EE C222/ME C237 - Spring’18 - Lecture 5 Notes1 1

Licensed under a Creative Commons

Attribution-NonCommercial-ShareAlike
Murat Arcak 4.0 International License.
January 31 2018

Center Manifold Theory

Khalil (Section 8.1), Sastry (Section
7.6.1)
ẋ = f ( x ) f (0) = 0 (1)

∂ f 
Suppose A , has k eigenvalues will zero real parts, and
∂x  x=0
m = n − k eigenvalues with negative real parts.
" #
y
Define = Tx such that
z
" #
−1 A1 0
TAT =
0 A2

where the eigenvalues of A1 have zero real parts and the eigenvalues
of A2 have negative real parts.
Rewrite ẋ = f ( x ) in the new coordinates:

ẏ = A1 y + g1 (y, z)
(2)
ż = A2 z + g2 (y, z)

∂gi ∂gi
gi (0, 0) = 0, ∂y (0, 0) = 0, ∂z (0, 0) = 0, i = 1, 2.
Theorem 1: There exists an invariant manifold z = h(y) defined in a
neighborhood of the origin such that

∂h
h (0) = 0 (0) = 0.
∂y

z = h(y)

Reduced System: ẏ = A1 y + g1 (y, h(y)) y ∈ Rk

Theorem 2: If y = 0 is asymptotically stable (resp., unstable) for the

reduced system, then x = 0 is asymptotically stable (resp., unstable)
for the full system ẋ = f ( x ).
 ee c222/me c237 - spring’18 - lecture 5 notes 2

Characterizing the Center Manifold

Define w , z − h(y) and note that it satisfies

∂h  
ẇ = A2 z + g2 (y, z) − A1 y + g1 (y, z) .
∂y

The invariance of z = h(y) means that w = 0 implies ẇ = 0. Thus, the

expression above must vanish when we substitute z = h(y):
∂h  
A2 h(y) + g2 (y, h(y)) − A1 y + g1 (y, h(y)) = 0.
∂y

To find h(y) solve this differential equation for h as a function on y.

If the exact solution is unavailable, an approximation is possible. For

scalar y, expand h(y) as

h ( y ) = h 2 y 2 + · · · + h p y p + O ( y p +1 )

where h1 = h0 = 0 because h(0) = ∂h p +1 )

∂y (0) = 0. The notation O ( y
refers to the higher order terms of power p + 1 and above.

Example:
ẏ = yz
ż = −z + ay2 a 6= 0

This is of the form (2) with g1 (y, z) = yz, g2 (y, z) = ay2 , A2 = −1.
Thus h(y) must satisfy

∂h
−h(y) + ay2 − yh(y) = 0.
∂y

Try h(y) = h2 y2 + O(y3 ):

0 = −h2 y2 + O(y3 ) + ay2 − (2h2 y + O(y2 ))y(h22 + O(y3 ))

= ( a − h2 ) y2 + O ( y3 )
=⇒ h2 = a

Reduced System: ẏ = y( ay2 + O(y3 )) = ay3 + O(y4 ).

If a < 0, the full systems is asymptotically stable. If a > 0 unstable.
 ee c222/me c237 - spring’18 - lecture 5 notes 3

Discrete-Time Models and a Chaos Example

CT: ẋ (t) = f ( x (t)) DT: xn+1 = f ( xn ) n = 0, 1, 2, . . .

f (x∗ ) = 0 f ( x ∗ ) = x ∗ (“fixed point”)

Asymptotic stability criterion:

 Asymptotic stability criterion:

<λi ( A) < 0 where A , ∂∂xf  |λi ( A)| < 1 where A , ∂∂xf 
 
x=x∗ x=x∗
f 0 ( x ∗ ) < 0 for first order system | f 0 ( x ∗ )| < 1 for first order system

These criteria are inconclusive if the respective inequality is not strict,

but for first order systems we can determine stability graphically:

Cobweb Diagrams for First Order Discrete-Time Systems

Example: xn+1 = sin( xn ) has unique fixed point at 0. Stability test

above inconclusive since f 0 (0) = 1. However, the "cobweb" diagram
below illustrates the convergence of iterations to 0:

y=x

x1 y = f (x)
x2

x2 x1 x0

In discrete time, even first order systems can exhibit oscillations:

f (x) xn
q q

p p
x n
p q
 ee c222/me c237 - spring’18 - lecture 5 notes 4

Detecting Cycles Analytically

f ( p) = q f (q) = p =⇒ f ( f ( p)) = p f ( f (q)) = q

For the existence of a period-2 cycle, the map f ( f (·)) must have two
fixed points in addition to the fixed points of f (·).
Period-3 cycles: fixed points of f ( f ( f (·))).

Chaos in a Discrete Time Logistic Growth Model

x n +1 = r (1 − x n ) x n (3)
Range of interest: 0 ≤ x ≤ 1 ( x n > 1 ⇒ x n +1 < 0 )

r/4

x
0 1

We will study the range 0 ≤ r ≤ 4 so that f ( x ) = r (1 − x ) x maps [0, 1]

onto itself.
(
x ∗ = 0 and
Fixed points: x = r (1 − x ) x ⇒
x ∗ = 1 − 1r if r > 1.
r ≤ 1: x ∗ = 0 unique and stable fixed point

x
0 1

r > 1: x = 0 unstable because f 0 (0) = r > 1

x
0 1 1
1− r
 ee c222/me c237 - spring’18 - lecture 5 notes 5

Note that a transcritical bifurcation occurred at r = 1, creating the

new equilibrium
1
x∗ = 1 − .
r
Evaluate its stability using f 0 ( x ∗ ) = r (1 − 2x ∗ ) = 2 − r.

r < 3 ⇒ | f 0 ( x ∗ )| < 1 (stable)

r > 3 ⇒ | f 0 ( x ∗ )| > 1 (unstable).
At r = 3, a period-2 cycle is born:

x = f ( f ( x ))
= r (1 − f ( x )) f ( x )
= r (1 − r (1 − x ) x )r (1 − x ) x
= r2 x (1 − x )(1 − r + rx − rx2 )

0 = r2 x (1 − x )(1 − r + rx − rx2 ) − x
Factor out x and ( x − 1 + 1r ), find the roots of the quotient:
p
r + 1 ∓ (r − 3)(r + 1)
p, q =
2r

y=x

f ( f ( x ))

x
0 1
p 1− r q 1
√
This period-2 cycle is stable when r < 1 + 6 = 3.4494:

d
= f 0 ( f ( p)) f 0 ( p) = f 0 ( p) f 0 (q) = 4 + 2r − r2

f ( f ( x ))
dx x= p
√
|4 + 2r − r2 | < 1 ⇒ 3 < r < 1 + 6 = 3.4494
At r = 3.4494, a period-4 cycle is born!

“period doubling bifurcations”

r
0 1 3 3.44
 ee c222/me c237 - spring’18 - lecture 5 notes 6

r1 =3 period-2 cycle born

r2 = 3.4494 period-4 cycle born
r3 = 3.544 period-8 cycle born
r4 = 3.564 period-16 cycle born
..
.
r∞ = 3.5699
After r > r∞ , chaotic behavior for a window of r, followed by win-
dows of periodic behavior (e.g., period-3 cycle around r = 3.83).
Below is the cobweb diagram for r = 3.9 which is in the chaotic
regime:

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1