Nonlinear Systems Notes ( EE 553)

Lecture 4
K. Hariche
IGEE/UMBB

1 Review
dx1
= f1 (x1 ; x2 )dt
Given a second order nonlinear system and if f1 ; f2 are
dx2
= f2 (x1 ; x2 )dt
smooth enough , we can linearize it around an equilibrium point. The eigenval-
ues of the Jacobian determine the nature of the equilibrium point in all cases
except for the case of a center ( pure complex conjugate pair ).
dx1
dt = x2
Example 1 Consider dx2 ; Clearly the unique equi-
dt = x1 + x2 (1 x1 2 )
0 f1
librium point is Xe = : The Jacobian of f = is then JX f (X) =
0 f2
0 1 0 1
and A = JX f (0) = whose eigenvalues
1 2x1 x2 1 x1 2 1 1
p
1 j 3
are 1;2 = 2 indicating that the origin is an unstable focus.
dx1
x2 x1 (x21 + x22 )
dt =
Example 2 Consider now dx2 and the equilibrium
x1 x2 (x21 + x22 )
dt =
0 0 1
point Xe = : Here A = JX f (0) = with eigenvalues
0 1 0
1;2 = j . While the linear model indicates a center we cannot conclude for
the nonlinear system.

2 Features of the indirect approach

f (X) = f (Xe ) + JX f (Xe ):[X Xe ] + R(X; Xe )

where R(X) is the remainder term containing higher order terms which are
negligeable if X is "close enough" to Xe :
It follows that f (X) ' f (Xe )+JX f (Xe ):[X Xe ] or dX
dt ' JX f (Xe ):[X
e dXe e or dXe ' AX: e
Xe ] and if we let X = X Xe we obtain dt ' JX f (Xe )X dt
Analyzing the indirect approach, we can see some fundamental limitations
in the approach:

1
 The approach is applicable only for " smooth nonlinearities" ,i.e cases
where f1 ; f2 are continuously di¤erentiable.
Even when applicable, the conclusions that can be made are only local,
that is valid only in the close neighborhood of the equilibrium point. This
limits the analysis to only small perturbations about the equilibrium point.
It gives no hint on the behaviour outside this small region.
No conclusion can be made if the linear model indicates a center ( pure
conjugate pair of eigenvalues ).
It must also be noticed that we have as many linearized models as existing
equilibrium points in the system.
The above calls for an alternative approach to overcome these limitations.

3 Alternative approaches
dx1
dt = f1 (x1 ; x2 ); x1 (0) = x10
Given the second order nonlinear system dx2 ,
dt = f2 (x1 ; x2 ); x2 (0) = x20
other alternative approaches may be:
Analytic solution of the di¤erential system
Numerical solution
Graphical solution
In most practical cases, we cannot obtain an analytic or exact solution of
the di¤erential system because of its nonlinear nature. The numerical solution
consists in applying some numerical techniques to solve the set of nonlinear
di¤erential equations. Most of these techniques convert the system into a set
of di¤erence equations that can be solved iteratively starting at some initial
condition. The graphical approach is similar to the numerical approach but
the idea here is to construct state trajectories or state portraits using numrical
techniques.
As an example of a numerical method used to construct trajectories and
state portraits, consider the well-known Euler’s method. The idea here is
to approximate the derivative dx dt by a …rst di¤erence to convert a di¤erential
equation onto a di¤erence one.
By de…nition dxdti (t)
= lim t!0 xi (t+ t)t xi (t) and if we assume( t "small
x1 (t+ t) x1 (t)
dxi (t) xi (t+ t) xi (t) t ' f1 (x1 ; x2 ); x1 (0) = x10
enough" , we can write dt ' t leading to the system x2 (t+ t) x2 (t)
t ' f2 (x1 ; x2 ); x2 (0) = x20
x1 (t + t) ' x1 (t) +t:f1 (x1 ; x2 ); x1 (0) = x10
or which can be solved
x2 (t + t) ' x2 (t) +t:f2 (x1 ; x2 ); x2 (0) = x20
x10
iteratively starting at the initial condition X(0) =
x20
The graphical approach derived from Euler’s method attempts to construct
trajectories using this approach.

2
4 Euler’s graphical approach
x1 (t + t) ' x1 (t) + t:f1 (x1 ; x2 ) x1 (t + t) x1 (t) ' t:f1 (x1 ; x2 )
From Euler’s method we have or
x2 (t + t) ' x2 (t) + t:f2 (x1 ; x2 ) x2 (t + t) x2 (t) ' t:f2 (x1 ; x2 )
x1 ' t:f1 (x1 ; x2 ) x2 :f2 (x1 ;x2 )
or leading to x1 = :f1 (x1 ;x2 ) if we eliminate time
x2 ' t:f2 (x1 ; x2 )
to obtain a relationship betweem x2 and x1
Clearly :f 2 (x1 ;x2 )
:f1 (x1 ;x2 ) = s(x1 ; x2 ) is the slope of the state trajectory at X =
x1
and we can write x2 ' s(x1 ; x2 ) x1 The idea is that if we select
x2
x1
x1 "small enough", we can approximate the trajectory at X = by a
x2
"short " directed segment along the tangent line at this point;. The direction
of the segment is determined from dx dt = fi (x1 ; x2 ):
i

The procedure to plot state trajectories can be summarized as:

x10
Pick an initial point X0 =
x20
:f1 (x10 ;x20 )
Compute the slope s(x10 ; x20 ) = :f2 (x10 ;x20 )

x10
Draw a short segment of slope s(x10 ; x20 ) at
x20
Determine the direction of the segment
Pick the end point of the short segment as a new initial point and restart
the process.
:f2 (x1 ;x2 )
In s(x1 ; x2 ) = :f 1 (x1 ;x2 )
we consider two possible cases : :f1 (x1 ; x2 ) 6= 0
and :f1 (x1 ; x2 ) = 0
case 1: f1 (x1 ; x2 ) 6= 0
If f2 (x1 ; x2 ) 6= 0, s(x1 ; x2 ) is a …nite nonzero slope. Otherwise s(x1 ; x2 ) = 0
and a have a horizontal short sement.
case 1: f1 (x1 ; x2 ) = 0
If f2 (x1 ; x2 ) 6= 0, s(x1 ; x2 ) is an in…nite slope ( vertical short segment),
otherwise we are dealing with an equilibrium point.
Euler’s graphical approach plots one trajecttory at a time corresponding to
a selected initial point in the state plane. It follows that the procedure has to
be repeated over and over for a su¢ cient number of initial points to obtain a
phase portrait.
Clearly, the shorter the drawn directed segment the more accurate will be
the state trajectory but the expense of more computations required. One can
modulate the length of the short segment according to regions of interest ( We
may want greater accuracy in some particular region ).