Lectures on Dynamic Systems and
Control
Mohammed Dahleh Munther A. Dahleh George Verghese
Department of Electrical Engineering and Computer Science
1
Massachuasetts Institute of Technology
1
c �
Chapter 13
Internal (Lyapunov) Stability
13.1 Intro duction
We have already seen some examples of b oth stable and unstable systems. The ob jective of
chapter is to formalize the notion of internal stability for general nonlinear state-space this
mo dels. Apart from de�ning the various notions of stability, we de�ne an entity known as a
various stability notions. Lyapunov function and relate it to these
13.2 Notions of Stability
For a general undriven system
x_ (t) � f (x(t)� 0� t) (C T ) (13.1)
x(k + 1) � f (x(k )� 0� k ) (D T )� (13.2)
we say that a p oint x is an equilibrium point from time t for the CT system ab ove if f (x� 0� t) �
0
, and is an equilibrium 0� 8t � t p oint from time k for the DT system ab ove if f ( x� 0� k ) �
0 0
, it will remain there for all x at time t or k � . If the system is started in the state � x� 8k � k
0 0 0
have multiple equilibrium p oints (or equilibria). (Another class time. Nonlinear systems can
sp ecial solutions for nonlinear systems are periodic solutions, but we shall just fo cus on of
We would like to characterize the stability of the equilibria in some fashion. equilibria here.)
For example, do es the state tend to return to the equilibrium p oint after a small p erturbation
away from it� Do es it remain close to the equilibrium p oint in some sense� Do es it diverge�
The most fruitful notion of stability for an equilibrium p oint of a nonlinear system is
given by the de�nition b elow. We shall assume that the equilibrium p oint of interest is at
the origin, since if 6� 0, a simple translation can always b e applied to obtain an equivalent x
system with the equilibrium at 0.
De�nition 13.1 A system is called asymptotical ly stable around its equilibrium p oint at the
following two conditions: origin if it satis�es the
, then : 1. Given any � � 0� 9� � 0 such that if kx(t )k � � kx(t)k � �� 8 t � t
0 1 0 1
, then 2. 9� � 0 such that if kx(t )k � � x(t) ! 0 as t ! 1.
0 2 2
The �rst condition requires that the state tra jectory can b e con�ned to an arbitrarily
centered at the equilibrium p oint and of radius �, when released from an arbitrary small \ball"
. This is called stability in su�ciently small (but p ositive) radius � initial condition in a ball of
1
p ossible to have stability in the sense of Lyapunov without the sense of Lyapunov (i.s.L.). It is
having asymptotic stability, in which case we refer to the equilibrium p oint as marginal ly
stable. Nonlinear systems also exist that satisfy the second requirement without b eing stable
following example shows. An equilibrium p oint that is not stable i.s.L. is termed i.s.L., as the
unstable.
Example 13.1 (Unstable Equilibrium Point That Attracts All Tra jectories)
Consider the second-order system with state variables x and x whose dynamics
1 2
describ ed in p olar co ordinates via the equations are most easily
r_ � r (1 ; r )
2
_
� � sin (� �2) (13.3)
q
2 2
where the radius r is given by r � x + x and the angle � by 0 � � �
1 2
) � (x �x 2� . (You might try obtaining a state-space description directly arctan
2 1
.) It is easy to see that there are precisely involving x and x two equilibrium
2 1
p oints: one at the origin, and the other at r � 1, � � 0. We leave you to verify
simulation from various initial conditions) with rough calculations (or computer
tra jectories of the system have the form shown in the �gure b elow. that the
Evidently all tra jectories (except the trivial one that starts and stays at the origin)
However, this equilibrium p oint is not stable i.s.L., end up at r � 1, � � 0.
b ecause these tra jectories cannot b e con�ned to an arbitrarily small ball around
p oint when they are released from arbitrary p oints with any ball the equilibrium
how small) around this equilibrium. (no matter y
unit circle
x
Figure 13.1: System Tra jectories
13.3 Stability of Linear Systems
We may apply the preceding de�nitions to the LTI case by considering a system with a
p oint diagonalizable A matrix (in our standard notation) and u � 0. The unique equilibrium
0, provided A has no eigenvalue at 0 (resp ectively 1) in the CT (resp ectively DT) is at x �
every p oint in the entire eigenspace corresp onding to this eigenvalue is an case. (Otherwise
Now equilibrium.)
At
x_ (t) � e x(0)
2 3
t �
1
e
6 7
.
.
� V W x(0) (C T ) (13.4)
4 5
.
� t
n
e
k
x(k ) � A x(0)
2 3
k
�
1
6 7
.
.
� V W x(0) (D T ) (13.5)
4 5
.
k
�
n
Hence, it is clear that in continuous time a system with a diagonalizable A is asymptotically
stable i�
) � Re(� 0� i 2 f1� : : : � ng� (13.6)
i
while in discrete time the requirement is that
j � 1 j� i 2 f1� : : : � ng� (13.7)
i
) j � 1 (DT), the system is not asymptotically stable, but Note that if Re(� � 0 (CT) or j�
i i
is marginally stable.
Exercise: For the nondiagonalizable case, use your understanding of the Jordan form to show
stability are the same as in the diagonalizable case. For that the conditions for asymptotic
) � 0, with marginal stability, we require in the CT case that Re(� equality holding for at
i
eigenvalue� furthermore, every eigenvalue whose real part equals 0 should have its least one
multiplicity equal to its algebraic multiplicity, i.e., all its asso ciated Jordan blo cks geometric
b e of size 1. (Verify that the presence of Jordan blo cks of size greater than one for should
eigenvalues would lead to the state variables growing polynomial ly with these imaginary-axis
stability in the DT case. time.) A similar condition holds for marginal
Stability of Linear Time-Varying Systems
Recall that the general unforced solution to a linear time-varying system is
x(t) � �(t� t )x(t )�
0 0
where �(t� � ) is the state transition matrix. It follows that the system is
1. stable i.s.L. at ) � k�(t� t )k � m(t 1. x � 0 if sup
0 0
t
2. asymptotically stable at . x � 0 if lim k�(t� t )k ! 0� 8t
0 0
t!1
These conditions follow directly from De�nition 13.1.
13.4 Lyapunov's Direct Metho d
General Idea
Consider the continuous-time system
x_ (t) � f (x(t)) (13.8)
with an equilibrium p oint at x � 0. This is a time-invariant (or \autonomous") system, since f
do es not dep end explicitly on t. The stability analysis of the equilibrium p oint in such a system
we cannot write a simple formula is a di�cult task in general. This is due to the fact that
tra jectory to the initial state. The idea b ehind Lyapunov's \direct" metho d is to relating the
prop erties of the equilibrium p oint (or, more generally, of the nonlinear system) by establish
how certain carefully selected scalar functions of the state evolve as the system state studying
evolves. (The term \direct" is to contrast this approach with Lyapunov's \indirect" metho d,
which attempts to establish prop erties of the equilibrium p oint by studying the b ehavior of
linearized system at that p oint. We shall study this next Chapter.) the
Consider, for instance, a continuous scalar function V (x) that is 0 at the origin and
p ositive elsewhere in some ball enclosing the origin, i.e. V (0) � 0 and V (x) � 0 for x 6� 0 in
_
this ball. Such a V (x) may b e thought of as an \energy" function. Let V (x) denote the time
derivative of V (x) along any tra jectory of the system, i.e. its rate of change as x(t) varies
according to (13.8). If this derivative is negative throughout the region (except at the origin),
over time. In this case, b ecause the then this implies that the energy is strictly decreasing
lower b ounded by 0, the energy must go to 0, which implies that all tra jectories energy is
converge to the zero state. We will formalize this idea in the following sections.
Lyapunov Functions
n
De�nition 13.2 Let V b e a continuous map from R to R . We call V (x) a local ly positive
(lp d) function around x � 0 if de�nite
1. V (0) � 0.
2. V (x) � 0� 0 � kxk � r for some r .
Similarly, the function is called lo cally positive semide�nite (lpsd) if the strict inequality on
by V (x) � 0. The function V (x) is lo cally the function in the second condition is replaced
negative de�nite (lnd) if ;V (x) is lp d, and lo cally negative semide�nite (lnsd) if ;V (x) is
may b e useful in forming a mental picture of an lp d function V (x) is to think of lpsd. What
having \contours" of constant V that form (at least in a small region around the origin) it as
a nested set of closed surfaces surrounding the origin. The situation for n � 2 is illustrated
in Figure 13.2.
V(x)=c 2
V(x)=c 3
V(x)=c 1
Figure 13.2: Level lines for a Lyapunov function, where c � c � c .
1 2 3
Throughout our treatment of the CT case, we shall restrict ourselves to V (x) that have
continuous �rst partial derivatives. (Di�erentiability will not b e needed in the DT case |
continuity will su�ce there.) We shall denote the derivative of such a V with resp ect to time
_
along a trajectory of the system (13.8) by V (x(t)). This derivative is given by
dV (x) (x) dV
_
V (x(t)) � x_ � f (x)
dx dx
dV (x)
where is a row vector | the gradient vector or Jacobian of V with resp ect to x |
dx
@ V
containing the comp onent-wise partial derivatives .
@ x
i
_
De�nition 13.3 Let V b e an lp d function (a \candidate Lyapunov function"), and let V b e
_
its derivative along tra jectories of system (13.8). If V is lnsd, then V is called a Lyapunov
function of the system (13.8).
Lyapunov Theorem for Lo cal Stability
Theorem 13.1 If there exists a Lyapunov function of system (13.8), then x � 0 is a stable
_
equilibrium p oint in the sense of Lyapunov. If in addition V (x) � 0, 0 � kxk � r for some
1
_
, i.e. if V is lnd, then x � 0 is an asymptotically stable equilibrium r p oint.
1
Pro of: First, we prove stability in the sense of Lyapunov. Supp ose � � 0 is given. We need
�nd a � � 0 such that for all kx(0)k � � , it follows that kx(t)k � �� 8t � 0. The Figure to
19.6 illustrates the constructions of the pro of for the case n � 2. Let � � min(�� r ). De�ne
1
ε1
r δ
Figure 13.3: Illustration of the neighb orho o ds used in the pro of
m � min V (x):
kxk��
1
Since V (x) is continuous, the ab ove m is well de�ned and p ositive. Cho ose � satisfying
� � � � such that for all kxk � � , V (x) � m. Such a choice is always p ossible, again 0
1
b ecause of the continuity of V (x). Now, consider any x(0) such that kx(0)k � � , V (x(0)) � m,
_
and let x(t) b e the resulting tra jectory. V (x(t)) is non-increasing (i.e. V (x(t)) � 0) which
. (x(t)) � m. We will show that this implies that kx(t)k � � Supp ose there results in V
1
t such that kx(t )k � � , then by continuity we must have that at an earlier time t , exists
1 1 1 2
kV (x)k � m � V (x(t )), which is a contradiction. Thus stability kx(t )k � � , and min
2 2 1
kxk��
1
Lyapunov holds. in the sense of
_
To prove asymptotic stability when V is lnd, we need to show that as t ! 1, V (x(t)) !
by continuity of V , kx(t)k ! 0. Since V (x(t)) is strictly decreasing, and V (x(t)) � 0 0� then,
we know that V (x(t)) ! c, with c � 0. We want to show that c is in fact zero. We can argue
by contradiction and supp ose that c � 0. Let the set S b e de�ned as
n
S � fx 2 R jV (x) � cg �
and let B b e a ball inside S of radius �,
�
B � f x 2 S jkxk � �g :
�
Supp ose x(t) is a tra jectory of the system that starts at x(0), we know that V (x(t)) is
� recall that 2 B (x(t)) � c for all t. Therefore, x(t) � decreasing monotonically to c and V
�
n
for which V (x) � c. In the �rst part of B � S which is de�ned as all the elements in R
�
pro of, we have established that if kx(0)k � � then kx(t)k � �. We can de�ne the largest the
derivative of V (x) as
_
;� � max V (x):
��kxk��
_
Clearly ;� � 0 since V (x) is lnd. Observe that,
Z
t
_
V (x(t) � V (x(0)) + V (x(� ))d�
0
� V (x(0)) ; � t�
which implies that V (x(t)) will b e negative which will result in a contradiction establishing
must b e zero. the fact that c
Example 13.2 Consider the dynamical system which is governed by the di�er-
ential equation
x_ � ;g (x)
where g (x) has the form given in Figure 13.4. Clearly the origin is an equilibrium
p oint. If we de�ne a function
Z
x
V (x) � g (y )dy
0
then it is clear that V (x) is lo cally p ositive de�nite (lp d) and
2
_
V (x) � ;g (x)
which is lo cally negative de�nite (lnd). This implies that x � 0 is an asymptotically
p oint. stable equilibrium g(x)
-1 1 x
Figure 13.4: Graphical Description of g (x)
Lyapunov Theorem for Global Asymptotic Stability
The region in the state space for which our earlier results hold is determined by the region
over which V (x) serves as a Lyapunov function. It is of sp ecial interest to determine the
p oint, i.e. the set of initial \basin of attraction" of an asymptotically stable equilibrium
subsequent tra jectories end up at this equilibrium p oint. An equilibrium conditions whose
p oint is global ly asymptotical ly stable (or asymptotically stable \in the large") if its basin of
entire state space. attraction is the
If a function V (x) is p ositive de�nite on the entire state space, and has the additional
_
prop erty that jV (x)j % 1 as kxk % 1, and if its derivative V is negative de�nite on the
entire state space, then the equilibrium p oint at the origin is globally asymptotically stable.
We omit the pro of of this result. Other versions of such results can b e stated, but are also
omitted.
Example 13.3
Consider the nth-order system
x_ � ;C (x)
0
with the prop erty that C (0) � 0 and x C (x) � 0 if x 6� 0. Convince yourself that
p oint of the system is at 0. Now consider the candidate the unique equilibrium
Lyapunov function
0
V (x) � x x
which satis�es all the desired prop erties, including jV (x)j % 1 as kxk % 1.
Evaluating its derivative along tra jectories, we get
0 0
_
V (x) � 2x x_ � ;2x C (x) � 0 for x �6 0
Hence, the system is globally asymptotically stable.
Example 13.4 Consider the following dynamical system
x_ � ;x + 4x
1 1 2
3
x_ � ;x ; x :
2 1
2
The only equlibrium p oint for this system is the origin x � 0. To investigate the
2 2
stability of the origin let us prop ose a quadratic Lyapunov function V � x + ax ,
1 2
p ositive constant to b e determined. It is clear that V is p ositive where a is a
2
de�nite on the entire state space R . In addition, V is radially unb ounded, that
jV (x)j % 1 as kxk % 1. The derivative of V along the tra jectories is it satis�es
given by of the system is
" #
h i
;x + 4x
1 2
_
V � 2x 2ax
1 2
3
;x ; x
1
2
2 4
� ;2x + (8 ; 2a)x x ; 2ax :
1 2
1 2
, and the If we cho ose a � 4 then we can eliminate the cross term x x derivative
1 2
b ecomes of V
2 4
_
V � ;2x ; 8x �
1 2
which is clearly a negative de�nite function on the entire state space. Therefore
we conclude that x � 0 is a globally asymptotically stable equilibrium p oint.
Example 13.5 A highly studied example in the area of dynamical systems and
3
chaos is the famous Lorenz system, which is a nonlinear system that evolves in R
whose equations are given by
x_ � � (y ; x)
y_ � r x ; y ; xz
z_ � xy ; bz �
where � , r and b are p ositive constants. This system of equations provides an
approximate mo del of a horizontal �uid layer that is heated from b elow. The
warmer �uid from the b ottom rises and thus causes convection currents. This
approximates what happ ens in the atmosphere. Under intense heating this mo del
b ehaviour. However, in this example we would like to exhibits complex dynamical
stability of the origin under the condition r � 1, which is known not to analyze the
2 2 2
, , and + � y + � z , where � � b ehaviour. Le us de�ne V � � x lead to complex
2 3 1 2 1
� are p ositive constants to b e determined. It is clear that V is p ositive de�nite
3
3
on R and is radially unb ounded. The derivative of V along the tra jectories of the
given by system is
2 3
� (y ; x)
h i
6 7
_
y z V � 2� x 2� 2� r x ; y ; xz
4 5
1 2 3
xy ; bz
2 2 2
y � ;2� � x ; 2� ; 2� bz
1 2 3
� ) + +xy (2� + 2r � (2� ; 2� )xy z :
1 2 2 3
1
_
If we cho ose � � � � 1 and � � then the V b ecomes
2 3 1
�
� �
2 2 2
_
V � ;2 x + y + 2bz ; (1 + r )xy
" #
� � � �
2
1 1 + r
2 2 2
� ;2 x ; (1 + r )y + 1 ; ( ) y + bz :
2 2
1+r
_
Since 0 � r � 1 it follows that 0 � � 1 and therefore V is negative de�nite on
2
3
the entire state space R . This implies that the origin is globally asymptotically
stable.
Example 13.6 (Pendulum)
The dynamic equation of a p endulum comprising a mass M at the end of a rigid
ro d of length R is but massless
�
M R � + M g sin � � 0
where � is the angle made with the downward direction, and g is the acceleration
_
due to gravity. To put the system in state-space form, let x � � , and x � � �
1 2
then
x_ � x
1 2
g
x_ � ; sin x
2 1
R
Take as a candidate Lyapunov function the total energy in the system. Then
1
2 2
V (x) � ) � kinetic + M R x + M g R (1 ; cos x p otential
1
2
2
" #
dV
x
2
2
_
V � ] f (x) � [M g R sin x M R x
2 1
g
; sin x
dx
1
R
� 0
Hence, V is a Lyapunov function and the system is stable i.s.L. We cannot conclude
asymptotic stability with this analysis.
Consider now adding a damping torque prop ortional to the velo city, so that the
state-space description b ecomes
x_ � x
1 2
g
x_ � ;D x ; sin x
2 2 1
R
With this change, but the same V as b efore, we �nd
2 2
_
V � ;D M R x � 0:
2
From this we can conclude stability i.s.L. We still cannot directly conclude asymp-
_
_
totic stability. Notice however that V � 0 ) � � 0. Under this condition,
� �
� ;(g �R ) sin � : Hence, � 6� 0 if � �6 k � for integer k , i.e. if the p endulum is not �
vertically down or vertically up. This implies that, unless we are at the b ottom or
� _
_
�6 0 when V � 0, so � will not remain at velo city, we shall have � top with zero
Lyapunov function will b egin to decrease again. The only place 0, and hence the
velo city, hanging vertically down or the system can end up, therefore, is with zero
vertically up, i.e. at one of the two equilibria. The formal pro of of this standing
invariant set theorem") is b eyond the scop e result in the general case (\LaSalle's
of this course.
The conclusion of lo cal asymptotic stability can also b e obtained directly through
- alternative choice of Lyapunov function. Consider the Lyapunov function can an
didate
1 1
2 2
V (x) � x + (x + x ) + 2(1 ; cos x ):
2 1 1
2
2 2
It follows that
2 2
_
_
V � ;(x + x sin x ) � ; ; (� + � sin � ) � 0:
1 1
2
2 2
_ _ _
_
Also, � + � sin � � 0 ) � � 0� � sin � � 0 ) � � 0� � � 0: Hence, V is strictly
negative in a small neighb orho o d around 0. This proves asymptotic stability.
Discrete-Time Systems
Essentially identical results hold for the system
x(k + 1) � f (x(k )) (13.9)
_
provided we interpret V as
4
_
V (x) � V (f (x)) ; V (x) �
i.e. as
state ) ; V (present state ) V (next
Example 13.7 (DT System)
Consider the system
x (k )
2
x (k + 1) �
1
2
1 + x (k )
2
x (k )
1
x (k + 1) �
2
2
1 + x (k ) 2
which has its only equilibrium at the origin. If we cho ose the quadratic Lyapunov
function
2 2
V (x) � x + x
1 2
we �nd
� �
1
_
V (x(k )) � V (x(k )) ; 1 � 0
2
2
[1 + x (k )]
2
from which we can conclude that the equilibrium p oint is stable i.s.L. In fact,
ab ove relations more carefully (in the same style as we did for the examining the
p endulum with damping), it is p ossible to conclude that the equilibrium p oint is
global ly asymptotical ly stable. actually
Notes
The system in Example 2 is taken from the eminently readable text by F. Verhulst, Nonlinear
Di�erential Equations and Dynamical Systems, Springer-Verlag, 1990.
Exercises
Exercise 13.1 Consider the horizontal motion of a particle of unit mass sliding under the in�uence
gravity on a frictionless wire. It can b e shown that, if the wire is b ent so that its height h is given of
by h(x) � V (x), then a state-space mo del for the motion is given by
�
x_ � z
d
z_ � ; V (x)�
�
dx
4 2
Supp ose V (x) � x ; �x .
�
(a) Verify that the ab ove mo del has (z � x) � (0� 0) as equilibrium p oint for any � in the interval
r
� �
�
;1 � � � 1, and it also has (z � x) � 0� � p oints when � is in the interval as equilibrium
2
0 � � � 1.
(b) Verify that the linearized mo del ab out any of the equilibrium p oints is neither asymptotically
any � in the interval ;1 � � � 1. stable nor unstable for
Exercise 13.2 Consider the dynamic system describ ed b elow:
2
y + cy y� + a y_ + a � u + u_ �
1 2
where y is the output and u is the input.
(a) Obtain a state-space realization of dimension 2 that describ es the ab ove system.
(b) If a � 3� a � 2� c � 2, show that the system is asymptotically stable at the origin.
1 2
(c) Find a region (a disc of non-zero radius) around the origin such that every tra jectory, with an
converges to zero as t approaches in�nity. This is known as initial state starting in this region,
a region of attraction.
Exercise 13.3 Consider the system
(x) dP
x_ (t) � ;
dx
(x) has continuous �rst partial derivatives. The function P (x) is referred to as the potential where P
x b e a gradient system. Let � b e an isolated lo cal function of the system, and the system is said to
) minimum of P (x), i.e. P (x � P (x) for 0 � kx ; � � x � r , some r . k
(a) Show that x� is an equilibrium p oint of the gradient system.
(b) Use the candidate Lyapunov function
V (x) � P (x) ; P (x� )
to try and establish that x� is an asymptotically stable equilibrium p oint.
Exercise 13.4 The ob jective of this problem is to analyze the convergence of the gradient algorithm
n
�
! R and assume that x is a lo cal minimum� i.e., lo cal minimum of a function. Let f : R for �nding a
� �
) � f (x) for all x close enough but not equal to x . Assume that f is continuously di�erentiable. f (x
n
T
: R ! R b e the gradient of f : g Let
@ g @ g
T
g � ( : : : ) :
@ x @ x
1 n
�
It follows from elementary Calculus that g (x ) � 0.
�
, then it is argued that the solution to the dynamic system: go o d estimate of x If one has a
x_ � ;g (x) (13.10)
�
with x(0) close to x will give x(t) such that
�
lim x(t) � x :
t!1
(a) Use Lyapunov stability analysis metho ds to give a precise statement and a pro of of the ab ove
argument.
(b) System 13.10 is usually solved numerically by the discrete-time system
x(k + 1) � x(k ) ; �(x )g (x )� (13.11)
k k
n
) is some function from where �(x R ! R . In certain situations, � can b e chosen as a constant
k
choice is not always go o d. Use Lyapunov stability analysis metho ds for function, but this
) so that give a p ossible choice for �(x discrete-time systems to
k
�
lim x(k + 1) � x :
k !1
(c) Analyze directly the gradient algorithm for the function
1
T
f (x) � x Qx� Q Symmetric, Positive De�nite:
2
�
Show directly that system 13.10 converges to zero (� x ). Also, show that � in system 13.11
b e chosen as a real constant, and give tight b ounds on this choice. can
Exercise 13.5 (a) Show that any (p ossibly complex) square matrix M can b e written uniquely as
0 0
� H and S � ;S . skew-Hermitian matrix S , i.e. H the sum of a Hermitian matrix H and a
0
.) Note that if M is real, then this decomp osition (Hint: Work with combinations of M and M
skew-symmetric matrix. expresses the matrix as the sum of a symmetric and
0 0
(b) With M , H , and S as ab ove, show that the real part of the quadratic form x M x equals x H x,
0 0 0
M x equals x S x. (It follows that if M and x are real, then x M x � x and the imaginary part of
0
H x.) x
0
(c) Let V (x) � x M x for real M and x. Using the standard de�nition of dV (x)�dx as a Jacobian
row vector in this case | whose j th entry is @ V (x)�@ x , show that matrix | actually just a
j
dV (x)
0
� 2x H
dx
where H is the symmetric part of M , as de�ned in part (a).
(d) Show that a Hermitian matrix always has real eigenvalues, and that the eigenvectors asso ciated
eigenvalues are orthogonal to each other. with distinct
Exercise 13.6 Consider the (real) continuous-time LTI system x_ (t) � Ax(t).
(a) Supp ose the (continuous-time) Lyapunov equation
0
P A + A P � ;I (3:1)
has a symmetric, p ositive de�nite solution P . Note that (3.1) can b e written as a linear system
entries of P , so solving it is in principle straightforward� go o d numerical of equations in the
algorithms exist.
0
Show that the function V (x) � x P x serves as a Lyapunov function, and use it to deduce the
stability of the equilibrium p oint of the LTI system ab ove, i.e. to deduce that global asymptotic
eigenvalues of A are in the op en left-half plane. (The result of Exercise 13.5 will b e helpful the
_
in computing V (x).)
What part (a) shows is that the existence of a symmetric, p ositive de�nite solution of (3.1) is
given LTI system is asymptotically stable. The existence of such su�cient to conclude that the
b e necessary, as we show in what follows. [Instead of ;I on the a solution turns out to also
right side of (3.1), we could have had ;Q for any p ositive de�nite matrix Q. It would still b e
p ositive true that the system is asymptotically stable if and only if the solution P is symmetric,
We leave you to mo dify the arguments here to handle this case.] de�nite.
(b) Supp ose the LTI system ab ove is asymptotically stable. Now de�ne
Z
1
0
A t At
P � R (t)dt � R (t) � e e (3:2)
0
The reason the integral exists is that the system is asymptotically stable | explain this in more
Show that P is symmetric and p ositive de�nite, and that it is the unique solution of the detail!
Lyapunov equation (3.1). You will �nd it helpful to note that
Z
1
dR (t)
R (1) ; R (0) � dt
dt
0
The results of this problem show that one can decide whether a matrix A has all its eigenvalues
op en left-half plane without solving for all its eigenvalues. We only need to test for the in the
p ositive de�niteness of the solution of the linear system of equations (3.1). This can b e simpler.
Exercise 13.7 This problem uses Lyapunov's direct metho d to justify a key claim of his indirect
method: if the linearized mo del at an equilibrium p oint is asymptotically stable, then this equilibrium
p oint of the nonlinear system is asymptotically stable. (We shall actually only consider an equilibrium
p oint at the origin, but the approach can b e applied to any equilibrium p oint, after an appropriate
change of variables.)
time-invariant continuous-time nonlinear system given by Consider the
x_ (t) � Ax(t) + h(x(t)) (4:1)
where A has all its eigenvalues in the op en left-half plane, and h(:) represents \higher-order terms", in
kh(x)k�kxk ! 0 as kxk ! 0. the sense that
(a) Show that the origin is an equilibrium p oint of the system (4.1), and that the linearized mo del at
x_ (t) � Ax(t). the origin is just
0
(b) Let P b e the p ositive de�nite solution of the Lyapunov equation in (3.1). Show that V (x) � x P x
quali�es as a candidate Lyapunov function for testing the stability of the equilibrium p oint at
_
the origin in the system (4.1). Determine an expression for V (x), the rate of change of V (x)
tra jectories of (4.1) along
0 2
(c) Using the fact that x x � kxk , and that kP h(x)k � kP kkh(x)k, how small a value (in terms of
_
kP k) of the ratio kh(x)k�kxk will allow you to conclude that V (x(t)) � 0 for x(t) �6 0� Now
you can indeed limit kh(x)k�kxk to this small a value by cho osing a small enough argue that
_
neighb orho o d of the equilibrium. In this neighb orho o d, therefore, V (x(t)) � 0 for x(t) �6 0. By
Lyapunov's direct metho d, this implies asymptotic stability of the equilibrium p oint.
0
Exercise 13.8 For the discrete-time LTI system x(k + 1) � Ax(k ), let V (x) � x P x, where P is a
p ositive de�nite matrix. What condition will guarantee that V (x) is a Lyapunov function symmetric,
involving A and P will guarantee asymptotic stability of the system� for this system� What condition
your answers in terms of the p ositive semide�niteness and de�niteness of a matrix.) (Express MIT OpenCourseWare http://ocw.mit.edu
6.241J / 16.338J Dynamic Systems and Control Spring 2011
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.