ODE Yanovsky

Ordinary Dierential Equations: Graduate Level Problems
and Solutions
Igor Yanovsky
1
Ordinary Dierential Equations Igor Yanovsky, 2005 2
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.
Contents
1 Preliminaries 5
1.1 Gronwall Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Trajectories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Linear Systems 7
2.1 Existence and Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Fundamental Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2.1 Distinct Eigenvalues or Diagonalizable . . . . . . . . . . . . . . . 7
2.2.2 Arbitrary Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Asymptotic Behavior of Solutions of Linear Systems with Constant Co-
ecients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4 Variation of Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.5 Classication of Critical Points . . . . . . . . . . . . . . . . . . . . . . . 12
2.5.1 Phase Portrait . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.7 Stability and Asymptotic Stability . . . . . . . . . . . . . . . . . . . . . 23
2.8 Conditional Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.9 Asymptotic Equivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.9.1 Levinson . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3 Lyapunovs Second Method 27
3.1 Hamiltonian Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2 Lyapunovs Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2.1 Stability (Autonomous Systems) . . . . . . . . . . . . . . . . . . 29
3.3 Periodic Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4 Invariant Sets and Stability . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.5 Global Asymptotic Stability . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.6 Stability (Non-autonomous Systems) . . . . . . . . . . . . . . . . . . . . 41
3.6.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4 Poincare-Bendixson Theory 42
5 Sturm-Liouville Theory 48
5.1 Sturm-Liouville Operator . . . . . . . . . . . . . . . . . . . . . . . . . . 48
5.2 Existence and Uniqueness for Initial-Value Problems . . . . . . . . . . . 48
5.3 Existence of Eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
5.4 Series of Eigenfunctions . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.5 Lagranges Identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.6 Greens Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.7 Self-Adjointness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.8 Orthogonality of Eigenfunctions . . . . . . . . . . . . . . . . . . . . . . . 66
5.9 Real Eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.10 Unique Eigenfunctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.11 Rayleigh Quotient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.12 More Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
6 Variational (V) and Minimization (M) Formulations 97
7 Euler-Lagrange Equations 103
7.1 Rudin-Osher-Fatemi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
7.1.1 Gradient Descent . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
7.2 Chan-Vese . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
7.3 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
8 Integral Equations 110
8.1 Relations Between Dierential and Integral Equations . . . . . . . . . . 110
8.2 Greens Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
9 Miscellaneous 119
10 Dominant Balance 124
11 Perturbation Theory 125
1 Preliminaries
Cauchy-Peano.
_
du
dt
= f(t, u) t
0
t t
1
u(t
0
) = u
0
(1.1)
f(t, u) is continuous in the rectangle R = (t, u) : t
0
t t
0
+ a, [u u
0
[ b.
M = max
R
[f(t, u)[, and = min(a,
b
M
). Then u(t) with continuous rst derivative
s.t. it satises (1.1) for t
0
t t
0
+.
Local Existence via Picard Iteration.
f(t, u) is continuous in the rectangle R = (t, u) : t
0
t t
0
+ a, [u u
0
[ b.
Assume f is Lipschitz in u on R.
[f(t, u) f(t, v)[ L[u v[
M = max
R
[f(t, u)[, and = min(a,
b
M
). Then a unique u(t), with u,
du
dt
continuous
on [t
0
, t
0
+ ], (0, ] s.t. it satises (1.1) for t
0
t t
0
+ .
Power Series.
du
dt
= f(t, u)
u(0) = u
0
u(t) =
j=0
1
j!
d
j
u
dt
j
(0)t
j
i.e.
d
2
u
dt
2
(0) = (f
t
+ f
u
f)[
0
Fixed Point Iteration.
[x
n
x
[ k
n
[x
0
x
[ k < 1
[x
n+1
x
n
[ k
n
[x
1
x
0
[ k < 1
[x
x
n
[ = lim
m
[x
m
x
n
[ k
n
(1 + k +k
2
+ )[x
1
x
0
[ =
k
n
1 k
[x
1
x
0
[
Picard Iteration. Approximates (1.1). Initial guess: u
0
(t) = u
0
u
n+1
(t) = Tu
n
(t) = u
0
+
t
_
t
0
f(s, u
n
(s))ds.
Dierential Inequality. v(t) piecewise continuous on t
0
t t
0
+ a.
u(t) and
du
dt
continuous on some interval. If
du
dt
v(t)u(t)
u(t) u(t
0
)e
t
_
t
0
v(s)ds
Proof. Multiply both sides by e
t
_
t
0
v(s)ds
. Then
d
dt
[e
t
_
t
0
v(s)ds
u(t)] 0.
1.1 Gronwall Inequality
Gronwall Inequality. u(t), v(t) continuous on [t
0
, t
0
+a]. v(t) 0, c 0.
u(t) c +
t
_
t
0
v(s)u(s)ds
u(t) c e
t
_
t
0
v(s)ds
t
0
t t
0
+ a
Proof. Multiply both sides by v(t):
u(t)v(t) v(t)
_
c +
t
_
t
0
v(s)u(s)ds
_
Denote A(t) = c +
t
_
t
0
v(s)u(s)ds
dA
dt
v(t)A(t). By dierential inequality and
hypothesis:
u(t) A(t) A(t
0
)e
t
_
t
0
v(s)ds
= ce
t
_
t
0
v(s)ds
.
Error Estimates. f(t, u(t)) continuous on R = (t, u) : [t t
0
[ a, [u u
0
[ b
f(t, u(t)) Lipschitz in u: [f(t, A) f(t, B)[ L[A B[
u
1
(t), u
2
(t) are
1
,
2
approximate solutions
du
1
dt
= f(t, u
1
(t)) + R
1
(t), [R
1
(t)[
1
du
2
dt
= f(t, u
2
(t)) + R
2
(t), [R
2
(t)[
2
[u
1
(t
0
) u
2
(t
0
)[
[u
1
(t) u
2
(t)[ ( + a(
1
+
2
))e
aL
t
0
t t
0
+a
Generalized Gronwall Inequality. w(s), u(s) 0
u(t) w(t) +
t
_
t
0
v(s)u(s)ds
u(t) w(t) +
t
_
t
0
v(s)w(s) e
t
_
s
v(x)dx
ds
Improved Error Estimate (Fundamental Inequality).
[u
1
(t) u
2
(t)[ e
L(tt
0
)
+
(
1
+
2
)
L
(e
L(tt
0
)
1)
1.2 Trajectories
Let K D compact. If for the trajectory Z = (t, z(t)) : < t < ) we have that
< , then Z lies outside of K for all t suciently close to .
2 Linear Systems
2.1 Existence and Uniqueness
A(t), g(t) continuous, then can solve
y
= A(t)y +g(t) (2.1)

y(t
0
) = y
0
For uniqueness, need RHS to satisfy Lipshitz condition.
2.2 Fundamental Matrix
A matrix whose columns are solutions of y
= A(t)y is called a solution matrix.

A solution matrix whose columns are linearly independent is called a fundamental
matrix.
F(t) is a fundamental matrix if:
1) F(t) is a solution matrix;
2) det F(t) ,= 0.
Either det M(t) ,= 0 t R, or det M(t) = 0 t R.
F(t)c is a solution of (2.1), where c is a column vector.
If F(t) is a fundamental matrix, can use it to solve:
y
(t) = A(t)y(t), y(t

0
) = y
0
i.e. since F(t)c[
t
0
= F(t
0
)c = y
0
c = F
1
(t
0
)y
0

y(t) = F(t)F(t
0
)
1
y
0
2.2.1 Distinct Eigenvalues or Diagonalizable
F(t) = [e
1
t
v
1
, . . . , e
nt
v
n
] e
At
= F(t)C
2.2.2 Arbitrary Matrix
i) Find generalized eigenspaces X
j
= x : (A
j
I)
n
j
x = 0;
ii) Decompose initial vector = v
1
+ + v
k
, v
j
X
j
,
solve for v
1
, . . . , v
k
in terms of components of
y(t) =
k
j=1
e
j
t
_
n
j
1
i=0
t
i
i!
(A
j
I)
i
_
v
j
(2.2)
iii) Plug in = e
1
, . . . , e
n
successively to get y
1
(t), . . . , y
n
(t) columns of F(t).
Note: y(0) = , F(0) = I.
2.2.3 Examples
Example 1. Show that the solutions of the following system of dierential equations
remain bounded as t :
u
= v u
v
= u
Proof. 1)
_
u
v
_
=
_
1 1
1 0
__
u
v
_
. The eigenvalues of A are
1,2
=
1
2

3
2
i, so
the eigenvalues are distinct diagonalizable. Thus, F(t) = [e
1
t
v
1
, e
2
t
v
2
] is a funda-
mental matrix. Since Re(
i
) =
1
2
< 0, the solutions to y
= Ay remain bounded as
t .
2) u
= v
= u u
,
u
+u
+ u = 0,
u
+ (u
)
2
+u
u = 0,
1
2
d
dt
((u
)
2
) + (u
)
2
+
1
2
d
dt
(u
2
) = 0,
1
2
((u
)
2
) +
1
2
(u
2
) +
_
t
t
0
(u
)
2
dt = const,
1
2
((u
)
2
) +
1
2
(u
2
) const,
(u
, u) is bounded.
Example 2. Let A be the matrix given by: A =
1 0 3
2 1 2
0 0 2
. Find the eigenvalues,

the generalized eigenspaces, and a fundamental matrix for the system y
(t) = Ay.
Proof. det(AI) = (1 )
2
(2 ). The eigenvalues and their multiplicities:
1
= 1, n
1
= 2;
2
= 2, n
2
= 1.
Determine subspaces X
1
and X
2
, (A
j
I)
n
j
x = 0.
(AI)
2
x = 0 (A 2I)x = 0
To nd X
1
:
(AI)
2
x =
0 0 3
2 0 2
0 0 1
0 0 3
2 0 2
0 0 1
x =
0 0 3
0 0 8
0 0 1
x
1
x
2
x
3
0
0
0
.
x
3
= 0, x
1
, x
2
arbitrary X
1
=
_
, any , C
_
. dimX
1
= 2.
To nd X
2
:
(A2I)x =
1 0 3
2 1 2
0 0 0
x =
1 0 3
0 1 8
0 0 0
x
1
x
2
x
3
0
0
0
.
x
3
= , x
1
= 3, x
2
= 8 X
2
=
_
3
8
1
, any C
_
. dimX
2
= 1.
Need to nd v
1
X
1
, v
2
X
2
, such that initial vector is decomposed as = v
1
+v
2
.
3
8
.
v
1
=
1
3
3
2
8
3
0
, v
2
=
3
3
8
3
.
y(t) =
k
j=1
e
j
t
_
n
j
1
i=0
t
i
i!
(A
j
I)
i
_
v
j
= e
1
t
(I +t(A I))v
1
+e
2
t
v
2
= e
t
(I + t(A I))v
1
+ e
2t
v
2
= e
t
(I + t(AI))
1
3
3
2
8
3
0
+e
2t
3
3
8
3
= e
t
1 0 3t
2t 1 2t
0 0 1 +t
1
3
3
2
8
3
0
+e
2t
3
3
8
3
.
Note: y(0) = =
.
To nd a fundamental matrix, putting successively equal to
1
0
0
0
1
0
0
0
1
in this formula, we obtain the three linearly independent solutions that we use as
columns of the matrix. If =
1
0
0
, y
1
(t) = e
t
1
2t
0
. If =
0
1
0
, y
2
(t) =
e
t
0
1
0
.
If =
0
0
1
, y
3
(t) = e
t
3
6t 8
0
+e
2t
3
8
1
. The fundamental matrix is

F(t) = e
At
=
e
t
0 3e
t
+ 3e
2t
2te
t
e
t
(6t 8)e
t
+ 8e
2t
0 0 e
2t
Note: At t = 0, F(t) reduces to I.

2.3 Asymptotic Behavior of Solutions of Linear Systems with Con-
stant Coecients
If all
j
of A are such that Re(
j
) < 0, then every solution (t) of the system y
= Ay
approaches zero as t . [(t)[

Ke
t
or [e
At
[ Ke
t
.
If, in addition, there are
j
such that Re(
j
) = 0 and are simple, then [e
At
[ K, and
hence every solution of y
= Ay is bounded.
Also, see the section on Stability and Asymptotic Stability.
Proof.
1
,
2
, . . . ,
k
are eigenvalues and n
1
, n
2
, . . . , n
k
are their corresponding multi-
plicities. Consider (2.2), i.e. the solution y satisfying y(0) = is
y(t) = e
tA
=
k
j=1
e
j
t
_
n
j
1
i=0
t
i
i!
(A
j
I)
i
_
v
j
.
Subdivide the right hand side of equality above into two summations, i.e.:
1)
j
, s.t. n
j
= 1, Re(
j
) 0;
2)
j
, s.t. n
j
2, Re(
j
) < 0.
y(t) =
k
j=1
e
j
t
v
j
. .
(n
j
=1) Re(
j
)0
+
k
j=1
e
j
t
_
I + t(A
j
I) + +
t
n
j
1
(n
j
1)!
(A
j
I)
n
j
1
v
j
. .
(n
j
2) Re(
j
)<0
.
[y(t)[
k
j=1
[e
j
t
I[[v
j
[
. .
Re(
j
)0
+

Ke
t
. .
=max(Re(
j
), Re(
j
)<0)
c
k
j=1
[v
j
[ +

Ke
t
ck max
j
[v
j
[ +

Ke
t
max(ck,

K)
. .
const indep of t
_
max
j
[v
j
[
. .
indep of t
+ e
t
..
0 as t
_
K.
2.4 Variation of Constants
Derivation: Variation of constants is a method to determine a solution of y
= A(t)y+
g(t), provided we know a fundamental matrix for the homogeneous system y
= A(t)y.
Let F be a fundamental matrix. Look for solution of the form (t) = F(t)v(t), where
v is a vector to be determined. (Note that if v is a constant vector, then satises
the homogeneous system and thus for the present purpose v(t) c is ruled out.)
Substituting (t) = F(t)v(t) into y
= A(t)y + g(t), we get
(t) = F
(t)v(t) + F(t)v
(t) = A(t)F(t)v(t) +g(t)

Since F is a fundamental matrix of the homogeneous system, F
(t) = A(t)F(t). Thus,

F(t)v
(t) = g(t),
v
(t) = F
1
(t)g(t),
v(t) =
_
t
t
0
F
1
(s)g(s)ds.
Therefore, (t) = F(t)
_
t
t
0
F
1
(s)g(s)ds.
Variation of Constants Formula: Every solution y of y
= A(t)y + g(t) has the

form:
y(t) =
h
(t) +
p
(t) = F(t)c + F(t)
t
_
t
0
F
1
(s)g(s)ds
where
p
is the solution satisfying initial condition
p
(t
0
) = 0 and
h
(t) is that solution
of the homogeneous system satisfying the same initial condition at t
0
as y,
h
(t
0
) = y
0
.
F(t) = e
At
is the fundamental matrix of y
= Ay with F(0) = I. Therefore, every

solution of y
= Ay has the form y(t) = e

At
c for a suitably chosen constant vector c.
y(t) = e
(tt
0
)A
y
0
+
t
_
t
0
e
(ts)A
g(s)ds
That is, to nd the general solution of (2.1), use (2.2) to get a fundamental matrix
F(t).
Then, add
t
_
t
0
e
(ts)A
g(s)ds = F(t)
t
_
t
0
F
1
(s)g(s)ds to F(t)c.
2.5 Classication of Critical Points
y
= Ay. Change of variable y = Tz, where T is nonsingular constant matrix (to be

determined). z
= T
1
ATz The solution is passing through (c
1
, c
2
) at t = 0.
1)
1
,
2
are real. z
=
_

1
0
0
2
_
z
z =
_
c
1
e
1
t
c
2
e
2
t
_
a)
2
>
1
> 0 z
2
(t) = c(z
1
(t))
p
, p > 1 Improper Node (tilted toward z
2
-axis)
b)
2
<
1
< 0 z
2
(t) = c(z
1
(t))
p
, p > 1 Improper Node (tilted toward z
2
-axis)
c)
2
=
1
, A diagonalizable z
2
= cz
1
Proper Node
d)
2
< 0 <
1
z
1
(t) = c(z
2
(t))
p
, p < 0 Saddle Point
2)
2
=
1
, A non-diagonalizable, z
=
_
1
0
_
z
z =
_
e
t
te
t
0 e
t
__
c
1
c
2
_
=
_
c
1
+ c
2
t
c
2
_
e
t
Improper Node
3)
1,2
= i. z
=
_

_
z
z = e
t
_
c
1
cos(t) +c
2
sin(t)
c
1
sin(t) +c
2
cos(t)
_
Spiral Point
2.5.1 Phase Portrait
Locate stationary points by setting:
du
dt
= f(u, v) = 0
dv
dt
= g(u, v) = 0
(u
0
, v
0
) is a stationary point. In order to classify a stationary point, need to nd
eigenvalues of a linearized system at that point.
J(f(u, v), g(u, v)) =
_
f
u
f
v
g
u
g
v
_
.
Find
j
s such that det(J[
(u
0
,v
0
)
I) = 0.
2.6 Problems
Problem (F92, #4). Consider the autonomous dierential equation
v
xx
+v v
3
v
0
= 0
in which v
0
is a constant.
a) Show that for v
2
0
<
4
27
, this equation has 3 stationary points and classify their type.
b) For v
0
= 0, draw the phase plane for this equation.
Proof. a) We have
v
+v v
3
v
0
= 0.
In order to nd and analyze the stationary points of an ODE above, we write it as a
rst-order system.
y
1
= v,
y
2
= v
.
y
1
= v
= y
2
= 0,
y
2
= v
= v +v
3
+v
0
= y
3
1
y
1
+v
0
= 0.
The function f(y
1
) = y
3
1
y
1
= y
1
(y
2
1
1) has zeros y
1
= 0, y
1
= 1, y
1
= 1.
See the gure.
Its derivative f
(y
1
) = 3y
2
1
1 has zeros y
1
=
1
3
, y
1
=
1
3
.
At these points, f(
1
3
) =
2
3
3
, f(
1
3
) =
2
3
3
.
If v
0
= 0, y
2
is exactly this function f(y
1
), with 3 zeros.
v
0
only raises or lowers this function. If [v
0
[ <
2
3
3
,
i.e. v
2
0
<
4
27
, the system would have 3 stationary points:
Stationary points: (p
1
, 0), (p
2
, 0), (p
3
, 0),
with p
1
< p
2
< p
3
.
y
1
= y
2
= f(y
1
, y
2
),
y
2
= y
3
1
y
1
+v
0
= g(y
1
, y
2
).
In order to classify a stationary point, need to nd eigenvalues of a linearized system
at that point.
J(f(y
1
, y
2
), g(y
1
, y
2
)) =
_
f
y
1
f
y
2
g
y
1
g
y
2
_
=
_
0 1
3y
2
1
1 0
_
.
For (y
1
, y
2
) = (p
i
, 0) :
det(J[
(p
i
,0)
I) =
1
3p
2
i
1
=
2
3p
2
i
+ 1 = 0.
=
_
3p
2
i
1.
At y
1
= p
1
<
1
3
,
< 0 <
+
. (p
1
,0) is Saddle Point.
At
1
3
< y
1
= p
2
<
1
3
,
C, Re(
) = 0. (p
2
,0) is Stable Concentric
Circles.
At y
1
= p
3
>
1
3
,
< 0 <
+
. (p
3
,0) is Saddle Point.
b) For v
0
= 0,
y
1
= y
2
= 0,
y
2
= y
3
1
y
1
= 0.
Stationary points: (1, 0), (0, 0), (1, 0).
J(f(y
1
, y
2
), g(y
1
, y
2
)) =
_
0 1
3y
2
1
1 0
_
.
For (y
1
, y
2
) = (0, 0) :
det(J[
(0,0)
I) =
1
1
=
2
+ 1 = 0.
= i.
(0,0) is Stable Concentric Circles (Center).
For (y
1
, y
2
) = (1, 0) :
det(J[
(1,0)
I) =
1
2
=
2
2 = 0.
2.
(-1,0) and (1,0) are Saddle Points.
Problem (F89, #2). Let V (x, y) = x
2
(x 1)
2
+y
2
. Consider the dynamical system
dx
dt
=
V
x
,
dy
dt
=
V
y
.
a) Find the critical points of this system and determine their linear stability.
b) Show that V decreases along any solution of the system.
c) Use (b) to prove that if z
0
= (x
0
, y
0
) is an isolated minimum of V then z
0
is an
asymptotically stable equilibrium.
Proof. a) We have
x
= 4x
3
+ 6x
2
2x
y
= 2y.
_
x
= x(4x
2
6x + 2) = 0
y
= 2y = 0.
Stationary points: (0, 0),
_
1
2
, 0
_
, (1, 0).
J(f(y
1
, y
2
), g(y
1
, y
2
)) =
_
f
x
f
y
g
x
g
y
_
=
_
12x
2
+ 12x 2 0
0 2
_
.
For (x, y) = (0, 0) :
det(J[
(0,0)
I) =
2 0
0 2
= (2 )(2 ) = 0.
y
= Ay,
1
=
2
< 0, A diagonalizable.
(0,0) is Stable Proper Node.
For (x, y) =
_
1
2
, 0
_
:
det(J[
(
1
2
,0)
I) =
1 0
0 2
= (1 )(2 ) = 0.
1
= 2,
2
= 1.
1
< 0 <
2
.
_
1
2
,0
_
is Unstable Saddle Point.
For (x, y) = (1, 0) :
det(J[
(1,0)
I) =
2 0
0 2
= (2 )(2 ) = 0.
y
= Ay,
1
=
2
< 0, A diagonalizable.
(1,0) is Stable Proper Node.
b) Show that V decreases along any solution of the system.
dV
dt
= V
x
x
t
+V
y
y
t
= V
x
(V
x
) + V
y
(V
y
) = V
2
x
V
2
y
< 0.
c) Use (b) to prove that if z
0
= (x
0
, y
0
) is an isolated minimum of V then z
0
is an
asymptotically stable equilibrium.
Lyapunov Theorem: If V (y) that is positive denite and for which V
(y) is negative
denite in a neighborhood of 0, then the zero solution is asymptotically stable.
Let W(x, y) = V (x, y) V (x
0
, y
0
). Then, W(x
0
, y
0
) = 0.
W(x, y) > 0 in a neighborhood around (x
0
, y
0
), and
dW
dt
(x, y) < 0 by (b). (
dV
dt
(x, y) < 0
and
dV
dt
(x
0
, y
0
) = 0).
(x
0
, y
0
) is asymptotically stable.
Problem (S98, #1). Consider the undamped pendulum, whose equation is
d
2
p
dt
2
+
g
l
sin p = 0.
a) Describe all possible motions using a phase plane analysis.
b) Derive an integral expression for the period of oscillation at a xed energy E,
and nd the period at small E to rst order.
c) Show that there exists a critical energy for which the motion is not periodic.
Proof. a) We have
y
1
= p
y
2
= p
.
y
1
= p
= y
2
= 0
y
2
= p
=
g
l
sin p =
g
l
sin y
1
= 0.
Stationary points: (n, 0).
y
1
= y
2
= f(y
1
, y
2
),
y
2
=
g
l
siny
1
= g(y
1
, y
2
).
J(f
1
(y
1
, y
2
), f
2
(y
1
, y
2
)) =
_
f
1
y
1
f
1
y
2
f
2
y
1
f
2
y
2
_
=
_
0 1
g
l
cos y
1
0
_
.
For (y
1
, y
2
) = (n, 0), n-even:
det(J[
(n,0)
I) =
g
l

=
2
+
g
l
= 0.
i
_
g
l
C, g > 0, (n,0), n-even, are Stable Centers.
g
l
R, g < 0. (n,0), n-even, are Unstable Saddle Points.
For (y
1
, y
2
) = (n, 0), n-odd:
det(J[
(n,0)
I) =
1
g
l

=
2
g
l
= 0.
_
g
l
R, g > 0, (n,0), n-odd, are Unstable Saddle Points.
i
_
g
l
C, g < 0, (n,0), n-odd, are Stable Centers.
b) We have
p
+
g
l
sin p = 0,
p
+
g
l
p
sin p = 0,
1
2
d
dt
(p
)
2
g
l
d
dt
(cos p) = 0,
1
2
(p
)
2
g
l
cos p =

E.
E =
1
2
(p
)
2
+
g
l
(1 cos p).
Since we assume that [p[ is small, we could replace sinp by p, and perform similar
calculations:
p
+
g
l
p = 0,
p
+
g
l
p
p = 0,
1
2
d
dt
(p
)
2
+
1
2
g
l
d
dt
(p)
2
= 0,
1
2
(p
)
2
+
1
2
g
l
p
2
= E
1
,
(p
)
2
+
g
l
p
2
= E = constant.
Thus,
(p
)
2
E
+
p
2
lE
g
= 1,
which is an ellipse with radii

E on p
-axis, and
_
lE
g
on p-axis.
We derive an Integral Expression for the Period of oscillation at a xed energy E.
Note that at maximum amplitude (maximum displacement), p
= 0.
Dene p = p
max
to be the maximum displacement:
E =
1
2
(p
)
2
+
g
l
(1 cos p),
p
=
_
2E
2g
L
(1 cos p),
_ T
4
0
p
_
2E
2g
L
(1 cos p)
dt =
_ T
4
0
dt =
T
4
,
T = 4
_ T
4
0
p
_
2E
2g
L
(1 cos p)
dt.
_
T = 4
_
pmax
0
dp
_
2E
2g
L
(1 cos p)
_
Making change of variables: = p(t), d = p
(t)dt, we obtain
T(p
max
) = 4
_
pmax
0
d
_
2E
2g
L
(1 cos )
.
Problem (F94, #7).
The weakly nonlinear approximation to the pendulum equation ( x = sinx) is
x = x +
1
6
x
3
. (2.3)
a) Draw the phase plane for (2.3).
b) Prove that (2.3) has periodic solutions x(t) in the neighborhood of x = 0.
c) For such periodic solutions, dene the amplitude as a = max
t
x(t). Find an integral
formula for the period T of a periodic solution as a function of the amplitude a.
d) Show that T is a non-decreasing function of a.
Hint: Find a rst integral of equation (2.3).
Proof. a)
y
1
= x
y
2
= x
.
y
1
= x
= y
2
= 0
y
2
= x
= x +
1
6
x
3
= y
1
+
1
6
y
3
1
= 0.
Stationary points: (0, 0), (
6, 0), (
6, 0).
y
1
= y
2
= f(y
1
, y
2
),
y
2
= y
1
+
1
6
y
3
1
= g(y
1
, y
2
).
J(f(y
1
, y
2
), g(y
1
, y
2
)) =
_
f
y
1
f
y
2
g
y
1
g
y
2
_
=
_
0 1
1 +
1
2
y
2
1
0
_
.
For (y
1
, y
2
) = (0, 0):
det(J[
(0,0)
I) =
1
1
=
2
+ 1 = 0.
= i. (0,0) is Stable Center.

For (y
1
, y
2
) = (
6, 0):
det(J[
(
6,0)
I) =
1
2
=
2
2 = 0.
2. (
6,0) are Unstable Saddle Points.

b) Prove that x = x+
1
6
x
3
has periodic solutions x(t) in the neighborhood of x = 0.
We have
x = x +
1
6
x
3
,
x x = xx +
1
6
x
3
x,
1
2
d
dt
( x
2
) =
1
2
d
dt
(x
2
) +
1
24
d
dt
(x
4
),
d
dt
_
x
2
+x
2
1
12
x
4
_
= 0.
E = x
2
+x
2
1
12
x
4
.
Thus the energy is conserved.
For E > 0 small enough, consider x =
_
E x
2
+
1
12
x
4
. For small E, x
E.
Thus, there are periodic solutions in a neighborhood of 0.
c) For such periodic solutions, dene the amplitude as a = max
t
x(t). Find an Integral
Formula for the Period T of a periodic solution as a function of the amplitude a.
Note that at maximum amplitude, x = 0. We have
E = x
2
+ x
2
1
12
x
4
,
x =
_
E x
2
+
1
12
x
4
,
_ T
4
0
x
_
E x
2
+
1
12
x
4
dt =
_ T
4
0
dt =
T
4
,
T = 4
_ T
4
0
x
_
E x
2
+
1
12
x
4
dt.
Making change of variables: = x(t), d = x(t)dt, we obtain
T(a) = 4
_
a
0
d
_
E
2
+
1
12
4
.
d) Show that T is a non-decreasing function of a.
dT
da
= 4
d
da
_
a
0
d
_
E
2
+
1
12
4
.
Problem (S91, #1). Consider the autonomous ODE
d
2
x
dt
2
+ sinx = 0.
a) Find a nontrivial function H(x,
dx
dt
) that is constant along each solution.
1
b) Write the equation as a system of 2 rst order equations. Find all of the stationary
points and analyze their type.
c) Draw a picture of the phase plane for this system.
Proof. a) We have
x + sin x = 0.
Multiply by x and integrate:
x x + xsin x = 0,
1
2
d
dt
( x
2
) +
d
dt
(cos x) = 0,
x
2
2
cos x = C,
H(x, x) =
x
2
2
cos x.
H(x, x) is constant along each solution. Check:
d
dt
H(x, x) =
H
x
x +
H
x
x = (sinx) x + x(sinx) = 0.
b,c)
2
1
Note that H does not necessarily mean that it is a Hamiltonian.
2
See S98 #1a.
2.7 Stability and Asymptotic Stability
y
= f(y) (2.4)
An equilibrium solution y
0
of (2.4) is stable if , () such that whenever any
solution (t) of (2.4) satises [(t
0
) y
0
[ < , we have [(t) y
0
[ < .
0
of (2.4) is asymptotically stable if it is stable, and
0
> 0, such that whenever any solution (t) of (2.4) satises [(t
0
) y
0
[ <
0
, we
have lim
t
[(t) y
0
[ = 0.
y
= f(t, y) (2.5)
A solution (t) of (2.5) is stable if , t
0
0, (, t
0
) > 0 such that whenever
any solution (t) of (2.5) satises [(t
0
) (t
0
)[ < , we have [(t) (t)[ < , t t
0
.
A solution (t) of (2.5) is asymptotically stable if it is stable, and
0
> 0,
such that whenever any solution (t) of (2.5) satises [(t
0
) (t
0
)[ <
0
, we have
lim
t
[(t) (t)[ = 0.
Re(
j
) 0, and when Re(
j
) = 0,
j
is simple y 0 is stable
Re(
j
) < 0 y 0 is asymptotically stable
e
A(tt
0
)
a fundamental matrix. K > 0, > 0, s.t. [e
A(tt
0
)
[ Ke
(tt
0
)
Re(
0
) > 0 y 0 is unstable.
y
= (A+B(t))y (2.6)
Theorem. Re(
j
) < 0, B(t) continuous for 0 t < and such that
_
0
[B(s)[ds <
. Then the zero solution of (2.6) is asymptotically stable.
Proof. y
= (A+B(t))y = Ay +B(t)y
. .
g(t)
, g(t) is an inhomogeneous term.
Let (t) be a solution to the ODE with (t
0
) = y
0
.
By the variation of constants formula:
(t) = e
A(tt
0
)
y
0
+
_
t
t
0
e
A(ts)
B(s)(s)ds
Note: (t
0
) = y
0
y
0
= e
t
0
A
= e
t
0
A
y
0
= e
t
0
A
(t
0
).
[(t)[ [e
A(tt
0
)
[[y
0
[ +
_
t
t
0
[e
A(ts)
[[(s)[[B(s)[ds
Re(
j
) < 0 K, > 0, such that
[e
A(tt
0
)
[ Ke
(tt
0
)
, t
0
t <
[e
A(ts)
[ Ke
(ts)
, t
0
s <
[(t)[ Ke
(tt
0
)
[y
0
[ +K
_
t
t
0
e
(ts)
[(s)[[B(s)[ds
e
t
[(t)[
. .
u(t)
Ke
t
0
[y
0
[
. .
c
+K
_
t
t
0
e
s
[(s)[
. .
u(s)
[B(s)[
. .
v(s)
ds
By Gronwall Inequality:
e
t
[(t)[ Ke
t
0
[y
0
[e
K
_
t
t
0
|B(s)|ds
[(t)[ Ke
(tt
0
)
[y
0
[e
K
_
t
t
0
|B(s)|ds
But K
_
t
t
0
[B(s)[ds M
0
< e
K
_
t
t
0
|B(s)|ds
e
M
0
= M
1
,
[(t)[ KM
1
e
(tt
0
)
[y
0
[ 0, as t .
Thus, the zero solution of y
= (A+ B(t))y is asymptotically stable.

Theorem. Suppose all solutions of y
= Ay are bounded. Let B(t) be continuous for

0 t < , and
_
0
[B(s)[ds < . Show all solutions of y
= (A+B(t))y are bounded

on t
0
< t < .
Proof.
y
= Ay (2.7)
y
= (A+B(t))y (2.8)
Solutions of (2.7) can be written as e
tA
c
0
, where e
tA
is the fundamental matrix.
Since all solutions of (2.7) are bounded, [e
tA
c
0
[ c, 0 t < .
Now look at the solutions of non-homogeneous equation (2.8). By the variation of
constants formula and the previous exercise,
(t) = e
A(tt
0
)
y
0
+
_
t
t
0
e
A(ts)
B(s)(s)ds
[(t)[ [e
A(tt
0
)
[[y
0
[ +
_
t
t
0
[e
A(ts)
[[(s)[[B(s)[ds c[y
0
[ + c
_
t
t
0
[(s)[[B(s)[ds
By Gronwall Inequality,
[(t)[ c[y
0
[e
c
_
t
t
0
|B(s)|ds
.
But
_
t
t
0
[B(s)[ds < c
_
t
t
0
[B(s)[ds < M
0
, e
c
_
t
t
0
|B(s)|ds
M
1
.
[(t)[ c[y
0
[M
1

K.
Thus, all solutions of (2.8) are bounded.
Claim: The zero solution of y
= (A+ B(t))y is stable.

0
is stable if , () such that whenever any solution (t)
satises [(t
0
) y
0
[ < , we have [(t) y
0
[ < .
We had [(t)[ c[
0
[M
1
. Choose [(t
0
)[ small enough such that , () such that
[(t
0
)[ < <

CM
1
[(t) 0[ = [(t)[ c[(t
0
)[M
1
< cM
1
< .
Thus, the zero solution of y
= (A+B(t))y is stable.
y
= (A+B(t))y + f(t, y) (2.9)

Theorem. i) Re(
j
) < 0, f(t, y) and
f
y
j
(t, y) are continuous in (t, y).
ii) lim
|y|0
|f(t,y)|
|y|
= 0 uniformly with respect to t.
iii) B(t) continuous. lim
t
B(t) = 0.
Then the solution y 0 of (2.9) is asymptotically stable.
2.8 Conditional Stability
y
= Ay +g(y) (2.10)
Theorem. g,
g
y
j
continuous, g(0) = 0 and lim
|y|0
|g(y)|
|y|
= 0. If the eigenvalues of
A are , with , > 0, then a curve C in the phase plane of original equation
passing through 0 such that if any solution (t) of (2.10) with [(0)[ small enough starts
on C, then (t) 0 as t . No solution (t) with [(0)[ small enough that does
not start on C can remain small. In particular, 0 is unstable.
2.9 Asymptotic Equivalence
x
= A(t)x (2.11)
y
= A(t)y +f(t, y) (2.12)

The two systems are asymptotically equivalent if to any solution x(t) of (2.11) with
x(t
0
) small enough there corresponds a solution y(t) of (2.12) such that
lim
t
[y(t) x(t)[ = 0
and if to any solution y(t) of (2.12) with y(t
0
) small enough there corresponds a solution
x(t) of (2.11) such that
lim
t
[ y(t) x(t)[ = 0
2.9.1 Levinson
Theorem. A is a constant matrix such that all solutions of x
= Ax are bounded on
0 t < . B(t) is a continuous matrix such that
_
0
[B(s)[ds < . Then, the systems
x
= Ax and y
= (A+B(t))y are asymptotically equivalent.

3 Lyapunovs Second Method
Lagranges Principle. If the rest position of a conservative mechanical system has
minimum potential energy, then this position corresponds to a stable equilibrium. If the
rest position does not have minimum potential energy, then the equilibrium position is
unstable.
3.1 Hamiltonian Form
A system of 2 (or 2n) equations determined by a single scalar function H(y, z)
(or H(y
1
, . . . , y
n
, z
1
, . . . , z
n
)) is called Hamiltonian if it is of the form
H(y, z) y
=
H
z
z
=
H
y
H(y
1
, . . . , y
n
, z
1
, . . . , z
n
) y
i
=
H
z
i
z
i
=
H
y
i
(i = 1, . . . , n) (3.1)
Problem. If = (
1
, . . . ,
2n
) is any solution of the Hamiltonian system (3.1), then
H(
1
, . . . ,
2n
) is constant.
Proof. Need to show
dH
dt
= 0.
Can relabel: H(
1
, . . . ,
n
,
n+1
, . . . ,
2n
) = H(y
1
, . . . , y
n
, z
1
, . . . z
n
).
dH
dt
=
d
dt
H(
1
, . . . ,
n
,
n+1
, . . . ,
2n
)
=
H
1
d
1
dt
+ +
H
n
d
n
dt
+
H
n+1
d
n+1
dt
+ +
H
2n
d
2n
dt
=
n
i=1
H
i
d
i
dt
+
n
i=1
H
n+i
d
n+i
dt
=
n
i=1
H
y
i
dy
i
dt
+
n
i=1
H
z
i
dz
i
dt
= (by (3.1)) =
n
i=1
H
y
i
H
z
i
+
n
i=1
H
z
i
_
H
y
i
_
= 0.
Thus, H(
1
, . . . ,
2n
) is constant.
Problem (F92, #5). Let x = x(t), p = p(t) be a solution of the Hamiltonian
system
dx
dt
=

p
H(x, p), x(0) = y
dp
dt
=

x
H(x, p), p(0) = .
Suppose that H is smooth and satises
H
x
(x, p)
C
_
[p[
2
+ 1
H
p
(x, p)
C.
Prove that this system has a nite solution x(t), p(t) for < t < .
Proof.
x(t) = x(0) +
_
t
0
dx
ds
ds,
[x(t)[ [x(0)[ +
_
t
0
dx
ds
ds = [x(0)[ +
_
t
0
H
p
ds [x(0)[ + C
_
t
0
ds = [x(0)[ +Ct.
Thus, x(t) is nite for nite t.
p(t) = p(0) +
_
t
0
dp
ds
ds,
[p(t)[ [p(0)[ +
_
t
0
dp
ds
ds = [p(0)[ +
_
t
0
H
x
ds [p(0)[ +C
_
t
0
_
[p[
2
+ 1 ds
[p(0)[ +C
_
t
0
(1 + [p[) ds = [p(0)[ + Ct +C
_
t
0
[p[ ds
([p(0)[ +Ct)e
_
t
0
C ds
([p(0)[ + Ct)e
Ct
,
where we have used Gronwall (Integral) Inequality.
3
Thus, p(t) is nite for nite t.
3
Gronwall (Dierential) Inequality: v(t) piecewise continuous on t0 t t0 + a.
u(t) and
du
dt
continuous on some interval. If
du
dt
v(t)u(t)
u(t) u(t0)e
_
t
t
0
v(s)ds
Gronwall (Integral) Inequality: u(t), v(t) continuous on [t0, t0 + a]. v(t) 0, c 0.
u(t) c +
_
t
t
0
v(s)u(s)ds
u(t) c e
_
t
t
0
v(s)ds
t0 t t0 + a
3.2 Lyapunovs Theorems
Denitions: y
= f(y)
The scalar function V (y) is said to be positive denite if V (0) = 0 and V (y) > 0 for
all y ,= 0 in a small neighborhood of 0.
The scalar function V (y) is negative denite if V (y) is positive denite.
The derivative of V with respect to the system y
= f(y) is the scalar product

V
(y) = V f(y)
d
dt
V (y(t)) = V f(y) = V
(y)
along a solution y the total derivative of V (y(t)) with respect to t coincides with
the derivative of V with respect to the system evaluated at y(t).
3.2.1 Stability (Autonomous Systems)
If V (y) that is positive denite and for which V
(y) 0 in a neighborhood of 0, then

the zero solution is stable.
If V (y) that is positive denite and for which V
(y) is negative denite in a neighbor-

hood of 0, then the zero solution is asymptotically stable.
If V (y), V (0) = 0, such that V
(y) is either positive denite or negative denite, and

every neighborhood of 0 contains a point a ,= 0 such that V (a)V
(a) > 0, then the 0

solution is unstable.
Problem (S00, #6).
a) Consider the system of ODEs in R
2n
given in vector notation by
dx
dt
= f([x[
2
)p and
dp
dt
= f
([x[
2
)[p[
2
x,
where x = (x
1
, . . . , x
n
), p = (p
1
, . . . , p
n
), and f > 0, smooth on R. We use the nota-
tion x p = x
1
p
1
+ + x
n
p
n
, [x[
2
= x x and [p[
2
= p p.
Show that [x[ is increasing with t when p x > 0 and decreasing with t when p x < 0,
and that H(x, p) = f([x[
2
)[p[
2
is constant on solutions of the system.
b) Suppose
f(s)
s
has a critical value at s = r
2
. Show that solutions with x(0) on the
shpere [x[ = r and p(0) perpendicular to x(0) must remain on the sphere [x[ = r for all
t. [Compute
d(px)
dt
and use part (a)].
Proof. a)
Consider p x > 0:
Case : p > 0, x > 0
dx
dt
> 0 x = [x[ is increasing.
Case : p < 0, x < 0
dx
dt
< 0 x = [x[ is decreasing [x[ is increasing.
Consider p x < 0:
Case : p > 0, x < 0
dx
dt
> 0 x = [x[ is increasing [x[ is decreasing.
Case : p < 0, x > 0
dx
dt
< 0 x = [x[ is decreasing.
Thus, [x[ is increasing with t when p x > 0 and decreasing with t when p x < 0.
To show H(x, p) = f([x[
2
)[p[
2
is constant on solutions of the system, consider
dH
dt
=
d
dt
_
f([x[
2
)[p[
2
_
= f
([x[
2
) 2x x[p[
2
+f([x[
2
) 2p p
= f
([x[
2
) 2xf([x[
2
)p[p[
2
+ f([x[
2
) 2p
_
f
([x[
2
)[p[
2
x
_
= 0.
Thus, H(x, p) is constant on solutions of the system.
b) G(s) =
f(s)
s
has a critical value at s = r
2
. Thus,
G
(s) =
sf
(s) f(s)
s
2
,
G
(r
2
) = 0 =
r
2
f
(r
2
) f(r
2
)
r
4
,
0 = r
2
f
(r
2
) f(r
2
).
Since p(0) and x(0) are perpendicular, p(0) x(0) = 0.
d(p x)
dt
= x
dp
dt
+p
dx
dt
= f
([x[
2
)[p[
2
[x[
2
+f([x[
2
)[p[
2
= [p[
2
_
f([x[
2
) f
([x[
2
)[x[
2
_
,
d(p x)
dt
(t = 0) = [p[
2
_
f(r
2
) f
(r
2
)r
2
_
= [p[
2
0 = 0.
Also,
d(px)
dt
= 0 holds for all [x[ = r. Thus, p x = C for [x[ = r. Since, p(0) x(0) = 0,
p x = 0. Hence, p and x are always perpendicular, and solution never leaves the sphere.
Note: The system
dx
dt
= f([x[
2
)p and
dp
dt
= f
([x[
2
)[p[
2
x,
determined by H(x, p) = f([x[
2
)[p[
2
is Hamiltonian.
x =
H
p
= 2f([x[
2
)[p[, p =
H
x
= 2xf
([x[
2
)[p[
2
.
Example 1. Determine the stability property of the critical point at the origin for the
following system.
y
1
= y
3
1
+ y
1
y
2
2
y
2
= 2y
2
1
y
2
y
3
2
Try V (y
1
, y
2
) = y
2
1
+cy
2
2
.
V (0, 0) = 0; V (y
1
, y
2
) > 0, y ,= 0 V is positive definite.
V
(y
1
, y
2
) =
dV
dt
= 2y
1
y
1
+ 2cy
2
y
2
= 2y
1
(y
3
1
+y
1
y
2
2
) + 2cy
2
(2y
2
1
y
2
y
3
2
)
= 2y
4
1
2cy
4
2
+ 2y
2
1
y
2
2
4cy
2
1
y
2
2
.
If c =
1
2
, V
(y
1
, y
2
) = 2y
4
1
y
4
2
< 0, y ,= 0; V
(0, 0) = 0
V
negative definite.
Since V (y
1
, y
2
) is positive denite and V
(y
1
, y
2
) is negative denite, the critical point
at the origin is asymptotically stable.
following system.
y
1
= y
3
1
y
3
2
y
2
= 2y
1
y
2
2
+ 4y
2
1
y
2
+ 2y
3
2
Try V (y
1
, y
2
) = y
2
1
+cy
2
2
.
V (0, 0) = 0; V (y
1
, y
2
V
(y
1
, y
2
) =
dV
dt
= 2y
1
y
1
+ 2cy
2
y
2
= 2y
1
(y
3
1
y
3
2
) + 2cy
2
(2y
1
y
2
2
+ 4y
2
1
y
2
+ 2y
3
2
)
= 2y
4
1
2y
1
y
3
2
+ 4cy
1
y
3
2
+ 8cy
2
1
y
2
2
+ 4cy
4
2
.
If c =
1
2
, V
(y
1
, y
2
) = 2y
4
1
+ 4y
2
1
y
2
2
+ 2y
4
2
> 0, y ,= 0; V
(0, 0) = 0
V
positive definite.
Since V
(y
1
, y
2
) is positive denite and V (y)V
(y) > 0, y ,= 0, the critical point at

the origin is unstable.
following system.
y
1
= y
3
1
+ 2y
3
2
y
2
= 2y
1
y
2
2
Try V (y
1
, y
2
) = y
2
1
+cy
2
2
.
V (0, 0) = 0; V (y
1
, y
2
V
(y
1
, y
2
) =
dV
dt
= 2y
1
y
1
+ 2cy
2
y
2
= 2y
1
(y
3
1
+ 2y
3
2
) + 2cy
2
(2y
1
y
2
2
)
= 2y
4
1
+ 4y
1
y
3
2
4cy
1
y
3
2
.
If c = 1, V
(y
1
, y
2
) = 2y
4
1
0, y; V
(y) = 0 for y = (0, y

2
).
V
is neither positive definite nor negative definite.

Since V is positive denite and V
(y
1
, y
2
) 0 in a neighborhood of 0, the critical point
at the origin is at least stable.
V is positive denite, C
1
, V
(y
1
, y
2
) 0, y. The origin is the only invariant
subset of the set E = y[V
(y) = 0 = (y
1
, y
2
) [ y
1
= 0. Thus, the critical point at
the origin is asymptotically stable.
Problem (S96, #1).
Construct a Liapunov function of the form ax
2
+cy
2
for the system
x = x
3
+ xy
2
y = 2x
2
y y
3
,
and use it to show that the origin is a strictly stable critical point.
Proof. We let V (x, y) = ax
2
+cy
2
.
V
(x, y) =
dV
dt
= 2ax x + 2cy y = 2ax(x
3
+ xy
2
) + 2cy(2x
2
y y
3
)
= 2ax
4
+ 2ax
2
y
2
4cx
2
y
2
2cy
4
= 2ax
4
+ (2a 4c)x
2
y
2
2cy
4
.
For 2a 4c < 0, i.e. a < 2c, we have V
(x, y) < 0. For instance, c = 1, a = 1.

Then, V (0, 0) = 0; V (x, y) > 0, (x, y) ,= (0, 0) V is positive denite.
Also, V
(0, 0) = 0; V
(x, y) = 2ax
4
2x
2
y
2
2cy
4
< 0, (x, y) ,= (0, 0)
V
is negative denite.
Since V (x, y) is positive denite and V
(x, y) is negative denite, the critical point at

the origin is asymptotically stable.
Example 4. Consider the equation u
+ g(u) = 0, where g is C
1
for [u[ < k, k > 0,
and ug(u) > 0 if u ,= 0. Thus, by continuity, g(0) = 0. Write the equation as a system
y
1
= y
2
y
2
= g(y
1
)
and the origin is an isolated critical point. Set
V (y
1
, y
2
) =
y
2
2
2
+
_
y
1
0
g()d.
Thus, V (0, 0) = 0 and since g() > 0,
_
y
1
0
g()d > 0 for 0 < [y
1
[ < k.
Therefore, V (y
1
, y
2
) is positive denite on = (y
1
, y
2
) [ [y
1
[ < k, [y
2
[ < .
V
(y
1
, y
2
) =
dV
dt
= y
2
y
2
+ g(y
1
)y
1
= y
2
(g(y
1
)) + g(y
1
)y
2
= 0.
Since V is positive denite and V
(y
1
, y
2
) 0 in a neighborhood of 0, the critical point
at the origin is stable.
Example 5. The Lienard Equation. Consider the scalar equation
u
+u
+ g(u) = 0
or, written as a system,
y
1
= y
2
y
2
= g(y
1
) y
2
where g is C
1
, ug(u) > 0, u ,= 0. Try
V (y
1
, y
2
) =
y
2
2
2
+
_
y
1
0
g()d.
V is positive denite on = (y
1
, y
2
) [ [y
1
[ < k, [y
2
[ < .
V
(y
1
, y
2
) =
dV
dt
= y
2
y
2
+ g(y
1
)y
1
= y
2
(g(y
1
) y
2
) +g(y
1
)y
2
= y
2
2
.
Since V
(y
1
, y
2
) 0 in , the solution is stable. But V
(y
1
, y
2
) is not negative denite
on (V
(y
1
, y
2
) = 0 at all points (y
1
, 0)). Even though the solution is asymptotically
stable, we cannot infer this here by using Lyapunovs theorems.
4
4
See the example in Invariant Sets and Stability section.
3.3 Periodic Solutions
Problem. Consider the 2-dimensional autonomous system y
= f(y) where f(y)

C
1
(R
2
). Let R
2
be simply connected, such that y , we have div f (y) ,= 0.
Show that the ODE system has no periodic solutions in .
Proof. Towards a contradiction, assume ODE system has a periodic solution in . Let
be a boundary on .
y
= f(y)
_
y
1
= f
1
(y
1
, y
2
),
y
2
= f
2
(y
1
, y
2
).
n = (n
1
, n
2
) = (y
2
, y
1
) is the normal to . Recall Divergence Theorem:
_
f n ds =
__
div f dA.
Let y be a periodic solution with period T, i.e. y(t +T) = y(t). Then, a path traversed
by a solution starting from t = a to t = a +T is . Then, is a closed curve.
_
f n ds =
_
(f
1
n
1
+f
2
n
2
) ds =
_
a+T
a
(y
1
y
2
y
2
y
1
) dt = 0
__
div f dA = 0.
However, by hypothesis, div f(y) ,= 0 and f C
1
. Therefore, div f C
0
, and ei-
ther div f > 0 or div f < 0 on . Thus,
__
div f dA > 0 or
__
div f dA < 0, a
contradiction.
Example. Show that the given system has no non-trivial periodic solutions:
dx
dt
= x +y + x
3
y
2
,
dy
dt
= x + 2y + x
2
y +
y
3
3
.
Proof.
dx
dt
= f
1
(x, y),
dy
dt
= f
2
(x, y).
divf(x, y) =
f
1
x
+
f
2
y
= (1 + 2x
2
) + (2 +x
2
+ y
2
) = 3 + 3x
2
+ y
2
> 0.
By the problem above, the ODE system has no periodic solutions.
Problem (F04, #5).
Consider a generalized Volterra-Lotka system in the plane, given by
x
(t) = f(x(t)), x(t) R

2
, (3.2)
where f(x) = (f
1
(x), f
2
(x)) = (ax
1
bx
1
x
2
ex
2
1
, cx
2
+dx
1
x
2
fx
2
2
) and a, b, c, d, e, f
are positive constants. Show that
div(f ) ,= 0 x
1
> 0, x
2
> 0,
where (x
1
, x
2
) = 1/(x
1
x
2
). Using this observation, prove that the autonomous system
(3.2) has no closed orbits in the rst quadrant.
Proof.
f =
_
ax
1
bx
1
x
2
ex
2
1
x
1
x
2
cx
2
+dx
1
x
2
fx
2
2
x
1
x
2
_
=
_
ax
1
2
b ex
1
x
1
2
cx
1
1
+ d fx
1
1
x
2
_
,
div(f) =

x
1
(ax
1
2
b ex
1
x
1
2
) +

x
2
(cx
1
1
+d fx
1
1
x
2
) = ex
1
2
fx
1
1
,= 0,
for x
1
, x
2
> 0, f, e > 0.
Towards a contradiction, assume ODE system has a closed orbit in the rst quad-
rant. Let be a bounded domain with an orbit that is .
Let x be a periodic solution with a period T, i.e. x(t +T) = x(t).
n = (n
1
, n
2
) = (x
2
, x
1
) is the normal to . By Divergence Theorem,
_
div(f) dx =
_
(f) ndS =
_
(f
1
n
1
+f
2
n
2
) dS
=
_
a+T
a
(x
1
x
2
x
2
x
1
) dt = 0.
Since f C
1
in , then div(f) C
0
in .
Thus, the above result implies div(f) = 0 for some (x
1
, x
2
) ,
which contradicts the assumption.
Problem (F04, #4).
Prove that each solution (except x
1
= x
2
= 0) of the autonomous system
_
x
1
= x
2
+ x
1
(x
2
1
+x
2
2
)
x
2
= x
1
+x
2
(x
2
1
+x
2
2
)
blows up in nite time. What is the blow-up time for the solution which starts at the
point (1, 0) when t = 0?
Proof. We have r
2
= x
2
1
+x
2
2
. Multiply the rst equation by x
1
and the second by x
2
:
x
1
x
1
= x
1
x
2
+x
2
1
(x
2
1
+ x
2
2
),
x
2
x
2
= x
1
x
2
+ x
2
2
(x
2
1
+x
2
2
).
Add equations:
x
1
x
1
+ x
2
x
2
= (x
2
1
+x
2
2
)(x
2
1
+ x
2
2
),
1
2
(x
2
1
+x
2
2
)
= (x
2
1
+x
2
2
)(x
2
1
+ x
2
2
),
1
2
(r
2
)
= r
4
,
rr
= r
4
,
r
= r
3
,
dr
dt
= r
3
,
dr
r
3
= dt,
1
2r
2
= t +C,
r =
1
2(t +C)
.
Thus, solution blows up at t = C. We determine C.
Initial conditions: x
1
(0) = 1, x
2
(0) = 0 r(0) = 1.
1 = r(0) =
_
1
2C
,
C =
1
2
,
r =
_
1
2t 1
=
_
1
1 2t
.
Thus, the blow-up time is t =
1
2
.
3.4 Invariant Sets and Stability
A set K of points in phase space is invariant with respect to the system y
= f(y) if
every solution of y
= f(y) starting in K remains in K for all future time.

A point p R
n
is said to lie in the positive limit set L(C
+
) (or is said to be a limit
point of the orbit C
+
) of the solution (t) i for the solution (t) that gives C
+
for
t 0, a sequence t
n
+ as n such that lim
n
(t
n
) = p.
Remark: V
0, S
= y R
n
: V (y) .
For every the set S
, in fact, each of its components, is an invariant set with respect

to y
= f(y).
Reasoning: if y
0
S
and (t, y
0
) is solution
d
dt
V ((t, y
0
)) = V
((t, y
0
)) 0
V ((t, y
0
)) V ((0, y
0
)), t 0
(t, y
0
) S
, t 0
S
invariant (as its components).

If the solution (t, y
0
) is bounded for t 0 L(C
+
) is a nonempty closed,
connected, invariant set. Moreover, the solution (t, y
0
) L(C
+
) as t .
V is C
1
. V
0 on . Let y
0
and (t, y
0
) be bounded with (t, y
0
) ,
t 0. Assume that L(C
+
) lies in . Then, V
(y) = 0 at all points of L(C

+
).
V positive denite, C
1
, V
0. Let the origin be the only invariant subset of the set

y[V
(y) = 0. Then the sero solution is asymptotically stable.

V nonnegative, C
1
, V
0, V (0) = 0. Let M be the largest invariant subset of

y[V
(y) = 0. Then all bounded solutions approach the set M as t .

L(C
+
) contains a closed (periodic) orbit L(C
+
) contains no other points.
The limit set can not be a closed disk topologically.
Example. The Lienard Equation. Consider the scalar equation
u
+f(u)u
+g(u) = 0
where f(u) > 0 for u ,= 0 and ug(u) > 0 for u ,= 0. Written as a system,
y
1
= y
2
y
2
= f(y
1
)y
2
g(y
1
)
V (y
1
, y
2
) =
y
2
2
2
+
_
y
1
0
g()d.
V (0, 0) = 0; V (y
1
, y
2
) > 0, y ,= 0, so V is positive denite.
V
(y
1
, y
2
) =
dV
dt
= y
2
y
2
+g(y
1
)y
1
= y
2
(f(y
1
)y
2
g(y
1
)) +g(y
1
)y
2
= f(y
1
)
. .
>0
y
2
2
..
0
0.
The zero solution is at least stable by one of Lyapunovs theorems.
V
(y
1
, 0) = 0 on y
1
axis E = y [ V
(y) = 0 = y [ (y
1
, 0) E is y
1
-axis.
A set of points in phase space is invariant if every solution that starts in remains
in for all time.
On y
1
-axis (y
2
= 0):
dy
1
dt
= 0
dy
2
dt
= g(y
1
) =
_
> 0, y
1
> 0,
< 0, y
1
< 0.
=
_
< 0, y
1
> 0,
> 0, y
1
< 0.
The solution can remain on E (y
2
= 0) only if y
2
= g(y
1
) = 0.
Thus, (0, 0) is the largest (and only) invariant subset of E = y [ V
(y) = 0.
Since V is positive denite, C
1
on R
2
, V
0, y R
2
, and the origin is the only
invariant subset of E, the zero solution is asymptotically stable.
Example. Van Der Pol Equation. Region of Asymptotic Stability.
Determine an estimate of the region of asymptotic stability in the phase plane for
u
+(1 u
2
)u
+ u = 0, > 0, a constant.
Proof. Recall the Lienard equation: u
+f(u)u
+g(u) = 0. In our case,

f(u) = (1 u
2
), g(u) = u.
Similar to assumptions made for the Lienard equation, we have
g(0) = 0, ug(u) = u
2
> 0, u ,= 0. Let F(u) =
_
u
0
f()d.
F(u) =
_
u
0
f()d =
_
u
0
(1
2
)d = u
u
3
3
.
Find a > 0 such that uF(u) > 0 for 0 < [u[ < a:
uF(u) = u
2
u
4
3
> 0 0 < [u[ <
3 = a. (3.3)
Here, we employ a dierent equivalent system than we had done in previous examples,
y
1
= u,
y
2
= u
+F(u), which gives

y
1
= y
2
F(y
1
),
y
2
= y
1
.
Dene G(y
1
) =
_
y
1
0
g() d =
_
y
1
0
d =
y
2
1
2
.
Choose V (y
1
, y
2
) =
y
2
2
2
+G(y
1
) =
y
2
2
2
+
y
2
1
2
V (y
1
, y
2
) is positive denite on R
2
.
V
(y
1
, y
2
) = y
2
y
2
+y
1
y
1
= y
2
(y
1
) +y
1
(y
2
F(y
1
)) = y
1
F(y
1
) 0
on the strip = (y
1
, y
2
) [
3 < y
1
<
3, < y
2
< , by (3.3).
Thus, the origin is stable.
V
= y
1
F(y
1
) = 0 for y
1
= 0 (y
2
axis)
E = y [ V
(y) = 0 = (y
1
, y
2
) [ y
1
= 0. On E : y
1
= y
2
, y
2
= 0.
Thus, 0 is the only invariant subset of E, and the zero solution is asymptotically stable.
Consider the curves V (y
1
, y
2
) = (
y
2
1
2
+
y
2
2
2
= ) for
3 < y
1
<
3 with increasing
values of , beginning with = 0. These are closed curves symmetric about the y
1
-axis.
Since V (y
1
, y
2
) =
y
2
2
2
+
y
2
1
2
, V (y
1
, y
2
) rst makes contact with the boundary of at
one of the points (
3, 0) or (
3, 0). The best value of

= min(G(
3), G(
3)) =
min(
3
2
,
3
2
) =
3
2
and C
= (y
1
, y
2
) [
y
2
2
2
+
y
2
1
2
<

= (y
1
, y
2
) [ y
2
1
+y
2
2
< 3.
Every solution that starts in C
approaches the origin.

5
3.5 Global Asymptotic Stability
Theorem. Let there exist a scalar function V (y) such that:
(i) V (y) is positive denite on all R
n
;
(ii) V (y) as [y[ ;
(iii) V
(y) 0 on R
n
;
(iv) 0 it the onlty invariant subset of E = y [ V
(y) = 0.
Then 0 is globally asymptotically stable.
Corollary. V (y) satises (i) and (ii) above, and V
(y) is negative denite.

Then 0 is globally asymptotically stable.
5
Brauer, Nohel, Theorem 5.5, p. 214.
3.6 Stability (Non-autonomous Systems)
y
= f(t, y)
The scalar function V (t, y) is positive denite if V (t, 0) = 0, t and W(y) positive
denite, s.t. V (t, y) W(y) in = (t, y) : t 0, [y[ b, b > 0.
The scalar function V (t, y) is negative denite if V (t, y) is positive denite.
V
(t, y) =
d
dt
V (t, y(t)) =
V
t
+V f(t, y)
If there exists a scalar function V (t, y) that is positive denite and for which V
(t, y) 0
in , then the zero solution is stable.
If there exists a scalar function V (t, y) that is positive denite, satises an innitesimal
upper bound (i.e. lim
0
+ sup
t0,|y|
[V (t, y)[ = 0), and for which V
(t, y) is negative
denite, then the zero solution is asymptotically stable.
3.6.1 Examples
V (t, y) = y
2
1
+(1 +t)y
2
2
y
2
1
+y
2
2
= W(y) V positive denite on = (t, y) : t
0)
V (t, y) = y
2
1
+ ty
2
2
y
2
1
+ ay
2
2
= W(y) V positive denite on = (t, y) : t
a, a > 0)
V (t, y) = y
2
1
+
y
2
2
1+t
. Since V (t, 0, a
2
) =
a
2
2
1+t
0 as t V not positive denite
even though V (t, y) > 0 for y ,= 0.
4 Poincare-Bendixson Theory
A segment without contact with respect to a vector eld V : R
n
R
n
is a nite,
closed segment L of a straight line, s.t:
a) Every point of L is a regular point of V ;
b) At no point of L the vector eld V has the same direction as L.
Poincare-Bendixson Theorem. Let C
+
be a positive semi-orbit contained in a closed
and bounded set K R
2
. If its limit set L(C
+
) contains no critical points of vector
eld

f, then L(C
+
) is a periodic orbit. Also, either:
i) C = L(C
+
), or
ii) C approaches L(C
+
) spirally from either inside or outside.
Corollary. If C
+
is a semiorbit contained in an invariant compact set K in which
f has no critical points, then K contains a periodic orbit. Such a set cannot be
equivalent to a disk.
Example. Prove that the second order dierential equation
z
+ (z
2
+ 2(z
)
2
1)z
+ z = 0 (4.1)
has a non-trivial periodic solution.
Proof. Write (4.1) as a rst-order system:
y
1
= y
2
,
y
2
= y
1
(y
2
1
+ 2y
2
2
1)y
2
.
Let V (y
1
, y
2
) =
1
2
y
2
1
+
1
2
y
2
2
V
(y
1
, y
2
) = y
1
y
1
+ y
2
y
2
= y
1
y
2
+ y
2
(y
1
(y
2
1
+ 2y
2
2
1)y
2
)
= y
2
2
(y
2
1
+ 2y
2
2
1)
Use Poincare-Bendixson Theorem: If C
+
is a semiorbit contained in an invariant
compact set K in which f has no critical points, then K contains a periodic orbit.
Setting both equations of the system to 0, we see that (0, 0) is the only critical point.
Choose a compact set K = (y
1
, y
2
) [
1
4
y
2
1
+ y
2
2
4 and show that it is invariant.
V
= V

f. Need V
[
out
< 0, V
in
> 0.
Check invariance of K:
V
[
out
= y
2
2
(y
2
1
+ 2y
2
2
1) <
..
need
0,
Need: y
2
1
+ 2y
2
2
1 > 0,
y
2
1
+ 2y
2
2
1 y
2
1
+ y
2
2
1 = 4 1 = 3 > 0.
V
in
= y
2
2
(y
2
1
+ 2y
2
2
1) >
..
need
0,
Need: y
2
1
+ 2y
2
2
1 < 0,
y
2
1
+ 2y
2
2
1 2y
2
1
+ 2y
2
2
1 = 2(
1
4
) 1 =
1
2
< 0.
K is an invariant set. (0, 0) / K.
Thus K contains a periodic orbit.
Polar Coordinates. Sometimes it is convenient to use polar coordinates when ap-
plying Poincare-Bendixson theorem.
y
1
= f
1
(y
1
, y
2
)
y
2
= f
2
(y
1
, y
2
)
V =
y
2
1
2
+
y
2
2
2
V
=
dV
dt
= y
1
y
1
+ y
2
y
2
= r cos f
1
(r, ) + r sin f
2
(r, ).
Example. Polar Coordinates. Consider the system
y
1
= y
2
+ y
1
(1 y
2
1
y
2
2
),
y
2
= y
1
+y
2
(1 y
2
1
y
2
2
).
Proof. Let V (y
1
, y
2
) =
y
2
1
2
+
y
2
2
2
.
V
(y
1
, y
2
) = y
1
y
1
+y
2
y
2
= r cos f
1
(r, ) + r sin f
2
(r, )
= r cos (r sin + r cos (1 r
2
)) + r sin (r cos + r sin(1 r
2
))
= r
2
cos sin +r
2
cos
2
(1 r
2
) r
2
cos sin + r
2
sin
2
(1 r
2
)
= r
2
(1 r
2
).
Use Poincare-Bendixson Theorem: If C
+
is a semiorbit contained in an invariant
compact set K in which f has no critical points, then K contains a periodic orbit.
Setting both equations of the system to 0,
we see that (0, 0) is the only critical point.
Choose a compact set K = (y
1
, y
2
) [
1
4
y
2
1
+ y
2
2
4
and show that it is invariant.
V
= V

f. Need V
[
out
< 0, V
in
> 0.
V
[
out
= r
2
(1 r
2
) = 4(1 4) < 0.
V
in
= r
2
(1 r
2
) =
1
4
(1
1
4
) > 0.
Example. Show that the autonomous system
du
dt
= u v u
3
uv
2
dv
dt
= u +v v
3
u
2
v
has (a) a unique equilibrium point, (b) which is unstable, and (c) a unique closed
solution curve.
Proof. a) Set above equations to 0 and multiply the rst by v and the second by u:
uv v
2
u
3
v uv
3
= 0
u
2
+uv uv
3
u
3
v = 0 u
2
+ v
2
= 0 u
2
= v
2
u = 0, v = 0.
Thus, (0, 0) is a unique equilibrium point.
b) Let V (u, v) =
1
2
u
2
+
1
2
v
2
, V is positive denite in R
2
.
V
(u, v) = uu
+ vv
= u(u v u
3
uv
2
) + v(u + v v
3
u
2
v)
= (u
2
+ v
2
) (u
2
+v
2
)
2
= (u
2
+ v
2
)(1 (u
2
+ v
2
)).
V
(u, v) is positive denite in a small neighborhood of (0, 0), i.e. V
is positive denite
on = (u, v) [ u
2
+ v
2
=
1
2
. Thus (0, 0) is unstable.
c) To show that the ODE system has a closed solution curve, use Poincare-Bendixson
theorem: If C
+
is a semiorbit contained in an invariant compact set K in which f
has no critical points, then K contains a periodic orbit.
Choose a compact set K = (u, v) [
1
2
u
2
+ v
2
2 and show that it is invariant.
V
= V

f. Need V
[
out
< 0, V
in
> 0.
V
[
out
= (u
2
+v
2
)(1 (u
2
+v
2
)) = 2(1 2) = 2 < 0.
V
in
= (u
2
+v
2
)(1 (u
2
+v
2
)) =
1
2
(1
1
2
) =
1
4
> 0.
To show uniqueness of a periodic orbit, suppose is
the orbit of a periodic solution in K.
_
dV = 0,
dV =
dV
dt
dt = V
dt
dt = 0.
V
(u, v) = (u
2
+v
2
)(1 (u
2
+v
2
))
(u
2
+v
2
)(1 (u
2
+v
2
)) dt = 0.
u
2
+v
2
= 1 is a periodic orbit.
Suppose there is another periodic orbit in K. We know that the following integral
should be equal to 0 for a closed curve :
_
(u
2
+ v
2
)
. .
=0
(1 (u
2
+ v
2
))
. .
oscillates about 0 as going around
dt = 0.
In order for integral above to be equal to 0, (1 (u
2
+v
2
)) should change sign as going
around. At some point a, = (u, v) [ u
2
+ v
2
= 1 and
2
dened by the second
solution would intersect. But this is impossible, since at that point, there would be
more than one possible solution. contradiction. Thus, the system has unique closed
solution curve.
Problem (S99, #8). Consider the pair of ordinary dierential equations
dx
1
dt
= x
2
dx
2
dt
= x
1
+ (1 x
2
1
x
2
2
)x
2
a) Show any nontrivial solution has the property that x
2
1
+ x
2
2
decreases in time if its
magnitude is greater than one and increases in time if its magnitude is less than one.
b) Use your work in (a) to show that on a periodic orbit, the integral
_
_
1 x
2
1
(t) x
2
2
(t)
_
x
2
2
(t) dt = 0.
c) Consider the class of solutions x
1
= sin(t +c), x
2
= cos(t +c). Show that these are
the only periodic orbits, for c any constant.
Hint: Use (b) to show that any periodic solution for which 1x
2
1
x
2
2
,= 0 must be such
that 1 x
2
1
x
2
2
changes sign on the orbit and use (a) to show this is impossible.
Proof. a) (0, 0) is the only equilibrium point.
Let V (x
1
, x
2
) =
1
2
x
2
1
+
1
2
x
2
2
; V is positive denite on R
2
.
V
(x
1
, x
2
) = x
1
x
1
+x
2
x
2
= x
1
x
2
+x
2
(x
1
+(1x
2
1
x
2
2
)x
2
) = (1x
2
1
x
2
2
)x
2
2
(4.2)
V
(x
1
, x
2
) 0 inside and V
(x
1
, x
2
) 0 outside the unit circle in the phase plane.
Since V
= 0 on x
2
= 0 (x
1
-axis), it can not be concluded
that the statement to be proved is satised.
Let r =
1
2
x
2
1
+
1
2
x
2
2
in (4.2), then
V
(x
1
, x
2
) =
d
dt
_
1
2
x
2
1
+
1
2
x
2
2
_
= (1 x
2
1
x
2
2
)x
2
2
,
dr
dt
= (1 2r)x
2
2
=
_
< 0, 2r > 1
> 0, 2r < 1
=
_
< 0, x
2
1
+x
2
2
> 1,
> 0, x
2
1
+x
2
2
< 1.
Thus, r (and thus, x
2
1
+ x
2
2
) decreases if x
2
1
+ x
2
2
> 1 and increases if x
2
1
+x
2
2
< 1.
If r =
1
2
,
dr
dt
= 0, so x
2
1
+x
2
2
= 1 is a circular orbit.
b) The only periodic orbit is x
2
1
+x
2
2
= 1 where V
= 0:
_
dV = 0,
dV =
dV
dt
dt = V
dt
dt = 0.
_
_
1 x
2
1
x
2
2
_
x
2
2
dt = 0.
c) The class of solutions x
1
= sin(t + c), x
2
= cos(t + c) satisfy x
2
1
+ x
2
2
= 1, and
therefore, are periodic orbits, for c any constant. Suppose there is another periodic
orbit. We know that the following integral should be equal to 0 for a closed curve :
_
_
1 x
2
1
x
2
2
_
. .
oscillates about 0 as going around
x
2
2
..
=0
dt = 0.
In order for integral above to be equal to 0, 1 x
2
1
x
2
2
should change sign as going
around.
At some point a, = (x
1
, x
2
) [ x
2
1
+ x
2
2
= 1 and
2
dened by the second solution
would intersect. But this is impossible, since at that point, there would be more than
one possible solution. contradiction. Thus, the system has a unique closed solution
curve.
Also, by (a), we can conclude that solution curves either increase or decrease in time
if the magnitude of x
2
1
+x
2
2
is not one. Thus, they approach the only periodic solution
x
2
1
+x
2
2
= 1.
5 Sturm-Liouville Theory
Denition. The dierential equation
(py
+ qy +ry = 0, a x b (5.1)
c
1
y(a) + c
2
y
(a) = 0, c
3
y(b) +c
4
y
(b) = 0
is called a Sturm-Liouville equation. A value of the parameter for which a non-
trivial solution (y ,= 0) exists is called an eigenvalue of the problem and correspond-
ing nontrivial solution y(x) of (5.1) is called an eigenfunction which is associated with
that eigenvalue. Problem (5.1) is also called an eigenvalue problem.
The coecients p, q, and r must be real and continuous everywhere and p > 0 and
r > 0 everywhere.
5.1 Sturm-Liouville Operator
Consider the Sturm-Liouville dierential operator
Ly = (py
+qy
_
L =
d
dx
_
p
d
dx
_
+q
_
(5.2)
where p > 0, r > 0, and p
, q and r are continuous on [a, b]. The dierential equation

(5.1) takes the operational form
Ly +ry = 0, a x b (5.3)
c
1
y(a) + c
2
y
(a) = 0, c
3
y(b) +c
4
y
(b) = 0.
5.2 Existence and Uniqueness for Initial-Value Problems
Theorem
6
. Let P(x), Q(x) and R(x) be continuous on [a, b]. If x
0
is a point in this
interval and y
0
and y
1
are arbitrary numbers, then the initial-value problem
y
+P(x)y
+Q(x)y = R(x)
y(x
0
) = y
0
, y
(x
0
) = y
1
has a unique solution on [a, b].
Note. The unique solution of the initial-value problem with R(x) = 0, y(x
0
) =
y
(x
0
) = 0, is the trivial solution.
5.3 Existence of Eigenvalues
Theorem
7
. The Sturm-Liouville problem (5.1) has an innite number of eigenvalues,
which can be written in increasing order as
1
<
2
< . . . <
n
< . . . , such that
lim
n
n
= . The eigenfunctions y
n
(x) corresponding to
n
has exactly n1 zeros
in (a, b).
6
Bleecker and Csordas, Theorem 1, p. 260.
7
5.4 Series of Eigenfunctions
Theorem
8
. The eigenfunctions
n
(x) form a complete set, meaning that any piece-
wise smooth function f(x) can be represented by a generalized Fourier series of eigen-
functions:
f(x)
n=1
a
n
n
(x).
5.5 Lagranges Identity
We calculate uL(v) vL(u), where u and v are any two functions. Recall that
L(u) = (pu
+ qu and L(v) = (pv
+qv,
and hence
uL(v) vL(u) = u(pv
+ quv v(pu
quv = u(pv
v(pu
.
The right hand side is manipulated to an exact dierential:
uL(v) vL(u) =
_
p(uv
vu
.
5.6 Greens Formula
The integral form of the Lagranges identity is known as Greens formula.
_
b
a
_
uL(v) vL(u)
dx = p
_
uv
vu
b
a
for any functions u and v.
8
Haberman, edition 4, Theorem 4, p. 163.
5.7 Self-Adjointness
With the additional restriction that the boundary terms vanish,
p
_
uv
vu
b
a
= 0,
we get
_
b
a
_
uL(v) vL(u)
dx = 0. (5.4)
In fact, in the regular Sturm-Liouville eigenvalue problems, the boundary terms
vanish.
9
When (5.4) is valid, we say that L is self-adjoint.
Denition
10
. Let L and L
denote the linear, second-order dierential operators de-

ned by
Ly = p
2
(x)y
+p
1
(x)y
+p
0
(x)y,
L
y = (yp
2
(x))
(yp
1
(x))
+yp
0
(x).
Then L
is called the adjoint of L and the dierential equation L
y = 0 is called
the adjoint equation. The operator L is said to be self-adjoint, if L = L
. A
homogeneous, linear, second order ODE is said to be in self-adjoint form if the ODE
has the form
(p(x)y
+q(x)y = 0.
Note: The linear, second-order dierential operator
Ly = p
2
(x)y
+ p
1
(x)y
+ p
0
(x)y
is self-adjoint (L = L
) if and only if p
2
(x) = p
1
(x), i.e.,
Ly = (p
2
(x)y
+p
0
(x)y.
Proof. The adjoint L
is given by
L
y = (yp
2
(x))
(yp
1
(x))
+yp
0
(x) = y
p
2
+ 2y
2
+yp
2
p
1
y p
1
y
+yp
0
= p
2
y
+ (2p
2
p
1
)y
+ (p
2
p
1
+p
0
)y.
Thus, L = L
2p
2
p
1
= p
1
, or p
2
= p
1
.
9
Haberman, p. 176.
10
Bleecker and Csordas, p. 264.
Problem (F91, #6). Consider the boundary value problem
x
d
2
w
dx
2
+ (a x)
dw
dx
= w
w(L) = w(R) = 0,
where a, L(> 0) and R(> L) are real constants.
By casting the problem in self-adjoint form shows that the eigenfunctions, w
1
and
w
2
, corresponding to dierent eigenvalues,
1
and
2
, are orthogonal in the sense that
_
R
L
e
x
x
a1
w
1
w
2
dx =
_
R
L
e
x
x
a
dw
1
dx
dw
2
dx
dx = 0.
Show also that
i
=
_
R
L
e
x
x
a
(
dw
i
dx
)
2
dx
_
R
L
e
x
x
a1
w
2
i
dx
and hence that all eigenvalues are positive.
Proof. A homogeneous, linear, second order ODE is said to be in self-adjoint form
if the ODE has the form
(p(x)u
+q(x)u = 0.
We have
Lu = xu
+ (a x)u
.
Multiply the equation by v so that it becomes of self-adjoint form:
vLu = xvu
+ (a x)vu
.
Thus, we need
(pu
= xvu
+ (a x)vu
,
pu
+p
= xvu
+ (a x)vu
.
Thus, p = xv, and
(xv)
= (a x)v,
xv
+ v = av xv,
v
v
=
a x 1
x
,
v
v
=
a 1
x
1,
ln v = (a 1) ln x x,
ln v = ln x
a1
x,
v = e
lnx
a1
e
x
= x
a1
e
x
.
Thus, the self-adjoint form is
(xvu
+ uv = 0, or
(x
a
e
x
u
+ x
a1
e
x
u = 0.
Let
m
,
n
, be the eigenvalues and u
m
, u
n
be the corresponding eigenfunctions.
We have
(x
a
e
x
u
m
)
+
m
x
a1
e
x
u
m
= 0, (5.5)
(x
a
e
x
u
n
)
+
n
x
a1
e
x
u
n
= 0. (5.6)
Multiply (5.5) by u
n
and (5.6) by u
m
and subtract equations from each other
u
n
(x
a
e
x
u
m
)
+
m
x
a1
e
x
u
n
u
m
= 0,
u
m
(x
a
e
x
u
n
)
+
n
x
a1
e
x
u
m
u
n
= 0.
(
m

n
)x
a1
e
x
u
m
u
n
= u
m
(x
a
e
x
u
n
)
u
n
(x
a
e
x
u
m
)
,
= [x
a
e
x
(u
m
u
n
u
n
u
m
)]
.
Integrating over (L, R) gives
(
m

n
)
_
R
L
x
a1
e
x
u
m
u
n
dx = [x
a
e
x
(u
m
u
n
u
n
u
m
)]
R
L
= 0,
Since
n
,=
m
, u
n
(x) and u
m
(x) are orthogonal on [L, R].
To show that u
m
and u
n
are orthogonal with respect to x
a1
e
x
, consider
_
R
L
x
a
e
x
u
m
u
n
dx = x
a
e
x
u
m
u
n
[
R
L
_
R
L
(x
a
e
x
u
m
)
u
n
dx
=
_
R
L
(x
a
e
x
u
m
)
u
n
dx = =
m
_
1
0
x
a1
e
x
u
m
u
n
dx = = 0.
We now show that eigenvalues are positive. We have
(x
a
e
x
u
+x
a1
e
x
u = 0.
Multiplying by u and integrating, we get
_
R
L
u(x
a
e
x
u
+x
a1
e
x
u
2
dx = 0,
x
a
e
x
uu
[
R
L
. .
=0
_
R
L
x
a
e
x
u
2
dx +
_
R
L
x
a1
e
x
u
2
dx = 0,
=
_
R
L
x
a
e
x
u
2
dx
_
R
L
x
a1
e
x
u
2
dx
0.
The equality holds only if u
0, which means u = C. Since u(0) = u(1) = 0, then

u 0, which is not an eigenfunction. Thus, > 0.
Problem (F01, #2). Consider the dierential operator
L =
_
d
dx
_
2
+ 2
_
d
dx
_
+ (x)
in which is a real-valued function. The domain is x [0, 1], with Neumann boundary
conditions
du
dx
(0) =
du
dx
(1) = 0.
a) Find a function = (x) for which L is self-adjoint in the norm
[[u[[
2
=
_
1
0
u
2
dx.
b) Show that L must have a positive eigenvalue if is not identically zero and
_
1
0
(x) dx 0.
Proof. a) Lu = u
+ 2u
+ (x)u. L is self-adjoint in the above norm, if

_
1
0
_
uL(v) vL(u)
dx = 0, or
_
1
0
uL(v)dx =
_
1
0
vL(u)dx,
_
1
0
u(v
+ 2v
+ (x)v)dx =
_
1
0
v(u
+ 2u
+(x)u)dx,
_
1
0
v
..
g
u
..
f
dx + 2
_
1
0
uv
dx +
_
1
0
(x)uv dx =
_
1
0
u
..
g
v
..
f
dx + 2
_
1
0
vu
dx +
_
1
0
(x)uv dx,
v
u[
1
0
_
1
0
v
(u
+ u
) dx + 2
_
1
0
uv
dx = u
v[
1
0
_
1
0
u
(v
+ v
) dx + 2
_
1
0
vu
dx.
Boundary terms are 0 due to boundary conditions. Cancelling out other terms, we get
_
1
0
uv
dx + 2
_
1
0
uv
dx =
_
1
0
vu
dx + 2
_
1
0
vu
dx,
uv
+ 2uv
= vu
+ 2vu
,
(vu
uv
= 2(vu
uv
= 2. Thus,
= ae
2x
.
b) Divide by u and integrate:
u
+ 2u
+(x) u = u,
_
1
0
u
u
dx + 2
_
1
0
u
u
dx +
_
1
0
(x) dx =
_
1
0
dx,
_
1
0
1
u
..
f
u
..
g
dx + 2
_
1
0
u
u
dx +
_
1
0
(x) dx = ,
1
u
u
[
1
0
_
1
0
1
u
2
u
. .
f
..
g
dx + 2
_
1
0
u
u
dx +
_
1
0
(x) dx = ,
_
1
0
u
2
u
2
dx + 2
_
1
0
u
u
dx +
_
1
0
(x) dx = .
In order to have > 0, we must prove that there exists u(x) such that
_
1
0
_
_
u
u
_
2
+ 2
u
u
_
dx > 0.
We can choose to have
_
u
u
_
2
+ 2
u
u
> 0,
which means that
u
u
> 0 or
u
u
< 2. For example, if u(x) = e
cx
with c > 0, we
have
u
u
= c > 0.
For such u(x), > 0.
Problem (F99, #7). Consider the dierential operator
L =
_
d
dx
_
2
+ 2
_
d
dx
_
.
The domain is x [0, 1], with boundary conditions u(0) = u(1) = 0.
a) Find a function = (x) for which L is self-adjoint in the norm
[[u[[
2
=
_
1
0
u
2
dx.
b) If a < 0 show that L +aI is invertible.
c) Find a value of a, so that (L +aI)u = 0 has a nontrivial solution.
Proof. a) Ly = y
+ 2y
. L is self-adjoint in the above norm, if

_
1
0
_
uL(v) vL(u)
dx = 0, or
_
1
0
uL(v)dx =
_
1
0
vL(u)dx,
_
1
0
u(v
+ 2v
)dx =
_
1
0
v(u
+ 2u
)dx,
_
1
0
v
..
g
u
..
f
dx + 2
_
1
0
uv
dx =
_
1
0
u
..
g
v
..
f
dx + 2
_
1
0
vu
dx,
v
u[
1
0
_
1
0
v
(u
+u
) dx + 2
_
1
0
uv
dx = u
v[
1
0
_
1
0
u
(v
+v
) dx + 2
_
1
0
vu
dx.
Boundary terms are 0 due to boundary conditions. Cancelling out other terms, we get
_
1
0
uv
dx + 2
_
1
0
uv
dx =
_
1
0
vu
dx + 2
_
1
0
vu
dx,
uv
+ 2uv
= vu
+ 2vu
,
(vu
uv
= 2(vu
uv
),
= 2,
Thus,
= ae
2x
.
b) L +aI is invertible if the following holds:
(L+ aI)u = 0 u 0.
[ u = 0 (L +aI)u = 0.
[ We have
(L +aI)u = 0,
Lu +au = 0,
u
+ 2u
+au = 0.
Multiply by u and integrate:
_
1
0
uu
dx +
_
1
0
2uu
dx +
_
1
0
au
2
dx = 0,
uu
[
1
0
. .
=0
_
1
0
(u
)
2
dx + 2uu[
1
0
. .
=0
_
1
0
2u
u dx
. .
=0, since
_
1
0
2u
u=
_
1
0
2u
u
+
_
1
0
au
2
dx = 0,
_
1
0
(u
)
2
dx +
_
1
0
au
2
dx = 0,
_
1
0
_
(u
)
2
+ au
2
_
. .
0, (a<0)
dx = 0.
Thus, u 0.
[ We could also solve the equation directly and show u 0.
(L +aI)u = 0,
Lu +au = 0,
u
+ 2u
+au = 0,
u = ce
sx
, (anzats)
u(x) = c
1
e
(1+
1a)x
+c
2
e
(1
1a)x
,
u(0) = 0 = c
1
+c
2
c
1
= c
2
.
u(1) = 0 = c
1
e
1+
1a
c
1
e
1
1a
,
0 = c
1
e
1
(e
1a
e
1a
),
c
1
= 0 c
2
= 0 u 0.
c) We want to nd a value of a, so that (L +aI)u = 0 has a nontrivial solution.
u
+ 2u
+au = 0,
u(x) = c
1
e
(1+
1a)x
+ c
2
e
(1
1a)x
.
Let a = 1 +
2
. Then
u(x) = c
1
e
(1+
2
)x
+c
2
e
(1
2
)x
= c
1
e
(1+i)x
+ c
2
e
(1i)x
= c
1
e
x
e
ix
+ c
2
e
x
e
ix
= c
1
e
x
(cos x + i sinx) + c
2
e
x
(cos x i sinx),
u(0) = 0 = c
1
+c
2
c
1
= c
2
.
u(x) = c
1
e
x
(cos x + i sinx) c
1
e
x
(cos x i sinx) = 2ic
1
e
x
sinx.
u(1) = 0.
Let c
1
= i. Then, u(x) = 2e
x
sinx, is a nontrivial solution.
Problem (F90, #6). Consider the dierential-dierence operator
Lu = u
(x) + u
(x 1) + 3u(x)
dened on 0 < x < 3/2 along with the boundary conditions u(x) 0 on 1 x 0
and u(3/2) = 0. Determine the adjoint operator and the adjoint boundary conditions.
Hint: Take the inner product to be (u, v)
_
3/2
0
u(x)v(x) dx.
Proof. The adjoint operator of L is L
, such that
_ 3
2
0
_
uLv vL
dx = H(x)
3
2
0
.
_ 3
2
0
uLv dx =
_ 3
2
0
u
_
v
(x) +v
(x 1) + 3v(x)
_
dx
=
_ 3
2
0
u(x)v
(x) +
_ 3
2
0
u(x)v
(x 1) + 3
_ 3
2
0
u(x)v(x) =
Change of variables: y = x 1, dy = dx, then
_ 3
2
0
u(x)v
(x 1) dx =
_ 1
2
1
u(y + 1)v
(y) dy =
_ 1
2
1
u(x + 1)v
(x) dx.
=
_ 3
2
0
u(x)v
(x) +
_ 1
2
1
u(x + 1)v
(x) dx + 3
_ 3
2
0
u(x)v(x)
= u(x)v
(x)
3
2
0
. .
=0
_ 3
2
0
u
(x)v
(x) +u(x + 1)v(x)
1
2
1
. .
=0
_ 1
2
1
u
(x + 1)v(x) + 3
_ 3
2
0
u(x)v(x)
= u
(x)v(x)
3
2
0
. .
=0
+
_ 3
2
0
u
(x)v(x)
_ 1
2
1
u
(x + 1)v(x) + 3
_ 3
2
0
u(x)v(x)
=
_ 3
2
0
u
(x)v(x)
_ 1
2
1
u
(x + 1)v(x) + 3
_ 3
2
0
u(x)v(x)
=
_ 3
2
0
u
(x)v(x)
_ 3
2
0
u
(x + 1)v(x) + 3
_ 3
2
0
u(x)v(x) (if u 0 for x [1, 0], [
1
2
,
3
2
])
=
_ 3
2
0
_
u
(x)v(x) u
(x + 1)v(x) + 3u(x)v(x)
_
dx
=
_ 3
2
0
v
_
u
(x) u
(x + 1) + 3u(x)
_
dx =
_ 3
2
0
vL
u dx.
Thus, the adjoint boundary conditions are u 0 for 1 x 0,
1
2
x
3
2
, and
L
u = u
(x) u
(x + 1) + 3u(x).
Problem (S92, #2). Consider the two point boundary value problem
y
+ a(x)y
+b(x)y
+c(x)y = F 0 x 1
with boundary conditions
y(0) = 0, y
(0) = y
(0), y(1) = 0, y
(1) = y
(1).
Here a, b, c are real C
-smooth functions and , are real constants.

a) Derive necessary and sucient conditions for a, b, c, , such that the problem is
self-adjoint.
Proof. a) METHOD I: L is self-adjoint if
L = L
,
y
+ ay
+ by
+cy = y
(ay)
(by)
+ cy,
ay
+ by
= (ay)
(by)
,
ay
+ by
= a
y 3a
3a
ay
y by
,
2ay
+ 3a
+ (3a
+ 2b)y
+ (a
+ b
)y = 0,
a = 0, b = 0, c arbitrary.
METHOD II: L is self-adjoint if
(Lu[v) = (u[Lv), or
_
1
0
uL(v) dx =
_
1
0
vL(u) dx.
In the procedure below, we integrate each term of uL(v) by parts at most 4 times to
get
_
1
0
uL(v) dx =
_
1
0
vL(u) dx + F(x),
and set F(x) = 0, which determines the conditions on a, b and c.
_
1
0
uL(v) dx =
_
1
0
u(v
+av
+bv
+ cv) dx
=
_
1
0
uv
+
_
1
0
auv
+
_
1
0
buv
+
_
1
0
cuv
= uv
[
1
0
. .
=0
_
1
0
u
+auv
[
1
0
. .
=0
_
1
0
(a
uv
+ au
) + buv[
1
0
. .
=0
_
1
0
(b
uv +bu
v) +
_
1
0
cuv
= u
[
1
0
+
_
1
0
u
uv
[
1
0
. .
=0
+
_
1
0
(a
uv
+a
) au
[
1
0
+
_
1
0
(a
+au
)
_
1
0
(b
uv +bu
v) +
_
1
0
= u
[
1
0
+u
[
1
0
_
1
0
u
+a
uv[
1
0
. .
=0
_
1
0
(a
uv + a
v) + a
v[
1
0
. .
=0
_
1
0
(a
v +a
v)
au
[
1
0
+a
v[
1
0
. .
=0
_
1
0
(a
v +a
v) +au
v[
1
0
. .
=0
_
1
0
(a
v + au
v)
_
1
0
(b
uv + bu
v) +
_
1
0
cuv
= u
[
1
0
+u
[
1
0
u
v[
1
0
. .
=0
+
_
1
0
u
v
_
1
0
(a
uv +a
v)
_
1
0
(a
v +a
v)
au
[
1
0
_
1
0
(a
v +a
v)
_
1
0
(a
v +au
v)
_
1
0
(b
uv +bu
v) +
_
1
0
cuv
= u
[
1
0
+u
[
1
0
au
[
1
0
+
_
1
0
(u
u a
au
u bu
+cu)v
= u
[
1
0
+u
[
1
0
au
[
1
0
+
_
1
0
(u
u 3a
3a
au
u bu
+cu)v
=
_
1
0
v(u
+au
+ bu
+ cu)
u
[
1
0
+ u
[
1
0
au
[
1
0
+
_
1
0
(a
u 3a
3a
2au
u 2bu
)v
=
_
1
0
vL(u) dx u
[
1
0
+u
[
1
0
au
[
1
0
+
_
1
0
_
(a
)u (3a
+ 2b)u
3a
2au
_
v.
Thus, L is self-adjoint if
_
1
0
_
(a
)u (3a
+ 2b)u
3a
2au
_
v = 0, or
a = 0, b = 0, c arbitrary. Also, need
u
(1)v
(1) +u
(0)v
(0) + u
(1)v
(1) u
(0)v
(0) au
[
1
0
. .
=0, (a=0)
= 0,
u
(1)v
(1) + u
(0)v
(0) + u
(1)v
(1) u
(0)v
(0) = 0.
Thus, , = 0.
Note that both Methods I and II give the same answers. However, we need to use
Method II in order to obtain information about boundary conditions.
b) Assume that c(x) = c
0
is constant and that the problem is self-adjoint. Deter-
minte the eigenvalues and eigenfunctions and show that they form a complete
orthonormal set.
From part (a), we have
y
+ c
0
y = F 0 x 1
y(0) = 0, y
(0) = 0, y(1) = 0, y
(1) = 0.
The eigenvalue problem is
y
+c
0
y = y,
y
( c
0
)y = 0.
To determine eigenfunctions, try y = a cos(c
0
)
1
4
x+b sin(c
0
)
1
4
x. Initial conditions
give
y(0) = a = 0 y = b sin( c
0
)
1
4
x,
y(1) = b sin( c
0
)
1
4
= 0 ( c
0
)
1
4
= n
n
= n
4
4
+ c
0
.
Thus, the eigenvalues and eigenfunctions are
n
= n
4
4
+c
0
, y
n
= sin(
n
c
0
)
1
4
x = sinnx, n = 1, 2, . . . .
We could also use the table to nd out that the eigenfunctions are y = sin
nx
L
=
sinnx. We have
y
+c
0
y = y,
(sinnx)
+c
0
sin nx = sinnx,
n
4
4
sinnx +c
0
sin nx = sinnx,
n
4
4
+c
0
=
n
. |
The normalized eigenfunctions form an orthonormal set
_
1
0
(
2 sinnx) (
2 sinmx) dx =
_
0 n ,= m
1 n = m
Any smooth function f can be written in terms of eigenfunctions f(x) =

n=1
a
n
2 sin nx.
c) Use the eigenfunctions to construct the Greens function.
We have
y
+ c
0
y = F(x), (5.7)
y(0) = 0, y
(0) = 0, y(1) = 0, y
(1) = 0. (5.8)
The related eigenvalue problem is
y
+ c
0
y = y
y(0) = 0, y
(0) = 0, y(1) = 0, y
(1) = 0.
The eigenvalues are
n
= n
4
4
+c
0
, and the corresponding eigenfunctions are sinnx,
n = 1, 2, . . ..
Writing y =

a
n
n
=

a
n
sin nx and inserting into (5.7), we get
n=1
_
a
n
n
4
4
sinnx + c
0
a
n
sinnx
_
= F(x),
n=1
a
n
(n
4
4
+c
0
) sinnx = F(x),
_
1
0
n=1
a
n
(n
4
4
+ c
0
) sinnx sinmx dx =
_
1
0
F(x) sin mx dx,
a
n
(n
4
4
+c
0
)
1
2
=
_
1
0
F(x) sin nx dx,
a
n
=
2
_
1
0
F(x) sin nx dx
n
4
4
+c
0
.
y(x) =

a
n
sinnx =
n=1
2
_
1
0
F() sinnx sin n d
n
4
4
+ c
0
,
y =
_
1
0
F()
_
2
n=1
sin nx sinn
n
4
4
+c
0
_
. .
= G(x,)
d.
See a less complicated problem, y
= f, in Poisson Equation subsection of Eigenvalues

of the Laplacian section (PDEs).
Problem (S91, #5). Dene the operator
Lu = u
xxxx
+a(x)u
xx
+b(x)u
x
+c(x)u
for 0 x 2 with boundary conditions
u = u
xx
= 0 on x = 0, 2.
a) Find conditions on the functions a, b and c so that L is self-adjoint.
b) For a = b = 0 and c = constant, nd the fundamental solution for the PDE
u
t
= Lu
as a Fourier series in x.
Proof. a) METHOD I: L is self-adjoint if
L = L
,
u
+ au
+bu
+cu = u
+ (au)
(bu)
+cu,
au
+ bu
= (au)
(bu)
,
au
+ bu
= a
u + 2a
+ au
u bu
,
0 = a
u + 2a
u 2bu
,
0 = (a
)u + 2(a
b)u
,
a
= b, c arbitrary.
METHOD II: L is self-adjoint if
(Lu[v) = (u[Lv), or
_
2
0
uL(v) dx =
_
2
0
vL(u) dx.
In the procedure below, we integrate each term of uL(v) by parts at most 4 times to
get
_
2
0
uL(v) dx =
_
2
0
vL(u) dx + F(x),
and set F(x) = 0, which determines the conditions on a, b and c.
_
2
0
uL(v) dx =
_
2
0
u(v
+av
+ bv
+ cv) dx
=
_
2
0
uv
+
_
2
0
auv
+
_
2
0
buv
+
_
2
0
cuv
= uv
[
2
0
. .
=0
_
2
0
u
+auv
[
2
0
. .
=0
_
2
0
(a
uv
+ au
) + buv[
2
0
. .
=0
_
2
0
(b
uv +bu
v) +
_
2
0
cuv
= u
[
2
0
. .
=0
+
_
2
0
u
uv[
2
0
. .
=0
+
_
2
0
(a
uv +a
v) au
v[
2
0
. .
=0
+
_
2
0
(a
v + au
v)
_
2
0
(b
uv + bu
v) +
= u
[
2
0
. .
=0
_
2
0
u
+
_
2
0
(a
uv + a
v) +
_
2
0
(a
v +au
v)
_
2
0
(b
uv + bu
v) +
_
2
0
cuv
= u
v[
2
0
. .
=0
+
_
2
0
(u
v +a
uv + a
v + a
v +au
v b
uv bu
v +cuv)
=
_
2
0
v(u
+ au
+ bu
+cu) +
_
2
0
(a
uv + 2a
v b
uv 2bu
v)
=
_
2
0
vL(u) dx +
_
2
0
(a
uv + 2a
v b
uv 2bu
v).
Thus, L is self-adjoint if
_
2
0
(a
u + 2a
u 2bu
)v = 0, or a
= b, c arbitrary.
b) For a = b = 0 and c = constant, nd the fundamental solution for the PDE
u
t
= Lu
as a Fourier series in x.
We have u
t
= Lu = u
cu. We rst need to nd eigenfunctions and eigenvalues.

The eigenvalue problem is
u
+cu = u,
u
( c)u = 0,
u = u
xx
= 0 on x = 0, 2.
To determine eigenfunctions, try u = a cos(c)
1
4
x+b sin(c)
1
4
x. Initial conditions:
u(0) = a = 0 u = b sin( c)
1
4
x,
u(2) = 0 = b sin( c)
1
4
2 = 0 ( c)
1
4
2 = n
n
=
n
4
16
+c.
n
=
n
4
16
+c, u
n
= sin(
n
c)
1
4
x = sin
nx
2
, n = 1, 2, . . . .
Let u(x, t) =
n=1
u
n
(t) sin
nx
2
.
u(x, t) =
n=1
_
u
n
(t) sin
nx
2
+u
n
(t)
n
4
16
sin
nx
2
+cu
n
(t) sin
nx
2
_
= 0,
u
n
(t) + u
n
(t)
n
4
16
+cu
n
(t) = 0,
u
n
(t) +
_
n
4
16
+c
_
u
n
(t) = 0,
u
n
(t) = c
n
e
(
n
4
16
+c)t
.
u(x, t) =
n=1
c
n
e
(
n
4
16
+c)t
sin
nx
2
.
In order to determine c
n
we need initial conditions u(x, 0) = f(x). Then
11
u(x, 0) =
n=1
c
n
sin
nx
2
dx = f(x).
c
n
=
_
2
0
f(x) sin
nx
2
dx,
c
n
=
1
_
2
0
f(x) sin
nx
2
dx.
u(x, t) =
n=1
c
n
e
(
n
4
16
+c)t
sin
nx
2
=
n=1
1
_
2
0
f() sin
n
2
e
(
n
4
16
+c)t
sin
nx
2
d.
11
ChiuYens solutions list G(x, t; x0, t0) =
n=1
1
sin
nx
0
2
e
(
n
4
16
+c)(tt
0
)
sin
nx
2
. Similar result may
be found in Haberman, p. 383.
u(x, t) =
_
2
0
f()
n=1
1
sin
n
2
sin
nx
2
e
(
n
4
16
+c)t
. .
= G(x,t;x
0
,t
0
)
d.
5.8 Orthogonality of Eigenfunctions
Denition
12
. A positive, continuous function r(x) dened on [a, b] is called a weight
function. Two continuous functions f(x) and h(x) dened on [a, b] are said to be
orthogonal on [a, b] with respect to the weight function r(x), if
_
b
a
f(x)h(x)r(x)dx = 0.
Theorem
13
. Let
m
and
n
be two distinct eigenvalues of the Sturm-Liouville problem
(5.3). Then the corresponding eigenfunctions y
m
(x) and y
n
(x) are orthogonal on [a, b]
with respect to the weight function r(x).
_
b
a
y
m
(x)y
n
(x)r(x)dx = 0.
Proof. We have the relations
(py
m
)
+ qy
m
+
m
ry
m
= 0, (5.9)
(py
n
)
+ qy
n
+
n
ry
n
= 0. (5.10)
Multiply (5.9) by y
n
and (5.10) by y
m
14
(
n

m
)ry
m
y
n
= y
n
(py
m
)
y
m
(py
n
)
= [p(y
n
y
m
y
m
y
n
)]
. (5.11)
Integrating both sides of (5.11) over (a, b) gives
(
n

m
)
_
b
a
y
m
y
n
r = [p(y
n
y
m
y
m
y
n
)]
b
a
.
The boundary conditions in (5.3) ensure that the right side vanishes (e.g. if c
2
,= 0, then
y
(a) =
c
1
c
2
y(a), and y
n
(a)y
m
(a) y
m
(a)y
n
(a) = y
n
(a)
c
1
c
2
y
m
(a) +y
m
(a)
c
1
c
2
y
n
(a) = 0).
Thus,
(
n

m
)
_
b
a
y
m
y
n
r = 0.
Since
n
,=
m
, y
n
(x) and y
m
(x) are orthogonal on [a, b] with respect to the weight
function r(x).
12
Bleecker and Csordas, p. 266.
13
14
Note an important identity:
yn(py
m
)
ym(py
n
)
= [p(yny
m
ymy
n
)]
.
Problem (S90, #3). Consider the eigenvalue problem
a(x)
d
2
u(x)
dx
2
= u(x), 0 < x < 1,
with the boundary conditions u(0) = 0, u
(1) = 0. Here 0 < c

1
a(x) c
2
is a smooth
function on [0, 1]. Let
n
, n = 1, . . ., be the eigenvalues and
n
(x) be the corresponding
eigenfunctions. Prove that there is a weight (x) such that
_
1
0
m
(x)
n
(x)(x) dx = 0 for m ,= n.
Proof. Rewrite the equation as
u
1
a(x)
u = 0.
Let
m
,
n
, be the eigenvalues and u
m
, u
n
be the corresponding eigenfunctions. We
have
u
m

m
1
a(x)
u
m
= 0, (5.12)
u
n

n
1
a(x)
u
n
= 0. (5.13)
Multiply (5.12) by u
n
and (5.13) by u
m
u
n
u
m
=
m
1
a(x)
u
m
u
n
,
u
m
u
n
=
n
1
a(x)
u
n
u
m
.
(
m

n
)
1
a(x)
u
m
u
n
= u
n
u
m
u
m
u
n
= (u
n
u
m
u
m
u
n
)
.
Integrating over (0, 1) gives
(
m

n
)
_
1
0
1
a(x)
u
m
u
n
dx = [u
n
u
m
u
m
u
n
]
1
0
= 0.
Since
n
,=
m
, u
n
(x) and u
m
(x) are orthogonal on [0, 1] with respect to the weight
function (x) =
1
a(x)
.
5.9 Real Eigenvalues
Theorem
15
. For any regular Sturm-Liouville problem, all the eigenvalues are real.
Proof. We can use orthogonality of eigenfunctions to prove that the eigenvalues are
real. Suppose that is a complex eigenvalue and (x) the corresponding eigenfunction
(also allowed to be complex since the dierential equation dening the eigenfunction
would be complex):
L() + r = 0. (5.14)
Thus, the complex conjugate of (5.14) is also valid:
L() + r = 0, (5.15)
15
Haberman, edition 4, p. 178.
assuming that r is real. Since the coecients of a linear operator L =
d
dx
_
p
d
dx
_
+q are
real, L() = L(). Thus,
L() + r = 0.
If satises boundary conditions with real coecients, for example c
1
(a)+c
2
(a) = 0,
then satises the same boundary conditions, c
1
(a) + c
2
(a) = 0. Equation (5.14)

and the boundary conditions show that satises the Sturm-Liouville problem, but
with eigenvalue being . Thus, if is a complex eigenvalue with corresponding
eigenfunction , then is also an eigenvalue with corresponding eigenfunc-
tion .
Using orthogonality of eigenfunctions, and are orthogonal (with weight r). Thus,
_
b
a
r dx = 0.
Since = [[
2
0 and r > 0, the integral above is 0. In fact, the integral can equal
0 only if 0, which is prohibited since is an eigenfunction. Thus, = , and hence
is real.
5.10 Unique Eigenfunctions
Theorem. Consider the Sturm-Liouville problem (5.3). If y
1
(x) and y
2
(x) are two
eigenfunctions corresponding to the same eigenvalue , then y
1
(x) = y
2
(x), a x b,
for some nonzero constant , (i.e., y
1
(x) and y
2
(x) are linearly dependent).
Proof.
16
Method 1: Suppose that there are two dierent eigenfunctions y
1
and y
2
corresponding to the same eigenvalue . In this case,
L(y
1
) + ry
1
= 0,
L(y
2
) + ry
2
= 0.
0 = y
2
_
L(y
1
) +ry
1
_
y
1
_
L(y
2
) +ry
2
_
= y
2
L(y
1
) y
1
L(y
2
) =
_
p
_
y
2
y
1
y
1
y
2
_
_
,
where the Lagranges identity was used in the last equality. It follows that
p
_
y
2
y
1
y
1
y
2
_
= constant.
This constant is evaluated from the boundary conditions and is equal to 0 if the bound-
ary conditions are of the Sturm-Liouville type. Thus,
y
2
y
1
y
1
y
2
= 0.
This is equivalent to
d
dx
(
y
1
y
2
) = 0, and hence for these boundary conditions
y
2
= cy
1
.
Thus, the two eigenfunctions are dependent; the eigenfunction is unique.
Proof.
17
Method 2: Consider the function
w(x) = y
2
(a)y
1
(x) y
1
(a)y
2
(x),
and suppose that
y
1
(a)
2
+y
2
(a)
2
,= 0. (5.16)
Then w(x) satises the following initial-value problem
Lw + rw = 0 a x b
_
(pw
+ qw + rw = 0

w(a) = w
(a) = 0.
Check that w(x) indeed satises the initial-value problem:
(pw
+ qw +rw =
_
p
_
y
2
(a)y
1
(x) y
1
(a)y
2
(x)
_
+q
_
y
2
(a)y
1
(x) y
1
(a)y
2
(x)
_
+r
_
y
2
(a)y
1
(x) y
1
(a)y
2
(x)
_
= y
2
(a)
_
(py
1
(x))
+ qy
1
(x) + ry
1
(x)
_
y
1
(a)
_
(py
2
(x))
+ qy
2
(x) + ry
2
(x)
_
= 0,
since y
1
and y
2
are eigenfunctions. Also,
w(a) = y
2
(a)y
1
(a) y
1
(a)y
2
(a) =
c
1
c
2
y
2
(a)y
1
(a) +
c
1
c
2
y
1
(a)y
2
(a) = 0,
w
(a) = y
2
(a)y
1
(a) y
1
(a)y
2
(a) = 0. |
16
Haberman, edition 4, p. 179.
17
By the uniqueness theorem for initial-value problems, w(x) 0. Therefore,
y
2
(a)y
1
(x) y
1
(a)y
2
(x) 0, a x b. (5.17)
Since y
1
(x) and y
2
(x) are eigenfunctions, y
1
(x) and y
2
(x) are not identically 0. Hence,
(5.16) and (5.17) imply that y
1
(a)y
2
(a) ,= 0. Thus, by (5.17), y
1
(x) = y
2
(x), where
= y
1
(a)/y
2
(a).
Remark: In the theorem above, we showed that, for the Sturm-Liouville prob-
lem (5.3), there is only one linearly independent eigenfunction associated with each
eigenvalue . For this reason, is said to be simple.
5.11 Rayleigh Quotient
Theorem
18
. Any eigenvalue can be related to its eigenfunction by the Rayleigh quo-
tient:
=
p
[
b
a
+
_
b
a
_
p(
)
2
q
2
dx
_
b
a

2
r dx
.
Proof. The Rayleigh quotient can be derived from the Sturm-Liouville dierential equa-
tion,
(p
+ q +r = 0, (5.18)
by multiplying (5.18) by and integrating:
_
b
a
_
(p
+ q
2
_
dx +
_
b
a
r
2
dx = 0.
Since
_
b
a
r
2
> 0, we can solve for :
=
_
b
a
_
(p
q
2
dx
_
b
a
r
2
dx
.
Integrating by parts gives
=
p
[
b
a
+
_
b
a
_
p(
)
2
q
2
dx
_
b
a
r
2
dx
.
Note: Given the equation:
1
x
(xf
+f = 0,
we can obtain
=
_
1
0
xf
2
dx
_
1
0
xf
2
dx
0.
18
Haberman, edition 4, Theorem 6, p. 189.
We can establish the Rayleigh-Ritz principle, namely that
F(f) =
_
1
0
x(f
)
2
dx
_
1
0
xf
2
dx
is an upper bound on the smallest eigenvalue.
Let f(x) =

a
n
f
n
, where f
n
s are eigenfunctions. Then,
F(f) =
_
1
0
x(f
)
2
dx
_
1
0
xf
2
dx
=
_
1
0
x(
a
n
f
n
)
2
dx
_
1
0
x(
a
n
f
n
)
2
dx
(by orthogonality)
=
a
2
n
_
1
0
xf
2
n
dx
a
2
n
_
1
0
xf
2
n
dx
= =
a
2
n
n
_
1
0
xf
2
n
dx
a
2
n
_
1
0
xf
2
n
dx
>
min
a
2
n
_
1
0
xf
2
n
dx
a
2
n
_
1
0
xf
2
n
dx
=
min
.
5.12 More Problems
Example. Determine the eigenvalues and eigenfunctions of the Sturm-Liouville prob-
lem
y
+y = 0, 0 x L
y(0) = 0, y(L) = 0.
Proof. Note that we get this equation from (5.1) with p 1, q 0, r 1, a = 0,
b = L. We consider the three cases > 0, = 0, < 0.
If = 0, the ODE reduces to y
= 0. Try y(x) = Ax + B.
Applying the rst boundary condition gives y(0) = 0 = B. The second boundary
condition gives y(L) = 0 = AL, or A = 0. Therefore, the only solution for this case is
the trivial solution, y(x) 0, which is not an eigenfunction, and therefore, 0 is not an
eigenvalue.
If < 0, or =
2
, the ODE becomes
y
2
y = 0.
The anzats y = e
sx
gives s
2
2
= 0, or s = . Thus the general solution is
y(x) = Ae
x
+ Be
x
.
Applying the rst boundary condition gives
y(0) = 0 = A+B, or B = A.
The second boundary condition gives
y(L) = 0 = A(e
L
e
L
) = 2AsinhL, or A = 0.
Thus, the only solution is the trivial solution, y(x) 0, which is not an eigenfunction,
and therefore, there are no negative eigenvalues.
If > 0, try = +
2
y
+
2
y = 0,
with the anzats y = e
sx
, which gives s = i with the family of solutions
y(x) = Asinx +B cos x.
Applying the rst boundary condition gives
y(0) = 0 = B.
The second boundary condition gives
y(L) = 0 = AsinL.
Since we want nontrivial solutions, A ,= 0, and we set Asin L = 0, obtaining L = n.
Thus the eigenvalues and the corresponding eigenfunctions are
=
n
=
_
n
L
_
2
, y
n
(x) = A
n
sin
_
nx
L
_
.
Also, the eigenfunctions can always be used to represent any piecewise smooth func-
tion f(x),
f(x)
n=1
a
n
y
n
(x).
Thus, for our example,
f(x)
n=1
a
n
sin
nx
L
. |
Problem (F98, #3). Consider the eigenvalue problem
d
2
dx
2
+ = 0,
(0)
d
dx
(0) = 0, (1) +
d
dx
(1) = 0.
a) Show that all eigenvalues are positive.
b) Show that there exist a sequence of eigenvalues =
n
, each of which satises
tan
=
2
1
.
Proof. a) Method . 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) = e

sx
, which gives s =
= R.
Hence, the family of solutions is (x) = Ae
x
+Be
x
. Also,
(x) = Ae
x
Be
x
.
The boundary conditions give
(0)
(0) = 0 = A+ B A+ B = A(1 ) + B(1 +), (5.19)

(1) +
(1) = 0 = Ae
+Be
+Ae
Be
= Ae
(1+) +Be
(1). (5.20)
From (5.19) and (5.20) we get
1 +
1
=
A
B
and
1 +
1
=
B
A
e
2
, or
A
B
= e
.
From (5.19), =
A+ B
A B
and thus,
A
B
= e
A+B
BA
, which has no solutions.
Method . Multiply the equation by and integrate from 0 to 1.
_
1
0
dx +
_
1
0
2
dx = 0,
[
1
0
_
1
0
(
)
2
dx +
_
1
0
2
dx = 0,
=
(1)
(1) + (0)
(0) +
_
1
0
(
)
2
dx
_
1
0

2
dx
=
(1)
2
+(0)
2
+
_
1
0
(
)
2
dx
_
1
0

2
dx
.
Thus, > 0 for not identically 0.
b) Since > 0, the anzats = e
sx
gives s = i
and the family of solutions takes

the form
(x) = Asin(x
) +B cos(x
).
Then,
(x) = A
cos(x
) B
sin(x
). The rst boundary condition gives

(0)
(0) = 0 = B A
B = A
.
Hence, (x) = Asin(x
) + A
cos(x
). The second boundary condition gives

(1) +
(1) = 0 = Asin(
) + A
cos(
) + A
cos(
) A sin(
)
= A
_
(1 ) sin(
) + 2
cos(
A ,= 0 (since A = 0 implies B = 0 and = 0, which is not an eigenfunction). Therefore,

(1 ) sin(
) = 2
cos(
), and thus tan(
) =
2
1
.
Problem (F02, #2). Consider the second order dierential operator L dened by
Lu = u
+xu
for 0 < x < with boundary conditions
u(0) = u() = 0.
a) For = 0 nd the leading (i.e. smallest) eigenvalue
0
and the corresponding
eigenfunction
0
for L.
b) For > 0 look for the eigenvalues and eigenfunctions to have an expansion of the
form
=
0
+
1
+O(
2
),
=
0
+
1
+O(
2
).
Find formulas for
1
and
1
(your formulas will contain denite integrals which you
do not need to evaluate).
Proof. a) Since = 0, the eigenvalue problem for =
2
becomes
u
+
2
u = 0.
The equation has solutions in the form
u(x) = Asinx + Bcos x.
The rst boundary condition gives u(0) = 0 = B, and the second gives u() = 0 =
Asin. Since we are looking for nontrivial solutions, A ,= 0 and sin = 0, which
gives = 1, 2, 3, . . .. Thus, the smallest eigenvalue and the corresponding eigenfunction
are
0
= 1,
0
= sinx.
b) For > 0, we have
u
+ xu u = 0,
(
0
+
1
)
+ x(
0
+
1
) (
0
+
1
)(
0
+
1
) = 0,
1
+x
0
+
2
x
1

0
1
= 0.
Drop O(
2
) terms. Since
0
+
0
0
= 0,
1
+x
0
0
= 0,
1
+ x
0
1

1
0
= 0,
1
+ x sinx
1

1
sinx = 0,
1
+
1
= x sinx
1
sinx.
Multiplying by
0
and using orthogonality of the eigenfunctions
19
, we get
_

0
1
dx +
_

0
1
dx
. .
=0
=
_

0
(x sin
2
x
1
sin
2
x)dx,
1
[
0

_

0
1
dx =
_

0
(x sin
2
x
1
sin
2
x)dx, (integration by parts)
0 =
_

0
(x sin
2
x
1
sin
2
x)dx,
1
_

0
sin
2
x dx =
_

0
x sin
2
x dx,
19
Bleecker and Csordas, p. 267, p. 274.
1
=
_
0
x sin
2
x dx
_
0
sin
2
x dx
Since
1
is known, we should be able to solve the ODE
1
+
1
= x sinx
1
sin x by
the variation of parameters.
Problem (F00, #5). Consider the eigenvalue problem on the interval [0, 1],
y
(t) + p(t)y(t) = y(t),

y(0) = y(1) = 0.
a) Prove that all eigenvalues are simple.
b) Prove that there is at most a nite number of negative eigenvalues.
a) In order to show that is simple, need to show that there is only one linearly
independent eigenfunction associated with each eigenvalue .
Proof. Method 1: Let y
1
(x) and y
2
(x) be two eigenfunctions corresponding to the
same eigenvalue . We will show that y
1
and y
2
are linearly dependent. We have
y
1
+ py
1
y
1
= 0,
y
2
+ py
2
y
2
= 0.
0 = y
2
_
y
1
+py
1
y
1
_
y
1
_
y
2
+ py
2
y
2
_
= y
1
y
2
y
2
y
1
= [y
1
y
2
y
2
y
1
]
,
where Lagranges identity was used in the last equality. It follows that
y
1
y
2
y
2
y
1
= constant.
Using boundary conditions,
_
y
1
y
2
y
2
y
1
_
(0) = 0.
Therefore, y
1
y
2
y
2
y
1
0. This is equivalent to
_
y
2
y
1
_
= 0, and hence
y
2
= cy
1
.
Thus the two eigenfunctions are dependent; the eigenfunction is unique, and simple.
Proof. Method 2: Let y
1
(x) and y
2
(x) be two eigenfunctions corresponding to the
same eigenvalue . We will show that y
1
and y
2
are linearly dependent. We only
consider the case with
y
1
(0)
2
+y
2
(0)
2
,= 0. (5.21)
Consider the function
w(x) = y
2
(0)y
1
(x) y
1
(0)y
2
(x),
Then w(x) satises the following initial-value problem
w
+ pw w = 0 0 x 1
w(0) = w
(0) = 0.
Check that w(x) indeed satises the initial-value problem:
w
+pw w =
_
y
2
(0)y
1
(x) y
1
(0)y
2
(x)
+p
_
y
2
(0)y
1
(x) y
1
(0)y
2
(x)
_
y
2
(0)y
1
(x) y
1
(0)y
2
(x)
= y
2
(0)
_
y
1
(x) +py
1
(x) y
1
(x)
1
(0)
_
y
2
(x) +py
2
(x) y
2
(x)
= 0,
since y
1
and y
2
are eigenfunctions. Also,
w(0) = y
2
(0)y
1
(0) y
1
(0)y
2
(0) = y
2
(0) 0 y
1
(0) 0 = 0,
w
(0) = y
2
(0)y
1
(0) y
1
(0)y
2
(0) = 0. |
Then, by the uniqueness theorem for initial-value problems, w(x) 0. Therefore,
y
2
(0)y
1
(x) y
1
(0)y
2
(x) 0, 0 x 1. (5.22)
Since y
1
(x) and y
2
(x) are eigenfunctions, y
1
(x) and y
2
(x) are not identically 0. Hence,
(5.21) and (5.22) imply that y
1
(0)y
2
(0) ,= 0. Thus, by (5.22), y
1
(x) = y
2
(x), where
= y
1
(0)/y
2
(0).
y
(t) + p(t)y(t) = y(t),

y(0) = y(1) = 0.
b) Prove that there is at most a nite number of negative eigenvalues.
We need to show that the eigenvalues are bounded below
<
0
<
1
<
2
< . . . ; with
n
as n .
Multiply the equation by y and integrate:
yy
+py
2
= y
2
,
_
1
0
yy
dt +
_
1
0
py
2
dt =
_
1
0
y
2
dt,
yy
1
0
. .
=0
+
_
1
0
(y
)
2
dt +
_
1
0
py
2
dt =
_
1
0
y
2
dt,
=
_
1
0
(y
)
2
dt +
_
1
0
py
2
dt
_
1
0
y
2
dt
.
The Poincare inequality gives:
_
1
0
y
2
dt C
_
1
0
(y
)
2
dt, or
_
1
0
y
2
dt C
_
1
0
(y
)
2
dt.
Thus, we have
=
_
1
0
(y
)
2
dt +
_
1
0
py
2
dt
_
1
0
y
2
dt
_
1
0
(y
)
2
dt max
0x1
[p[
_
1
0
y
2
dt
_
1
0
y
2
dt
1
C
_
1
0
y
2
dt max
0x1
[p[
_
1
0
y
2
dt
_
1
0
y
2
dt
=
_
1
C
max[p[
_
_
1
0
y
2
dt
_
1
0
y
2
dt
=
1
C
max[p[.
Problem (S94, #6). Consider the eigenvalue problem
d
2
u
dx
2
+v(x)u = u on [0, 1]
with the boundary conditions
du
dx
(0) =
du
dx
(1) = 0. Show that if
_
1
0
v(x) dx = 0 then
there is a negative eigenvalue, unless v(x) 0.
Proof. Divide by u and integrate:
u
+v(x)u = u,
_
1
0
u
u
dx +
_
1
0
v(x) dx
. .
=0
=
_
1
0
dx,
_
1
0
1
u
..
f
u
..
g
dx = ,
1
u
u
[
1
0
+
_
1
0
1
u
2
u
. .
f
..
g
dx = ,
0 >
_
1
0
u
2
u
2
dx = .
Thus, < 0.
Problem (S95, #1). Find the eigenfunctions/eigenvalues for the following op-
erator
Lf =
d
2
dx
2
f + 4f < x <
f 2 periodic.
Find all solutions (periodic or non-periodic) for the problems
a) Lf = cos x,
b) Lf = cos 2x.
Proof. To nd eigenfunctions and eigenvalues for L, consider
f
+ 4f + f = 0,
f
+ ( + 4)f = 0.
The anzats f = e
sx
gives s
2
+ ( + 4) = 0, or s =
4.
Case 1: 4 < 0 s = i
+ 4
. .
R
.
Thus, eigenfunctions are cos
+ 4 x, sin
+ 4 x. To make these 2 periodic, need

n =
_
n
+ 4
n
+4 = n
2

n
= 4+n
2
, n = 0, 1, 2, . . . (note: 4 < 0).
n
= 4 +n
2
, cos nx, n = 0, 1, 2, . . . , sinnx, n = 1, 2, . . . .
For example, with n = 1, eigenvalues and eigenfunctions are:
1
= 3, cos x, sin x.
Note that <
0
<
1
<
2
< . . . ; with
n
as n .
Case 2: 4 = 0, ( + 4 = 0)
f
= 0 f = ax + b. Since a ,= 0 does not satisfy periodicity (being a linear

function), a = 0. Since an eigenfunction can not be 0 everywhere b ,= 0. Thus,
= 4, f = b ,= 0
is 2 periodic.
Case 3: 4 > 0 s =
4
Eigenfunctions e
4x
, e
4x
are not 2-periodic.
As in F92 #3, could take f(x) =

a
n
e
inx
, 2periodic. Then
f
+ 4f +f = 0,
n
2
+ 4 + = 0,
n
= 4 + n
2
.
e
inx
, n = 0, 1, 2, . . ., are eigenfunctions.
a) f
+ 4f = cos x. We rst solve the homogeneous equation f
+ 4f = 0. Sub-
stitution f = e
sx
gives s
2
+ 4 = 0. Hence, s
1,2
= 2i and the superposition principle
gives:
f
h
(x) = Acos 2x + Bsin2x.
Find a particular solution of the inhomogeneous equation f
+ 4f = cos x.
Try f(x) = C cos x +Dsinx. Then,
C cos x Dsinx + 4C cos x + 4Dsinx = cos x,
3C cos x + 3Dsinx = cos x,
C =
1
3
, D = 0.
Thus,
f
p
(x) =
1
3
cos x.
f(x) = f
h
(x) +f
p
(x) = Acos 2x +B sin2x +
1
3
cos x.
b) f
+ 4f = cos 2x. In part (a), we already found

f
h
(x) = Acos 2x + Bsin2x.
to be a homogeneous equation. To nd a particular solution of the inhomogeneous
equation, we try
f
p
(x) = Cx cos 2x +Dx sin2x,
f
p
(x) = 2Cx sin 2x +C cos 2x + 2Dx cos 2x +Dsin2x,
f
p
(x) = 4Cx cos 2x 2C sin 2x 2C sin2x 4Dx sin2x + 2Dcos 2x + 2Dcos 2x
= 4Cx cos 2x 4C sin 2x 4Dx sin2x + 4Dcos 2x.
Substitution into f
+ 4f = cos 2x gives:
4Cx cos 2x 4C sin 2x 4Dx sin2x + 4Dcos 2x + 4Cx cos 2x + 4Dx sin2x = cos 2x,
which gives 4C sin 2x + 4Dcos 2x = cos 2x, or C = 0, D =
1
4
.
Thus,
f
p
(x) =
1
4
x sin2x,
f(x) = f
h
(x) +f
p
(x) = Acos 2x +B sin2x +
1
4
x sin2x.
Problem (F92, #3). Denote
Lf =

4
f
x
4
+ 3
2
f
x
2
+f for 0 < x <
for f satisfying
f =

2
x
2
f = 0 for x = 0 and x = .
a) Find the eigenfunctions and eigenvalues for L.
b) Solve the problem
t
f = Lf
f(x, t = 0) =
e
ix
1
1
2
e
ix

e
ix
1
1
2
e
ix
with the boundary conditions above.
Proof. a) In order to nd eigenfunctions and eigenvalues for L, consider
f
+ 3f
+f = f.
Let f(x) =
a
0
2
+
n=1
a
n
cos nx +b
n
sinnx.
f(0) = f() = 0 a
n
= 0, n = 0, 1, 2, . . ..
f
(0) = f
() = 0 a
n
= 0, n = 0, 1, 2, . . ..
f(x) =
n=1
b
n
sinnx.
Thus, the eigenfunctions are sin nx, n = 1, 2, . . .. We have
(sinnx)
+ 3(sinnx)
+ sinnx = sinnx,
(n
4
3n
2
+ 1) sinnx = sinnx,
n
4
3n
2
+ 1 =
n
.
n
= n
4
3n
2
+ 1, f
n
(x) = sin nx, n = 1, 2, . . . .
b) We have
f
t
= f
xxxx
+ 3f
xx
+f,
f(x, 0) =
e
ix
1
1
2
e
ix

e
ix
1
1
2
e
ix
Let f(x, t) =

f
n
(t) sinnx. Then
n=1
f
n
(t) sinnx =
n=1
f
n
(t)n
4
sin nx 3f
n
(t)n
2
sin nx + f
n
(t) sinnx,
f
n
(t) = (n
4
3n
2
+ 1)f
n
(t),
f
n
(t) (n
4
3n
2
+ 1)f
n
(t) = 0,
f
n
(t) = c
n
e
(n
4
3n
2
+1)t
,
f(x, t) =
n=1
c
n
e
(n
4
3n
2
+1)t
sin nx.
Using initial conditions, we have
f(x, 0) =
n=1
c
n
sinnx =
e
ix
1
1
2
e
ix

e
ix
1
1
2
e
ix
=
n=0
e
ix
_
1
2
e
ix
_
n
n=0
e
ix
_
1
2
e
ix
_
n
=
n=0
1
2
n
e
ix(n+1)
n=0
1
2
n
e
ix(n+1)
=
n=0
1
2
n
_
e
ix(n+1)
e
ix(n+1)
_
=
n=0
1
2
n1
i
2i
_
e
ix(n+1)
e
ix(n+1)
_
=
n=0
i
2
n1
sin((n + 1)x)
=
n=1
i
2
n2
sinnx.
Thus, c
n
= i/2
n2
, n = 1, 2, . . ., and
f(x, t) =
n=1
i
2
n2
e
(n
4
3n
2
+1)t
sinnx.
Problem (W02, #2). a) Prove that
_

0
[u(x)[
2
dx
_

0
du
dx
2
dx
for all continuously dierentiable functions u satisfying u(0) = u() = 0.
b) Consider the dierential operator
Lu =
d
2
u
dx
2
+ q(x)u, 0 < x <
with the boundary conditions u(0) = u() = 0. Suppose q is continuous on [0, ] and
q(x) > 1 on [0, ]. Prove that all eigenvalues of L are positive.
Proof. a) Use eigenvalues of the Laplacian for u
+u = 0, u(0) = u() = 0.
Then
n
= sinnx,
n
= n
2
, n = 1, 2, . . ..
Then
_

0
u
2
dx =
_

0
_
m
a
m
m
__
n
a
n
n
_
dx =

n
a
2
n
_

0
sin
2
nx dx
=

n
a
2
n
_

0
1 cos 2nx
2
dx =

n
a
2
n
_
x
2

1
4n
sin 2nx
_
0
=

2
n
a
2
n
,
_

0
(u
)
2
dx = u()u
() u(0)u
(0)
_

0
uu
dx =
_

0
uu
dx
=
_

0
_
m
a
m
m
__
n
a
n
n
_
dx
=

n
n
a
2
n
_

0
sin
2
nx dx =

2
n
a
2
n
.
Since
n
= n
2
, n = 1, 2, . . .
n
1, so
20
_

0
u
2
dx =

2
n
a
2
n

2
n
a
2
n
=
_

0
(u
)
2
dx.
b) We have
u
+ q(x)u u = 0,
uu
+q(x)u
2
u
2
= 0,
_

0
uu
dx +
_

0
q(x)u
2
dx
_

0
u
2
dx = 0,
uu
0
+
_

0
(u
)
2
dx +
_

0
q(x)u
2
dx
_

0
u
2
dx = 0,
_

0
(u
)
2
dx +
_

0
q(x)u
2
dx =
_

0
u
2
dx.
Since q(x) > 1, and using result from part (a),
0
..
(a)
_

0
(u
)
2
dx
_

0
u
2
dx <
..
q>1
_

0
(u
)
2
dx +
_

0
q(x)u
2
dx =
_

0
u
2
dx.
Since
_
0
u
2
dx 0, we have > 0.
20
See similar Poincare Inequality PDE problem.
Problem (F02, #5; F89, #6). a) Suppose that u is a continuously dierentiable
function on [0, 1] with u(0) = 0. Starting with u(x) =
_
x
0
u
(t) dt, prove the (sharp)

estimate
max
[0,1]
[u(x)[
2
_
1
0
[u
(t)[
2
dt. (5.23)
b) For any function p dene p
(x) = minp(x), 0.
21
Using the inequality (5.23), if
p is continuous on [0, 2], show that all eigenvalues of
Lu = u
+pu on [0, 2]
with u(0) = u(2) = 0 are strictly positive if
_
2
0
p
(t) dt < 1.
Proof. a) By the Fundamental Theorem of Calculus,
_
x
0
u
(t) dt = u(x) u(0) = u(x),

max
[0,1]
[u(x)[ =
_
1
0
u
(t) dt

_
1
0
[u
(t)[ dt [[1[[
L
2
_
_
1
0
[u
(t)[
2
dt
_1
2
=
_
_
1
0
[u
(t)[
2
dt
_1
2
,
max
[0,1]
[u(x)[
2
_
1
0
[u
(t)[
2
dt.
b) We have
u
+pu = u,
_
2
0
uu
dt +
_
2
0
pu
2
dt =
_
2
0
u
2
dt,
uu
[
2
0
. .
=0
+
_
2
0
[u
[
2
dt +
_
2
0
pu
2
dt =
_
2
0
u
2
dt.
If we dene p
+
(x) = maxp(x), 0 and p
(x) = minp(x), 0, then p = p

+
p
.
21
Note that p+ and p are dened by
p+(x) =
_
p(x) for p(x) 0
0 for p(x) < 0,
p(x) =
_
0 for p(x) 0
|p(x)| for p(x) < 0.
_
2
0
[u
[
2
dt +
_
2
0
p
+
u
2
dt
_
2
0
p
u
2
dt =
_
2
0
u
2
dt,
_
2
0
[u
[
2
dt
_
2
0
p
u
2
dt
_
2
0
u
2
dt,
max
[0,2]
[u[
2
_
2
0
p
u
2
dt
_
2
0
u
2
dt,
max
[0,2]
[u[
2
max
[0,2]
[u[
2
_
2
0
p
dt
_
2
0
u
2
dt,
max
[0,2]
[u[
2
_
1
_
2
0
p
dt
. .
<1
_

_
2
0
u
2
dt,
c
2
max
[0,2]
[u[
2

_
2
0
u
2
dt.
Thus, > 0.
Problem (F95, #6). Dene
Ly(x) = y
(x) + q(x)y(x) on (0, a).

Denote q
(x) = min(q(x), 0). We seek conditions on q
(x) so that L will be nonnegative

denite on C
0
(0, a), i.e.,
(L, ) =
_
a
0
(x) L(x) dx 0 C
0
(0, a). (5.24)
Find optimal conditions on q
(x) so that (5.24) holds.

Can q
(x) be unbounded and (5.24) still hold?

Proof. Dene q
+
= max(q(x), 0). We have
_
a
0
(x) L(x) dx =
_
a
0
(
+ q) dx =
_
a
0
(
+ q
2
) dx
=
[
a
0
. .
=0
+
_
a
0
(
)
2
+ q
2
dx =
_
a
0
(
)
2
dx +
_
a
0
q
2
dx

_
a
_
2
_
a
0
2
dx +
_
a
0
q
2
dx
_
a
_
2
_
a
0
2
dx +
_
a
0
q
2
dx
=
_
a
0
__
a
_
2
+q
2
dx
..
need
0.
Thus, if (
a
)
2
+ q
0, L will be nonnegative denite on C
0
(0, a).
Proof of :
Use eigenvalues of the Laplacian for
+ = 0, (0) = (a) = 0.
Then
n
= sin(
n
a
)x,
n
= (
n
a
)
2
, n = 1, 2, . . .. We have
_
a
0
2
dx =
_
a
0
_
m
a
m
m
__
n
a
n
n
_
dx =

n
a
2
n
_
a
0
sin
2
_
nx
a
_
dx,
_
a
0
(
)
2
dx =
[
a
0
_
a
0
dx =
_
a
0
dx
=
_
a
0
_
m
a
m
m
__
n
a
n
n
_
dx
=

n
n
a
2
n
_
a
0
sin
2
_
nx
a
_
dx.
_
a
_
2
_
a
0
2
dx =
_
a
_
2
n
a
2
n
_
a
0
sin
2
_
nx
a
_
dx
n
a
2
n
_
a
0
sin
2
_
nx
a
_
dx
=
_
a
0
(
)
2
dx.
_
a
0
(
)
2
dx
_
a
_
2
_
a
0
2
dx.
Problem (W04, #4). Consider boundary value problem on [0, ]:
y
(x) +p(x)y(x) = f(x), 0 < x < ,

y(0) = 0, y
() = 0.
Find the smallest
0
such that the boundary value problem has a unique solution
whenever p(x) >
0
for all x. Justify your answer.
Proof. Suppose y
1
and y
2
are two solutions of the problem. Let w = y
1
y
2
. Then
w
+ pw = 0, 0 < x < ,
w(0) = 0, w
() = 0.
Multiply by w and integrate
_
1
0
ww
dx +
_

0
pw
2
dx = 0,
ww
0
. .
=0
+
_

0
(w
)
2
dx +
_

0
pw
2
dx = 0,
_

0
(w
)
2
dx +
_

0
pw
2
dx = 0.
We will derive the Poincare inequality for this boundary value problem.
Use eigenvalues of the Laplacian for w
+w = 0, w(0) = w
() = 0.
Expand w in eigenfunctions: w =

n
a
n
n
. Then
n
(x) = a
n
cos
n
x+b
n
sin
n
x.
Boundary conditions give:
n
=
_
n +
1
2
_
2
,
n
(x) = sin
_
n +
1
2
_
x, n = 0, 1, 2, . . . . Then,
_

0
w
2
dx =
_

0
_
m
a
m
m
__
n
a
n
n
_
dx =

n
a
2
n
_

0
2
n
(x) dx,
_

0
(w
)
2
dx = ww
0

_

0
ww
dx =
_

0
ww
dx
=
_

0
_
m
a
m
m
__
n
a
n
n
_
dx =

n
n
a
2
n
_

0
2
n
dx.
Thus, the Poincare inequality is:
1
4
_

0
w
2
dx =
1
4
n
a
2
n
_

0
2
n
dx
n
a
2
n
_

0
2
n
dx =
_

0
(w
)
2
dx.
Thus, from :
0 =
_

0
(w
)
2
dx +
_

0
pw
2
dx
1
4
_

0
w
2
dx +
_

0
pw
2
dx =
_

0
_
1
4
+p
_
w
2
dx.
If
1
4
+ p(x) > 0, (p(x) >
1
4
), x, then w 0, and we obtain uniqueness.
Problem (F97, #5). a) Prove that all eigenvalues of the Sturm-Liouville problem
d
dx
_
p(x)
du
dx
_
+u(x) = 0, 0 < x < a,
u(0) = 0,
du(a)
dx
+hu(a) = 0,
are positive. Here h > 0, p(x) > 0 and continuous on [0, a].
b) Show that the same is true when h < 0 and [h[ is suciently small.
Proof. a) Let be an eigenfunction. We have
(p
+ = 0. (5.25)
Multiply (5.25) by and integrate from 0 to a,
_
a
0
_
(p
+
2
_
dx = 0.
Since
_
a
0

2
dx > 0, we can solve for :
=

_
a
0
(p
dx
_
a
0

2
dx
.
Integrating by parts and plugging in the boundary conditions give
=
p
[
a
0
+
_
a
0
p(
)
2
dx
_
a
0

2
dx
=
hp(a)
2
(a) +
_
a
0
p(x)(
(x))
2
dx
_
a
0

2
(x) dx
0.
To show that > 0, assume = 0. Then the ODE becomes
(pu
= 0 p(x) u
(x) = C, a constant.
Then
p(a) u
(a) = hp(a) u(a) = C

Wrong assumption follows: u = 0.
b) h < 0. For [h[ is suciently small, i.e.
[hp(a)
2
(a)[ <
_
a
0
p(x)(
(x))
2
dx,
we have
=
hp(a)
2
(a) +
_
a
0
p(x)(
(x))
2
dx
_
a
0

2
(x) dx
> 0.
Problem (S93, #7). a) Show that the general solution of
1
x
d
dx
_
x
df
dx
_
= f, (5.26)
where is a constant, is a linear combination of f
1
and f
2
, where
f
1
= O(1), f
2
= O(lnx), x 0.
Proof. a) We use the method of dominant balance. We have
1
x
(xf
= f,
1
x
(xf
+f
) = f,
f
+
1
x
f
= f,
xf
+f
= xf.
As x 0, f
(x) 0, i.e. f(x) C. (Incomplete)

b) Consider the eigenvalue problem posed by (5.26) and the conditions
f(0) = O(1), f(1) = 0. (5.27)
Assuming that the spectrum of is discrete, show that the eigenfunctions belonging to
dierent are orthogonal:
_
1
0
xf
i
f
j
dx =
_
1
0
x
df
i
dx
df
j
dx
dx = 0,
i
,=
j
,
and that all eigenvalues are positive.
Proof. Rewrite the equation as
1
x
(xf
+f = 0.
Let
m
,
n
, be the eigenvalues and f
m
, f
n
be the corresponding eigenfunctions. We
have
1
x
(xf
m
)
+
m
f
m
= 0, (5.28)
1
x
(xf
n
)
+
n
f
n
= 0. (5.29)
Multiply (5.28) by f
n
and (5.29) by f
m
f
n
1
x
(xf
m
)
+
m
f
n
f
m
= 0,
f
m
1
x
(xf
n
)
+
n
f
m
f
n
= 0.
(
m

n
)f
m
f
n
= f
m
1
x
(xf
n
)
f
n
1
x
(xf
m
)
,
(
m

n
)xf
m
f
n
= f
m
(xf
n
)
f
n
(xf
m
)
= [x(f
m
f
n
f
n
f
m
)]
.
Integrating over (0, 1) gives
(
m
n
)
_
1
0
xf
m
f
n
dx = [x(f
m
f
n
f
n
f
m
)]
1
0
= 1 (f
m
f
n
f
n
f
m
)(1) 0 (f
m
f
n
f
n
f
m
)(0) = 0,
since f
m
(1) = f
n
(1) = 0. Since
n
,=
m
, f
n
(x) and f
m
(x) are orthogonal on [0, 1].
To show that f
m
and f
n
are orthogonal with respect to x, consider
_
1
0
xf
m
f
n
dx = xf
m
f
n
[
1
0
_
1
0
(xf
m
)
f
n
dx
= 1 f
m
(1)f
n
(1) 0 f
m
(0)f
n
(0)
_
1
0
(xf
m
)
f
n
dx
=
_
1
0
(xf
m
)
f
n
dx = =
m
_
1
0
xf
m
f
n
dx = = 0.
We now show that eigenvalues are positive. We have
1
x
(xf
+f = 0,
(xf
+ xf = 0.
Multiplying by f and integrating, we get
_
1
0
f(xf
dx +
_
1
0
xf
2
dx = 0,
fxf
[
1
0
. .
=0
_
1
0
xf
2
dx +
_
1
0
xf
2
dx = 0,
=
_
1
0
xf
2
dx
_
1
0
xf
2
dx
0.
The equality holds only if f
0, which means f = C. Since f(1) = 0, then f 0,

which is not an eigenfunction. Thus, > 0.
c) Let f(x) be any function that can be expanded as a linear combination of eigenfunc-
tions of (5.26) and (5.27). Establish the Rayleigh-Ritz principle, namely that
F(f) =
_
1
0
x(f
)
2
dx
_
1
0
xf
2
dx
is an upper bound on the smallest eigenvalue.
Proof. Let f(x) =

a
n
f
n
, where f
n
s are eigenfunctions. Then,
F(f) =
_
1
0
x(f
)
2
dx
_
1
0
xf
2
dx
=
_
1
0
x(
a
n
f
n
)
2
dx
_
1
0
x(
a
n
f
n
)
2
dx
(by orthogonality)
=
_
1
0
x
a
2
n
f
2
n
dx
_
1
0
x
a
2
n
f
2
n
dx
=
a
2
n
_
1
0
xf
2
n
dx
a
2
n
_
1
0
xf
2
n
dx
(by )
=
a
2
n
n
_
1
0
xf
2
n
dx
a
2
n
_
1
0
xf
2
n
dx
>
min
a
2
n
_
1
0
xf
2
n
dx
a
2
n
_
1
0
xf
2
n
dx
=
min
.
Thus,
min
< F(f), i.e. F(f) is an upper bound on
min
.
d) The Bessel function J
0
(r) is O(1) at r = 0 and obeys
1
r
d
dr
_
r
dJ
0
dr
_
= J
0
.
Obtain an upper bound for the smallest positive zero of J
0
.
Problem (F90, #8). Consider the dierential equation
u
xx
+ (1 +x
2
)u = u,
u(0) = u(a) = 0.
a) Find a variational characterization for the eigenvalues
i
, i = 1, 2, . . ..
b) Show that the eigenvalues are all positive, i.e.
i
> 0.
c) Consider the problem for two dierent values of a: a = a
1
and a = a
2
with a
1
<
a
2
. Show that the eigenvalues
1
(a
1
) for a = a
1
is larger than (or equal to) the rst
eigenvalues
1
(a
2
) for a
2
, i.e.
1
(a
1
)
1
(a
2
) for a
1
< a
2
.
d) Is this still true for i > 1, i.e. is
i
(a
1
)
i
(a
2
) for a
1
< a
2
?
Proof. a) We have
u
+ (1 + x
2
)u = u,
_
a
0
uu
dx +
_
a
0
(1 + x
2
)u
2
dx =
_
a
0
u
2
dx,
uu
[
a
0
. .
=0
+
_
a
0
(u
)
2
dx +
_
a
0
(1 + x
2
)u
2
dx =
_
a
0
u
2
dx,
=
_
a
0
_
(u
)
2
+ (1 + x
2
)u
2
_
dx
_
a
0
u
2
dx
.
b) =
_
a
0
_
(u
)
2
+ (1 + x
2
)u
2
_
dx
_
a
0
u
2
dx
> 0, if u not identically 0.
c)
1
(a
1
) = min
[0,a
1
]
_
a
1
0
_
(u
)
2
+ (1 +x
2
)u
2
_
dx
_
a
1
0
u
2
dx
,
1
(a
2
) = min
[0,a
2
]
_
a
2
0
_
(u
)
2
+ (1 +x
2
)u
2
_
dx
_
a
2
0
u
2
dx
.
The minimum value in a small interval is greater then or equal to the minimum value
in the larger interval. Thus,
1
(a
1
)
1
(a
2
) for a
1
< a
2
.
We may also think of this as follows: We can always make a 0 extension of u from a
1
to
a
2
. Then, we can observe that the minimum of for such extended functions would
be greater.
d)
6 Variational (V) and Minimization (M) Formulations
Consider
(D)
_
u
(x) = f(x) for 0 < x < 1,

u(0) = u(1) = 0,
(V) Find u V, s.t. a(u, v) = L(v) v V,
(M) Find u V, s.t. F(u) F(v) v V, (F(u) = min
vV
F(v)).
V = v : v C
0
[0, 1], v
piecewise continous and bounded on [0, 1], and v(0) = v(1) =

0.
F(v) =
1
2
a(v, v) L(v)
(D) (V) (M)
(D) (V)
Multiply the equation by v V , and integrate over (0, 1):
u
= f(x),
_
1
0
u
v dx =
_
1
0
fv dx,
u
v[
1
0
. .
=0
+
_
1
0
u
dx =
_
1
0
fv dx,
_
1
0
u
dx =
_
1
0
fv dx,
a(u, v) = L(v) v V.
(V) (M)
We have a(u, v) = L(v), v V . Suppose v = u + w, w V . We have
F(v) = F(u +w) =
1
2
a(u +w, u +w) L(u +w)
=
1
2
a(u, u) +a(u, w) +
1
2
a(w, w) L(u) L(w)
=
1
2
a(u, u) L(u)
. .
=F(u)
+
1
2
a(w, w) +a(u, w) L(w)
. .
=0, by
F(u).
(M) (V)
We have F(u) F(u +v), for any v V , since u +v V . Thus, the function
g() F(u + v) =
1
2
a(u +v, u + v) L(u + v)
=
1
2
a(u, u) + a(u, v) +

2
2
a(v, v) L(u) L(v),
has a minimum at = 0 and hence g
(0) = 0. We have
g
() = a(u, v) +a(v, v) L(v),

0 = g
(0) = a(u, v) L(v),

a(u, v) = L(v).
(V) (D)
We have
_
1
0
u
dx
_
1
0
fv dx = 0 v V.
Assume u
exists and is continuous, then

u
v[
1
0
..
=0
_
1
0
u
v dx
_
1
0
fv dx = 0,
_
1
0
(u
+f)v dx = 0 v V.
Since u
+ f is continuous, then
(u
+f)(x) = 0 0 < x < 1.

We can show that (V) is uniquely determined if a(u, v) = (u
, v
) =
_
1
0
u
dx.
Suppose u
1
, u
2
V and
(u
1
, v
) = L(v) v V,
(u
2
, v
) = L(v) v V.
Subtracting these equations gives
(u
1
u
2
, v
) = 0 v V.
Choose v = u
1
u
2
V . We get
(u
1
u
2
, u
1
u
2
) = 0,
_
1
0
(u
1
u
2
)
2
dx =
_
1
0
(u
1
u
2
)
2
dx = 0,
which shows that
(u
1
u
2
)
(x) = 0 u
1
u
2
= constant.
The boundary conditions u
1
(0) = u
2
(0) = 0 give u
1
(x) = u
2
(x), x [0, 1].
Problem (F91, #4). Consider a boundary value problem in a bounded plane domain
:
_

2
u
x
2
+

2
u
y
2
= f(x, y) in ,
u
n
+a(s)u = 0 on ,
(6.1)
where a(s) is a smooth function on .
a) Find the variational formulation of this problem, i.e. nd a functional F(v) dened
on smooth functions in the such that the Euler-Lagrange equation for this func-
tional is equivalent to (6.1).
Proof. a) (D) (M)
We will proceed as follows: (D) (V) (M). We have
_
u = f in ,
u
n
+ a(s)u = 0 on .
(D) (V)
Multiply the equation by v V , and integrate over :
u = f,
_
uv dx =
_
fv dx,
_
u
n
v ds
_
u v dx =
_
fv dx,
a(s)uv ds
_
u v dx =
_
fv dx,
_
u v dx +
_
a(s)uv ds
. .
a(u,v)
=
_
fv dx
. .
L(v)
.
(V) (M)
a(u, v) = L(v),
a(u, v) =
_
u v dx +
_
a(s)uv ds,
L(v) =
_
fv dx,
F(v) =
1
2
a(v, v) L(v).
F(v) =
1
2
_
[v[
2
dx +
1
2
_
a(s)v
2
ds +
_
fv dx.
We show that F(v), dened as , minimizes the functional.
We have a(u, v) = L(v), v V . Suppose v = u + w, w V . We have
F(v) = F(u +w) =
1
2
a(u +w, u +w) L(u +w)
=
1
2
a(u, u) +a(u, w) +
1
2
a(w, w) L(u) L(w)
=
1
2
a(u, u) L(u)
. .
=F(u)
+
1
2
a(w, w) +a(u, w) L(w)
. .
=0, by
F(u).
b) Prove that if a(s) > 0, then the solution of (6.1) is unique in the class of smooth
functions in .
Proof. Let u
1
, u
2
be two solutions of (6.1), and set w = u
1
u
2
. Then
a(u
1
, v) = L(v),
a(u
2
, v) = L(v),
a(w, v) = 0.
Let v = w V . Then,
a(w, w) =
_
[w[
2
dx +
_
a(s)w
2
ds = 0.
Since a(s) > 0, w 0.
We can also begin from considering
_
w = 0 in ,
w
n
+ a(s)w = 0 on .
Multiplying the equation by w and integrating, we obtain
_
wwdx = 0,
_
w
w
n
ds
_
[w[
2
dx = 0,
a(s)w
2
ds
_
[w[
2
dx = 0,
_
a(s)w
2
ds +
_
[w[
2
dx = 0.
Since a(s) > 0, w 0.
Problem (W04, #2). Let C
2
() be the space of all twice continuously dierentiable
functions in the bounded smooth closed domain R
2
. Let u
0
(x, y) be the function
that minimizes the functional
D(u) =
_ _
_
_
u(x, y)
x
_
2
+
_
u(x, y)
y
_
2
+ f(x, y)u(x, y)
_
dxdy +
_
a(s) u
2
(x(s), y(s)) ds,
where f(x, y) and a(s) are given continuous functions.
Find the dierential equation and the boundary condition that u
0
satises.
Proof. (M) (D)
We will proceed as follows: (M) (V) (D). We have
F(v) =
_
([v[
2
+fv) dx +
_
a(s) v
2
ds,
F(v) =
1
2
a(v, v) L(v),
a(u, v) = 2
_
u v dx + 2
_
a(s)uv ds,
L(v) =
_
fv dx.
(M) (V) Since u
0
minimizes F(v) we have
F(u
0
) F(v), v V.
Thus, the function
g() F(u
0
+v) =
1
2
a(u
0
+v, u
0
+ v) L(u
0
+ v)
=
1
2
a(u
0
, u
0
) +a(u
0
, v) +

2
2
a(v, v) L(u
0
) L(v),
has a minimum at = 0 and hence g
(0) = 0. We have
g
() = a(u
0
, v) +a(v, v) L(v),
0 = g
(0) = a(u
0
, v) L(v),
a(u
0
, v) = L(v).
2
_
u
0
v dx + 2
_
a(s)u
0
v ds =
_
fv dx.
(V) (D)
2
_
u
0
v dx + 2
_
a(s)u
0
v ds =
_
fv dx,
2
_
u
0
n
v ds 2
_
u
0
v dx + 2
_
a(s)u
0
v ds =
_
fv dx,
_
(2u
0
+f)v dx + 2
_
_
u
0
n
+ a(s)u
0
_
v ds = 0.
If
u
0
n
+a(s)u
0
= 0, we have
_
(2u
0
+f)v dx = 0 v V.
Since 2u
0
+f is continuous, then 2u
0
+ f = 0.
_
2u
0
+ f = 0, x ,
u
0
n
+ a(s)u
0
= 0, x .
See the preferred solution in the Euler-Lagrange Equations section.
7 Euler-Lagrange Equations
Consider the problem of determining a C
1
function u(x) for which the integral
E =
_
J(x, u, u) dx
takes on a minimum value.
Suppose u(x) is the actual minimizing function, and choose any C
1
function (x).
Since u is the minimizer
E(u +) E(u), .
E(u +) has a minimum at = 0. Thus,
dE
d
(u + )
=0
= 0.
7.1 Rudin-Osher-Fatemi
E =
_
[u[ +(u f)
2
dx.
dE
d
(u + )
=0
=
d
d
_
[(u +)[ + (u + f)
2
dx
=0
=
_
(u + )
[(u + )[
+ 2(u + f) dx
=0
=
_
u
[u[
+ 2(u f) dx
=
_
u
[u[
nds
_
_
u
[u[
_
dx +
_
2(u f) dx
=
_
u
[u[
nds
_
_
u
[u[
_
2(u f)
_
dx = 0.
Choose C
1
c
(). The Euler-Lagrange equations
22
are
_
u
[u[
_
2(u f) = 0 on ,
22
Hildebrands (p.124-128) denition of Euler-Lagrange equations in one dimension:
_
x
2
x
1
_
J
y

d
dx
_
J
y
_
dx +
_
J
y
(x)
_
x
2
x
1
= 0.
d
dx
_
J
y
_
=
J
y
.
_
J
y
_
x=x
1
= 0,
_
J
y
_
x=x
2
= 0.
In n dimensions:
x (pJ) = uJ on ,
pJ n = 0 on ,
where p = u = (ux, uy).
u n = 0 on .
7.1.1 Gradient Descent
If we want to nd a local minimum of a function f in R
1
, we have
dx
dt
=
df
dx
.
To minimize the energy E (in R
2
), we would have
du
dt
=
dE(u)
du
.
Also, consider
E =
_
[u[ +(u f)
2
dx.
We want E(u(x, t)) to decrease, that is,
d
dt
E(u(x, t)) 0, for all t.
Assume u n = 0 on . We have
d
dt
E(u(x, t)) =
d
dt
_
[u[ +(u f)
2
dx
=
_
u u
t
[u[
+ 2(u f) u
t
dx
=
_
_
u
[u[
_
u
t
+ 2(u f) u
t
dx
=
_
u
t
_
_
u
[u[
_
+ 2(u f)
. .
_
dx 0.
To ensure that holds, we need to choose u
t
to be negative of , or
u
t
=
_
u
[u[
_
2(u f).
7.2 Chan-Vese
F
CV
=
_
()[[ dx +
_
(1 H()) dx
+
1
_
[u
0
c
1
[
2
(1 H()) dx +
2
_
[u
0
c
2
[
2
H() dx.
dF
CV
d
( +)
=0
=
d
d
_
( + )[(+)[ dx +
d
d
_
(1 H( +)) dx
+
1
d
d
_
(u
0
c
1
)
2
(1 H( + )) dx +
2
d
d
_
(u
0
c
2
)
2
H( + ) dx
=0
=
_
( +) [( +)[ + ( +)
( + )
[(+ )[

_
dx
+
_
( +) dx
+
1
_
(u
0
c
1
)
2
(H
( + )) dx
+
2
_
(u
0
c
2
)
2
H
( + ) dx
=0
=
_
() [[ +()

[[

_
dx

_
() dx

1
_
(u
0
c
1
)
2
H
() dx
+
2
_
(u
0
c
2
)
2
H
() dx
=
_
() [[ dx +
_
()
[[
n
ds

_
()

[[
dx
_
()
x
_

[[
_
dx

_
() dx

1
_
(u
0
c
1
)
2
() dx
+
2
_
(u
0
c
2
)
2
() dx
=
_
()
[[
n
ds
+
_
()
_
_

[[
_

1
(u
0
c
1
)
2
+
2
(u
0
c
2
)
2
_
dx = 0.
Choose C
1
c
(). The Euler-Lagrange equations are
()
_
_

[[
_
+ +
1
(u
0
c
1
)
2

2
(u
0
c
2
)
2
_
= 0 on ,
()
[[
n
= 0 on .
7.3 Problems
The problem below was solved in the previous section. However, the approach below
is preferable.
Problem (W04, #2). Let C
2
() be the space of all twice continuously dierentiable
functions in the bounded smooth closed domain R
2
. Let u
0
(x, y) be the function
that minimizes the functional
D(u) =
_ _
_
_
u(x, y)
x
_
2
+
_
u(x, y)
y
_
2
+ f(x, y)u(x, y)
_
dxdy +
_
a(s) u
2
(x(s), y(s)) ds,
where f(x, y) and a(s) are given continuous functions.
Find the dierential equation and the boundary condition that u
0
satises.
Proof. Suppose u(x) is the actual minimizing function, and choose any C
1
function
(x).
F(u +) F(u), .
F(u + ) has a minimum at = 0. Thus,
dF
d
(u +)
=0
= 0.
F(u) =
_
([u[
2
+ fu) dx +
_
a(s) u
2
ds,
dF
d
(u +)
=0
=
d
d
_
_
[(u + )[
2
+ f (u +)
_
dx +
d
d
_
a(s) (u + )
2
ds
=0
=
_
_
2 (u +) + f
_
dx +
_
2 a(s) (u + ) ds
=0
=
_
_
2 u +f
_
dx +
_
2 a(s) u ds
=
_
2
u
n
ds
_
_
2 u f
_
dx +
_
2 a(s) u ds
= 2
_
_
u
n
+a(s)u
_
ds
_
_
2 u f
_
dx = 0.
The Euler-Lagrange equations are
_
2u = f, x ,
u
n
+a(s)u = 0, x .
Problem (F92, #7). Let a
1
and a
2
be positive constants with a
1
,= a
2
and dene
a(x) =
_
a
1
for 0 < x <
1
2
a
2
for
1
2
< x < 1
and let f(x) be a smooth function. Consider the functional
F(u) =
_
1
0
a(x)u
2
x
dx
_
1
0
f(x)u(x) dx
in which u is continuous on [0, 1], twice dierentiable on [0,
1
2
] and [
1
2
, 1], and has a
possible jump discontinuity in u
x
at x =
1
2
. Find the Euler-Lagrange equation for
u(x) that minimizes the functional F(u). In addition nd the boundary conditions on
u at x = 0, x =
1
2
and x = 1.
Proof. Suppose u(x) is the actual minimizing function, and choose any C
1
function
(x).
F(u +) F(u), .
dF
d
(u +)
=0
= 0.
F(u) =
_ 1
2
0
a
1
u
2
x
dx +
_
1
1
2
a
2
u
2
x
dx
_
1
0
f(x)u(x) dx
dF
d
(u +)
=0
=
d
d
_ 1
2
0
a
1
(u
x
+
x
)
2
dx +
d
d
_
1
1
2
a
2
(u
x
+
x
)
2
dx
d
d
_
1
0
f(x)(u + ) dx
=0
=
_ 1
2
0
2a
1
(u
x
+
x
)
x
dx +
_
1
1
2
2a
2
(u
x
+
x
)
x
dx
_
1
0
f(x) dx
=0
=
_ 1
2
0
2a
1
u
x
x
dx +
_
1
1
2
2a
2
u
x
x
dx
_
1
0
f(x) dx
= 2a
1
u
x
1
2
0
_ 1
2
0
2a
1
u
xx
dx + 2a
2
u
x
1
1
2
_
1
1
2
2a
2
u
xx
dx
_
1
0
f(x) dx
= 2a
1
u
x
1
2
0
+ 2a
2
u
x
1
1
2
_
1
0
2a(x)u
xx
dx
_
1
0
f(x) dx.
Thus,
a
1
u
x
_
1
2
_
_
1
2
_
a
1
u
x
(0)(0) + a
2
u
x
(1)(1) a
2
u
x
(
1
2
)
_
1
2
_
= 0.
_
1
0
_
2a(x)u
xx
+ f(x)
_
dx = 0.
2a(x)u
xx
+ f(x) = 0,
u
x
(0) = 0,
u
x
(1) = 0,
a
1
u
x
_
1
2

_
= a
2
u
x
(
1
2
+).
The process of nding Euler-Lagrange equations (given the minimization functional) is
equivalent to (D) (V) (M).
Problem (F00, #4). Consider the following functional
F(v) =
_ _ _
_
3
j,k=1
_
v
j
x
k
_
2
+
_
3
j=1
v
2
j
(x) 1
_
2
_
dx,
where x = (x
1
, x
2
, x
3
) R
3
, v(x) = (v
1
(x), v
2
(x), v
3
(x)), R
3
bounded, and > 0
is a constant. Let u(x) = (u
1
(x), u
2
(x), u
3
(x)) be the minimizer of F(v) among all
smooth functions satisfying the Dirichlet condition, u
k
(x) =
k
(x), k = 1, 2, 3. Derive
the system of dierential equations that u(x) satises.
Proof. (M) (D)
Suppose u(x) is the actual minimizing function, and choose any C
1
function (x) =
(
1
(x),
2
(x),
3
(x)).
F(u +) F(u), .
dF
d
(u +)
=0
= 0.
dF
d
(u +)
=0
=
d
d
_
_
3
j,k=1
_
u
j
x
k
+
j
x
k
_
2
+
_
3
j=1
(u
j
+
j
)
2
1
_
2
_
dx
=0
=
_
_
2
3
j,k=1
_
u
j
x
k
+
j
x
k
_
j
x
k
+ 2
_
3
j=1
(u
j
+
j
)
2
1
_
2
3
j=1
(u
j
+
j
)
j
_
dx
=0
=
_
_
2
3
j,k=1
_
u
j
x
k
_
j
x
k
+ 4
_
3
j=1
u
2
j
1
_
3
j=1
u
j
j
_
dx
=
_
_
2
_
u
1

1
+ u
2

2
+ u
3

3
_
+ 4
_
u
2
1
+u
2
2
+u
2
3
1
__
u
1
1
+ u
2
2
+u
3
3
_
_
dx
=
_
_
2
_
u
1
n

1
+
u
2
n

2
+
u
3
n

3
_
ds +
_
_
2
_
u
1
1
+ u
2
2
+ u
2
3
_
dx
+ 4
_
u
2
1
+u
2
2
+u
2
3
1
__
u
1
1
+ u
2
2
+u
3
3
_
_
dx = 0.
If we assume that u
2
1
+u
2
2
+ u
2
3
1 = 1, we have
dF
d
(u +)
=0
=
_
_
2
_
u
1
n

1
+
u
2
n

2
+
u
3
n

3
_
ds +
_
_
2
_
u
1
1
+ u
2
2
+ u
2
3
_
dx
+ 4
_
u
1
1
+u
2
2
+ u
3
3
_
_
dx = 0.
u
i
+ 2u
i
= 0, in
u
i
n
= 0, i = 1, 2, 3, on .
8 Integral Equations
Fredholm Equation: (x)y(x) = F(x) +
_
b
a
K(x, )y()d
Volterra Equation: (x)y(x) = F(x) +
_
x
a
K(x, )y()d
When 0, the equation is said to be an integral equation of the rst kind.
When 1, the equation is said to be an integral equation of the second kind.
d
dx
_
B(x)
A(x)
F(x, )d =
_
B
A
F(x, )
x
d +F(x, B(x))
dB
dx
F(x, A(x))
dA
dx
.
8.1 Relations Between Dierential and Integral Equations
Example 1. Consider the boundary-value problem
y
+y = 0,
y(0) = y(L) = 0.
After the rst integration over (0, x), we obtain
y
(x) =
_
x
0
y() d + C,
where C represents the unknown value of y
(0). A second integration over (0, x) gives

y(x) =
_
x
0
ds
..
v
_
s
0
y() d
. .
u
+Cx + D =
_
_
s
_
s
0
y() d
_
x
0
_
x
0
sy(s) ds
_
+Cx + D
=
_
x
_
x
0
y() d
_
x
0
y() d
_
+Cx +D =
_
x
0
(x )y() d + Cx + D.
y(0) = 0 gives D = 0. Since y(L) = 0, then
y(L) = 0 =
_
L
0
(L )y() d +CL,
C =

L
_
L
0
(L )y() d.
If the values of C and D are introduced into , this relation takes the form
y(x) =
_
x
0
(x )y() d +
x
L
_
L
0
(L )y() d
=
_
x
0
(x )y() d +
_
x
0
x
L
(L )y() d +
_
L
x
x
L
(L )y() d
=
_
x
0
L
(L )y() d +
_
L
x
x
L
(L )y() d.
Thus,
y(x) =
_
L
0
K(x, )y() d
where
K(x, ) =
_
L
(L ), < x
x
L
(L ), > x
Note, K(x, ) is symmetric: K(x, ) = K(, x).
The kernel K is continuous at x = .
Example 2. Consider the boundary-value problem
y
+Ay
+ By = 0,
y(0) = y(1) = 0.
Integrating over (0, x) twice, we obtain
y
(x) = Ay(x) B
_
x
0
y() d + C,
y(x) = A
_
x
0
y() d B
_
x
0
ds
..
v
_
s
0
y() d
. .
u
+Cx + D
= A
_
x
0
y() d B
_
_
s
_
s
0
y() d
_
x
0
_
x
0
sy(s) ds
_
+ Cx +D
= A
_
x
0
y() d B
_
x
_
x
0
y() d
_
x
0
y() d
_
+Cx +D
=
_
x
0
_
A B(x )
_
y() d + Cx + D.
y(0) = 0 gives D = 0. Since y(1) = 0, then
y(1) = 0 =
_
1
0
_
A B(1 )
_
y() d + C,
C =
_
1
0
_
A+B(1 )
_
y() d.
If the values of C and D are introduced into , this relation takes the form
y(x) =
_
x
0
_
AB(x )
_
y() d + x
_
1
0
_
A +B(1 )
_
y() d
=
_
x
0
_
AB(x )
_
y() d +
_
x
0
_
Ax + Bx(1 )
_
y() d +
_
1
x
_
Ax +Bx(1 )
_
y() d
=
_
x
0
_
A(x 1) +B(1 x)
_
y() d +
_
1
x
_
Ax + Bx(1 )
_
y() d
Thus,
y(x) =
_
1
0
K(x, )y() d
where
K(x, ) =
_
A(x 1) + B(1 x), < x
Ax +Bx(1 ), > x
Note, K(x, ) is not symmetric: K(x, ) ,= K(, x), unless A = 0.
The kernel K is not continuous at x = , since
lim
x
+
A(x1)+B(1x) = A(1)+B(1) ,= A+B(1) = lim
x
Ax+Bx(1).
8.2 Greens Function
Given the dierential operator
L =
d
dx
_
p
d
dx
_
+ q,
consider the dierential equation
Ly +F(x) = 0, a x b
c
1
y(a) + c
2
y
(a) = 0, c
3
y(b) +c
4
y
(b) = 0
where F may also depend upon x indirectly through y(x), F(x) = F(x, y(x)).
We construct a Greens function G which, for a given number , is given by
u(x) when x < and by v(x) when x > , and which has the following four properties:
The functions u and v satisfy the equation LG = 0 in their intervals of denition;
that is Lu = 0 when x < , and Lv = 0 when x > .
u satises the boundary condition at x = a, and v that at x = b.
G is continuous at x = ; that is u() = v().
v
() u
() = 1/p().
When G(x, ) exists, the original formulation of the problem can be transformed to
y(x) =
_
b
a
G(x, )F()d. |
Thus, conditions and imply
G =
_
u(x), x < ,
v(x), x > .
(8.1)
where u and v satisfty respective boundary conditions, and conditions and deter-
mine additional properties of u and v (i.e. constants in terms of ):
c
2
v() c
1
u() = 0, (8.2)
c
2
v
() c
1
u
() =
1
p()
. (8.3)
Example. Transform the problem
d
2
y
dx
2
+ y +y
2
= f(x),
y(0) = 0, y(1) = 0
to a nonlinear Fredholm integral equation in each of the two following ways. Use
a) Ly = y
.
b) Ly = y
+y.
Proof. a) We have
y
..
Ly
+y +y
2
f(x)
. .
F(x)
= 0
Ly = y
= 0 y = ax + b
u(x) = ax +b, v(x) = cx +d.
u(0) = 0 = b u(x) = ax.
v(1) = 0 = c + d v(x) = c(x 1).
Determine a and c in terms of :
u() = v(),
a = c( 1),
=
c
ca
.
v
() u
() = c a =
1
p()
= 1,
c = , a = 1 . Thus,
G =
_
u(x), x < ,
v(x), x > .
=
_
x(1 ), x < ,
(1 ), x > .
y(x) =
_
1
0
G(x, )F() d =
_
1
0
G(x, )f() d +
_
1
0
G(x, )
_
y() +y
2
()
d
b) We have
y
+y
. .
Ly
+y
2
f(x)
. .
F(x)
= 0
Ly = y
+y = 0 y = Acos x +B sinx
u(x) = a cos x +b sinx, v(x) = c cos x +d sinx.
u(0) = 0 = a u(x) = b sinx.
v(1) = 0 = c cos 1 +d sin1 v(x) = d(sinx
sin1
cos 1
cos x).
Determine b and d in terms of :
u() = v(),
b sin = d(sin
sin1
cos 1
cos ),
b = d(1
sin 1
cos 1
cos
sin
).
v
() u
() = d(
sin1
cos1
sin + cos b cos ) =
1
p()
= 1.
After some algebra,
u(x) =
sin(1) sinx
sin 1
,
v(x) =
sin(1x) sin
sin1
.
G =
_
sin(1) sinx
sin1
, x < ,
sin(1x) sin
sin1
, x > .
y(x) =
_
1
0
G(x, )F() d =
_
1
0
G(x, )f() d +
_
1
0
G(x, )y
2
() d
Problem (W02, #1). Consider the second order dierential operator L dened by
Ly =
d
2
y
dx
2
y.
Find the Greens function (= solution operator kernel) for the boundary value problem
Ly = f on 0 < x < 1, y(1) = y(0) = 0.
Proof. Ly = y
y = 0 y = Ae
x
+Be
x
u(x) = ae
x
+ be
x
, v(x) = ce
x
+ de
x
.
u(0) = 0 = a + b u(x) = a(e
x
e
x
).
v(1) = 0 = ce
1
+ de
1
v(x) = d(e
x
e
2x
).
Determine a and d in terms of :
u() = v(),
a(e
) = d(e
e
2
),
a = d
e
e
2
e
.
v
() u
() = d(e
+ e
2
) a(e
) =
1
p()
= 1.
Plugging in into , we get
d(e
+e
2
) d
e
e
2
e
(e
) = 1,
e
+ e
2
+
e
e
2
e
(e
+ e
) =
1
d
,
(e
+e
2
)
e
+
e
e
2
e
(e
+ e
) =
1
d
,
1 e
2
+e
22
e
2
e
+
1 +e
2
e
22
e
2
e
=
1
d
,
2 2e
2
e
=
1
d
,
d =
e
2(e
2
1)
.
a = d
e
e
2
e
=
e
2(e
2
1)

e
e
2
e
=
e
e
2
2(e
2
1)
.
G =
_
e
e
2
2(e
2
1)
(e
x
e
x
), x < ,
e
2(e
2
1)
(e
x
e
2x
), x > .
Example. Show that the Greens function G(x, ) associated with the expression
d
2
y
dx
2
y
over the innite interval (, ), subject to the requirement that y be bounded as
x , is of the form
G(x, ) =
1
2
e
|x|
.
Proof. Ly = y
y = 0 y = Ae
x
+Be
x
u(x) = ae
x
+ be
x
, v(x) = ce
x
+ de
x
.
Since y is bounded as x , a = 0 u(x) = be
x
.
Since y is bounded as x +, d = 0 v(x) = ce
x
.
Determine b and c in terms of :
u() = v(),
be
= ce
,
b = ce
2
.
v
() u
() = ce
be
=
1
p()
= 1.
c =
1be
=
1ce
2
e
=
1ce
= e
c,
c =
1
2
e
b =
1
2
e
. Thus,
G(x, ) =
_
be
x
, x <
ce
x
, x >
=
_
1
2
e
e
x
, x <
1
2
e
e
x
, x >
=
_
1
2
e
x
, x <
1
2
e
x
, x >
=
_
1
2
e
|x|
, x <
1
2
e
|x|
, x >
=
_
1
2
e
|x|
, x < ,
1
2
e
|x|
, x > .
G(x, ) =
1
2
e
|x|
Problem (W04, #7). For the two-point boundary value problem Lf = f
xx
f on
< x < with lim
x
f(x) = lim
x
f(x) = 0, the Greens function G(x, x
)
solves LG = (x x
) in which L acts on the variable x.

a) Show that G(x, x
) = G(x x
).
b) For each x
, show that
G(x, x
) =
_
a
e
x
for x < x
,
a
+
e
x
for x
< x,
in which a
are functions that depend only on x
.
c) Using (a), nd the x
dependence of a
.
d) Finish nding G(x, x
) by using the jump conditions to nd the remaining unknowns

in a
.
Proof. a) We have
Lf = f
xx
f,
LG = G(x, x
)
xx
G(x, x
) = (x x
),
??? G(x, x
) = G(x x
).
b, c, d) Lf = f
f = 0 y = Ae
x
+Be
x
u(x) = ae
x
+ be
x
, v(x) = ce
x
+ de
x
.
Since lim
x
f(x) = 0, a = 0 u(x) = be
x
.
Since lim
x+
f(x) = 0, d = 0 v(x) = ce
x
.
Determine b and c in terms of :
u() = v(),
be
= ce
,
b = ce
2
.
v
() u
() = ce
be
=
1
p()
= 1.
c =
1be
=
1ce
2
e
=
1ce
= e
c,
c =
1
2
e
b =
1
2
e
. Thus,
G(x, ) =
_
be
x
, x <
ce
x
, x >
=
_
1
2
e
e
x
, x <
1
2
e
e
x
, x >
=
_
1
2
e
x
, x <
1
2
e
x
, x >
=
_
1
2
e
|x|
, x <
1
2
e
|x|
, x >
=
_
1
2
e
|x|
, x < ,
1
2
e
|x|
, x > .
G(x, ) =
1
2
e
|x|
9 Miscellaneous
Problem (F98, #1). Determine such that the dierential equation
d
2
dx
2
+ = +x
2
, (9.1)
with (0) = 0 and () = 0 has a solution.
Proof. Solve the homogeneous equation
+ = 0. Subsitution = e
sx
gives
s
2
+ 1 = 0. Hence, s
1,2
= i and the superposition principle gives the family of
solutions:
h
(x) = Acos x + Bsinx.
Find a particular solution of the inhomogeneous equation
+ = +x
2
.
Try (x) = ax
2
+bx +c. Substitution into (9.1) gives
ax
2
+bx + 2a +c = +x
2
.
By equating coecients, a = 1, b = 0, c = 2. Thus,
p
(x) = x
2
+ 2.
Use the principle of the complementary function to form the family of solutions:
(x) =
h
(x) +
p
(x) = Acos x + Bsin x +x
2
+ 2.
(0) = 0 = A + 2,
() = 0 = A +
2
+ 2.
Thus, A =

2
2
, which gives = 2

2
2
.
Problem (S92, #5). Consider the initial value problem for the ODEs
y
= y y
3
, y
= y +y
3
, t 0,
with initial data
y(0) =
1
2
.
Investigate whether the solutions stay bounded for all times. If not compute the blow-
up time.
Proof. a) We solve the initial value problem.
dy
dt
= y y
3
= y(1 y
2
) = y(1 y)(1 +y),
dy
y(1 y)(1 + y)
= dt,
_
1
y
+
1
2
1
1 y

1
2
1
1 + y
_
dy = dt,
ln y
1
2
ln (1 y)
1
2
ln (1 + y) = t + c
1
,
ln y
1
2
ln ((1 y)(1 +y)) = t + c
1
,
ln y ln ((1 y)(1 + y))
1
2
= t + c
1
,
ln
_
y
((1 y)(1 +y))
1
2
_
= t + c
1
,
y
((1 y)(1 + y))
1
2
= c
2
e
t
,
y
(1 y
2
)
1
2
= c
2
e
t
.
Initial condition y(0) =
1
2
, we obtain c
2
=
1
3
. Thus
y
(1 y
2
)
1
2
=
1
3
e
t
,
y
2
1 y
2
=
1
3
e
2t
,
y =
1
3e
2t
+ 1
.
As t y 1. Thus, the solutions stay bounded for all times.
We can also observe from the image above that at y(0) =
1
2
,
dx
dt
> 0. Thus y 1 as
t .
b) We solve the initial value problem.
dy
dt
= y +y
3
,
y
3
y
= y
2
+ 1.
Let v = y
2
, then v
= 2y
3
y
. We have
1
2
v
v = 1 v
+ 2v = 2 v = ce
2t
1,
y
2
= v = ce
2t
1 y =
1
ce
2t
1
y =
1
5e
2t
1
.
The solution blows up at t =
1
2
ln 5.
Problem (S94, #4).
Suppose that
1
(t) and
2
(t) are any two solutions of the linear dierential equation
f
+ a
1
(t)f
+a
2
(t)f = 0. (9.2)
a) Show that
1
(t)
2
(t)
2
(t)
1
(t) = ce
_
t
a
1
(s) ds
for some constant c.
b) For any solution
1
(t), show that
(t) =
1
(t)
_
t
e
_
s
a
1
(r) dr
1
1
(s)
2
ds
is also a solution and is independent of
1
, on any interval in which
1
(t) ,= 0.
Proof. a) Suppose
1
and
2
are two solutions of (9.2). Then
1
+a
1
1
+a
2
1
= 0,
2
+a
1
2
+a
2
2
= 0.
1
[
2
+a
1
2
+a
2
2
]
2
[
1
+a
1
1
+ a
2
1
] = 0,
2

2
1
+ a
1
[
1
1
] = 0.
Let w =
1
1
. Then, w
=
1
2

2
1
. Thus,
w
+ a
1
(t)w = 0
w
w
= a
1
(t) w = c e
_
t
a
1
(s) ds
.
1
= c e
_
t
a
1
(s) ds
.
b) Let (t) =
1
(t)v(t) for some non-constant function v(t), which we will nd.
Since (t) is a solution of (9.2), we have
+ a
1
+ a
2
= 0,
(
1
v)
+a
1
(
1
v)
+a
2
1
v = 0,
1
v + 2
1
v
+
1
v
+ a
1
1
v +a
1
1
v
+a
2
1
v = 0,
1
v
+ [2
1
+a
1
1
]v
+ [
1
+ a
1
1
+ a
2
1
. .
=0
]v = 0,
1
v
+ [2
1
+a
1
1
]v
= 0,
v
=
2
1
+a
1
1
= 2
1
a
1
,
ln v
= 2 ln
1
_
t
a
1
(s) ds + c
1
,
v
= c
1
2
1
e
_
t
a
1
(s) ds
,
v = c
_
t
1
2
1
e
_
s
a
1
(r) dr
ds.
(t) =
1
(t)v(t) = c
1
(t)
_
t
1
1
(s)
2
e
_
s
a
1
(r) dr
ds.
(t) is a solution independent of
1
(t).
Problem (W03, #7). Under what conditions on g, continuous on [0, L], is there a
solution of
2
u
x
2
= g,
u(0) = u(L/3) = u(L) = 0?
Proof. We have
u
xx
= g(x),
u
x
=
_
x
0
g() d + C,
u(x) =
_
x
0
_

0
g(s) ds d + Cx + D.
0 = u(0) = D. Thus,
u(x) =
_
x
0
_

0
g(s) ds d + Cx.
0 = u(L) =
_
L
0
_

0
g(s) ds d + CL,
0 = u(L/3) =
_ L
3
0
_

0
g(s) ds d +
CL
3
.
and are the conditions on g.
10 Dominant Balance
Problem (F90, #4). Use the method of dominant balance to nd the asymptotic
behavior at t = for solutions of the equation
f
tt
+t
3
f
2
t
4f = 0.
Proof. Assume f = ct
n
as t , where need to nd n and c. Then
n(n 1)ct
n2
+n
2
c
2
t
3
t
2n2
4ct
n
= 0,
n(n 1)ct
n2
+ n
2
c
2
t
2n+1
4ct
n
= 0.
The 2
nd
and the 3
rd
terms are dominant. In order to satisfy the ODE for t , set
2n + 1 = n n = 1,
n
2
c
2
= 4c c
2
4c = 0 c = 4.
f 4t
1
, as t .
Problem (S91, #3). Find the large time behavior for solutions of the equation
d
2
dt
2
f +
d
dt
f + f
3
= 0
using the method of dominant balance.
Proof.
23
Assume f = ct
n
as t , where need to nd n and c. Then
n(n 1)ct
n2
+nct
n1
+ c
3
t
3n
= 0.
The 2
nd
and the 3
rd
terms are dominant. In order to satisfy the ODE for t , set
n 1 = 3n n =
1
2
,
nc +c
3
= 0
1
2
c + c
3
= 0 c =
1
2
.
f
1
2
t
1
2
, as t .
23
ChiuYens solutions show a dierent approach, but they are wrong.
11 Perturbation Theory
Problem (F89, #5a). Solve the following ODE for u(x) by perturbation theory
_
u
xx
= u
2
0 x 1
u(0) = 0, u(1) = 1
(11.1)
for small . In particular, nd the rst two terms of u as an expansion in powers of
the parameter .
Proof. We write u = u
0
(x) +u
1
(x) +O(
2
) as 0 and nd the rst two terms u
0
and u
1
. We have
u = u
0
+u
1
+ O(
2
),
u
2
=
_
u
0
+ u
1
+O(
2
)
_
2
= u
2
0
+ 2u
0
u
1
+O(
2
).
Plugging this into (11.1), we obtain
u
0xx
+ u
1xx
+ O(
2
) =
_
u
2
0
+ 2u
0
u
1
+O(
2
)
_
,
u
0xx
+ u
1xx
+ O(
2
) = u
2
0
+O(
2
).
O(1) terms:
u
0xx
= 0,
u
0
= c
0
x +c
1
,
u
0
(0) = c
1
= 0,
u
0
(1) = c
0
= 1,
u
0
= x.
O() terms:
u
1xx
= u
2
0
,
u
1xx
= u
2
0
,
u
1xx
= x
2
,
u
1
=
x
4
12
+c
2
x +c
3
,
u
1
(0) = c
3
= 0,
u
1
(1) =
1
12
+c
2
= 0 c
2
=
1
12
,
u
1
=
x
4
12

1
12
x.
u(x) = x +
_
x
4
12

1
12
x
_
+O(
2
).
Problem (F89, #5b). For the dierential equation
u
xx
= u
2
+x
3
u
3
(11.2)
look for any solution which are bounded for x near +. Determine the behavior u for
x near + for any such solutions.
Hint: Look for the dominant behavior of u to be in the form x
n
.
Proof. Let u = cx
n
. Plugging this into (11.2), we obtain
n(n 1)cx
n2
= c
2
x
2n
+c
3
x
3
x
3n
,
n(n + 1)cx
n2
= c
2
x
2n
+c
3
x
33n
.
Using the method of dominant balance, we want to cancel two terms such that the
third term is 0 at + compared to the other two. Let
n 2 = 3 3n,
n =
5
2
.
Also,
5
2
_
5
2
+ 1
_
c = c
3
,
c =
35
2
.
u(x) =
35
2
x
5
2
.
Problem (F03, #6a). For the cubic equation
3
x
3
2x
2
+ 2x 6 = 0 (11.3)
write the solutions x in the asymptotic expansion x = x
0
+x
1
+ O(
2
) as 0.
Find the rst two terms x
0
and x
1
for all solutions x.
Proof. As 0,
x = x
0
+x
1
+ O(
2
),
x
2
=
_
x
0
+ x
1
+O(
2
)
_
2
= x
2
0
+ 2x
0
x
1
+ O(
2
),
x
3
=
_
x
0
+ x
1
+O(
2
)
_
3
=
_
x
2
0
+ 2x
0
x
1
+ O(
2
)
__
x
0
+ x
1
+O(
2
)
_
= x
3
0
+ 3x
2
0
x
1
+ O(
2
).
3
(x
3
0
+ 3x
2
0
x
1
+ O(
2
)) 2(x
2
0
+ 2x
0
x
1
+ O(
2
)) + 2(x
0
+x
1
+O(
2
)) 6 = 0.
As 0, we ignore the O(
2
) terms:
2x
2
0
O(
2
) + 2x
0
+ 2x
1
+O(
2
) 6 = 0,
x
2
0
+x
0
+ x
1
3 +O(
2
) = 0. (11.4)
As 0, x
2
0
+x
0
+x
1
3 +O(
2
) x
0
3 = 0. Thus, x
0
= 3.
Plugging this value of x
0
into (11.4), we obtain
9 + 3 + x
1
3 + O(
2
) = 0,
9 +x
1
+ O(
2
) = 0,
x
2
= 9.
x = 3 + 9 +O(
2
).
Problem (F03, #6b). For the ODE
_
u
t
= u u
3
,
u(0) = 1,
(11.5)
write u = u
0
(t) +u
1
(t) +
2
u
2
(t) +O(
3
) as 0. Find the rst three terms u
0
, u
1
and u
2
.
Proof. We have u = u
0
+u
1
+
2
u
2
+ O(
3
) as 0.
u
3
=
_
u
0
+u
1
+
2
u
2
+ O(
3
)
_
3
= u
3
0
+ 3u
2
0
u
1
+ 3
2
u
2
0
u
2
+ 3
2
u
0
u
2
1
+ O(
3
).
u
0t
+ u
1t
+
2
u
2t
+ O(
3
)
= u
0
+u
1
+
2
u
2
+ O(
3
)
_
u
3
0
+ 3u
2
0
u
1
+ 3
2
u
2
0
u
2
+ 3
2
u
0
u
2
1
+O(
3
)
_
,
u
0t
+ u
1t
+
2
u
2t
+ O(
3
) = u
0
+u
1
+
2
u
2
u
3
0
3
2
u
2
0
u
1
+O(
3
),
O(1) terms:
u
0t
= u
0
,
u
0
= c
0
e
t
.
O() terms:
u
1t
= u
1
u
3
0
,
u
1t
= u
1
u
3
0
,
u
1t
u
1
= c
3
0
e
3t
,
u
1
= c
1
e
t
1
2
c
3
0
e
3t
.
O(
2
) terms:
24
2
u
2t
=
2
u
2
3
2
u
2
0
u
1
,
u
2t
= u
2
3u
2
0
u
1
,
u
2t
u
2
= 3c
2
0
e
2t
_
c
1
e
t
1
2
c
3
0
e
3t
_
,
u
2t
u
2
= 3c
2
0
c
1
e
t
e
2t
+
3
2
c
5
0
e
2t
e
3t
,
u
2
= c
2
e
t
3
2
c
2
0
c
1
e
t
e
2t
+
3
8
c
5
0
e
2t
e
3t
.
Thus,
u(t) = c
0
e
t
+
_
c
1
e
t
1
2
c
3
0
e
3t
_
+
2
_
c
2
e
t
3
2
c
2
0
c
1
e
t
e
2t
+
3
8
c
5
0
e
2t
e
3t
_
+O(
3
).
Initial condition gives
u(0) = c
0
+
_
c
1
1
2
c
3
0
_
+
2
_
c
2
3
2
c
2
0
c
1
+
3
8
c
5
0
_
+O(
3
) = 1.
Thus, c
0
= 1, c
1
=
1
2
, c
2
=
3
8
, and
u(t) = e
t
+
1
2
_
e
t
e
3t
_
+
2
_
3
8
e
t
3
4
e
t
e
2t
+
3
8
e
2t
e
3t
_
+O(
3
).
24
Solutions to ODEs in u1 and u2 are obtained by adding homogeneous and particular solutions.

ODE Yanovsky

Uploaded by

Document Information

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

ODE Yanovsky

Uploaded by

Copyright:

Available Formats

Ordinary Dierential Equations: Graduate Level Problems

= A(t)y +g(t) (2.1)

= A(t)y is called a solution matrix.

(t) = A(t)y(t), y(t

. Find the eigenvalues,

. The fundamental matrix is

Note: At t = 0, F(t) reduces to I.

= A(t)y + g(t), we get

(t) = A(t)F(t)v(t) +g(t)

(t) = A(t)F(t). Thus,

= A(t)y + g(t) has the

= Ay with F(0) = I. Therefore, every

= Ay has the form y(t) = e

= Ay. Change of variable y = Tz, where T is nonsingular constant matrix (to be

= i. (0,0) is Stable Center.

6,0) are Unstable Saddle Points.

= (A+ B(t))y is asymptotically stable.

= Ay are bounded. Let B(t) be continuous for

= (A+B(t))y are bounded

= (A+ B(t))y is stable.

= (A+B(t))y + f(t, y) (2.9)

= A(t)y +f(t, y) (2.12)

= (A+B(t))y are asymptotically equivalent.

= f(y) is the scalar product

(y) 0 in a neighborhood of 0, then

(y) is negative denite in a neighbor-

(y) is either positive denite or negative denite, and

(a) > 0, then the 0

(y) > 0, y ,= 0, the critical point at

(y) = 0 for y = (0, y

is neither positive definite nor negative definite.

(x, y) < 0. For instance, c = 1, a = 1.

(x, y) is negative denite, the critical point at

= f(y) where f(y)

(t) = f(x(t)), x(t) R

= f(y) starting in K remains in K for all future time.

, in fact, each of its components, is an invariant set with respect

invariant (as its components).

(y) = 0 at all points of L(C

0. Let the origin be the only invariant subset of the set

(y) = 0. Then the sero solution is asymptotically stable.

0, V (0) = 0. Let M be the largest invariant subset of

(y) = 0. Then all bounded solutions approach the set M as t .

+g(u) = 0. In our case,

+F(u), which gives

3, 0). The best value of

approaches the origin.

(y) is negative denite.

(u, v) is positive denite in a small neighborhood of (0, 0), i.e. V

, q and r are continuous on [a, b]. The dierential equation

+ qu and L(v) = (pv

denote the linear, second-order dierential operators de-

is called the adjoint of L and the dierential equation L

0, which means u = C. Since u(0) = u(1) = 0, then

+ (x)u. L is self-adjoint in the above norm, if

. L is self-adjoint in the above norm, if

(x) +u(x + 1)v(x)

-smooth functions and , are real constants.

= f, in Poisson Equation subsection of Eigenvalues

cu. We rst need to nd eigenfunctions and eigenvalues.

(1) = 0. Here 0 < c

(a) = 0. Equation (5.14)

= 0. Try (x) = Ax +B.

If < 0, try (x) = e