Nonlinear Systems Notes ( EE 553)

Lecture 3
K. Hariche
IGEE/UMBB

1 Introduction
In these few next lectures, we will consider second order nonlinear systems. The
reason for it is that a very large class of dynamic physical systems belong to this
class. Furthermore, the dynamic behaviour of these systems can be represented
graphically on a plane ( phase plane analysis )
We will review second order linear systems and extend the discussion to
second order nonlonear systems.

2 Phase plane analysis concepts

Consider the second order nonlinear system described by the state space equa-
dx1
tion dt = f1 (x1 ; x2 ); x1 (0) = x10 ; The phase plane ( state plane) of the
dx2
dt = f2 (x1 ; x2 ); x2 (0) = x20
system is the 2 dimensional plane x2 (t) versus x1 (t) where state trajecto-
ries are plotted.
x1 (t)
A state trajectory is a collection of points in the state plane as
x2 (t)
x1 (0)
t varies over 0 1 starting from some initial point ; Clearly a
x2 (0)
state trajectory is a curve in the x2 versus x1 that can be de…ned by x2 = F (x1 ).
Each trajectory , in fact, corresponds to a particular solution of the state space
x1 dx2
equation. The slope of the trajectory at is then s(x1 ; x2 ) = dx =
x2 1
dx2
f2 (x1 ;x2 )
dt
dx1 = f1 (x1 ;x2 ) .
dt
Every state trajectory corresponds to a given initial state ( initial condition
) and for a "su¢ cient" number of initial points, we obtain a collection of state
trajectories describibng the dynamic behaviour of the system called a state
portrait.
An equilibrium point of the system can be interpreted as a particular state
trajectory: if the state is initiated at an equilibrium point it remainn there
forever under no external force. If a system reaches an equilibrium point, it will
also remain there.

1
3 Second order linear systems
The idea is to have a grasp on the type of state trajectories and phase por-
traits encountered when linearizing a nonlinear system around some point (
equilibrium point ).
Consider the second order linear system dX dt = AX ; X(0) = X0 where
a11 a12 x1
A = and X = : Clearly an equilibrium point of the
a21 a22 x2
system satis…es AX = 0 meaning that any equilibrium point Xe of the system
is a vector of the null space ( Kernel ) of A i.e Xe 2 N (A): If A is nonsingular,
then the unique equilibrium point is the origin X = 0:

The solution of the state sapce equation is straightforward in this case and
is given by X(t) = eAt X(0) but we are interested in state trajectories, i.e curves
x2 = F (x1 ) in the x2 versus x1 plane.
To do so, we need to eliminate time t from the solution to get a relationship
x2 versus x1 . This can be cumbersome if we use original coordinate system. To
simplify things, we convert the original system to diagonal or Jordan form.
From linear algebra, the matrix A can be converted to diagonal form if A
has a complete set of linearly independent eigenvectors and to Jordan form
otherwise.
If M is the modal matrix ( whose columns are linearly independent vectors
) , we can use the similarity transformation X = M 1 X to obtain the new
state space equation dX dt = X ;X(0) = X 0 where = M 1 AM is either in
1 0 1
diagonal or Jordan form i.e in the form = or = :
0 2 0
1 0
Consider …rst the diagonal form = : The eigenvalues may be
0 2
repeated ( 1 = 2 = ) or distinct. If distinct, they may real or complex
conjugate.
case1: 1 ; 2 real distinct.
dx1
The new system is then dt = 1 x1 ; x1 (0) = x10 whose solution is
dx2
dt = 2 x2 ; x2 (0) = x20
x1 (t) = x10 e 1 t
; Eliminating time t we can obtain a relationship x2 versus
x2 (t) = x20 e 2 t
x1 : let us assume for the moment that 1 ; 2 6= 0: We have e 1 t = xx10 1
or 1 t =
1
x1 1 x1 x1 1
ln x10 and t = 1
ln x10 = ln x10 and upon plugging in the expression of
2

x2 , we get x2 = x20 xx10

1
1
. This expression de…nes a collection of trajectories
in the x2 versus x1 plane, each of them corresponding to a particular initial
condition. Clearly the type of trajectory depends on the sign of the exponent
2
1
. The exponent is positive if 1 ; 2 have the same sign and negative if they
have opposite signs.

If 1; 2 have the same sign with 1; 2 0 , we obtain a collection of

2
 trajectories ( parabola like ) converging to the origin. In this case, the
origin is said to be a stable node. The direction of the trajectory can
be determined from dx dt = 1 x1 : for x1
1
0 ( left half plane ) we have
dx1
dt 0 , hence x1 increases toward the origin and for x1 0 ( right half
plane ) we have dxdt
1
0 hence x1 decreases toward the origin
If 1 ; 2 have the same sign with 1 ; 2 0 , we obtain a collection of
trajectories of the same shape but diverging from the origin and the origin
is called an unstable node. The direction of the trajectory is determined
in the same way as above.
If 1 ; 2 have opposite signs sign , we have a collection of curves ( hyper-
bola type) approaching the origin, then departing from it without reaching
it and the origin is said to be a saddle.

case2: 1 ; 2 real repeated

Clearly this results in the expression .x2 = x20 xx10 1
which indicates that
the trajectories are straight lines converging to the origin if 1 = 2 0 ( stable
node) or departing from the origin if 1 = 2 0 ( unstable node).
Remark: We have assumed 1 ; 2 6= 0: If 1 = 0; 2 6= 0 we have
x1 (t) = x10
meaning a collection of vertical lines. On the other hand if
x2 (t) = x20 e 2 t
If 1 6= 0; 2 = 0 , we obtain a collection of horizontal lines.
1 = + j!
case 3: 1 ; 2 is a complex conjugate pair i:e
1 = j!
To avoid manipulating complex numbers we use the equivalent representa-
dx1
! dt = x1 + wx2 ; x1 (0) = x10
tion = leading to dx2
! dt = !x1 + x2 ; x2 (0) = x20
r2 = x1 2 + x2 2
Using polar coordinates , we obtain the new decoupled
tan = xx21
dr
dt = r :whose solution is r(t) = r(0)e t
system d :
dt = ! (t) = !t + (0)
If 0 : the trajectory rotates around the origin with increasing radius,
thus diverging from it and the origin is called an unstable focus.
If 0 : the trajectory rotates around the origin with decreasing radius,
thus converging to it and the origin is called a stable focus.
If = 0 : the trajectories are a collection of circles with center the origin.
The latter is called a center.
1
Consider now the Jordan form = leading to the system
0
dx1
dt = x1 + x2 ; x1 (0) = x10 ; Solving and eliminationg time t will lead to
dx2
dt = 2 x2 ; x2 (0) = x20
an expression that is a bit cumbersome. The trajectories are characteristic of a
node: if 0 the trajectories converge to the origin and the latter is a stable
node, otherwise for 0 the trajectories depart from the origin and the latter
is an unstable node.

3
 In summary , for a second order linear system, the state trajectories are
characteristic of :

1. A stable node for 1 0; 2 0

2. An unstable node for 1 0; 2 0
3. A saddle for 1 2 0
4. A stable focus for 0
5. An unstable focus for 0
6. A center for =0

The eigenvalues 1 ; 2 determine then the nature of the origin ( equilibrium

point ) and the shape of the trajectories around it. The following shows the
types of trajectories encountered:

4
4 Second order nonlinear systems
dx1
= f1 (x1 ; x2 ); x1 (0) = x10
dt
Consider the second order nonlinear system dx2
= f2 (x1 ; x2 ); x2 (0) = x20
dt
and assume, for simplicity purposes , that the equilibrium point is the origin
X = 0:
If the functions f1 ; f2 are di¤erentiable we can expand them in Taylor’s series
near the origin as:
2 2
@f1 @f1
f1 (x1 ; x2 ) = f1 (0; 0)+ @x1
(0; 0)x1 + @x2
(0; 0)x2 + @@xf21 (0; 0)x21 + @@xf21 (0; 0)x22 +
1 2
@ 2 f1
@x1 @x2 (0; 0)x1 x2 +
2 2
@f2 @f2
f2 (x1 ; x2 ) = f2 (0; 0)+ @x1
(0; 0)x1 + @x2
(0; 0)x2 + @@xf22 (0; 0)x21 + @@xf22 (0; 0)x22 +
1 2
@ 2 f2
@x1 @x2 (0; 0)x1 x2 +
Separating linear and nonlinear parts we have:
@f1 @f1
f1 (x1 ; x2 ) = f1 (0; 0) + @x 1
(0; 0)x1 + @x 2
(0; 0)x2 + R1 (x1 ; x2 )
@f2 @f2
f2 (x1 ; x2 ) = f2 (0; 0) + @x1 (0; 0)x1 + @x2 (0; 0)x2 + R2 (x1 ; x2 )
Clearly if we operate "su¢ ciently close" to the origin the remainder terms
R1 (x1 ; x2 ); R2 (x1 ; x2 ) will be negligeable and we can approximate the system
by the
( linear model
dx1 @f1 @f1
dt ' f1 (0; 0) + @x1 (0; 0)x1 + @x2 (0; 0)x2 ; x1 (0) = x10
dx2 @f @f2
dt = f2 (0; 0) + @x1 (0; 0)x1 + @x2 (0; 0)x2 ; x2 (0) = x20
2

5
 Now f1 (0; 0)(= f2 (0; 0) = 0 since the origin is assumed to be an equilibrium
dx1 @f1 @f1
point leading to dt ' @x1 (0; 0)x1 + @x2 (0; 0)x2 ; x1 (0) = x10 and in matrix
dx2 @f2 @f2
dt = @x1 (0; 0)x1 + @x2 (0; 0)x2 ; x2 (0) = x20
form !
dx1 @f1 @f1
@x1 (0; 0) @x2 (0; 0) x1
dt
dx2 = @f2 @f2 or dXdt = AX where we
dt @x1 (0; 0) @x2 (0; 0) x2
!
@f1 @f1
@x1 (0; 0) @x2 (0; 0) f1
recognize A = @f2 @f2 as the Jacobian of f = eval-
@x1 (0; 0) @x2 (0; 0)
f2
uated at the equilibrium poinf X = 0:
Using this linearized model, we can use the eigenvalues of A to determine
the nature of the equilibrium point X = 0 and the shape of the trajectories in
a small neighborhood around it. Clearly, unlike the linear case, the conclusions
that can be made are only local in the vicinity of X = 0:
This indirect approach ( linearization ) is possible only if f1 ; f2 are smooth
enough making the approach not applicable for hard nonlinearities.
It must be pointed out that the location of the eigenvalues 1 ; 2 determines
the nature of the equilibrium point X = 0 for all cases except for the case of
1 = +j!
a center where they are a pure conjugate pair i.e . The pair is
2 = j!
located exactly on the j! axis as a result of an approximation ( neglecting
remainders ); For the nonlinear system , we have no idea and we cannot conclude.