Topic 10.

Nonlinear Systems and

Their Linearization
Instructor: Prof. Kwang-Hyun Cho

Department of Bio and Brain Engineering

Korea Advanced Institute of Science and Technology

Web: http://sbie.kaist.ac.kr

BiS352 System Modeling in Bioengineering Fall 2013

Linearity and Nonlinearity

The models thus far are developed from systems that can be described
approximately by linear, time-invariant differential equations.

An assumption of linearity was implicit in the development of these

models.

We will formally define the terms linear and nonlinear and show how to
distinguish between the two.

Then, we will learn how to approximate a nonlinear system to a linear

system so that we can apply the modeling techniques previously
covered.

Most of the materials used in this slide are from Control Systems Engineering (4th ed, by Nise).

Kwang-Hyun Cho

1
 Linearity
A linear system possesses two properties: superposition and
homogeneity.
 The property of superposition means that the output response of a system
to the sum of inputs is the sum of the responses to the individual inputs.
Thus, if an input r1(t) yields an output c1(t) and an input r2(t) yields an
output c2(t), then an input r1(t) + r2(t) yields an output c1(t) + c2(t).

 The property of homogeneity describes the response of the system to a

multiplication of the input by a scalar.
Specifically, in a linear system, the property of homogeneity is
demonstrated if for an input of r1(t) that yields an output of c1(t), an input
of Ar1(t) yields an output of Ac1(t); that is, multiplication of an input by a
scalar yields a response that is multiplied by the same scalar.

Kwang-Hyun Cho

Nonlinearity

We can visualize linearity as shown in Figure 2.45.

Figure 2.45

 Figure 2.45(a) is a linear system where output is always ½ the input, or

f(x) = 0.5x, regardless of the value of x. Thus the two properties of linear
systems apply.

 Figure 2.45(b) where f(x) = 2-2e-x is a nonlinear system.

Kwang-Hyun Cho

2
 Nonlinearity

Figure 2.46

 Figure 2.46 shows some examples of physical nonlinearities.

 An electronic amplifier is linear over a specific range but exhibits
nonlinearity called saturation at high input voltages.
 A motor that does not respond at very low input voltages due to frictional
forces exhibits a nonlinearity called dead zone.
 Gears that do not fit tightly exhibit a nonlinearity called backlash: the
input moves over a small range without the output responding.

Kwang-Hyun Cho

Mathematical description of dynamic systems

Input and output description (Black box model)

 Only the relation between the input and output data is considered. Inner
states are not observed.

Input u(t) y(t) Output y( t )  M ( u( t ), t )

M
where M(·) is a nonlinear
operator (or nonlinear mapping).
State space description (White box model)
 Both input/output data and the inner states are considered at the same
time. It can help the understanding of the inner situation of systems.

x  f ( x ( t ), u( t ), t )
y  g ( x ( t ), u( t ), t )

where f and g are nonlinear functions with

respect to x(t) and u(t).
Kwang-Hyun Cho

3
 Properties of nonlinear dynamic systems
Autonomous nonlinear system (time-invariant)
 A nonlinear system is said to be autonomous if f does not depend explicitly
on time t, i.e., if the system’s state equation can be written as
x  f ( x ( t ), u( t ))
Equilibrium point
 A state x* is an equilibrium state (or equilibrium point) of the system if once
x(t) is equal to x*, it remains equal to x* for all future time under the
condition that there is no input.
x *  f ( x * , u( t ))  0 with u(t) =0
 A linear system x  Ax can have only a unique equilibrium when A is a
nonsingular matrix; equilibrium points of infinity, which are contained in the
null-space of the matrix A, when A is a singular matrix.

 A nonlinear system can have several (or infinitely many) isolated equilibrium
points.
For example, x1  x2 Therefore, the equilibrium x2  0
b g points are given by x1   k , k  
x2   x2  sin x1
MR 2 R Kwang-Hyun Cho

Properties of nonlinear dynamic systems

Stability and instability
 An equilibrium point x=0 is said to be stable if, for any scalar R>0, there
exists another scalar r>0, such as if ║x(0)║<r, then ║x(t)║<R for all t≥0.
Otherwise, the equilibrium point is unstable.
 Essentially, stability (also called stability in the sense of Lyapunov, or
Lyapunov stability) means that the system trajectory can be kept arbitrarily
close to the origin by starting sufficiently close to it.
Asymptotical Stability and Marginal Stability
 An equilibrium point is said to be asymptotically stable if it is stable, and if,
in addition, there exists some r>0 such that ║x(0)║<r implies x(t)→0 as
t→∞.
 An equilibrium point is said to be marginally stable if it is Lyapunov stable
but not asymptotically stable.
R
1 2
Asymptotically stable Curve 1
Marginally stable Curve 2 0 r
Unstable Curve 3 X(0)
3

Kwang-Hyun Cho

4
 Properties of nonlinear dynamic systems
Local Stability and Global Stability
 If asymptotic stability holds for any initial states, the equilibrium point is
said to be asymptotically stable in the large. It is also called globally
asymptotically stable.

Kwang-Hyun Cho

Properties of nonlinear dynamic systems

Positive function and Lyapunov’s function

 A scalar continuous function V(x) (with V(0)=0) is said to be locally positive

definite if in a ball ║x║<R for any x≠0 it gives V(x)>0.
Or, if V(x) ≥ 0, then it is called locally positive semi-definite.

 If V(0)=0 and the above property holds over the whole state space, then
V(x) is said to be globally positive definite.

 If, in a ball ║x║<R, the function V(x) is positive definite and has
continuous partial derivative, and if its time derivative along any state
trajectory of a system x  f ( x ) is negative semi-definite, i.e., V ( x)  0
then V(x) is said to be a Lyapunov function for the system.

Kwang-Hyun Cho

5
 Properties of nonlinear dynamic systems
Lyapunov’s Direct Method to determine the stability
 Theorem (Local Stability): If, in a ball ║x║<R, there exists a scalar
function V(x) with continuous first partial derivatives such that
V(x) is positive definite (locally in ║x║<R)

V(x) is negative semi-definite (locally in ║x║<R)

then the equilibrium point x*=0 is stable. If the derivative V(x) is locally
negative definite in ║x║<R, then the stability is asymptotic.

x2

V=V3 V3 > V2 > V1

V=V2
0 x1
V=V1 x(t)

Typical shape of a positive definite The trajectory of the system crossing

function V(x1,x2) the contour curves of V(x1,x2)
Kwang-Hyun Cho

Properties of nonlinear dynamic systems

 Theorem (Global Stability): Assume that there exists a scalar function V
of the state x, with continuous first order derivatives such that:
V(x) is positive definite

V(x) is negative definite
V(x) → ∞ as ║x║ → ∞
Then, the equilibrium point x*=0 is globally asymptotically stable.
 Example: A nonlinear mass-damper spring system
Let us consider the nonlinear system in the below figure whose dynamic
equation is
mx  bx x  k0 x  k1 x3  0
with bx x representing nonlinear dissipation or damping, and k0 x  k1 x
3

representing nonlinear spring term. Assume that the mass is pulled away
from the natural length of the spring by a large distance, and then released.
Will the resulting motion be stable?

Kwang-Hyun Cho

6
Properties of nonlinear dynamic systems
 Example: A nonlinear mass-damper spring system (Cont.)
It is very difficult to answer this question using the definitions of stability,
because the general solution of this nonlinear equation is unavailable.
The total mechanical energy of the system is the sum of its kinetic energy
and its potential energy:
1 1 1 1
V ( x)  mx 2    k0 x  k1 x3 dx  mx 2  k0 x 2  k1 x 4  0, for x  0 and x  0
x

2 0 2 2 4
It is obvious that the energy function V(x) is positive definite for any position x.
The rate of energy variation during the system’s motion is obtained easily by
differentiating V(x) and using the dynamical equation of the system

   k0 x  k1 x 3  x  b x  0
3
V ( x)  mxx

Therefore, V(x) is a Lyapunov function of the system. The above equation

implies that the energy of the system, starting from some initial value, is
continuously dissipated by the damper until the mass settles down, i.e., until
x  0 . Physically, it is easy to see that the mass must finally settle down at
the natural length of the spring, because it is subjected to a non-zero spring
force at any position other than the natural length. Thus, the system is
globally asymptotically stable. Kwang-Hyun Cho

Properties of nonlinear dynamic systems

 Limit Cycles
Nonlinear systems can display oscillations of fixed amplitude and fixed
period without external excitation. These oscillations are called limit cycles,
or self-excited oscillations.
The second order nonlinear differential equation mx  2c  x  1 x  kx  0
2

where m, c, and k are positive constant, is the famous Van der Pol equation.

Let x1  x and x2  x
The corresponding state space description can be described as
x1  x2
x2    kx1  2c  x12  1 x2  / m

Kwang-Hyun Cho

7
Properties of nonlinear dynamic systems
The Matlab code for simulating the system can be shown as

m = 1;
c = 2;
k = 10;
[T1,Y1] = ode45(@(t,y) Vandepolsystem(t,y,m,c,k), [0 20], [-3 0]);
[T2,Y2] = ode45(@(t,y) Vandepolsystem(t,y,m,c,k), [0 20],[5 0]);
plot(T1,Y1(:,1),'-.', T2,Y2(:,1),'-','LineWidth', 5)
xlabel('t', 'FontSize', 50, 'FontWeight', 'bold')
ylabel('x(t)', 'FontSize', 50, 'FontWeight', 'bold')
title('Response of Van der Pol Oscillator', 'FontSize', 40, 'FontWeight', 'bold')
set(gca, 'FontSize', 30, 'FontWeight', 'bold')
legend('The initial condition 1: x(0)=0, x''(0)=0', 'The initial condition 2: x(0)=5,
x''(0)=0')
legend(gca,'boxoff‘)

function dy = Vandepolsystem(t,y,m,c,k)
dy = zeros(2,1); % a column vector
dy(1) = y(2);
dy(2) = ( -k*y(1)-2*c*(y(1)^2-1)*y(2) )/m;

Kwang-Hyun Cho

Properties of nonlinear dynamic systems

It displays a sustained and stable oscillation irrespective of initial conditions.

Response of Van der Pol Oscillator

5
The initial condition 1: x(0)=0, x'(0)=0
4 The initial condition 2: x(0)=5, x'(0)=0

2
x(t)

-1

-2

-3
0 5 10 15 20
t
m=1, c=2, and k=10

Kwang-Hyun Cho

8
 Properties of nonlinear dynamic systems
The phase plane method is concerned with the graphical study of second-order
autonomous systems described by
x1  f1 ( x1 , x2 )
x2  f 2 ( x1 , x2 )
where x1 and x2 are the states of the system, and f1 and f2 are nonlinear functions
of the states. Geometrically, the state space of the system is a plane having x1 and
x2 as coordinates. We will call this plane the phase plane.
With time t varied from zero to infinity, the solution x(t) =[x1(t), x2(t)] can be
represented geometrically as a curve in the phase plane. Such a curve is called a
phase plane trajectory, i.e., phase portrait of a system.
Phase Portrait of Van der Pol Oscillator
10

The values of parameters determine

dx/dt

0 the exact shape of the limit cycle in

the phase plane.
-5
The parameter set 1: m=1, c=2, k=10
The parameter set 2: m=3, c=6, k=2
-10
-3 -2 -1 0 1 2 3
x Kwang-Hyun Cho

Properties of nonlinear dynamic systems

 Bifurcations
As the parameters of nonlinear dynamic systems are changed, the stability
of the equilibrium point can change and so can the number of equilibrium
points. Values of these parameters at which the qualitative nature of the
system’s motion changes are known as critical or bifurcation values. The
phenomenon of bifurcation, i.e., quantitative change of parameters leading
to qualitative change of system properties, is the topic of bifurcation theory.
For example, a system is described by the so-called undamped Duffing
equation
x   x  x3  0

We can plot the equilibrium points as a function of the parameter α. As α
varies from positive to negative, one equilibrium point splits into three points
( xe  0,  ,   ) as shown in the following figure. f(t)
xe
stable

stable
A pitchfork bifurcation unstable
0 α

stable
Kwang-Hyun Cho

9
Properties of nonlinear dynamic systems
 Chaos
For stable linear systems, small differences in the initial conditions can only
cause small differences in the output. Nonlinear systems, however, can
display a phenomenon called chaos, by which we mean that system output
is extremely sensitive to the initial conditions.

The very famous ‘Butterfly Effect ’, which is essentially a chaotic

phenomenon, means that a very small change of initial state may cause a
result of disaster.

The essential feature of chaos is the unpredictability of the system output.

Even if we have an exact model of a nonlinear system and an extremely
accurate computer, the system’s response in the long-run still cannot be
well predicted.

Chaos must be distinguished from random motion. In random motion, the

system model or input contain uncertainty and, as a result, the time
variation of the output cannot be predicted exactly (only statistical
measures are available). In chaotic motion, on the other hand, the involved
problem is deterministic, and there is little uncertainty in the system model.
Kwang-Hyun Cho

Properties of nonlinear dynamic systems

As an example of chaotic behavior, let us consider a simple nonlinear
system
x  0.1x  x5  6sin t

which may represent a lightly-damped, sinusoidally forced mechanical
structure undergoing large elastic deflections. The following figure shows
the response of the system corresponding to two almost identical initial
conditions. Due to the presence of the strong nonlinearity in x5, the two
responses are radically different after some time.
Chaotic Response of the System
3
The initial condition 1: x(0)=2, x'(0)=3
The initial condition 2: x(0)=2.01, x'(0)=3.01
2

1
x(t)

-1

-2

-3
0 20 40 60 80 100
t Kwang-Hyun Cho

10
Properties of nonlinear dynamic systems
The corresponding state space description can be described as
x1  x  x2
x2  
x  6sin t  0.1x2  x15
The Matlab code for simulating the system can be shown as

[T1,Y1] = ode45(@chaoticsystem1, [0 100], [2 3]);

[T2,Y2] = ode45(@chaoticsystem1, [0 100],[2.01 3.01]);
plot(T1,Y1(:,1),'-.', T2,Y2(:,1),'-','LineWidth', 5)
xlabel('t', 'FontSize', 50, 'FontWeight', 'bold')
ylabel('x(t)', 'FontSize', 50, 'FontWeight', 'bold')
title('Chaotic Response of the System', 'FontSize', 40, 'FontWeight', 'bold')
set(gca, 'FontSize', 30, 'FontWeight', 'bold')
legend('The initial condition 1: x(0)=2, x''(0)=3', 'The initial condition 2:
x(0)=2.01, x''(0)=3.01')
legend(gca,'boxoff')

function dy = chaoticsystem1(t, y)
dy = zeros(2,1); % a column vector
dy(1) = y(2);
dy(2) = 6*sin(t) - 0.1*y(2) - y(1)^5;

Kwang-Hyun Cho

Linearization
A designer can often make a linear approximation to a nonlinear system.
Linear approximations simplify the analysis and design of a system and
are used as long as the results yield a good approximation to reality.
 For example, a linear relationship can be established at a point on the
nonlinear curve if the range of input values about that point is small and the
origin is translated to that point.
The process of linearizing nonlinear systems is important, because by
linearizing nonlinear equations, it is possible to apply numerous linear
analysis methods that will produce information on the behavior of those
systems.
 The linearization procedure presented here is based on the expansion of the
nonlinear function into a Taylor series about the operating point and the
retention of only the linear terms.
 Because we neglect higher order terms in the Taylor series expansion,
these neglected terms must be small enough; that is, the variables must
deviate only slightly from the operating condition.
Linearization procedure
 The first step is to recognize the nonlinear component and write the
nonlinear differential equation.

Kwang-Hyun Cho

11
 Linearization
 When we linearize a nonlinear differential equation, we linearize it for small-
signal inputs about the steady-state solution obtained while the small signal
input is considered as nonexistent. This steady-state solution is called
equilibrium and is selected for the further linearization process.
 Next we linearize the nonlinear differential equation, and then we take the
Laplace transform of the linearized differential equation, assuming zero initial
condition.
 Finally, we separate input and output variables and form the transfer function.
Let us first see how to linearize a function.

 If we assume a nonlinear system operating at

point A, [x0, f(x0)] in Figure 2.47, small changes in
the input can be related to the changes in the
output about the point by way of the slope of the
curve at the point A.
 Thus, if the slope of the curve at point A is ma,

then small excursions of the input about point A, δx,

yield small changes in the output, δf(x), related by
the slope at A. Thus,

Figure 2.47 [ f ( x )  f ( x0 )]  m a ( x  x0 )
Kwang-Hyun Cho

Linearization

 From which  f ( x )  ma x

and f ( x )  f ( x0 )  m a ( x  x0 )  f ( x0 )  m a  x Eq. 2.180

 This relationship is shown graphically in Figure 2.47, where a new set of

axes, δx and δf(x), is created at the point A, and f(x) is approximately equal
to f(x0), the ordinate of the new origin, plus small excursions, maδx, away
from point A.

Example 2.26

Linearize f ( x)  5cos x about x  x0 
2
 We first find that the derivative of f(x) is
df/dx = -5sinx.
 At the given point, the derivative is -5.
Also, at that point, f(x0) = 0. Thus from
Eq. 2.180, the system can be represented
as f(x) = -5δx for small excursions of x
about π/2. Figure 2.48
Kwang-Hyun Cho

12
 Linearization

The previous discussion can be formalized using the Taylor series

expansion, which expresses the value of a function in terms of the
value of that function at a particular point, the excursion away from
that point, and derivatives evaluated at that point.
The Taylor series is shown in Eq. 2.181.
df ( x  x0 ) d 2 f ( x  x0 ) 2
f ( x )  f ( x0 )    Eq. 2.181
dx x  x0 1! dx 2 x  x0
2!
For small excursions of x from x0, we can neglect the higher-order
terms. The resulting-approximation yields a straight-line relationship
between the change in f(x) and the excursions away from x0. We get:
df
f ( x )  f ( x0 )  ( x  x0 ) Eq. 2.182
dx x  x0
or

 f ( x)  m x  x0
x Eq. 2.183

Kwang-Hyun Cho

Linearization

Example 2.27 Linearizing a differential equation

 Problem: Linearize Eq. 2.184 for small excursions about x = π/4.
d 2x dx
2  cos x  0 Eq. 2.184
dt 2 dt
 Solution : The presence of the term cos(x) makes this equation nonlinear.
Let x = δx + π/4, where δx is the small recursion.
     
d 2  x   d  x   Eq. 2.185
 4  4   
2  cos   x    0
dt 2 dt  4

But  
   
d 2  x   d  x  
4 d x
2
   4  d x Eq. 2.187
2 Eq. 2.186 
dt dt 2 dt dt
By substituting following equations into Eq. 2.182
      
f ( x )  cos   x  , f ( x 0 )  f    cos   , ( x  x0 )   x
 4  4 4 
we get:
      d cos x  
cos   x    cos     x   sin   x Eq. 2.188
 4 4 dx x
 4
4 Kwang-Hyun Cho

13
 Linearization

Solve the last equation:

       2 2
cos   x    cos    sin   x   x Eq. 2.189
 4 4  4  2 2

Substituting Eqs. 2.186-2.189 into Eq. 2.185 yields the following linearized
differential equation:
d 2 x d x 2 2 Eq. 2.190
2
2  x
dt dt 2 2
 Even though the nonlinear Eq. 2.184 is homogeneous, the linerized Eq.
2.190 is not homogeneous. It has a forcing function on its right-hand side.
 Another observation about Eq. 2.190 is the negative sign on the left-hand
side. The study of differential equations tells us that since the roots of the
characteristic equation are positive, the homogeneous solution grows
instead of diminishing to zero. Thus, this system linearized about x = pi/4 is
not stable.

Kwang-Hyun Cho

Linearization
Example 2.28 Transfer function-nonlinear electrical network
 v(t) is a small alternative voltage source.
 Find the transfer function VL(s) / V(s), for the electrical network shown in
Figure 2.49, which contains a nonlinear resistor whose voltage-curent
relationship is defined by:
ir  2 e 0.1vr
 Solution : We will use Kirchhoff’s voltage law to sum the voltages in the
loop to obtain the nonlinear differential equation.
 Applying Kirchhoff’s voltage law around the loop, where ir = i, yields

1
ir  2 e 0.1vr v r  10 ln ir
2
di 1
L  10 ln i  20  v ( t ) Eq. 2.191
dt 2

 Next, let us evaluate the equilibrium solution.

First, set the small-signal source, v(t), equal to
zero. Now we evaluate the steady-state
current.
Figure 2.49
Kwang-Hyun Cho

14
 Linearization

 At the steady state the voltage across the inductor will be zero, since vL(t)
= Ldi/dt and di/dt is zero at the steady state, given a constant battery
source.
 Hence, the resistor voltage is 20 V. Using the characteristics of the resistor,
we find that ir = i = 14.78 A. This current, i0, is the equilibrium value of the
network current.
 Hence : i  i0   i
 Substituting this current into Eq. 2.191 yields:
d ( i0   i ) 1
L  10 ln ( i0   i )  20  v ( t ) Eq. 2.192
dt 2
1
 Use Eq. 2.182 to linearize ln ( i0   i ) :
2

 1 
d  ln i 
1 1  2  1 1
ln ( i0   i )  ln i0  i  i  i Eq. 2.193
2 2 dt i i  i0 i0
i  i0

Kwang-Hyun Cho

Linearization

or 1 i 1
ln ( i0   i )  ln 0   i Eq. 2.194
2 2 i0
 Substituting into Eq. 2.192, the linearized equaion becomes:

d i  i 1 
L  10  ln 0   i   20  v ( t ) Eq. 2.195
dt  2 i 0 
 Letting L = 1 and i0 = 14.78, the final linearized differential equation is:
d i
 0.677  i  v ( t ) Eq. 2.196
dt
 Taking the Laplace transform with zero initial conditions and solving for δi(s),
we get:
V (s)
 i(s)  Eq. 2.197
s  0.677
 The voltage across the inductor about the equilibrium point is:
d d i
v L (t )  L ( i0   i )  L Eq. 2.198
dt dt
Kwang-Hyun Cho

15
 Linearization

 Taking the Laplace transform yields:

V L ( s )  Ls i ( s )  s i ( s ) Eq. 2.199

 Substituting Eq. 2.197 into Eq. 2.199 yields:

V (s)
VL ( s)  s Eq. 2.200
s  0.677
 From which the final transfer function is:
VL (s) s
 Eq. 2.201
V (s) s  0.677
for small excursions about i = 14.78 or, equivalently, about v(t) = 0.

Kwang-Hyun Cho

Linearization

Linearization of z = f(x,y) about a point (x, y, z)

 Next, we consider a nonlinear system whose output z is a function of two
inputs x and y, or
z  f ( x, y )
 To obtain a linear mathematical model for this nonlinear system about an
operating point (x, y, z) , we expand Eq. 7-24 into a Taylor series about that
point. Then:
 f f 
z  f ( x, y )  f ( x , y )   (x  x )  ( y  y )
  x x x , y y y x x , y y 
1  2 f 2 f 2 f 
  ( x  x )2  ( x  x )( y  y )  2 ( y  y )2   
2!  x2 xy x x , y y y 
 x x , y y x x , y y

 Near this point, the higher order terms may be neglected. Noting that
z  f ( x , y ). We find that a linear mathematical model of this nonlinear
system near the operating point (x, y, z ) is:

Kwang-Hyun Cho

16
 Linearization

z  z  a( x  x )  b( y  y )
where
f f
a b
x x x , y y
y x x , y y

 It is important to remember that in the present linearization procedure, the

deviations of the variables from the operating condition must be sufficiently
small. Otherwise, the procedure does not apply.

Example 7-3
 Linearize the nonlinear equation:
z  xy
in the region 5 ≤ x ≤ 7, 10 ≤ y ≤ 12. Find the error if the linearized
equation is used to calculate the value of z when x = 5 and y = 10.
 Since the region considered is given as above, choose x  6, y  11
 Then, z  xy  66. Let us obtain a linearized equation for the nonlinear
equation near this point.
Kwang-Hyun Cho

Linearization

 Expanding the nonlinear equation into a Taylor series about this point and
neglecting the higher order terms, we have:
z  z  a( x  x )  b( y  y )
where
f f
a  y  11 b x 6
x x x , y y
y x x , y y

 Hence, the linearized equation is:

z  66  11( x  6)  6( y  11)
or
z  11x  6 y  66
 When x = 5 and y = 10, the value of z given by the linearized equation is:

z  11x  6 y  66  49
 The exact value of z is z = xy = 50. The error is thus 50 - 49 = 1. In terms
of percentage, the error is 2%.

Kwang-Hyun Cho

17
 Linearization – State space representation

A prime advantage of the state-space representation over the transfer

function representation is the ability to represent systems with
nonlinearities.

The ability to represent nonlinear systems does not imply the ability to
solve their state equations for the state variables and the output.

If we are interested in small perturbations about an equilibrium point,

we can linearize the state equations about the equilibrium point.

The key to linearize about an equilibrium point is, once again, the
Taylor series.

Let us consider an example of 2nd-order nonlinear system.

x1  f1 ( x1 , x2 )
x 2  f 2 ( x1 , x2 )

Kwang-Hyun Cho

Linearization – State space representation

We can have its Taylor series at the point ( x1 , x2 ) as follows
x1  f1 ( x1 , x2 )  a11 ( x1  x1 )  a12 ( x2  x2 )  Higher Order Terms
x 2  f 2 ( x1 , x2 )  a21 ( x1  x1 )  a22 ( x2  x2 )  Higher Order Terms

where f1 ( x1 , x2 ) f1 ( x1 , x2 )

a11  , a12 
x1 x  x ,x  x
x2 x  x ,x  x
1 1 2 2 1 1 2 2

f 2 ( x1 , x2 ) f ( x , x )
a21  , a22  2 1 2
x1 x  x ,x  x
x2 x  x ,x  x
1 1 2 2 1 1 2 2

If ( x1 , x2 ) is an equilibrium point of the system, we also have

f1 ( x1 , x2 )  0
f 2 ( x1 , x2 )  0

In the following example we write the state equations for a simple

pendulum, showing that we can represent a nonlinear system in a
state space; then we linearize the pendulum about its equilibrium point.
Kwang-Hyun Cho

18
 Linearization – State space representation
Example 3.7
 Consider the simple pendulum shown in Figure 3.14(a). Mg is the weight, T
is an applied torque in the θ direction, and L is the length of the pendulum.
 Assume the mass is evenly distributed, with the center of mass at L/2. Then
linearize the state equation about the pendulum’s equilibrium point – the
vertical position with zero angular velocity.
 First draw a free-body diagram as shown in Figure 3.14(c). Summing the
torques, we get:
d 2 MgL
J  sin   T Eq. 3.79
dt 2 2
 Here J is the
momentum of inertia
of the pendulum
around the point of
rotation.
 Select the state

variables x1 and x2 as
phase variables.

Figure 3.14
Kwang-Hyun Cho

Linearization – State space representation

 Letting x1 = θ and x2 = dθ/dt, we write the state equations as:

x1  x 2 Eq. 3.80(a)
MgL T
x 2   sin x1  Eq. 3.80(b)
2J J
 Thus, we have represented a nonlinear system in a state space. It is
interesting to note that the nonlinear Eqs. 3.80(a)-(b) represent a valid and
complete model of the pendulum in the state space even under nonzero
initial conditions and even if parameters, such as mass, are time varying.
 However, if we want to apply classical techniques and convert these state
equations to a transfer function, we must linearize them.

 Let us proceed now to linearize the equation about the equilibrium point, x1
= 0, x2 = 0, that is θ = 0 and dθ/dt = 0. Let x1 and x2 be perturbed about
the equilibrium point, or
x1  0   x1 Eq. 3.81(a)

x2  0   x2 Eq. 3.81(b)

Kwang-Hyun Cho

19
 Linearization – State space representation

 Using Eq. 2.182, we obtain:

d sin x1
sin x1  sin 0   x1   x1 Eq. 3.82
dx x1  0

from which
sin x1   x1 Eq. 3.83

 Substituting Eqs. 3.81-3.83 into Eq. 3.80 yields the following state
equations:
 x1   x 2 Eq. 3.84(a)
MgL T
 x 2    x1  Eq. 3.84(b)
2J J
which are linear and a good approximation to Eqs. 3.80(a)-(b) for small
excursions away from the equilibrium point.

Kwang-Hyun Cho

20