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

Licensed under a Creative Commons

Attribution-NonCommercial-ShareAlike
Murat Arcak 4.0 International License.
January 17 2018

Nonlinear Systems

ẋ = Ax + Bu −→ ẋ = f ( x, u) (1)

Analysis: We use the shorthand notation ẋ = f ( x )

d
for dt x (t) = f ( x (t)).
ẋ = f ( x ) f : Rn → Rn time-invariant (autonomous)
ẋ = f (t, x ) f : R × Rn → Rn time-varying (non-autonomous)

Design:

ẋ = f ( x, u) u to be designed as a function of x.

Equilibria

x = x ∗ is an equilibrium for ẋ = f ( x ) if f ( x ∗ ) = 0.

Example: Linear system ẋ = Ax.

If A is nonsingular, x ∗ = 0 is the unique equilibrium.
If A is singular, the nullspace defines a continuum of equilibria.

Example: Logistic growth model in population dynamics

r
 x
ẋ = f ( x ) = r 1 − x, r > 0, K > 0 (2)
| {z K }
growth rate
x
x > 0 denotes the population and K is called the carrying capacity. K

For systems with a scalar state variable x ∈ R, stability can be deter-

mined from the sign of f ( x ) around the equilibrium. In this example f (x)

f ( x ) > 0 for x ∈ (0, K ), and f ( x ) < 0 for x > K; therefore

x=0 unstable equilibrium x

x=K asymptotically stable.

 ee c222/ me c237 - spring’18 - lecture 1 notes 2

Linearization

Local stability properties of x ∗ can be determined by linearizing the

vector field f ( x ) at x ∗ :

∗ ∗ ∂ f 
f ( x + x̃ ) = f ( x ) + x̃ + higher order terms (3)
| {z } ∂x  x= x∗
= 0 | {z }
, A
Thus, the linearized model is:
x̃˙ = A x̃. (4)

If <λi ( A) < 0 for each eigenvalue λi of A, then x ∗ is asymp. stable.

If <λi ( A) > 0 for some eigenvalue λi of A, then x ∗ is unstable.

Example: Logistic growth model above:

f (x)

f 0 (0) > 0 f 0 (K ) < 0

unstable stable
Caveats:

1. Only local properties can be determined from the linearization.

Example: The logistic growth model linearized at x = 0 (ẋ = rx)
would incorrectly predict unbounded growth of x (t). In reality,
x (t) → K.

2. If <λi ( A) ≤ 0 with equality for some i, then linearization is

inconclusive as a stability test. Higher order terms determine
stability.
Example: f ( x ) = x3 vs. f ( x ) = − x3

x x

f 0 (0) = 0 in each case, but one is stable and the other is unstable.
 ee c222/ me c237 - spring’18 - lecture 1 notes 3

Second order example: Pendulum

`mθ̈ = −k`θ̇ − mg sin θ (5) θ

" #
θ `
Define x = . State space: S1 × R.
θ̇

ẋ1 = x2
(6) mg sin θ
k g
ẋ2 = − x2 − sin x1
m ` mg

Equilibria: (0, 0) and (π, 0)

 

 0 1
(stable) at x1 = 0

" # 
 g k

− −

∂f 0 1 
= =  ` `
∂x − g` cos x1 − k` 
 0 1
(unstable) at x1 = π

  
 g k
−


` `

Phase portrait: plot of x1 (t) vs. x2 (t) for 2nd order systems

x2 Figure 1: Phase portrait of the pendu-

lum for the undamped case k = 0.

0 π 2π x1

Essentially Nonlinear Phenomena

1. Finite Escape Time

Example: ẋ = x2
d −1
x = − x −2 ẋ = −1
dt
1 1
⇒ − = −t (7)
x (t) x (0) x
1
⇒ x (t) = 1
x (0)
−t

For linear systems, x (t) → ∞ cannot happen in finite time.

x (0)
t
1
x (0)
 ee c222/ me c237 - spring’18 - lecture 1 notes 4

2. Multiple Isolated Equilibria

Linear systems: either unique equilibrium or a continuum
Pendulum: two isolated equilibria (one stable, one unstable)
“Multi-stable” systems: two or more stable equilibria
Example: bistable switch

ẋ1 = − ax1 + x2 x1 : concentration of protein

x12 (8)
ẋ2 = 1+ x12
− bx2 x2 : concentration of mRNA

a > 0, b > 0 are constants. State space: R≥0 × R≥0 .

This model describes a positive feedback where the protein en-
x12
coded by a gene stimulates more transcription via the term 1+ x12
.
Single equilibrium at the origin when ab > 0.5. If ab < 0.5, the line
where ẋ1 = 0 intersects the sigmoidal curve where ẋ2 = 0 at two
other points, giving rise to a total of three equilibria:

x2

ẋ1 = 0
ẋ2 = 0

x1

stable unstable stable

(gene off) (saddle point) (gene on)