International Journal of Control, Energy and Electrical Engineering (CEEE)

Copyright – IPCO-2014
Vol.1, pp. 17-22

Construction of Lyapunov function to examine

Robust Stability for Linear System
Beisenbi Mamyrbek#1, Yermekbayeva Janar#2
#
Department of System Analysis and Control, L.N.Gumilyov Eurasian National University
010000 Kazakhstan, Astana, 2, Mirzoyana str.
1 2
beisenbi_ma@enu.kz,beisenbi@mail.ru, erjanar@gmail.com

Abstract —In the recent years, there is a growing interest to the There is provided a method for building Lyapunov function
robust stability approach for representation and control of linear based on geometric interpretation of Lyapunov’s direct
and nonlinear complex systems. This paper detailed describes a (second) method [5] on gradient of dynamic systems in
new approach to ensure robustness for research linear system. relation to some potential function in the state space of
The robust stability of control systems where the controlled plant
dynamic system.
possesses dynamics is relevant today. The originality of this
paper focuses on robust stability analysis of system with design of
Lyapunov function and defines conditions. We describe a method
for construct the Lyapunov function for linear system via
II. MATHEMATICAL SYSTEM
applying geometric interpretation. Finally we made comparative The control system is given by the linear equation.
analysis of examples and for all the examples the stability
conditions of the system executed. The results proved that the .
(1)
area of robust stability can be extended using the proposed x = Ax + Bu, x ∈ R n , u ∈ R m
approach. This work presents some theoretical fundamental and y = Cx, y ∈ R l
practical results and general problem of robust stability is
defined.
Keywords — Stability, Linear systems, Robust control, Control The controller is described by the equation
theory, Lyapunov function.
u = −Kx (2)
I. INTRODUCTION or u i = − k i1 x1 − k i 2 x 2 − ... − k in x n , i = 1,2,..., m
Contemporary control approach is characterized by
Description of parameters
increased dynamics requiring stability and quality control of
processes conditioned by multiple restrictions and incomplete
data. A∈ R n × n , B ∈ R n × m , C ∈ R l × n , K ∈ R m × n
Key task is to investigate the problem of behavioral
change of control systems at ultimate parameters changes and Matrices of the object, control, output and coefficients
control law synthesis that provides for optimal protection from of control system, x ( t ) ∈ R n - state vector, u(t ) ∈ Rm - vector
indefiniteness in research of object properties. control, y(t ) ∈ Rl - vector output of the system.
The robustness is assumed as an ability to maintain We can provide equation (1) in expanded form:
system availability in a condition of parametric and non-
parametric indeterminateness [1,2] describing control objects. .

The most important idea in the study of robust stability is to x1 = a11 x1 + a12 x 2 + ... + a1n x n + b11u1 + b12 u 2 + ... + b1m u m
.
specify constraints for changes in control system parameters x 2 = a 21 x1 + a 22 x 2 + ... + a 2 n x n + b21u1 + b22 u2 + ... + b2 m u m (3)
that preserve stability. ... ...
Studies [3,4] etc dedicated to study robust stability of .
x n = a n1 x1 + a n 2 x2 + ... + a nn x n + bn1u1 + bn 2 u2 + ... + bnm u m
control systems. Nevertheless, many of them focused on
studies of robust stability of linear continuous and discrete Let us denote G = A − BK matrix of the closed system
control systems specifically of characteristic polynomial, and the system (3) in matrix-vector form, we can write
frequency characteristics and Lyapunov matrix equation.
For the purpose of studying the system dynamics and .

their control, we considered models based on observations of x = G x , x(t ) ∈ R n ,

input and output signals of the object and the representing its m
g ij = aij − ∑ bik k kj
behavior in the state space as most suitable. k =1

Therefore equation (3) can be written as

17
 International Journal of Control, Energy and Electrical Engineering (CEEE)
Copyright – IPCO-2014
Vol.1, pp. 17-22

But in this case the integral curves have to be infinitely

 .  b1k k k 1  x1 +  a12 − ∑ b1k k k 2  x2 + ... +  a1n − ∑ b1k k kn  xn close to the origin, i.e. unperturbed motion is asymptotically
m m m

 x 1 =  a11 − ∑ k =1   k =1   k =1  stable [2].

.
 x 2 =  a 21 − ∑ b2 k k k 1  x1 +  a 22 − ∑ b2 k k k 2  x2 + ... +  a 2 n − ∑ b2 k k kn  xn
m m m
(4) Thus, from the geometric interpretation point of view
  k =1   k =1   k =1 
 ... the second method of Lyapunov, the study of stability is
.  m
  m
  m
 reduced to the construction of a family of closed surfaces
 x n =  a n1 − ∑ bnk k k1  x1 +  a n 2 − ∑ bnk k k 2  x2 + ... +  a nn − ∑ bnk k kn  xn
  k =1   k =1   k =1  surrounding the origin. As the integral curves have property to
intersect each of these surfaces, then stability of the
unperturbed motion will be set [2].
III.GEOMETRICAL INTERPRETATION OF LYAPUNOV Let us consider, that the expression dV ( x ) dt < 0
FUNCTION
means, that
Certain functions or so-called Lyapunov functions are dV ( x ) ∂V ( x ) dx dx
= = grad V ( x ) cos α < 0
,
used as basic instruments in Lyapunov direct method and
based on two Lyapunov theorems. Lyapunov theorems have dt ∂x dt dt
simple geometric interpretation. This interpretation not only
determine main contents of the theorem but also could be used i.e. scalar product of the gradient vector Lyapunov
to solve problems in building Lyapunov function. functions grad V ( x ) by the velocity vector dx dt for the
The direct method is a great advantage in the case of asymptotic stability of the system must be less than zero.
nonlinear systems. The method of constructing a Lyapunov This condition will be true if the angle α between the
function for stability determination is called the second gradient of the Lyapunov function grad V ( x ) and the
method of Lyapunov. We use the «second method of
velocity vector dx dt forms an obtuse angle 90 0 < α ≤ 180 0 .
Lyapunov» or the «direct method» as applied to linear
systems. The gradient vector of the Lyapunov function is always
The geometric meaning of a Lyapunov function used directed from the origin toward the highest growth of
for determining the system stability around the zero Lyapunov functions.
equilibrium and it is shown schematically in Figures 1. Also note that, in the study of stability [1] the origin
corresponds to the stationary states of the system or the set of
the system. The state equation (1) or (4) shall be made in
respect to deviations from the steady state
X s (x = ∆x = X (t ) − X s (t ) ) .
Therefore the left side of (1) or (4), dx/ dt expresses the
velocity vector changes and deviations. We can assume that
the velocity vector of deviations submitted to the stability of a
system to the origin.
Components of the gradient vector Lyapunov functions in the
Fig. 1. Lyapunov function opposite direction, but they are equal in absolute value. Then,
if the Lyapunov function V (x) is specified as a vector of
The geometric identification of stable states is reduced functions V (V1 ( x ),V2 ( x ),...,Vn ( x )) then gradient vector
to creating a family of closed surfaces that surround the zero
equilibrium of coordinates. The system state moves across Lyapunov function can be written as
contour curves: each integrated curve can cross each of these ∂ V ∂ x = − dx dt = − ( A − BK )x.
surfaces. Vector components of the gradient of a potential
We suppose that there exists a positive definite function V ( x1 ,..., x n ) are given in the form of vector
function V ( x1 , x2 ,..., xn ) for which (dV dt < 0 ) , and consider Lyapunov functions with components
any integral curve of (3), coming out at the initial time of any (V1 ( x 1 , x 2 ,..., x n ), V 2 ( x1 , x 2 ,..., x n ),..., V n ( x1 , x 2 ,..., x n ) ) we
point of the origin. write in the form:
If dV dt is a function with negative definite
(dV dt < 0 ), then every integral curve starting from a
sufficiently small neighborhood of the origin, will be sure to
cross each of the surfaces V ( x1 (t ), x2 (t ),.., x n (t )) = C , C = const
of the outside to the inside, as the V ( x1 (t ), x2 (t ),.., x n (t )) = C
function is continuously decreasing.

18
 International Journal of Control, Energy and Electrical Engineering (CEEE)
Copyright – IPCO-2014
Vol.1, pp. 17-22

 dx1 ∂V1 ( x )
2
∂V1 ( x ) ∂V ( x )  
bik kk1  x1 +  ai 2 − ∑ bik kk 2  x2 +
m m

 − dt = ∂x +  ai1 − ∑
+ ... + 1
∂x 2 ∂x n dVi ( x)   
 1
(5)
= − k =1
 m

k =1
 , (8)
 − dx 2 = ∂V2 ( x ) + ∂V2 ( x )
+ ... +
∂V2 ( x ) dt +, , , ,+ ain − ∑ bik kkn  xn 
 dt ∂ x1 ∂x 2 ∂x n   k =1  
 ....
 dx n ∂Vn ( x ) ∂Vn ( x ) ∂V ( x ) i = 1,2,..., n
− = + + ... + n
 dt ∂x1 ∂x 2 ∂x n
From the expressions (8) that the total time derivative
of the vector-Lyapunov V i ( x ) functions in the performance of
In this system by substituting values of the components of
the velocity vector we get: the initial assumptions resulting from the geometric
interpretation of a theorem A.M. Lyapunov will be negative
sign function. This means that the conditions for asymptotic
stability of the system will always be performed (4).
 ∂V1 ( x) ∂V1 ( x) ∂V1 ( x)
 ∂x + ∂x +,...,+ ∂x = Now, using components of the gradient vector we will
 1 2 n
restore components of the vector Lyapunov functions:
−  a11 − ∑ b1k k k 1  x1 −  a12 − ∑ b1k k k 2  x2 −,...,− a1n − ∑ b1k k kn  xn ,
m m m

  k =1   k =1   k =1 

Vi ( x1 , x2 ,..., xn ) = − ai1 − ∑ bik kk1  x12 −  ai 2 − ∑ bik kk 2  x22 −,
m m

 (6)
 ∂V2 ( x) + ∂V2 ( x) +,...,+ ∂V2 ( x) =
 k =1   k =1 
− ...,− ain − ∑ bik kkn  xn2 ,
m
 ∂x ∂x2 ∂xn
 
1
 k =1 
m
 x −  a − m b k  x −,...,− a − m b k  x
  21 ∑ 2 k k 1  1  22 ∑ 2 k k 2  2
− a − b k  2 n ∑ 2 k kn  n
  k =1   k =1   k =1  i = 1,2,..., n
 ...
 ...
 ∂Vn ( x) ∂Vn ( x) ∂Vn ( x) The positive definiteness of all components of the
 + +,...,+ = vector Lyapunov function will be expressed by
 ∂x1 ∂x2 ∂xn
 −  a − ∑ b k  x −  a − ∑ b k  x −,...,− a − ∑ b k  x
m m m

  n1 k =1 nk k 1  1  n 2 k =1 nk k 2  2
−  aij − ∑ bik k kj  > 0, i = 1,2,..., n, j = 1,2,.., n

nn
k =1
nk kn

n m
(9)
 k =1 
From here we can find the components of the gradient vector
for the component vector functions This condition characterized superstability of
transposed matrix of a closed system [4].

(V 1
( x 1 , x 2 ,..., x n ), V 2 ( x 1 , x 2 ,..., x n ),..., V n ( x 1 , x 2 ,..., x n ) )
IV.RADIUS OF THE ROBUSTNESS

 ∂V1 ( x)  ∂V ( x)
b1k kk 1  x1 , 1 = − a12 − ∑ b1k kk 2  x2 ,
m m

 ∂x = − a11 − ∑ k =1  ∂x2  k =1 

A. Define the Radius of the Robustness.
 ∂V 1

= − a1n − ∑ b1k kkn  xn

( x) m
 ..., 1 Let us investigate the robust stability of the vector-
 ∂xn  k =1 
 ∂V2 ( x) ∂V ( x) Lyapunov functions. Then let us transform the condition of
 m

= − a21 − ∑ b2 k kk 1  x1 , 2 = − a22 − ∑ b2 k k k 2  x2 ,
m
(7)
 robust stability of the components of the vector Lyapunov
 ∂x1  k =1  ∂x2  k =1 
..., ∂V2 ( x) = − a − m b k  x function. For this, we can turn to a parametric family of
  2 n ∑ 2 k kn  n coefficients the vector-Lyapunov functions, such as the
∂xn  k =1 
 interval family, defined as [4]:
 ∂V ( x) ...
  ∂V ( x)
= − an 2 − ∑ bnk k k 2  x2 ,
m m
 n = − an1 − ∑ bnk kk1  x1 , n
 ∂x1  k =1  ∂x2  k =1  dij = dij0 + ∆ij , ∆ij ≤ γ mij , i, j = 1,2,...,n
 ∂Vn ( x)  a − m b k x
..., = − nn ∑ nk kn  n
 ∂xn  k =1  where the nominal rate

Total time derivative of the components of the vector

d ij0 = − aij0 − ∑ bik0 k kj0 
m

Lyapunov function Vi (x) given by the equation of motion (1)  k =1 

and (4) is determined by
corresponds to a positive-definite Lyapunov functions,
i.e.

19
 International Journal of Control, Energy and Electrical Engineering (CEEE)
Copyright – IPCO-2014
Vol.1, pp. 17-22


min −  a ij0 − ∑ bik0 k kj0  > 0
m
a 22 − b2 k 2 − a11 + b1k1 > 0
σ ( D 0 ) = min 
i j
 k =1   12
( a − b k
1 2
)( a 21
− b2 k1 ) − ( a11 − b1k 1)( a 22 − b2 k 2 ) > 0
Now, we require that the positivity condition
coefficients stored for all functions of the family: We investigate the stability of the system using the idea
of Lyapunov functions.
−  a ij0 − ∑ bik0 k kj0  + ∆ ij > 0, i = 1,2,..., n; j = 1,2,..., n
m
Let us investigate the components of the gradient
 k =1  vector components vector functions V1 ( x1 , x 2 ) and V2 ( x1 , x 2 ) :

Clearly, this inequality holds for all admissible ∆ ij ∂V1 ( x1 , x2 ) ∂V ( x , x )

= −( a11 − b1k1 ) x1 , 1 1 2 = +(a12 − b1k 2 ) x2
if and only if ∂x1 ∂x2
∂V2 ( x1 , x2 ) ∂V2 ( x1 , x2 )
= −(a 21 − b2 k1 ) x1 , = +(a 22 − b2 k 2 ) x2
−  aij0 − ∑ bik0 kkj0  + γ mij > 0,
m
∂x1 ∂x2
 k =1 
i = 1,2,..., n; j = 1,2,..., n
We discover the total time derivative of the Lyapunov
function by the formula (8):
i.e. when
−  aij0 − ∑ bik0 k kj0 
m

  (10) dV ( x1 , x2 )
= −[(a11 − b1k1 ) x1 + (a12 − b1k 2 ) x2 ] −
k =1 2
γ < γ * = min min
i j
mij dt
− [(a21 − b2 k 2 ) x1 + (a22 − b2 k 2 ) x2 ] < 0
2

In particular, if mij = 1 (scale factors of a member of

Lyapunov functions are the same), then The next step - discovering vector Lyapunov functions

γ * = σ ( D0 ) (11) 1
V1 ( x1 , x2 ) = − (a11 − b1k1 )x12 + 1 (a12 − b1 k 2 )22
2 2
1 1
Thus, the stability radius of interval family of positive V2 ( x1 , x2 ) = − (a 21 − b2 k1 )x12 + (a 22 − b2 k 2 )x22
definite functions is the smallest value of the coefficients of 2 2
the vector Lyapunov functions.
Conditions for the stability of the system obtained in
the form:
B.Example for Second Order System − (a11 − b1 k 1 ) > 0, (a12 − b 1 k 2 ) > 0,
− (a 21 − b2 k 1 ) > 0, (a 22 − b2 k 2 ) > 0
As an example, we consider the second order system.
Let n = 2, m = 1, i.e., and
. a a12
x = Ax + Bu , A = 11 , − (a − b k ) > (a
11 1 1 21
− b2 k1 ),
a 21 a 22 (a − b k ) > −(a
22 2 2 12 −b1 k2 )

b1
B=b= , u = − Kx, K = k = k1 k2 From this we can get a system of inequalities
b2
a22 − b2 k 2 − a11 + b1k1 > 0
.
x 1 = (a11 − b1 k1 )x1 − (a12 − b1 k 2 )x 2 (a 12 − b1k 2 )(a 21 − b2 k1 ) − (a11 − b1k1 )(a22 − b2 k 2 ) > 0
.
x 2 = (a 21 − b2 k1 )x1 − (a 22 − b2 k 2 )x 2 . Thus, from (9) and (10) we can determine the radius of
robust stability of a second order system, if system parameters
Then are uncertain:
a11 − b1 k1 − ( a12 − b1k 2 )
G = A + BK = − (a11 − b1k1 ), (a12 − b1k 2 ), 
a 21 − b2 k1 − ( a 22 − b2 k 2 )
γ * = min 
 − (a21 − b2 k1 ), (a22 − b2 k 2 )

With inequality [4] characteristic equation has roots

C.Experimental results
with negative real parts.

20
 International Journal of Control, Energy and Electrical Engineering (CEEE)
Copyright – IPCO-2014
Vol.1, pp. 17-22

Then, as an example, we define the following initial

conditions, find conditions for the stability of the second
order system and the radius and transients.
When the initial settings are follow:

− 11.6 45 9.2
A= ,B = , K = 2 0 .001
11.7 0.4 7

In this case, the radius will be equal ( γ * = 0.3860 ). Fig. 4.The transition process, exp.3.
The overall the transition process of the system shows on the
Figure 2. For 4-d case, when the initial settings are follow:

− 11.6 55 9.9
A= ,B = , K = 2 0 .003
11.7 0.4 7.3

In this case, the radius will be equal ( γ * = 0.3562 ).

The overall the transition process of the system shows on the
Figure 5.

Fig. 2.The transition process, exp.1.

The second case, when the initial settings are follow:

− 11.6 45 9.2
A= ,B = , K = 2 0 .001
11.7 0.4 7.3

In this case, the radius will be equal ( γ * = 0.3854 ).

The overall the transition process of the system shows Fig. 5.The transition process, exp.4.
on the Figure 3.

The short analysis of examples #3 and #4 we can see on

Figure 6 and complex analysis of examples #1- #4 shown on
Figure 7. For all the examples given initial values and the
stability conditions of the system are executed.

Fig. 3.The transition process, exp.2.

The third case, when the initial settings are follow:

− 11.6 55 9.2
A= ,B = , K = 2 0 .003
11.7 0.4 7.3
Fig. 6.The transition process, exp.3-4.
In this case, the radius will be equal ( γ * = 0.3562 ).
The overall the transition process of the system shows on the
Figure 4.

21
 International Journal of Control, Energy and Electrical Engineering (CEEE)
Copyright – IPCO-2014
Vol.1, pp. 17-22

[2] Malkin I.G. The theory of stability of motion.2-d Publishing.

Nauka. Moscow, (in Russian), 540 p., 1966.
[3] Siljak D.D. Parameter space methods for robust control design:
a guided tour. IEEE Transactions on Automatic Control, V.AC-
34, N.7,pp. 674-688, 1989.
[4] Polyak B., Scherbakov P.Robust Stability and Control [in
Russian], Nauka, ISBN 5-02-002561-5, Moscow, 303 p, 2002.
[5] Beisеnbi M.A., Kulniyazova K.S. Research of robust stability in
the control systems with Lyapunov A.M. direct method.
Proceedings of 11-th Inter-University Conference on
Mathematics and Mechanics. Astana, Kazakhstan. pp.50-56,
2007.
Fig. 7.The transition process, exp.1-4. [6] Beisеnbi M.A. Methods of increasing the potential of robust
stability control systems. L.N.Gumilyov Eurasian National
The experimental results of the proposed robust control University (ISBN 978-601-7321-83-3). Astana, Kazakhstan ,
system and comparing the satisfied results obtained. 352 p, 2011.
[7] Nikulin V., Beisеnbi M., Ainagulova A., Аbitova G. Design of
Control Systems for Nonlinear Control Laws with Increased
V.CONCLUSIONS Robust Stability. The Third International Conference on CSDM
Paris, France, 2012.
[8] Nikulin V., Beisеnbi M., Ainagulova A., Аbitova G. Design of
To conclude, theory robust stability performs an Control System Based on Functions of Catastrophe. The
important role in the theory of control of dynamic objects [6]. International Journal of Art & Sciences’ (IJAS) International
The main task in studying robust stability is to specify Conference for Academic Disciplines, Proceedings of the IJAS
constraints on the change of control system parameters that 2012. Massachusetts, USA, 2012.
preserve stability. These restrictions determined by the range [9] Nikulin V., Beisеnbi M., V. Skormin., Ainagulova A. Аbitova.
of stability at undetermined and selected, i.e. varying G. Control System with High Robust Stability Characteristics
parameters [7,8]. Based on Catastrophe Function. 17th Annual IEEE
This work provides for construction of Lyapunov International Conference on the Engineering of Complex
function in a form of vector-function so that its gradient shall Computer Systems (ICECCS 2012), Proceedings of the ICECCS
/ France, Paris, 2012.
be equal to components of velocity vector (right side of the [10] Nikulin V., Beisеnbi M., Аbitova G. Design of Complex
equation) but with opposite sign. The study of robust stability Automation System for Effective Control of Technological
of the system is founded on the basis of Lyapunov’s direct Processes of Industry. 3td International Conference on Industrial
method. Stability region resulted in simplest equations with Engineering and Operations Management (IEOM 2012),
undetermined parameters of control objects and selected Proceedings of the IEOM 2012/ Istanbul, Turkey, 2012.
parameters of regulator. [11] Yermekbayeva J.J., Beisеnbi M., The Research of the Robust
This article described geometric interpretation of Stability in Linear System. Proceeding Engineering and
Lyapunov function in detail. Conditions of robust stability Technology , vol 3., pp.142-147, 2013.
were identified. Robust stability of Lyapunov function was [12] Beisenbi M.A Abdrakhmanova L.G. Research of dynamic
properties of control systems with increased potential of robust
examined by applying interval matrix family of coefficients. stability in a class of two-parameter structurally stable maps by
The radius of stability coefficients interval family of positive Lyapunov function. International Conference on Computer,
definite functions is equal to the smallest value of the Network and Communication Engineering (ICCNCE 2013). –
coefficients of the vector Lyapunov functions. Detailed case Published by Atlantis Press, p.201-203,2013.
study was presented. Besides, theoretical results obtained in [13] Beisenbi M.A, Yermekbayeva J., Omarov A., Abitova G. The
this paper provide considerable contribution to the theory of Control of Population Tumor Cells via Compensatory Effect.
stability at certain indefiniteness, i.e. to the theory of robust Applied Mathematics & Information Sciences. An International
stability of linear control systems . Jounal. – NSP, (Thomson Reuters, IF-0,731), 2013.
Practical importance of achieved results from one [14] Beisenbi M.,Uskenbayeva G.The New Approach of Design
side provides for application of common applied approaches Robust Stability for Linear Control System, Proceeding of
International Conference On Advances in Computing and
to define region of robust stability [9-12] that will preserve Information Technology - ACIT 2014, Bangkok, Thailand,
dynamic safety of the object under investigation, from the 2014.
other side these studies important for providing an opportunity [15] Beisenbi M., Abdrakhmanova L The new method of research of
to design more effective control systems [13-15]. the systems with increased potential with robust stability,
Proceeding of International Conference On Advances in
Computing and Information Technology - ACIT 2014,
REFERENCES Bangkok, Thailand, 2014.
[1] Barbashin E.A. Introduction in the theory of stability. Publishing
Nauka, Moscow, (Russian), 255 p., 1967.

22