Capitolo 0. INTRODUCTION 2.

Trajectories in the state space

To show the typical behaviors of the state space trajectories as a function of the
position of the system poles, we will refer to the following example, a dynamic
system with a mass, a spring and a dumper.
The differential equation describing the dynamics of the system composed by
a mass M , a spring K, a dumper b and an external force F is the following :

K b
F = M ẍ + bẋ + Kx → x− ẍ = −
ẋ + F
M M
An equivalent state space dynamic description of the system is the following:
    

 ẋ(t) = 0 1 0
K b x(t) + 1 u(t)
(1) − M − M M


y(t) = [ 1 0 ] x(t)
where    
x1 x
x= =
x2 ẋ
The dynamic behaviors of the system is a function of the numerical values of
the parameters. We will consider three different cases.

————–

Case I. Setting M = 1, K = 2 and b = 3 one obtains the following system:

    
 0 1 0
ẋ(t) = x(t) + u(t)
(2) −2 −3 1

y = [ 1 0 ] x(t)
The characteristic polynomial of matrix A is:
 
λ −1
det(λI − A) = det = λ2 + 3λ + 2 = (λ + 1)(λ + 2)
2 λ+3
The eigenvalues of matrix A are:
λ1 = −1, λ2 = −2
Zanasi Roberto - System Theory. A.A. 2015/2016
Capitolo 2. CANONICAL FORMS 2.2

The two distinct eigenvalues λ1 and λ2 generate two linearly independent ei-
genvalues v1 and v2 which can be determined solving the following linear
homogeneous systems
(A − λ1I)v1 = o, (A − λ2I)v2 = o
that is    
1 1 1
v1 = o → v1 =
−2 −2 −1
and    
2 1 1
v2 = o → v2 =
−2 −1 −2
Let us consider the state space transformation x = T x where
   
1 1 2 1
T = [v1 v2] = , T−1 = .
−1 −2 −1 −1
The new system matrix A has a diagonal form with the eigenvalues located
on the main diagonal:
     
2 1 0 1 1 1 −1 0
A = T−1AT = =
−1 −1 −2 −3 −1 −2 0 −2
The exponential of the diagonal matrix At is:
 −t 
e 0
eAt =
0 e−2t
The free evolution of system (2) starting from the initial condition x0 is:
−1
x(t) = eAtx0 = e(TAT )t x0 = TeAtT−1x0
   −t  
1 1 e 0 2 1
= x0
−1 −2 0 e−2t −1 −1
 
2e−t − e−2t e−t − e−2t
= x0
−2e−t + 2e−2t −e−t + 2e−2t
Computation of eAt in Matlab:
A=[ 0 1; ... +- -+
-2 -3]; | 2 exp(-t) - exp(-2 t), exp(-t) - exp(-2 t) |
syms t; --> | |
Eat=expm(A*t); | 2 exp(-2 t) - 2 exp(-t), 2 exp(-2 t) - exp(-t) |
pretty(Eat) +- -+

Zanasi Roberto - System Theory. A.A. 2015/2016

Capitolo 2. CANONICAL FORMS 2.3

The behavior of the state space trajectories x(t) in the “original” space is the
following:

1.5

0.5
x2

−0.5

−1

−1.5
v2 v1

−2
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2

x1
The two red trajectories correspond to the two subspaces Uλ1 and Uλ2 associa-
ted to the eigenvectors v1 and v2. The same trajectories in the transformed
space x(t) have the following behavior:

2
Stable Node

1.5

v2
0.5

v1
x2

−0.5

−1

−1.5

−2
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2
x1

Trajectories of this type are referred to as: “Stable Node”.

Zanasi Roberto - System Theory. A.A. 2015/2016
Capitolo 2. CANONICAL FORMS 2.4

Case II. Setting M = 1, K = 1 and b = 2 the system becomes:

    
 0 1 0
ẋ(t) = x(t) + u(t)
(3) −1 −2 1

y = [ 1 0 ] x(t)
The characteristic polynomial of matrix A is: det(λI − A) = λ2 + 2λ + 1 =
(λ + 1)2. The matrix A has only one eigenvalue, λ1,2 = −1, with algebraic
multiplicity 2. The corresponding eigenvector v1 can be determined solving
the following homogeneous system:
   
1 1 1
(A − λ1I)v1 = v1 = o → v1 =
−1 −1 −1
A second order generalized eigenvector v2 is then determined solving the
following linear system:
     
1 1 1 0
(A − λ1I)v2 = v1 ↔ v2 = → v2 =
−1 −1 −1 1
let us now consider the following state space transformation x = Tx where:
   
1 0 1 0
T = [v1 v2] = , T−1 =
−1 1 1 1
The new state matrix A has the following form:
   −t −t 
−1 1 e te
A = T−1AT = → eAt =
0 −1 0 e−t
The free evolution of system (3) starting from the initial condition x0 is:
−1
x(t) = eAtx0 = eTAT tx0 = TeAtT−1x0
   −t −t   
1 0 e te 1 0
= x0
−1 1 0 e−t 1 1
 −t −t −t

e + te te
= −t x0
−te e − te−t
−t

Computation of eAt in Matlab:

A= [0 1; ... +- -+
-1 -2]; | exp(-t) (t + 1), t exp(-t) |
syms t; --> | |
Eat=simplify(expm(A*t)); | -t exp(-t), -exp(-t) (t - 1) |
pretty(Eat) +- -+

Zanasi Roberto - System Theory. A.A. 2015/2016

Capitolo 2. CANONICAL FORMS 2.5

The trajectories x(t) in the state space (x1 , x2) is the following:
Stable degenerate node
2

1.5

0.5
x2

−0.5

−1

−1.5
v1
−2
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2

x1
The red trajectory corresponds to the eigenvector v1. The behavior of trajec-
tories x(t) in the transformed space (x̄1, x̄2) is the following:
Stable degenerate node
2

1.5

0.5

v1
x̄2

−0.5

−1

−1.5

−2
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2

x̄1
Trajectories of this type are called: “Stable degenerate node”.

Zanasi Roberto - System Theory. A.A. 2015/2016

Capitolo 2. CANONICAL FORMS 2.6

Case III. Choosing M = 1, K = 2 and b = 2 one obtains the following

system:     
 0 1 0
ẋ(t) = x(t) + u(t)
(4) −2 −2 1

y = [ 1 0 ] x(t)
The characteristic polynomial is now the following:
 
λ −1
det(λI − A) = det = λ2 + 2λ + 2 = (λ + 1)2 + 1
2 λ+2
Matrix A has two complex conjugate eigenvalues λ1,2 = −1 ± j to which
corresponds two complex conjugate eigenvectors v1,2. The eigenvector v1 can
be determined solving the following homogeneous system:
   
1−j 1 1+j
(A − λ1I)v1 = v1 = o → v1 =
−2 −1 − j −2
Using a state space transformation x = Tx based on the complex conjugate
eigenvectors v1,2, it is possible to “diagonalize” the matrix A:
   
1+j 1−j −1 −1 + j 0
T = [v1 v2] = , A = T AT =
−2 2 0 −1 − j
The obtained transformed system is complex and difficult to use. In these cases
it is preferable to use the “real Jordan form”.
In this case the columns of the transformation matrix T are equal to the real
part w1 and the imaginary part w2 of the previously calculated eigenvector v1:
       1

1 1 1 1 0 −
v1 = +j T = [w1 w2] = , T−1 = 2
−2 0 −2 0 1 12
| {z } | {z }
w1 w2
The obtained transformed matrix A has the following structure:
 1
    
0 − 0 1 1 1 −1 1
A = T−1AT = 2 =
1 12 −2 −2 −2 0 −1 −1
which is a particular value of the following general case:
 
σ ω
A=
−ω σ

Zanasi Roberto - System Theory. A.A. 2015/2016

Capitolo 2. CANONICAL FORMS 2.7

The free evolution of system (4) starting from the initial condition x0 is:
−1
x(t) = eAtx0 = eTAT tx0 = TeAtT−1x0
 σt σt

e cos ωt e sin ωt
= T σt σt T−1x0
−e sin ωt e cos ωt
In this case it is λ1 = σ + jω = −1 + j and the following solution is obtained:
   −t  
1 1 e cos t e−t sin t 0 − 12
x(t) = x0
−2 0 −e−t sin t e−t cos t 1 12
 −t 
e (cos t + sin t) e−t sin t
= x0
−2e−t sin t e−t(cos t − sin t)
The behavior of trajectories x(t) in the state space (x1, x2) is the following:

1
Stable focus
0.8

0.6

0.4

0.2
x2

−0.2

−0.4

−0.6

−0.8

−1
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1

x1
The trajectories tend to the origin with a spiral behavior. Trajectories of this
type are referred to as: “Stable focus”.
Computation of eAt in Matlab:
A= [0 1; ... +- -+
-2 -2]; | exp(-t) (cos(t) + sin(t)), exp(-t) sin(t) |
syms t; --> | |
Eat=simplify(expm(A*t)); | -2 exp(-t) sin(t), exp(-t) (cos(t) - sin(t)) |
pretty(Eat) +- -+

Zanasi Roberto - System Theory. A.A. 2015/2016

Capitolo 2. CANONICAL FORMS 2.8

The behavior of trajectories x(t) in the transformed state space is the following:
1
Stable focus
0.8

0.6

0.4

0.2
x̄2

−0.2

−0.4

−0.6

−0.8

−1
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1

x̄1
————–
The trajectories of a second order system having a negative eigenvalue λ1 < 0
and a positive eigenvalue λ2 > 0 are called “Saddle” and have the following
behavior:
2
Saddle

1.5

0.5
v2
x2

−0.5

−1

v1
−1.5

−2
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2

x1
The two red trajectories correspond to the eigenvectors v1 and v2 associated
to the two eigenvalues λ1 and λ2.
Zanasi Roberto - System Theory. A.A. 2015/2016
Capitolo 2. CANONICAL FORMS 2.9

• Typical behavior of trajectories belonging to an “Unstable focus”:

2
Unstable focus

1.5

 
1 −1 1

A=
1 1 0.5
x2
0

−0.5

λ1,2 = 1 ± j −1

−1.5

−2
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2

x1

• Typical behavior of trajectories belonging to an “Unstable node”:

1
Unstable node
0.8

0.6
 
1 0
A= 0.4
v1
0 2 0.2
x2

−0.2

−0.4

λ1 = 1, λ2 = 2 −0.6

−0.8
v2
−1
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1

x1

In this case the dominant eigenvalue is λ2 = 2 and therefore all the trajectories
tend to become parallel to the eigenvector v2.

Zanasi Roberto - System Theory. A.A. 2015/2016

Capitolo 2. CANONICAL FORMS 2.10

Free trajectories in a three dimensional space.

The free trajectories of a linear system having a real eigenvalue λ3 = −3 which

is dominant with respect to the other two complex eigenvalues λ1,2 = −5±5 j:

6 v3
 
−5 5 0 5

A =  −5 −5 0  C
4
0 0 −3
3
z

λ1,2 = −5 ± 5j 1

0
λ3 = −3 5 5

0 0

−5 −5
y x
All the trajectories tend towards the dominant eigenvector v3 = [0 0 1]T . The
free evolution of the “output” signal y(t) is:
Evoluzione libera. C=[ −1 1 1 ]
14

12

10

y(t) = c x(t) 8
y

  4
c= −1 1 1
2

−2
0 0.2 0.4 0.6 0.8 1 1.2 1.4
t

All the free evolutions of the “output” signal y(t) have an “aperiodic” behavior
which is determined by the dominant eigenvalue λ3 = −3.
Zanasi Roberto - System Theory. A.A. 2015/2016
Capitolo 2. CANONICAL FORMS 2.11

The free trajectories of a linear system having two complex eigenvalues λ1,2 =
−1 ± 5 j dominant with respect to the third eigenvalue λ3 = −3 are:

6 v3
 
−1 5 0 5

A =  −5 −1 0  4 C
0 0 −3
3
z

λ1,2 = −1 ± 5j 1

λ3 = −3 0
4 4
2 2
0 0
−2 −2
−4 −4

y x
All the trajectories tend towards the plane (x, y) determined by the two com-
plex eigenvectors v1 and v2 associated to the eigenvalues λ1,2 = −1 ± 5 j. In
this case the free evolutions of the output signal y(t) are:
Evoluzione libera. C=[ −1 1 1 ]
14

12

10

y(t) = c x(t) 8

6
y

4
 
c= −1 1 1 2

−2

−4
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
t

All the free evolutions of the output signal y(t) have a stable decreasing oscil-
latory behavior determined by the two complex eigenvalues λ1,2 = −1 ± 5 j.
The state space trajectories have been projected along the c vector.
Zanasi Roberto - System Theory. A.A. 2015/2016