Linear Algebra: Lecture notes from Kolman and

Hill 9th edition.

Taylan Şengül
December 30, 2019

Please let me know of any mistakes in these notes.

Contents
Chapter 2. Flows on the line 1
2.1. A Geomtric Way of Thinking . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2.2. Fixed Point and Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.3. Population Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.4. Linear Stability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.5. Existence and Uniqueness Theory . . . . . . . . . . . . . . . . . . . . . . . 5
2.6. Impossibility of Oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.7. Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Chapter 2 Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

Chapter 3. Bifurcations 7
3.1. Saddle-Node Bifurcation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.2. Transcritical Bifurcation: To do . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.4. Pitchfork Bifurcation: To do . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.7. Insect Outbreak: To do . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Chapter 3 Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

Chapter 5. Two dimensional linear systems 9

5.1. Definitions and Examples: To do . . . . . . . . . . . . . . . . . . . . . . . . 9
5.2. Classification of Linear Systems: To do . . . . . . . . . . . . . . . . . . . . . 9
5.3. Love Affairs: To do . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Chapter 5 Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

Chapter 6. Two dimensional nonlinear systems fixed points 9

6.1 Phase Portraits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
6.3. Fixed Points and Linearization . . . . . . . . . . . . . . . . . . . . . . . . . 9
6.4. Rabbits vs Sheep . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
6.5. Conservative Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
6.7. Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
 6.8. Index Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Chapter 6 Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

Chapter 7. Limit Cycles. 18

7.0. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
7.2 Ruling Out Closed Orbits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
7.3 Poincaré-Bendixson Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
7.5 Relaxation Oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Chapter 7 Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

Chapter 2. Flows on the line

2.1. A Geomtric Way of Thinking
We will study
f : A ⊂ R → R, x:I⊂R→A (1)

ẋ(t) = f (x(t)),
Consider
ẋ = sin x, x(0) = x0
We can solve it.
Z
dx
= dt =⇒ t = csc xdx = − ln |csc x + cot x| + C
sin x
The solution is
csc x0 + cot x0
t=
csc x + cot x
π
Not very useful. Question: For x0 = 4, limt→∞ x(t) =?

Geomtric idea: x = position, ẋ = velocity of an imaginary particle.

The particle moves to right if f (x) = ẋ > 0 and to left if f (x) = ẋ < 0.
The particle stays if f (x) = ẋ = 0. Such points are called fixed points.

Figure 1: Phase portrait of ẋ = sin x

 Figure 2: The solution of ẋ = sin x, x(0) = π/4

A fixed point x∗ is called an (asymptotically) stable fixed point if any solution

with initial condition near x∗ tends to x∗ as t → ∞. Otherwise it is called an unstable
fixed point.

2.2. Fixed Point and Stability

Example. Find all fixed points of
ẋ = x2 − 1
and classify their stability.

Figure 3: f (x) = x2 − 1

Example. Sketch the phase portrait of

ẋ = x − cos x

and determine the stability of all fixed points. Hint: sketch y = x and y = cos x
separately.

Figure 4: f (x) = x − cos x

Example. Consider
ẋ = ax (2)

Show that x∗ = 0 is a stable fixed point if a < 0 and an unstable fixed point if a > 0.
2.3. Population Growth
Example. Population Growth. Most basic model

Ṅ = rN, N (0) = N0

N (t) is the population of a species at time t and r > 0 is a constant indicating the
growth ability.
The solution is N (t) = ert N0 . This solution is not realistic, no species can always
grow exponentially.
A more realistic population growth model is called the logistic equation
 
N
Ṅ = rN 1− (3)
K
N


K > 0 is a constant, called the carrying capacity. The growth rate is r 1 − K which
decreases linearly with N and is negative for N > K .

N


Figure 5: f (N ) = rN 1 − K

Picturing the solutions. Note that Ṅ increases with N when 0 < N < K/2 and
decreases when N > K/2. The solution is concave up, when 0 < N < K/2 and
concave down when N > K/2.

N


Figure 6: The solutions of ẋ = rN 1 − K with different initial conditions.

Lab tests have shown that the logistic growth model gives good results for the
growth of simple organisms such as bacteria under ideal conditions. However, for
more complex species, it fails.

2.4. Linear Stability Analysis

Let x∗ be a fixed point of (1), that is f (x∗ ) = 0. Then x∗ is stable if f 0 (x∗ ) < 0, unstable
if f 0 (x∗ ) > 0.
Proof. Let x(t) be a solution with an initial condition sufficiently close to x∗ .

η(t) = x(t) − x∗

If limt→∞ η(t) = 0 then x∗ is stable, and unstable otherwise.

Taylor’s expansion:

f 00 (x∗ ) 2
η̇ = ẋ = f (x∗ + η) = f (x∗ ) + f 0 (x∗ )η + η + · · · = f 0 (x∗ )η + O(η 2 )
2!
since f (x∗ ) = 0. Here O(η 2 ) denotes the remaining terms with η k , k ≥ 2. Now if
f 0 (x∗ ) 6= 0 and if η is “small enough” then O(η 2 ) part can be ignored:

η̇ ≈ f 0 (x∗ )η (4)

This equation is called the linearization of (1) at the fixed point x = x∗ .

For small η , η(t) ≈ exp(f 0 (x∗ ))C .

lim η(t) = 0, if f 0 (x∗ ) < 0

t→∞

When f 0 (x∗ ) > 0, as t increases η becomes larger and limt→∞ η(t) 6= 0. (Note: when
f 0 (x∗ ) > 0, η(t) ≈ exp(f 0 (x∗ ))C is no lorger valid for large t as ignoring the O(η 2 ) terms
is not valid when η is not small.)

Example. Logistic equation: f (N ) = rN (1 − N/K) with fixed points N ∗ = 0 and

N ∗ = K . Then f 0 (0) = r > 0, so that N ∗ = 0 is unstable and f 0 (K) = −r < 0 so that
N ∗ = K is stable.

Example. What can be said about stability of a fixed point when f 0 (x∗ ) = 0?
Nothing. Consider

(a) ẋ = −x3 , (b) ẋ = x3 , (c) ẋ = x2 , (d) ẋ = 0,

All these systems have a fixed point x∗ = 0 with f 0 (0) = 0. However the stability is
different in each case. In (c) x∗ = 0 is called half-stable. In (d), the x-axis is a line of
fixed points. Each fixed point is marginally stable, the perturbations do not decay to
zero, but they do not grow either.
Such examples seem artifical but they arise naturally in the context of bifurcations.
 Figure 7: The graphs of (a)−x3 , (b) x3 , (c) x2 , (d) 0.

2.5. Existence and Uniqueness Theory

Example. Show that the solution to ẋ = x1/3 with x(0) = 0 is not unique.
One solution is x(t) = 0. Another solution is
Z Z
x−1/3 dx = dt

2 3/2
which gives 32 x2/3 = t + C . The initial condition gives C = 0. Hence x(t) =


3t is
another solution.

When uniqueness fails, our geometric approach fails. The system can not choose
how to evolve.
Problem. Show that the above system has infinitely many solutions.
The reason of non-uniqueness: the slope f 0 (0) = ∞, the fixed point 0 is very unsta-
ble.
Existence and uniqueness theorem. If f and f 0 are continuous then the IVP
(initial value problem)
ẋ = f (x), x(0) = x0
has a unique solution x(t) defined on some interval (−τ, τ ).

Example. Consider the IVP

ẋ = 1 + x2 , x(0) = 0

By the above theorem, a unique solution exists. We can find the solution by sepa-
ration of variables:
Z Z
dx
dx = dt =⇒ arctan x = t + C
1 + x2
By the initial condition, x(t) = tan t. The solution exists only for −π/2 < t < π/2
as x(t) → ±∞ as t → ±π/2. There is no differentiable function defined on a larger
interval which solves the IVP. This phenemenon is called finite time blow-up.

2.6. Impossibility of Oscillations

In all our examples so far, all trajectories either approached a fixed point, or diverged
to ±∞. In fact, those are the only things that can happen for a vector field on the real
line.
The solutions can not oscillate in 1D systems.

Proof. Suppose on the contrary x(t) = x(t + T ) for some T > 0 and x(t) 6= x(t + s) for
all 0 < s < T . Suppose F 0 = f (such F always exists when f is continuous)
Z t+T  2 Z t+T Z t+T
dx dx d
0 6= dt f (x) dt = (F (x(t))) dt = F (x(t + T )) − F (x(t)) = 0
t dt t dt t dt

2.7. Potentials
Consider
dV
ẋ = − .
dx
Then  2
d dV dx dV
V (x(t)) = =− ≤0
dt dx dt dx
Thus for any solution x(t), V (x(t)) is non-increasing in t. It is strictly decreasing if x(t)
is not an equilibrium solution. Thus x(t) tends to minima of V which are stable fixed
points. Maxima of V are unstable fixed points.

Example.
ẋ = −x
Then V (x) = − −xdx = x2 . (choose C = 0). The solution tends to the fixed point
R

which is the minima of V which is x = 0.

Example. A bistable system: ẋ = x − x3 . This time V = − 12 x2 + 14 x4 known as double

well potential. There are two stable fixed points ±1 which are minima of V and an
unstable fixed point 0 which a maxima of V .

Figure 8: V = − 12 x2 + 14 x4
Chapter 2 Homework
• 2.2: 1, 3, 7, 8, 9

• 2.3: 1 (Solve the logistic equation (3)), 3, 4

• 2.4: 1, 3, 7, 8

• 2.5: 3, 4

• 2.6: 2

• 2.7: 1, 6

Chapter 3. Bifurcations
The qualitative structure of the flow can change as parameters are varied such as fixed
points can be created or destroyed, or their stability can change. These qualitative
changes in the dynamics are called bifurcations, and the parameter values at which
they occur are called bifurcation points.

3.1. Saddle-Node Bifurcation

In a saddle-node bifurcation, as a parameter is varied, two fixed points move toward
each other, collide, and mutually annihilate.

ẋ = r + x2 (5)

where r ∈ R is a parameter. (5) is the prototypical example of a saddle-node bifurca-

tion.

Figure 9: Fixed points and their stability for (5). r = 0 is a bifurcation point.
 Figure 10: Bifurcation diagram of (5).

Example. Give the linear stability analysis of (5).

√
Solution. There are two fixed points x∗ = ± r for r > 0, 1 for r = 0 and none for
√ √ √ √ √
r < 0. For r < 0, f 0 (− r) = 2 r > 0 and x∗ = − r is unstable, f 0 ( r) = −2 r < 0
√
and x∗ = r is stable.

Example. Show that ẋ = r − x − ex undergoes a saddle-node bifurcation as r is varied.

Find the bifurcation point.

At the bifurcation point, y = r − x and y = ex become tangent.

d −x d
e−x = r − x, and e = (r − x)
dx dx
From the second equation, e−x = 1 so x = 0. The first equation yields r = 1.

3.2. Transcritical Bifurcation: To do

3.4. Pitchfork Bifurcation: To do

3.7. Insect Outbreak: To do

Chapter 3 Homework
• 3.1: 1, 3

• 3.2: 1, 4

• 3.4: 1, 3, 7, 8

• 3.7: 1, 2
Chapter 5. Two dimensional linear systems
5.1. Definitions and Examples: To do

5.2. Classification of Linear Systems: To do

5.3. Love Affairs: To do

Chapter 5 Homework
1. 5.1: 1, 2, 10ad.

2. 5.2: 1, 2, 3, 6, 9, 11.

Chapter 6. Two dimensional nonlinear systems fixed

points
6.1 Phase Portraits
Watch: https://www.youtube.com/watch?v=9yh9DmNqdk4

Example (Example 6.1.1).

ẋ = x + e−y
(6)
ẏ = −y
Sketch the phase portrait.
Solution. The only fixed point is (−1, 0). To determine its stability, note that y(t) → 0
as t → ∞ and for large t, ẋ ≈ x + 1 which has an exponentially growing solution and
the fixed point is unstable.

Figure 11: Phase portrait of (6).

6.3. Fixed Points and Linearization

ẋ = f (x, y)
ẏ = g(x, y)
Suppose (x∗ , y ∗ ) is a fixed point, that is

f (x∗ , y ∗ ) = g(x∗ , y ∗ ) = 0

Consider small deviations,

u(t) = x(t) − x∗ , v(t) = y(t) − y ∗

∂f ∗ ∗ ∂f
u̇ = ẋ = f (x, y) = f (x∗ + u, y ∗ + v) = f (x∗ , y ∗ ) + u (x , y ) + v (x∗ , y ∗ ) + h.o.t.
∂x ∂y
∂g ∂g
v̇ = ẋ = g(x, y) = g(x∗ + u, y ∗ + v) = g(x∗ , y ∗ ) + u (x∗ , y ∗ ) + v (x∗ , y ∗ ) + h.o.t.
∂x ∂y
Let the Jacobian matrix be " #
∂f ∂f
∂x ∂y
A= ∂g ∂g
∂x ∂y

Since f (x∗ , y ∗ ) = g(x∗ , y ∗ ) = 0, if we neglect h.o.t, we have linearization about x∗ .

  " ∂f ∂f
# 
u̇ u
= ∂x
∂g
∂y
∂g
v̇ ∂x ∂y
v

Question. Does linearization give qualitatively correct dynamics near x∗ ? Answer.

Yes if the real parts of the eigenvalues of the Jacobian matrix are non-zero. That is
when x∗ is a saddle, a node or a spiral. But borderline cases (degenerate node, star,
center, non-isolated fixed point) can be altered.
Here is an example where the center of linear system is altered by nonlinear terms.

Example.

ẋ = −y + ax x2 + y
2
ẏ = x + ay x2 + y 2
Show that the linearized system incorrectly predicts that the origin is a center for all
values of a, whereas in fact the origin is a stable spiral if a < 0 and an unstable spiral
if a > 0.
Solution.
The Jacobian is  
0 −1
A=
1 0
which has trace τ = 0 and determinant ∆ = 1 > 0. Thus the origin is a center.
To analyze the nonlinear system, we change variables to polar coordinates. Let
x = r cos θ and y = r sin θ.

x2 + y 2 = r2 =⇒ xẋ + y ẏ = rṙ

rṙ = ar4 =⇒ ṙ = ar3

y xẏ − y ẋ
θ = arctan =⇒ θ̇ =
x r2
ṙ = ar3
(7)
θ̇ = 1
 Figure 12: Behavior of (7)

6.4. Rabbits vs Sheep

Lotka-Volterra model of competition: x(t) is the population of rabbits, y(t) is the popu-
lation of sheep. x, y ≥ 0.
ẋ = x(3 − x − 2y)
ẏ = y(2 − x − y)
If y = 0, rabbits grow as ẋ = x(3 − x), this is logistic growth. The effect of sheep on
rabbits is −2xy .
Fixed points: (0, 0), (3, 0), (0, 2), (1, 1). Jacobian matrix is
 
3 − 2x − 2y −2x
A=
−y 2 − x − 2y

For (0, 0): λ1 = 3, λ2 = 2 unstable node. Trajectories leave along λ2 = 2 direction

e2 = (0, 1)T .

For (0, 2): λ1 = −1, λ2 = −2 stable node. Trajectories approach along λ1 = −1

1
eigendirection which is −2

For (3, 0): λ1 = −3, λ2 = −1 stable node.

 For (1, 1): τ = −2, ∆ = −1, a saddle

Combining all together

The phase portrait has an interesting biological interpretation. It shows that one
species generally drives the other to extinction. Trajectories starting below the stable
manifold lead to eventual extinction of the sheep, while those starting above lead to
eventual extinction of the rabbits. This dichotomy occurs in other mod- els of compe-
tition and has led biologists to formulate the principle of competitive exclusion, which
states that two species competing for the same limited resource typically cannot coex-
ist.

6.5. Conservative Systems

Watch: https://www.youtube.com/watch?v=3s2lmZspEU8
Consider mechanical system with 1 degree of freedom
dV
mẍ = F (x) = −
dx
 
dV d 1
mẍẋ + ẋ = 0 =⇒ mẋ + V (x) = 0
dx dt 2
Hence the total energy
1
E= mẋ2 + V (x) ≡ C
2
For ẋ = f (x), E(x) is a conserved quantity if E(x) is a continuous, real valued
function that is constant on trajectories, that is dE
dt = 0, and not identically constant
on any open set (otherwise E(x) = 42 is a conserved quantity). If a system has a
conserved quantity, then it is called a conservative system.
Note that E is constant on trajectories mean that trajectories must lie on the level
sets of E (sets on which E is constant).

Example. A conservative system cannot have any attracting/repelling fixed points.

Solution. Suppose x∗ were an attracting fixed point. Then E(x∗ ) would be constant
on the basin of attraction of x∗ .

Example. Consider the double well potential V (x) = − 21 x2 + 41 x4 with mass m = 1.

ẍ = x − x3 .

ẋ = y
ẏ = x − x3
The equilibria are (0, 0), (±1, 0). The Jacobian is
 
0 1
A=
1 − 3x2 0

At (0, 0), ∆ = −1 which is a saddle. At (±1, 0), τ = 0, ∆ = 2 which are centers. We know
that small nonlinear terms can destroy centers predicted by the linear approximation.
But here they are actually centers.
Recall E = 12 y 2 − 12 x2 + 14 x4 . Trajectories lie on the sets E = c, c ∈ R. So we need
plot the contours of E = c.
To sketch this graph one can use Mathematica:

ContourPlot[y^2/2 - x^2/2 + x^4/4,{x,-2, 2}, {y -2, 2}]

To manually sketch:

1. Near (0, 0), ignore x4 term,

E ≈ y 2 /2 − x2 /2 = C

which are hyperbolas.

2. Near (1, 0), (x + 1)2 ≈ 4 and

E = y 2 /2 − 1/4(x − 1)2 (x + 1)2 ≈ y 2 /2 − (x − 1)2 = C 0

which are ellipses.

3. The vector field is vertical, ẋ = 0 on the line y = 0. The vector field is horizontal,
ẏ = 0 on the lines x = −1, x = 0, x = 1.

4. The contours have symmetries x → −x and y → −y . So the picture will be

symmetric with respect to both x and y axes.
 1 2
Figure 13: Contours of E = 2y − 12 x2 + 41 x4

Thus solutions of the system are typically periodic, except for the equilibrium solu-
tions and two very special trajectories: which start and end at (0, 0) called homoclinic
orbit. They are common in conservative systems but are rare otherwise.

Figure 14: The graph of V (x) = − 12 x2 + 14 x4 . Periodic orbits can be understood as the
oscillations of the particle in the double well. The homoclinic orbit corresponds to the
trajectory of a particle with just enough energy to end up (in infinite time) at the top.

Figure 15: Graph of E = 21 y 2 − 12 x2 + 14 x4 . The local minima of E correspond to centers

at (±1, 0) while the saddle of E corresponds to the saddle (0, 0).

Theorem 1. Let ẋ = f (x) be a system with a conserved quantity E . At the local

isolated minima/maxima of E (since the contours of E are closed by the Morse Lemma)
the system has a fixed point which is center. At the isolated saddles of E the system
has a saddle.
∂E ∂E
Recall that the critical point of E(x, y) are given by ∂x = ∂y = 0. Moreover let

Exx Exy
∆E =
Eyx Eyy
1. If ∆E > 0 and Exx > 0 at (x0 , y0 ) then E has an isolated local minimum at (x0 , y0 ).

2. If ∆E > 0 and Exx < 0 at (x0 , y0 ) then E has an isolated local maximum at (x0 , y0 ).

3. If ∆E < 0 then E has an isolated saddle at (x0 , y0 ).

6.7. Pendulum
Example. Pendulum: θ̈ + sin θ = 0. Let v = θ̇ . This is a conservative system with
potential − sin θ = −V 0 (θ) and V (θ) = − cos θ and energy E = v 2 /2 − cos θ . Critical
points of E occur at ∂E ∂E
∂θ = 0, ∂v = 0 which are θ = 0, sin θ = 0. The critical points are
(0, nπ), n ∈ Z.
cos θ 0
∆E = = cos θ
0 1

E has a local minimum at (θ, v) = (0, 2nπ), n ∈ Z. E has saddles at (θ, θ̇) = (0, π +
2nπ), n ∈ Z.

Figure 16: Graph of E = θ̇ 2 /2 − cos θ .

6.8. Index Theory

https://www.youtube.com/watch?v=O2fcpxLT5wk Index theory is a method that pro-
vides global information about the phase portrait.
Let ẋ = f (x), x ∈ R2 . Consider a simple (not intersecting itself) closed curve C
which does not pass through any fixed points of f . Note that C is not necessarily a
trajectory. At each point on C , we can define the angle φ = arctan(ẏ/ẋ).

The index of the closed curve with respect to the vector field f is

1
IC = [φ]
2π C
where [φ]C is the net change in angle as the curve is transversed counterclockwise.
Note that IC must be an integer which is the number of net revolutions.
 Figure 17: IC = 1.

Figure 18: IC = −1.

Example. Given ẋ = x2 y , ẏ = x2 − y 2 find IC where C is the unit circle.

Figure 19: IC = 0.

Properties of the Index

1. If C can be continuously deformed into C 0 without passing through a fixed point

then IC = IC 0 . proof. IC changes continuously and is an integer. It must be
constant.

2. If C does not enclose any fixed points then IC = 0. proof. By the previous
property, shrink C to a tiny circle C 0 without changing the index. But IC 0 = 0
since the vector field is almost constant on C 0 .

3. If we reverse the arrows by t → −t the index is unchanged. proof. All angles

change from φ to φ + π and the net change in φ stay the same.
 4. Index of a closed trajectory = +1.

Figure 20: Index of a closed trajectory = +1.

Index of a Point Suppose x∗ is an isolated fixed point of f . Then the index of I of

x is defined as IC where C is any closed curve that encloses x∗ and no other fixed
∗

points. By (1) above, index of a point is well-defined.

Index of a stable node is 1. By property (3) index of an unstable node is also 1.
Index of a saddle is -1. Index of a non fixed point is 0.
Spirals, centers, degenerate nodes and stars all have index 1. Thus, a saddle point
is truly a different animal from all the other familiar types of isolated fixed points.

Theorem 2. Any closed trajectory on R2 must enclose at least one fixed point.
Moreover if it encloses n fixed points with index Ii , then

IC = I1 + I2 + · · · In

Figure 21: Idea of the proof.

Proof.

We can use index theory to rule out closed trajectories.

Figure 22: A closed trajectory in the plane can enclose two centers and a saddle but
not two saddles and a center.
Example. Rabbit vs sheep system.

ẋ = x(3 − x − 2y)
ẏ = y(2 − x − y)

with x, y ≥ 0. As shown before there are 4 fixed points (0, 0) unstable node; (0, 2)
and (3, 0) stable nodes; (1, 1) saddle point. There can be no closed trajectory for this
system.

Figure 23: Rabbit sheep system can not have any closed trajectory.

Example. Show that the system ẋ = xe−x , ẏ = 1+x+y 2 has no closed orbits. Solution.
This system has no fixed points so it can not have any closed orbits.

Index theory can be generalized to 2-manifolds.

Chapter 6 Homework
1. 6.1: 1, 3, 14

2. 6.3: 1, 3, 12, 13, 15

3. 6.4: 1, 2, 3

4. 6.8: 1, 2, 3, 4, 5, 7, 11, 13

Chapter 7. Limit Cycles.

7.0. Introduction
A limit cycle is an isolated closed trajectory. Isolated means that neighboring trajecto-
ries are not closed; they spiral either toward or away from the limit cycle.
 Linear systems ẋ = Ax can not have limit cycles. They can have non-isolated
periodic orbits. Because if x(t) is a closed trajectory then so is cx(t) for every c.

Example. Consider ṙ = r(1 − r 2 ), θ̇ = 1 where r ≥ 0. The system has a limit cycle

x(t) = cos(t + θ0 ) and y(t) = sin(t + θ0 )

Figure 24: Stability of ṙ = r(1 − r 2 ).

Figure 25: Stable limit cycle of ṙ = r(1 − r 2 ), θ̇ = 1.

Figure 26: x coordinate of the solution of ṙ = r(1−r 2 ), θ̇ = 1 with some initial condition.
7.2 Ruling Out Closed Orbits
Ways to rule out closed orbit (closed trajectory):
Index theory: if there is a closed orbit then the sum of indices of fixed points
inside must be +1. See the example: rabbit vs sheep system.
Gradient systems: Closed orbits are impossible in gradient systems ẋ = −∇V .
Because on a trajectory V is constant. So if C is closed trajectory corresponding to
the periodic solution x(t + T ) = x(t) with T > 0 then
Z T Z T Z T
dV (x) 2
dt = ∇V · ẋdt = − kẋk dt < 0
0 dt 0 0

unless ẋ(t) = 0 for all 0 < t < T which implies the orbit is a fixed point and not a closed
trajectory.

Example. There are no closed orbits for ẋ = sin y , ẏ = x cos y . Solution. The system is
a gradient system with potential V = −x sin y .

Liapunov Functions Consider ẋ = f (x) with a fixed point at x∗ . Suppose

1. L is positive definite. L(x) > 0 for all x 6= x∗ and L(x∗ ) = 0.

2. L̇ is negative definite. L̇ < 0 for all x 6= x∗ .

Then x∗ is globally asymptotically stable: for all initial conditions x(t) → x∗ as t → ∞.

In particular the system has no closed orbits.
For the proof:
Z t
L(x(t)) = L(x(0)) + L̇(x(s))ds < L(x(0)).
0

So L(x(t)) is decreasing and minimum of L(x) is at x = 0. The actual proof is a bit

more involved.

Example. Show that the system ẋ = −x + 4y and ẏ = −x − y 3 has no closed orbits by

using a Liapunov function of the form L(x, y) = x2 + ay 2 , choosing a carefully.
Solution. Note that V (x) > 0 for all x 6= 0. V̇ = −2x2 + (8 − 2a)xy − 2ay 4 which is
negative-definite if a = 4.

Disadvantage of the Liapunov method. No general way of finding a Liapunov func-

tion.
Dulac’s criterion. Let ẋ = f (x) be a smooth vector field on a simply connected
(no holes inside) subset R of the plane. If there exists a smooth real valued function
g(x) such that ∇ · (g ẋ) has one sign on R then there are no closed orbits lying in R.
H
Proof. If C is a closed trajectory and n be its normal vector. Then C g ẋ · n = 0 since
ẋ · n = 0 on C . Let A be the region enclosed by C . By Green’s Theorem
ZZ I
∇ · (g ẋ)dA = g ẋ · n = 0
A C

which is a contradiction since ∇ · (g ẋ) has a single sign.

 Figure 27: Proof of Dulac’s criterion.

Example. Show ẋ = x(2 − x − y), ẏ = y(4x − x2 − 3) has no closed orbits on x, y > 0.

Solution. Pick g = 1/xy . See ∇ · (g ẋ) < 0 and the domain is simply-connected.

Disadvantage of the Dulac’s method.

No general method for finding g . It is hard to guess a g .
Try g = 1, 1/(xy), 1/(xa y b ), ...

Example. Study example 7.2.5.

7.3 Poincaré-Bendixson Theorem

In this section we show methods to show the existence of closed orbits.
Poincaré-Bendixson Theorem. Suppose

1. R is closed, bounded region in R2 .

2. ẋ = f (x) is smooth.

3. no fixed points in R.

4. there exists a trapped trajectory C : it starts in R and stays in R for all t.

Then either C is a closed trajectory or it spirals to a closed trajectory as t → ∞.

Figure 28: Region R in Poincaré-Bendixson Theorem will generally look like this be-
cause a closed trajectoy always encloseds a fixed point in the plane by the index theory.
Figure 29: The key to finding a trapped trajectory is to construct a trapping region
R where no trajectory can get out and all trajectories in R are trapped.

Example. Consider
ṙ = r(1 − r2 ) + µr cos θ
θ̇ = 1
Show that a closed orbit exists for all 0 < µ < 1.
Solution. Fix µ. Construct a trapping region by seeking two concentric circles with
radii rmin > 0 and rmax > 0 such ṙ < 0 at r = rmax and ṙ > 0 at r = rmin . This
guarantees that the annulus rmin < r < rmax is a trapping region.
Notice that when µ = 0, ṙ < 0 if r > 1 and ṙ > 0 if r < 1. Since

r(1 − r2 − µ) < ṙ < r(1 − r2 + µ)

choose p
2 2
1 − rmin − µ > 0 =⇒ 0 < rmin < 1 − µ =⇒ 0 < rmin < 1−µ
and choose p
2
1 − rmax + µ < 0 =⇒ rmax > 1 + µ.
The result follows by Poincaré-Bendixson.

Example. Glycolosis (breaking down of sugar to get energy).

ẋ = −x + ay + x2 y
ẏ = b − ay − x2 y

This is kinetic equation of glycolosis. x and y are concentrations of ADP and F6P,
a, b > 0.
Solution. Construct a trapping region by using nullclines (curves where ẋ = 0,
ẏ = 0.)
 Figure 30: Nullclines of glycolosis system.

Figure 31: Trapping region of glycolosis system. The right boundary has slope -1.

How to come up with the right upper vertex (b, b/a) of the trappring region? First
note that
ẏ
x, y  1 =⇒ ẋ ≈ x2 y, ẏ ≈ −x2 y, ≈ −1
ẋ
To better see
ẏ
ẋ + ẏ = b − x =⇒ < −1 if x > b.
ẋ
Can we conclude that there is a closed orbit inside the trapping region? No! There
is a fixed point in the region (at the intersection of the nullclines), and so the condi-
tions of the Poincaré-Bendixson theorem are not satisfied. But if this fixed point is a
repeller, then we can prove the existence of a closed orbit by considering the modified
“punctured” region. Do a linear stability analysis to find that at the fixed point, the
determinant is ∆ = a + b2 > 0 and the trace is τ > 0.

No Chaos in the Phase Plane by Poincaré-Bendixson

Dynamical possibilities in the phase plane are very limited: if a trajectory is confined
to a closed, bounded region that contains no fixed points, then the trajectory must
eventually approach a closed orbit. Nothing more complicated is possible.
In higher-dimensional systems, the Poincaré-Bendixson theorem no longer applies.
It may happen that trajectories may wander around forever in a bounded region with-
out settling down to a fixed point or a closed orbit.

7.5 Relaxation Oscillations

Mechanical and Electrical Vibrations

my 00 = −µy 0 − ky
This equation models both mechanical vibrations such as motion of a spring and elec-
trical vibrations in an electric circuit. In the case of mechanic spring, this follows from
the Newton’s Law. my 00 is the acceleration term, −ky is the restoring force due to the
spring, −µy 0 term is the damping term.
Show the following:

1. Without damping µ = 0, this is just harmonic oscillator: all orbits are periodic
except when y(0) = y 0 (0) = 0.

2. If µ > 0, then all solutions decay to zero.

3. If µ < 0 then all solutions tend to infinity as t to∞.

The Van der Pol oscillator

The Van der Pol oscillator was originally proposed by the Dutch electrical engineer
and physicist Balthasar van der Pol while he was working at Philips. Van der Pol found
stable oscillations,[2] which he subsequently called relaxation-oscillations and are now
known as a type of limit cycle in electrical circuits employing vacuum tubes.

ẍ + µ x2 − 1 ẋ + x = 0

1. Nonlinear damping: µ x2 − 1 ẋ. For µ > 0, if x2 − 1 > 0 then the damping is
positive and the solutions decay, when x2 − 1 < 0 then the damping is negative
and the solutions are pumped. So it is plausible that the system will settle down
to some oscillation.

2. Using Poincaré-Bendixson type argument, we can show that there exists a unique
stable limit cycle for all µ > 0. The proof is elobarate.
 Aim is to investigate: µ  1 (in this section), 0 < µ  1 (in the next section).
Trick: tricky change of variables.
Note:
  
2 d 1 3
ẍ + µẋ(x − 1) = ẋ + µ x −x
dt 3
Let
1 3
w = ẋ + µF (x), F (x) = x −x
3
From the equation
ẋ = w − µF (x)
ẇ = −x
Let
w
y=
µ
Then
ẋ = µ(y − F (x))
1
ẏ = − x
µ
For µ  1,

y − F (x) ∼ O(1) =⇒ |ẋ| ∼ O(µ)  1, |ẏ| ∼ O(µ−1 )  1.

Hence the velocity is enormous in the horizontal direction except on the cubic null-
cline.

Figure 32: Van der Pol system. The graph has a local min at B = (1, − 32 ) and local max
x3
at D = (−1, 32 ). Compute A by 3 − x = 2 which gives x = 2. So A = (2, 23 ). Similarly
C = (−2, − 23 ).

This analysis shows that the limit cycle has two widely separated time scales: the
crawls require ∆t ∼ O(µ). Why? Intution. On the crawls, y -speed is O(1/µ) and the
distance traveled is O(1). So the time = distance/speed = O(µ)  1. The jumps require
∆t ∼ O(µ−1 ).
 2

10 20 30 40 50

-1

-2

mu = 10;
sol = NDSolve[{y’’[t] + mu*(y[t]^2 - 1) y’[t] + y[t] == 0, y[0] == .5,
y’[0] == .5}, y, {t, 0, 5*mu}];
Plot[y[t] /. sol, {t, 0, 5*mu}]

Example 7.5.2 Estimate the period for µ  1.

The period T ∼ 2×(time from A to B ).
Between A and B , use the fact that w ≈ µF (x).

tB 1 1  
−1
Z Z Z
dt dw
µ(x2 − 1) dx = µ (3 − 2 ln 2)


T =2 dt ≈ 2 dx = 2
tA 2 dw dx 2 x

which is O(µ) as expected.

The formula can be refined. With much work, one can show that

T ≈ µ (3 − 2 ln 2) + 2αµ−1/3

where α ≈ 2.338 . . . is the smallest root of the Airy function.

Chapter 7 Homework
1. 7.1: 1, 3, 8.

2. 7.2: 1, 2, 6, 10, 12, 18

3. 7.3: 1, 3, 4, 5, 10.