Lectures on Dynamic Systems and

Control

Mohammed Dahleh Munther A. Dahleh George Verghese

Department of Electrical Engineering and Computer Science

Massachuasetts Institute of Technology1

1�
c
Chapter 15

External Input-Output Stability

15.1 Introduction
In this lecture, we introduce the notion of external, or input-output, stability for systems.
There are many connections between this notion of stability and that of Lyapunov stability
which we discussed in the previous two chapters. We will only make the connection in the LTI
case. In addition, we will point out the fact that the notion of input-output stability depends
in a non-trivial fashion on the way we measure the inputs and the outputs.

15.2 Signal Measures

The signals of interest to us are de�ned as maps from a time set into R n . A continuous-time
signal is a map from R ! R n , and a discrete-time signal is a map from Z ! R n . If n � 1 we
have a scalar signal, otherwise we have a vector-valued signal. It is helpful, in understanding
the various signal measures de�ned below, to visualize a discrete-time signal w(k) as just
a vector of in�nite (or, if our signal is de�ned only for non-negative time, then a vector of
semi-in�nite) length or dimension, concretely representing it as the array
0 . 1
.. C 0 w(0) 1

B
B C

B
B w(0) CCC or B@ w(1) CA : (15.1)
B w
@ . A(1)

.
..

..

Three of the most commonly used DT signal measures are then natural generalizations of
the �nite-dimensional vector norms (1-, 2- and 1-norms) that we have already encountered
in earlier chapters, generalized to such in�nite-dimensional vectors. We shall examine these
three measures, and a fourth that is related to the 2-norm, but is not quite a norm. We shall
also de�ne CT signal measures that are natural counterparts of the DT measures.
 The signal measures that we study below are:
1. peak magnitude (or 1-norm)�
2. energy (whose square root is the 2-norm)�
3. power (or mean energy, whose square root is the \rms" or root-mean-square value)�
4. \action" (or 1-norm).
Peak Magnitude: The 1-Norm
The 1-norm kwk1 of a signal is its peak magnitude, evaluated over all signal components
and all times :
kwk1 �4 max magnitude of w
4
� sup max
i
jwi (k)j � sup kw(k)k1 (for DT systems) (15.2)
k k
4
� sup max
t i
jwi (t)j � sup
t
kw(t)k1 (for CT systems) � (15.3)
where wi (k) indicates the i-th component of the signal vector w(k). Note that kw(k)k1
denotes the 1-norm of the signal value at time k, i.e. the familiar 1 norm of an n-vector,
namely the maximum magnitude among its components. On the other hand, the notation
kwk1 denotes the 1-norm of the entire signal. The \sup" denotes the supremum or least
upper bound, the value that is approached arbitrarily closely but never (i.e., at any �nite
time) exceeded. We use \sup" instead of \max" because over an in�nite time set the signal
magnitude may not have a maximum, i.e. a peak value that is actually attained | consider,
for instance, the simple case of the signal
1 ; 1 +1 jkj �
which does not attain its supremum value of 1 for any �nite k.
Note that the DT de�nition is the natural generalization of the standard 1-norm for
�nite-dimensional vectors to the case of our in�nite vector in (15.1), while the CT de�nition is
the natural counterpart of the DT de�nition. This pattern is typical for all the signal norms
we deal with, and we shall not comment on it explicitly again.

Example 15.1 Some bounded signals:

(a) For w(t) � 1, t 2 R � t � 0:
kwk1 � 1.

(b) For w(t) � at � t 2 Z:

kwk1 � 1 if jaj 6� 1 and kwk1 � 1 otherwise.

 The space of all signals with �nite 1-norm are generally denoted by `1 and L1
for DT and CT signals respectively. For vector-valued signals, the size of the
vector may be explicitly added to the symbol, e.g., `n1 . These form normed-vector
spaces.

Energy and the 2-Norm

The 2-norm of a signal is the square root of its \energy", which is in turn de�ned as the sum
(in DT) or integral (in CT) of the squares of all components over the entire time set:
kwk2 �4 square-root of energy in w
" # 12 "X # 12
4 X T
� w (k)w(k) � kw(k)k22 (for DT systems) (15.4)
k k
4
�Z � 21 �Z � 12
� wT (t)w(t) dt � kw(t)k22 dt (for CT systems) : (15.5)
t
Example 15.2 Some examples:
(a) For w(t) � e;at and time set t � 0, with a � 0:
kwk2 � p12a � 1

(b) For w(t) � 1 and time set t � 0:

kwk2 � 1

(c) For w(t) � cos !ot and time set t � 0:

kwk2 � 1.

These examples suggest that bounded-energy signals go to zero as time progresses. For
discrete-time signals, this expectation holds up: if kwk2 � 1, then kw(k)k ;! 0 as k ;! 1.
However, for continuous-time signals, the property of having bounded energy does not imply
that kw(t)k ;! 0 as t ;! 1, unless additional assumptions are made. This is because
continuous-time bounded energy signals can still have arbitrarily large excursions in ampli
tude, provided these excursions occur over su�ciently narrow intervals of time that the integral
of the square remains �nite | consider, for instance, a CT signal that is zero everywhere,
except for a triangular pulse of height k and base 1�k4 centered at every nonzero integer value
k. If the continuous-time signal w(t) is di�erentiable and both w and its derivative w_ have
bounded energy (which is not the case for the preceding triangular-pulse example), then it is
true that kw(t)k ;! 0 as t ;! 1. The reader may wish to verify this fact.
It is not hard to show that DT or CT signals with �nite 2-norms form a vector space.
On the vector space `2 (respectively L2 ) of DT (respectively CT) signals with �nite 2-norm,
one can de�ne a natural inner product as follows, between signals x and y :
"X #
4
hx � y i � xT (k)y(k) (for DT systems) (15.6)
k
 4
�Z �
� xT (t)y(t) dt (for CT systems) : (15.7)
(The 2-norm is then just the square root of the inner product of a signal with itself.) These
particular in�nite-dimensional inner-product vector spaces are of great importance in appli
cations, and are the prime examples of what are known as Hilbert spaces.
Power and RMS Value
Another signal measure of interest is the \power" or mean energy of the signal. One also often
deals with the square root of the power, which is commonly termed the \root-mean-square"
(or \rms") value. For a signal w for which the following limits exist, we de�ne the power by
2 3
4 1 X
N ; 1
Pw � lim 4
N !1 2N k�;(N ;1)
wT (k)w(k)5 (for discrete ; time systems) (15.8)
" ZL #
�4
lim 1
L!1 2L ;L
wT (t)w(t)dt (for continuous ; time systems) : (15.9)
(The above de�nitions assume that the time set is the entire time axis, but the necessary
modi�cations for other choices of ptime set should be obvious.) We shall use the symbol �w
to denote the rms value, namely Pw . The reason that �w is not a norm, according to the
technical de�nition of a norm, is that �w � 0 does not imply that w � 0.
Example 15.3 Some �nite-power signals:
(a) For w(t) � 1 :
�w � 1

(b) For w(t) such that kwk2 � 1:

�w � 0

(c) For w(t) � cos !0 t (with t 2 R or t 2 Z):

�w � p12 .

Example c) points out an important di�erence between bounded power and bounded energy
signals: unlike bounded energy signals, if �w � 1, the signal doesn't necessarily decay to
zero.
As a �nal comment on the de�nition of the power of a signal, we elaborate on the hint
in the preamble to our de�nition that the limit required by the de�nition may not exist for
certain signals. The limit of a sequence or function (in our case, the sequence or function is the
set of �nite-interval rms values, considered over intervals of increasing length) may not exist
even if the sequence or function stays bounded, as when it oscillates between two di�erent
�nite values. The following signal is an example of a CT signal that is bounded but does not
have a well-de�ned power, because the required limit does not exist:
(
t 2 [22k � 22k+1 ], for k � 0� 1� 2� : : :
w(t) � 10 ifotherwise
Also note that the desired limit may exist, but not be �nite. For instance, the limit of a
sequence is +1 if the values of the sequence remain above any chosen �nite positive number
for su�ciently large values of the index.
Action: The 1-Norm
The 1-norm of a signal is also sometimes termed the \action" of the signal, which is in turn
de�ned as the sum (in DT) or integral (in CT) of the 1-norm of the signal value at each time,
taken over the entire time set:
kwk1 �4 "action of w #
�4 X kw(k)k (for discrete ; time systems) (15.10)
1
�Z k �
4
� kw(t)k1 dt (for continuous ; time systems) : (15.11)
Recall that kw(k)k for the n-vector w(k) denotes the sum of magnitudes of its components.
The space of all signals with �nite 1-norm are generally denoted by `1 and L1 for DT
and CT signals respectively. These form normed-vector spaces.
We leave you to construct examples that show familiar signals of �nite and in�nite 1
norm.
Relationships Among Signal Measures
a) If w is a discrete-time sequence, then
kwk2 � 1 �) kwk1 � 1 (15.12)
but
kwk2 � 1 (
6 � kwk1 � 1 (15.13)
b) If w is a continuous-time signal, then
kwk2 � 1 �6 ) kwk1 � 1: (15.14)
and
kw k 2 � 1 (
6 � kw k 1 � 1 : (15.15)
c) If kwk1 � 1, then (when �w exists)
�w � kwk1
Item a) is true because of the relationship between energy and magnitude for discrete-time
signals. Since the energy of a DT signal is the sum of squared magnitudes, if the energy is
bounded, then the magnitude must be bounded. However, the converse is not true |take for
example, the signal w(k) � 1. As item b) indicates, though, bounded energy implies nothing
about the boundedness of magnitude for continuous time signals.
(Many more relationships of the above form can be stated.)

15.3 Input-Output Stability

At this point, it is important to make a connection between the stability of a system and its
input-output behavior. The most important notion is that of `p -stability (p-stability).
De�nition 15.1 A system with input signal u and output signal y that is obtained from u
through the action of an arbitrary operator H , so y � H (u), is `p-stable or p-stable (p �
1� 2� 1) if there exists a �nite C 2 R such that
kykp � C kukp (15.16)
for every input u.
A p-stable system is therefore characterized by the requirement that every input of �nite
p-norm gives rise to an output of �nite p-norm. For the case p � 1, this notion is known
as Bounded-Input Bounded-Output (BIBO) stability. We will see that BIBO stability is
equivalent to p-stability for �nite-dimensional LTI state-space systems, but not necessarily in
other cases.
Example 15.4 The system described by one integrator:
y_ � u
is not BIBO stable. A step input is mapped to a ramp which is unbounded. It is
not hard to see that this system is not p-stable for any p.
15.3.1 BIBO Stability of LTI Systems
A continuous-time LTI system may be characterized by its impulse response matrix, H( � ),
whose (i� j )th entry hij ( � ) is the impulse response from the j th input to the ith output. In
other words the input-output relation is given by
Z
y(t) � H(t ; � )u(� )d� :
Theorem 15.1 A CT LTI system with m inputs, p outputs, and impulse response matrix
H(t) is BIBO stable if and only if
m Z
X
max
1�i�p
jhij (t)j dt � 1:
j �1
Proof: The proof of su�ciency involves a straightforward computation of bounds. If u is an
input signal that satis�es kuk1 � 1, i.e. a bounded signal, then we have
Z
y(t) � H(t ; � )u(� )d��
and
���Z X ��
���
m

max j y ( t ) j � max �� h ( t ; � ) u ( � ) d�

i ij j

1�i�p i � �

2 Z j�1 3

� 4max
i

jhij (t ; � )j d� 5 max sup ju (t)j:

j t j

j
It follows that
2 Z 3
X

kyk1 � sup max jyi (t)j � 4max jhij (t)jdt5 kuk1 � 1:

t i i

j
In order to prove the converse of the theorem, we show that if the above integral is
in�nite then there exists a bounded input that will be mapped to an unbounded output. Let
us consider the case when p � m � 1, for notational simplicity (in the general case, we can
still narrow the focus to a single entry of the impulse response matrix). Denote the impulse
response by h(t) for this scalar case. If the integral
Z
jh(t)j dt

is unbounded then given any (large) M there exists an interval of length 2T such that
ZT
jh(t)j dt � M:
;T
Now by taking the input uM (t) as
(
uM (t) � sgn (h(;t)) ;T � t � T �

jtj � T
we obtain an output yM (t) that satis�es
ZT
sup jyM (t)j � yM (0) � h(0 ; � )uM (� ) d�
t ;T
ZT
� jh(0 ; � )j d�
;T
� M:
In other words, for any M � 0, we can have an input whose maximum magnitude is 1 and
whose corresponding output is larger than M . Therefore, there is no �nite constant C such
that the inequality (24.3) holds.
Further re�ection on the proof of Theorem 15.1 reveals that the constant kHk1 de�ned by
XZ
kHk1 � max
i
jhij (t)jdt
j
is the smallest constant C that satis�es the inequalty (24.3) when p � 1. This number is
called the `1 -norm of H(t). In the scalar case, this number is just the `1 ;norm of h( � ),
regarded as a signal.
The discrete-time case is quite similar to continuous-time where we start with a pulse
response matrix, H( � ), whose (i� j )th entry hij ( � ) is the pulse response from the j th input to
the ith output. The input-output relation is given by
X
y(t) � H(t ; � )u(� ) :
�
Theorem 15.2 A DT LTI system with m inputs, p outputs, and pulse response matrix H(t)
is BIBO stable if and only if
X
m X
max
1�i�p
jhij (t)j � 1:
j �1 t

In addition, the constant kHk1 de�ned by

XX
kHk1 � max
i
jhij (t)j
j t
is the smallest constant C that satis�es the inequalty (24.3) when p � 1. We leave the proof
of these facts to the reader.
Application to �nite-dimensional State-Space Models
Now consider the application to the following causal CT LTI system in state-space form (and
hence of �nite order) :
x_ � Ax + Bu (15.17)
y � Cx + Du (15.18)
The impulse response of this system is given by
H(t) � CeAtB + D�(t) for t � 0
which has Laplace transform
H (s) � C (sI ; A);1B + D
The system (15.18) is BIBO stable if and only if the poles of H (s) are in the open left half
plane. (We leave the proof to you.) This is in turn guaranteed if the system is asymptotically
stable, i.e. if A has all its eigenvalues in the open left half plane.
Example 15.5 BIBO Stability Doesn't Imply Asymptotic Stability
It is possible that a system be BIBO stable and not asymptotically stable. Consider
the system
� � � �
x_ � 01 10 x + 01 u
y � ( 1 ;1 ) x
This system is not stable since A has an eigenvalue at 1. Nevertheless, thanks
to a pole-zero cancellation, the only pole that H (s) has is at ;1, so the system
is BIBO stable. We shall have much more to say about such cancellations in the
context of reachability, observability, and minimality (the example here turns out
to be unobservable).
Marginal stability of an LTI system, i.e., stability in the sense of Lyapunov but without
asymptotic stability, is not su�cient to guarantee BIBO stability. For instance, consider a
simple integrator, whose transfer function is 1�s.
Time-Varying and Nonlinear Systems
Although there are results connecting Lyapunov stability with I/O stability for general time-
varying and nonlinear systems, they are not as powerful as the linear time-invariant case. In
particular, systems may be I/O stable with respect to one norm and not stable with respect
to another. Below are some examples illustrating these facts.
Example 15.6 A Time-Varying System
Consider the time-varying DT system given by:

y(t) � H (u)(t) � u(0):

H is obviously 1-stable with gain less than 1. However, it is not 2-stable.

Example 15.7 A Nonlinear System

Consider the nonlinear system given by:
x_ � ;x + exu� y � x:
The unforced system is linear and is asymptotically stable. On the other hand the
system is not I/O stable. To see this, consider the input u(t) � 1. Since ex � x,
x_ is always strictly positive, indicating that x is strictly increasing. Hence, for a
bounded input, the output is not bounded.
15.3.2 p-Stability of LTI Systems (optional)
In this section we will continue our analysis of the p-stability of systems described through
input-output relations. Let us start with the continuous-time case, and restrict ourselves to
single-input single-output. The input u(t) is related to the output y(t) by
Z
y(t) � h(t ; � )u(� )d�
where h(t) is the impulse response. The following theorem shows that the constant C in 24.3
is always bounded above by khk1 .
Theorem 15.3 If khk1 � 1 and kukp � 1 then kykp � 1 and furthermore
kykp � khk1 kukp :
Proof: In Theorem 15.1 we have already established this result for p � 1. In what follows
p � 1� 2. The output y(t) satis�es
�Z 1 �p �Z 1 �p
jy(t)jp � j(h � u)(t)jp � ��� h(t ; � )u(� ) d� ��� � jh(t ; � )j ju(� )j d�
;1 ;1
therefore,
Z1 Z 1 �Z 1 �p
kh � ukpp � j(h � u)(t)jp dt � jh(t ; � )j ju(� )j d� dt :
;1 ;1 ;1
Next we analyze the inner integral
Z1 Z1
jh(t ; � )j ju(� )j d� � jh(t ; � )j1�q jh(t ; � )j1�p ju(� )j d�
;1
�Z;11 �1�q �Z 1 �1�p
� jh(t ; � )j d� jh(t ; � )j ju(� )j d�
p
;1 ;1
where the last inequality follows from Minkowski's inequalities, and 1p + 1q � 1. Hence,
Z 1 �Z 1 �p�q �Z 1 �
kh � ukpp � jh(t ; � )j d� jh(t ; � )j ju(� )jp d� dt
Z;1
1
;1 �Z
1
;1 �
� (khk1 )p�q jh(t ; � )j ju(� )j d� dt
p
;1 Z Z ;1
1 1
� khkp�q
1 ;1 ;1 jh(t ; � )j ju(� )j d� dt
p

p�q
Z1 �Z 1 �
� khk1 ju(� )jp jh(t ; � )j dt d�
;1
Z ;1
� khkp�q +1 1 ju(� )jp d�
1 ;1
� khkp1 kukpp
Therefore
kh � ukp � khk1 kukp :

Recall that when p � 1, khk1 was the smallest constant for which the inequality kykp �
C kukp for all u. This is not the case for p � 2, and we will see later that a smaller constant
can be found. We will elaborate on these issues when we discuss systems' norms later on in
the course. The discrete-time case follows in exactly the same fashion.
Example 15.8 For a �nite-dimensional state-space model, a system H is p-stable
if and only if all the poles of of H (s) are in the LHP. This coincides with BIBO
stability.
 Exercises
Exercise 15.1 Non-causal Systems In this chapter, we only focused on causal operators, although
the results derived were more general. As an example, consider a particular CT LTI system with a
bi-lateral Laplace transform:
G(s) � (s ;s2)(

+2
s + 1) :

(a) Check the p-stability and causality of the system in the following cases:
(i) the ROC (Region of Convergence) is R1 � fs 2 C j Re(s) � ;1g where Re(s) denotes the
real part of s�
(ii)the ROC is R2 � fs 2 C j ; 1 � Re(s) � 2g�
(iii) the ROC is R3 � fs 2 C j Re(s) � 2g.

(b) In the cases where the system is not p-stable for p � 2 and p � 1, �nd a bounded input that
makes the output unbounded, i.e., �nd an input u 2 Lp that produces an output y 26 Lp , for

p � 2� 1.

Exercise 15.2 In nonlinear systems, p-stability may be satis�ed in only a local region around zero.
In that case, a system will be locally p-stable if:
kGukp � C kukp� for all u with kukp � �
Consider the system:
x_ � Ax + Bu
z � Cx + Du
y � g(y)

Where g is a continuous function on [;T� T ]. Which of the following systems is p-stable, locally
p-stable or unstable for p � 1:
(a) g(x) � cos x.
(b) g(x) � sin x.
(c) g(x) � Sat(x) where

� x jxj � 1
Sat(x) � 1 jxj � 1
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.