NCTU Department of Electrical and Computer Engineering
Senior Course
By Prof. Yon-Ping Chen
8-1
8. Phase Plane Method
A. Phase Plane
B. Natural Modes of LTI Systems
C. Phase Portraits
Phase plane method is an important graphical methods to deal with problems
related to a second-order autonomous system.
A. Phase Plane
First, let’s introduce the basic concepts of the phase plane by a second-order
autonomous system which is mathematically expressed as below:
( ) ( ) ( )( )txtxftx 2111 ,=& , ( ) 101 0 xx = (8-1)
( ) ( ) ( )( )txtxftx 2122 ,=& , ( ) 202 0 xx = (8-2)
without any inputs. It is known that the system states x1(t) and x2(t) are uniquely
determined by initial values ( ) 101 0 xx = and ( ) 202 0 xx = . Clearly, different initial
conditions ( ) 101 0 xx = and ( ) 202 0 xx = will result in different states x1(t) and x2(t) for
t>0. If we adopt the values of the two state variables as two axes, then a coordinate is
formed in Figure 8-1, called the phase plane. We can draw all the points (x1(t), x2(t))
for t≥0 to obtain a trajectory with respect to specified initial conditions ( )01x and
( )02x . Figure 8-1 shows an example of the trajectory (x1(t),x2(t)) for 0
By Prof. Yon-Ping Chen
8-2
Since each trajectory is uniquely determined by a set of initial conditions, that
means no intersection exists among all the trajectories, except at a singular point
(x1e,x2e) satisfying ( ) 01 =tx& and ( ) 02 =tx& for all t, i.e.,
( )txxf ee ,, 2110 = (8-3)
( )txxf ee ,, 2120 = (8-4)
In fact, (x1e,x2e) is an equilibrium point and not necessary to be unique. A set of
trajectories obtained from different initial conditions is called a phase portrait, as
shown in Figure 8-2, where P is an equilibrium point or a singular point.
B. Natural Modes of LTI Systems
The natural modes reflects the intrinsic properties of systems which are not
excited by any external inputs. In other words, the natural modes are the system
behavior caused only by its initial conditions. Let’s employ the simplest 2nd order
dynamic system as an example to explain how to determine the natural modes of an
LTI system. Consider the following 2nd order LTI system without any input:
( ) ( ) ( )txatxatx 2121111 +=& , x1(0)=x10 (8-5)
( ) ( ) ( )txatxatx 2221212 +=& , x2(0)=x20 (8-6)
or in a matrix form
⋅= tt xAx& , 0 0xx = (8-7)
x1
x2
Figure 8-2
t=0
P
NCTU Department of Electrical and Computer Engineering
Senior Course
By Prof. Yon-Ping Chen
8-3
where
=
2221
1211
aa
aa
A ,
( )
( )
=
0
0
0
2
1
x
x
x and
=
20
10
0
x
x
x . Clearly, the origin (0,0) in
the phase plane is an equilibrium point since the truth of 0x =
t leads to 0x =
t& .
From (1-8), the state vector can be solved as
0xx
A tet =
(8-8)
with te A defined as
LL
!n
t
!
t
te
nn
t AAAIA ++++=
2
22
(8-9)
a form similar to LL
!! n
xx
xe
n
x ++++=
2
1
2
. It is known that any square matrix A
can be transformed into a matrix J in diagonal form or Jordan form, expressed as
1−
= VJVA (8-10)
where V is invertible. It is easy to show that
1−
= VVJA nn n=0,1,2,… (8-11)
There are three typical forms of J, denoted as
=
2
1
0
0
λ
λJ ,
=
r
r
λ
λ
0
1J or
=
∗
c
c
λ
λ
0
1J (8-12)
where the disgonal terms are the eigenvalues of A and can be solved from its
characteristic equation
0
2221
1211
=
−−
−−
=−
aa
aa
λ
λλ AI (8-13)
That means they are the solutions of
( ) 02112221122112 =−++− aaaaaa λλ (8-14)
and both of them may be real or complex in conjugate and may be distinct or repeated.
Accordingly, there are three cases of the eigenvalues:
Case-1 distinct real eigenvalues, 21 λλ ≠
Case-2 repeated real eigenvalues, rλλλ == 21
Case-3 complex eigenvalues in conjugate, ωαλ jc += and ωαλ jc −=∗ .
It is known that for an eigenvalue λi there exists at least one eigenvector vi such that
iii vAv λ= (8-15)
NCTU Department of Electrical and Computer Engineering
Senior Course
By Prof. Yon-Ping Chen
8-4
For Case-1, based on (8-15) we have
=
=
222
111
vAv
vAv
λ
λ
(8-16)
which results in
[ ] [ ]
43421
4342143421
JV
vv
V
vvA
=
2
1
2121 0
0
λ
λ
(8-17)
Note that J is the first typical form in (8-12) and V is invertible and called the
eigenvectormatrix since it contains all the eigenvectors of A. Clearly, (8-17) can be
rearranged as 1 −= VJVA , a form given in (8-10). Besides, we have
1
2
11
0
0
−−
== VVVVJA
n
n
nn
λ
λ
(8-18)
Hence, the state vector in (8-8) is obtained as
0
022
2
2
22
1
1
0
22
0
22
0
2
1
0
0
2
10
0
2
1
2
2
xVV
xVV
xVJJJIV
x
AAAI
xx
1
1
1
A
−
−
−
=
+++
+++
=
++++=
++++=
=
t
t
nn
nn
t
e
e
!
t
t
!
t
t
!n
t
!
t
t
!n
t
!
t
t
et
λ
λ
λλ
λλ
L
L
LL
LL
(8-19)
or in the following form
( ) tt eetx 21 12111 λλ αα += (8-20)
( ) tt eetx 21 22212 λλ αα += (8-21)
where α11, α12, α21 and α22 depend on x1(0) and x2(0). The functions te 1λ and te 2λ are
called the natural modes of this system.
For Case-2 of repeated eigenvalue λr, we may obtain one or two eigenvectors
correspondingly. Let’s assume only one eigenvector is found, then we have
rrr vAv λ= (8-22)
NCTU Department of Electrical and Computer Engineering
Senior Course
By Prof. Yon-Ping Chen
8-5
It is known that we can create a new vector rv′ which satisfies
rrrr vvvA +′=′ λ (8-23)
and most importantly rv and rv′ are independent. As a result, wwe attain
[ ] [ ]
43421
4342143421
JV
vv
V
vvA
′=′
r
r
rrrr λ
λ
0
1
(8-24)
Note that J is the second typical form in (8-12) and V is invertible since rv and rv′
are independent. Clearly, (8-24) can be rearranged as 1 −= VJVA , same as (8-10).
Besides, we have
1
1
1
0
−
−
−
== VVVVJA
n
r
n
r
n
rnn n
λ
λλ
(8-25)
Hence, the state vector in (8-8) is obtained as
0
022
32
2
22
0
22
0
22
0
0
2
10
2
0
2
1
2
2
xVV
xVV
xVJJJIV
x
AAAI
xx
1
1
1
A
−
−
−
=
+++
+++++++
=
++++=
++++=
=
t
tt
r
r
r
r
r
r
nn
nn
t
r
rr
e
tee
!
t
t
!
t
tt
!
t
t
!n
t
!
t
t
!n
t
!
t
t
et
λ
λλ
λλ
λλλλ
L
LL
LL
LL
(8-26)
or in the following form
( ) tt rr teetx λλ βγ 111 += , (8-27)
( ) tt rr teetx λλ βγ 222 += , (8-28)
where γ1, γ2, β1 and β2 depend on x1(0) and x2(0). The functions treλ and trteλ are
called the natural modes of this system.
For Case-3 of complex eigenvalues in conjugate, we will skip the derivation and
just show the resulted state variables which are
( ) tsineBtcoseAtx tt ωω αα 111 += (8-29)
( ) tsineBtcoseAtx tt ωω αα 222 += (8-30)
NCTU Department of Electrical and Computer Engineering
Senior Course
By Prof. Yon-Ping Chen
8-6
where A1, A2, B1 and B2 depend on x1(0) and x2(0). The functions tcose t ωα and
tsine t ωα are called the natural modes of this system. Note that if α=0, i.e., the
eigenvalues are pure imaginary, then the natural modes are tcosω and tsinω .
It is clear that all the natural modes depend on the eigenvalues of the system
matrix A. If an eigenvalue is located on the left half complex plane, then the natural
mode corresponding this eigenvalue will converge to 0 as t→∞. On the other hand, if
an eigenvalue is located on the right half complex plane, then the related natural mode
will increase to ∞ as t→∞. As for the eigenvalue on the imaginary axis, its natural
mode will oscillate.
C. Phase Portraits
Now, let’s use some examples to draw phase portraits, a set of trajectories, and
discuss their related properties. Actually, different natural modes will result in
different trajectories and different phase portraits as well.
Consider the second-order LTI system in canonical form, which is described as
below:
( ) ( )txtx 21 =& , x1(0)=x10 (8-31)
( ) ( ) ( )txatxatx 2221212 +=& , x2(0)=x20 (8-32)
The system matrix is
=
2221
10
aa
A and the characteristic equation is
0
1
2122
2
2221
=−−=
−−
−
=− aa
aa
λλλ
λλ AI (8-33)
whose roots are
2
4 21
2
2222
21
aaa +±
=λλ , (8-34)
If 04 21
2
22 >+ aa , then λ1 and λ2 are real and distinct. Without loss of generality,
assume λ1>λ2, then there are three cases listed as below:
I. λ2<λ1<0, i.e., 04 21222 >+ aa , a22<0 and a21<0.
II. λ1>λ2>0, i.e., 04 21222 >+ aa , a22>0 and a21<0.
NCTU Department of Electrical and Computer Engineering
Senior Course
By Prof. Yon-Ping Chen
8-7
III. λ1>0>λ2, i.e., 04 21222 >+ aa and a21>0.
Their phase portraits are depicted in Figure 8-3(a)(b)(c), respectively. From (8-20),
the first state variable ( )tx1 is
( ) tt eetx 21 12111 λλ αα += (8-35)
and from (8-31), the second state variable is
( ) ( ) tt eetxtx 21 21211112 λλ λαλα +== & (8-36)
For some initial conditions x1(0) and x2(0), the coefficient α12 in (8-35) may be equal
to 0, then we have
( ) tetx 1111 λα= (8-37)
( ) ( )txetx t 111112 1 λλα λ == (8-38)
which implies
( ) ( ) 0112 =− txtx λ (8-39)
i.e., the related trajectory is the straight line L1 shown in Figure 8-3(a)(b)(c). Similarly,
if initial conditions result in α11=0, then
( ) tetx 2121 λα= (8-40)
( ) ( )txetx t 122122 2 λλα λ == (8-41)
which implies
( ) ( ) 0122 =− txtx λ (8-42)
i.e., the related trajectory is the straight line L2 shown in Figure 8-3(a)(b)(c).
In Case-I with λ2<λ1<0, its phase portrait is depicted in Figure 8-3(a). As t→∞,
due to the fact that 021 →>> tt ee λλ and 02 →teλ , all the trajectories except L2 in
the phase portrait must satisfy
( ) tetx 1111 λα→ (8-43)
( ) ( )txetx t 111112 1 λλα λ =→ (8-44)
i.e., ( ) ( ) 0112 →− txtx λ . In other words, all the trajectories except L2 in the phase
portrait will approach L1 as t→∞. Clearly, it is a stable system since the system state
will converge to the equilibrium point (0,0) as t→∞.
NCTU Department of Electrical and Computer Engineering
Senior Course
By Prof. Yon-Ping Chen
8-8
In Case-II and Case III, their phase portraits are depicted in Figure 8-3(b) and
Figure 8-3(c). As t→∞, due to the fact that tt ee 21 λλ >> and ∞→te 1λ , all the
trajectories except L2 in the phase portrait must satisfy
( ) tetx 1111 λα→ (8-45)
( ) ( )txetx t 111112 1 λλα λ =→ (8-46)
i.e., ( ) ( ) 0112 →− txtx λ . In other words, all the trajectories except L2 in the phase
portrait will approach L1 as t→∞. Clearly, it is an unstable system since the system
state will move to (∞, ∞) or (−∞, −∞) as t→∞, except L2 in Case-II. However, Case-II
is still treated as unstable because any extremely small disturbance will drive the
system to leave L2 and move to (∞, ∞) or (−∞, −∞).
If 04 21
2
22 =+ aa , then from (8-34) we have 2
22
21
a
== λλ . There are two cases
listed as below:
IV. λ2=λ1<0, i.e., 04 21222 =+ aa and a22<0.
x1
(a) (b)
Figure 8-4
x1
x2 x2
x1
x2
x1
(a) (b) (c)
Figure 8-3
L2
L1
x1
x2
L1
L2 x2 L1 L2
NCTU Department of Electrical and Computer Engineering
Senior Course
By Prof. Yon-Ping Chen
8-9
V. λ1=λ2>0, i.e., 04 21222 =+ aa and a22>0.
Both phase portraits are depicted in Figure 8-4(a)(b), respectively. In fact, they are
special cases of Case-I and Case-II when L1 and L2 are merged together. From the
phase portraits, it is easy to conclude that Case-IV is stable and Case-V is unstable.
If 04 21
2
22 <+ aa , then λ1 and λ2 are complex in conjugate. There are two cases
listed as below:
VI. λ2,λ1=±jω where ω>1, ω=1 or ω<1.
VII. λ2,λ1=α±jω where α>0 or α<0.
Their phase portraits are depicted in Figure-5(a)(b)(c) and Figure-6(a)(b). For Case-VI,
from (8-29) with α=0, the first state variable is
( ) ( )θωωω +=+= tcosCtsinBtcosAtx 111 (8-47)
and from (8-31) the second state variable is
( ) ( ) ( )θωω +−== tsinCtxtx 12 & (8-48)
where C is determined by the initial conditions x1(0) and x2(0). Clearly, it is an
oscillatory system. From (8-47) and (8-48), we have
( ) ( ) 22
2
22
1 C
tx
tx =+
ω
(8-49)
whose phase portraits are ellipses or circles, shown in Figure 8-5(a)(b)(c) respectively
for ω>1, ω=1 and ω<1. Note that the arrow in these figures is pointing from left to
right in the region x2(t)>0 since ( ) ( ) 021 >= txtx& , i.e., ( )tx1 is increased as t increases,
and from right to left in the region x2(t)<0 since ( ) ( ) 021 <= txtx& , i.e., ( )tx1 is
decreased as t increases. While passing through the ( )tx1 axis, i.e., ( ) ( ) 021 == txtx& ,
the arrow goes down or up vertically.
x1
x2
x1
x2
x1
x2
(a) (b) (c)
Figure 8-5
NCTU Department of Electrical and Computer Engineering
Senior Course
By Prof. Yon-Ping Chen
8-10
The above description of the arrow direction is also suitable for Case-I to
Case-V. For Case-VII, from (23), the first state variable is
( ) ( ) ( )θωωω αα +=+= tcoseCtsinBtcosAetx tt 111 (8-50)
and from (8-31) the second state variable is
( ) ( ) ( ) ( )
( )θω
θωαθωω
α
αα
′+′=
+++−==
tcoseC
tcoseCtsinCetxtx
t
tt
12 &
(8-51)
Clearly, it is a divergent oscillatory system for α>0 as shown in Figure 8-6(a) and an
attenuated oscillatory system for α<0 as shown in Figure 8-6(b).
Unlike an LTI system, the phase portrait of a nonlinear system is often more
complicated since it may have multiple equilibrium points or limit cycles. However,
the local behavior of a continuous nonlinear system around an equilibrium point can
be determined by linearization which results in an approximted LTI system near to the
equilibrium point. If the approximated LTI system is stable then the equilibrium point
is stable, otherwise it is unstable.
Now, consider a nonlinear system and determine its stability by the use of phase
plane method. Let the state equations be
( ) ( ) ( ) ( ) ( )( )
( ) ( ) ( ) ( ) ( )( )
−++−=
−++=
1
1
2
2
2
1212
2
2
2
1121
txtxtxtxtx
txtxtxtxtx
&
&
(8-52)
If ( ) ( ) 012221 =−+ txtx , then it is easy to solve the state variables as ( ) ( )θ+= tcostx1
and ( ) ( )θ+−= tsintx2 . In other words, if the system is initially located on the circle
( ) ( ) 012221 =−+ txtx , then the system state will keep in a circular trajectory. While the
initial condition is a little deviated from the circle, for example ( ) ( ) ε=−+ 12221 txtx ,
x1
(a) (b)
Figure 8-6
x1
x2 x2
NCTU Department of Electrical and Computer Engineering
Senior Course
By Prof. Yon-Ping Chen
8-11
its behavior is then governed by
( ) ( ) ( )
( ) ( ) ( )
+−=
+=
txtxtx
txtxtx
212
211
ε
ε
&
&
(8-53)
whose system matrix is
− ε
ε
1
1
. Correspondingly, its characteristic equation is
0122 =+− ελλ (8-54)
and the eigenvalues are 1
2
442 2
21 j±≈
−±
= ε
εελλ , . Clearly, for ε>0 the system
is in the region outside the circle and becomes unstable, i.e., it moves away the circle
to the infinity. For ε<0 the system is in the region inside the circle and becomes stable,
i.e., it moves away the circle to the origin. Figure 8-7 shows the phase portrait
including a limit cycle ( ) ( ) 012221 =−+ txtx .
Next, let’s consider a variable structure system, described by the following state
equations
( ) ( )
( ) ( ) ( )( )( ) ( )
+−=
=
txtxtxsigntx
txtx
1212
21
5152
..&
&
(8-55)
It is obvious that there are two structures of the system, one for ( ) ( ) 021 >txtx and the
other for ( ) ( ) 021 txtx , i.e., in the first and third quadrants of the phase plane, the system
is governed by
x1
x2
( ) ( ) 012221 =−+ txtx
Figure 8-7
NCTU Department of Electrical and Computer Engineering
Senior Course
By Prof. Yon-Ping Chen
8-12
( ) ( )
( ) ( )
−=
=
txtx
txtx
12
21
4
&
&
(8-56)
which performs as an ellipse. While ( ) ( ) 021 > % key in the following instructions
>> [t,x]=ode45(@vss,[0:0.01:10],[10 0])
>> plot(x(:,1),x(:,2)); xlabel(‘x1’); ylabel(‘x2’); grid
-5 0 5 10
-10
-5
0
5
x1
x
2
==========================================
It can be concluded that the variable structure system is stable, even though both of its
sub-systems are unstable.
Problems
P.8-1 Consider the following system:
( ) ( )txtx 21 =& , x1(0)=5
( ) ( ) ( )txtxtx 212 4 −−=& , x2(0)=0
NCTU Department of Electrical and Computer Engineering
Senior Course
By Prof. Yon-Ping Chen
8-13
Draw the trajectory on the phase plane from t=0 to 10 in MATLAB.