0 in D,
w=0
on D.
ww + a(x)w 2

ww dx + a(x)w 2 dx


|w|2 dx + a(x)w 2 dx


|w|2 dx
a(x)w 2 dx =


1
w 2 dx
a(x)w 2 dx
C


1
w 2 dx
|a(x)| w 2 dx
C


1
w 2 dx
|a(x)|
C


If |a(x)| < C1
b) Consider

F (v, u) =

= 0,
= 0,
= 0,

1
C

w 2 dx,

(by Poincare inequality)

0,
0,
0.

w 0.


(vu + a(x)vu) dx =

vf (x) dx = F (v).

F (v) is a bounded linear functional on v H 1,2 (D), D = .


|F (v)| ||f ||2||v||2 ||f ||2C||v||H 1,2 (D)
So by Riesz representation, there exists a solution u H01,2 (D) of


vu + a(x)vu dx =
vf (x) dx = F (v)
< u, v >=

v H01,2 (D).

Partial Dierential Equations

Igor Yanovsky, 2005

231

Problem (S91, #8). Dene the operator


Lu = uxx + uyy 4(r 2 + 1)u
in which r 2 = x2 + y 2 .
2
a) Show that = er satises L = 0.
b) Use this to show that the equation
Lu = f
u
=
n

in
on

has a solution only if




f dx =
ds(x).

Proof. a) Expressing Laplacian in polar coordinates, we obtain:


Lu =
L =
=

1
(rur )r 4(r 2 + 1)u,
r
1
1
2
2
(rr )r 4(r 2 + 1) = (2r 2 er )r 4(r 2 + 1)er
r
r
1
2
2
r2
2
r2
(4re + 2r 2re ) 4r 2 er 4er = 0. 
r
2

b) We have = er = ex

51

2 +y2

= ex ey . From part (a),

L = 0,
2
2
2
2
2

= n = (x , y ) n = (2xex ey , 2yex ey ) n = 2er (x, y) (y, x) = 0.


n
Consider two equations:
Lu = u 4(r 2 + 1)u,
L =  4(r 2 + 1).

Multiply the rst equation by and the second by u and subtract the two equations:
Lu = u 4(r 2 + 1)u,
uL = u 4(r 2 + 1)u,
Lu uL = u u.
Then, we start from the LHS of the equality we need to prove and end up with RHS:




f dx =
Lu dx =
(Lu uL) dx =
(u u) dx





u
u
u ) ds =
ds =
(

ds. 
=
n
n

51
The only shortcoming in the above proof is that we assume
n = (y, x), without giving an explanation why it is so.

Partial Dierential Equations

17.6

Igor Yanovsky, 2005

232

Self-Adjoint Operators

Consider an mth-order dierential operator



Lu =
a (x)Du.
||m

The integration by parts formula





uxk v dx =
uvnk ds
uvxk dx

n = (n1 , . . . , nn ) Rn ,

with u or v vanishing near is:




uxk v dx = uvxk dx.

We can repeat the integration by parts with any combination of derivatives


D = (/x1 )1 (/xn)n :



m
(D u)v dx = (1)
uD v dx,
(m = ||).

We have




 

(Lu) v dx =
a (x)D u v dx =
a (x) v D u dx

||m

(1)

||

||m

||m

D (a(x) v) u dx =

L (v) u dx,

for all u C () and v C0 .


The operator
m

L (v) =

(1)||D (a(x) v)

||m

is called the adjoint of L.


The operator is self-adjoint if L = L.
Also, L is self-adjoint if 52



vL(u) dx =

52

L = L

uL(v) dx.

(Lu|v) = (u|L v) = (u|Lv).

(1)||D (a(x) v) u dx

||m



L (v)

Partial Dierential Equations

Igor Yanovsky, 2005

233

Problem (F92, #6).


Consider the Laplace operator  in the wedge 0 x y with boundary conditions
f
=0
x
f
f

=0
x
y

on x = 0
on x = y.

a) For which values of is this operator self-adjoint?


b) For such a value of , suppose that
f = er

2 /2

cos

with these boundary conditions. Evaluate




f ds
r
CR
in which CR is the circular arc of radius R connecting the boundaries x = 0 and x = y.
Proof. a) We have
Lu = u = 0
u
=0
x
u
u

=0
x
y

on x = 0
on x = y.

The operator L is self-adjoint if:



(u Lv v Lu) dx = 0.

(u Lv v Lu) dx =
=
=
=

v
u
v
ds
u
n
n




v
u
u
v
v
ds +
v
ds
u
u
n
n
n
n
x=0
x=y




u (v n) v (u n) ds +
u (v n) v (u n) ds
x=0
x=y






u (vx , vy ) (1, 0) v (ux , uy ) (1, 0) ds
x=0




u (vx , vy ) (1/ 2, 1/ 2) v (ux, uy ) (1/ 2, 1/ 2) ds
+
x=y






u (0, vy ) (1, 0) v (0, uy ) (1, 0) ds


 x=0
= 0




+
u (vy , vy ) (1/ 2, 1/ 2) v (uy , uy ) (1/ 2, 1/ 2) ds
x=y


uv
vu
y ( 1) y ( 1) ds 
= 0.
2
2
x=y
(uv vu) dx =

need

Partial Dierential Equations

Igor Yanovsky, 2005

234

Thus, we need = 1 so that L is self-adjoint.


b) We have = 1. Using Greens identity and results from part (a), ( f
n = 0 on
x = 0 and x = y):






f
f
f
f
f
ds =
ds +
ds.
 f dx =
ds +
ds =
n
n

n
CR n
x=0 
x=y 
CR r
=0

Thus,


CR

f
ds =
r


 f dx =

er

2 /2

=0

cos r drd

1
2
1 R r2 /2
e
r dr = 1 (1 eR /2 ).
=
1
2 0
2

Partial Dierential Equations

Igor Yanovsky, 2005

235

Problem (F99, #1). Suppose that u = 0 in the weak sense in Rn and that there
is a constant C such that

|u(y)| dy < C, x Rn .
{|xy|<1}

Show that u is constant.


Proof. Consider Greens formula:



u
ds
u v dx =
v
v u dx

For v = 1, we have


u
ds =
u dx.
n

Let Br (x0 ) be a ball in Rn . We have





u
u
ds = r n1
(x0 + rx) ds
u dx =
0 =
n
Br (x0 )
Br (x0 )
|x|=1 r

1
n1
n
u(x0 + rx) ds.
= r
r n |x|=1

Thus, 1n |x|=1 u(x0 + rx) ds is independent of r. Hence, it is constant.
By continuity, as r 0, we obtain the Mean Value property:
1
u(x0 ) =
n


|x|=1

u(x0 + rx) ds.


If |xy|<1 |u(y)| dy < C x Rn , we have |u(x)| < C in Rn .
Since u is harmonic and bounded in Rn , u is constant by Liouvilles theorem.
53

Liouvilles Theorem: A bounded harmonic function dened on Rn is a constant.

53

Partial Dierential Equations

Igor Yanovsky, 2005

236

Problem (S01, #1). For bodies (bounded regions B in R3 ) which are not perfectly
conducting one considers the boundary value problem
3


u
0 = (x)u =
(x)
xj
xj
j=1

u = f

on B.

The function (x) is the local conductivity of B and u is the voltage. We dene
operator (f ) mapping the boundary data f to the current density at the boundary by
(f ) = (x)

u
,
n

and /n is the inward normal derivative (this formula denes the current density).
a) Show that is a symmetric operator, i.e. prove


g(f ) dS =
f (g) dS.
B

b) Use the positivity of (x) > 0 to show that is negative as an operator, i.e., prove

f (f ) dS 0.
B

Proof. a) Let

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

(x)v = 0
on ,
v=g
on .

v
u
,
(g) = (x) .
n
n
Since /n is inward normal derivative, Greens formula is:



u
dS

v
(x)
v

(x)u
dx
=
v (x)u dx.


n

(f ) = (x)

=g

We have




u
dS =
g(f ) dS =
g(x)
v (x)u dx
v (x)u dx


n


=0


v
dS +
u(x)
u (x)v dx
=


n


=0


v
dS =
f (x)
f (g) dS. 
=
n

b) We have (x) > 0.




f (f ) dS =



u
dS = u (x)u dx
u(x)
(x)u u dx


n

=0

= (x)|u|2 0. 
 

0

Partial Dierential Equations

Igor Yanovsky, 2005

237

Problem (S01, #4). The Poincare Inequality states that for any bounded domain
in Rn there is a constant C such that


2
|u| dx C
|u|2 dx

for all smooth functions u which vanish on the boundary of .


a) Find a formula for the best (smallest) constant for the domain in terms of the
eigenvalues of the Laplacian on , and
b) give the best constant for the rectangular domain in R2
= {(x, y) : 0 x a, 0 y b}.
Proof. a) Consider Greens formula:



u
ds
u v dx =
v
v u dx.

Setting u = v and with u vanishing on , Greens formula becomes:




|u|2 dx = uu dx.

&
Expanding u in the eigenfunctions of the Laplacian, u(x) = an n (x), the formula
above gives
 





2
|u| dx =
an n (x)
m am m (x) dx =
m an am
n m dx

n=1

m=1

n |an |2 .

m,n=1

n=1

Also,

|u| dx =

 

n=1

an n (x)

am m (x) =

m=1

|an |2 .

n=1

Comparing  and , and considering that n increases as n , we obtain







2
2
2
|u| dx = 1
|an |
n |an | =
|u|2 dx.
1

|u|2 dx

n=1

1
1

n=1

|u|2 dx,

with C = 1/1 .
b) For the rectangular domain = {(x, y) : 0 x a, 0 y b} R2 , the
eigenvalues of the Laplacian are

m2 n2
m, n = 1, 2, . . ..
mn = 2 2 + 2 ,
a
b

1

1
1 = 11 = 2 2 + 2 ,
a
b
1
1
1
.
= 2 1
C =
11
( a2 + b12 )

Partial Dierential Equations

Igor Yanovsky, 2005

238

Partial Dierential Equations

Igor Yanovsky, 2005

239

Problem (S01, #6). a) Let B be a bounded region in R3 with smooth boundary B.


The conductor potential for the body B is the solution of Laplaces equation outside
B
V = 0 in R3 /B
subject to the boundary conditions, V = 1 on B and V (x) tends to zero as |x| .
Assuming that the conductor potential exists, show that it is unique.
b) The capacity C(B) of B is dened to be the limit of |x|V (x) as |x| . Show
that

V
1
dS,
C(B) =
4 B n
where B is the boundary of B and n is the outer unit normal to it (i.e. the normal
pointing toward innity).
c) Suppose that B  B. Show that C(B  ) C(B).
Proof. a) Let V1 , V2 be two solutions of the boundary value problem. Dene W =
V1 V2 . Then W satises

W = 0 in R /B
W =0
on B

W 0 as |x| .
Consider Greens formula:



u
ds
u v dx =
v
v u dx.
B
B n
B
Setting W = u = v gives


2
|W | dx =
B

W
ds
W

n
B
=0

Thus, |W |2 = 0 W = const.


B

W W dx = 0.

=0

Since W = 0 on B, W 0, and V1 = V2 .

b & c) For (b)&(c), see the solutions from Ralstons homework (a few pages
down).

Partial Dierential Equations

Igor Yanovsky, 2005

240

Problem (W03, #2). Let L be the second order dierential operator L =  a(x)
in which x = (x1 , x2 , x3 ) is in the three-dimensional cube C = {0 < xi < 1, i = 1, 2, 3}.
Suppose that a > 0 in C. Consider the eigenvalue problem
Lu = u

for

xC

u=0

for

x C.

a) Show that all eigenvalues are negative.


b) If u and v are eigenfunctions for distinct eigenvalues and , show that u and v
are orthogonal in the appropriate product.
c) If a(x) = a1 (x1 ) + a2 (x2 ) + a3 (x3 ) nd an expression for the eigenvalues and eigenvectors of L in terms of the eigenvalues and eigenvectors of a set of one-dimensional
problems.
Proof. a) We have
u a(x)u = u.
Multiply the equation by u and integrate:
uu a(x)u2 = u2 ,


2
uu dx
a(x)u dx = u2 dx,




u
2
2
ds |u| dx
u
a(x)u dx = u2 dx,
n


 
=0

(|u|2 + a(x)u2 ) dx

< 0.
=
2
u dx


b) Let , , be the eigenvalues and u, v be the corresponding eigenfunctions. We have


u a(x)u = u.

(17.5)

v a(x)v = v.

(17.6)

Multiply (17.5) by v and (17.6) by u and subtract equations from each other
vu a(x)uv = uv,
uv a(x)uv = uv.
vu uv = ( )uv.
Integrating over gives




uv dx,
vu uv dx = ( )




v
u
u
uv dx.
v
dx = ( )
 n  n

=0

Since = , u and v are orthogonal on .

Partial Dierential Equations

Igor Yanovsky, 2005

c) The three one-dimensional eigenvalue problems are:


u1x1 x1 (x1 ) a(x1 )u1 (x1 ) = 1 u1 (x1 ),
u2x2 x2 (x2 ) a(x2 )u2 (x2 ) = 2 u2 (x2 ),
u3x3 x3 (x3 ) a(x3 )u3 (x3 ) = 3 u3 (x3 ).
We need to derive how u1 , u2 , u3 and 1 , 2 , 3 are related to u and .

241

Partial Dierential Equations

17.7

Igor Yanovsky, 2005

242

Spherical Means

Problem (S95, #4). Consider the biharmonic operator in R3


 2


2
2 2
2
 u
+
+
u.
x2 y 2 z 2
a) Show that 2 is self-adjoint on |x| < 1 with the following boundary conditions on
|x| = 1:
u = 0,
 u = 0.
Proof. a) We have
Lu = 2 u = 0
u = 0

on |x| = 1

u = 0

on |x| = 1.

The operator L is self-adjoint if:



(u Lv v Lu) dx = 0.

(u 2 v v 2 u) dx




v
u
ds u (v) dx
ds +
u
v
v (u) dx
=
n
n



 
 
=0



 =0
u
v
ds +
ds
v
uv dx +
u
vu dx = 0. 
=
n
n



 
 

(u Lv v Lu) dx =

=0

=0

Partial Dierential Equations

Igor Yanovsky, 2005

243

b) Denote |x| = r and dene the averages



u(x) ds,
S(r) = (4r 2)1

V (r) =

|x|=r

4 3
r
3

1 

|x|r

u(x) dx.

Show that
d
r
S(r) = V (r).
dr
3
Hint: Rewrite S(r) as an integral over the unit sphere before dierentiating; i.e.,

1
u(rx ) dx.
S(r) = (4)
|x |=1

c) Use the result of (b) to show that if u is biharmonic, i.e. 2 u = 0, then


r2
u(0).
6
Hint: Use the mean value theorem for u.
S(r) = u(0) +

b) Let x = x/|x|. We have


S(r) =
dS
dr

=
=

54




1
1
1
 2
u(x) dSr =
u(rx ) r dS1 =
u(rx ) dS1 .
4r 2 |x|=r
4r 2 |x |=1
4 |x |=1



1
1
1
u
u
u


(rx ) dS1 =
(rx ) dS1 =
(x) dSr
2
4 |x |=1 r
4 |x|=1 n
4r |x|=r n

1
u dx. 
4r 2 |x|r

where we have used Greens identity in the last equality. Also



1
r
V (r) =
u dx. 
3
4r 2 |x|r
c) Since u is biharmonic (i.e. u is harmonic), u has a mean value property. We
have

 
r
r 4 3 1
r
d
S(r) =
V (r) =
r
u(x) dx = u(0),
dr
3
3 3
3
|x|r
S(r) =

r2
r2
u(0) + S(0) = u(0) +
u(0).
6
6

54

Change of variables:
Surface integrals: x = rx in R3 :


u(x) dS =
u(rx ) r2 dS1 .
|x |=1

|x|=r

Volume integrals: = r in Rn :


h(x + ) d  =
h(x + r) rn d.
|  |<r

||<1

Partial Dierential Equations

Igor Yanovsky, 2005

244

Partial Dierential Equations

Igor Yanovsky, 2005

Problem (S00, #7). Suppose that u = u(x) for x R3 is biharmonic;


i.e. that 2 u (u) = 0. Show that

2 1
u(x) ds(x) = u(0) + (r 2 /6)u(0)
(4r )
|x|=r

through the following steps:


a) Show that for any smooth f ,


d
f (x) dx =
f (x) ds(x).
dr |x|r
|x|=r
b) Show that for any smooth f ,


d
2 1
2 1
(4r )
f (x) ds(x) = (4r )
n f (x, y) ds
dr
|x|=r
|x|=r
in which n is the outward normal to the circle |x| = r.
c) Use step (b) to show that


d
(4r 2 )1
f (x) ds(x) = (4r 2)1
f (x) dx.
dr
|x|=r
|x|r
d) Combine steps (a) and (c) to obtain the nal result.
Proof. a) We can express the integral in Spherical Coordinates:
 R  2 

f (x) dx =
f (, , r) r 2 sin d d dr.
|x|R

d
dr

 R  2 
d
f (x) dx =
f (, , r) r 2 sin d d dr = ???
dr
0
0
0
|x|R
 2 
f (, , r) R2 sin d d
=
0
0
f (x) dS.
=
|x|=R

55

55

Dierential Volume in spherical coordinates:


d3 = 2 sin d d d.

Dierential Surface Area on sphere:


dS = 2 sin d d.

245

Partial Dierential Equations


b&c) We have



1
d
f (x) dS
=
dr 4r 2 |x|=r
=
=
=

Igor Yanovsky, 2005

246






1
d
1 d
 2

f (rx ) r dS1 =
f (rx ) dS1
dr 4r 2 |x |=1
4 dr
|x |=1


1
f
f
1

(rx ) dS1 =
(rx ) dS1
4 |x |=1 r
4 |x |=1 n


1
1
f
(x) dS =
f n dS 
4r 2 |x|=r n
4r 2 |x|=r

1
f dx. 
4r 2 |x|r

Greens formula was used in the last equality.


Alternatively,





 2 
1
1
d
d
2
f (x) dS
=
f (, , r) r sin d d
dr 4r 2 |x|=r
dr 4r 2 0
0
  2 

1
d
f (, , r) sin d d
=
dr 4 0
0
 2 
f
1
(, , r) sin d d
=
4 0
0 r
 2 
1
f n sin d d
=
4 0
0
 2 
1
f n r 2 sin d d
=
4r 2 0
0

1
f n dS 
=
4r 2 |x|=r

1
f dx. 
=
4r 2 |x|=r
d) Since f is biharmonic (i.e. f is harmonic), f has a mean value property. From
(c), we have 56





1
1
r 1
d
f (x) ds(x)
=
f (x) dx =
f (x) dx
dr 4r 2 |x|=r
4r 2 |x|r
3 43 r 3 |x|r
r
f (0).
=
3

r2
1
f (0) + f (0).
f
(x)
ds(x)
=
4r 2 |x|=r
6

56

Note that part (a) was not used. We use exactly the same derivation as we did in S95 #4.

Partial Dierential Equations

Igor Yanovsky, 2005

247

Problem (F96, #4).


Consider smooth solutions of u = k2 u in dimension d = 2 with k > 0.
a) Show that u satises the following mean value property:
1
Mx (r) + Mx (r) k2 Mx (r) = 0,
r
in which Mx (r) is dened by
 2
1
Mx (r) =
u(x + r cos , y + r sin ) d
2 0
and the derivatives (denoted by  ) are in r with x xed.
b) For k = 1, this equation is the modied Bessel equation (of order 0)
1
f  + f  f = 0,
r
for which one solution (denoted as I0 ) is
 2
1
er sin d.
I0 (r) =
2 0
Find an expression for Mx (r) in terms of I0 .
Proof. a) Laplacian in polar coordinates written as:
1
1
u = urr + ur + 2 u .
r
r
Thus, the equation may be written as
1
1
urr + ur + 2 u = k2 u.
r
r
 2
1
u d,
Mx (r) =
2 0
 2
1

ur d,
Mx (r) =
2 0
 2
1
urr d.
Mx (r) =
2 0
Mx (r) +

1 
M (r) k2 Mx (r) =
r x
=


1
urr + ur k2 u d
r
0
 2
1  2
1
u d =
u 0 = 0. 

2
2r 0
2r 2

1
2

b) Note that w = er sin satises w = w, i.e.


1
1
w = wrr + wr + 2 w
r
r
1
1
2
r sin
= sin e
+ sin er sin + 2 (r sin er sin + r 2 cos2 er sin ) = er sin = w.
r
r
Thus,
 2
1
er sin d = ey I0 .
Mx (r) = ey
2 0

Partial Dierential Equations


57
57

Check with someone about the last result.

Igor Yanovsky, 2005

248

Partial Dierential Equations

17.8

Igor Yanovsky, 2005

249

Harmonic Extensions, Subharmonic Functions

Problem (S94, #8). Suppose that is a bounded region in R3 and that u = 1 on


. If u = 0 in the exterior region R3 / and u(x) 0 as |x| , prove the
following:
a) u > 0 in R3 /;
b) if (x) is a smooth function such that (x) = 1 for |x| > R and (x) = 0 near ,
then for |x| > R,

1
((u))(y)
u(x) =
dy.
4 R3 / |x y|
c) lim|x| |x|u(x) exists and is non-negative.
Proof. a) Let B r (0) denote the closed ball {x : |x| r}.
Given > 0, we can nd r large enough that B R1 (0) and maxxB R (0) |u(x)| < ,
1
since |u(x)| 0 as |x| .
Since u is harmonic in B R1 , it takes its maximum and minimum on the boundary.
Assume
min

xBR1 (0)

u(x) = a < 0

(where |a| < ).

We can nd an R2 such that maxxBR

(0) |u(x)|

< a2 ; hence u takes a minimum inside

B R2 (0) , which is impossible; hence u 0.


Now let V = {x : u(x) = 0} and let = minxV |x|. Since u cannot take a minimum
inside B R (0) (where R > ), it follows that u C and C = 0, but this contradicts
u = 1 on . Hence u > 0 in R3 .
b) For n = 3,
K(|x y|) =

1
1
1
.
|x y|2n =
(2 n)n
4 |x y|

Since (x) = 1 for |x| > R, then for x


/ BR , we have (u) = u = 0. Thus,

((u))(y)
1
dy

4 R3 / |x y|

((u))(y)
1
dy
=
4 BR / |x y|



1
1
  1
1
y (u) dy
u
dSy
y
=
4 BR /
|x y|
4 (BR/) n
|x y|




1
1
1
1

1
  1
u dy +
u dSy
u
dSy
=

4 BR /
|x y|
4 (BR/) n |x y|
4 (BR/) n
|x y|


1
1
u
dSy
= ??? = u(x)
u dSy
4R2 B
4R B n


 


0, as R

= u(x).
c) See the next problem.

0, as R

Partial Dierential Equations

Igor Yanovsky, 2005

250

Ralston Hw. a) Suppose that u is a smooth function on R3 and u = 0 for |x| > R.
1
If limx u(x) = 0, show that you can write u as a convolution of u with the 4|x|
and prove that limx |x|u(x) = 0 exists.
b) The conductor potential for R3 is the solution to the Dirichlet problem v =
0. The limit in part (a) is called the capacity of . Show that if 1 2 , then the
capacity of 2 is greater or equal the capacity of 1 .
Proof. a) If we dene

u(y)
1
dy,
v(x) =
4 R3 |x y|
then (u v) = 0 in all R3 , and, since v(x) 0 as |x| , we have lim|x| (u(x)
v(x)) = 0. Thus, u v must be bounded, and Liouvilles theorem implies that it is
identically zero. Since we now have

1
|x| u(y)
|x|u(x) =
dy,
4 R3 |x y|
and |x|/|x y| converges uniformly to 1 on {|y| R}, it follows that

1
u(y) dy.
lim |x|u(x) =
4 R3
|x|
b) Note that part (a) implies that the limit lim|x| |x|v(x) exists, because we can
apply (a) to u(x) = (x)v(x), where is smooth and vanishes on , but (x) = 1 for
|x| > R.
Let v1 be the conductor potential for 1 and v2 for 2 . Since vi as |x| and
vi = 1 on i , the max principle says that 1 > vi (x) > 0 for x R3 i . Consider
v2 v1 . Since 1 2 , this is dened in R3 2 , positive on 2 , and has limit 0 as
|x| . Thus, it must be positive in R3 2 . Thus, lim|x| |x|(v2 v1 ) 0.

Problem (F95, #4). 58 Let be a simply connected open domain in R2


and u = u(x, y) be subharmonic there, i.e. u 0 in . Prove that if
DR = {(x, y) : (x x0 )2 + (y y0 )2 R2 }
then
1
u(x0 , y0 )
2


0

u(x0 + R cos , y0 + R sin ) d.

Proof. Let

 2
1
u(x0 + R cos , y0 + R sin ) d,
2 0
w(r, ) = u(x0 + R cos , y0 + R sin ).

M (x0 , R) =

Dierentiate M (x0 , R) with respect to R:


 2
1
d
M (x0 , R) =
wr (R, )R d,
dr
2R 0
58

See McOwen, Sec.4.3, p.131, #1.

Partial Dierential Equations

Igor Yanovsky, 2005

251

59
59
See ChiuYens solutions and Sung Has solutions (in two places). Nicks solutions, as started above,
have a very simplistic approach.

Partial Dierential Equations

Igor Yanovsky, 2005

252

Ralston Hw (Maximum Principle).


Suppose that u C() satises the mean value property in the connected open set .
a) Show that u satises the maximum principle in , i.e.
either u is constant or u(x) < sup u for all x .
b) Show that, if v is a continuous function on a closed ball Br () and has the
mean value property in Br (), then u = v on Br () implies u = v in Br (). Does this
imply that u is harmonic in ?
Proof. a) If u(x) is not less than sup u for all x , then the set
K = {x : u(x) = sup u}

is nonempty. This set is closed because u is continuous. We will show it is also open.
This implies that K = because is connected. Thus u is constant on .
Let x0 K. Since is open, > 0, s.t. B (x0 ) = {x Rn : |x x0 | } . Let
sup u = M . By the mean value property, for 0 r


1
1
u(x
+
r)dS
,
and
0
=
(M u(x0 + r))dS .
M = u(x0 ) =
0

A(S n1 ) ||=1
A(S n1 ) ||=1
Sinse M u(x0 + r) is a continuous nonnegative function on , this implies M u(x0 +
r) = 0 for all S n1 . Thus u = 0 on B (x0 ).
b) Since u v has the mean value property in the open interior of Br (), by part
a) it satises the maximum principle. Since it is continuous on Br (), its supremum
over the interior of Br () is its maximum on Br (), and this maximum is assumed at a
point x0 in Br (). If x0 in the interior of Br (), then u v is constant ant the constant
must be zero, since this is the value of u v on the boundary. If x0 is on the boundary,
then u v must be nonpositive in the interior of Br ().
Applying the same argument to v u, one nds that it is either identically zero or
nonpositive in the interior of Br (). Thus, u v 0 on Br ().
Yes, it does follow that u harmonic in . Take v in the preceding to be the harmonic
function in the interior of Br () which agrees with u on the boundary. Since u = v on
Br (), u is harmonic in the interior of Br (). Since is open we can do this for every
. Thus u is harmonic in .

Partial Dierential Equations

Igor Yanovsky, 2005

253

Ralston Hw. Assume is a bounded open set in Rn and the Greens function, G(x, y),
for exists. Use the strong maximum principle, i.e. either u(x) < sup u for all x ,
or u is constant, to prove that G(x, y) < 0 for x, y , x = y.
Proof. G(x, y) = K(x, y) + (x, y). For each x , f (y) = (x, y) is continuous on ,
thus, bounded. So |(x, y)| Mx for all y . K(x y) as y x. Thus,
given Mx , there is > 0, such that K(x y) < Mx when |x y| = r and 0 < r .
So for 0 < r the Greens function with x xed satises, G(x, y) is harmonic on
Br (x), and G(x, y) 0 on the boundary of Br (x). Since we can choose r as
small as we wish, we get G(x, y) < 0 for y {x}.
Problem (W03, #6). Assume that u is a harmonic function in the half ball
D = {(x, y, z) : x2 + y 2 + z 2 < 1, z 0} which is continuously dierentiable, and satises u(x, y, 0) = 0. Show that u can be extended to be a harmonic function in the whole
ball. If you propose and explicit extension for u, explain why the extension is harmonic.
Proof. We can extend u to all of n-space by dening
u(x , xn ) = u(x , xn )
for xn < 0. Dene
1
(x) =
an


|y|=1

a2 |x|2
v(y)dSy
|x y|n

(x) is continuous on a closed ball B, harmonic in B.


Poisson kernel is symmetric in y at xn = 0. (x) = 0, (xn = 0).
is harmonic for x B, xn 0,with the same boundary values = u.
is harmonic u can be extended to a harmonic function on the interior of B.
Ralston Hw. Show that a bounded solution to the Dirichlet problem in a half
space is unique. (Note that one can show that a bounded solution exists for any
given bounded continuous Dirichlet data by using the Poisson kernel for the half space.)
Proof. We have to show that a function, u, which is harmonic in the half-space, continuous, equal to 0 when xn = 0, and bounded, must be identically 0. We can extend
u to all of n-space by dening
u(x , xn ) = u(x , xn )
for xn < 0. This extends u to a bounded harmonic function on all of n-space (by the
problem above). Liouvilles theorem says u must be constant, and since u(x , 0) = 0,
the constant is 0. So the original u must be identically 0.
Ralston Hw. Suppose u is harmonic on the ball minus the origin, B0 = {x R3 :
0 < |x| < a}. Show that u(x) can be extended to a harmonic function on the ball
B = {|x| < a} i lim|x|0 |x|u(x) = 0.
Proof. The condition lim|x|0 |x|u(x) = 0 is necessary, because harmonic functions are
continuous.
To prove the converse, let v be the function which is continuous on {|x| a/2},
harmonic on {|x| < a/2}, and equals u on {|x| = a/2}. One can construct v using the
Poisson kernel. Since v is continuous, it is bounded, and we can assume that |v| M .
Since lim|x|0 |x|u(x) = 0, given  > 0, we can choose , 0 < < a/2 such that
 < |x|u(x) <  when |x| < . Note that u, v 2/|x|, and v + 2/|x| are harmonic

Partial Dierential Equations

Igor Yanovsky, 2005

254

on {0 < |x| < a/2}. Choose b, 0 < b < min(, a/2), so that /b > M . Then on both
{|x| = a/2} and {|x| = b} we have v 2/|x| < u(x) < v + 2/|x|.
Thus, by
max principle these inequalities hold on {b |x| a/2}. Pick x with 0 < |x| a/2.
u(x) = v(x). v is the extension of u on {|x| < a/2}, and u is extended on {|x| < a}.

Partial Dierential Equations

18

Igor Yanovsky, 2005

Problems: Heat Equation

McOwen 5.2 #7(a). Consider

for x > 0, t > 0


ut = uxx
u(x, 0) = g(x)
for x > 0

u(0, t) = 0
for t > 0,
where g is continuous and bounded for x 0 and g(0) = 0.
Find a formula for the solution u(x, t).
Proof. Extend g to be an odd function on all of R:

g(x),
x0
g(x) =
g(x), x < 0.
Then, we need to solve

xx
u
t = u
u
(x, 0) = g(x)

for x R, t > 0
for x R.

The solution is given by: 60




(xy)2
1

u
(x, t) =
K(x, y, t)g(y) dy =
e 4t g(y) dy
4t
R


 0
2
(xy)
(xy)2
1
e 4t g(y) dy +
e 4t g(y) dy
=
4t 0




(xy)2
(x+y)2
1
4t
4t
e
g(y) dy
e
g(y) dy
=
4t 0
0

2

x +2xyy 2
x2 2xyy 2
1
4t
4t
e
e
g(y) dy
=
4t 0


xy

(x2 +y 2 )
xy
1
e 4t
e 2t e 2t g(y) dy.
=
4t 0


xy
(x2 +y 2 )
1
g(y) dy.
u(x, t) =
e 4t 2 sinh
2t
4t 0
Since sinh(0) = 0, we can verify that u(0, t) = 0.
60

In calculations, we use:

0

ey dy =


0

ey dy, and g(y) = g(y).

255

Partial Dierential Equations

Igor Yanovsky, 2005

256

McOwen 5.2 #7(b). Consider

for x > 0, t > 0

ut = uxx
u(x, 0) = g(x)
for x > 0

for t > 0,
ux (0, t) = 0
where g is continuous and bounded for x 0.
Find a formula for the solution u(x, t).
Proof. Extend g to be an even function

g(x),
x0
g(x) =
g(x), x < 0.
Then, we need to solve

xx
u
t = u
u
(x, 0) = g(x)

61

on all of R:

for x R, t > 0
for x R.

The solution is given by: 62




(xy)2
1
u
(x, t) =
K(x, y, t)g(y) dy =
e 4t g(y) dy
4t
R


 0
2
(xy)2
1
(xy)

4t
e 4t g(y) dy +
e
g(y) dy
=
4t 0




(xy)2
(x+y)2
1
4t
4t
e
g(y) dy +
e
g(y) dy
=
4t 0
0

2

x +2xyy 2
x2 2xyy 2
1
4t
4t
+e
e
g(y) dy
=
4t 0


xy

(x2 +y 2 )
xy
1
e 4t
e 2t + e 2t g(y) dy.
=
4t 0


xy
(x2 +y 2 )
1
g(y) dy.
u(x, t) =
e 4t 2 cosh
2t
4t 0
To check that the boundary condition holds, we perform the calculation:



xy 
1
d (x2 +y2 )
4t
e
g(y) dy
2 cosh
ux (x, t) =
2t
4t 0 dx



xy 

xy
(x2 +y 2 )
y
2x (x2 +y2 )
1

4t
4t
e
+e
g(y) dy,
2 cosh
2 sinh

=
4t
2t
2t
2t
4t 0

 
2
2
1
y
y4t
y4t
2 cosh 0 + e
2 sinh 0 g(y) dy = 0.
0e
ux (0, t) =
2t
4t 0

61

Even extensions are always continuous. Not true for odd extensions. g odd is continuous if g(0) =

0.
62

In calculations, we use:

0

ey dy =


0

ey dy, and g(y) = g(y).

Partial Dierential Equations

Igor Yanovsky, 2005

257

Problem (F90, #5).


The initial value problem for the heat equation on the whole real line is
t0

ft = fxx

f (t = 0, x) = f0 (x)
with f0 smooth and bounded.
a) Write down the Greens function G(x, y, t) for this initial value problem.
b) Write the solution f (x, t) as an integral involving G and f0 .
c) Show that the maximum values of |f (x, t)| and |fx(x, t)| are non-increasing
as t increases, i.e.
sup |f (x, t)| sup |f0 (x)|
x

sup |fx (x, t)| sup |f0x (x)|.


x

When are these inequalities actually equalities?


Proof. a) The fundamental solution
1 |xy|2
e 4t .
K(x, y, t) =
4t
The Greens function is:

63

n

(xy)2
2
1

4k(ts)
G(x, t; y, s) =
e
.
(2)n k(t s)

b) The solution to the one-dimensional heat equation is




|xy|2
1
K(x, y, t) f0(y) dy =
e 4t f0 (y) dy.
u(x, t) =
4t R
R
c) We have
sup |u(x, t)| =
x





 1


(xy)2
(xy)2 
1
4t
4t


f0 (y) dy

e
f
(y)
dy

e
0
 4t

4t R
R 2


y
1

e 4t f0 (x y) dy
4t R


y2
dy
1
y
e 4t dy
z = , dz =
sup |f0 (x)|
x
4t R
4t
4t

1
2
ez 4t dz
sup |f0 (x)|
x
4t R

1
2
ez dz = sup |f0 (x)|. 
sup |f0 (x)|
R
x
x
  
=

63
The Greens function for the heat equation on an innite domain; derived in R. Haberman using
the Fourier transform.

Partial Dierential Equations

Igor Yanovsky, 2005

258



1
2(x y) (xy)2
1
d  (xy)2
4t
4t
e
e

f0 (y) dy =

f0 (y) dy
4t
dy
4t R
4t
R


(xy)2
(xy)2
1
1 
4t

e
f0 (y)
+
e 4t f0y (y) dy,
=

4t
4t R



=0


2
1
1
2
(xy)
4t
sup |f0x(x)| e
sup |f0x(x)| ez 4t dz
dy =
sup |u(x, t)|
x
4t x
4t x
R
R
= sup |f0x(x)|. 
ux (x, t) =

These inequalities are equalities when f0 (x) and f0x (x) are constants, respectively.

Partial Dierential Equations

Igor Yanovsky, 2005

259

Problem (S01, #5). a) Show that the solution of the heat equation
< x <

ut = uxx ,

with square-integrable initial data u(x, 0) = f (x), decays in time, and there is a constant
independent of f and t such that for all t > 0
1


3
2
|f (x)|2 dx .
max |ux(x, t)| t 4
x

b) Consider the solution of the transport equation t + ux = 0 with square-integrable


initial data (x, 0) = 0 (x) and the velocity u from part (a). Show that (x, t) remains
square-integrable for all nite time


1
|(x, t)|2 dx eCt 4
|0(x)|2 dx,
R

where C does not depend on 0 .


Proof. a) The solution to the one-dimensional homogeneous heat equation is

(xy)2
1
e 4t f (y) dy.
u(x, t) =
4t R
Take the derivative with respect to x, we get 64


(xy)2
1
2(x y) (xy)2
1
4t
e
ux (x, t) =

f (y) dy = 3
(x y)e 4t f (y) dy.
4t
4t R
4t 2 R
|ux (x, t)|

=
=

1
3

 

2


(xy)
4t
f
(y)

y)e
 dy
(x

(Cauchy-Schwarz)

4t 2 R

1
(xy)2
1

2
(x y)2 e 2t dy ||f ||L2(R)
3
4t 2 R

 12
3
1

2
z2 
2e
(2t)

z
dz ||f ||L2(R)
3
4t 2 R

3
1
(2t) 4

2
2 z2
z
e
dz
||f ||L2(R)
3
4t 2  R 


dy
xy
z = , dz =
2t
2t

M <
34

= Ct

1
2

M ||f ||L2(R) = t 4 ||f ||L2(R) .

b) Note:





1
2

 1
(xy)2
1

2
(xy)


4t
e
f (y) dy 
e 2t dy ||f ||L2(R)
max |u| = max 
x
x
4t R
4t R

1



dy
1
xy
 ez2 2t dz 2 ||f ||L2(R)

z = , dz =
4t R
2t
2t
1

1

1
2
(2t) 4
2
ez dz ||f ||L2(R) = Ct 4 ||f ||L2(R).
=
1 1
2 2 t 2  R  
=

65
64
65

Cauchy-Schwarz: |(u, v)| ||u||||v||


See Yanas and Alans solutions.

in any norm, for example


1 
1
|uv|dx ( u2 dx) 2 ( v 2 dx) 2

Partial Dierential Equations

Igor Yanovsky, 2005

260

Problem (F04, #2).


Let u(x, t) be a bounded solution to the Cauchy problem for the heat equation

t > 0, x R, a > 0,
ut = a2 uxx ,
u(x, 0) = (x).
Here (x) C(R) satises
lim (x) = b,

lim (x) = c.

x+

Compute the limit of u(x, t) as t +, x R. Justify your argument carefully.


Proof. For a = 1, the solution to the one-dimensional homogeneous heat equation is

(xy)2
1

e 4t (y) dy.
u(x, t) =
4t R
We want to transform the equation to vt = vxx . Make a change of variables: x = ay.
u(x, t) = u(x(y), t) = u(ay, t) = v(y, t). Then,
vy = ux xy = aux ,
vyy

= auxx xy = a2 uxx ,

v(y, 0) = u(ay, 0) = (ay).


Thus, the new problem is:

t > 0, y R,
vt = vyy ,
v(y, 0) = (ay).
v(y, t) =

1
4t

(yz)2
4t

(az) dz.

Since is continuous, and limx+ (x) = b, limx (x) = c, we have


|(x)| < M,
|v(y, t)|
=

x R.
Thus,


z2
dz
z
M

e 4t dz
s = , ds =
4t R
4t
4t



M
2
2
M

es 4t ds =
es ds = M.
R
4t R
  


Integral in  converges uniformly lim = lim. For = (a):




(yz)2
z2
1
1
4t
e
(z) dz =
e 4t (y z) dz
v(y, t) =
4t
4t


2
1
es (y s 4t) 4t ds
=
4t


1
2
es (y s 4t) ds.
=

Partial Dierential Equations


lim v(y, t) =

t+

=
=

261

 0

1
2

e
lim (y s 4t) ds +
es lim (y s 4t) ds
t+
t+

0


 0
1

1
1
1
2
2

+ b
es c ds +
es b ds = c
0

2
2
c+b
.
2
1

Igor Yanovsky, 2005

s2

Partial Dierential Equations

Igor Yanovsky, 2005

Problem. Consider
ut = kuxx + Q,

0<x<1

u(0, t) = 0,
u(1, t) = 1.
What is the steady state temperature?
Proof. Set ut = 0, and integrate with respect to x twice:
kuxx + Q = 0,
Q
uxx = ,
k
Q
ux = x + a,
k
Q x2
u=
+ ax + b.
k 2
Boundary conditions give

Q
Q
x.
u(x) = x2 + 1 +
2k
2k

262

Partial Dierential Equations

18.1

Igor Yanovsky, 2005

263

Heat Equation with Lower Order Terms

McOwen 5.2 #11. Find a formula for the solution of



ut = u cu
in Rn (0, )
u(x, 0) = g(x)
on Rn .

(18.1)

Show that such solutions, with initial data g L2 (Rn ), are unique, even when c is
negative.
Proof. McOwen. Consider v(x, t) = ect u(x, t).

in Rn (0, )
vt = v
v(x, 0) = g(x)
on Rn .

The transformed problem is


(18.2)

Since g is continuous and bounded in Rn , we have




|xy|2
1
v(x, t) =
K(x, y, t) g(y) dy =
e 4t g(y) dy,
n
(4t) 2 Rn
Rn

|xy|2
1
e 4t ct g(y) dy.
u(x, t) = ect v(x, t) =
n
(4t) 2 Rn
u(x, t) is a bounded solution since v(x, t) is.
To prove uniqueness, assume there is another solution v  of (18.2). w = v v  satises

in Rn (0, )
wt = w
(18.3)
w(x, 0) = 0
on Rn .
Since bounded solutions of (18.3) are unique, and since w is a nontrivial solution, w is
unbounded. Thus, v  is unbounded, and therefore, the bounded solution v is unique.

Partial Dierential Equations


18.1.1

Igor Yanovsky, 2005

264

Heat Equation Energy Estimates

Problem (F94, #3). Let u(x, y, t) be a twice continuously dierential solution of


ut = u u3

in R2 , t 0

u(x, y, 0) = 0

in

u(x, y, t) = 0

in , t 0.

Prove that u(x, y, t) 0 in [0, T ].


Proof. Multiply the equation by u and integrate:
uut = uu u4 ,


uut dx =
uu dx
u4 dx,





u
1 d
2
2
ds |u| dx
u dx =
u
u4 dx,
2 dt
n




=0


1 d
2
2
||u||2 = |u| dx
u4 dx 0.
2 dt

Thus,
||u(x, y, t)||2 ||u(x, y, 0)||2 = 0.
Hence, ||u(x, y, t)||2 = 0, and u 0.

Partial Dierential Equations

Igor Yanovsky, 2005

265

Problem (F98, #5). Consider the heat equation


ut u = 0
in a two dimensional region . Dene the mass M as

u(x, t) dx.
M (t) =

a) For a xed domain , show M is a constant in time if the boundary conditions are
u/n = 0.
b) Suppose that = (t) is evolving in time, with a boundary that moves at velocity
v, which may vary along the boundary. Find a modied boundary condition (in terms
of local quantities only) for u, so that M is constant.
Hint: You may use the fact that



d
f (x, t) dx =
ft (x, t) dx +
n v f (x, t) dl,
dt (t)
(t)
(t)
in which n is a unit normal vector to the boundary .
Proof. a) We have

ut u = 0,
on
u
on .
n = 0,
d
M (t) = 0. We have 66
We want to show that dt




d
d
u
M (t) =
ds = 0. 
u(x, t) dx =
ut dx =
u dx =
dt
dt

b) We need
0 =

d
dt M (t)

= 0.



u(x, t) dx =
ut dx +
n v u ds
(t)
(t)
(t)



u
ds +
u dx +
n v u ds =
n v u ds
=
(t)
(t)
(t) n
(t)



u n ds +
n v u ds =
n (u + vu) ds.
=

d
M (t) =
dt

d
dt


(t)

(t)

Thus, we need:
n (u + vu) ds = 0,

66

on .

The last equality below is obtained from the Greens formula:




u
ds.
u dx =

(t)

Partial Dierential Equations

Igor Yanovsky, 2005

266

Problem (S95, #3). Write down an explicit formula for a function u(x, t) solving

in Rn (0, )
ut + b u + cu = u
(18.4)
u(x, 0) = f (x)
on Rn .
where b Rn and c R are constants.
Hint: First transform this to the heat equation by a linear change of the dependent
and independent variables. Then solve the heat equation using the fundamental solution.
Proof. Consider
u(x, t) = ex+tv(x, t).
ut = ex+t v + ex+tvt = (vt + v)ex+t,
u = ex+t v + ex+tv = (v + v)ex+t,


(u) = (v + v)ex+t = ( v + v)ex+t + (||2 v + v)ex+t

= v + 2 v + ||2 v)ex+t.
Plugging this into (18.4), we obtain
vt + v + b (v + v) + cv = v + 2 v + ||2 v,




vt + b 2 v + + b + c ||2 v = v.
In order to get homogeneous heat equation, we set
b
= ,
2
which gives

vt = v

|b|2
c,
4

in Rn (0, )

v(x, 0) = e 2 xf (x)

on Rn .

The above PDE has the following solution:



|xy|2
b
1
e 4t e 2 y f (y) dy.
v(x, t) =
n
(4t) 2 Rn
Thus,
u(x, t) = e

2
b
x( |b|4
2

+c)t

2
b
1
x( |b|4 +c)t
v(x, t) =
n e2
(4t) 2


Rn

|xy|2
4t

e 2 y f (y) dy.

Partial Dierential Equations

Igor Yanovsky, 2005

267

Problem (F01, #7). Consider the parabolic problem


ut = uxx + c(x)u

(18.5)

for < x < , in which


c(x) = 0

for

|x| > 1,

c(x) = 1

for

|x| < 1.

|u|

Find solutions of the form u(x, t) = etv(x) in which


Hint: Look for v to have the form
v(x) = aek|x|
v(x) = b cos lx

dx < .

for |x| > 1,


for |x| < 1,

for some a, b, k, l.
Proof. Plug u(x, t) = etv(x) into (18.5) to get:
etv(x) = etv  (x) + cetv(x),
v(x) = v (x) + cv(x),
v  (x) v(x) + cv(x) = 0.
For |x| > 1, c = 0. We look for solutions of the form v(x) = aek|x| .
v  (x) v(x) = 0,
ak2 ek|x| aek|x| = 0,
k2 = 0,

Thus, v(x) = c1 e

k2 = ,

k = .
x

+ c2 e

u(x, t) = aet e

x .

Since we want

|u|2 dx < :

For |x| < 1, c = 1. We look for solutions of the form v(x) = b cos lx.
v  (x) v(x) + v(x) = 0,
bl 2 cos lx + (1 )b cos lx = 0,
l 2 + (1 ) = 0,
l 2 = 1 ,

l = 1 .
Thus, (since cos(x) = cos x)

u(x, t) = bet cos (1 )x.
We want v(x) to be continuous on R, and at x = 1, in particular. Thus,


ae = b cos (1 ),


a = be cos (1 ).
Also, v(x) is symmetric:



2
2
|u| dx = 2
|u| dx = 2

|u| dx +


1

|u| dx < .

Partial Dierential Equations

Igor Yanovsky, 2005

268

Problem (F03, #3). The function


X2

h(X, T ) = (4T ) 2 e 4T

satises (you do not have to show this)


hT = hXX .
Using this result, verify that for any smooth function U

1 3
u(x, t) = e 3 t xt
U () h(x t2 , t) d

satises
ut + xu = uxx .
Given that U (x) is bounded and continuous everywhere on x , establish
that

U () h(x , t) d = U (x)
lim
t0

and show that u(x, t) U (x) as t 0. (You may use the fact that
1
2 .)


0

e d =

Proof. We change the notation: h K, U g, y. We have


X2
1
e 4T
K(X, T ) =
4T
We want to verify that

1 3
t xt
3
K(x y t2 , t) g(y) dy.
u(x, t) = e

satises
ut + xu = uxx .
We have
ut =
=
xu =
ux =
=
uxx =
=


d  1 t3 xt
e3
K(x y t2 , t) g(y) dy
dt




1 3
1 3
(t2 x) e 3 t xt K + e 3 t xt KX (2t) + KT g(y) dy,


1 3
x e 3 t xt K(x y t2 , t) g(y) dy,


d  1 t3 xt
e3
K(x y t2 , t) g(y) dy
dx



1 3
1 3
t e 3 t xt K + e 3 t xt KX g(y) dy,


1 3
1 3
d
t e 3 t xt K + e 3 t xt KX g(y) dy
dx



1 3
1 3
1 3
1 3
t2 e 3 t xt K t e 3 t xt KX t e 3 t xt KX + e 3 t xt KXX g(y) dy.

Partial Dierential Equations

Igor Yanovsky, 2005

269

Plugging these into , most of the terms cancel out. The remaining two terms cancel
because KT = KXX .
Given that g(x) is bounded and continuous on x , we establish that 67

lim
K(x y, t) g(y) dy = g(x).
Fix

t0
x0 R n ,

> 0. Choose > 0 such that

if |y x0 | < , y Rn .

Then if |x x0 | < 2 , we have: ( R K(x, t) dx = 1)








K(x

y,
t)
g(y)
dy

g(x
)

K(x

y,
t)
[g(y)

g(x
)]
dy



0 
0
R
R






K(x y, t) g(y) g(x0 ) dy +
K(x y, t) g(y) g(x0 ) dy 

B (x )
RB (x0 )


 0

|g(y) g(x0 )| <

K(xy,t) dy =

Furthermore, if |x x0 |

and |y x0 | , then

|y x| + |y x0 |.
2
2
1
Thus, |y x| 2 |y x0 |. Consequently,

K(x y, t) dy
 = + 2||g||L
RB (x0 )

|xy|2
C
e 4t dy
+
t RB (x0 )

|yx0 |2
C
e 16t dy
+
t RB (x0 )

r2
C
e 16t r dr + 0
as t 0+ .
= +
t
|y x0 | |y x| +

Hence, if |x x0 | <
67

and t > 0 is small enough, |u(x, t) g(x0)| < 2.

Evans, p. 47, Theorem 1 (c).

Partial Dierential Equations

Igor Yanovsky, 2005

270

Problem (S93, #4). The temperature T (x, t) in a stationary medium, x 0, is


governed by the heat conduction equation
2T
T
.
=
t
x2

(18.6)

Making the change of variable (x, t) (u, t), where u = x/2 t, show that
4t

2T
T
T
=
.
+ 2u
t
u2
u

(18.7)

Solutions of (18.7) that depend on u alone are called similarity solutions.

68

Proof. We change notation: the change of variables is (x, t) (u, ), where t = .


After the change of variables, we have T = T (u(x, t), (t)).
x
u=
2 t
=t
T
t
T
x
2T
x2

=
=
=

ut =

t = 1,

x
4t

3
2

1
ux = ,
2 t
x = 0.

uxx = 0,

T u T
+
,
u t

T u
,
u x

2 T u u T 2 u

T u

T
2 T
u 2
=
=
+
=
.
x x
x u x
u2 x x
u 
x2
u2 x
=0

Thus, (18.6) gives:


T u
+
u t
x
T

3 +
u
4t 2

T
4t

T
4t

68

=
=
=
=
=

2 T
u 2
,
u2 x
2 T
1 2
,
u2 2 t
1 2T
x T
,
+ 3
2
4t u
4t 2 u
2T
x T
,
+
u2
t u
2T
T
. 
+ 2u
2
u
u

This is only the part of the qual problem.

Partial Dierential Equations

19

Igor Yanovsky, 2005

271

Contraction Mapping and Uniqueness - Wave

Recall that the solution to



utt c2 uxx = f (x, t),
u(x, 0) = g(x), ut(x, 0) = h(x),

(19.1)

is given by adding together dAlemberts formula and Duhamels principle:



 x+ct
 t   x+c(ts)
1
1
1
u(x, t) = (g(x + ct) + g(x ct)) +
h() d +
f (, s) d ds.
2
2c xct
2c 0
xc(ts)

Problem (W02, #8). a) Find an explicit solution of the following Cauchy problem
 2
2
u
xu2 = f (t, x),
t2
(19.2)
u(0, x) = 0, u
x (0, x) = 0.
b) Use part (a) to prove the uniqueness of the solution of the Cauchy problem
 2
2
u
xu2 + q(t, x)u = 0,
t2
(19.3)
u(0, x) = 0, u
x (0, x) = 0.
Here f (t, x) and q(t, x) are continuous functions.
Proof. a) It was probably meant to give the ut initially. We rewrite (19.2) as

utt uxx = f (x, t),
u(x, 0) = 0, ut (x, 0) = 0.

(19.4)

Duhamels principle, with c = 1, gives the solution to (19.4):




 t   x+c(ts)
   x+(ts)
1 t
1
f (, s) d ds =
f (, s) d ds.
u(x, t) =
2c 0
2 0
xc(ts)
x(ts)

b) We use the Contraction Mapping Principle to prove uniqueness.


Dene the operator
 
1 t x+(ts)
q(, s) u(, s) d ds.
T (u) =
2 0 x(ts)
on the Banach space C 2,2 , || || .
We will show |T un T un+1 | < ||un un+1 || where < 1. Then {un }
n=1 :
un+1 = T (un ) converges to a unique xed point which is the unique solution of PDE.
  t  x+(ts)

1




q(, s) un (, s) un+1 (, s) d ds
|T un T un+1 | = 
2 0 x(ts)

1 t
||q||||un un+1 || 2(t s) ds

2 0
for small t.
t2 ||q||||un un+1 || ||un un+1 || ,
Thus, T is a contraction a unique xed point.
Since T u = u, u is the solution to the PDE.

Partial Dierential Equations

Igor Yanovsky, 2005

272

Problem (F00, #3). Consider the Goursat problem:

Find the solution of the equation


2u 2u

+ a(x, t)u = 0
t2
x2
in the square D, satisfying the boundary conditions
u|1 = ,

u|2 = ,

where 1 , 2 are two adjacent sides D. Here a(x, t), and are continuous functions.
Prove the uniqueness of the solution of this Goursat problem.
Proof. The change of variable = x + t, = x t
transforms the equation to
(, )
u = 0.
u
+ a
We integrate the equation:
 
 
u
(u, v) du dv =
a
(, ) u
du dv,
0 0
 0 0


(0, v) dv =
a
(, ) u
du dv,
u
(, v) u
0
0
0
 
a(, ) u
du dv.
u
(, ) = u
(, 0) + u
(0, ) u(0, 0)
0

We change the notation. In the new notation:


 x y
a(u, v)f (u, v) du dv,
f (x, y) = (x, y)
0

= + Kf,

= + K( + Kf ),

= +

K n ,

n=1

f = Kf
f = 0,
max |f | max |a| max |f |.

0<x<

For small enough , the operator K is a contraction. Thus, there exists a unique xed
point of K, and f = Kf , where f is the unique solution.

Partial Dierential Equations

20

Igor Yanovsky, 2005

273

Contraction Mapping and Uniqueness - Heat

The solution of the initial value problem



for t > 0, x Rn
ut = u + f (x, t)
u(x, 0) = g(x)
for x Rn .

(20.1)

is given by

u(x, t) =
where

K(x,
t) =

Rn

y, t) g(y) dy +
K(x

n e
(4t) 2

|x|2
4t

 t
0

Rn

y, t s) f (y, s) dy ds
K(x

for t > 0,
for t 0.

Problem (F00, #2). Consider the Cauchy problem


ut u + u2 (x, t) = f (x, t),

x RN , 0 < t < T

u(x, 0) = 0.
Prove the uniqueness of the classical bounded solution assuming that T is small
enough.
Proof. Let {un } be a sequence of approximations to the solution, such that
 t


2
=
K(x

y,
t

s)
f
(y,
s)

u
(y,
s)
dy ds.
S(un ) = un+1
n

use Duhamel s principle

Rn



We will show that S has a xed point |S(un ) S(un+1 )| |un un+1 |, < 1
{un } converges to a uniques solution for small enough T .
Since un , un+1 C 2 (Rn ) C 1 (t) |un+1 + un | M .
 t



K(x y, t s) u2 u2  dy ds
|S(un ) S(un+1 )|
n+1
n
0 Rn
 t




K(x y, t s) un+1 un  un+1 + un  dy ds
=
0 Rn
 t



K(x y, t s) un+1 un  dy ds
M
0
Rn
 t


un+1 (x, s) un (x, s) ds
M M1
0

M M1 T ||un+1 un || < ||un+1 un ||


2

for small T.
1

Thus, S is a contraction a unique xed point u C (R ) C (t) such that


u = limn un . u is implicitly dened as
 t


K(x y, t s) f (y, s) u2 (y, s) dy ds.
u(x, t) =
0

Rn

Partial Dierential Equations

Igor Yanovsky, 2005

274


Problem (S97, #3). a) Let Q(x) 0 such that x= Q(x) dx = 1,
and dene Q = 1 Q( x ). Show that (here denotes convolution)
||Q(x) w(x)||L ||w(x)||L .
In particular, let Qt (x) denote the heat kernel (at time t), then
||Qt(x) w1 (x) Qt (x) w2 (x)||L ||w1 (x) w2 (x)||L .
b) Consider the parabolic equation ut = uxx + u2 subject to initial conditions
u(x, 0) = f (x). Show that the solution of this equation satises
 t
Qts (x) u2 (x, s) ds.
(20.2)
u(x, t) = Qt (x) f (x) +
0

c) Fix t > 0. Let {un (x, t)}, n = 1, 2, . . . the xed point iterations for the solution of
(20.2)
 t
Qts (x) u2n (x, s) ds.
(20.3)
un+1 (x, t) = Qt (x) f (x) +
0

Let Kn (t) = sup0mn ||um (x, t)||L . Using (a) and (b) show that
 t
||un (x, s) un1 (x, s)||L ds.
||un+1 (x, t) un (x, t)||L 2 sup Kn ( )
0 t

Conclude that the xed point iterations in (20.3) converge if t is suciently small.
Proof. a) We have
||Q(x) w(x)||L



= 







Q (x y)w(y) dy
Q (x y)w(y) dy 





1
x y


Q
dy
Q (x y) dy = ||w||
||w||




y
dy
1
y
Q
dy
z = , dz =
= ||w||






Q(z) dz = ||w(x)||. 
= ||w||

Partial Dierential Equations

Igor Yanovsky, 2005

275

x
1 e 4t
4t

, the heat kernel. We have 69


 






||Qt(x) w1 (x) Qt (x) w2 (x)||L = 
Qt (x y)w1 (y) dy
Qt (x y)w2 (y) dy 

 



(xy)2
1  (xy)2

4t
4t
=
e
w1 (y) dy
e
w2 (y) dy 


4t


2

(xy)
1
4t 

e
w1 (y) w2 (y) dy
4t



(xy)2
1




e 4t dy

w1 (y) w2 (y)
4t



1
2
xy
dy
ez 4t dz
z = , dz =
= w1 (y) w2 (y)
4t
4t
4t


 1
2




=
w1 (y) w2 (y)
ez dz




Qt (x) =



= w1 (y) w2 (y) . 
69

Note:

Qt (x) dx =

1
4t

(xy)2
4t

dy =

1
4t

ez

1
4t dz =

ez dz = 1.

Partial Dierential Equations

Igor Yanovsky, 2005

276

b) Consider

ut = uxx + u2 ,
u(x, 0) = f (x).
We will show that the solution of this equation satises
 t
Qts (x) u2 (x, s) ds.
u(x, t) = Qt (x) f (x) +
0

 t
Qts (x) u2 (x, s) ds =
Qts (x y) u2 (y, s) dy ds
0
0
R
 t


Qts (x y) us (y, s) uyy (y, s) dy ds
=
0
R
 t


d
d
=
Qts (x y)u(y, s)
Qts (x y) u(y, s) Qts (x y)uyy (y, s) dy ds
ds
ds

0 R

Q0 (x y)u(y, t) dy
Qt(x y)u(y, 0) dy
=
R
R
 t

d2
d
Qts (x y) u(y, s) + 2 Qts (x y)u(y, s) dy ds

dy
0
R ds






= u(x, t)

= 0, since Qt satisf ies heat equation

Qt(x y)f (y) dy

Note:

lim Q(x, t) = 0 (x) = (x).



lim R Q(x y, t)v(y) dy = v(0).

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

t0+

t0+

Note that we used: D (f g) = (Df ) g = f (D g).


c) Let

un+1 (x, t) = Qt (x) f (x) +


||un+1 (x, t) un (x, t)||L

(a)

Also,

Qts (x) u2n (x, s) ds.

  t


 2
 
2

Qts (x) un (x, s) un1 (x, s) ds

0

 t


 2

Qts (x) u (x, s) u2 (x, s)  ds
0



t 

u2 (x, s) u2

n1




n1 (x, s) ds

 

un (x, s) un1 (x, s) un (x, s) + un1 (x, s) ds

0
 t




un (x, s) un1 (x, s) ds
sup un (x, s) + un1 (x, s)

t 

0 t

2 sup Kn ( )
0 t

t
0

||un (x, s) un1 (x, s)||L ds. 

||un+1 (x, t) un (x, t)||L 2t sup Kn ( ) ||un (x, s) un1 (x, s)||L .
0 t

Partial Dierential Equations

Igor Yanovsky, 2005

For t small enough, 2t sup0 t Kn ( ) < 1. Thus, T dened as


 t
T u = Qt (x) f (x) +
Qts (x) u2 (x, s) ds
0

is a contraction, and has a unique xed point u = T u.

277

Partial Dierential Equations

Igor Yanovsky, 2005

278

Problem (S99, #3). Consider the system of equations


ut = uxx + f (u, v)
vt = 2vxx + g(u, v)
to be solved for t > 0, < x < , and smooth initial data with compact support:
u(x, 0) = u0 (x),

v(x, 0) = v0 (x).

If f and g are uniformly Lipschitz continuous, give a proof of existence and uniqueness of the solution to this problem in the space of bounded continuous functions with
||u(, t)|| = supx |u(x, t)|.
Proof. The space of continuous bounded functions forms a complete metric space so
the contraction mapping
principle
applies.


First, let v(x, t) = w x2 , t , then
ut = uxx + f (u, w)
wt = wxx + g(u, w).
These initial value problems have the following solutions (K is the heat kernel):
 t


y, t s) f (u, w) dy ds,
K(x y, t) u0 (y) dy +
K(x
u(x, t) =
n
n
0
R
R
 t

y, t) w0(y) dy +
y, t s) g(u, w) dy ds.
K(x
K(x
w(x, t) =
Rn

Rn

By the Lipshitz conditions,


|f (u, w)| M1 ||u||,
|g(u, w)| M2 ||w||.
Now we can show the mappings, as dened below, are contractions:

 t

y, t s) f (u, w) dy ds,
K(x y, t) u0 (y) dy +
K(x
T1 u =
n
n
0
R
R

 t
y, t) w0(y) dy +
y, t s) g(u, w) dy ds.
K(x
K(x
T2 w =
Rn

Rn

 t




K(x
y, t s) f (un , w) f (un+1 , w) dy ds
0
Rn
 t



K(x
y, t s) un un+1  dy ds
M1
0
Rn

 t


y, t s)dy ds


sup un un+1
K(x
M1
0 x
Rn
 t




sup un un+1  ds M1 t sup un un+1 
M1
x
0 x

for small t.
< sup un un+1 

|T1 (un ) T1 (un+1 )|


y, t s) dy = 1.
We used the Lipshitz condition and R K(x
Thus, for small t, T1 is a contraction, and has a unique xed point. Thus, the solution
is dened as u = T1 u.
Similarly, T2 is a contraction and has a unique xed point. The solution is dened as
w = T2 w.

Partial Dierential Equations

21
21.1

Igor Yanovsky, 2005

279

Problems: Maximum Principle - Laplace and Heat


Heat Equation - Maximum Principle and Uniqueness

Let us introduce the cylinder U = UT = (0, T ). We know that harmonic (and


subharmonic) functions achieve their maximum on the boundary of the domain. For
the heat equation, the result is improved in that the maximum is achieved on a certain
part of the boundary, parabolic boundary:
= {(x, t) U : x or t = 0}.
Let us also denote by C 2;1 (U ) functions satisfying ut , uxi xj C(U ).
Weak Maximum Principle. Let u C 2;1 (U ) C(U ) satisfy u ut in U .
Then u achieves its maximum on the parabolic boundary of U:
max u(x, t) = max u(x, t).

(21.1)

Proof. First, assume u > ut in U . For 0 < < T consider


U = (0, ),

= {(x, t) U : x or t = 0}.

If the maximum of u on U occurs at x and t = , then ut (x, ) 0 and


u(x, ) 0, violating our assumption; similarly, u cannot attain an interior maximum
on U . Hence (21.1) holds for U : maxU u = max u. But max u max u
and by continuity of u, maxU u = lim T maxU u. This establishes (21.1).
Second, we consider the general case of u ut in U . Let u = v + t for > 0.
Notice that v u on U and v vt > 0 in U . Thus we may apply (21.1) to v:
max u = max(v + t) max v + T = max v + T max u + T.
U

Letting 0 establishes (21.1) for u.

Partial Dierential Equations

Igor Yanovsky, 2005

280

Problem (S98, #7). Prove that any smooth solution, u(x, y, t) in the unit box
= {(x, y) | 1 x, y 1}, of the following equation
ut = uux + uuy + u,

t 0, (x, y)
(x, y)

u(x, y, 0) = f (x, y),


satises the weak maximum principle,

max u(x, y, t) max{ max u(1, 1, t), max f (x, y)}.


0tT

[0,T ]

(x,y)

Proof. Suppose u satises given equation. Let u = v + t for > 0.

Then,

vt + = vvx + vvy + t(vx + vy ) + v.


Suppose v has a maximum at (x0 , y0 , t0 ) (0, T ). Then
vx = vy = vt = 0

= v

v > 0

v has a minimum at (x0 , y0 , t0 ), a contradiction.


Thus, the maximum of v is on the boundary of (0, T ).
Suppose v has a maximum at (x0 , y0 , T ), (x0 , y0 ) . Then
vx = vy = 0, vt 0

v

v > 0

v has a minimum at (x0 , y0 , T ), a contradiction. Thus,


max v max{ max v(1, 1, t), max f (x, y)}.
0tT

[0,T ]

(x,y)

Now
max u =

[0,T ]

max (v + t) max v + T

[0,T ]

[0,T ]

max{ max v(1, 1, t), max f (x, y)} + T


0tT

(x,y)

max{ max u(1, 1, t), max f (x, y)} + T.


0tT

Letting 0 establishes the result.

(x,y)

Partial Dierential Equations

21.2

Igor Yanovsky, 2005

281

Laplace Equation - Maximum Principle

Problem (S91, #6). Suppose that u satises


Lu = auxx + buyy + cux + duy eu = 0
with a > 0, b > 0, e > 0, for (x, y) , with a bounded open set in R2 .
a) Show that u cannot have a positive maximum or a negative minimum in the interior of .
b) Use this to show that the only function u satisfying Lu = 0 in , u = 0 on
and u continuous on is u = 0.
Proof. a) For an interior (local) maximum or minimum at an interior point (x, y), we
have
ux = 0,

uy = 0.

Suppose u has a positive maximum in the interior of . Then


u > 0,

uxx 0,

uyy 0.

With these values, we have


= 0,
auxx + buyy + cux + duy eu
      
0

=0

=0

<0

which leads to contradiction. Thus, u can not have a positive maximum in .


Suppose u has a negative minimum in the interior of . Then
u < 0,

uxx 0,

uyy 0.

With these values, we have


= 0,
auxx + buyy + cux + duy eu
      
0

=0

=0

>0

which leads to contradiction. Thus, u can not have a negative minimum in .


b) Since u can not have positive maximum in the interior of , then max u = 0 on .
Since u can not have negative minimum in the interior of , then min u = 0 on .
Since u is continuous, u 0 on .

Partial Dierential Equations

22

Igor Yanovsky, 2005

282

Problems: Separation of Variables - Laplace Equation

Problem 1: The 2D LAPLACE Equation on a Square.


Let = (0, ) (0, ), and use separation of variables to solve the boundary value
problem

0 < x, y <

uxx + uyy = 0
u(0, y) = 0 = u(, y)
0y

u(x, 0) = 0,
u(x, ) = g(x)
0 x ,
where g is a continuous function satisfying g(0) = 0 = g().

Proof. Assume u(x, y) = X(x)Y (y), then substitution in the PDE gives X Y + XY  =
0.
Y 
X 
=
= .
X
Y
From X  + X = 0, we get Xn (x) = an cos nx + bn sin nx.
give

u(0, y) = X(0)Y (y) = 0
X(0) = 0 = X().
u(, y) = X()Y (y) = 0

Boundary conditions

Thus, Xn (0) = an = 0, and


Xn (x) = bn sin nx, n = 1, 2, . . .. 
n2 bn sin nx + bn sin nx = 0,
n = n2 , n = 1, 2, . . .. 
With these values of n we solve Y  n2 Y = 0 to nd Yn (y) = cn cosh ny +
dn sinh ny.
Boundary conditions give
u(x, 0) = X(x)Y (0) = 0 Y (0) = 0 = cn .
Yn (x) = dn sinh ny. 
By superposition, we write
u(x, y) =

a
n sin nx sinh ny,

n=1

which saties the equation and the three homogeneous boundary conditions. The
boundary condition at y = gives
u(x, ) = g(x) =


g(x) sin mx dx =
0

an sin nx sinh n,

n=1


n=1


an sinh n

sin nx sin mx dx =
0

am sinh m.
2

Partial Dierential Equations


2
a
n sinh n =

g(x) sin nx dx.


0

Igor Yanovsky, 2005

283

Partial Dierential Equations

Igor Yanovsky, 2005

284

Problem 2: The 2D LAPLACE Equation on a Square. Let = (0, ) (0, ),


and use separation of variables to solve the mixed boundary value problem

in
u = 0
0<y<
ux (0, y) = 0 = ux (, y)

u(x, 0) = 0,
u(x, ) = g(x)
0 < x < .
Proof. Assume u(x, y) = X(x)Y (y), then substitution in the PDE gives X Y + XY  =
0.
X 
Y 
=
= .
X
Y
Consider X  + X = 0.
If = 0, X0 (x) = a0 x + b0 .
If > 0, Xn (x) = an cos nx + bn sin nx.
Boundary conditions give

ux (0, y) = X  (0)Y (y) = 0
X  (0) = 0 = X  ().
ux (, y) = X  ()Y (y) = 0
Thus, X0 (0) = a0 = 0, and Xn (0) = nbn = 0.
X0 (x) = b0 , Xn (x) = an cos nx, n = 1, 2, . . ..

n an cos nx + an cos nx = 0,
n = n2 , n = 0, 1, 2, . . .. 
With these values of n we solve Y  n2 Y = 0.
If n = 0, Y0 (y) = c0 y + d0 .
If n = 0, Yn (y) = cn cosh ny + dn sinh ny.
Boundary conditions give
u(x, 0) = X(x)Y (0) = 0 Y (0) = 0.
Thus, Y0 (0) = d0 = 0, and Yn (0) = cn = 0.
Y0 (y) = c0 y, Yn (y) = dn sinh ny, n = 1, 2, . . . . 
We have
0 y,
u0 (x, y) = X0 (x)Y0 (y) = b0 c0 y = a
n cos nx sinh ny.
un (x, y) = Xn (x)Yn (y) = (an cos nx)(dn sinh ny) = a
By superposition, we write
u(x, y) = a
0 y +

a
n cos nx sinh ny,

n=1

which saties the equation and the three homogeneous boundary conditions. The fourth
boundary condition gives
u(x, ) = g(x) = a
0 +


n=1

an cos nx sinh n,

Partial Dierential Equations

Igor Yanovsky, 2005





&
g(x) dx = 0 a
an cos nx sinh n dx = a
0 2 ,
0 +
n=1

&
g(x) cos mx dx = n=1
an sinh n 0 cos nx cos mx dx =

1
a
0 = 2
g(x) dx,
0

0
0

2
a
n sinh n =

g(x) cos nx dx.


0

am sinh m.

285

Partial Dierential Equations

Igor Yanovsky, 2005

286

Problem (W04, #5) The 2D LAPLACE Equation in an Upper-Half Plane.


Consider the Laplace equation
2u 2u
+ 2 = 0,
y > 0, < x < +
x2
y
u(x, 0)
u(x, 0) = f (x),
y
where f (x) C0 (R1 ).
Find a bounded solution u(x, y) and show that u(x, y) 0 when |x| + y .
Proof. Assume u(x, y) = X(x)Y (y), then substitution in the PDE gives X Y + XY  =
0.
Y 
X 
=
= .
X
Y

Consider X  + X = 0.
If = 0, X0 (x) = a0 x + b0 .

If > 0, Xn (x) = an cos n x + bn sin n x.


Since we look for bounded solutions as |x| , we have a0 = 0.
Consider Y  n Y = 0.
If n = 0, Y0 (y) = c0 y +d0 .

If n > 0, Yn (y) = cn e n y + dn e n y .
Since we look for bounded solutions as y , we have c0 = 0, dn = 0. Thus,
u(x, y) = a
0 +

n y

an cos




n x + bn sin n x .

n=1

Initial condition gives:


a0
f (x) = uy (x, 0) u(x, 0) =







( n + 1) a
n cos n x + bn sin n x .
n=1

f (x) C0 (R1 ), i.e. has compact support [L, L], for some L > 0. Thus the coecients
a
n , bn are given by
 L


f (x) cos n x dx = ( n + 1)
an L.
L
L

f (x) sin
L



n x dx = ( n + 1)bnL.

Thus, u(x, y) 0 when |x| + y .


70

70

Note that if we change the roles of X and Y in , the solution we get will be unbounded.

Partial Dierential Equations

Igor Yanovsky, 2005

287

Problem 3: The 2D LAPLACE Equation on a Circle.


Let be the unit disk in R2 and consider the problem

u = 0
in
u
on ,
n = h
where h is a continuous function.
Proof. Use polar coordinates (r, )


urr + 1r ur + r12 u = 0
u
r (1, ) = h()

for 0 r < 1, 0 < 2


for 0 < 2.

r 2 urr + rur + u = 0.
Let r = et , u(r(t), ).
ut = ur rt = et ur ,
utt = (et ur )t = et ur + e2t urr = rur + r 2 urr .
Thus, we have
utt + u = 0.
Let u(t, ) = X(t)Y (), which gives X  (t)Y () + X(t)Y  () = 0.
Y  ()
X (t)
=
= .
X(t)
Y ()
From Y  () + Y () = 0, we get Yn () = an cos n + bn sin n.
n = n2 , n = 0, 1, 2, . . ..
With these values of n we solve X  (t) n2 X(t) = 0.
X0 (r) = c0 log r + d0 .
If n = 0, X0 (t) = c0 t + d0 .
If n = 0, Xn (t) = cn ent + dn ent Xn (r) = cn r n + dn r n .
We have
u0 (r, ) = X0 (r)Y0 () = (c0 log r + d0 )a0 ,
un (r, ) = Xn (r)Yn () = (cn r n + dn r n )(an cos n + bn sin n).
But u must be nite at r = 0, so cn = 0, n = 0, 1, 2, . . ..
u0 (r, ) = d0 a0 ,
un (r, ) = dn r n (an cos n + bn sin n).
By superposition, we write
u(r, ) = a
0 +

r n (
an cos n + bn sin n).

n=1

Boundary condition gives


ur (1, ) =

n(
an cos n + bn sin n) = h().

n=1

The coecients an , bn for n 1 are determined from the Fourier series for h().
a0 is not determined by h() and therefore may take an arbitrary value. Moreover,

Partial Dierential Equations

Igor Yanovsky, 2005

288

 2
the constant term in the Fourier series for h() must be zero [i.e., 0 h()d = 0].
Therefore, the problem is not solvable for an arbitrary function h(), and when it is
solvable, the solution is not unique.

Partial Dierential Equations

Igor Yanovsky, 2005

Problem 4: The 2D LAPLACE Equation on a Circle.


Let = {(x, y) R2 : x2 + y 2 < 1} = {(r, ) : 0 r < 1, 0 < 2},
and use separation of variables (r, ) to solve the Dirichlet problem

u = 0
in
u(1, ) = g() for 0 < 2.
Proof. Use polar coordinates (r, )

for 0 r < 1, 0 < 2
urr + 1r ur + r12 u = 0
u(1, ) = g()
for 0 < 2.
r 2 urr + rur + u = 0.
Let r = et , u(r(t), ).
ut = ur rt = et ur ,
utt = (et ur )t = et ur + e2t urr = rur + r 2 urr .
Thus, we have
utt + u = 0.
Let u(t, ) = X(t)Y (), which gives X  (t)Y () + X(t)Y  () = 0.
Y  ()
X (t)
=
= .
X(t)
Y ()
From Y  () + Y () = 0, we get Yn () = an cos n + bn sin n.
n = n2 , n = 0, 1, 2, . . ..
With these values of n we solve X  (t) n2 X(t) = 0.
X0 (r) = c0 log r + d0 .
If n = 0, X0 (t) = c0 t + d0 .
If n = 0, Xn (t) = cn ent + dn ent Xn (r) = cn r n + dn r n .
We have
u0 (r, ) = X0 (r)Y0 () = (c0 log r + d0 )a0 ,
un (r, ) = Xn (r)Yn () = (cn r n + dn r n )(an cos n + bn sin n).
But u must be nite at r = 0, so cn = 0, n = 0, 1, 2, . . ..
u0 (r, ) = d0 a0 ,
un (r, ) = dn r n (an cos n + bn sin n).
By superposition, we write
u(r, ) = a
0 +

r n (
an cos n + bn sin n).

n=1

Boundary condition gives


u(1, ) = a
0 +


n=1

(
an cos n + bn sin n) = g().

289

Partial Dierential Equations


a
0 =
an =
bn =


1
g() d,
0

2
g() cos n d,
0

2
g() sin n d.
0

Igor Yanovsky, 2005

290

Partial Dierential Equations

Igor Yanovsky, 2005

291

Problem (F94, #6): The 2D LAPLACE Equation on a Circle.


Find all solutions of the homogeneous equation

Hint:

x2 + y 2 < 1,
uxx + uyy = 0,
u
u = 0,
x2 + y 2 = 1.
n

 = 1r r
(r r
) + r12
in polar coordinates.
2

Proof. Use polar coordinates (r, ):



urr + 1r ur + r12 u = 0
u
r (1, ) u(1, ) = 0

for 0 r < 1, 0 < 2


for 0 < 2.

Since we solve the equation on a circle, we have periodic conditions:

u(r, 0) = u(r, 2)

u (r, 0) = u (r, 2)

X(r)Y (0) = X(r)Y (2)




Y (0) = Y (2),

X(r)Y (0) = X(r)Y (2) Y  (0) = Y  (2).

Also, we want the solution to be bounded. In particular, u is bounded for r = 0.


r 2 urr + rur + u = 0.
Let r = et , u(r(t), ), we have
utt + u = 0.
Let u(t, ) = X(t)Y (), which gives X  (t)Y () + X(t)Y  () = 0.
Y  ()
X (t)
=
= .
X(t)
Y ()

From Y  () + Y () = 0, we get Yn () = an cos + bn sin .


Using periodic condition: Yn (0) = an ,



n = n
Yn (2) = an cos( n 2) + bn sin( n 2) = an
Thus, Yn () = an cos n + bn sin n.
With these values of n we solve X  (t) n2 X(t) = 0.
X0 (r) = c0 log r + d0 .
If n = 0, X0 (t) = c0 t + d0 .
nt
nt
Xn (r) = cn r n + dn r n .
If n = 0, Xn (t) = cn e + dn e
u must be nite at r = 0 cn = 0, n = 0, 1, 2, . . ..
u(r, ) = a
0 +

r n (
an cos n + bn sin n).

n=1

Boundary condition gives


a0 +
0 = ur (1, ) u(1, ) =

(n 1)(
an cos n + bn sin n).

n=1

Calculating Fourier coecients gives 2a0 = 0


an = 0, n = 2, 3, . . ..
(n 1)an = 0
a1 , b1 are constants. Thus,
u(r, ) = r(
a1 cos + b1 sin ).

a
0 = 0.

n = n2 .

Partial Dierential Equations

Igor Yanovsky, 2005

292

Problem (S00, #4).


a) Let (r, ) be polar coordinates on the plane,
i.e. x1 + ix2 = rei . Solve the boudary value problem
u = 0

in r < 1

u/r = f ()

on r = 1,

beginning with the Fourier series for f (you may assume that f is continuously differentiable). Give your answer as a power series in x1 + ix2 plus a power series in
x1 ix2 . There is a necessary condition on f for this boundary value problem to be
solvable that you will nd in the course of doing this.
b) Sum the series in part (a) to get a representation of u in the form
 2
N (r,  )f ( ) d .
u(r, ) =
0

Proof. a) Greens identity gives the necessary compatibility condition on f :





 2
u
u
d =
ds =
f () d =
u dx = 0.
0
r=1 r
n

Use polar coordinates (r, ):



urr + 1r ur + r12 u = 0
u
r (1, ) = f ()

for 0 r < 1, 0 < 2


for 0 < 2.

Since we solve the equation on a circle, we have periodic conditions:


u(r, 0) = u(r, 2)

u (r, 0) = u (r, 2)

X(r)Y (0) = X(r)Y (2)




Y (0) = Y (2),

X(r)Y (0) = X(r)Y (2) Y  (0) = Y  (2).

Also, we want the solution to be bounded. In particular, u is bounded for r = 0.


r 2 urr + rur + u = 0.
Let r = et , u(r(t), ), we have
utt + u = 0.
Let u(t, ) = X(t)Y (), which gives X  (t)Y () + X(t)Y  () = 0.
Y  ()
X (t)
=
= .
X(t)
Y ()

From Y  () + Y () = 0, we get Yn () = an cos + bn sin .


Using periodic condition: Yn (0) = an ,



n = n
Yn (2) = an cos( n 2) + bn sin( n 2) = an
Thus, Yn () = an cos n + bn sin n.
With these values of n we solve X  (t) n2 X(t) = 0.
X0 (r) = c0 log r + d0 .
If n = 0, X0 (t) = c0 t + d0 .

n = n2 .

Partial Dierential Equations

Igor Yanovsky, 2005

293

If n = 0, Xn (t) = cn ent + dn ent Xn (r) = cn r n + dn r n .


u must be nite at r = 0 cn = 0, n = 0, 1, 2, . . ..
u(r, ) = a
0 +

r n (
an cos n + bn sin n).

n=1

Since
ur (r, ) =

nr n1 (
an cos n + bn sin n),

n=1

the boundary condition gives


ur (1, ) =

n (
an cos n + bn sin n) = f ().

n=1

an =
bn =

 2
1
f () cos n d,
n 0
 2
1
f () sin n d.
n 0

 2
a
0 is not determined by f () (since 0 f () d = 0). Therefore, it may take an
arbitrary
 2 value. Moreover, the constant term in the Fourier series for f () must be zero
[i.e., 0 f ()d = 0]. Therefore, the problem is not solvable for an arbitrary function
f (), and when it is solvable, the solution is not unique.
b) In part (a), we obtained the solution and the Fourier coecients:
 2
1
f ( ) cos n d ,
an =
n 0
 2
bn = 1
f ( ) sin n d .
n 0
u(r, ) = a
0 +
= a
0 +
= a
0 +
= a
0 +
= a
0 +


n=1

r n (
an cos n + bn sin n)
r



n=1

rn
n
n=1

rn
n

1
n
 2


 1  2




f ( ) cos n d cos n +
f ( ) sin n d sin n
n 0




f ( ) cos n cos n + sin n sin n d

n=1
 2 

f ( ) cos n( ) d

rn
cos n(  ) f ( ) d .
n


n=1
N (r,  )

Partial Dierential Equations

Igor Yanovsky, 2005

294

Problem (S92, #6). Consider the Laplace equation


uxx + uyy = 0
for x2 + y 2 1. Denoting by x = r cos , y = r sin polar coordinates, let f = f () be
a given smooth function of . Construct a uniformly bounded solution which satises
boundary conditions
u=f

for x2 + y 2 = 1.

What conditions has f to satisfy such that


lim

x2 +y2

(x2 + y 2 )u(x, y) = 0?

Proof. Use polar coordinates (r, ):



for r 1
urr + 1r ur + r12 u = 0
u(1, ) = f ()
for 0 < 2.
Since we solve the equation on outside of a circle, we have periodic conditions:

u(r, 0) = u(r, 2)

X(r)Y (0) = X(r)Y (2)




Y (0) = Y (2),

X(r)Y (0) = X(r)Y (2) Y  (0) = Y  (2).

u (r, 0) = u(r, 2)

Also, we want the solution to be bounded. In particular, u is bounded for r = .


r 2 urr + rur + u = 0.
Let r = et , u(r(t), ), we have
utt + u = 0.
Let u(t, ) = X(t)Y (), which gives X  (t)Y () + X(t)Y  () = 0.
Y  ()
X (t)
=
= .
X(t)
Y ()

From Y  () + Y () = 0, we get Yn () = an cos + bn sin .


Using periodic condition: Yn (0) = an ,



n = n
Yn (2) = an cos( n 2) + bn sin( n 2) = an
Thus, Yn () = an cos n + bn sin n.
With these values of n we solve X  (t) n2 X(t) = 0.
X0 (r) = c0 log r + d0 .
If n = 0, X0 (t) = c0 t + d0 .
If n = 0, Xn (t) = cn ent + dn ent Xn (r) = cn r n + dn r n .
u must be nite at r = c0 = 0, dn = 0, n = 1, 2, . . ..
u(r, ) = a
0 +

r n (
an cos n + bn sin n).

n=1

Boundary condition gives


f () = u(1, ) = a
0 +


n=1

(
an cos n + bn sin n).

n = n2 .

Partial Dierential Equations

 2

a0 = 0 f () d,
2
 2

an = 0 f () cos n d,

 2

bn = 0 f () sin n d.

Igor Yanovsky, 2005

 2
1

0 = 2
f () d,
f0 = a
0
1 2
n = 0 f () cos n d,
fn = a

 2

f () sin n d.
fn = bn = 1

We need to nd conditions for f such that


lim

x2 +y2

(x2 + y 2 )u(x, y) = 0,

or

= 0,
lim r 2 u(r, ) 

need

lim r 2 f0 +


n=1


r n (fn cos n + fn sin n) 
= 0.
need

Since
lim




r 2n (fn cos n + fn sin n) = 0,

n>2

we need
2



2
r 2n (fn cos n + fn sin n) 
= 0.
lim r f0 +

n=1

Thus, the conditions are


fn , fn = 0,

n = 0, 1, 2.

need

295

Partial Dierential Equations

Igor Yanovsky, 2005

296

Problem (F96, #2): The 2D LAPLACE Equation on a Semi-Annulus.


Solve the Laplace equation in the semi-annulus

1 < r < 2, 0 < < ,


u = 0,

u(r, 0) = u(r, ) = 0,
1 < r < 2,

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

u(2, ) = 0,
0 < < .
Hint: Use the formula  =

1 2
r r (r r ) + r 2 2

for the Laplacian in polar coordinates.

Proof. Use polar coordinates (r, )


1
1
urr + ur + 2 u = 0
r
r
r 2 urr + rur + u = 0.

1 < r < 2, 0 < < ,

With r = et , we have
utt + u = 0.
Let u(t, ) = X(t)Y (), which gives X  (t)Y () + X(t)Y  () = 0.
Y  ()
X (t)
=
= .
X(t)
Y ()

From Y  () + Y () = 0, we get Yn () = an cos + bn sin .


Boundary conditions give
un (r, 0) = 0 = Xn (r)Yn (0) = 0, Yn (0) = 0,
un (r, ) = 0 = Xn (r)Yn () = 0, Yn () = 0.

= n n = n2 .
Thus, 0 = Yn (0) = an , and Yn () = bn sin = 0
Thus, Yn () = bn sin n, n = 1, 2, . . ..
With these values of n we solve X  (t) n2 X(t) = 0.
X0 (r) = c0 log r + d0 .
If n = 0, X0 (t) = c0 t + d0 .
If n > 0, Xn (t) = cn ent + dn ent Xn (r) = cn r n + dn r n .
We have,
u(r, ) =

Xn (r)Yn () =

n=1

(
cn r n + dn r n ) sin n.

n=1

Using the other two boundary conditions, we obtain


sin = u(1, ) =
0 = u(2, ) =


n=1

(
cn + dn ) sin n


c1 + d1 = 1,

cn + dn = 0, n = 2, 3, . . . .

(
cn 2n + dn 2n ) sin n

cn 2n + dn 2n = 0, n = 1, 2, . . ..

n=1

Thus, the coecients are given by


4
1
c1 = , d1 = ;
3
3
cn = 0, dn = 0.

Partial Dierential Equations

u(r, ) =

4
r

sin .
3r 3

Igor Yanovsky, 2005

297

Partial Dierential Equations

Igor Yanovsky, 2005

Problem (S98, #8): The 2D LAPLACE Equation on a Semi-Annulus.


Solve

1 < r < 2, 0 < < ,


u = 0,
u(r, 0) = u(r, ) = 0,
1 < r < 2,

u(1, ) = u(2, ) = 1,
0 < < .

Proof. Use polar coordinates (r, )


1
1
urr + ur + 2 u = 0
r
r
r 2 urr + rur + u = 0.

for 1 < r < 2, 0 < < ,

With r = et , we have
utt + u = 0.
Let u(t, ) = X(t)Y (), which gives X  (t)Y () + X(t)Y  () = 0.
Y  ()
X (t)
=
= .
X(t)
Y ()
From Y  () + Y () = 0, we get Yn () = an cos n + bn sin n.
Boundary conditions give
un (r, 0) = 0 = Xn (r)Yn (0) = 0, Yn (0) = 0,
un (r, ) = 0 = Xn (r)Yn () = 0, Yn () = 0.
Thus, 0 = Yn (0) = an , and Yn () = bn sin n.
n = n2 , n = 1, 2, . . ..
With these values of n we solve X  (t) n2 X(t) = 0.
X0 (r) = c0 log r + d0 .
If n = 0, X0 (t) = c0 t + d0 .
If n > 0, Xn (t) = cn ent + dn ent Xn (r) = cn r n + dn r n .
We have,
u(r, ) =

Xn (r)Yn () =

n=1

(
cn r n + dn r n ) sin n.

n=1

Using the other two boundary conditions, we obtain


u(1, ) = 1 =
u(2, ) = 1 =


n=1

(
cn + dn ) sin n,
(
cn 2n + dn 2n ) sin n,

n=1

which give the two equations for cn and dn :




(
cn + dn ),
sin n d =
2
0


(
cn 2n + dn 2n ),
sin n d =
2
0
that can be solved.

298

Partial Dierential Equations

Igor Yanovsky, 2005

299

Problem (F89, #1). Consider Laplace equation inside a 90 sector of a circular


annulus

u = 0
a < r < b, 0 < <
2
subject to the boundary conditions
u
(r, 0) = 0,

u
(a, ) = f1 (),
r

u
(r, ) = 0,

2
u
(b, ) = f2 (),
r

where f1 (), f2 () are continuously dierentiable.


a) Find the solution of this equation with the prescribed
boundary conditions using separation of variables.
Proof. a) Use polar coordinates (r, )
1
1
urr + ur + 2 u = 0
r
r
r 2 urr + rur + u = 0.

for a < r < b, 0 < <

,
2

With r = et , we have
utt + u = 0.
Let u(t, ) = X(t)Y (), which gives X  (t)Y () + X(t)Y  () = 0.
Y  ()
X (t)
=
= .
X(t)
Y ()

From Y  () + Y () = 0, we get Yn () = an cos + bn sin .


Boundary conditions give
un (r, 0) = Xn (r)Yn (0) = 0 Yn (0) = 0,

un (r, ) = Xn (r)Yn ( ) = 0 Yn ( ) = 0.


2
2
2


Yn () = an n sin n + bn n cos n . Thus, Yn (0) = bn n = 0
n 2 = n n = (2n)2 .
Yn ( 2 ) = an n sin n 2 = 0
Thus, Yn () = an cos(2n), n = 0, 1, 2, . . ..
In particular, Y0 () = a0 t + b0 . Boundary conditions give Y0 () = b0 .
With these values of n we solve X  (t) (2n)2 X(t) = 0.
X0 (r) = c0 log r + d0 .
If n = 0, X0 (t) = c0 t + d0 .
2nt
2nt
Xn (r) = cn r 2n + dn r 2n .
If n > 0, Xn (t) = cn e + dn e
u(r, ) = c0 log r + d0 +

(
cn r 2n + dn r 2n ) cos(2n).

n=1

Using the other two boundary conditions, we obtain

ur (r, ) =

c0 
+
(2n
cn r 2n1 + 2ndn r 2n1 ) cos(2n).
r
n=1

bn = 0.

Partial Dierential Equations

Igor Yanovsky, 2005

300


c0
+2
n(
cn a2n1 + dn a2n1 ) cos(2n),
a

f1 () = ur (a, ) =

n=1

c0
+2
b

f2 () = ur (b, ) =

n(
cn b2n1 + dn b2n1 ) cos(2n).

n=1

which give the two equations for cn and dn :



2

n(
cn a2n1 + dn a2n1 ),
f1 () cos(2n) d =
2
0

2

n(
cn b2n1 + dn b2n1 ).
f2 () sin(2n) d =
2
0

b) Show that the solution exists if and only if




2
2
f1 () d b
f2 () d = 0.
a
0

Proof. Using Greens identity, we obtain:



0 =


=


u dx =


=


=

u
n
 0

u
u
(b, ) d +
(a, ) d +

r
r
2

2
f2 () d +
f1 () d + 0 + 0
0


f1 () d +

b
a

u
(r, 0) dr +


b

u

r,
dr

f2 () d.

c) Is the solution unique?

Proof. No, since the boundary conditions are Neumann. The solution is unique only
up to a constant.

Partial Dierential Equations

Igor Yanovsky, 2005

Problem (S99, #4). Let u(x, y) be harmonic inside the unit disc,
with boundary values along the unit circle

1, y > 0
u(x, y) =
0, y 0.
Compute u(0, 0) and u(0, y).
Proof. Since u is harmonic, u = 0. Use polar coordinates (r, )

1
+ r12 u = 0
0 r < 1, 0 < 2

urr + r ur
1,
0<<

u(1, ) =
0,
2.
r 2 urr + rur + u = 0.
With r = et , we have
utt + u = 0.
Let u(t, ) = X(t)Y (), which gives X  (t)Y () + X(t)Y  () = 0.
Y  ()
X (t)
=
= .
X(t)
Y ()
From Y  () + Y () = 0, we get Yn () = an cos n + bn sin n.
n = n2 , n = 1, 2, . . ..
With these values of n we solve X  (t) n2 X(t) = 0.
X0 (r) = c0 log r + d0 .
If n = 0, X0 (t) = c0 t + d0 .
If n > 0, Xn (t) = cn ent + dn ent Xn (r) = cn r n + dn r n .
We have
u0 (r, ) = X0 (r)Y0 () = (c0 log r + d0 )a0 ,
un (r, ) = Xn (r)Yn () = (cn r n + dn r n )(an cos n + bn sin n).
But u must be nite at r = 0, so cn = 0, n = 0, 1, 2, . . ..
0 ,
u0 (r, ) = a
an cos n + bn sin n).
un (r, ) = r n (
By superposition, we write
u(r, ) = a
0 +

r n (
an cos n + bn sin n).

n=1

Boundary condition gives


u(1, ) = a
0 +


n=1

(
an cos n + bn sin n) =

1,
0,

0<<
2,

and the coecients an and bn are determined from the above equation.
71

71

See Yanas solutions, where Greens function on a unit disk is constructed.

301

Partial Dierential Equations

23

Igor Yanovsky, 2005

302

Problems: Separation of Variables - Poisson Equation

Problem (F91, #2): The 2D POISSON Equation on a Quarter-Circle.


Solve explicitly the following boundary value problem
uxx + uyy = f (x, y)
in the domain = {(x, y), x > 0, y > 0, x2 + y 2 < 1}
with boundary conditions
u=0
u
=0
x
u=0

for y = 0, 0 < x < 1,


for x = 0, 0 < y < 1,
for x > 0, y > 0, x2 + y 2 = 1.

Function f (x, y) is known and is assumed to be continuous.


Proof. Use polar coordinates (r, ):

urr + 1r ur + r12 u = f (r, )

u(r, 0) = 0
0 r < 1,

0 r < 1,
u (r, 2 ) = 0

u(1, ) = 0
0 2 .

0 r < 1, 0 <

We solve
r 2 urr + rur + u = 0.
Let r = et , u(r(t), ), we have
utt + u = 0.

Let u(t, ) = X(t)Y (), which gives X  (t)Y () + X(t)Y  () = 0.


Y  ()
X (t)
=
= .
X(t)
Y ()

From Y  () + Y () = 0, we get Yn () = an cos + bn sin .


conditions:



u(r, 0) = X(r)Y (0) = 0


Y
(0)
=
Y
= 0.
2
u (r, 2 ) = X(r)Y  ( 2 ) = 0



(0)
=
a
=
0,
and
Y
(
)
=

b
cos
n 2 = 0
Thus,
Y
n
n
n
n
n
2

n 2 = n 2 , n = 1, 2, . . . n = (2n 1)2 .

Thus, Yn () = bn sin(2n 1), n = 1, 2, . . .. Thus, we have


u(r, ) =


n=1

Xn (r) sin[(2n 1)].

Boundary

Partial Dierential Equations

Igor Yanovsky, 2005

303

We now plug this equation into  with inhomogeneous term and obtain




Xn(t) sin[(2n 1)] (2n 1)2 Xn (t) sin[(2n 1)]

= f (t, ),

n=1





Xn (t) (2n 1)2 Xn (t) sin[(2n 1)] = f (t, ),

n=1


 
Xn (t) (2n 1)2 Xn (t) =
4
Xn(t)

(2n 1) Xn (t) =

f (t, ) sin[(2n 1)] d,


0

f (t, ) sin[(2n 1)] d.

The solution to this equation is


Xn (t) = cn e(2n1)t + dn e(2n1)t + Unp (t),
Xn (r) = cn r

(2n1)

+ dn r

(2n1)

or

+ unp(r),

where unp is the particular solution of inhomogeneous equation.


u must be nite at r = 0 cn = 0, n = 1, 2, . . .. Thus,
u(r, ) =





dn r (2n1) + unp(r) sin[(2n 1)].

n=1

Using the last boundary condition, we have


0 = u(1, ) =





dn + unp(1) sin[(2n 1)],

n=1

(dn + unp (1)),


4
dn = unp (1).

0=

u(r, ) =





unp (1)r (2n1) + unp (r) sin[(2n 1)].

n=1

The method used to solve this problem is similar to section


Problems: Eigenvalues of the Laplacian - Poisson Equation:
1) First, we nd Yn () eigenfunctions.
2) Then, we plug in our guess u(t, ) = X(t)Y () into the equation utt + u = f (t, )
and solve an ODE in X(t).

Note the similar problem on 2D Poisson equation on a square domain. The problem is used by rst nding the eigenvalues and eigenfunctions of the Laplacian, and
then expanding f (x, y) in eigenfunctions, and comparing coecients of f with the general solution u(x, y).
Here, however, this could not be done because of the circular geometry of the domain.
In particular, the boundary conditions do not give enough information to nd explicit
representations for m and n . Also, the condition u = 0 for x > 0, y > 0, x2 +y 2 = 1

Partial Dierential Equations

Igor Yanovsky, 2005

can not be used.


72

72

ChiuYens solutions have attempts to solve this problem using Greens function.

304

Partial Dierential Equations

24

Igor Yanovsky, 2005

305

Problems: Separation of Variables - Wave Equation

Example (McOwen 3.1 #2). We considered the initial/boundary value problem and
solved it using Fourier Series. We now solve it using the Separation of Variables.

0 < x < , t > 0

utt uxx = 0
(24.1)
0<x<
u(x, 0) = 1,
ut (x, 0) = 0

u(0, t) = 0,
u(, t) = 0
t 0.
Proof. Assume u(x, t) = X(x)T (t), then substitution in the PDE gives XT  X  T = 0.
T 
X 
=
= .
X
T
From X  + X = 0, we get Xn (x) = an cos nx + bn sin nx.
give

u(0, t) = X(0)T (t) = 0
X(0) = X() = 0.
u(, t) = X()T (t) = 0

Boundary conditions

Thus, Xn (0) = an = 0, and Xn (x) = bn sin nx, n = n2 , n = 1, 2, . . . .


With these values of n , we solve T  +n2 T = 0 to nd Tn (t) = cn sin nt+dn cos nt.
Thus,
u(x, t) =
ut (x, t) =



n=1


cn sin nt + dn cos nt sin nx,

n
cn cos nt ndn sin nt sin nx.

n=1

Initial conditions give


1 = u(x, 0) =
0 = ut (x, 0) =


n=1

dn sin nx,
n
cn sin nx.

n=1

By orthogonality, we may multiply both equations by sin mx and integrate:




sin mx dx = dm ,
2
0

0 dx = n
cn ,
2
0
which gives the coecients
2
(1 cos n) =
dn =
n

4
n ,

0,

n odd,
n even,

and

Plugging the coecients into a formula for u(x, t), we get

4  cos(2n + 1)t sin(2n + 1)x


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

cn = 0.

Partial Dierential Equations

Igor Yanovsky, 2005

306

Example. Use the method of separation of variables to nd the solution to:

0<x<1

utt + 3ut + u = uxx ,


u(0, t) = 0, u(1, t) = 0,

u(x, 0) = 0, ut (x, 0) = x sin(2x).


Proof. Assume u(x, t) = X(x)T (t), then substitution in the PDE gives
XT  + 3XT  + XT = X T,
T
X 
T 
+3 +1 =
= .
T
T
X

From X  + X = 0, Xn (x) = an cos n x + bn sin n x. Boundary conditions


give

u(0, t) = X(0)T (t) = 0
X(0) = X(1) = 0.
u(1, t) = X(1)T (t) = 0

Thus, Xn (0) =
an = 0, and Xn (x)
= bn sin n x.
Xn (1) = bn sin n = 0. Hence, n = n, or n = (n)2 , n = 1, 2, . . ..
n = (n)2,

Xn (x) = bn sin nx.

With these values of n , we solve


T  + 3T  + T = n T,
T  + 3T  + T = (n)2 T,
T  + 3T  + (1 + (n)2 )T = 0.
st
We can solve
' this 2nd-order ODE with the following guess, T (t) = ce' to obtain
s = 32 54 (n)2 . For n 1, 54 (n)2 < 0. Thus, s = 32 i (n)2 54 .

32 t

Tn (t) = e




5
5
cn cos (n)2 t + dn sin (n)2 t .
4
4

u(x, t) = X(x)T (t) =


n=1

32 t




5
5
cn cos (n)2 t + dn sin (n)2 t sin nx.
4
4

Initial conditions give


0 = u(x, 0) =

cn sin nx.

n=1

By orthogonality, we may multiply this equations by sin mx and integrate:


 1
1
0 dx = cm
cm = 0.
2
0

Partial Dierential Equations

Igor Yanovsky, 2005

Thus,

307

5
(n)2 t sin nx.
4
n=1







3
5
5

5
32 t
32 t
2
2
2
dn e
cos (n) t sin nx,
(n)
sin (n) t + dn e
ut (x, t) =
2
4
4
4
n=1


5
sin nx.
dn
(n)2
x sin(2x) = ut (x, 0) =
4
u(x, t) =

32 t

dn e

sin

n=1

By orthogonality, we may multiply this equations by sin mx and integrate:



 1
5
1

,
x sin(2x) sin(mx) dx = dm
(m)2
2
4
0
 1
2
dn = '
x sin(2x) sin(nx) dx.
5
2
0
(n) 4
32 t

u(x, t) = e


n=1

dn sin

5
(n)2 t sin nx.
4

Problem (F04, #1). Solve the following initial-boundary value problem for the wave
equation with a potential term,

0 < x < , t < 0


utt uxx + u = 0
u(0, t) = u(, t) = 0
t>0

0 < x < ,
u(x, 0) = f (x), ut (x, 0) = 0
where


f (x) =

x
x

if x (0, /2),
if x (/2, ).

The answer should be given in terms of an innite series of explicitly given functions.
Proof. Assume u(x, t) = X(x)T (t), then substitution in the PDE gives
XT  X T + XT = 0,
X 
T 
+1 =
= .
T
X

From X  + X = 0, Xn (x) = an cos n x + bn sin n x. Boundary conditions


give

u(0, t) = X(0)T (t) = 0
X(0) = X() = 0.
u(, t) = X()T (t) = 0

an = 0, and Xn (x)= bn sin n x.


Thus, Xn (0) =
Xn () = bn sin n = 0. Hence, n = n, or n = n2 , n = 1, 2, . . ..
n = n2 ,

Xn (x) = bn sin nx.

Partial Dierential Equations

Igor Yanovsky, 2005

308

With these values of n , we solve


T  + T = n T,
T  + T = n2 T,
Tn + (1 + n2 )Tn = 0.
The solution to this 2nd-order ODE is of the form:


Tn (t) = cn cos 1 + n2 t + dn sin 1 + n2 t.

u(x, t) = X(x)T (t) =




cn cos

1 + n2 t + dn sin


1 + n2 t sin nx.

n=1

ut (x, t) =






cn ( 1 + n2 ) sin 1 + n2 t + dn ( 1 + n2 ) cos 1 + n2 t sin nx.

n=1

Initial conditions give


f (x) = u(x, 0) =
0 = ut(x, 0) =

cn sin nx.

n=1


dn ( 1 + n2 ) sin nx.

n=1

By orthogonality, we may multiply both equations by sin mx and integrate:




f (x) sin mx dx = cm ,
2
0


0 dx = dm
1 + m2 ,
2
0
which gives the coecients



2
2 2
2
f (x) sin nx dx =
x sin nx dx +
( x) sin nx dx
cn =
0
0

2









1

1 2
2
1
1
2
2


x cos nx +
cos nx + x cos nx
cos nx dx +
cos nx dx
=

n
0
n 0

n
n
n
2
2
2



n
1
1
n
2

cos
+ 2 sin
2 sin 0
=

2n
2
n
2
n



n
1
1
n
2
cos n + cos
+ cos n
cos
2 sin n + 2 sin
+

n
n
2
n
2n
2
n
n
2




2 1
4
n
n
n
2 1
+
=
sin
sin
sin
=
n2
2
n2
2
n2
2

n = 2k
0,
0,
n = 2k
4
=
=
n = 4m + 1
n1
2,
4
n

(1) 2 n2 ,
n = 2k + 1.
4
n = 4m + 3
n2 ,
dn = 0.
u(x, t) =



n=1

cn cos



1 + n2 t sin nx.

Partial Dierential Equations

25

Igor Yanovsky, 2005

309

Problems: Separation of Variables - Heat Equation

Problem (F94, #5).


Solve the initial-boundary value problem

0 < x < 2, t > 0

ut = uxx
2
0x2
u(x, 0) = x x + 1

u(0, t) = 1, u(2, t) = 3
t > 0.
Find limt+ u(x, t).
Proof. First, we need to obtain function v that satises vt = vxx and takes 0
boundary conditions. Let
v(x, t) = u(x, t) + (ax + b),

(25.1)

where a and b are constants to be determined. Then,


vt = ut ,
vxx = uxx .
Thus,
vt = vxx .
We need equation (25.1) to take 0 boundary conditions for v(0, t) and v(2, t):
v(0, t) = 0 = u(0, t) + b = 1 + b

b = 1,

v(2, t) = 0 = u(2, t) + 2a 1 = 2a + 2

a = 1.

Thus, (25.1) becomes


v(x, t) = u(x, t) x 1.

(25.2)

The new problem is

vt = vxx ,
v(x, 0) = (x2 x + 1) x 1 = x2 2x,

v(0, t) = v(2, t) = 0.
We solve the problem for v using the method of separation of variables.
Let v(x, t) = X(x)T (t), which gives XT  X  T = 0.
T
X 
=
= .
X
T

From X  + X = 0, we get Xn (x) = an cos x + bn sin x.


Using boundary conditions, we have

v(0, t) = X(0)T (t) = 0
X(0) = X(2) = 0.
v(2, t) = X(2)T (t) = 0

an = 0, and Xn (x)
Hence, Xn (0) =
= bn sin x.
2
Xn (2) = bn sin 2 = 0 2 = n n = ( n
2 ) .
Xn (x) = bn sin

nx
,
2

n =

n 2
2

Partial Dierential Equations

Igor Yanovsky, 2005

With these values of n , we solve T  +


Tn (t) = cn e(

v(x, t) =

n 2
) t
2

 n 2
2

T = 0 to nd

Xn (x)Tn (t) =

n=1

cn e(

n=1

n 2
) t
2

sin

nx
.
2

Coecients cn are obtained using the initial condition:


v(x, 0) =

cn sin

n=1


cn =

2
0

nx
= x2 2x.
2

nx
dx =
(x2 2x) sin
2

v(x, t) =

n=2k1

0
n is even,
32
(n)3 n is odd.

32 ( n )2 t
nx
.
e 2
sin
3
(n)
2

We now use equation (25.2) to convert back to function u:


u(x, t) = v(x, t) + x + 1.
u(x, t) =


n=2k1

32 ( n )2 t
nx
+ x + 1.
e 2
sin
(n)3
2

lim u(x, t) = x + 1.

t+

310

Partial Dierential Equations

Igor Yanovsky, 2005

311

Problem (S96, #6).


Let u(x, t) be the solution of the initial-boundary value problem for the heat equation

0 < x < L, t > 0

ut = uxx
u(x, 0) = f (x)
0xL

ux (0, t) = ux (L, t) = A
t>0
(A = Const).
Find v(x) - the limit of u(x, t) when t . Show that v(x) is one of the ininitely
many solutions of the stationary problem
vxx = 0

0<x<L

vx (0) = vx (L) = A.
Proof. First, we need to obtain function v that satises vt = vxx and takes 0
boundary conditions. Let
v(x, t) = u(x, t) + (ax + b),

(25.3)

where a and b are constants to be determined. Then,


vt = ut ,
vxx = uxx .
Thus,
vt = vxx .
We need equation (25.3) to take 0 boundary conditions for vx (0, t) and vx (L, t).
vx = ux + a.
vx (0, t) = 0 = ux (0, t) + a = A + a

a = A,

vx (L, t) = 0 = ux (L, t) + a = A + a

a = A.

We may set b = 0 (innitely many solutions are possible, one for each b).
Thus, (25.3) becomes
v(x, t) = u(x, t) Ax.
The new problem is

vt = vxx ,
v(x, 0) = f (x) Ax,

vx (0, t) = vx(L, t) = 0.
We solve the problem for v using the method of separation of variables.
Let v(x, t) = X(x)T (t), which gives XT  X  T = 0.
T
X 
=
= .
X
T

From X  + X = 0, we get Xn (x) = an cos x + bn sin x.


Using boundary conditions, we have

vx (0, t) = X  (0)T (t) = 0
X  (0) = X  (L) = 0.

vx (L, t) = X (L)T (t) = 0

(25.4)

Partial Dierential Equations

Igor Yanovsky, 2005

312

x + bn cos x.
Xn (x) = an sin

Hence, Xn (0)


= bn n= 0 bn =
0; and Xn (x) = an cos x.
2
Xn (L) = an sin L = 0 L = n n = ( n
L ) .
nx
,
L

Xn (x) = an cos

n 2

n =

L
 n 2

With these values of n , we solve T  +


T0 (t) = c0 ,

v(x, t) =

n 2
) t
L

Tn (t) = cn e(

T = 0 to nd

, n = 1, 2, . . ..

Xn (x)Tn (t) = c0 +

n=1

cn e(

n 2
) t
L

cos

n=1

nx
.
L

Coecients cn are obtained using the initial condition:


v(x, 0) = c0 +

cn cos

n=1


L
c0 =
L
cn =
2

nx
= f (x) Ax.
L


(f (x) Ax) dx =

nx
dx
(f (x) Ax) cos
L

1
v(x, t) =
L


0


1 L
AL
c0 =
,
f (x) dx
L 0
2

1 L
nx
cn =
dx.
(f (x) Ax) cos
L 0
L

AL2
f (x) dx
2

f (x) dx

n 2
AL 
nx
+
.
cn e( L ) t cos
2
L
n

We now use equation (25.4) to convert back to function u:


u(x, t) = v(x, t) + Ax.
1
u(x, t) =
L

L
0

f (x) dx

lim u(x, t) = Ax + b,

t+

n 2
AL 
nx
+
+ Ax.
cn e( L ) t cos
2
L
n

b arbitrary.

To show that v(x) is one of the ininitely many solutions of the stationary problem
vxx = 0

0<x<L

vx (0) = vx (L) = A,
we can solve the boundary value problem to obtain v(x, t) = Ax+b, where b is arbitrary.

Partial Dierential Equations

Igor Yanovsky, 2005

313

Heat Equation with Nonhomogeneous Time-Independent BC in N-dimensions.


The solution to this problem takes somewhat dierent approach than in the last few problems, but is similar.
Consider the following initial-boundary value problem,

x , t 0
ut = u,
u(x, 0) = f (x),
x

u(x, t) = g(x),
x , t > 0.
Proof. Let w(x) be the solution of the Dirichlet problem:

w = 0,
x
w(x) = g(x),
x
and let v(x, t) be the solution of the IBVP for the heat equation with homogeneous
BC:

x , t 0

vt = v,
v(x, 0) = f (x) w(x),
x

v(x, t) = 0,
x , t > 0.
Then u(x, t) satises
u(x, t) = v(x, t) + w(x).
lim u(x, t) = w(x).

Partial Dierential Equations

Igor Yanovsky, 2005

314

Nonhomogeneous Heat Equation with Nonhomogeneous Time-Independent


BC in N dimensions.
Describe the method of solution of the problem

x , t 0

ut = u + F (x, t),
u(x, 0) = f (x),
x

u(x, t) = g(x),
x , t > 0.
Proof. We rst nd u1 , the solution to the homogeneous heat equation (no F (x, t)).
Let w(x) be the solution of the Dirichlet problem:

w = 0,
x
w(x) = g(x),
x
and let v(x, t) be the solution of the IBVP for the heat equation with homogeneous
BC:

x , t 0

vt = v,
v(x, 0) = f (x) w(x),
x

v(x, t) = 0,
x , t > 0.
Then u1 (x, t) satises
u1 (x, t) = v(x, t) + w(x).
lim u1 (x, t) = w(x).

The solution to the homogeneous equation with 0 boundary conditions is given by


Duhamels principle.

for t > 0, x Rn
u2t = u2 + F (x, t)
(25.5)
u2 (x, 0) = 0
for x Rn .
Duhamels principle gives the solution:
 t
y, t s) F (y, s) dy ds
K(x
u2 (x, t) =
0

Rn

Note: u2 (x, t) = 0 on may not be satised.


 t
u(x, t) = v(x, t) + w(x) +
0

Rn

y, t s) F (y, s) dy ds.
K(x

Partial Dierential Equations

Igor Yanovsky, 2005

Problem (S98, #5). Find the solution of

t 0, 0 < x < 1,

ut = uxx ,
u(x, 0) = 0,
0 < x < 1,

t
u(0, t) = 1 e , ux (1, t) = et 1,

315

t > 0.

Prove that limt u(x, t) exists and nd it.


Proof. First, we need to obtain function v that satises vt = vxx and takes 0
boundary conditions. Let
v(x, t) = u(x, t) + (ax + b) + (c1 cos x + c2 sin x)et ,

(25.6)

where a, b, c1 , c2 are constants to be determined. Then,


vt = ut (c1 cos x + c2 sin x)et ,
vxx = uxx + (c1 cos x c2 sin x)et .
Thus,
vt = vxx .
We need equation (25.6) to take 0 boundary conditions for v(0, t) and vx (1, t):
v(0, t) = 0 = u(0, t) + b + c1 et
= 1 et + b + c1 et .
Thus, b = 1, c1 = 1, and (25.6) becomes
v(x, t) = u(x, t) + (ax 1) + (cos x + c2 sin x)et .

(25.7)

vx (x, t) = ux (x, t) + a + ( sin x + c2 cos x)et ,


vx (1, t) = 0 = ux (1, t) + a + ( sin 1 + c2 cos 1)et
= 1 + a + (1 sin 1 + c2 cos 1)et .
Thus, a = 1, c2 =

sin 11
cos 1 ,

and equation (25.7) becomes

v(x, t) = u(x, t) + (x 1) + (cos x +

sin 1 1
sin x)et .
cos 1

(25.8)

Initial condition tranforms to:


v(x, 0) = u(x, 0) + (x 1) + (cos x +
The new problem is

vt = vxx ,
v(x, 0) = (x 1) + (cos x +

v(0, t) = 0,
vx (1, t) = 0.

sin 1 1
sin 1 1
sin x) = (x 1) + (cos x +
sin x).
cos 1
cos 1

sin 11
cos 1

sin x),

We solve the problem for v using the method of separation of variables.


Let v(x, t) = X(x)T (t), which gives XT  X  T = 0.
T
X 
=
= .
X
T

Partial Dierential Equations

Igor Yanovsky, 2005

316

From X  + X = 0, we get Xn (x) = an cos x + bn sin x.


Using the rst boundary condition, we have
X(0) = 0.

Hence, Xn (0) = an = 0, and Xn (x) = bn sin x. We also have


v(0, t) = X(0)T (t) = 0

vx (1, t) = X  (1)T (t) = 0 X  (1) = 0.

bn cos x,
Xn (x) =

Xn (1) =
bn cos = 0,

cos = 0,

= n + .
2
Thus,


x,
Xn (x) = bn sin n +
2

2
n = n +
.
2

2
With these values of n , we solve T  + n + 2 T = 0 to nd
2

Tn (t) = cn e(n+ 2 ) t .

v(x, t) =

Xn (x)Tn (t) =

n=1


n=1

bn sin n + x e(n+ 2 )2 t.
2

We now use equation (25.8) to convert back to function u:


u(x, t) = v(x, t) (x 1) (cos x +
u(x, t) =

sin 1 1
sin x)et .
cos 1

bn sin n + x e(n+ 2 )2 t (x 1) (cos x + sin 1 1 sin x)et .


2
cos 1
n=1

Coecients bn are obtained using the initial condition:


bn sin n + x (x 1) (cos x + sin 1 1 sin x).
u(x, 0) =
2
cos 1
n=1

Finally, we can check that the dierential equation and the boundary conditions are
satised:
u(0, t) = 1 (1 + 0)et = 1 et . 


bn n + cos n + x e(n+ 2 )2 t 1 + (sin x sin 1 1 cos x)et ,
ux (x, t) =
2
2
cos 1
n=1

sin 1 1
cos 1)et = 1 + et . 
cos 1


(n+ )2 t
sin 1 1
2
2
xe
sin x)et = uxx. 
=
bn n +
sin n +
+ (cos x +
2
2
cos 1

ux (1, t) = 1 + (sin 1
ut

n=1

Partial Dierential Equations

Igor Yanovsky, 2005

317

Problem (F02, #6). The temperature of a rod insulated at the ends with an exponentially decreasing heat source in it is a solution of the following boundary value
problem:

2t

for (x, t) [0, 1] R+


ut = uxx + e g(x)
ux (0, t) = ux (1, t) = 0

u(x, 0) = f (x).
Find the solution to this problem by writing u as a cosine series,
u(x, t) =

an (t) cos nx,

n=0

and determine limt u(x, t).


Proof. Let g accept an expansion in eigenfunctions
g(x) = b0 +


bn cos nx

with

bn = 2

n=1

g(x) cos nx dx.


0

Plugging  in the PDE gives:


a0 (t) +

an (t) cos nx =

n=1

n2 2 an (t) cos nx + b0 e2t + e2t

n=1

bn cos nx,

n=1

which gives

a0 (t) = b0 e2t ,
an (t) + n2 2 an (t) = bn e2t ,

n = 1, 2, . . . .

Adding homogeneous and particular solutions of the above ODEs, we obtain the solutions

a0 (t) = c0 b20 e2t ,
2 2
bn
2t
, n = 1, 2, . . . ,
an (t) = cn en t 2n
2 2 e
for some constants cn , n = 0, 1, 2, . . .. Thus,
u(x, t) =


2 2
cn en t
n=0


bn
2t
e
cos nx.
2 n2 2

Initial condition gives



cn
u(x, 0) =
n=0


bn
cos nx = f (x),
2 n2 2

As, t , the only mode that survives is n = 0:


u(x, t) c0 +

b0
2

as t .

Partial Dierential Equations

Igor Yanovsky, 2005

318

Problem (F93, #4). a) Assume f, g C . Give the compatibility conditions which


f and g must satisfy if the following problem is to possess a solution.
u = f (x)
u
(s) = g(s)
n

x
s .

Show that your condition is necessary for a solution to exist.


b) Give an explicit solution to

ut = uxx + cos x
ux (0, t) = ux (2, t) = 0

u(x, 0) = cos x + cos 2x

x [0, 2]
t>0
x [0, 2].

c) Does there exist a steady state solution to the problem in (b) if


ux (0) = 1

ux (2) = 0

Explain your answer.


Proof. a) Integrating the equation and using Greens identity gives:




u
ds =
f (x) dx =
u dx =
g(s) ds.

b) With
v(x, t) = u(x, t) cos x
the problem above transforms to

vt = vxx
vx (0, t) = vx(2, t) = 0

v(x, 0) = cos 2x.


We solve this problem for v using the separation of variables. Let v(x, t) = X(x)T (t),
which gives XT  = X T .
T
X 
=
= .
X
T

get
X
(x)
=
a
cos
x
+
b
sin
x.
From X  +X = 0, we
n
n
n

Xn (x) = n an sin x + n bn cos x.


Using boundary conditions, we have

vx (0, t) = X  (0)T (t) = 0
X  (0) = X  (2) = 0.

vx (2, t) = X (2)T (t) = 0

= n bn= 0, and Xn (x) =an cos n x.


Hence, Xn (0)

n = n2 n = ( n2 )2 . Thus,
Xn (2) = n an sin n 2 = 0
Xn (x) = an cos

nx
,
2

n =

n 2
2

Partial Dierential Equations

Igor Yanovsky, 2005

With these values of n , we solve T  +

 n 2
2

319

T = 0 to nd

n 2

Tn (t) = cn e( 2 ) t .

v(x, t) =

Xn (x)Tn (t) =

n=0

n 2
t

a
n e( 2 )

n=0

cos

nx
.
2

Initial condition gives


v(x, 0) =

an cos

n=0

nx
= cos 2x.
2

n = 0, n = 4. Hence,
Thus, a
4 = 1, a
v(x, t) = e4t cos 2x.
u(x, t) = v(x, t) + cos x = e4t cos 2x + cos x.

c) Does there exist a steady state solution to the problem in (b) if


ux (0) = 1

ux (2) = 0

Explain your answer.


c) Set ut = 0. We have

uxx + cos x = 0
ux (0) = 1, ux (2) = 0.

x [0, 2]

uxx = cos x,
ux = sin x + C,
u(x) = cos x + Cx + D.
Boundary conditions give:
1 = ux (0) = C,
0 = ux (2) = C

contradiction

There exists no steady state solution.

We may use the result we obtained in part (a) with


need


u
ds,
f (x) dx =

n
 2
cos x dx = ux (2) ux (0) = 
1 .

 0 
given
=0

uxx = cos x = f (x).

We

Partial Dierential Equations

Igor Yanovsky, 2005

320

Problem (F96, #7). Solve the parabolic problem



 
 
1 12
u
u
=
,
0 x , t > 0
0 2
v xx
v t
u(x, 0) = sin x,

u(0, t) = u(, t) = 0,

v(x, 0) = sin x,

v(0, t) = v(, t) = 0.

Prove the energy estimate (for general initial data)




2
2
[u (x, t) + v (x, t)] dx c
[u2 (x, 0) + v 2 (x, 0)] dx
x=0

x=0

for come constant c.


Proof. We can solve the second equation for v and then use the value of v to solve the
rst equation for u. 73
We have

0 x , t > 0
vt = 2vxx,
v(x, 0) = sin x,

v(0, t) = v(, t) = 0.
Assume v(x, t) = X(x)T (t), then substitution in the PDE gives XT  = 2X T .
X 
T
=2
= .
T
X
X 

'

2X

2x

+
= 0, we get Xn (x) = an cos
+ bn sin
From
Boundary conditions give

v(0, t) = X(0)T (t) = 0
X(0) = X() = 0.
v(, t) = X()T (t) = 0
'
Thus, Xn (0) = an = 0, and Xn (x) = bn sin 2 x.
'
'
Xn () = bn sin 2 = 0. Hence 2 = n, or = 2n2 .
= 2n2 ,

'

2 x.

Xn (x) = bn sin nx.


2

With these values of n , we solve T  + 2n2 T = 0 to get Tn (t) = cn e2n t.


Thus, the solution may be written in the form
v(x, t) =

2t

a
n e2n

sin nx.

n=1

From initial condition, we get


v(x, 0) =

a
n sin nx = sin x.

n=1

n = 0, n = 2, 3, . . ..
Thus, a
1 = 1, a
v(x, t) = e2t sin x.
73
Note that if the matrix was fully inseparable, we would have to nd eigenvalues and eigenvectors,
just as we did for the hyperbolic systems.

Partial Dierential Equations

Igor Yanovsky, 2005

321

We have

1 2t

sin x,
0 x , t > 0
ut = uxx 2 e
u(x, 0) = sin x,

u(0, t) = u(, t) = 0.
&
Let u(x, t) =
n=1 un (t) sin nx. Plugging this into the equation, we get

un (t) sin nx

n=1


n=1

1
n2 un (t) sin nx = e2t sin x.
2

For n = 1:
1
u1 (t) + u1 (t) = e2t .
2
Combining homogeneous and particular solution of the above equation, we obtain:
1
u1 (t) = e2t + c1 et .
2
For n = 2, 3, . . .:
un (t) + n2 un (t) = 0,
2

un (t) = cn en t .
Thus,
u(x, t) =




1
2
2
e2t + c1 et sin x +
cn en t sin nx = e2t sin x +
cn en t sin nx.
2
2

n=2

n=1

From initial condition, we get

u(x, 0) =


1
sin x +
cn sin nx = sin x.
2
n=1

Thus, c1 = 12 , cn = 0, n = 2, 3, . . ..
u(x, t) =

1
sin x (e2t + et ).
2

To prove the energy estimate (for general initial data)




2
2
[u (x, t) + v (x, t)] dx c
[u2 (x, 0) + v 2 (x, 0)] dx
x=0

x=0

for come constant c, we assume that


u(x, 0) =


n=1

an sin nx,

v(x, 0) =


n=1

bn sin nx.

Partial Dierential Equations

Igor Yanovsky, 2005

322

The general solutions are obtained by the same method as above

u(x, t) =


1 2t
2
e sin x +
cn en t sin nx,
2
n=1

v(x, t) =


2t

bn e2n

sin nx.

n=1



 1 2t
2  
2
2
n2 t
e sin x +
[u (x, t) + v (x, t)] dx =
cn e
sin nx +
bn e2n t sin nx dx
x=0
x=0 2
n=1
n=1




(b2n + a2n )
sin2 nx dx
[u2 (x, 0) + v 2 (x, 0)] dx.

n=1

x=0

x=0

Partial Dierential Equations

26

Igor Yanovsky, 2005

323

Problems: Eigenvalues of the Laplacian - Laplace

The 2D LAPLACE Equation (eigenvalues/eigenfuctions of the Laplacian).


Consider

in
uxx + uyy + u = 0
(26.1)
u(0, y) = 0 = u(a, y)
for 0 y b,

u(x, 0) = 0 = u(x, b)
for 0 x a.
Proof. We can solve this problem by separation of variables.
Let u(x, y) = X(x)Y (y), then substitution in the PDE gives X  Y + XY  + XY = 0.
X  Y 
+
+ = 0.
X
Y
Letting = 2 + 2 and using boundary conditions, we nd the equations for X and
Y:
X  + 2 X = 0

Y  + 2 Y = 0

X(0) = X(a) = 0

Y (0) = Y (b) = 0.

The solutions of these one-dimensional eigenvalue problems are


m
n
m =
n =
a
b
mx
ny
Yn (y) = sin
,
Xm (x) = sin
a
b
where m, n = 1, 2, . . . . Thus we obtain solutions of (26.1) of the form
mn = 2

m2
a2

n2
b2

umn (x, y) = sin

ny
mx
sin
,
a
b

where m, n = 1, 2, . . . .
Observe that the eigenvalues {mn }
m,n=1 are positive. The smallest eigenvalue 11
has only one eigenfunction u11 (x, y) = sin(x/a) sin(y/b); notice that u11 is positive
in . Other eigenvalues may correspond to more than one choice of m and n; for
example, in the case a = b we have nm = nm . For this , there are two linearly
independent eigenfunctions. However, for a particular value of there are at most
nitely many linearly independent eigenfunctions. Moreover,
 b a
 b a
ny
m x
n y
mx
sin
sin
sin
dx dy
umn (x, y) umn (x, y) dx dy =
sin
a
b
a
b
0
0
0
0

 
ny
n y
a b
ab
sin
sin
dy
if m = m and n = n
2 0
b
b
4
=
=
0 if m = m or n = n .
0
orthogonal. We could normalize each umn by a
In particular, the {umn } are pairwise

scalar multiple (i.e. multiply by 4/ab) so that ab/4 above becomes 1.
Let us change the notation somewhat so that each eigenvalue n corresponds to a
particular eigenfunction n (x). If we choose an orthonormal basis of eigenfunctions in
each eigenspace, we may arrange that {n }
n=1 is pairwise orthonormal:


1 if m = n
n (x)m(x) dx =
0 if m = n.

Partial Dierential Equations

Igor Yanovsky, 2005

In this notation, the eigenfunction expansion of f (x) dened on becomes


f (x)


n=1


an n (x),

where

an =

f (x)n (x) dx.

324

Partial Dierential Equations

Igor Yanovsky, 2005

325

Problem (S96, #4). Let D denote the rectangular


D = {(x, y) R2 : 0 < x < a, 0 < y < b}.
Find the eigenvalues of the following Dirichlet problem:
( + )u = 0

in D

u=0

on D.

Proof. The problem may be rewritten as

in

uxx + uyy + u = 0
u(0, y) = 0 = u(a, y)
for 0 y b,

u(x, 0) = 0 = u(x, b)
for 0 x a.
We may assume that the eigenvalues are positive, = 2 + 2 . Then,
mn = 2

m2
a2

n2
b2

umn (x, y) = sin

ny
mx
sin
,
a
b

m, n = 1, 2, . . ..

Problem (W04, #1). Consider the dierential equation:


2 u(x, y) 2 u(x, y)
+
+ u(x, y) = 0
x2
y 2

(26.2)

in the strip {(x, y), 0 < y < , < x < +} with boundary conditions
u(x, 0) = 0, u(x, ) = 0.

(26.3)

Find all bounded solutions of the boundary value problem (26.4), (26.5) when
a) = 0,

b) > 0,

c) < 0.

Proof. a) = 0. We have
uxx + uyy = 0.
Assume u(x, y) = X(x)Y (y), then substitution in the PDE gives
X Y + XY  = 0.
Boundary conditions give

u(x, 0) = X(x)Y (0) = 0
u(x, ) = X(x)Y () = 0

Y (0) = Y () = 0.

Method I: We have
Y 
X 
=
= c,
c > 0.
X
Y

cx +
bn sin cx.
From X  + cX = 0, we have Xn (x) = an cos

From Y  cY = 0, we have Yn (y) = cn e cy + dn e cy .


Y (0) = cn + dn = 0 cn = dn .

Partial Dierential Equations

Igor Yanovsky, 2005

Y () = cn e c cn e c = 0
u(x, y) = X(x)Y (y) = 0.
Method II: We have
X 
Y 
=
= c,
X
Y

Yn (y) = 0.

cn = 0

c > 0.

From X  cX = 0, we have Xn (x) = an e cx + bn e cx .


Since we look for bounded solutions for < x < , an = bn = 0

From Y  + cY = 0, we have Yn (y) = cn cos cy + dn sin cy.


Y (0) = cn = 0,

c = n c = n2 .
Y () = dn sin c = 0
Yn (y) = dn sin nx = 0.
u(x, y) = X(x)Y (y) = 0.

Xn (x) = 0.

b) > 0. We have
X  Y 
+
+ = 0.
X
Y
Letting = 2 + 2 , and using boundary conditions for Y , we nd the equations:
X  + 2 X = 0

Y  + 2 Y = 0
Y (0) = Y () = 0.

The solutions of these one-dimensional eigenvalue problems are


Xm (x) = am cos m x + bm sin m x.
n = n, Yn (y) = dn sin ny, where m, n = 1, 2, . . ..
u(x, y) =

326

umn (x, y) =

m,n=1

(am cos m x + bm sin m x) sin ny.

m,n=1

c) < 0. We have
uxx + uyy + u = 0,
u(x, 0) = 0, u(x, ) = 0.
u 0 is the solution to this equation. We will show that this solution is unique.
Let u1 and u2 be two solutions, and consider w = u1 u2 . Then,
w + w = 0,
w(x, 0) = 0, w(x, ) = 0.
Multiply the equation by w and integrate:
ww + w 2 = 0,

ww dx + w 2 dx = 0,




w
ds |w|2 dx + w 2 dx = 0,
w
n


 
=0


2
|w| dx = w 2 dx .
   
 


Partial Dierential Equations


Thus, w 0 and the solution u(x, y) 0 is unique.

Igor Yanovsky, 2005

327

Partial Dierential Equations

Igor Yanovsky, 2005

328

Problem (F95, #5). Find all bounded solutions


for the following boundary value problem in the strip
0 < x < a, < y < ,
( + k2 )u = 0
u(0, y) = 0,

(k = Const > 0),


ux (a, y) = 0.

In particular, show that when ak ,


the only bounded solution to this problem is u 0.
Proof. Let u(x, y) = X(x)Y (y), then we have X  Y + XY  + k2 XY = 0.
X  Y 
+
+ k2 = 0.
X
Y
Letting k2 = 2 + 2 and using boundary conditions, we nd:
X  + 2 X = 0,

Y  + 2 Y = 0.

X(0) = X (a) = 0.
The solutions of these one-dimensional eigenvalue problems are
(m 12 )
,
a
(m 12 )x
Yn (y) = cn cos n y + dn sin n y,
Xm (x) = sin
a
where m, n = 1, 2, . . . . Thus we obtain solutions of the form
m =

2
kmn


=

(m 12 )
a

2



(m 12 )x
cn cos n y+dn sin n y ,
umn (x, y) = sin
a

+n2 ,

where m, n = 1, 2, . . . .
u(x, y) =

umn (x, y) =

m,n=1

sin

m,n=1



(m 12 )x
cn cos n y + dn sin n y .
a

We can take an alternate approach and prove the second part of the question. We
have
X Y + XY  + k2 XY = 0,
X 
Y 
=
+ k 2 = c2 .

Y
X
We obtain Yn (y) = cn cos cy + dn sin cy.

The second equation gives

X  + k2 X = c2 X,
X  + (k2 c2 )X = 0,

Xm (x) = am e

c2 k2 x

+ bm e

c2 k2 x

Thus, Xm (x) is bounded only if k2 c2 > 0, (if k2 c2 = 0, X  = 0, and Xm (x) =


am x + bm , BCs give Xm (x) = x, unbounded), in which case


Xm (x) = am cos k2 c2 x + bm sin k2 c2 x.

Partial Dierential Equations

Igor Yanovsky, 2005

329

Boundary conditions give Xm (0) = am = 0.





Xm
(x) = bm k2 c2 cos k2 c2 x,



(a) = bm k2 c2 cos k2 c2 a = 0,
Xm


m = 1, 2, . . .,
k2 c2 a = m ,
2

1 2
m
,
k 2 c2 =
a
2

1 2
+ c2 ,
m
k2 =
a
2

1 2
a2 k2 > 2 m
,
2

1
,
m = 1, 2, . . ..
ak > m
2
Thus, bounded solutions exist only when ak > 2 .

Problem (S90, #2). Show that the boundary value problem


2 u(x, y) 2 u(x, y)
+
+ k2 u(x, y) = 0,
x2
y 2

(26.4)

where < x < +, 0 < y < , k > 0 is a constant,


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

(26.5)

has a bounded solution if and only if k 1.

Proof. We have
uxx + uyy + k2 u = 0,
X Y + XY  + k2 XY = 0,
Y 
X 
=
+ k 2 = c2 .

X
Y
We obtain Xm (x) = am cos cx + bm sin cx.

The second equation gives

Y  + k2 Y = c2 Y,
Y  + (k2 c2 )Y = 0,

c2 k2 y

Yn (y) = cn e

+ dn e

c2 k2 y

Thus, Yn (y) is bounded only if k2 c2 > 0, (if k2 c2 = 0, Y  = 0, and Yn (y) = cn y+dn ,


BCs give Y 0), in which case


Yn (y) = cn cos k2 c2 y + dn sin k2 c2 y.
Boundary conditions
give Yn (0) = cn = 0.

k 2 c2 = n
Yn () = dn sin k2 c2 = 0
2
2
2
k = n + c , n = 1, 2, . . .. Hence, k > n, n = 1, 2, . . ..

k2 c2 = n2

Partial Dierential Equations

Igor Yanovsky, 2005

Thus, bounded solutions exist if k 1.


Note: If k = 1, then c = 0, which gives trivial solutions for Yn (y).
u(x, y) =


m,n=1

Xm (x)Yn (y) =


m,n=1

sin ny Xm (x).

330

Partial Dierential Equations

Igor Yanovsky, 2005

331

McOwen, 4.4 #7; 266B Ralston Hw. Show that the boundary value problem

a(x)u + b(x)u = u
in
u=0
on
has only trivial solution with 0, when b(x) 0 and a(x) > 0 in .
Proof. Multiplying the equation by u and integrating over , we get



2
u au dx +
bu dx =
u2 dx.

Since (uau) = u au + a|u|2 , we have






2
2
(uau) dx +
a|u| dx +
bu dx = u2 dx.

(26.6)

Using divergence theorem, we obtain






u
2
2
ds +

u a
a|u| dx +
bu dx =
u2 dx,
n

=0



2
2
a |u| dx +
b u dx = 

u2 dx,



>0

Thus, u = 0 in , and u is constant. Since u = 0 on , u 0 on .


Similar Problem I: Note that this argument also works with Neumann B.C.:

a(x)u + b(x)u = u
in
u/n = 0
on
Using divergence theorem, (26.6) becomes




u
2
2
ua
ds +
a|u| dx +
bu dx = u2 dx,
n


=0



2
2
a
|u|
dx
+
b
u
dx
=

u2 dx.




>0

Thus, u = 0, and u = const on . Hence, we now have




2
b
u
dx
=

u2 dx,



which implies = 0. This gives the useful information that for the eigenvalue problem74

a(x)u + b(x)u = u
u/n = 0,
= 0 is an eigenvalue, its eigenspace is the set of constants, and all other s are
positive.
74

In Ralstons Hw#7 solutions, there is no - sign in front of a(x)u below, which is probably a
typo.

Partial Dierential Equations

Igor Yanovsky, 2005

332

Similar Problem II: If 0, we show that the only solution to the problem below
is the trivial solution.

u + u = 0
in
u=0
on


u2 dx = 0,

uu dx +


u
u
ds

n

=0

|u| dx + 

u2 dx = 0.

Thus, u = 0 in , and u is constant. Since u = 0 on , u 0 on .

Partial Dierential Equations

27

Igor Yanovsky, 2005

333

Problems: Eigenvalues of the Laplacian - Poisson

The ND POISSON Equation (eigenvalues/eigenfunctions of the Laplacian).


Suppose we want to nd the eigenfunction expansion of the solution of
u = f

in

u=0

on ,

when f has the expansion in the orthonormal Dirichlet eigenfunctions n :


f (x)


an n (x),

where

an =

n=1

&

Proof. Writing u =

n cn n =

n=1

f (x)n (x) dx.

cn n and inserting into u = f , we get

an n (x).

n=1

Thus, cn = an /n , and
u(x) =


an n (x)
n=1

The 1D POISSON Equation (eigenvalues/eigenfunctions of the Laplacian).


For the boundary value problem
u = f (x)
u(0) = 0,

u(L) = 0,

the related eigenvalue problem is


 =
(0) = 0,

(L) = 0.

The eigenvalues are n = (n/L)2 , and the corresponding eigenfunctions are sin(nx/L),
n = 1, 2, . . .. &
&
Writing u =
cn n = cn sin(nx/L) and inserting into u = f , we get


0

n=1

L
n=1

cn

n 2
L

cn

n 2
L

sin

nx
L

= f (x),

 L
mx
nx
mx
sin
dx =
dx,
sin
f (x) sin
L
L
L
0
 L

n 2 L
nx
=
dx,
f (x) sin
cn
L
2
L
0
L
2 0 f (x) sin(nx/L) dx
.
cn =
L
(n/L)2

Partial Dierential Equations




u(x) =

u=

Igor Yanovsky, 2005

cn sin(nx/L) =
L
n=1

L
0

334

f () sin(nx/L) sin(n/L) d
,
(n/L)2

2  sin(nx/L) sin(n/L)
f ()
d.
L
(n/L)2
n=1



= G(x,)

See similar, but more complicated, problem in Sturm-Liouville Problems (S92, #2(c)).

Partial Dierential Equations

Igor Yanovsky, 2005

335

Example: Eigenfunction Expansion of the GREENs Function.


Suppose we x x and attempt to expand the Greens function G(x, y) in the orthonormal
eigenfunctions n (y):



an (x)n (y),
where
an (x) =
G(x, z)n(z) dz.
G(x, y)

n=1

Proof. We can rewrite u + u = 0 in , u = 0 on , as an integral equation



u(x) + G(x, y)u(y) dy = 0.



m=1

&

cn n (x). Plugging this into , we get


 


cm m (x) +
an (x)n (y)
cm m (y) dy = 0,

Suppose, u(x) =

n=1

cm m (x) +

m=1


n=1

an (x)


cm

m=1

m=1

n (y)m(y) dy = 0,

cn n (x) +

n=1

an (x)cn = 0,

n=1



cn n (x) + an (x) = 0,

n=1

an (x) =
Thus,
G(x, y)


n=1

75

n (x)n (y)
.
n

See the section: ODE - Integral Equations.

n (x)
.
n

75

Partial Dierential Equations

Igor Yanovsky, 2005

336

The 2D POISSON Equation (eigenvalues/eigenfunctions of the Laplacian).


Solve the boundary value problem

for 0 < x < a, 0 < y < b

uxx + uyy = f (x, y)


(27.1)
u(0, y) = 0 = u(a, y)
for 0 y b,

u(x, 0) = 0 = u(x, b)
for 0 x a.
f (x, y) C 2 , f (x, y) = 0 if x = 0, x = a, y = 0, y = b,

ny
mx
2 

sin
.
cmn sin
f (x, y) =
a
b
ab m,n=1

Proof. First, we nd eigenvalues/eigenfunctions of the Laplacian.

in

uxx + uyy + u = 0
u(0, y) = 0 = u(a, y)
for 0 y b,

u(x, 0) = 0 = u(x, b)
for 0 x a.
Let u(x, y) = X(x)Y (y), then substitution in the PDE gives X  Y + XY  + XY = 0.
X  Y 
+
+ = 0.
X
Y
Letting = 2 + 2 and using boundary conditions, we nd the equations for X and
Y:
X  + 2 X = 0

Y  + 2 Y = 0

X(0) = X(a) = 0

Y (0) = Y (b) = 0.

The solutions of these one-dimensional eigenvalue problems are


m
n
n =
m =
a
b
mx
ny
Yn (y) = sin
,
Xm (x) = sin
a
b
where m, n = 1, 2, . . .. Thus we obtain eigenvalues and normalized eigenfunctions of
the Laplacian:
mn = 2

m2
a2

n2
b2

mx
ny
2
sin
,
mn (x, y) = sin
a
b
ab

where m, n = 1, 2, . . . . Note that


f (x, y) =

cmn mn .

m,n=1

Second, writing u(x, y) =

&

cmn mn and inserting into u = f , we get

mn cmn mn (x, y) =

m,n=1


m,n=1

.
Thus, cmn = cmn
mn
u(x, y) =


cmn
mn (x, y),

mn
n=1

cmn mn (x, y).

Partial Dierential Equations

Igor Yanovsky, 2005

with mn , mn (x) given above, and cmn given by


 b a
 b a 

f (x, y)mn dx dy =
cm n m n mn dx dy = cmn .
0

0 m ,n =1

337

Partial Dierential Equations

28

Igor Yanovsky, 2005

338

Problems: Eigenvalues of the Laplacian - Wave

In the section on the wave equation, we considered an initial boundary value problem
for the one-dimensional wave equation on an interval, and we found that the solution could be obtained using Fourier series. If we replace the Fourier series by an
expansion in eigenfunctions, we can consider an initial/boundary value problem for the
n-dimensional wave equation.
The ND WAVE Equation (eigenvalues/eigenfunctions of the Laplacian).
Consider

for x , t > 0

utt = u
u(x, 0) = g(x), ut (x, 0) = h(x)
for x

u(x, t) = 0
for x , t > 0.
Proof. For g, h C 2 () with g = h = 0 on , we have eigenfunction expansions
g(x) =

an n (x)

and

h(x) =

n=1

bn n (x).

n=1

Assume the solution u(x, t) may be expanded in the eigenfunctions with coecients
&
depending on t: u(x, t) =
n=1 un (t)n (x). This implies

un (t)n (x) =

n=1
un (t)

n un (t)n (x),

n=1

+ n un (t) = 0

for each n.

Since n > 0, this ordinary dierential equation has general solution




Thus,
un (t) = An cos n t + Bn sin n t.



 
An cos n t + Bn sin n t n (x),
u(x, t) =
ut (x, t) =
u(x, 0) =
ut (x, 0) =

n=1



 

n An sin n t + n Bn cos n t n (x),

n=1

An n (x)
n=1



= g(x),

n Bn n (x) = h(x).

n=1

Comparing with , we obtain


An = an ,

bn
Bn = .
n

Thus, the solution is given by


u(x, t) =



 

bn
an cos n t + sin n t n (x),
n
n=1

Partial Dierential Equations


with


an =
bn =

g(x)n(x) dx,
h(x)n (x) dx.

Igor Yanovsky, 2005

339

Partial Dierential Equations

Igor Yanovsky, 2005

340

The 2D WAVE Equation (eigenvalues/eigenfunctions of the Laplacian).


Let = (0, a) (0, b) and consider

for x , t > 0
utt = uxx + uyy
(28.1)
for x
u(x, 0) = g(x), ut (x, 0) = h(x)

u(x, t) = 0
for x , t > 0.

Proof. First, we nd eigenvalues/eigenfunctions of the Laplacian.

in

uxx + uyy + u = 0
u(0, y) = 0 = u(a, y)
for 0 y b,

u(x, 0) = 0 = u(x, b)
for 0 x a.
Let u(x, y) = X(x)Y (y), then substitution in the PDE gives X  Y + XY  + XY = 0.
X  Y 
+
+ = 0.
X
Y
Letting = 2 + 2 and using boundary conditions, we nd the equations for X and
Y:
X  + 2 X = 0

Y  + 2 Y = 0

X(0) = X(a) = 0

Y (0) = Y (b) = 0.

The solutions of these one-dimensional eigenvalue problems are


m
n
n =
m =
a
b
mx
ny
Yn (y) = sin
,
Xm (x) = sin
a
b
where m, n = 1, 2, . . .. Thus we obtain eigenvalues and normalized eigenfunctions of
the Laplacian:
mn = 2

m2
a2

n2
b2

mx
ny
2
sin
,
mn (x, y) = sin
a
b
ab

where m, n = 1, 2, . . . .
Second, we solve the Wave Equation (28.1) using the space eigenfunctions.
For g, h C 2 () with g = h = 0 on , we have eigenfunction expansions 76
g(x) =

an n (x)

and

h(x) =

n=1

Assume u(x, t) =

n=1

&

n=1

un (t)n (x).

un (t) + n un (t) = 0


76

This implies

for each n.

In 2D, n is really mn , and x is (x, y).

bn n (x).

Partial Dierential Equations

Igor Yanovsky, 2005

Since n > 0, this ordinary dierential equation has general solution




un (t) = An cos n t + Bn sin n t.
Thus,



 

An cos n t + Bn sin n t n (x),
u(x, t) =
ut (x, t) =
u(x, 0) =
ut (x, 0) =

n=1



 

n An sin n t + n Bn cos n t n (x),

n=1

An n (x)
n=1



= g(x),

n Bn n (x) = h(x).

n=1

Comparing with , we obtain


An = an ,

bn
Bn = .
n

Thus, the solution is given by






bmn
u(x, t) =
sin mn t mn (x),
amn cos mn t +
mn
m,n=1

with mn , mn (x) given above, and



g(x)mn (x) dx,
amn =

h(x)mn (x) dx.
bmn =

341

Partial Dierential Equations

Igor Yanovsky, 2005

342

McOwen, 4.4 #3; 266B Ralston Hw. Consider the initial-boundary value problem

for x , t > 0

utt = u + f (x, t)
u(x, t) = 0
for x , t > 0

for x .
u(x, 0) = 0, ut (x, 0) = 0
Use Duhamels principle and an expansion of f in eigenfunctions to obtain a (formal)
solution.
Proof. a) We expand u in terms of the Dirichlet eigenfunctions of Laplacian in
.
n + n n = 0

in ,

n = 0

an (t)n (x),

an (t) =

fn (t)n (x),

fn (t) =

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


n=1




n=1

an (t)


=

on .

n (x)u(x, t) dx.
n (x)f (x, t) dx.


n (x)utt dx =
n (u + f ) dx =
n u dx +
n f dx




n u dx +
n f dx = n
n u dx +
n f dx = n an (t) + fn (t).

  


=

fn


an (0) =
an (0) =

n (x)u(x, 0) dx = 0.
n (x)ut(x, 0) dx = 0.

77

Thus, we have an ODE which is converted and solved by Duhamels principle:



n + n an = 0
 t
a
an + n an = fn (t)

a
(t)
=
an (t s, s) ds.
an (0) = 0
an (0, s) = 0
n

0


an (0) = 0
an (0, s) = fn (s)

With the anzats a


n (t, s) = c1 cos n t + c2 sin n t, we get c1 = 0, c2 = fn (s)/ n ,
or

sin n t
.
an (t, s) = fn (s)
n
Duhamels principle gives

 t
 t
sin( n (t s))

a
n (t s, s) ds =
fn (s)
ds.
an (t) =
n
0
0




n (x) t

u(x, t) =
fn (s) sin( n (t s)) ds.
n 0
n=1
  u


n
We used Greens formula: n n
u
n ds = (n u n u) dx.
On , u = 0; n = 0 since eigenfunctions are Dirichlet.
77

Partial Dierential Equations

Igor Yanovsky, 2005

343

Problem (F90, #3). Consider the initial-boundary value problem

0 x , t 0

utt = a(t)uxx + f (x, t)


u(0, t) = u(, t) = 0
t0

0 x ,
u(x, 0) = g(x), ut (x, 0) = h(x)
where the coecient a(t) = 0.
a) Express (formally) the solution of this problem by the method of eigenfunction expansions.
b) Show that this problem is not well-posed if a 1.
Hint: Take f = 0 and prove that the solution does not depend continuously on the
initial data g, h.
Proof. a) We expand u in terms of the Dirichlet eigenfunctions of Laplacian in
.
n xx + n n = 0 in ,

n (0) = n () = 0.

That gives us the eigenvalues and eigenfunctions of the Laplacian: n = n2 , n (x) =


sin nx.
Assume



un (t)n (x),
un (t) =
n (x)u(x, t) dx.
u(x, t) =
f (x, t) =
g(x) =
h(x) =

n=1


fn (t)n (x),

fn (t) =

n=1

gn =

n (x)f (x, t) dx.

n (x)g(x) dx.


hn n (x),

hn =

n=1

un (t)

gn n (x),

n=1

n (x)h(x) dx.

n (x)utt dx =
n (a(t)uxx + f ) dx = a(t)
n uxx dx +
n f dx





n xxu dx +
n f dx = n a(t)
n u dx +
n f dx
= a(t)

  

fn

= n a(t)un (t) + fn (t).




n (x)u(x, 0) dx =
n (x)g(x) dx = gn .
un (0) =


n (x)ut(x, 0) dx =
n (x)h(x) dx = hn .
un (0) =

Thus, we have an ODE which is converted and solved by Duhamels principle:



un + n a(t)un = fn (t)

un (0) = gn


un (0) = hn .

Partial Dierential Equations

Igor Yanovsky, 2005

344

Note: The initial data is not 0; therefore, the Duhamels principle is not applicable.
Also, the ODE is not linear in t, and its solution is not obvious. Thus,
u(x, t) =

un (t)n (x),

n=1

where un (t) are solutions of .

Partial Dierential Equations

Igor Yanovsky, 2005

345

b) Assume we have two solutions, u1 and u2 , to the PDE:

u2tt + u2xx = 0,
u1tt + u1xx = 0,
u1 (0, t) = u1 (, t) = 0,
u2 (0, t) = u2 (, t) = 0,

u1 (x, 0) = g1 (x), u1t (x, 0) = h1 (x);


u2 (x, 0) = g2 (x), u2t (x, 0) = h2 (x).
Note that the equation is elliptic, and therefore, the maximum principle holds.
In order to prove that the solution does not depend continuously on the initial data
g, h, we need to show that one of the following conditions holds:
max |u1 u2 | > max |g1 g2 |,

max |ut1 ut2 | > max |h1 h2 |.

That is, the dierence of the two solutions is not bounded by the dierence of initial
data.
By the method of separation of variables, we may obtain
u(x, t) =
u(x, 0) =
ut (x, 0) =


n=1


n=1

(an cos nt + bn sin nt) sin nx,


an sin nx = g(x),
nbn sin nx = h(x).

n=1

Not complete.

We also know that for elliptic equations, and for Laplace equation in particular, the
value of the function u has to be prescribed on the entire boundary, i.e. u = g on
, which is not the case here, making the problem under-determined. Also, ut is
prescribed on one of the boundaries, making the problem overdetermined.

Partial Dierential Equations

29

Igor Yanovsky, 2005

346

Problems: Eigenvalues of the Laplacian - Heat

The ND HEAT Equation (eigenvalues/eigenfunctions of the Laplacian).


Consider the initial value problem with homogeneous Dirichlet condition:

for x , t > 0
ut = u
u(x, 0) = g(x)
for x

u(x, t) = 0
for x , t > 0.
Proof. For g C 2 () with g = 0 on , we have eigenfunction expansion
g(x) =

an n (x)

n=1

Assume the solution u(x, t)&may be expanded in the eigenfunctions with coecients
depending on t: u(x, t) =
n=1 un (t)n (x). This implies

un (t)n (x)

n=1
un (t)

= n

un (t)n (x),

n=1

+ n un (t) = 0,

which has the general solution

Thus,
un (t) = An en t.


An en t n (x),
u(x, t) =
u(x, 0) =

n=1

An n (x) = g(x).

n=1

Comparing with , we obtain An = an . Thus, the solution is given by


u(x, t) =

an en tn (x),

n=1


with

an =

g(x)n (x) dx.

Also
u(x, t) =

n t

an e

n=1

 

n=1

n (x) =




n=1


g(y)n(y) dy en t n (x)

en t n (x)n (y) g(y) dy

K(x,y,t),



heat kernel

Partial Dierential Equations

Igor Yanovsky, 2005

347

The 2D HEAT Equation (eigenvalues/eigenfunctions of the Laplacian).


Let = (0, a) (0, b) and consider

for x , t > 0

ut = uxx + uyy
(29.1)
u(x, 0) = g(x)
for x

u(x, t) = 0
for x , t > 0.

Proof. First, we nd eigenvalues/eigenfunctions of the Laplacian.

in
uxx + uyy + u = 0
u(0, y) = 0 = u(a, y)
for 0 y b,

u(x, 0) = 0 = u(x, b)
for 0 x a.
Let u(x, y) = X(x)Y (y), then substitution in the PDE gives X  Y + XY  + XY = 0.
X  Y 
+
+ = 0.
X
Y
Letting = 2 + 2 and using boundary conditions, we nd the equations for X and
Y:
X  + 2 X = 0

Y  + 2 Y = 0

X(0) = X(a) = 0

Y (0) = Y (b) = 0.

The solutions of these one-dimensional eigenvalue problems are


m
n
n =
m =
a
b
mx
ny
Yn (y) = sin
,
Xm (x) = sin
a
b
where m, n = 1, 2, . . .. Thus we obtain eigenvalues and normalized eigenfunctions of
the Laplacian:
mn = 2

m2
a2

n2
b2

mx
ny
2
sin
,
mn (x, y) = sin
a
b
ab

where m, n = 1, 2, . . . .
Second, we solve the Heat Equation (29.1) using the space eigenfunctions.
For g C 2 () with g = 0 on , we have eigenfunction expansion
g(x) =

an n (x).

n=1

Assume u(x, t) =

&

n=1

un (t)n (x).

un (t) + n un (t) = 0,

This implies

which has the general solution

Thus,
un (t) = An en t.


An en t n (x),
u(x, t) =
u(x, 0) =

n=1


n=1

An n (x) = g(x).

Partial Dierential Equations

Igor Yanovsky, 2005

Comparing with , we obtain An = an . Thus, the solution is given by


u(x, t) =

amn emn t mn (x),

m,n=1

with mn , mn given above and amn =

g(x)mn (x) dx.

348

Partial Dierential Equations

Igor Yanovsky, 2005

349

Problem (S91, #2). Consider the heat equation


ut = uxx + uyy
on the square = {0 x 2, 0 y 2} with
periodic boundary conditions and with initial data
u(0, x, y) = f (x, y).
a) Find the solution using separation of variables.
Proof. First, we nd eigenvalues/eigenfunctions of the Laplacian.

in

uxx + uyy + u = 0
u(0, y) = u(2, y)
for 0 y 2,

u(x, 0) = u(x, 2)
for 0 x 2.
Let u(x, y) = X(x)Y (y), then substitution in the PDE gives X  Y + XY  + XY = 0.
X  Y 
+
+ = 0.
X
Y
Letting = 2 + 2 and using periodic BCs, we nd the equations for X and Y :
X  + 2 X = 0

Y  + 2 Y = 0

X(0) = X(2)

Y (0) = Y (2).

The solutions of these one-dimensional eigenvalue problems are


m = m

n = n
imx

Yn (y) = einy ,

Xm (x) = e

where m, n = . . . , 2, 1, 0, 1, 2, . . .. Thus we obtain eigenvalues and normalized eigenfunctions of the Laplacian:


mn = m2 + n2

mn (x, y) = eimx einy ,

where m, n = . . . , 2, 1, 0, 1, 2, . . ..
Second, we solve the
using the space eigenfunctions.
&Heat Equationimx
Assume u(x, y, t) = m,n= umn (t)e einy . This implies
umn (t) + (m2 + n2 )umn (t) = 0,
2 +n2 )t

un (t) = cmn e(m


u(x, y, t) =


m,n=

which has the general solution

Thus,
2 +n2 )t

cmn e(m

eimx einy .

Partial Dierential Equations

Igor Yanovsky, 2005

u(x, y, 0) =


cmn eimx einy = f (x, y),

m,n=

f (x, y)eimxeiny dxdy =


0

350

m,n=

2 


= 2
1
4 2

cmn einy ein y dy = 4 2 cmn .

cmn =

cmn eimx einy eim x ein y dxdy


0

n=
2  2
0

f (x, y)eimx einy dxdy = fmn .

Partial Dierential Equations

b) Show that the integral

Igor Yanovsky, 2005

u2 (x, y, t) dxdy is decreasing in t, if f is not constant.

Proof. We have
ut = uxx + uyy
Multiply the equation by u and integrate:
uut = uu,
1 d 2
u = uu,

 2 dt
1 d
2
u dxdy =
uu dxdy =
2 dt


=

351

u
u
ds
n

 
=0, (periodic BC)

|u|2 dxdy

|u|2 dxdy 0.

Equality is obtained only when u = 0 u = constant f = constant.


If f is not constant, u2 dxdy is decreasing in t.

Partial Dierential Equations

Igor Yanovsky, 2005

352

Problem (F98, #3). Consider the eigenvalue problem


d2
+ = 0,
dx2
d
(0) = 0,
(0)
dx

(1) +

d
(1) = 0.
dx

a) Show that all eigenvalues are positive.


b) Show that there exist a sequence of eigenvalues = n , each of which satises

2
.
tan =
1
c) Solve the following initial-boundary value problem on 0 < x < 1, t > 0
u
2u
=
,
t
x2
u
(0, t) = 0,
u(0, t)
x
u(x, 0) = f (x).

u(1, t) +

u
(1, t) = 0,
x

You may call the relevant eigenfunctions n (x) and assume that they are known.
Proof. a) If = 0, the ODE reduces to  = 0. Try (x) = Ax + B.
From the rst boundary condition,
(0)  (0) = 0 = B A

B = A.

Thus, the solution takes the form (x) = Ax+A. The second boundary condition gives
(1) +  (1) = 0 = 3A

A = B = 0.

Thus the only solution is 0, which is not an eigenfunction, and 0 not an eigenvalue.


If < 0, try (x) = esx , which gives s = = R.


Hence, the family of solutions is (x) = Aex + Bex . Also,  (x) = Aex Bex .
The boundary conditions give
(0)  (0) = 0 = A + B A + B = A(1 ) + B(1 + ),

(29.2)

(1) +  (1) = 0 = Ae + Be + Ae Be = Ae (1 + ) + Be (1 ). (29.3)


From (29.2) and (29.3) we get
A
1+
=
1
B

and

From (29.2), =

1+
B
= e2 ,
1
A
A+B
AB

or

A
= e .
B

A+B
A
= e BA , which has no solutions. 
B

gives s = i and the family of solutions takes

and thus,

b) Since > 0, the anzats = esx


the form

(x) = A sin(x ) + B cos(x ).

Then,  (x) = A cos(x ) B sin(x ). The rst boundary condition gives

B = A .
(0)  (0) = 0 = B A

Partial Dierential Equations

Igor Yanovsky, 2005

353

Hence, (x) = A sin(x ) + A cos(x ). The second boundary condition gives

(1) +  (1) = 0 = A sin( ) + A cos( ) + A cos( ) A sin( )


= A (1 ) sin( ) + 2 cos( )
A = 0 (since A = 0 implies B = 0 and = 0, which is notan eigenfunction). Therefore,

(1 ) sin( ) = 2 cos( ), and thus tan( ) = 21 .


c) We may assume that the eigenvalues/eigenfunctins of the Laplacian, n and n (x),
are known. We solve the Heat Equation using the space eigenfunctions.

ut = uxx ,
u(1, t) + ux (1, t) = 0,
u(0, t) ux(0, t) = 0,

u(x, 0) = f (x).
For f , we have an eigenfunction expansion
f (x) =

an n (x).

n=1

Assume u(x, t) =
un (t)

&

n=1

un (t)n (x).

+ n un (t) = 0,

This implies

which has the general solution

Thus,
un (t) = An en t.


An en t n (x),
u(x, t) =
u(x, 0) =

n=1

An n (x) = f (x).

n=1

Comparing with , we have An = an . Thus, the solution is given by


u(x, t) =

an en tn (x),

n=1

with

an =

f (x)n (x) dx.

Partial Dierential Equations

Igor Yanovsky, 2005

354

Problem (W03, #3); 266B Ralston Hw. Let be a smooth domain in three
dimensions and consider the initial-boundary value problem for the heat equation

for x , t > 0

ut = u + f (x)
u/n = 0
for x , t > 0

u(x, 0) = g(x)
for x ,
in which f and g are known smooth functions with
for x .

g/n = 0

a) Find an approximate formula for u as t .


Proof. We expand u in terms of the Neumann eigenfunctions of Laplacian in .
n + n n = 0

n
= 0 on .
n

in ,

Note that here 1 = 0 and 1 is the constant V 1/2 , where V is the volume of .
Assume



an (t)n (x),
an (t) =
n (x)u(x, t) dx.
u(x, t) =
f (x) =
g(x) =

n=1

fn n (x),

n=1

fn =


gn n (x),

gn =

n=1

an (t)


=

n (x)f (x) dx.


n (x)g(x) dx.

n (x)ut dx =
n (u + f ) dx =
n u dx +
n f dx




n u dx +
n f dx = n
n u dx +
n f dx = n an + fn .

  


=


an (0) =
78

fn


n (x)u(x, 0) dx =

n g dx = gn .

Thus, we solve the ODE:



an + n an = fn
an (0) = gn .

For n = 1, 1 = 0, and we obtain a1 (t) = f1 t + g1 .


For n 2, the homogeneous solution is anh = cen t . The anzats for a particular
solution is anp = c1 t + c2 , which gives c1 = 0 and c2 = fn /n . Using the initial
condition, we obtain

fn n t fn
+
.
e
an (t) = gn
n
n
  u


n
We used Greens formula: n n
u
n ds = (n u n u) dx.
n
= 0;
= 0 since eigenfunctions are Neumann.
On , u
n
n
78

Partial Dierential Equations

u(x, t) = (f1 t + g1 )1 (x) +

Igor Yanovsky, 2005




gn

n=2


If f1 = 0
If f1 = 0


f (x) dx = 0 ,

f (x) dx = 0 ,

fn n t fn
+
e
n (x).
n
n

lim u(x, t) = g1 1 +

lim u(x, t) f1 1 t.


fn n
n=2

355

Partial Dierential Equations

Igor Yanovsky, 2005

b) If g 0 and f > 0, show that u > 0 for all t > 0.

356

Partial Dierential Equations

Igor Yanovsky, 2005

357

Problem (S97, #2). a) Consider the eigenvalue problem for the Laplace operator
 in R2 with zero Neumann boundary condition

in
uxx + uyy + u = 0
u
on .
n = 0
Prove that 0 = 0 is the lowest eigenvalue and that it is simple.
b) Assume that the eigenfunctions n (x, y) of the problem in (a) form a complete
orthogonal system, and that f (x, y) has a uniformly convergent expansion
f (x, y) =

fn n (x, y).

n=0

Solve the initial value problem


ut = u + f (x, y)
subject to initial and boundary conditions
u(x, y, 0) = 0,

u
u| = 0.
n

What is the behavior of u(x, y, t) as t ?


c) Consider the problem with Neumann boundary conditions

in
vxx + vyy + f (x, y) = 0
v
on .
n v = 0
When does a solution exist? Find this solution, and nd its relation with the behavior
of lim u(x, y, t) in (b) as t .

Proof. a) Suppose this eigenvalue problem did have a solution u with 0.


Multiplying u + u = 0 by u and integrating over , we get


uu dx + u2 dx = 0,




u
u
ds
|u|2 dx + u2 dx = 0,
n


=0


2
|u| dx = 

u2 dx,

Thus, u = 0 in , and u is constant in . Hence, we now have



u2 dx.
0 = 

For nontrivial u, we have = 0. For this eigenvalue problem, = 0 is an eigenvalue,


its eigenspace is the set of constants, and all other s are positive.

Partial Dierential Equations

Igor Yanovsky, 2005

b) We expand u in terms of the Neumann eigenfunctions of Laplacian in .


n + n n = 0
u(x, y, t) =

an (t)n (x, y),

n=1

an (t)

n (x, y)ut dx =
n (u + f ) dx =
n u dx +
n f dx





=
n u dx +
n f dx = n
n u dx +
n f dx = n an + fn .

  
=

fn


an (0) =
80

79

n
= 0 on .
n

an (t) =
n (x, y)u(x, y, t) dx.

in ,

358

n (x, y)u(x, y, 0) dx = 0.

Thus, we solve the ODE:



an + n an = fn
an (0) = 0.

For n = 1, 1 = 0, and we obtain a1 (t) = f1 t.


For n 2, the homogeneous solution is anh = cen t . The anzats for a particular
solution is anp = c1 t + c2 , which gives c1 = 0 and c2 = fn /n . Using the initial
condition, we obtain
an (t) =

fn n t fn
e
+
.
n
n

u(x, t) = f1 1 t +


n=2


If f1 = 0
If f1 = 0

fn n t fn
e
+
n (x).
n
n


f (x) dx = 0 ,

lim u(x, t) =


f (x) dx = 0 ,


fn n
n=2

lim u(x, t) f1 1 t.

c) Integrate v + f (x, y) = 0 over :






1
f dx =
v dx = v dx =

where we used 1 divergence theorem and


solution exists only if

f dx = 0.

v
ds =2 0,
n

Neumann boundary conditions. Thus, the

79

We use dx dy dx.
  u


n
We used Greens formula: n n
u
ds = (n u n u) dx.
n
n
= 0;
= 0 since eigenfunctions are Neumann.
On , u
n
n
80

Partial Dierential Equations

Igor Yanovsky, 2005

359

&
&
Assume v(x, y) =
n=0 an n (x, y). Since we have f (x, y) =
n=0 fn n (x, y), we
obtain



n an n +
fn n = 0,

n=0

n=0

n an n + fn n = 0,
fn
.
an =
n

v(x, y) =

&

fn
n=0 ( n )n (x, y).

Partial Dierential Equations

29.1

Igor Yanovsky, 2005

360

Heat Equation with Periodic Boundary Conditions in 2D


(with extra terms)

Problem (F99, #5). In two spatial dimensions, consider the dierential equation
ut = u 2 u
with periodic boundary conditions on the unit square [0, 2]2.
a) If = 2 nd a solution whose amplitude increases as t increases.
b) Find a value 0 , so that the solution of this PDE stays bounded as t , if < 0 .
Proof. a) Eigenfunctions of the Laplacian.
The periodic boundary conditions imply a Fourier Series solution of the form:

amn (t)ei(mx+ny) .
u(x, t) =
m,n

ut =

amn (t)ei(mx+ny) ,

m,n

u = uxx + uyy =


(m2 + n2 ) amn (t)ei(mx+ny) ,
m,n

2 u = uxxxx + 2uxxyy + uyyyy =


=

(m4 + 2m2 n2 + n4 ) amn (t)ei(mx+ny)

m,n

(m2 + n2 )2 amn (t)ei(mx+ny) .

m,n

Plugging this into the PDE, we obtain


amn (t) = (m2 + n2 )amn (t) (m2 + n2 )2 amn (t),
amn (t) [(m2 + n2 ) (m2 + n2 )2 ]amn (t) = 0,
amn (t) (m2 + n2 )[ (m2 + n2 )]amn (t) = 0.
The solution to the ODE above is
2

amn (t) = mn e(m +n )[(m +n )]t.



2
2
2
2
u(x, t) =
mn e(m +n )[(m +n )]t ei(mx+ny)
  .
m,n

oscillates

When = 2, we have

2
2
2
2
mn e(m +n )[2(m +n )]t ei(mx+ny) .
u(x, t) =
m,n

We need a solution whose amplitude increases as t increases. Thus, we need those


mn > 0, with
(m2 + n2 )[2 (m2 + n2 )] > 0,
2 (m2 + n2 ) > 0,
2 > m2 + n2 .
Hence, mn > 0 for (m, n) = (0, 0), (m, n) = (1, 0), (m, n) = (0, 1).
Else, mn = 0. Thus,
u(x, t) = 00 + 10 et eix + 01 et eiy = 1 + et eix + et eiy
= 1 + et (cos x + i sin x) + et (cos y + i sin y).

Partial Dierential Equations

Igor Yanovsky, 2005

b) For 0 = 1, the solution  stays bounded as t .

361

Partial Dierential Equations

Igor Yanovsky, 2005

362

Problem (F93, #1).


Suppose that a and b are constants with a 0, and consider the equation
ut = uxx + uyy au3 + bu

(29.4)

in which u(x, y, t) is 2-periodic in x and y.


a) Let u be a solution of (29.4) with

||u(t = 0)|| =

2  2

|u(x, y, t = 0)|2 dxdy 1/2 < .

Derive an explicit bound on ||u(t)|| and show that it stays nite for all t.
b) If a = 0, construct the normal modes for (29.4); i.e. nd all solutions of the form
u(x, y, t) = et+ikx+ily .
c) Use these normal modes to construct a solution of (29.4) with a = 0 for the initial
data
u(x, y, t = 0) =

1
1

1 ix
2e

1
1 12 eix

Proof. a) Multiply the equation by u and integrate:


ut = u au3 + bu,
uut = uu au4 + bu2 ,



4
uut dx =
uu dx
au dx +
bu2 dx,






u
1 d
2
2
4
ds
u dx =
u
|u| dx
au dx +
bu2 dx,
2 dt
n









=0, u periodic on [0,2]2

d
||u||22 2b ||u||22,
dt
||u||22 ||u(x, 0)||22 e2bt ,
||u||2 ||u(x, 0)||2 ebt ebt .
Thus, ||u|| stays nite for all t.
b) Since a = 0, plugging u = et+ikx+ily into the equation, we obtain:
ut
t+ikx+ily

= uxx + uyy + bu,

= (k2 l 2 + b) et+ikx+ily ,
= k2 l 2 + b.

Thus,
2

ukl = e(k l +b)t+ikx+ily ,



2
2
akl e(k l +b)t+ikx+ily .
u(x, y, t) =
k,l

Partial Dierential Equations

Igor Yanovsky, 2005

c) Using the initial condition, we obtain:



1
1
akl ei(kx+ly) =
u(x, y, 0) =
1 ix +
1 ix
1 2e
1 2e
k,l
=
=


1
k=0


k=0

= 2+

eix

k
+

1 ikx
e +
2k


k=1


1

k=0


k=0

ikx

ak e

k=


1 ikx
1 ikx
e
+
e .
k
k
2
2
k=1



1 ikx
1 ikx
= 2+
e +
e ,
k
k
2
2
k=1

a0 = 2;

a0 = 2;

k=1

1
, k > 0;
2k
1
ak = |k| , k =
 0.
2
ak =

u(x, y, t) = 2ebt +

ak =

+


81

1
, k<0
2k

1 (k2 +b)t+ikx
e
.
|k|
2
k=, k=0

81

Note a similar question formulation in F92 #3(b).

1 ikx
e
,
2k

Thus, l = 0, and we have

eix

363

Partial Dierential Equations

Igor Yanovsky, 2005

364

Problem (S00, #3). Consider the initial-boundary value problem for u = u(x, y, t)
ut = u u
for (x, y) [0, 2]2, with periodic boundary conditions and with
u(x, y, 0) = u0 (x, y)
in which u0 is periodic. Find an asymptotic expansion for u for t large with terms
tending to zero increasingly rapidly as t .
Proof. Since we have periodic boundary conditions, assume

umn (t) ei(mx+ny) .
u(x, y, t) =
m,n

Plug this into the equation:




umn (t) ei(mx+ny) =
(m2 n2 1) umn (t) ei(mx+ny) ,
m,n

m,n

umn (t)

= (m2 n2 1) umn (t),


2

umn (t) = amn e(m n 1)t ,



2
2
amn e(m +n +1)t ei(mx+ny) .
u(x, y, t) =
m,n

Since u0 is periodic,

u0mn ei(mx+ny) ,
u0 (x, y) =

u0mn

m,n

1
=
4 2

2
0

u0 (x, y) ei(mx+ny) dxdy.

Initial condition gives:



amn ei(mx+ny) = u0 (x, y),
u(x, y, 0) =
m,n


m,n

u(x, y, t) =

amn ei(mx+ny) =

u0mn ei(mx+ny) ,

m,n

amn = u0mn .
2 +n2 +1)t

u0mn e(m

ei(mx+ny) .

m,n
2 +n2 +1)t

u0mn e(m

2 +n2 +1)t

ei(mx+ny) 0 as t , since e(m

0 as t .

Partial Dierential Equations

30

Igor Yanovsky, 2005

Problems: Fourier Transform

Problem (S01, #2b). Write the solution of initial value problem




1 0
Ux = 0,
Ut
5 3
for general initial data
 (1)
 

u (x, 0)
f (x)
=
0
u(2)(x, 0)

as an inverse Fourier transform.

You may assume that f is smooth and rapidly decreasing as |x| .


Proof. Consider the original system:
(1)

= 0,
ut u(1)
x
(2)

(2)
= 0.
ut 5u(1)
x 3ux

Take the Fourier transform in x. The transformed initial value problems are:
(1)

u(1) = 0,
u
t i
(2)

u(1) 3i
u(2) = 0,
u
t 5i

u
(1)(, 0) = f(),
u
(2)(, 0) = 0.

Solving the rst ODE for u


(1) gives:
u
(1)(, t) = f()eit.

With this u
(1), the second initial value problem becomes
(2)

it

u(2) = 5i f()e
,
u
t 3i

u
(2)(, 0) = 0.

The homogeneous solution of the above ODE is:


(2)

u
h (, t) = c1 e3it .
(2)

With u
p = c2 eit as anzats for a particular solution, we obtain:
ic2 eit 3ic2 eit = 5i f()eit,
2ic2 eit = 5i f()eit,
5
c2 = f().
2
5
it
u
(2)
p (, t) = f ()e .
2
5
(2)
3it
h (, t) + u
(2)
f()eit.
u
(2)(, t) = u
p (, t) = c1 e
2
We nd c1 using initial conditions:
5
5
c1 = f().
u
(2)(, 0) = c1 f() = 0
2
2
Thus,


5

u
(2)(, t) = f() e3it eit .
2

365

Partial Dierential Equations

Igor Yanovsky, 2005

u(1)(x, t) and u(2)(x, t) are be obtained by taking inverse Fourier transform:



 (1)

1
(1)
u (x, t) = u
 (, t)
=
eix f() eit d,
2 Rn

 (2)



1
5
 (, t)
=
eix f() e3it eit d.
u(2)(x, t) = u
2
2 Rn

366

Partial Dierential Equations

Igor Yanovsky, 2005

367

Problem (S02, #4). Use the Fourier transform on L2 (R) to show that
du
+ cu(x) + u(x 1) = f
dx

(30.1)

has a unique solution u L2 (R) for each f L2 (R) when |c| > 1 - you may assume
that c is a real number.
 by
Proof. u L2 (R). Dene its Fourier transform u

1
eix u(x) dx
for R.
u
() =
2 R
*
du
() = i
u().
dx

We can nd u(x
1)() in two ways.

Let u(x
1) = v(x), and determinte v():
 
y


u(x
1)() = v() =
=



1
1

eix v(x) dx =
ei(y+1) u(y) dy
2 R
2 R

1

eiy ei u(y) dy = ei u
().

2 R

We can also write the denition for u


() and substitute x 1 later in calculations:


1
1
iy
e
u(y) dy =
ei(x1) u(x 1) dx
u
() =
2 R
2 R

1

eix ei u(x 1) dx = ei u(x
1)(),
=
2 R

().

u(x
1)() = ei u
Substituting into (30.1), we obtain
() = f(),
i
u() + c
u() + ei u
f()
.
u
() =
i + c + ei




f()
 = 1 f B,
= fB
u(x) =
i
i + c + e
2
1
 =
,
where B
i + c + ei



1
eix
1

=
d.
B=
i + c + ei
2 R i + c + ei
For |c| > 1, u
() exists for all R, so that u(x) = (
u()) and this is unique by the
Fourier Inversion Theorem.

Note that in Rn ,  becomes



u(x
1)() = v() =
=



1
1
ix
e
v(x) dx =
ei(y+1) u(y) dy
n
n
n
n
(2) 2 R
(2) 2 R

&
1


eiy ei1 u(y) dy = ei1 u
() = e(i j j ) u
().
n
(2) 2 Rn

Partial Dierential Equations

Igor Yanovsky, 2005

368

Problem (F96, #3). Find the fundamental solution for the equation
ut = uxx xux.

(30.2)

Hint: The Fourier transform converts this problem into a PDE which can be solved
using the method of characteristics.
 by
Proof. u L2 (R). Dene its Fourier transform u

1
eix u(x) dx
for R.
u
() =
2 R
(),
u
*x () = i u
2
() = 2 u
(). 
u

xx () = (i) u

We nd x
ux () in two steps:
Multiplication by x:



d
1

u
().
eix ixu(x) dx =
ixu() =
d
2 R
d

().
xu(x)()
= i u
d
Using the previous result, we nd:






1
1  ix
1
ix

e
e
xu

xux (x) dx =
(i)eix x + eix u dx
xux (x)() =

2 R
2
2 R



=0


1
1
eix u dx
= i eix x u dx
2
2
R
R

 d
d

() u
() = u
() u
().
= i xu(x)()
u
() = i i u
d
d
d

u
() u
(). 
xu
x (x)() =
d
Plugging these into (30.2), we get:

u
(, t) = 2 u
(, t) u
(, t) ,
(, t) u
t
d
2
 + 
u + u
,
u
t = u
u = ( 2 1)
u.
u
t 
We now solve the above equation by characteristics.
We change the notation: u
 u, t y, x. We have
uy xux = (x2 1)u.
dx
dt
dy
dt
dz
dt

x = c1 et ,

(c1 = xet )

y = t + c2 ,

dz
= (c21 e2t 1)dt
z
x2
x2
x2
1
+ t + c3 =
+ y c2 + c3 z = ce 2 +y .
log z = c21 e2t + t + c3 =
2
2
2
=

(x2 1)z = (c21 e2t 1)z

Partial Dierential Equations

Igor Yanovsky, 2005

Changing the notation back, we have


2

u
(, t) = ce 2

+t

Thus, we have
2

u
(, t) = ce 2

+t

We use Inverse Fourier Tranform to get u(x, t): 82




2
1
1
ix
u(x, t) =
e u
(, t) d =
eix ce 2 +t d
2 R
2 R


2
2
c
c
eix e 2 d = et
eix+ 2 d
= et
2
2
R
R
2
(+ix)2
2ix+
x2
c t
c
t
e 2 d = e
e 2 d e 2
= e
2
2
R
R

y2
x2
x2
x2
c
c
= et e 2
e 2 dy = et e 2 2 = c et e 2 .
2
2
R
u(x, t) = c et e

x2
2

Check:
ut = c et e

x2
2

x2
2

ux = c et xe ,
 x2
x2 
uxx = c et e 2 + x2 e 2 .
Thus,
ut = uxx xux ,
 x2
x2 
x2
= c et e 2 + x2 e 2 x c et xe 2 . 
c et e
x2
2

82

We complete the square for powers of exponentials.

369

Partial Dierential Equations

Igor Yanovsky, 2005

370

Problem (W02, #4). a) Solve the initial value problem


u
u 
+
ak (t)
+ a0 (t)u = 0,
t
xk
n

x Rn ,

k=1

u(0, x) = f (x)
where ak (t), k = 1, . . ., n, and a0 (t) are continuous functions, and f is a continuous
function. You may assume f has compact support.
b) Solve the initial value problem
u
u 
+
ak (t)
+ a0 (t)u = f (x, t),
t
xk
n

x Rn ,

k=1

u(0, x) = 0
where f is continuous in x and t.
Proof. a) Use the Fourier transform to solve this problem.

1
eix u(x, t) dx
for R.
u
(, t) =
n
2
n
(2)
R

u
= ik u
.
xk
Thus, the equation becomes:

&
 + a0 (t)
u = 0,
u
t + i nk=1 ak (t)k u

u
(, 0) = f (),
or

 + a0 (t)
u = 0,
u
t + i a(t)  u



.
u
t = i a(t) + a0 (t) u

This is an ODE in u
 with solution:
t

u
(, t) = ce 0 (i a(s)+a0 (s)) ds ,
t

u
(, t) = f() e 0 (i a(s)+a0 (s)) ds .

u
(, 0) = c = f().

Thus,

Use the Inverse Fourier transform to get u(x, t):




t
(f g)(x)

=
,
u(x, t) = u
(, t) = f() e 0 (i a(s)+a0 (s)) ds
n
(2) 2
t

where 
g() = e 0 (i a(s)+a0 (s)) ds .


 t

1
1

ix
e 
g() d =
eix e 0 (i a(s)+a0 (s)) ds d.
g(x) =
n
n
(2) 2 Rn
(2) 2 Rn
 
  t (i a(s)+a
(f g)(x)
1
 0 (s)) ds 
i(xy)
u(x, t) =
=
e
e 0
d f (y) dy.
n
n
(2) Rn Rn
(2) 2


b) Use Duhamels Principle and the result from (a).


 t
U (x, t s, s) ds,
where U (x, t, s) solves
u(x, t) =
0

U
U 
+
ak (t)
+ a0 (t)U = 0,
t
xk
n

k=1

U (x, 0, s) = f (x, s).

Partial Dierential Equations



u(x, t) =
0

U (x, t s, s) ds =

Igor Yanovsky, 2005


1
(2)n

 t
0


Rn

Rn

371

  ts


ei(xy) e 0 (i a(s)+a0 (s)) ds d f (y, s) dy ds.

Partial Dierential Equations

Igor Yanovsky, 2005

Problem (S93, #2). a) Dene the Fourier transform




eix f (x) dx.
f () =

372

83

State the inversion theorem. If

, || < a,
1

f () =
2 , || = a,

0, || > a,
where a is a real constant, what f (x) does the inversion theorem give?
b) Show that

(x),
f (x
b) = eib f
where b is a real constant. Hence, using part (a) and Parsevals theorem, show that

1 sin a(x + z) sin a(x + )
sin a(z )
dx =
,

x+z
x+
z
where z and are real constants.
Proof. a) The inverse Fourier transform for f L1 (Rn ):

1

eix f (x) dx
for R.
f () =
2
Fourier Inversion Theorem: Assume f L2 (R). Then
1
f (x) =
2

ix

1
f() d =
2

ei(yx) f (y) dy d = (f) (x).

Parsevals theorem (Plancherels theorem) (for this denition of the Fourier


transform). Assume f L1 (Rn ) L2 (Rn ). Then f, f L2 (Rn ) and
1 
||f||L2 (Rn ) = ||f ||L2 (Rn ) = ||f ||L2(Rn ) ,
2


1
2
|f (x)| dx =
|f()|2 d.
2

Also,

f (x) g(x) dx =

We can write

,

f () =
0,
83

1
2

or

g() d.
f() 

|| < a,
|| > a.

Note that the Fourier transform is dened incorrectly here. There should be - sign in eix .
Need to be careful, since the consequences of this denition propagate throughout the solution.

Partial Dierential Equations

Igor Yanovsky, 2005

373


 a
 a

1
1
1
1
eix f() d =
0 d +
eix d +
0 d
f (x) = (f()) =
2
2
2 a
2 a


1  ix =a
1  iax
sin ax
1 a ix
e
e
. 
e
d =
=
eiax =
=
2 a
2ix
2ix
x
=a
g():
b) Let f (x
b) = g(x), and determinte 
 
y


f (x
b)() = 
g() =


eix g(x) dx =
R

R
iy ib

ei(y+b) f (y) dy

f (y) dy = eib f(). 

sin ax
x

(from (a)), we have


With f (x) =


1
1
sin a(x + z) sin a(x + s)
dx =
f (x + z)f (x + s) dx

x+z
x+s


1
f (x + z s)f (x ) dx
=


1 1
+ z s)f
(x ) d
f (x
=
2

1
f
() ei(zs) f
() d
=
2 2
 a
2
1

ei(zs) d
f
()
=
2 2 a
 a
1
2 ei(zs) d
=
2 2 a

1 a i(zs)
e
d
=
2 a
 i(zs) =a
1
e
=
=a
2i(z s)
=

(x = x + s, dx = dx)


(Parsevals)
part (b)

sin a(z s)
ei(zs)a ei(zs)a
=
. 
2i(z s)
zs

Partial Dierential Equations

Igor Yanovsky, 2005

374

Problem (F03, #5). State Parsevals relation for Fourier transforms.


Find the Fourier transform f() of


eix/2 y, |x| y
f (x) =
0,
|x| > y,
in which y and are constants.
Use this in Parsevals relation to show that

sin2 ( )y
d = y.
( )2

What does the transform f() become in the limit y ?


Use Parsevals relation to show that

1 sin( )y sin( )y
sin( )y
=
d.
( )
( )
( )
 by
Proof. f L2 (R). Dene its Fourier transform u

1
eix f (x) dx
for R.
f() =
2 R
Parsevals theorem (Plancherels theorem):
Assume f L1 (Rn ) L2 (Rn ). Then f, f L2 (Rn ) and
 L2 (Rn ) = ||f ||L2 (Rn ) = ||f ||L2(Rn ) ,
||f||


Also,

|f (x)| dx =

 2 d.
|f()|

f (x) g(x) dx =

or

g() d.
f() 

Find the Fourier transform of f :



 y
 y
ix
1
1
1
ix
ix e


e
f (x) dx =
e
ei()x dx
f () =
dx =
2 y
2 2y y
2 R
2 y

x=y
 i()y

1
1
1

ei()x
e
=
ei()y
=
2 2y i( )
2i 2y( )
x=y
sin y( )

.

=
2y( )
Parsevals theorem gives:


2

|f()| d =

|f (x)|2 dx,

 y  2ix

e
sin2 y( )
dx,
d =
2
2
2y( )
y 4y


y
sin2 y( )
d =
dx,
( )2
2 y


sin2 y( )
d = y.

( )2

Partial Dierential Equations

Igor Yanovsky, 2005

We had
f() =

sin y( )

.
2y( )

We make change of variables: =  . Then, =  + . We have


sin y(  )
,
f() = f(  + ) =
(  )
sin y( )
.
f( + ) =
( )

or

We will also use the following result.


Let f( + a) = 
g(), and determinte 
g() :
  




1
1

ix

e g() d =
eix( a) f(  ) d 
2 R
2 R
ixa
= e
f (x).

f( + a) = g() =

Using these results, we have



1 sin( )y sin( )y
d =
( )
( )
=
=
=
=
=
=
=


1  2  
( 2y)
f () f( + ) d


f (x) e()ixf (x) dx
2y


f (x)2 e()ix dx
2y

 y  2ix
e
e()ix dx
2y
4y
y

1 y ()ix
e
dx
2 y
 ()ixx=y
1
e
x=y
2i( )
 ()iy

1
e
e()iy
2i( )
sin( )y
.


375

Partial Dierential Equations

Igor Yanovsky, 2005

Problem (S95, #5). For the Laplace equation



 2
2

+
f = 0
f
x2 y 2

376

(30.3)

in the upper half plane y 0, consider


the Dirichlet problem f (x, 0) = g(x);

f (x, 0) = h(x).
the Neumann problem y
Assume that f , g and h are 2 periodic in x and that f is bounded at innity.
Find the Fourier transform N of the Dirichlet-Neumann map. In other words,
nd an operator N taking the Fourier transform of g to the Fourier transform of h; i.e.
hk .
N
gk = 

Proof. We solve the problem by two methods.


Fourier Series.
Since f is 2-periodic in x, we can write

f (x, y) =

an (y) einx.

n=

Plugging this into (30.3), we get the ODE:




n2 an (y)einx + an (y)einx

= 0,

n=

an (y) n2 an (y) = 0.


Initial conditions give: (g and h are 2-periodic in x)

f (x, 0) =

n=

fy (x, 0) =

inx

an (0)e

= g(x) =

an (0)einx = h(x) =

n=


n=

gn einx


an (0) = 
gn .


hn einx

an (0) = 
hn .

n=

Thus, the problems are:


an (y) n2 an (y) = 0,
an (0) = gn ,
hn .
a (0) = 
n

(Dirichlet)
(Neumann)

a0 (y) = b0 y + c0 .
an (y) = bn eny + cn eny , n = 1, 2, . . .;

ny
ny
, n = 1, 2, . . .;
a0 (y) = b0 .
an (y) = nbn e ncn e

Since f is bounded at y = , we have:


bn = 0

for n > 0,

cn = 0

for n < 0,

b0 = 0,

c0 arbitrary.

Partial Dierential Equations

Igor Yanovsky, 2005

377

n > 0:
an (y)

cn eny ,

an (0)

cn = 
gn ,

(Dirichlet)

ncn

(Neumann)

an (0)

= 
hn .
= 
hn .

n
gn
n < 0:
an (y)

bn eny ,

an (0)

an (0)

bn = 
gn ,
nbn = 
hn .

(Dirichlet)
(Neumann)

hn .
n
gn = 

|n|
gn = 
hn ,

n = 0.

n = 0 : a0 (y) = c0 ,
g0 ,
a0 (0) = c0 = 

h0 .
a0 (0) = 0 = 

(Dirichlet)
(Neumann)

Note that solution f (x, y) may be written as


f (x, y) =

inx

an (y) e

= a0 (y) +

n=

= c0 +

=

1


an (y) e

n=
1


bn eny einx +

n=

an (y) einx

n=1

cn eny einx

n=1

&
gn e e +
gn eny einx ,
n= 
n=1 
&1
&


hn ny inx
e .
c0 + n= hnn eny einx +
n=1 n e
g0 +

&1

inx

ny inx

(Dirichlet)
(Neumann)

Fourier Transform. The Fourier transform of f (x, y) in x is:



1

eix f (x, y) dx,
f (, y) =
2

1
eix f(, y) d.
f (x, y) =
2
(i)2 f(, y) + fy y(, y) = 0,
fyy 2 f = 0.

The solution to this ODE is:

f(, y) = c1 ey + c2 ey .
For > 0, c1 = 0; for < 0, c2 = 0.
fy (, y) = c2 ey ,
c2 ey ,


1
1
eix f (x, 0) dx =
eix g(x) dx = 
g(),
c2 = f(, 0) =
2
2


1
1
ix

e
fy (x, 0) dx =
eix h(x) dx = 
h().
c2 = fy (, 0) =
2
2

g() = 
h().

>0:

f(, y) =

(Dirichlet)
(Neumann)

Partial Dierential Equations

Igor Yanovsky, 2005

378

fy (, y) = c1 ey ,
c1 ey ,


1
1
ix


c1 = f (, 0) =
e
f (x, 0) dx =
eix g(x) dx = 
g(),
2
2


1
1
eix fy (x, 0) dx =
eix h(x) dx = 
h().
c1 = fy (, 0) =
2
2


g() = h().

<0:

f(, y) =

||
g() = 
h().

(Dirichlet)
(Neumann)

Partial Dierential Equations

Igor Yanovsky, 2005

Problem (F97, #3). Consider the Dirichlet problem in the half-space


n 2:

379
xn > 0,

u
+ k2 u = 0,
xn > 0
xn
x = (x1 , . . ., xn1 ).
u(x , 0) = f (x ),

u + a

Here a and k are constants.


Use the Fourier transform to show that for any f (x ) L2 (Rn1 ) there exists a
solution u(x , xn ) of the Dirichlet problem such that

|u(x , xn)|2 dx C
Rn

for all 0 < xn < +.


Proof.

84

Denote = (  , n). Transform in the rst n 1 variables:


2u

u 
 
( , xn ) + a
( , xn) + k2 u
(  , xn ) = 0.
2
xn
xn

(  , xn) +
|  |2 u

Thus, the ODE and initial conditions of the transformed problem become:

u
xn xn + a
uxn + (k2 |  |2 )
u = 0,
u
(  , 0) = f(  ).
With the anzats u
 = cesxn , we obtain s2 + as + (k2 |  |2 ) = 0, and

a a2 4(k2 |  |2 )
.
s1,2 =
2
Choosing only the negative root, we obtain the solution: 85


u
( , xn ) = c( ) e

u
(  , xn ) = f(  ) e

a2 4(k2 |  |2 )
xn
2

a2 4(k2 |  |2 )
xn
2

Parsevals theorem gives:


||u||2L2(Rn1 )

||
u||2L2 (Rn1 )


=
=

Rn1

|
u(  , xn)|2 d 

  aa2 4(k2 | |2 ) x 2 
n
f( ) e
2
d

Rn1
 2 2 n1
||f||
L (R
)

Thus,


=

u
(  , 0) = c = f(  ).

||f ||2L2(Rn1 )


Rn1

  2 
f( ) d

C,

since f (x ) L2 (Rn1 ). Thus, u(x , xn) L2 (Rn1 ).

84

Note that the last element of x = (x , xn ) = (x1 , . . . , xn1 , xn ), i.e. xn , plays a role of time t.
As such, the PDE may be written as
u + utt + aut + k2 u = 0.
85

Note that a > 0 should have been provided by the statement of the problem.

Partial Dierential Equations

Igor Yanovsky, 2005

380

Problem (F89, #7). Find the following fundamental solutions


2 G(x, y, t)
G(x, y, t)
G(x, y, t)
= a(t)
+ c(t)G(x, y, t)
+ b(t)
2
t
x
x
G(x, y, 0) = (x y),

a)

where a(t), b(t), c(t) are continuous functions on [0, +], a(t) > 0 for t > 0.

b)

n

G
G
(x1 , . . . , xn , y1 , . . . , yn , t) =
ak (t)
t
xk

for t > 0,

k=1

G(x1 , . . . , xn, y1 , . . . , yn , 0) = (x1 y1 )(x2 y2 ) . . . (xn yn ).

Proof. a) We use the Fourier transform to solve this problem.


Transform the equation in the rst variable only. That is,

1

eix G(x, y, t) dx.
G(, y, t) =
2 R
The equation is transformed to an ODE, that can be solved:
 y, t) + i b(t) G(,
 y, t) + c(t) G(,
 y, t),
 t(, y, t) = a(t) 2 G(,
G


 y, t),
 t(, y, t) =
a(t) 2 + i b(t) + c(t) G(,
G
t

 y, t) = c e
G(,

0 [a(s)

2 +i b(s)+c(s)] ds

We can also transform the initial condition:


1


 y, 0) = (x
= eiy .
G(,
y)() = eiy ()
2
Thus, the solution of the transformed problem is:
t
 y, t) = 1 eiy e 0 [a(s) 2 +i b(s)+c(s)] ds .
G(,
2
The inverse Fourier transform gives the solution to the original problem:



1
 y, t) d


=
eix G(,
G(x, y, t) = G(, y, t)
2
R

 1

t
1
2
eix eiy e 0 [a(s) +i b(s)+c(s)] ds d
=
2 R
2

t
1
2
ei(xy) e 0 [a(s) +i b(s)+c(s)] ds d.

=
2 R
y = (y1 , . . . , yn ). Transform in x:
b) Denote x = (x1 , . . . , xn ), 

1

 ,
 y, t) =
eix G(x, y, t) dx.
G(
n
(2) 2 Rn
The equation is transformed to an ODE, that can be solved:
 y, t) =
 t(,
G

n

k=1

 ,
 y, t) = c ei
G(

 y , t),
 ,
ak (t) ik G(
 t &n
k=1 ak (s) k ] ds
0[

for t > 0

Partial Dierential Equations

Igor Yanovsky, 2005

We can also transform the initial condition:




 )
 =
 ,
 y , 0) = (x1 y1 )(x2 y2 ) . . . (xn yn ) () = eiy (
G(

381

1
i
y 
.
n e
2
(2)

Thus, the solution of the transformed problem is:


 &
1
i
y  i 0t [ n
 ,
 y, t) =
k=1 ak (s) k ] ds .
e
G(
n e
(2) 2
The inverse Fourier transform gives the solution to the original problem:



1
  
 ,
 y, t) =
eix G(
, y, t) d
G(x, y, t) = G(
n
(2) 2 Rn

 1

 &
1

i
y  i 0t [ n

k=1 ak (s) k ] ds d
eix
e
=
n
n e
(2) 2 Rn
(2) 2

 &
1
i(
x
y ) i 0t [ n

k=1 ak (s) k ] ds d.
e
e

=
(2)n Rn

Partial Dierential Equations

Igor Yanovsky, 2005

382

Problem (W02, #7). Consider the equation

2
2
+

+
u=f
x2n
x21

in Rn ,

(30.4)

where f is an integrable function (i.e. f L1 (Rn )), satisfying f (x) = 0 for |x| R.
Solve (30.4) by Fourier transform, and prove the following results.
2
n
a) There
 is a solution of (30.4) belonging to L (R ) if n > 4. 2 n
b) If Rn f (x) dx = 0, there is a solution of (30.4) belonging to L (R ) if n > 2.
Proof.
u = f,
() = f(),
|| u
1
u
() = 2 f(),
||

f()
.
u(x) =
||2
2

Rn ,

a) Then

||
u||L2 (Rn) =

Rn

|f()|2
d
||4

1
2



||<1

1

2
|f()|2
|f()|2
d +
d
.
4
4
||
||1 ||





A

 2 B, so B < .
Notice, ||f ||2 = ||f||
Use polar coordinates on A.
 1
 1

2
|f()|2
|f|
n1
 2 r n5 dSn1 dr.
d =
r
dSn1 dr =
|f|
A =
4
4
||
r
0
0
||<1
Sn1
Sn1
If n > 4,

 2 dSn1 = ||f||2 < .


|f|
2

Sn1
1

u||L2 (Rn ) = (A + B) 2 < .


||u||L2(Rn) = ||

Partial Dierential Equations

Igor Yanovsky, 2005

b) We have
u(x, t) =
=
=
=
=

f()

1
ix f ()
=

e
d

n
||2
||2
(2) 2
 ix



e
1
1
iy
e
f
(y)
dy
d

n
n
(2) 2 ||2 (2) 2


 ei(xy)
1
f
(y)
d dy

(2)n
||2

 1 

ei(xy)r n1
1
f
(y)
r
dS
dr
dy

n1
(2)n
r2
0
Sn1


 1 

1
i(xy)r n3

f
(y)
e
r
dS
dr
dy.
n1
(2)n
0
Sn1



M < , if n>2.

|u(x, t)| =



1 

M
f
(y)
dy
 < .

(2)n

383

Partial Dierential Equations

Igor Yanovsky, 2005

384

Problem (F02, #7). For the right choice of the constant c, the function
F (x, y) = c(x + iy)1 is a fundamental solution for the equation
u
u
+i
=f
x
y

in R2 .

Find the right choice of c, and use your answer to compute the Fourier transform
(in distribution sense) of (x + iy)1 .
Proof.

86

 =
F1 (x, y) =




+i
i
.
x
y x
y
1
2

log |z| is the fundamental solution of the Laplacian. z = x + iy.

F1 (x, y) = ,




+i
i
F (x, y) = .
x
y x
y
hx + ihy = ei(x1 +y2 ) .
Suppose h = h(x1 + y2 ) or h = cei(x1+y2 ) .


c i1 ei(x1 +y2 ) i2 2 ei(x1 +y2 ) = ic(1 i2 ) ei(x1 +y2 ) ei(x1 +y2 ) ,
ic(1 i2 ) = 1,
1
,
c =
i(1 i2 )
1
ei(x1 +y2 ) .
h(x, y) =
i(1 i2 )
Integrate by parts:




1
1
1
() =
+i
dxdy
ei(x1 +y2 )
x + iy
i(1 i2 ) x
y (x + iy) 0
R2
1
1
=
.
=
i(1 i2 )
i(2 + i1 )

86

Alan solved in this problem in class.

Partial Dierential Equations

31

Igor Yanovsky, 2005

385

Laplace Transform

If u L1 (R+ ), we dene its Laplace transform to be


L[u(t)] = u# (s) =

est u(t) dt

(s > 0).

In practice, for a PDE involving time, it may be useful to perform a Laplace transform
in t, holding the space variables x xed.
The inversion formula for the Laplace transform is:
u(t) = L

1
[u (s)] =
2i
#

c+i

est u# (s) ds.

ci

Example: f (t) = 1.

 1
t=
1
L[1] =
est 1 dt = est
=
s
s
t=0
0
Example: f (t) = eat .


at
st at
e e dt =
L[e ] =
0

e(as)t dt =

for s > 0.

1  (as)t t=
1
e
=
as
sa
t=0

Convolution: We want to nd an inverse Laplace transform of


L

1
s

for s > a.

1
.
s2 +1

 t
 1
1
2
1 sin t dt = 1 cos t.
= f g =
s
s
+
1
0
   
L[f ]

L[g]

Partial Derivatives: u = u(x, t)





t=
st
st
e ut dt = e u(x, t)
+s
est u dt = sL[u] u(x, 0),
L[ut] =
t=0
0
 0

t=
st
st
e utt dt = e ut
+s
est ut dt = ut (x, 0) + sL[ut]
L[utt] =
t=0

0
2

= s L[u] su(x, 0) ut (x, 0),




L[u],
est ux dt =
L[ux] =
x
0

2
est uxx dt =
L[u].
L[uxx] =
x2
0
Heat Equation: Consider

in U (0, )
ut u = 0
u=f
on U {t = 0},
and perform a Laplace transform with respect to time:

est ut dt = sL[u] u(x, 0) = sL[u] f (x),
L[ut] =
0
est u dt = L[u].
L[u] =
0

Partial Dierential Equations


Thus, the transformed problem is:
have
v + sv = f

Igor Yanovsky, 2005

386

sL[u] f (x) = L[u]. Writing v(x) = L[u], we

in U.

Thus, the solution of this equation with RHS f is the Laplace transform of the solution
of the heat equation with initial data f .

Partial Dierential Equations

Igor Yanovsky, 2005

Table of Laplace Transforms: L[f ] = f # (s)


a
,
L[sin at] = 2
s + a2
s
,
L[cos at] = 2
s + a2
a
L[sinh at] = 2
,
s a2
s
,
L[cosh at] = 2
s a2
b
L[eat sin bt] =
,
(s a)2 + b2
sa
,
L[eat cos bt] =
(s a)2 + b2
n!
L[tn ] = n+1 ,
s
n!
,
L[tn eat ] =
(s a)n+1
eas
L[H(t a)] =
,
s
L[H(t a) f (t a)] = eas L[f ],

s>0
s>0
s > |a|
s > |a|
s>a
s>a
s>0
s>a
s>0

L[af (t) + bg(t)] = aL[f ] + bL[g],


L



t
0

L[f (t) g(t)] = L[f ] L[g],



g(t t) f (t ) dt = L[f ] L[g],
 df
= sL[f ] f (0),
L
dt

 d2 f
df
2


L[f
]

sf
(0)

f
(0),
f
=
=
s
L
dt2
dt
 dn f
= sn L[f ] sn1 f (0) . . . f n1 (0),
L
dtn
1 #
s
f
,
L[f (at)] =
a
a
L[ebtf (t)] = f # (s b),
d
L[tf (t)] = L[f ],
 ds
 f (t)

=
f # (s ) ds ,
L
t
s

 t
1
L[f ],
f (t ) dt =
L
s
0
1

L[J0 (at)] = (s2 + a2 ) 2 ,


L[(t a)] = esa .
Example: f (t) = sin t. After integrating by parts twice, we obtain:


est sin t dt = 1 s2
est sin t dt,
L[sin t] =
0
0
1
st
e
sin t dt =
.

1 + s2
0

387

Partial Dierential Equations

Igor Yanovsky, 2005

Example: f (t) = tn .

 tn est n 
n
n
st n
L[t ] =
e t dt =
+
est tn1 dt = L[tn1 ]
s
s 0
s
0
0


n!
n!
n n1
L[tn2 ] = . . . = n L[1] = n+1 .
=
s
s
s
s

388

Partial Dierential Equations

Igor Yanovsky, 2005

Problem (F00, #6). Consider the initial-boundary value problem


ut uxx + au = 0,

t > 0, x > 0

u(x, 0) = 0,

x>0

u(0, t) = g(t),

t > 0,

where g(t) is continuous function with a compact support, and a is constant.


Find the explicit solution of this problem.
Proof. We solve this problem using the Laplace transform.

#
est u(x, t) dt
(s > 0).
L[u(x, t)] = u (x, s) =
0


L[ut] =

est ut dt =



t=
est u(x, t)
+s
t=0

est u dt

= su# (x, s) u(x, 0) = su# (x, s),


(since u(x, 0) = 0)

2
#
est uxx dt =
u (x, s),
L[uxx] =
x2
0

est g(t) dt = g # (s).
L[u(0, t)] = u# (0, s) =
0

Plugging these into the equation, we obtain the ODE in u# :


su# (x, s)


2 #
u (x, s) + au# (x, s) = 0.
x2

(u# )xx (s + a)u# = 0,


u# (0, s) = g # (s).

This initial value problem has a solution:

u# (x, s) = c1 e

s+a x

+ c2 e

s+a x

Since we want u to be bounded as x , we have c1 = 0, so

u# (x, s) = c2 e

s+a x

s+a x

u# (x, s) = g # (s)e

u# (0, s) = c2 = g # (s),

thus,

To obtain u(x, t), we take the inverse Laplace transform of u# (x, s):



s+a x
= gf
u(x, t) = L1 [u# (x, s)] = L1 g # (s) e 

  
L[g]

L[f ]


 1  c+i



1 s+a x
st s+a x
e e
ds ,
= g
= gL e
2i ci
 t
 1  c+i 


u(x, t) =
g(t t )
est e s+a x ds dt .
2i ci
0

389

Partial Dierential Equations

Igor Yanovsky, 2005

390

Problem (F04, #8). The function y(x, t) satises the partial dierential equation
2y
y
+
+ 2y = 0,
x xt
and the boundary conditions
x

y(0, t) = eat ,

y(x, 0) = 1,

where a 0. Find the Laplace transform, y(x, s), of the solution, and hence derive
an expression for y(x, t) in the domain x 0, t 0.
Proof. We change the notation: y u. We have

xux + uxt + 2u = 0,
u(x, 0) = 1, u(0, t) = eat .
The Laplace transform is dened as:

#
est u(x, t) dt
L[u(x, t)] = u (x, s) =

(s > 0).

est ux dt = x(u# )x ,



t=
st
st
e uxt dt = e ux (x, t)
+s
est ux dt
L[uxt] =

L[xux] =

st

xux dt = x

t=0

= s(u )x ux (x, 0) = s(u )x ,




est eat dt =
L[u(0, t)] = u# (0, s) =
0

(since u(x, 0) = 0)

e(s+a)t dt =

1 (s+a)t t=
e
s+a
t=0

1
.
s+a

Plugging these into the equation, we obtain the ODE in u# :



(x + s)(u# )x + 2u# = 0,
1
,
u# (0, s) = s+a
which can be solved:
2
(u# )x
=
#
x+s
u

log u# = 2 log(x + s) + c1 u# = c2 elog(x+s)

From the initial conditions:


c2
1
u# (0, s) = 2 =
s
s+a
u# (x, s) =

c2 =

c2
.
(x + s)2

s2
.
s+a

s2
.
(s + a)(x + s)2

To obtain u(x, t), we take the inverse Laplace transform of u# (x, s):
 c+i




s2
1
s2
st
e
=
ds.
u(x, t) = L1 [u# (x, s)] = L1
(s + a)(x + s)2
2i ci
(s + a)(x + s)2
 c+i


1
s2
u(x, t) =
est
ds.
2i ci
(s + a)(x + s)2

Partial Dierential Equations

Igor Yanovsky, 2005

391

Problem (F90, #1). Using the Laplace transform, or any other convenient method,
solve the Volterra integral equation
 x
sin(x y)u(y) dy.
u(x) = sin x +
0

Proof. Rewrite the equation:


 t
sin(t t )u(t ) dt ,
u(t) = sin t +
0

u(t) = sin t + (sin t) u.

Taking the Laplace transform of each of the elements in :



#
est u(t) dt,
L[u(t)] = u (s) =
0

L[sin t] =

1
,
1 + s2

L[(sin t) u] = L[sin t] L[u] =

u#
.
1 + s2

Plugging these into the equation:


u# =

1
u#
u# + 1
+
=
.
2
2
1+s
1+s
1 + s2

u# (s) =

1
.
s2

To obtain u(t), we take the inverse Laplace transform of u# (s):


1
u(t) = L1 [u# (s)] = L1 2 = t.
s
u(t) = t.

Partial Dierential Equations

Igor Yanovsky, 2005

392

Problem (F91, #5). In what follows, the Laplace transform of x(t) is denoted
either by x(s) or by Lx(t). Show that, for integral n 0,
L(tn ) =

n!
sn+1

Hence show that

1
LJ0 (2 ut) = eu/s ,
s
where
J0 (z) =


(1)n ( 1 z)2n
2

n!n!

n=0

is a Bessel function. Hence show that




 

1
1
L
.
J0 (2 ut)x(u) du = x
s
s
0

(31.1)

Assuming that
1
,
LJ0 (at) =
2
a + s2
prove with the help of (31.1) that if t 0
 


1
t
.
J0 (au)J0 (2 ut) du = J0
a
a
0
Hint: For the last part, use the uniqueness of the Laplace transform.
Proof.

L[tn ] =

 tn est n 
n
st
n
e
t
dt
=

+
est tn1 dt = L[tn1 ]

s
s
s
0
0



f
g
=0

n!
n!
n
n 1 n2
L[t ] = . . . = n L[1] = n+1 .
s
s
s
s





(1)n un tn
(1)nun
(1)n un
=
L[tn ] =
LJ0 (2 ut) = L
n!n!
n!n!
n!sn+1

=

L

n=0

1
(1)
s n=0 n!


J0 (2 ut) x(u) du

x# (s) =


0

eus x(u) du.

u n
s

=
where

n=0

1 u
e s.
s

n=0

1
L[J0 (2 ut)] x(u) du =
s
0
1 #
1
x
,

s
s


0

e s x(u) du

Partial Dierential Equations

32

Igor Yanovsky, 2005

393

Linear Functional Analysis

32.1

Norms

|| || is a norm on a vector space X if


i) ||x|| = 0 i x = 0.
ii) ||x|| = || ||x|| for all scalars .
iii) ||x + y|| ||x|| + ||y|| (the triangle inequality).
The norm induces the distance function d(x, y) = ||x y|| so that X is a metric space,
called a normed vector space.

32.2

Banach and Hilbert Spaces

A Banach space is a normed vector space that is complete in that norms metric. I.e.
a complete normed linear space is a Banach space.
A Hilbert space is an inner product space for which the corresponding normed space
is complete. I.e. a complete inner product space is a Hilbert space.
Examples: 1) Let K be a compact set of Rn and let C(K) denote the space of continuous
functions on K. Since every u C(K) achieves maximum and minimum values on K,
we may dene
||u|| = max |u(x)|.
xK

|| || is indeed a norm on C(K) and since a uniform limit of continuous functions is


continuous, C(K) is a Banach space. However, this norm cannot be derived from an
inner product, so C(K) is not a Hilbert space.
2) C(K) is not a Banach space with || ||2 norm. (Bell-shaped functions on [0, 1] may
converge to a discontinuous -function). In general, the space of continuous functions
on [0, 1], with the norm || ||p, 1 p < , is not a Banach space, since it is not
complete.
3) Rn and C n are real and complex Banach spaces (with a Eucledian norm).
4) Lp are Banach spaces (with || ||p norm).
5) The space of bounded real-valued functions on a set S, with the sup norm || ||S are
Banach spaces.
6) The space of bounded continuous real-valued functions on a metric space X is a
Banach space.

32.3

Cauchy-Schwarz Inequality

|(u, v)| ||u||||v|| in any norm, for example


1
1
|a(u, v)| a(u, u) 2 a(v, v) 2



1 
1
|v|dx = |v| 1 dx = ( |v|2 dx) 2 ( 12 dx) 2

32.4


1 
1
|uv|dx ( u2 dx) 2 ( v 2 dx) 2

H
older Inequality

|uv| dx ||u||p||v||q,

which holds for u Lp () and v Lq (), where


uv L1 ().

1
p

1
q

= 1. In particular, this shows

Partial Dierential Equations

32.5

Igor Yanovsky, 2005

394

Minkowski Inequality
||u + v||p ||u||p + ||v||p,

which holds for u, v Lp (). In particular, it shows u + v Lp ().


Using the Minkowski Inequality, we nd that || ||p is a norm on Lp ().
The Riesz-Fischer theorem asserts that Lp () is complete in this norm, so Lp() is a
Banach space under the norm || ||p.
If p = 2, then L2 () is a Hilbert space with inner product

(u, v) = uv dx.

denotes the functions that, along with


Example: Rn bounded domain, C 1 ()
is
Then C 1 ()
their rst-order derivatives, extend continuously to the compact set .
a Banach space under the norm
||u||1, = max(|u(x)| + |u(x)|).

Note that C 1 () is not a Banach space since ||u||1, need not be nite for u C 1 ().

32.6

Sobolev Spaces

A Sobolev space is a space of functions whose distributional derivatives (up to some


xed order) exist in an Lp -space.
Let be a domain in Rn , and let us introduce

(32.1)
< u, v >1 = (u v + uv) dx,

||u||1,2 =



(|u| + |u| ) dx

< u, u >1 =

1
2

(32.2)

when these expressions are dened and nite. For example, (32.1) and (32.2) are dened
for functions in C01 (). However, C01 () is not complete under the norm (32.2), and so
does not form a Hilbert space.
Divergence Theorem


 dx

div A
A n dS =

Trace Theorem
uL2() CuH 1 ()

smooth or square

Poincare Inequality
1p
up Cup


2
|u(x)| dx C
|u(x)|2 dx

u C01 (), H01,2 ()

u u p up

u H01,p()

i.e. p = 2

Partial Dierential Equations


1
u =
||

Igor Yanovsky, 2005

395


u(x) dx

(Average value of u over ), || is the volume of

Notes
u
u
u
= u n = n1
+ n2
n
x1
x2


|u|
u dx
|u| dx =

ab

a+b
2

ab

a2 + b2
2

u2
uu = ( )
2


2
(uxy ) dx =
uxxuyy dx

|u|2 = u2x1 + u2x2

||u||||u||

u H02 ()

||u||2 + ||u||2
2

square

Problem (F04, #6). Let q C01 (R3 ). Prove that the vector eld

q(y)(x y)
1
dy
u(x) =
4 R3 |x y|3
enjoys the following properties: 87
a) u(x) is conservative;
b) div u(x) = q(x) for all x R3 ;
c) |u(x)| = O(|x|2 ) for large x.
Furthermore, prove that the proverties (1), (2), and (3) above determine the vector eld
u(x) uniquely.
Proof. a) To show that 
u(x) is conservative, we need to show that curl u = 0.

The curl of V is another vector eld dened by



e1 e2 e3
V3 V2 V1 V3 V2 V1



1 2 3

.
=
curl V = V = det
x2 x3 x3
x1 x1
x2
V1 V2 V3
Consider
 (x) =
V

(x1 , x2 , x3 )
x
=
3 .
2
|x|3
(x1 + x22 + x23 ) 2

Then,
u(x) =
curl u(x) =
curl V (x) = curl
=


1
q(y) V (x y) dy,
4 R3

1
q(y) curlx V (x y) dy.
4 R3
(x1 , x2 , x3 )
3

(x21 + x22 + x23 ) 2


32 2x2 x3

(x21 + x22 + x23 ) 2


= (0, 0, 0).
87

McOwen, p. 138-140.

32 2x3 x2

5 ,

(x21 + x22 + x23 ) 2

32 2x3 x1

(x21 + x22 + x23 ) 2

32 2x1 x3

5 ,

(x21 + x22 + x23 ) 2

32 2x1 x2

(x21 + x22 + x23 ) 2

Partial Dierential Equations

Igor Yanovsky, 2005


1
u(x) is conservative. 
Thus, curl u = 4
R3 q(y) 0 dy = 0, and 
1
b) Note that the Laplace kernel in R3 is 4r
.



q(y)(x y)
q(r) r
q(r)
1
1
dr = q.
dy =
r dr =
u(x) =
3
3
4 R3 |x y|
4 R3 r
R3 4r
c) Consider
1
F (x) =
4


R3

q(y)
dy.
|x y|

F (x) is O(|x|1 ) as |x| .


Note that u = F , which is clearly O(|x|2 ) as |x| .

396