ELEC-E8101: Digital and Optimal Control

7. Stability of discrete-time systems

Themistoklis Charalambous
Office number: 2561, TUAS Building

Email: [email protected]

Office hours: Wednesday 16.00-18.00

 In the previous lecture…

We:

• Discretized a continuous-time system in a state-space form to its discrete

counterpart

• Computed the transfer function of a discrete-time system expressed in a

state-space representation

• Derived the stability conditions for the state space form

 Local stability
• Different forms of stability: suppose x1[k] and x2[k] be solutions to a system
with initial conditions x1[k0] and x2[k0], respectively.

2 / 29

- Local asymptotic stability: The solution x1[k] is asymptotically stable if it is

stable and if δ can be chosen such that:

kx2 [k0 ] x1 [k0 ]k < ) kx2 [k] x1 [k]k ! 0, as k ! 1

- (Local)
Stability Stability
concepts of a particular solution (non-linear and/or time-varying systems):
in pictures
The solution x1[k] is stable if for a given ε > 0, there exists δ(ε, k0) > 0, such that:

kx2 [k0 ] x1 [k0 ]k < ) kx2 [k] x1 [k]k < ", 8k k0

δ δ

x eq x eq x eq
ε

Globally asymptotically stable Locally asymptotically stable Locally stable

 Stability of discrete-time LTI systems
• Consider the following discrete-time LTI system:

x[k + 1] = x[k], x[0] = ↵

• To investigate the stability of the system above, its initial value is perturbed:
stems Stability concepts in picture
x0 [k + 1] = x0 [k], x0 [0] = ↵0
tem

x•t+1
Then, the
= f (x t)
difference x̃ = x x0 satisfies:
(1)

i.e. x eq = f (x eq x̃[k
). + 1] = x[k + 1] x0 [k + 1]
= x[k] x0 [k]
globally asymptotically stable if, for every
eq = x̃[k], x̃[0] = ↵ ↵0
hat xt ! x as t ! 1. x eq

locally• asymptotically
This implies that if the
stable solution
near x is stable,
x eq if there is a then
kx0 x eq every
k  other
) xt solution
! x eq asistalso
! 1. stable. For LTI systems,
stability is a property of the system and not of a
(locally)special
stable solution.
if for every (small) " > 0, there Globally asymptotically stable Loc

x eq k  ) kxt x eq k  " for all t 0.

 Stability of discrete-time LTI systems
• The solution to the discrete-time LTI system:

x[k + 1] = x[k], x[0] = ↵

is given by
k
x[k + 1] = x[k] = ( x[k 1]) = . . . = x[0]

• If Φ is diagonalizable: x[k + 1] = k
x[0] = V ⇤k V 1
x[0]

1
• We define y[k] = V x[k] and hence

y[k + 1] = ⇤k y[0], y[0] = V 1

x[0] = V 1
↵

• If Φ is NOT diagonalizable: the solution is instead a linear combination of the

terms
pi [k] ki

where pi [k] are polynomials in k of order one less than the multiplicity of the
corresponding eigenvalue
 Stability of discrete-time LTI systems
• To get asymptotic stability, all solutions must go to 0 as k increases to infinity

Asymptotic stability theorem: The discrete-time LTI system

x[k + 1] = x[k], x[0] = ↵

is asymptotically stable if and only if all eigenvalues of Φ are strictly inside

the unit disk, i.e.,
| i | < 1, i = 1, 2, . . . , n

• If Φ has unique eigenvalues on the circumference of the unit circle with all
other eigenvalues being located inside → steady-state output will perform
oscillations of finite amplitude - system is marginally stable
System

Unstable Stable

Marginally Asymptotically
Stable Stable
 Stability of discrete-time LTI systems

• Distance from origin is a measure of decay rate

1. Distance from origin is a measure of decay rate.

• Complex 2.poles
Complexjust inside
poles unitunit
just inside circle
circle give lightly-damped
give lightly oscillation. Oscillation
damped oscillation.
is possible for real ispoles
3. Oscillation onfornegative
possible real poles on real axisreal axis.
negative

x x
x x

xx x xx
x x
x x

Time responses as a function of the poles location

 Learning outcomes

By the end of this lecture, you should be able to:

• Understand the different notions and forms of stability

• Analyze the stability properties of discrete dynamical systems using:

‣ Jury’s stability criterion,

‣ Nyquist stability criterion, and

‣ Lyapunov’s stability theory

 Stability tests

• The stability analysis of digital control systems is similar to the stability

analysis of analog systems and all the known methods can be applied to
digital control systems with some modifications

• There are good numerical algorithms to compute the eigenvalues; in MATLAB:

>> eig(A)

• Also important to have algebraic or graphical methods for investigating

stability. The most prevalent methods are:

- Direct numerical or algebraic computation of the eigenvalues of Φ

- Methods based on properties of characteristic polynomials (e.g., Jury’s criterion)

- The root locus method

- The Nyquist stability criterion

- The Bode stability criterion

- Lyapunov's method
 Jury’s stability criterion
• Explicit calculation of the eigenvalues of a matrix cannot be done conveniently
by hand for systems of order higher than 2

• In some cases it is easy to calculate the characteristic equation

(z) , |zI | = a0 z n + a1 z n 1
+ . . . + an 1z + an = 0

and investigate its roots

• It is the discrete time analogue of the Routh–Hurwitz stability criterion:

- The Jury stability criterion requires that the system poles are located inside
the unit circle centered at the origin

- The Routh-Hurwitz stability criterion requires that the poles are in the left
half of the complex plane
 Jury’s stability criterion

• The following test is useful for determining if the characteristic equation χ(z)
has all its roots inside the unit disc

a0 a1 ··· an an
rows are the coefficients of χ(z)
1
1stand 2nd
an
in forward and reverse order, respectively an an 1 ··· a1 a0 bn =
a0
3rd row obtained by multiplying the 2nd row n 1
by bn and subtracting this from the 1st a0 an1 1
··· ann 1
1
ann 1
1
ann 1
ann 2
··· an0 1
bn 1 =
4th row is the third row in reverse order 1 1
an0 1
..
.
...

The scheme is then repeated a00

until there are 2n +1 rows,
which includes a single term where
aki 1
= aki bk akk i
akk
bk = k
a0
 Jury’s stability criterion

Jury’s stability test: If a0 > 0 then the characteristic polynomial χ(z) has all
its roots inside the unit circle if and only if
ak0 > 0, k = 0, 1, 2, . . . , n 1

If no a0 = 0 , then the number of negative a0 is equal to the roots of χ(z)

k k

outside the unit circle.

0
• Remark: If 0
a k
> 0, k = 0, 1, 2, . . . , n 1, then the condition 0 > 0 can be
a
shown to be equivalent to the conditions
(1) > 0
( 1)n ( 1) > 0

These conditions constitute necessary conditions for stability and hence

can be used before forming the table.
 In-class exercise

• Consider the system with the following characteristic equation:

(z) = 3z 2 + 2z + 1
Is the system stable?

Solution:
a0 a1 a2 3 2 1
1
a2 a1 a0 1 2 3 b2 =
3
1 8 1 4
a10 a11 3 1⇥ 3 = 3 2 2⇥ 3 = 3
4 8
4/3 1
a11 a10 3 3 b1 = =
8/3 2
8 1 4
a00 3 2 ⇥ 3 = 2

The system is stable: a00 > 0, a10 > 0, a0 > 0

In fact, the roots can easily be solved directly: – 0.3333 ± 0.4714j

 Example using symbolic calculations
• The Jury criterion can easily be used in the case of a small system, when a
computer is not at hand.

• The real power of the Jury criterion comes into action in symbolic calculations.
Stability can be determined as a function of one or even more parameters.

• For example, consider the system with the following characteristic equation:

(z) = z 2 + a1 z + a2

a0 a1 a2 1 a1 a2
a2
a2 a1 a0 a2 a1 1 b2 = = a2
1
a10 a11 1 (a2 )2 a1 (1 a2 )
a1 (1 a2 ) a1
a11 a10 a1 (1 a2 ) 1 (a2 ) 2 b1 = 2
=
1 (a2 ) 1 + a2
a00 2 (a1 )2 (1 a2 )
1 (a2 ) 1+a2
 Example using symbolic calculations

• The system is stable if

(
1 (a2 )2 > 0
2 (a1 )2 (1 a2 )
1 (a2 ) 1+a2 >0

• Factoring differences of squares in the first inequality

1 (a2 )2 = (1 a2 )(1 + a2 ) > 0

• Doing algebraic manipulations to the second inequality

2 (a1 )2 (1 a2 ) (1 a2 )(1 + a2 )2 (a1 )2 (1 a2 )

1 (a2 ) =
1 + a2 1 + a2

(1 a2 )
= [(1 + a2 )2 (a1 )2 ]
1 + a2

(1 a2 )(1 + a2 a1 )(1 + a2 + a1 )
= >0
1 + a2
 Example using symbolic calculations

• The set of inequalities can be simplified to (triangle rule):

9 (
(1 a2 )(1 + a2 ) > 0 = (1 a2 )(1 + a2 ) > 0
(1 a2 )(1 + a2 a1 )(1 + a2 + a1 ) )
he Jury stability criterion
1 + a2
> 0; (1 + a2 a1 )(1 + a2 + a1 ) > 0

8 8 2
he polynomial A( z )
characteristic >
< 1 < a2 < 1 >z az a
< 1 < a21 < 1 2
) a 1 1 < a2 OR a 1 1 > a2
>
: >
:
he values α1 and α2asuch
1 1 <that:
a2 a 1 1 > a2

a2 infeasible
1

a1 1 a1

1 -2 -1 1 2

-1
 Stability in the frequency domain

• The frequency response of G(s) is: G(j!), ! 2 [0, 1). It can be graphically
presented:

- in the complex plane as the Nyquist curve, or,

- as amplitude/phase curves as a function of frequency - Bode diagram

• Correspondingly, for a discrete-time system H(z) the frequency response is

H(ejω), where !h 2 [0, ⇡). This can also be presented graphically as a
discrete-time Nyquist curve or discrete-time Bode diagram

• Note that it is sufficient to consider the map in the interval 
 !h 2 [ ⇡, ⇡)

because the function H(ejω) is periodic with period 2⇡/h
 Stability in the frequency domain

• Consider, for example, the continuous-time system with transfer function:

1
G(s) = 2
s + 1.4s + 1
• Zero-order-hold sampling of the system with h = 0.4 gives the discrete-time
system
0.066z + 0.055
G(z) = 2Stability in frequency domain
in frequency domain z 1.45z + 0.571
sys=tf(1,[1 1.4 1])
Bode Diagrams
Nyquist Diagrams Transfer function: N

0.8 >> sys = 1tf(1,[1 1.4 1]);
 0

---------------
0.6 >> sysd = c2d(sys,0.4,’zoh');
 continuous
s^2 + 1.4 s + 1 -20
0.4 >> w=logspace(2,1,1000)’;


 sysd=c2d(sys,0.4)

Phase (deg); Magnitude (dB)

0.2 -40
Imaginary Axis

discrete
0 >> figure(1); hold on;

-60
Transfer
>> function:
nyquist(sys,w)

-0.2

>>0.06609 z + 0.05481
nyquist(sysd,w)
 0
-0.4 continuous ---------------------
-0.6

 z^2 - 1.45 z + 0.5712 -50

-0.8
discrete >> figure(2); hold on;
 -100 continuous

-0.2 0 0.2 0.4 0.6 0.8 1

>>Sampling
bode(sys,w)

time: 0.4 -150

Real Axis >> bode(sysd,w) -200

discrete
w=logspace(-2,1,1000)'; -250
bode(sys,w) 10-1 100

hold Frequency (rad/sec)

bode(sysd,w)
 Nyquist diagram

YREF(z) X Y(z) YREF(z) X Y(z)

+ C(z) P(z) + L(z)
- ⌘ -

• Nyquist diagram: A plot of the open-loop frequency response of L(z), with the
imaginary part Im(L(e jω)) plotted against the real part Re(L(e jω)) on an Argand
diagram (complex plane).
! jω
"
Im L(e )
∠L(ejω1 )
! jω
"
Re L(e )

|L(ejω1 )|

ω = ω1
 jω
How to find L(e )

• As we have already seen, rational z-transform can be written as

(z µ1 )(z µ2 ) . . . (z µM )
L(z) = c
(z p1 )(z p2 ) . . . (z pN )

• To compute the frequency response of L(ej! ) , we compute L(z) at z = ej!

• But z = ej! represents a point on the circle’s perimeter

by the produc
Im(z) to exp(j!) (A1

• Let Ai = |ej! pi | and Bi = |ej!

given by the su
µi | exp(j!), minus
unit
it circle
i l poles to exp(j
(M-N) !
pd11 x
j! j!
|e µ 1 | |e µ2 | . . . |ej! µM | X
|L(ej! )| = |c| j! , N M A1 z=exp(
|e p1 | |ej! p2 | . . . |ej! pN |
B2 B1
B1 B2 . . . BM !!
= |c| o
O o
O
<latexit sha1_base64="COchiK1BUgCh4UMd4KCR4O3lCz8=">AAAB7XicbVDLSgNBEOyNrxhfUY9eBoPgKeyKqMeAF48RzAOSJcxOepMxszPLzKwQQv7BiwdFvPo/3vwbJ8keNLGgoajqprsrSgU31ve/vcLa+sbmVnG7tLO7t39QPjxqGpVphg2mhNLtiBoUXGLDciuwnWqkSSSwFY1uZ37rCbXhSj7YcYphQgeSx5xR66RmVyU4oL1yxa/6c5BVEuSkAjnqvfJXt69YlqC0TFBjOoGf2nBCteVM4LTUzQymlI3oADuOSpqgCSfza6fkzCl9EivtSloyV39PTGhizDiJXGdC7dAsezPxP6+T2fgmnHCZZhYlWyyKM0GsIrPXSZ9rZFaMHaFMc3crYUOqKbMuoJILIVh+eZU0L6qBXw3uLyu1qzyOIpzAKZxDANdQgzuoQwMYPMIzvMKbp7wX7937WLQWvHzmGP7A+/wBjJWPDw==</latexit>
<latexit

A1 A2 . . . AN -1 cμ11 cμ22 1
A2

\L(ej! ) = \(ej! µ1 ) + \(ej! µ2 ) + . . . + \(ej! µM ) x

X
Thus when
dp2
the magnit

\(ej! p1 ) \(ej! p2 ) ... \(ej! pN ) 2 When exp(j

the magnit
 Discrete-time Nyquist stability criterion

YREF(z) X Y(z) YREF(z) X Y(z)

+ C(z) P(z) + L(z)
- ⌘ -

• Definition (asymptotic stability of a feedback system):


We say that the closed-loop system is asymptotically stable if the closed-loop
transfer function L(z)/(1 + L(z)) is asymptotically stable.

• Closed-loop poles ⌘ poles of L(z)/(1 + L(z)) ⌘ zeros of 1 + L(z)=0

• This corresponds to all the roots lying in the unit circle, i.e., if the characteristic
equation has zeros outside the unit circle, the closed loop system is unstable
 Discrete-time Nyquist stability criterion
Discrete-time Nyquist stability criterion: The closed-loop system 


 YREF(z) X Y(z)

 + L(z)
-


will be stable if (and only if) the number of clockwise encirclements N of the
point −1 by L(e jω) as ω increases from 0 to 2π is equal to
N =Z P
where
Z : # of zeros of χ(z)
: 1 + L(z) = 0 outside the unit circle
P : # of poles of χ(z) : 1 + L(z) = 0 outside the unit circle

• The open loop poles are the same as the poles of the characteristic equation

• The zeros of χ(z) determine the stability of the system so that if the
characteristic equation has zeros outside the unit circle, then the closed
loop system is unstable. The stability criterion is thus obtained by setting
Z = 0 and by demanding that the Nyquist curve encircles the point −1 P times
counterclockwise
 Discrete-time Nyquist stability criterion

• The criterion becomes simple, if the open loop pulse transfer function L(z) has
no poles outside the unit circle. Then, the Nyquist curve must not encircle the
point –1 at all.
! "
Im L(ejω )
! "
Im L(ejω ) ! "
Im L(ejω )

! " ! "
Re L(ejω ) Re L(ejω )
! jω
"
Re L(e )
+ + +
−1 −1 −1

L(z) L(z) L(z)

) ) )
1 + L(z) 1 + L(z) 1 + L(z)
Asymptotically Marginally
Unstable
Stable Stable
 Example

• A discrete process (h = 1) is controlled with a proportional controller, which has

gain K, as shown below
YREF(z) X Y(z) 0.4
+ K H(z) H(z) =
- (z 0.5)(z 0.2)

• The discrete Nyquist diagram is constructed with MATLAB

Nyquist Diagram
1

0.8
• By zooming in at the intersection wit the real
-0.4416
axis, the point is approximately -0.4416
0.6

0.4
Imaginary Axis

0.2

-0.2
• The magnitude can thus be multiplied with
-0.4
(1/0.4416) to reach the critical point −1
-0.6

-0.8

-1
-1 -0.5 0
Real Axis
0.5 1 • The controlled system is stable when
1
>> sysd=zpk([ ],[0.2 0.5],0.4,1);
 K< ⇡ 2.26
>> nyquist(sysd) 0.4416
 Example

• Stability can be determined by direct calculus from the pulse transfer function

0.4
H(z) =
(z 0.5)(z 0.2)

• Substitute z with ej!h = ej! = cos(!) + j sin(!) (Euler formula)

0.4
• Then: H(e ) =
j!
(ej! 0.5)(ej! 0.2)

0.4
=
(cos ! + j sin ! 0.5)(cos ! + j sin ! 0.2)

0.4
=
(cos2 ! sin2 ! 0.7 cos ! + 0.1) + (2 cos ! sin ! 0.7 sin !)j

0.4
=
(2 cos2 ! 0.7 cos ! 0.9) + (2 cos ! sin ! 0.7 sin !)j
 Example

• Setting the imaginary part equal to 0 the intersection point with the real axis is
obtained: (
sin ! = 0
2 cos ! sin ! 0.7 sin ! = 0 ) sin !(2 cos ! 0.7) = 0 )
cos ! = 0.35

• The frequency 0 describes the start point in the Nyquist curve and the
frequency arccos(0.35) the intersection point with the real axis.

• Substitute it to the frequency response function

⇣ ⌘ 0.4 4
jarccos(0.35)
H e = = ⇡ 0.444
(2(0.35)2 0.7(0.35)
<latexit sha1_base64="zTv+lua8F1TOL0e+maLjjnA7w8s=">AAACVXicbZHPS+NAFMcnWXW1/qrr0ctgEdpDQ1K7qAuC4MWjwlaFppbJ9KUdnWTCzItYQv5JL4v/iRfBadODqz4Y+PD9vse8+U6USWHQ918c98fS8srP1bXa+sbm1nZ959e1Ubnm0ONKKn0bMQNSpNBDgRJuMw0siSTcRA/nM//mEbQRKv2L0wwGCRunIhacoZWGdXkRSoixCXfFfZgwnOikYJpzZcqm7x3+bpWhFuMJtk7DWDNe+F63LJqdyrvrtH3vqGJLJ62SnrarPtt1UoYsy7R6orRN7Vy3O6w3fM+fF/0KwQIaZFGXw/pzOFI8TyBFLpkx/cDPcGAXRMEllLUwN5Ax/sDG0LeYsgTMoJinUtIDq4xorLQ9KdK5+nGiYIkx0ySynbOHm8/eTPzO6+cYHw8KkWY5Qsqri+JcUlR0FjEdCQ0c5dQC41rYXSmfMJsK2o+o2RCCz0/+CtcdL7B81W2c/VnEsUr2yD5pkoAckTNyQS5Jj3DyTF4dx3Gdf86bu+SuVK2us5jZJf+Vu/0OBj6taA==</latexit>
0.9) 9

• The gain of the controller K can be multiplied by the factor (1/0.444) in order
the crossing at point −1 to take place. The controlled system is stable, when
9
K < ⇡ 2.25
4
 Example

• Find the values of K for which the following system

YREF(z) X Y(z)
+ K G(z)
-

1
is stable for G(z) = .
Example.
z 2
Solution: 1 1
G(z) = ) G(e ) = j✓
1z 2 j! 1
ej✓ 2
G(z) = ) G(e ) =
z 2 ej! 2
z-plane
Im Im
ee j!
j✓ One unstable pole
-1/3

-1 1
x
2 Re -1 Re
One counter clockwise
encirclement of point
-1/K
1 1 1 1
Closed-loop stable:
Closed-loop stable 11 <
< << ) 11<< K
, K <
< 3
3
KK 3 3
1 z 2+K
check: 1 + K = ) 1<K<3
z 2 z 2
 Open-loop poles on the unit circle

Closing the plot:

Im
Open-loop poles on the unit circle
• To obtain a closed curve for the Nyquist plot x
(and hence correctly count the encirclements), it
is customary to indent the path of ejω around
the poles on the unit circle with a small semi- 1 Re
Closing
circular the locus
excursion outside the unit circle. x

In order to obtain a closed curve for the Nyquist locus and hence
• Then,correctly
the opencount
loopthe encirclements,
poles on the unit it is customary
circle to indent
are counted the stable
as being path in
of z around
the stability the poles on the unit circle with a small semi-circular
criterion.
excursion outside the unit circle. Then the open loop poles on the
unit circle are counted as being stable in the stability criterion.

• The “right turn” in the z-plane gives a “right turn” in the Argand diagram and a
The “right turn” in the z-plane then gives a “right turn” in the G-plane
circular arc of large radius is produced.
and a circular arc of large radius is produced which continues for m⇡
radians where m is the multiplicity of the pole on the unit circle.
 Open-loop poles on the unit circle
Asymptotes:

• If there is an open-loop pole of multiplicity one on the unit circle then the
Nyquist diagram will be asymptotic to a straight line as it tends to infinity.

• It is possible to find the asymptote along which it tends. Suppose G(z) has a
pole at z = 1, i.e.,
1
G(z) = F (z)
z 1

where F(z) has no poles at z = 1. Then for z ≈ 1, expand F(z) in a Taylor series
to give:
⇢
1 0 1 00
G(z) = F (1) + (z 1)F (1) + F (z)(z 1)2 + . . .
z 1 2
F (1)
⇡ + F 0 (1)
(z 1)
 Open-loop poles on the unit circle

• But
large as ! ! 0
j! cos !2 j sin !2
1 1 e 2 1 j
= ! ! ! = ! = ! =
ej! 1 j j
e (e 2
2 j
e 2) 2j sin 2 2j sin 2 2 2 tan !2

• Hence
1 F (1)
G(e ) ⇡ j!
F (1) + F 0 (1) j
2 2 tan !2
constant real part

• For multiple poles on the unit circle the asymptotic behavior is more complex
and requires more terms in the Taylor series expansion.
 2 2 tan(✓/2)
Example
Hence the asymptote as ✓ ! 0 will be a straight line with a constant
real part of 12 F (1) + F 0 (1).
For multiple poles on the unit circle the asymptotic behaviour is more
• complex
Find theand requires
values ofmore terms
K for in thethe
which Taylor series expansion.
following system 37

YREF(z) X Y(z)
+ K G(z)
-
Example.
1 1
is stable
G(z)for
= G(z) = . = 1 , F 0 (z) =
. Then, F (z)
1
z(z 1) z(z 1) z z2
1
and F (1) + F 0 (1) = 1.5
Solution: 2 Im
1
Nyquist K
z-plane diagram
Im

x 1x -1.5 -1 0.5
Re Re
right turn

Closed-loop stable for

1 1
Closed-loop
K
< stable for:
1 , 0 < K < 1 < 1)0<K<1
K 38
 Gain and phase margins
• L(ejω) being close to -1 without encircling it is undesirable for 2 main reasons:

1. It implies that a closed-loop pole will be close to the imaginary axis and
that the closed-loop system will be oscillatory

2. If our plant P(z) is the transfer function of an inaccurate model, then the
“true” Nyquist diagram might actually encircle −1
! jω "
Im L(e )
• The gain margin (GM) is a measure of how
much the gain can be increased before the
closed-loop system becomes unstable
! jω
" 1 1
Re L(e ) GM = =
−1 −α ↵ |L(ej!pc )|
ω = ωpc θ
• The phase margin (PM) is a measure of
ω = ωgc
how much phase lag can be added before
the closed-loop system becomes unstable
L(ejω )
P M = ✓ = ⇡ + \L(ej!gc )
 Gain and phase margins via Bode plots

Gain plot

GM

Phase plot

PM

!gc !pc

• Phase crossover frequency ωpc: frequency at which the phase equals 180o

• Gain crossover frequency ωgc: frequency at which the magnitude equals 1

 Lyapunov’s stability theory
• A useful tool for determining the stability of linear and nonlinear dynamical
systems

• Allows to guarantee asymptotic stability, regions of stability, etc.

• Main idea: introduce a generalized energy function, called the Lyapunov

function, which is zero at the equilibrium and positive elsewhere

• Aleksandr Mikhailovich Lyapunov developed the theory for differential

equations in 1900, but the corresponding theory has also been
developed for difference equations.

• Lyapunov well known for his development of the stability theory of a

dynamical system, as well as for his contributions to mathematical
physics and probability theory.

• The theory passed into oblivion, until N. G. Chetaev realized the

magnitude of the discovery made by Lyapunov. The interest in it
skyrocketed during the Cold War period when it was found to be
applicable to the stability of aerospace guidance systems which
typically contain strong nonlinearities not treatable by other methods.
 Lyapunov’s stability theory
• Geometric illustration of Lyapunov’s theorem:

- Lyapunov function
Lyapunov V(x)in(left)

theorem pictures
- Level curves for Lyapunov function V(x) in the 2-dimensional case (right). All level
Lyapunov function (left); level sets and one state trajectory (right)
curves encircle the origin and do not intersect any other level curve.
10

8
CV,α = {x | V (x) ≤ α}
6

x2
0

−2

−4

−6
x(1) x(2)
x(0)
−8

−10
−10 −5 0 5 10
x1

• Stability:
• once the state enters a level set, it never leaves.
- Once the state enters a level set, it does not leave. Bounded level sets ensure
• bounded level sets ensure bounded trajectories (stability).
bounded trajectories (stability)

• a few additional conditions on V will ensure asymptotic stability

- The equilibrium will be asymptotically stable if we can show that the Lyapunov
7 / 29
function decreases along the trajectories of the system.
 • Basic idea: introduce energy measure, show that does not increase

Lyapunov function
• Allows to guarantee asymptotic stability, regions of local stability, etc
• Sometimes both necessary and sufficient

• V(x) is a Lyapunov function for the system

x[k + 1] = f (x[k]), f (0) = 0
5 / 29

if
1. V(x) is continuous in x and V(0) = 0.

2. V(x) ≥ 0 for all x.

Lyapunov theorem in pictures

3. ΔV(x) = V(x[k + 1]) − V(x[k]) is negative definite.

Lyapunov function (left); level sets and one state trajectory (right)
10

Condition 3 implies that the dynamics of the 8

6
CV,α = {x | V (x) ≤ α}

system is such that the solution always moves 4

toward curves with lower values 2

x2
0

−2

−4

−6
x(1) x(2)
x(0)
−8

−10
−10 −5 0 5 10
x1

• once the state enters a level set, it never leaves.

 Stability theorem of Lyapunov

• From the geometric interpretation → the existence of a Lyapunov function

ensures asymptotic stability

• The following theorem is a precise statement of this fact:

Theorem: The solution x[k] = 0 is asymptotically stable if there exists a

Lyapunov function to the system
x[k + 1] = f (x[k]), f (0) = 0

Further, if
0 < (kxk) < V (x)

where (kxk) ! 1 as kxk ! 1 , then the solution is asymptotically stable for

all initial conditions.

• Main obstacle to using the Lyapunov theory is finding a suitable Lyapunov

function
 Application to Linear systems
• For the linear system
x[k + 1] = x[k], x[0] = ↵
it is straightforward to determine quadratic Lyapunov functions. Take:
V (x) = xT P x

as a candidate Lyapunov function. The increment of V is then given by

V (x) = V (x[k + 1]) V (x[k])
= V ( x[k]) V (x[k])
= xT [k] T
P x[k] xT [k]P x[k]
= xT [k] T
P P x[k]
= xT [k]Qx[k]
• For V to be a Lyapunov function is necessary and sufficient that there exists a
positive definite matrix P that satisfies the equation:
T
P P = Q
where Q is positive definite. The equation above is called the Lyapunov equation
 Application to Linear systems

• There is always a solution to the Lyapunov equation when the linear system is
stable. Matrix P is positive definite if Q is positive definite

• One way of determining a Lyapunov function for a linear system is to choose a

positive definite matrix Q and solve the Lyapunov equation. If the solution P is
positive definite then the system is asymptotically stable.
90 Analysis of Discrete-Time Systems Chap. 3

0.4 0
• Example: x[k + 1] = x[k]
0.4 0.6
 
1 0 1.19 0.25
Using Q = )P =
0 1 0.25 2.06
2 0 .
....
00.

Level curves for V (x) = xT P x

Trajectories for some starting values of x
-2 o 2
State Xl

Figure 3.9 Level curves of V(x} and trajectories for different initial values
of the system in Example 3.6. The sampling points are indicated by dots.
 Learning outcomes

By the end of this lecture, you should be able to:

• Understand the different notions and forms of stability

• Analyze the stability properties of discrete dynamical systems using:

‣ Jury’s stability criterion,

‣ Nyquist stability criterion, and

‣ Lyapunov’s stability theory