Partial Differential Equations: Graduate Level Problems and Solutions

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.
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
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
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
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
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 =

A=
a b
c d

,
1
=
det(A)
Igor Yanovsky, 2005
d b
c a
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, . . .
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.
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
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.
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 .
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)

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
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
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
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.

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 .
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
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.

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.
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
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
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.
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.
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
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).
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).
Igor Yanovsky, 2005
We can solve (7.7) with nonhomogeneous initial conditions,

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.
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.
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.

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
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.
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.
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
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.
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
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.
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|).
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
= 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)


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)

= 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
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)
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)
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

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

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)

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

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

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)
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


log |x y| = log
and
log |x y| = .
n
0, (|u| is f inite)
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

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

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)

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

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

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
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
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.
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 .
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) .
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
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.

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

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
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
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.
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.
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.
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.
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,
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

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
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
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
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)).
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
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)

u3x
+ uy + cu = 0,
3
p3
+ q + cz = 0.
F (x, y, z, p, q) =
3
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
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
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 ,
x<0
x>0
t=
Igor Yanovsky, 2005
1
1
1
=
= 3.
2

min{3h(s) h (s)}
min f (s)
3a
64
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 .
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).
Igor Yanovsky, 2005
b) The characteristics would intersect when 1 f (s)xyeuy = 0. Thus,

tc =
1
f (s)xeutc
67
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).
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.
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,
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
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
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 < +.

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.
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
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
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.
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
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
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
Igor Yanovsky, 2005
77
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.

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<
Igor Yanovsky, 2005
78
Igor Yanovsky, 2005
79
1
u2 = 0,
t>0
ut +
2
x
with initial data
u(x, 0) = h(x) =
2
0
if 0 < x < 1,
if otherwise.

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,
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.
Igor Yanovsky, 2005
80
Igor Yanovsky, 2005
81
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.

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.
Igor Yanovsky, 2005
82
Igor Yanovsky, 2005
83
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.
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 ,
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) =
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.
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.
F (x, y, z, p, q) =
x2
p2
q+
= 0.
2
2
is parameterized by : (s, 0, s, (s), (s)).

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.
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) =
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.
F (x, y, z, p, q) =
x2
p2
+q+
= 0.
2
2

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.
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).
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)).
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 .
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.
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.
F (x, y, z, p, q) = p2 + q 2 1 = 0.
2
is parameterized by : (s, s2 , 0, (s), (s)).

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=
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.
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)

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.
Igor Yanovsky, 2005
94

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
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 )

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

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)

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.
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
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 )

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

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)
Igor Yanovsky, 2005
97

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
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.
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
f
g
24
Chain Rule: u(x1 , x2 , x3 ) = (f (x1, x2 , x3 ), g(x1 , x2 , x3 )), then ux1 =
+
.
f x1
g x1
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
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
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.
Igor Yanovsky, 2005
101
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.

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.
Igor Yanovsky, 2005
Part (b) of the problem is solved in the Fourier Transform section.
103
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,
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.

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)
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.

.
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.

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.

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)

(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)
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).
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)
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))
1
1
3
5 f (x + 3t) 5 f ( 2 (x 2t))
2
1
3
5 f (x + 3t) + 10 f ( 2 (x 2t))

.
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.

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)

v (1)(x, t) = F (x + t),
v (2)(x, t) = G(x 3t).
(14.13)

(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.
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),
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.
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

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 )
Igor Yanovsky, 2005

110
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.

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)
v (1)(x, t) = F (x + 2t),
v (2)(x, t) = G(x 2t).
(14.18)
Igor Yanovsky, 2005
112

(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),

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)
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

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:
Igor Yanovsky, 2005
113

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.
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)
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.

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)
Igor Yanovsky, 2005
115

(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 .
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.
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,
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.

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)
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
(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
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.
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)
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.

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)

v (1)(x, t) = F (x + t),
v (2)(x, t) = G(x t).
(14.30)
Igor Yanovsky, 2005
119

(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.
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.

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.
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).
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.

1 0
Vx = 0,
Vt +
0 2
(14.36)

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.

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
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

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
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.
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.
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

Conditions give ux = 0 at x = . Thus
U+2
U2
+ R+ =
+ R ,
2
2
U+ R+ = U R .
Igor Yanovsky, 2005
129
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)

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.
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

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.
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,

we have
u(x, t) =

k=0
that is, u(t) A.
u
k ei
kx
2
Igor Yanovsky, 2005
135
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

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 +
ux uxxxxx dx = 0.

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
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
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.
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)
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)
a0 (t)
+
an (t) cos nx + bn (t) sin nx.
u(x, t) =
2
n=1
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.

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)
4 cos(2n + 1)t sin(2n + 1)x

.
u(x, t) =
n=0
(2n + 1)
Igor Yanovsky, 2005
143
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
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)
a0 (t)
nx
nx
u(x, t) =
+
+ bn (t) sin
.
an (t) cos
2
L
L
n=1
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 .
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.
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
a0 (t)
+
an (t) cos nx + bn (t) sin nx.
u(x, t) =
2
n=1

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.

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)
Igor Yanovsky, 2005
2 (cos n 1) cos nt cos nx
u(x, t) = +
.
2
n=1
n2
147
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
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)

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
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.

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.
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).
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
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.
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
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.
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 =
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
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
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
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 ).
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.
Igor Yanovsky, 2005
163
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.
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.

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 ).

at that point.
(
)

f
f
0
1
y1
y2
.
=
J(f (y1 , y2 ), g(y1, y2 )) =
g
g
2y1 1 c
y
y
1

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).
Igor Yanovsky, 2005
167
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.
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 ).
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

.
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
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
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 =

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

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.
Igor Yanovsky, 2005
173
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
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
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.
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.
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.
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.
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
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)
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.
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 .

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

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.
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.
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
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
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
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.
Igor Yanovsky, 2005
191
area of intersection of the sphere of radius ct with the support of the functions g and
h.
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
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
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
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
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.

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, )).
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( ).
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

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
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.
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

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.
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
17.2
Igor Yanovsky, 2005
205
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
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.
Igor Yanovsky, 2005
207
Thus,
w = c1 er + c2 er ,
er
er
1
w(r) = c1 + c2
.
K =
r
r
r
= 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)

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
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|).
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) .
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
= 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)

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
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|).
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)
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.
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.
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 ) =
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.
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
= 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

+
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)


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 =

Igor Yanovsky, 2005

K(|x|) v dx = lim
0
K(|x|)2 v dx = v(0).
215
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,

t
R(t) = R0 e 2 .
Igor Yanovsky, 2005
217
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
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 )
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.
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
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 ;

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().
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

u
ds
u v dx =
v
v u dx.

|w|2 dx =
=0,
ds
w
n

Dirichlet or Neumann
w w dx.

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?
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.
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,
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

u
ds
u v dx =
v
v u dx.
D
D n
D

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.
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
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| .
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).
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.
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)
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
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
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
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
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 )
Igor Yanovsky, 2005
238
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| .

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).
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 .
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
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
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
Igor Yanovsky, 2005
244
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

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.
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

57
57
Check with someone about the last result.
Igor Yanovsky, 2005
248
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
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.
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.
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 .
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
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}.
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.

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
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.

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).
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.
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.
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
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.
=


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
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
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.

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.
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)
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.
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 < .
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.
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).
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.
19
Igor Yanovsky, 2005
271
Contraction Mapping and Uniqueness - Wave
Recall that the solution to

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.
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.
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
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||
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.
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
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
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.
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.
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)
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 .
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.
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

2
a
n sinh n =
g(x) sin nx dx.

0
Igor Yanovsky, 2005
283
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 < .
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.

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.
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
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,
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
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 .
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.
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).
u(r, ) = a
0 +
r n (
an cos n + bn sin n).
n=1

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,
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.
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.

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.
Y ()
X (t)
=
= .
X(t)
Y ()
n = n2 , n = 0, 1, 2, . . ..
X0 (r) = c0 log r + d0 .
If n = 0, X0 (t) = c0 t + d0 .
We have
u0 (r, ) = X0 (r)Y0 () = (c0 log r + d0 )a0 ,
u0 (r, ) = d0 a0 ,
un (r, ) = dn r n (an cos n + bn sin n).
u(r, ) = a
0 +
r n (
n=1

u(1, ) = a
0 +

n=1
(
an cos n + bn sin n) = g().
289

a
0 =
an =
bn =

1
g() d,
0

2
g() cos n d,
0

2
g() sin n d.
0
Igor Yanovsky, 2005
290
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.
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.
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 (
n=1

a0 +
0 = ur (1, ) u(1, ) =
(n 1)(
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 .
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.
utt + u = 0.
Y ()
X (t)
=
= .
X(t)
Y ()


n = n
X0 (r) = c0 log r + d0 .
If n = 0, X0 (t) = c0 t + d0 .
n = n2 .
Igor Yanovsky, 2005
293

u must be nite at r = 0 cn = 0, n = 0, 1, 2, . . ..
u(r, ) = a
0 +
r 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, )
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?

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.
utt + u = 0.
Y ()
X (t)
=
= .
X(t)
Y ()


n = n
X0 (r) = c0 log r + d0 .
If n = 0, X0 (t) = c0 t + d0 .
u must be nite at r = c0 = 0, dn = 0, n = 1, 2, . . ..
u(r, ) = a
0 +
r n (
n=1

f () = u(1, ) = a
0 +

n=1
(
n = n2 .
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
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.

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.
Y ()
X (t)
=
= .
X(t)
Y ()

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, . . ..
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.
u(r, ) =
4
r
sin .
3r 3
Igor Yanovsky, 2005
297
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 < < .

1
1
urr + ur + 2 u = 0
r
r
r 2 urr + rur + u = 0.
for 1 < r < 2, 0 < < ,
utt + u = 0.
Y ()
X (t)
=
= .
X(t)
Y ()
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, . . ..
X0 (r) = c0 log r + d0 .
If n = 0, X0 (t) = c0 t + d0 .
We have,
u(r, ) =
Xn (r)Yn () =
n=1
(
cn r n + dn r n ) sin n.
n=1

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
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
utt + u = 0.
Y ()
X (t)
=
= .
X(t)
Y ()

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
ur (r, ) =
c0
+
(2n
cn r 2n1 + 2ndn r 2n1 ) cos(2n).
r
n=1
bn = 0.
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.
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.
utt + u = 0.
Y ()
X (t)
=
= .
X(t)
Y ()
n = n2 , n = 1, 2, . . ..
X0 (r) = c0 log r + d0 .
If n = 0, X0 (t) = c0 t + d0 .
We have
u0 (r, ) = X0 (r)Y0 () = (c0 log r + d0 )a0 ,
0 ,
u0 (r, ) = a
un (r, ) = r n (
u(r, ) = a
0 +
r n (
n=1

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
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.

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.
utt + u = 0.

Y ()
X (t)
=
= .
X(t)
Y ()

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
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
Igor Yanovsky, 2005
can not be used.

72
72
ChiuYens solutions have attempts to solve this problem using Greens function.
304
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

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.
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

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
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.
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

f (x) = u(x, 0) =
0 = ut(x, 0) =
cn sin nx.
n=1

dn ( 1 + n2 ) sin nx.
n=1

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.
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.

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.


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
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
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.
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).
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.
T
X
=
= .
X
T


vx (0, t) = X (0)T (t) = 0
X (0) = X (L) = 0.

vx (L, t) = X (L)T (t) = 0
(25.4)
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

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

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.
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).
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
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.

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),

T
X
=
= .
X
T
Igor Yanovsky, 2005
316

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

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
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 .
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.


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
Igor Yanovsky, 2005
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
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

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.
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.
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
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.
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
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.

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 .
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.

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 .


Thus, w 0 and the solution u(x, y) 0 is unique.
Igor Yanovsky, 2005
327
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.
(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.
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
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
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.
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 .
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

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)).
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
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.
X Y
+
+ = 0.
X
Y
Y:
X + 2 X = 0
Y + 2 Y = 0
X(0) = X(a) = 0
Y (0) = Y (b) = 0.

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).
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
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

with

an =
bn =
g(x)n(x) dx,
h(x)n (x) dx.
Igor Yanovsky, 2005
339
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.
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.
X Y
+
+ = 0.
X
Y
Y:
X + 2 X = 0
Y + 2 Y = 0
X(0) = X(a) = 0
Y (0) = Y (b) = 0.

m
n
n =
m =
a
b
mx
ny
Yn (y) = sin
,
Xm (x) = sin
a
b
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).
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
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
Igor Yanovsky, 2005
343
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 .
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 .
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.
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
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.
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.
X Y
+
+ = 0.
X
Y
Y:
X + 2 X = 0
Y + 2 Y = 0
X(0) = X(a) = 0
Y (0) = Y (b) = 0.

m
n
n =
m =
a
b
mx
ny
Yn (y) = sin
,
Xm (x) = sin
a
b
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
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).
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
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.
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.
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).

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=
Thus,
2 +n2 )t
cmn e(m
eimx einy .
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 .
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.
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
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
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.
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
u
n ds = (n u n u) dx.
n
= 0;
= 0 since eigenfunctions are Neumann.
On , u
n
n
78
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
Igor Yanovsky, 2005
b) If g 0 and f > 0, show that u > 0 for all t > 0.
356
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.
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
u
ds = (n u n u) dx.
n
n
= 0;
= 0 since eigenfunctions are Neumann.
On , u
n
n
80
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).
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).
Igor Yanovsky, 2005
b) For 0 = 1, the solution stays bounded as t .
361
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
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
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 .
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
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
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
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
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
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).

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.
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.
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
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
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
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.
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)
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)
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.
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
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
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) = ||
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
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.
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

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 .
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
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
Igor Yanovsky, 2005

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
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
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.
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
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
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

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
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

Partial Differential Equations: Graduate Level Problems and Solutions

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Partial Differential Equations: Graduate Level Problems and Solutions

Uploaded by

Copyright:

Available Formats

Partial Dierential Equations: Graduate Level Problems

Partial Dierential Equations

Igor Yanovsky, 2005

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

Partial Dierential Equations

Igor Yanovsky, 2005

2 Simple Eigenvalue Problem

4 Eigenvalues of the Laplacian:

Partial Dierential Equations

Igor Yanovsky, 2005

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

11 Problems: Quasilinear Equations

13 Problems: General Nonlinear Equations

15 Problems: Gas Dynamics Systems

16 Problems: Wave Equation

17 Problems: Laplace Equation

Partial Dierential Equations

Igor Yanovsky, 2005

18 Problems: Heat Equation

20 Contraction Mapping and Uniqueness - Heat

21 Problems: Maximum Principle - Laplace and Heat

23 Problems: Separation of Variables - Poisson Equation

24 Problems: Separation of Variables - Wave Equation

25 Problems: Separation of Variables - Heat Equation

26 Problems: Eigenvalues of the Laplacian - Laplace

27 Problems: Eigenvalues of the Laplacian - Poisson

28 Problems: Eigenvalues of the Laplacian - Wave

29 Problems: Eigenvalues of the Laplacian - Heat

32 Linear Functional Analysis

Partial Dierential Equations

cos 2t = cos2 t sin2 t

Igor Yanovsky, 2005

sin 2t = 2 sin t cos t

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

Partial Dierential Equations

Igor Yanovsky, 2005

Partial Dierential Equations

Igor Yanovsky, 2005

Simple Eigenvalue Problem

Partial Dierential Equations

Igor Yanovsky, 2005

Eigenvalues of the Laplacian: Quick Guide

utt + 3ut + u = uxx .

Partial Dierential Equations

Igor Yanovsky, 2005

with initial conditions

In particular, at t = 0 we obtain the condition

The characteristic equations are

y and t are interchanged here

Partial Dierential Equations

Igor Yanovsky, 2005

From (5.2), we have

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

Partial Dierential Equations

Igor Yanovsky, 2005

Weak Solutions for Quasilinear Equations

Consider shocks for an equation

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

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

Fans and Rarefaction Waves

For Burgers equation

Partial Dierential Equations

Igor Yanovsky, 2005