Lecture 3

◮ Phase-plane analysis
◮ Classification of singularities
◮ Stability of periodic solutions

Material
◮ Glad and Ljung: Chapter 13
◮ Khalil: Chapter 2.1–2.3
◮ Lecture notes
Today’s Goal

You should be able to

◮ sketch phase portraits for two-dimensional systems
◮ classify equilibria into nodes, focus, saddle points, and center
points.
◮ analyze limit cycles through Poincaré maps
First glipse of phase plane portraits: Consider the system

ẋ1 = x21 + x2
ẋ2 = −x1 − x2

x1 ’ = x12 + x2
x2 ’ = − x1 − x2

0
x2

−1

−2

−3

−4

−2 −1 0 1 2 3 4
x1
First glipse of phase plane portraits: Consider the system

ẋ1 = x21 + x2
ẋ2 = −x1 − x2

x1 ’ = x12 + x2
x2 ’ = − x1 − x2

0
x2

−1

−2

−3

−4

−2 −1 0 1 2 3 4
x1

Flow-interpretation: To each point (x1 , x2 ) in the plane there is an

associated flow-direction dx
dt = f (x1 , x2 )
First glipse of phase plane portraits: Consider the system

ẋ1 = x21 + x2
ẋ2 = −x1 − x2

x1 ’ = x12 + x2
x2 ’ = − x1 − x2

0
x2

−1

−2

−3

−4

−2 −1 0 1 2 3 4
x1

In the point (x1 , x2 ) = (1, 2) the vector field is pointing in the

direction (12 + 2, −1 − 2) =(3, −3).
Linear Systems Revival

   
d x1 x
=A 1
x
dt 2 x2
Analytic solution: x(t) = eAt x(0).
If A is diagonalizable, then
 
At Λt −1
  eλ1 t 0  −1
e =Ve V = v1 v2 λ t v 1 v 2
0 e 2

where v1 , v2 are the eigenvectors of A (Av1 = λ1 v1 etc).

Matlab:
>> [V,Lambda]=eig(A)
Example: Two real negative eigenvalues

Given the eigenvalues λ1 < λ2 < 0, with corresponding

|{z} |{z}
faster slower
eigenvectors v1 and v2 , respectively.

Solution: x(t) = c1 eλ1 t v1 + c2 eλ2 t v2

Fast eigenvalue/vector: x(t) ≈ c1 eλ1 t v1 + c2 v2 for small t.

Moves along the fast eigenvector for small t

Slow eigenvalue/vector: x(t) ≈ c2 eλ2 t v2 for large t.

Moves along the slow eigenvector towards x = 0 for large t
Example—Stable Node

 
−1 1
ẋ = x
0 −2
 
  1 −1
(λ1 , λ2 ) = (−1, −2) and v1 v2 =
0 1
v1 is the slow direction and v2 is the fast.
Equilibrium Points for Linear Systems
stable node unstable node saddle point
Imλi = 0 : λ1 , λ2 < 0 λ1 , λ2 > 0 λ1 < 0 < λ2

Imλi 6= 0 : Reλi < 0 Reλi > 0 Reλi = 0

stable focus unstable focus center point

x2

x1

Im λ
Re λ
Example—Unstable Focus

 
σ −ω
ẋ = x, σ, ω > 0, λ1,2 = σ ± iω
ω σ
   −1
At 1 1 eσt eiωt 0 1 1
x(t) = e x(0) = x(0)
−i i 0 eσt e−iωt −i i
p
In polar coordinates r = x21 + x22 , θ = arctan x2 /x1
(x1 = r cos θ, x2 = r sin θ):

ṙ = σr
θ̇ = ω
Example- unstable focus cont’d

λ1,2 = 1 ± i λ1,2 = 0.3 ± i

Equilibrium Points for Linear Systems
stable node unstable node saddle point
Imλi = 0 : λ1 , λ2 < 0 λ1 , λ2 > 0 λ1 < 0 < λ2

Imλi 6= 0 : Reλi < 0 Reλi > 0 Reλi = 0

stable focus unstable focus center point

x2

x1

Im λ
Re λ
4 minute exercise

What is the phase portrait if λ1 = λ2 ?

Hint: For λ1 = λ2 = λ there are two different cases: only one

linearly independent eigenvector or all vectors are eigenvectors
Star Node or Multi-Tangent Node
Case I: If  
λ 0
ẋ = x, rank (λI − A) = 0
0 λ
then the solution is

x1 (t) = x1 (0)eλt
x2 (t) = x2 (0)eλt
One Tangent Node
Case II: If  
λ 1
ẋ = x, rank (λI − A) = 1
0 λ
then the solution is

x1 (t) = x1 (0)eλt + tx2 (0)eλt

x2 (t) = x2 (0)eλt
 T
There is only one eigenvector: v1 = αv2 = 1 0 .
Linear Time-Varying Systems (warning)

Warning: Pointwise “Left Half-Plane eigenvalues” of A(t)

(i.e., time-varying systems) do NOT impose stability!!!
 
−1 + α cos2 t 1 − α sin t cos t
A(t) = , α>0
−1 − α sin t cos t −1 + α sin2 t

Pointwise eigenvalues are given by

√
α − 2 ± α2 − 4
λ(t) = λ =
2
which are in the LHP for 0 < α < 2 (and here even constant). However,
 (α−1)t 
e cos t e−t sin t
x(t) = x(0),
−e(α−1)t sin t e−t cos t

which is an unbounded solution for α > 1.

Phase-Plane Analysis for Nonlinear Systems

Close to equilibria “nonlinear system” ≈ “linear system”.

Theorem Assume
ẋ = f (x)
is linearized at x0 so that

ẋ = Ax + g(x),

where g ∈ C 1 and g(x)−g(x 0)

kx−x0 k → 0 as x → x0 .
If ż = Az has a focus, node, or saddle point, then ẋ = f (x) has
the same type of equilibrium at the origin.
If the linearized system has a center, then the nonlinear system has
either a center or a focus.
How to Draw Phase Portraits

If done by hand then

1. Find equilibria (also called singularities)
2. Sketch local behavior around equilibria
dx1 ẋ1
3. Sketch (ẋ1 , ẋ2 ) for some other points. Use that dx2 = ẋ2 .
4. Try to find possible limit cycles
5. Guess solutions

Matlab: pptool6/pptool7, dfield6/dfield7, dee,

ICTools, etc.
PPTool and some other tools for Matlab is available on or via
http://www.control.lth.se/course/FRTN05
Phase-Locked Loop

A PLL tracks phase θin (t) of a signal sin (t) = A sin[ωt + θin (t)].

sin “θout ”
Phase Filter VCO
Detector

θin e K θ̇out 1 θout

sin(·)
− 1 + sT s
Singularity Analysis of PLL

Let x1 (t) = θout (t) and x2 (t) = θ̇out (t).

Assume K, T > 0 and θin (t) = θin constant.

ẋ1 = x2
ẋ2 = −T −1 x2 + KT −1 sin(θin − x1 )

Singularities are (θin + nπ, 0), since

ẋ1 = 0 ⇒ x2 = 0
ẋ2 = 0 ⇒ sin(θin − x1 ) = 0 ⇒ x1 = θin + nπ
Singularity Classification of Linearized System

Linearization gives the following characteristic equations:

n even:
λ2 + T −1 λ + KT −1 = 0
K > (4T )−1 gives stable focus
0 < K < (4T )−1 gives stable node
n odd:
λ2 + T −1 λ − KT −1 = 0
Saddle points for all K, T > 0
Phase-Plane for PLL



K = 1/2, T = 1: Focus 2kπ, 0 , saddle points (2k + 1)π, 0
Summary

Phase-plane analysis limited to second-order systems (sometimes it

is possible for higher-order systems to fix some states)
Many dynamical systems of order three and higher not fully
understood (chaotic behaviors etc.)
Bonus — Discrete Time

Many results are parallel (observability, controllability,...)

Example: The difference equation

xk+1 = f (xk )

is asymptotically stable at x∗ if the linearization

∂f
has all eigenvalues in |λ| < 1
∂x x∗

(that is, within the unit circle).

Example (cont’d): Numerical iteration

xk+1 = f (xk )

to find fixed point

x∗ = f (x∗ )

When does the iteration converge?

f (x) x x
x f (x) x ?
f (x)

f (x)

x∗ x
∗
x∗ x∗ x∗
Periodic Solutions: x(t + T ) = x(t)

Example of an asymptotically stable periodic solution:

ẋ1 = x1 − x2 − x1 (x21 + x22 )
(1)
ẋ2 = x1 + x2 − x2 (x21 + x22 )
Periodic solution: Polar coordinates.

Let
x1 = r cos θ ⇒ dx1 = cos θdr − r sin θdθ
x2 = r sin θ ⇒ dx2 = sin θdr + r cos θdθ
⇒     
ṙ 1 r cos θ r sin θ ẋ1
=
θ̇ r − sin θ cos θ ẋ2
Now
ẋ1 = r(1 − r2 ) cos θ − r sin θ
ẋ2 = r(1 − r2 ) sin θ + r cos θ
which gives
ṙ = r(1 − r2 )
θ̇ = 1
Only r = 1 is a stable equilibrium!
A system has a periodic solution if for some T > 0

x(t + T ) = x(t), ∀t ≥ 0

Note that a constant value for x(t) by convention not is regarded

as periodic.
◮ When does a periodic solution exist?
◮ When is it locally (asymptotically) stable? When is it globally
asymptotically stable?
Poincaré map (“Stroboscopic map”)

ẋ = f (x), x ∈ Rn
ϕt (q) is the solution starting in q after time t.
Σ ⊂ Rn−1 is a hyperplane transverse to ϕt .
The Poincaré map P : Σ → Σ is

P (q) = ϕτ (q) (q), τ (q) is the first return time

P (q)
q ϕt (q)
Limit Cycles

If a simple periodic orbit pass through q ∗ , then P (q ∗ ) = q ∗ .

Such an orbit is called a limit cycle.
q ∗ is called a fixed point of P.

P (q ∗ ) = q ∗

Does the iteration qk+1 = P (qk ) converge to q ∗ ?

Locally Stable Limit Cycles

∂P
The linearization of P around q ∗ gives a matrix W = ∂q q ∗ so

(qk+1 − q ∗ ) ≈ W (qk − q ∗ ),

if qk is close to q ∗ .
◮ If all |λi (W )| < 1, then the corresponding limit cycle is locally
asymptotically stable.
◮ If |λi (W )| > 1, then the limit cycle is unstable.
Linearization Around a Periodic Solution

The linearization of

ẋ(t) = f (x(t))

around x0 (t) = x0 (t + T ) is

˙
x̃(t) = A(t)x̃(t)
∂f

A(t) = x0 (t) = A(t + T )
∂x
P is the map from the solution at t = 0 to t = τ (q).
Example—Stable Unit Circle

Rewrite (1) in polar coordinates:

ṙ = r(1 − r 2 )
θ̇ = 1

Choose Σ = {(r, θ) : r > 0, θ = 2πk}.

The solution is
 
−2 −2t −1/2
ϕt (r0 , θ0 ) = [1 + (r0 − 1)e ] , t + θ0

First return time from any point (r0 , θ0 ) ∈ Σ is τ (r0 , θ0 ) = 2π.

Example—Stable Unit Circle

The Poincaré map is

P (r0 ) = [1 + (r0−2 − 1)e−2·2π ]−1/2

r0 = 1 is a fixed point.
The limit cycle that corresponds to r(t) = 1 and θ(t) = t is locally
asymptotically stable, because
dP  
W = (1) = e−4π
dr0
and
dP
|W | = (1) = |e−4π | < 1
dr0
Example—The Hand Saw

Can we stabilize the inverted pendulum by vertical oscillations?

The Hand Saw—Poincaré Map

ẋ1 = x2
 
1 2
ẋ2 = g + aω sin x3 sin x1
ℓ
ẋ3 (t) = ω

Choose Σ = {x3 = 2πk}.

Σ
The Hand Saw–Poincaré Map

q ∗ = 0 and T = 2π/ω. No explicit expression for P . It is, however,

easy to determine W numerically. Do two (or preferably many
more) different simulations with different, small, initial conditions
x(0) = y and x(0) = z.
Solve W through (least squares solution of)
   
x(T )
 x(T )  = W y z 

x(0)=y x(0)=z

This gives for a = 1cm, ℓ = 17cm, ω = 180

 
 1.37 0.035
W =  

−3.86 0.630

which has eigenvalues (1.047, 0.955). Unstable.

W is stable for ω > 183
The Hand Saw—Stability Condition

Make the assumptions that

ℓ≫a and aω 2 ≫ g

Then some calculations show that the Poincaré map is stable at

q ∗ = 0 when √
2gℓ
ω>
a
a = 1 cm and ℓ = 17 cm give ω > 182.6 rad/s (29 Hz).
The Hand Saw—Simulation
Simulation results give good agreement

0.5 ω = 183
0

−0.5

−1
0 10 20 30 40 50 60 70 80

1000

500 ω = 182

0
0 10 20 30 40 50 60 70 80
Next Lecture

◮ Lyapunov methods for stability analysis

Lyapunov generalized the idea of: If the total energy is dissipated
along the trajectories (i.e the solution curves), the system must be
stable.

Benefit: Might conclude that a system is stable or asymptotically

stable without solving the nonlinear differential equation.
Lab 1: sign-up starts on Monday