MTH212 (Chap3)

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 25

Ordinary Differential Equations and Control

Zili Wu

November 9, 2022
Contents

1 The Laplace Transform 3

2 Systems of Linear First-order Differential Equations 4

3 Control Theory 5
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.2 Transfer Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.2.1 Proper Rational Functions . . . . . . . . . . . . . . . . . . . . . . . . 7
3.2.2 Poles and Zeros of Rational Functions . . . . . . . . . . . . . . . . . 8
3.3 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.3.1 Asymptotic stability . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.3.2 Bounded-input bounded-output stability . . . . . . . . . . . . . . . . 9
3.3.3 Hurwitz stability criterion . . . . . . . . . . . . . . . . . . . . . . . . 11
3.3.4 BIBO stabilization by output feedback . . . . . . . . . . . . . . . . . 15
3.3.5 Integral control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.4 Controllability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.5 Observability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2
Chapter 1

The Laplace Transform

3
Chapter 2

Systems of Linear First-order


Differential Equations

4
Chapter 3

Control Theory

3.1 Introduction
Following an initial value problem of a nonhomogeneous system, we will study in this chapter
the following so-called linear, autonomous, single-input, single-output system
ẋ(t) = Ax(t) + bu(t), x(0) = ξ, A ∈ Rn×n , b ∈ Rn ; (3.1)
y(t) = cT x(t) + du(t), c ∈ Rn , d ∈ R, (3.2)
where, for convenience, we denote X(t) and X ′ (t) by x(t) and ẋ(t) respectively, u is referred
to as a scalar-valued input (or control), x the state of the system, and y a scalar-valued
output. In this system, equation (3.1) represents a dynamical system (a way of describing
the passage in time of all points of a given space); once a control u(t) is given, the state x(t)
is completely and uniquely determined; consequently, the output y(t) comes out following
equation (3.2). So such a system is often used to model some real-life phenomena, for
example, the US economy, disease in a patient, air temperature in a classroom.
Control Theory Example 1 Example 2 Example 3
Object x(t) the US economy Illness/Disease Air temperature
(Dynamic Process) in a patient in the classroom
Control u Interest rate of USD Medication in Radiators
(Input) patient treatment in a heating system
Observation y US inflation Medical biopsies Temperature sensors
(Output) in a heating system
To control an object is to influence its behavior so as to achieve some desired objective.
The object to be controlled is usually a dynamic process (a process evolving in time). The
controls are inputs to the process which can influence its evolution. The evolution of the
process may be monitored by observations or outputs. A central concept in control theory
is that of feedback: the use of the information available from the observations or outputs to
generate the controls or inputs so as to cause the process to evolve in some desirable manner.
In what follows we list the topics we want to discuss for a linear, autonomous, single-
input, single-output system in this chapter:
Transfer Functions; Stability; Controllability; Observability.

5
6 CHAPTER 3. CONTROL THEORY

3.2 Transfer Functions


The concept of transfer functions plays an important role in studying control theory. It
comes from equations (3.1) and (3.2), where (3.1) is a special nonhomogeneous system.
Recall that for a nonhomogeneous system we have the following result:

Theorem 3.2.1. Let Φ(t) be a fundamental matrix of X ′ (t) = AX(t). Then the general
solution of X ′ (t) = AX(t) + F (t) is given by
 Z t 
−1
X(t) = Φ(t) C + Φ (s)F (s) ds ,
t0

where C is an arbitrary n × 1 matrix. In particular, the initial value problem

X ′ (t) = AX(t) + F (t) subject to X(t0 ) = X0

has the unique solution


 Z t 
−1 −1
X(t) = Φ(t) Φ (t0 )X0 + Φ (s)F (s) ds .
t0

Given X0 and F , the above formula gives the unique solution to the above nonhomoge-
neous system. When X0 = ξ and F (t) = bu(t), this result determines the unique solution of
(3.1).
For a linear, autonomous, single-input, single-output system (3.1−3.2), by Theorem 3.2.1,
the solution of (3.1) is
 Z t 
tA −sA
x(t) = e ξ+ e bu(s) ds = etA ξ + etA ∗ bu(t).
0

Therefore, by (3.2), the output is

y(t) = cT [etA ξ + etA ∗ bu(t)] + du(t).

If ξ = 0, then y(t) = [cT etA b + dδ(t)] ∗ u(t). Taking Laplace transform, we obtain the Laplace
transform of y(t) as follows

ŷ(s) = [cT (sI − A)−1 b + d]û(s), that is, (3.3)


ŷ(s) = G(s)û(s), where G(s) = cT (sI − A)−1 b + d. (3.4)

This means that, with zero initial data, the function G(s) relates the Laplace transform of
the input u(t) to that of the output y(t). Thus G(s) is said to be the transfer function of
the system.
Recall that the Dirac delta function δ has the Laplace transform L(δ(t)) = 1. If we take
u(t) = δ(t) and f (t) = cT etA b, then

G(s) = L{f (t)} + dL{δ(t)} = L{(f + dδ)(t)}.


3.2. TRANSFER FUNCTIONS 7

From this we see that the transfer function G is the Laplace transform of the function
g(t) := (f + dδ)(t) = cT etA b + dδ(t). (3.5)
Note that g is completely determined by (3.1), (3.2) and impulsive input δ(t). So the function
g is said to be the impulse response of system (3.1 − 3.2). It is easy to see that with ξ = 0
and u(t) = δ(t) the output y(t) of system (3.1 − 3.2) satisfies that
y(t) = (g ∗ δ)(t) = g(t).
Based on (3.4), the transfer function G(s) is determined by A, b, c, d, which completely
describes the system, so it is worth further studying. Now
G(s) = cT (sI − A)−1 b + d
1 N (s)
= [cT adj(sI − A)b] + d = ,
|sI − A| D(s)
where adj M denotes the adjoint matrix of a square matrix M , N (s) and D(s) are polyno-
mials of s. Note that each entry of adj(sI − A) is a polynomial of degree not greater than
n − 1. In addition, since d and the entries of A, b and c are all real numbers, the polynomials
N (s) and D(s) have real coefficients. Thus G(s) is a rational function whose numerator and
denominator have real coefficients.

3.2.1 Proper Rational Functions


For convenience, we introduce some notation and terminology:
ˆ R[s] denotes the set of polynomials in the complex variable s with real coefficients.
n o
ˆ R(s) := Q P
: P, Q ∈ R[s], Q ̸= 0 , the set of rational functions whose numerators and
denominators have real coefficients.
P
Definition 3.2.1. For a rational function R := Q ∈ R(s), R is said to be proper if the
degree of P is less than or equal to that of Q; R is said to be strictly proper if the degree
of P is less than that of Q.
Example 3.2.2. It is easy to see that
s2 +1 s3 +1
3s2 +4
is proper; ss+1
2 +5 is strictly proper; s2 +5
is not proper.
The following property is immediate.
Example 3.2.3. The transfer function G(s) = cT (sI − A)−1 b + d is proper. Moreover, G is
strictly proper if and only if d = 0. Indeed, since
1
G(s) = cT (sI − A)−1 b + d = [cT adj(sI − A)b] + d
|sI − A|
1
and |sI−A| [cT adj(sI − A)b] is strictly proper (Why?), G is proper and lims→∞ G(s) = d.
Thus G is strictly proper if and only if d = 0.
Proposition 3.2.2. Let R ∈ R(s). Then
ˆ R is proper iff lims→∞ |R(s)| < ∞.
ˆ R is strictly proper iff lims→∞ R(s) = 0.
8 CHAPTER 3. CONTROL THEORY

3.2.2 Poles and Zeros of Rational Functions


Definition 3.2.4. A number z ∈ C is a zero of R ∈ R(s) provided that lims→z R(s) = 0.
A number p ∈ C is a pole of R ∈ R(s) provided that lims→p |R(s)| = +∞.

Example 3.2.5. Find the poles and zeros of the following rational functions:

ˆ R(s) = s−1
s−2
;

ˆ R(s) = s−1
s2 −3s+2
;
(s−1)2
ˆ R(s) = s2 −3s+2
.
P
For a rational function R := Q ∈ R(s), to find its zeros z or poles p, we usually cancel the
common factors of P and Q (if there exist) before considering lims→z R(s) or lims→p |R(s)|
so that the numerator and denominator of the resulting rational function have no common
factors, that is, the numerator and denominator are coprime as defined below.

Definition 3.2.6. Two polynomials are said to be coprime provided that they have no
common linear factors.

Example 3.2.7. Discuss whether the following polynomials are coprime or not:

ˆ P (s) = s − 1 and Q(s) = s2 − 3s + 2;

ˆ P (s) = s − 1 and Q(s) = s2 − 5s + 6.

Proposition 3.2.3. For the transfer function

G(s) = cT (sI − A)−1 b + d, where A ∈ Rn×n , b ∈ Rn , c ∈ Rn , d ∈ R,

(i) if p ∈ C is a pole of G, then p is an eigenvalue of A.

(ii) The converse of statement (i) is not true in general.

Proof. (i) Let p ∈ C be a pole of G. We prove (i) by contradiction. Suppose that p ∈ C


were not an eigenvalue of A. Then the determinant |pI − A| =
̸ 0 and

1
G(p) = [cT adj(pI − A)b] + d ∈ C.
|pI − A|

It follows that

1 T

lim |G(s)| = lim
[c adj(sI − A)b] + d = |G(p)| < +∞,
s→p s→p |sI − A|

which contradicts the assumption for p.


(ii) Let p be an eigenvalue of A. If c = 0 or b = 0, then G(s) = d which has no poles and
so p cannot be a pole of G.
3.3. STABILITY 9

The converse of statement (i) is not true in general as the following counterexample
shows. Let    
−1 1 1
A= , b=c= , d = 0.
0 2 0
Then
 −1
T −1 s + 1 −1
T
G(s) = c (sI − A) b + d = c b
0 s−2
  
1   s−2 1 1
= 10
(s + 1)(s − 2) 0 s+1 0
s−2 1
= = .
(s + 1)(s − 2) s+1
From this we see that s = 2 is an eigenvalue of A but not a pole of G.

3.3 Stability
3.3.1 Asymptotic stability
For the homogeneous system

ẋ(t) = Ax(t), A ∈ Rn×n , (3.6)

it is easy to see that x(t) = etA 0 is a solution of (3.6): this is usually said to be the
equilibrium solution of (3.6).
Definition 3.3.1. System (3.6) is said to be asymptotically stable provided that for each
ξ the solution x(t) = etA ξ of the initial-value problem

ẋ(t) = Ax(t), x(0) = ξ

satisfies limt→∞ x(t) = limt→∞ etA ξ = 0. This is equivalent to saying that for each ξ the
solution vector etA ξ approaches the equilibrium solution as t → ∞.
Recall Corollary 2.2.5 and Theorem 2.2.6. Based on them, every solution of (3.6) ap-
proaches the equilibrium as t → ∞ if and only if Re(λ) < 0 for all eigenvalues λ.
Proposition 3.3.1. System (3.6) is asymptotically stable if and only if Re(λ) < 0 for all
eigenvalues λ of A.

3.3.2 Bounded-input bounded-output stability


Consider the linear, autonomous, single-input, single-output system with ξ = 0, that is,

ẋ(t) = Ax(t) + bu(t), x(0) = 0, A ∈ Rn×n , b ∈ Rn ; (3.7)


y(t) = cT x(t) + du(t), c ∈ Rn , d ∈ R, (3.8)

with impulse response g(t) := (f + dδ)(t) = cT etA b + dδ(t).


10 CHAPTER 3. CONTROL THEORY

Definition 3.3.2. System (3.7) − (3.8) is said to be bounded-input, bounded-output


(BIBO) stable provided that for each bounded input function u(t) the corresponding output
function y = g∗u is bounded, that is, there exists a constant γ > 0 such that for each bounded
input function u(t) the output function y = g ∗ u satisfies

sup |y(t)| ≤ γ sup |u(t)|.


t≥0 t≥0

Next result states that the BIBO stability of the system (3.7)−(3.8) can be characterized
by the poles of its transfer function G(s).
Theorem 3.3.2. System (3.7) − (3.8) is BIBO stable if and only if each pole of the transfer
function G(s) = cT (sI − A)−1 b + d has negative real part.
Proof. For convenience, denote g0 (t) := cT etA b and G0 (s) = L{g0 (t)}. Then

g(t) = cT etA b + dδ(t) = g0 (t) + dδ(t),


G(s) = G0 (s) + d, and
y(t) = (g ∗ u)(t) = (g0 ∗ u)(t) + du(t).

To show the necessity, suppose that system (3.7) − (3.8) is BIBO stable. Then there
exists γ > 0 such that each u(t) with supt≥0 |u(t)| = 1 satisfies

|y(m)| = |(g ∗ u)(m)| ≤ γ for all m ∈ N.

For each m ∈ N, define a function um by



 +1 for 0 ≤ t ≤ m with g0 (m − t) ≥ 0;
um (t) = −1 for 0 ≤ t ≤ m with g0 (m − t) < 0;
0 for t > m.

Then 
|g0 (m − t)|, if 0 ≤ t ≤ m;
g0 (m − t)um (t) =
0, if t > m.
And hence
Z m Z m Z m
|g0 (t)|dt = |g0 (m − t)|dt = g0 (m − t)um (t)dt = (g0 ∗ um )(m)
0 0 0
= (g − dδ) ∗ um (m) = (g ∗ um )(m) − dum (m)
≤ |(g ∗ um )(m)| + |d| ≤ γ + |d| for all m ∈ N.
R∞
From this we see that 0 |g0 (t)|dt ≤ γ + |d|. Therefore for any s ∈ C with Re(s) ≥ 0 we
have
Z ∞ Z ∞ Z ∞
−st −tRe(s)

|G0 (s)| =
g0 (t)e dt ≤ |g0 (t)|e dt ≤ |g0 (t)|dt ≤ γ + |d|< ∞.
0 0 0

This implies that G0 has no poles p with Re(p) ≥ 0. Since G(s) and G0 (s) have the same
set of poles, G has no poles p with Re(p) ≥ 0. Hence every pole of G has negative real part.
3.3. STABILITY 11

Next we prove the sufficiency. If G has no poles, then G = G0 + d = d and hence


ŷ(s) = G(s)û(s) = dû(s), from which we obtain y(t) = du(t). So the BIBO property holds
with γ = d.
Now suppose that G has at least one pole. Let pi (i = 1, . . . , l) denote the distinct poles
of G with multiplicities mi . Expanding G0 into partial fractions, we have
mi
l X
X cij
G0 (s) = for some constants cij ∈ C.
i=1 j=1
(s − pi )j

Note that g0 (t) = L−1 {G0 (s)} = li=1 m


P P i
j=1 gij (t), where

cij tj−1 epi t


 
−1 cij
gij (t) := L = for j = 1, . . . , mi and for i = 1, . . . , l.
(s − pi )j (j − 1)!
From this it follows that
Z ∞ Z ∞X mi
l X ∞ mi
l X
|cij |tj−1 eRe(pi )t
Z X
|g0 (t)| dt ≤ |gij (t)| dt = dt := L ∈ R (why?)
0 0 i=1 j=1 0 i=1 j=1
(j − 1)!

since Re(pi ) < 0 for i = 1, . . . , l.


Let u be a bounded input and write U := supt≥0 |u(t)|. Then the corresponding output
satisfies
|y(t)| = |(g ∗ u)(t)| ≤ |(g0 ∗ u)(t)| + |du(t)| = |(u ∗ g0 )(t)| + |du(t)|
Z t Z t
≤ |u(t − τ )g0 (τ )| dτ + |d||u(t)| ≤ U |g0 (τ )| dτ + U |d|
0 0
≤ (L + |d|)U for all t ≥ 0.
Upon taking γ := L + d, we obtain
sup |y(t)| ≤ γU = γ sup |u(t)|
t≥0 t≥0

and hence system (3.7) − (3.8) is BIBO stable.


Next result is immediate from Proposition 3.2.3 and Theorem 3.3.2 and what we can
directly use in practice if A is given.
Proposition 3.3.3. If every eigenvalue of A has negative real part, then system (3.7)−(3.8)
is BIBO stable.

3.3.3 Hurwitz stability criterion


For a given matrix A ∈ Rn×n , Proposition 3.3.1 states that system (3.6) is asymptotically
stable if and only if Re(λ) < 0 for all eigenvalues λ of A, that is, every root of the char-
acteristic equation |sI − A| = 0 has negative real part. Based on Theorem 3.3.2, system
(3.7) − (3.8) is BIBO stable if and only if each pole of the transfer function
N (s)
G(s) = cT (sI − A)−1 b + d =
D(s)
12 CHAPTER 3. CONTROL THEORY

has negative real part (where N (s) and D(s) are coprime if G(s) ̸= d). This implies that
every root of the polynomial equation D(s) = 0 has negative real part. The stability of the
above systems reduces to determining if every root of a relevant polynomial equation has a
negative real part. To discuss the stability of a system, we essentially check if every root of
a polynomial equation has negative real part.
Since the roots of a polynomial equation are completely determined by its coefficients,
it is natural to ask if there are more efficient methods for checking the stability of a system
directly from the coefficients of a polynomial instead of its zeros. What the Routh-Hurwitz
criterion provides us is a method of determining the location of zeros of a polynomial of real
coefficients with respect to the left half and right half of the s-plane based on its coefficients,
without actually solving for the zeros. To study this method, the following concept is needed.
Definition 3.3.3. A polynomial P ∈ R[s] is said to be stable if every root of P (s) = 0 has
a negative real part.
For example, P1 (s) = s2 + 4s + 3 = (s + 1)(s + 3) is stable but P2 (x) = s2 + 2s − 3 =
(s − 1)(s + 3) is not stable.
With the above concept, the following proposition gives us a necessary condition for a
polynomial to be stable.
Proposition 3.3.4. If the polynomial P ∈ R[s] given by

P (s) = an sn + an−1 sn−1 + · · · + a1 s + a0 , where an ̸= 0,

is stable, then all the coefficients am , m = 0, . . . , n, are non-zero and have the same sign.
Proof. By assumption, am ∈ R for m = 0, . . . , Pn and each root rm = αm + iβm of P (s) = 0
satisfies αm < 0 for m = 1, . . . , n and hence nm=1 αm < 0. We prove the conclusion by
induction.
For n = 1, P (s) = a1 s + a0 , where a1 ̸= 0. If r = α + iβ is the root of P (s) = 0,
then a1 (α + iβ) + a0 = 0 and, by assumption, α < 0. This implies that a1 α + a0 = 0 (i.e.,
a0 = (−α)a1 ) and β = 0 and hence a1 and a0 are both non-zero and have the same sign.
For n = 2, P (s) = a2 s2 + a1 s + a0 , where a2 ̸= 0. If rk = αk + iβk with αk , βk ∈ R for
k = 1, 2 are roots of P (s) = 0, then

a2 s2 + a1 s + a0 = P (s) = a2 [s − (α1 + iβ1 )][s − (α2 + iβ2 )]


= a2 {s2 − [(α1 + α2 ) + i(β1 + β2 )]s + [(α1 α2 − β1 β2 ) + i(α1 β2 + α2 β1 )]},

from which we obtain that β1 + β2 = 0 and α1 β2 + α2 β1 = 0, that is,

α1 = α2 < 0 and β1 = −β2 ̸= 0 or α1 + α2 < 0 and β1 = −β2 = 0.

Hence
a2 s2 + a1 s + a0 = P (s) = a2 [s2 − (α1 + α2 )s + (α1 α2 + β12 )].
This implies that a2 , a1 , a0 are nonzero and have the same sign.
Assume that all the coefficients of a stable polynomial of degree n are non-zero and have
the same sign for n = 1, 2, . . . , k. Then, for a stable polynomial P ∈ R[s] of degree n = k + 1,
there exists a root r = α + iβ of P (s) = 0 such that α < 0 and either β = 0 or β ̸= 0.
3.3. STABILITY 13

If β = 0, then r = α and

P (s) = an (s − α)(s − r1 )(s − r2 ) · · · (s − rk ) = an [(s − α)Pk (s)],

where Pk (s) is a polynomial whose coefficients are all real-valued constants and whose zeros
all have negative real parts. By assumption, the k + 1 coefficients of Pk (s) are all positive.
Hence all the coefficients am (m = 0, . . . , k + 1) of P (s) are non-zero and have the same sign.
If β ̸= 0, then r = α − iβ is also a root of P (s) = 0 and

P (s) = ak+1 [s2 − 2αs + (α2 + β 2 )](s − r1 )(s − r2 ) · · · (s − rk−1 )


= an [s2 − 2αs + (α2 + β 2 )]Pk−1 (s),

where Pk−1 (s) = 0 is a polynomial equation whose k coefficients are all real-valued constants
and whose roots all have negative real parts. By assumption, the coefficients of Pk−1 (s) are
all positive. Hence all the coefficients am (m = 0, . . . , k + 1) of P (s) are non-zero and have
the same sign. By induction, the proposition is valid.

In addition to the above proof, we can consider the following ratios:

ˆ an−1
P
an
=− all roots

ˆ an−2
P
an
= products of the roots taken two at a time

ˆ an−3
P
an
= − products of the roots taken three at a time

ˆ ···

ˆ a0
an
= (−1)n products of all the roots.

These ratios must be positive if all the roots of P (s) = 0 have negative real parts which
happens when P is stable.
Based on Proposition 3.3.4, the cubic polynomial P (s) = s3 − s2 + s + 1 is not stable.
How about the polynomial Q(s) = 2s3 + s2 + s + 1? Stable?
Not just a necessary condition, the Hurwitz criterion provides an equivalent condition
for stability of P ∈ R[s] in terms of its coefficients.

Theorem 3.3.5. For a polynomial P ∈ R[s] given by

P (s) = a0 + a1 s + · · · + an−1 sn−1 + an sn , where an ̸= 0,

let
an−1 an
0 0 ··· 0 0 0
an−3 an−2 an−1 an
··· 0 0 0
an−5 an−4 an−3 an−2
··· 0 0 0
Hn := |hij | = ... .. .. .. .. .. .. .. ,

. . . . . . .
0
0 0 0 ··· a2 a3 a4
0
0 0 0 ··· a0 a1 a2
0 0 0 0 ··· 0 0 a0
14 CHAPTER 3. CONTROL THEORY

where hij = an−(2i−j) for 0 ≤ 2i − j ≤ n, hij = 0 otherwise. (Note that am = 0 for m < 0
and m > n.) Denote the principal subdeterminants of H by Hm for m = 1, . . . , n as below

an−1 an 0
an−1 an
H1 = an−1 , H2 = , H3 = an−3 an−2 an−1 , · · · , Hn = a0 Hn−1 .

an−3 an−2
an−5 an−4 an−3

The polynomial P is stable if and only if Hm > 0 for all m = 1, . . . , n.


Proof. For the proof, we browse
http://en.wikipedia.org/wiki/Routh-Hurwitz stability criterion/,
http://en.wikipedia.org/wiki/Derivation of the Routh array/.
Example 3.3.4. Determine conditions on the coefficients which are necessary and sufficient
for stability of the cubic polynomial P (s) ∈ R[s]:
P (s) = as3 + bs2 + cs + d, a > 0.
Solution. For the given cubic polynomial P (s), it can be written as
P (s) = d + cs + bs2 + as3 ,
so the corresponding determinant is

a2 a3 0 b a 0

H3 := |hij | = a0 a1 a2 = d c b
0 0 a0 0 0 d

whose principal subdeterminants are



b a
H1 = b, H2 = = bc − ad, H3 = dH2 = d(bc − ad).
d c
So P (s) is stable iff b > 0, d > 0, and bc − ad > 0.
Example 3.3.5. For what values of the real parameter k is the following polynomial stable?
P (s) = s4 + 6s3 + 11s2 + 6s + k.
Solution. The given polynomial P (s) can be written as
P (s) = k + 6s + 11s2 + 6s3 + s4 ,
so corresponding determinant is

a3 a4 0 0 6 1 0 0

a1 a2 a3 a4 6 11 6 1
H := |hij | = =
0 a 0 a 1 a2
0 k 6 11

0 0 0 a0 0 0 0 k

with principal subdeterminants


H1 = 6, H2 = 60, H3 = 6 × 60 − 6 × 6k = 36(10 − k), H4 = kH3 = 36k(10 − k).
So P (s) is stable iff 10 − k > 0 and k(10 − k) > 0, that is, 0 < k < 10.
3.3. STABILITY 15

3.3.4 BIBO stabilization by output feedback


Consider system (3.7) − (3.8) with d = 0 whose transfer function G is given by

G(s) = cT (sI − A)−1 b.

Suppose that the system is not BIBO stable. A question arises:


Can such a system be rendered BIBO stable through the use of feedback?
Take a new input v(t) by v(t) = u(t) + ky(t), where k ∈ R. Then u(t) = v(t) − ky(t),
where −ky(t) is an output feedback component.

Taking Laplace transform and with reference to the Figure, we obtain

ŷ(s) = G(s)û(s) = G(s)[v̂(s) − kŷ(s)],

from which it follows that


G(s)
ŷ(s) = v̂(s).
1 + kG(s)
So the transfer function Gk of the feedback system from new input v to output y is given by
G(s)
Gk (s) = .
1 + kG(s)
Now our question is whether there exists a value k such that the feedback system is BIBO
stable?
Theorem 3.3.6. Suppose that
(i) G(s) has no zero with non-negative real part and
(ii) lim|s|→∞ sG(s) =: L > 0.
Then there exists k ∗ ≥ 0 such that, for each k > k ∗ ,
G(s)
Gk (s) :=
1 + kG(s)
is the transfer function of a BIBO stable system.
Proof. Write G(s) = N (s)
D(s)
, where N (s) and D(s) are coprime polynomials. Without loss of
generality, we may assume that the leading term of D(s) has coefficient 1 (called monic).
By hypothesis (ii), G(s) must be strictly proper with deg(D) = deg(N ) + 1. Therefore, for
some n ∈ N and real constants a0 , . . . , an , b0 , . . . , bn , we have

N (s) = an sn + · · · + a1 s + a0 , an ̸= 0, and D(s) = sn+1 + bn sn + · · · + b1 s + b0 .


16 CHAPTER 3. CONTROL THEORY

By hypothesis (ii), we also have an = L > 0.


Next, we consider the rational function

1 s D(s) s an D(s) − sN (s)


R(s) := − = − =
G(s) an N (s) an an N (s)
n+1
an (s + bn s + · · · + b0 ) − (an sn+1 + an−1 sn + · · · + a0 s)
n
=
an (an sn + · · · + a0 )
n
(an bn − an−1 )s + sum of terms with degrees < n
=
a2n sn + sum of terms with degrees < n
an bn − an−1
→ as s → ∞. (What does this imply?)
a2n

This implies that R(s) is proper and its poles coincide with the zeros of G(s). Denote

C+
0 := {s ∈ C : Re(s) ≥ 0} (the closed right-half complex plane).

By hypothesis (i), R(s) has no pole in C+


0 . Thus

0 ≤ sup |R(s)| =: k ∗ < ∞. (Why? Think!)


s∈C+
0

Let k > k ∗ . Then Gk (s) must have no pole with non-negative real part. Otherwise
suppose that there were p ∈ C+
0 such that lims→p |Gk (s)| = ∞. Then, since

G(s) 1 1
Gk (s) = = = ,
1 + kG(s) (1/G(s)) + k R(s) + (s/an ) + k
p
R(p) + + k = 0 and p = −an [k + R(p)].
an
Now
|Re(R(p))| ≤ |R(p)| ≤ sup |R(s)| = k ∗ < k, so
s∈C+
0

0 < k − |Re(R(p))| ≤ k + Re(R(p)).


It follows that
0 ≤ Re(p) = −an [k + Re(R(p))] < 0.
This is a contradiction. Thus every pole of Gk (s) has a negative real part, and hence, for
each k > k ∗ , by Theorem 3.3.2, Gk is the transfer function of a BIBO stable system.

Example 3.3.6. Consider the case


   
1 0 1 1
A = 0 0 1 , b = 0 = c,
 d = 0.
1 −1 −1 0

Since the top left element E11 (s) of the matrix


3.3. STABILITY 17

 −1  
s − 1 0 −1 E11 (s) E12 (s) E13 (s)
(sI − A)−1 =  0 s −1  := E21 (s) E22 (s) E23 (s) is
−1 1 s + 1 E31 (s) E32 (s) E33 (s)
2 2
s +s+1 s +s+1
E11 (s) = = 3 ,
|sI − A| s −s−1
 
E11 (s)
s2 + s + 1
G(s) = cT (sI − A)−1 b = cT E21 (s) = E11 (s) = ,
s3 − s − 1
E31 (s)

where we use only the top left element E11 of (sI − A)−1 because of the structure of b and c.
It is easy to see that G(s) satisfies the conditions of Theorem 3.3.6. So there exists k ∗
such that, for each k > k ∗ ,
G(s)
Gk (s) =
1 + kG(s)
is the transfer function of a BIBO stable system. Now

s2 + s + 1 s2 + s + 1
Gk (s) = = .
s3 − s − 1 + k(s2 + s + 1) s3 + ks2 + (k − 1)s + (k − 1)

By Theorem 3.3.5, the denominator polynomial is stable iff k > 1 =: k ∗ . (Check!). Therefore
for each k > 1, by Theorem 3.3.2, Gk (s) is the transfer function of a BIBO stable system.

3.3.5 Integral control


Consider system (3.7) − (3.8) with transfer function G given by

G(s) = cT (sI − A)−1 b + d

under integral control action


Z t
u(t) = k [r(τ ) − y(τ )] dτ ,
0

where k is a real parameter and r is a reference input. The control object is to cause the
output y to approach the reference input r in the sense that y(t) − r(t) → 0 as t → ∞.
By Laplace transform, we have
k
û(s) = [r̂(s) − ŷ(s)]
s
and the overall controlled system takes the form
18 CHAPTER 3. CONTROL THEORY

Note that
k
ŷ(s) = G(s)û(s) = G(s)[r̂(s) − ŷ(s)].
s
Solving this for ŷ(s) yields
kG(s)
ŷ(s) = r̂(s).
s + kG(s)
The transfer function Fk from the reference input r̂ to the output ŷ is given by
ŷ(s) kG(s)
Fk (s) = = .
r̂(s) s + kG(s)
No matter whether system (3.7) − (3.8) is BIBO stable or not, Fk may become the transfer
function of a BIBO stable system on some conditions.
For example, if
1
G(s) = ,
(s + 1)(s + 2)(s + 3)
then
k
Fk (s) = 4 .
s + 6s3 + 11s2 + 6s + k
Corresponding to the denominator of Fk , the Hurwitz determinant is

a3 a4 0 0 6 1 0 0

a1 a2 a3 a4 6 11 6 1
H = |hij | = =
0 a0 a1 a2 0 k 6 11

0 0 0 a0 0 0 0 k
with principal subdeterminants
H1 = 6, H2 = 60, H3 = 6 × 60 − 6 × 6k = 36(10 − k), H4 = kH3 = 36k(10 − k).
Thus the denominator of Fk is stable (and Fk is the transfer function of a BIBO stable
system) iff 0 < k < 10 := k ∗ .
In particular, if the reference input is constant: r(t) = r0 for all t ≥ 0, then r̂(s) = r0 /s.
Now for k ∈ (0, k ∗ ), let fk = L−1 (Fk ). Since Fk is the transfer function of a BIBO stable
system, each pole of Fk has negative real part and hence fk is of exponential order (since it
contains only factors e(x+iy)t with x < 0). Thus
lim y(t) = lim fk (t) ∗ r(t) = lim fk (t) ∗ r0 u(t)
t→+∞ t→+∞ t→+∞
Z t
= r0 lim fk (t) ∗ u(t) = r0 lim fk (t − τ ) dτ
t→+∞ t→+∞ 0
Z t Z ∞
= r0 lim fk (τ ) dτ = r0 fk (τ ) dτ =r0 Fk (0) = r0 ,
t→+∞ 0 0

where u(t) is the unit step function. This shows that the control objective is achieved. More
generally we have the following result:
3.4. CONTROLLABILITY 19

Theorem 3.3.7. Suppose that G(s) is the transfer function of a BIBO stable system satis-
fying G(0) > 0. Then there exists k ∗ > 0 such that for all k ∈ (0, k ∗ ) the feedback system
is BIBO stable. If also the reference input is constant r(t) = r0 for all t ≥ 0, then for all
k ∈ (0, k ∗ ) the output y(t) → r0 as t → ∞.
Proof. Complete the proof as an exercise.

3.4 Controllability
In this section, we will still consider the linear, autonomous, single-input, single-output
system:

ẋ(t) = Ax(t) + bu(t), A ∈ Rn×n , b ∈ Rn , (3.9)


y(t) = cT x(t) + du(t), c ∈ Rn , d ∈ R. (3.10)

For convenience we denote this system by Σ. If d = 0, the system Σ is also denoted by


(c, A, b). The class of admissible input functions u will be denoted by U. Usually U is
basically the class of piecewise continuous or locally integrable functions. It is important
that U is required to satisfy the slicing property, i.e., if u1 ∈ U and u2 ∈ U, then for any
θ > 0 the function u3 , defined by u3 (t) := u1 (t)(0 ≤ t < θ) and u3 (t) := u2 (t)(θ ≤ t), is in U.
In system Σ, the solution x(t) of ODE (3.9) subject to x(0) = x0 is
 Z t 
tA −sA
xu (t, x0 ) = e x0 + e bu(s) ds .
0

Geometrically, this implies that the point xu (t, x0 ) can be reached at time t from the point
x0 . The corresponding output y(t) is given by
 Z t 
T tA −sA
yu (t, x0 ) = c e x0 + e bu(s) ds + du(t)
0
Z t
T tA
= c e x0 + cT e(t−s)A bu(s) ds + du(t).
0

Obviously, xu (t, x0 ) is uniquely determined by x0 and u and so is yu (t, x0 ). Here the appear-
ance of a convolution suggests the use of the Laplace transformation. Consequently, when
x0 = 0, we obtain
ŷ(s) = [cT (sI − A)−1 b + d]û(s) = G(s)û(s).
Note that G(s) is a proper rational function and for the case d = 0 it is strictly proper.
Recall that a pole of G must be an eigenvalue of A but its converse is not valid because
there may be pole-zero cancellation to get G (see Proposition 3.2.3). This shows that the
eigenvalues of A have not fully been used in studying of the system Σ. In other words,
the transfer function G has not completely characterized the system Σ. So it is necessary
to introduce more concepts for studying this system. The concepts of controllability and
observability are what we need for further research.
In this section, we mainly deal with the typical controllability problem which involves
the determination of the control u(t) such that the state xu (t, x0 ) has the desired properties.
20 CHAPTER 3. CONTROL THEORY

Definition 3.4.1. The linear system given by (3.9) is said to be controllable if, for any
initial state x0 and any final state x1 (in Rn ) at the final moment t = T , there exists a control
u(t) such that the corresponding trajectory xu (t, x0 ) of (3.9) satisfies the condition

xu (0, x0 ) = x0 , xu (T, x0 ) = x1 .

ˆ By definition, the linear system (3.9) is controllable iff every point in Rn is reachable
from any point in Rn in a given time interval [0, T ].

To characterize the controllability of (3.9), we introduce more concepts for the case x0 = 0.
Definition 3.4.2. At time t = T , the reachable space WT is the set of points x1 for which
there exists a control u such that xu (T, 0) = x1 , i.e., every point in WT can be reached at
the moment T from the initial state 0. It is easy to see that
Z T 
(T −s)A
WT = e bu(s) ds |u ∈ U ⊆ Rn .
0

System Σ is said to be reachable at time T if WT = Rn , that is, every point in Rn can be


reached at the moment T from the origin.

ˆ If system Σ is reachable at time T , then, for any ponits x0 and x1 in Rn , we have


x1 − eT A x0 ∈ Rn = WT . And hence
Z T
TA TA TA
x1 = e x0 + (x1 − e x0 ) = e x0 + e(T −s)A bu(s) ds = xu (T, x0 ) for some u ∈ U,
0

that is, the point x1 can be reached at time T from x0 .


ˆ Linear system (3.9) is controllable iff it is reachable at time T .

Definition 3.4.3. A point x0 is said to be null controllable at T if the origin can be


reached from x0 , that is, if there exists a control u such that
Z T
TA
e x0 + e(T −s)A bu(s) ds = 0 (that is, xu (T, x0 ) = 0).
0

ˆ A point x0 is null controllable at T by the control u iff −eT A x0 is reachable at T from


0 by u.

Now we present an explicit expression for the reachable space WT .


Theorem 3.4.1. Let η ∈ Rn and T > 0. Then the following statements are equivalent:
(i) η ⊥ WT (that is, η T x = 0 for all x ∈ WT ).
(ii) η T etA b = 0 for all t ∈ [0, T ].
(iii) η T Ak b = 0 for k = 0, 1, 2, . . . .
(iv) η T (b Ab · · · An−1 b) = 0.
3.4. CONTROLLABILITY 21

Proof. (i) ⇔ (ii): If η ⊥ WT , then


Z T
η T e(T −s)A bu(s) ds = 0 for all u ∈ U.
0

T
Taking u(s) = bT e(T −s)A η for s ∈ [0, T ], we obtain
Z T
(η T e(T −s)A b)2 ds = 0,
0

which implies (ii). Conversely, if (ii) is valid, then


Z T
η T e(T −s)A bu(s) ds = 0 for all u ∈ U
0

and hence (i) follows.


(ii) ⇐ (iii) is immediate from the definition of etA . To prove (ii) ⇒ (iii), we consider the
function f (t) = η T etA b (t ∈ (−∞, ∞)) which satisfies 0 = f (k) (0) = η T Ak b for k = 0, 1, 2, . . . .
The implication (iii) ⇒ (iv) is obvious, so it remains to show (iv) ⇒ (iii). According to
the Cayley-Hamilton Theorem (which states if p(t) is the characteristic polynomial for an
n × n matrix A then p(A) is the zero matrix), An is a linear combination of I, A, . . . , An−1 .
By induction, for any k > n, Ak is a linear combination of I, A, . . . , An−1 . Hence (iv) implies
(iii).

Now for T > 0, if WT = Rn , then, for η ∈ Rn , by Theorem 3.4.1, η ⊥ WT iff


η T (b Ab · · · An−1 b) = 0. This implies that η = 0 and hence [b Ab · · · An−1 b] has rank n.
Thus the following result is obtained.

Theorem 3.4.2. The linear system given by (3.9) is controllable iff the following control-
lability matrix
S = [b Ab · · · An−1 b]
has rank n.

Example 3.4.4. Let   


−2 −6 −3
A= and b= .
2 5 2
Prove that the linear system given by (3.9) is not controllable.
Solution. Since     
−2 −6 −3 −6
Ab = = ,
2 5 2 4
the controllable matrix
 
−3 −6 −3 −6
S = [b Ab] = and det(S) =
= 0.
2 4 2 4

Thus the rank of S is less than 2 and hence the system is not controllable.
22 CHAPTER 3. CONTROL THEORY

3.5 Observability
Following the controllability of the system in previous section, we include (3.10) in our study
in this section for discussing the observability of Σ. The questions concerning observability
involve to what extent it is possible to reconstruct the state x when the input u and the
output y are given. We begin with the following concept:

Definition 3.5.1. Two states x0 and x1 are said to be indistinguishable on the interval
[0, T ] if for any input u we have

yu (t, x0 ) = yu (t, x1 ) for all t ∈ [0, T ].

From the definition, the expression of the output, and the Cayley-Hamilton theorem, we
see that

(i) x0 and x1 are indistinguishable if they give rise to the same output for each input u.

(ii) x0 and x1 are indistinguishable on [0, T ] iff cT etA x0 = cT etA x1 for all t ∈ [0, T ].

(iii) cT etA x0 = cT etA x1 for all t ∈ [0, T ] iff cT etA (x0 − x1 ) = 0 for all t ∈ [0, T ].

(iv) x0 and x1 are indistinguishable iff v := x0 − x1 and 0 are indistinguishable.

(v) cT etA v = 0 for all t ∈ [0, T ] iff cT Ak v = 0 for all k = 0, 1, 2, · · · .

(vi) cT Ak v = 0 (k = 0, 1, 2, · · · ) iff cT Ak v = 0 (k = 0, 1, 2, · · · , n − 1).

ˆ These statements show that the indistinguishability of two vectors does not depend on T .

ˆ For x0 and x1 , if they are the same, then they are certainly indistinguishable on the
interval [0, T ]. This is a trivial case.

ˆ The indistinguishability of two vectors is mainly discussed for distinct vectors.

ˆ If x0 and x1 are not indistinguishable on the interval [0, T ], then they must be distinct
and there exists some input u such that

yu (t, x0 ) ̸= yu (t, x1 ) for some t ∈ [0, T ].

In such a case, x0 and x1 are said to be distinguishable.

Definition 3.5.2. The linear system given by (3.9) and (3.10) is said to be observable if
any two distinct states are distinguishable.

By the definition, the system given by (3.9) and (3.10) is observable iff every nonzero
state v and 0 are distinguishable, that is, every nonzero v and 0 are not indistinguishable.
Thus every nonzero v is not a solution of equations in (vi) and hence next result follows
immediately.
3.5. OBSERVABILITY 23

Theorem 3.5.1. The linear system given by (3.9) and (3.10) is observable iff the following
observability matrix  
cT
 cT A 
 
 cT A2 
V = 
 .. 
 . 
T n−1
c A
has rank n.

Example 3.5.3. Let


     
−10 2 1 1 1
A =  −3 −5 2 , b = 0
 and c = −1 .
1 −1 0 1 0

Prove that the linear system given by (3.9) and (3.10) is not observable.
Proof. Since
 
−10 2 1
cT A = 1 −1 0  −3 −5 2 = −7
   
7 −1 ,
1 −1 0
 
T 2 T
  −10 2 1  
c A = (c A)A = −7 7 −1  −3 −5 2 = 48 −48 7 ,
1 −1 0

the observability matrix is


   
cT 1 −1 0
V =  cT A  = −7 7 −1  .
cT A2 48 −48 7

Now,
1 −1 0

det(V ) = −7 7 −1 = 0,
48 −48 7
so the system is not observable.
24 CHAPTER 3. CONTROL THEORY

Autonomous, single-input, single-output (ASISO) system 


x(t) − state of the system
(
ẋ(t) = Ax(t) + bu(t), x(0) = ξ, A ∈ Rn×n , b ∈ Rn

u(t) − a scalar-value input
y(t) = cT x(t) + du(t), c ∈ Rn , d ∈ R 
y(t) − a scalar-value output

( Rt
x(t) = etA [ξ + 0 e−sA bu(s) ds] = etA ξ + etA ∗ bu(t)
y(t) = cT [etA ξ + etA ∗ bu(t)] + du(t)

ŷ(s) = cT (sI − A)−1 ξ + [cT (sI − A)−1 b + d]û(s)


Special Case
ŷ(s) = [cT (sI − A)−1 b + d]û(s) = G(s)û(s) ξ = 0, G(s)-transfer function of the system

G(s) = cT (sI − A)−1 b + d ξ = 0 and u(t) = δ(t)


y(t) = (g ∗ δ)(t) = g(t) := cT etA b + dδ(t) impulse response of the system

G(s) = cT (sI − A)−1 b + d


1
= |sI−A| {cT adj(sI − A)b} + d How to find adjA, the adjoint matrix of A?
=N (s)
D(s)
(proper and coprime) (strictly) proper, zeroes, poles, coprime
G is strictly proper iff d = 0
Prop 3.2.3(pole of G ⇒ eigenvalues of A)

ẋ(t) = Ax(t), x(t) = 0 -equilibrium solution asymptotically stable, Proposition 3.3.1


( (
ẋ(t) = Ax(t) + bu(t), x(0) = 0, A ∈ Rn×n , b ∈ Rn g(t) := cT etA b + dδ(t),
y(t) = cT x(t) + du(t), c ∈ Rn , d ∈ R y =g∗u
BIBO stable, Theorem 3.3.2, Prop 3.3.3

Stable polynomial Proposition 3.3.4, Theorem 3.3.5


( (
ẋ(t) = Ax(t) + bu(t), x(0) = 0, A ∈ Rn×n , b ∈ Rn G(s) = cT (sI − A)−1 b
y(t) = cT x(t), c ∈ Rn The system is not BIBO stable
v(t) = u(t) + ky(t)
Theorem 3.3.6
( (
ẋ(t) = Ax(t) + bu(t), x(0) = 0, A ∈ Rn×n , b ∈ Rn G(s) = cT (sI − A)−1 b + d
Rt
y(t) = cT x(t) + du(t), c ∈ Rn , d ∈ R u(t) = k 0 [r(τ ) − y(τ )] dτ

ASISO system is controllable, S = [b Ab · · · An−1 b] S has rank n (Theorem 3.4.2)


ASISO system is observable, V = [cT cT A . . . cT An−1 ]T V has rank n (Theorem 3.5.1)
Bibliography

[1] Walter Rudin, Principles of Mathematical Analysis, Mc Graw Hill Companies, Inc.,
1976.

[2] Jerzy Zabczyk, Mathematical Control Theory, Birkhauser Boston, 2008.

[3] Dennis Zill and Michael R. Cullen, Differential Equations with Boundary-value Prob-
lems, BrooksCole, Cengage Learning, 2009.

25

You might also like