52nd IEEE Conference on Decision and Control

December 10-13, 2013. Florence, Italy

Lyapunov Stability for Continuous-time 2D Nonlinear Systems

Hamid Reza Shaker

Abstract— This paper deals with the stability of continuous- recently in [17]. The stability notions for 1D systems were
time 2D nonlinear systems in the Roesser form. The concepts developed for 2D systems while the studied nonlinearity is
from 1D Lyapunov stability theory are extended to 2D nonlinear general and the initial conditions are given on the positive
systems. To check the stability a direct Lyapunov method is
developed. While the direct Lyapunov method has been recently semiaxes. The direct Lyapunov method was also developed
proposed for discrete-time 2D nonlinear systems, to the best for nonlinear 2D systems in coherence with the proposed
of our knowledge what proposed in this paper are the first notions. The recent stability results in [17], just support
results of this kind on stability of continuous-time 2D nonlinear discrete-time 2D nonlinear systems. The continuous-time 2D
systems. Analogous to 1D systems, a sufficient condition for systems are important in modelling physical processes, in
the stability is the existence of a certain type of the Lyapunov
function. A new technique for constructing Lyapunov functions particular those which are described by partial differential
for 2D nonlinear systems is proposed. The proposed method equations[18]. This has been shown in [19]-[21]. To the
is based on the sum of squares decomposition, therefore it best of knowledge stability results similar to those in [17]
formulates the Lyapunov function search algorithmically. In do not exist for continuous-time 2D nonlinear systems in
this way polynomial nonlinearities can be handled exactly and Roesser form. This will be the focus of our paper. The 2D
a large class of other nonlinearities can be treated introducing
some auxiliary variables and constrains. Lyapunov stability theory in [17], even though addresses
in a simple way the important issue of stability based on
I. I NTRODUCTION the existence of a Lyapunov function, it does not provide
any coherent methodology for constructing such a function.
Over the past few decades there have been growing interest
In this work, however, we propose a method based on
in multidimensional systems. This is because of their clear
the sum of squares decomposition (SOS). This formulates
links to a wide variety of applications arising in both theory
the Lyapunov function search algorithmically and supports
and practice. The first studies in this context was mainly
polynomial and rational nonlinearities. A large class of other
focused on linear systems. For linear multidimensional sys-
nonlinearities can also be treated introducing some auxiliary
tems, polynomial functions of multiple independent variables
variables and constrains within this framework.
were used as an alternative for a transfer function with one
The paper is organized as follows. In Section II, we
independent variable. The notion of stability which plays an
review some definitions, notations and concepts related to
important role in control and systems theory was studied for
the stability for 1D systems. Section III devoted to Lyapunov
these multidimensional systems in terms of BIBO stability
stability for continuous-time 2D nonlinear systems. New
[1]. As a result a well-known stability criterion was proposed
theorem is proposed to check stability of such systems.
stating that the transfer function is BIBO stable, if there
Furthermore, a technique based on sum of square decompo-
are no poles outside the stability region. Several tests to
sition is proposed in this section for constructing Lyapunov
check stability were proposed based on this result [2], [3].
functions. In Section IV, the results are further illustrated
It is even extended to support exponential stability in [4].
with the help of a numerical example. Finally, the last section
While this stability conditions were thought to be necessary
concludes the paper.
and sufficient conditions, Goodman showed in [5] that the
conditions are only necessary conditions in the case that the
II. S TABILITY, LYAPUNOV F UNCTION AND S UM OF
transfer function does not show a particular type of the so-
S QUARE D ECOMPOSITION (SOS) FOR 1D S YSTEMS
called nonessential singularity. Roesser [6] and Fornasini-
Marchesini [7], [8] introduced state-space models for linear Stability analysis is among the key issues in dynamics
multidimensional systems. They proposed a definition of and control. Over the years this issue has received a lot of
asymptotic stability for their model as well. Since then with attention [23]-[26]. In what follows we briefly review some
the help of the development of techniques based on LMI’s results related to the stability notion. Consider a continuous-
and Lyapunov techniques, there was a growing interest in 2D time 1D nonlinear systems:
systems (e. g. [9]-[17]). Most of the results on the stability
of 2D systems that have been proposed so far are devoted to ẋ = f (x) (1)
linear systems and just few results address stability of non-
where:f : D → Rn is locally Lipschitz map from a domain
linear systems. The stability analysis for a general discrete-
D ⊂ Rn into Rn . Without loss of generality we assume
time 2D nonlinear systems in Roesser form was addressed
that the origin x = 0 is the equilibrium. In the following
H. R. Shaker is an Associate Professor at Aalesund University College, the basic definition for stability is presented.
N-6025 Aalesund, Norway [email protected]

978-1-4673-5717-3/13/$31.00 ©2013 IEEE 4586

Definition 1[22]. The equilibrium point x = 0 for (1) III. LYAPUNOV S TABILITY AND S UM OF S QUARE
is: D ECOMPOSITION (SOS) FOR C ONTINUOUS -T IME 2D
• stable in the sense of Lyapunov if for each ε > 0, there N ONLINEAR S YSTEMS
is δ = δ(ε) > 0 such that: Consider a continuous-time 2D nonlinear systems in the
kx(0)k < δ ⇒ kx(t)k < ε, ∀t ≥ 0. (2) following Roesser model:
h
 
∂x (t1 ,t2 )
• unstable if it is not stable. ∂t1
h v
we are now ready to present Lyapunov stability theory for  = f (x (t1 , t2 ), x (t1 , t2 )) (8)
 

1D systems. ∂xv (t1 ,t2 )
∂t2

xh
 
Theorem 1[22]. Let D ⊂ Rn be a domain containing where: f is a function with f (0, 0) = 0, x = ,
the equilibrium x = 0. Let V : D → R be a continuously xv
differentiable function such that: xh ∈ Rn , xv ∈ Rm and (t1 , t2 ) ∈ R2 and the initial
V (0) = 0 conditions which are given on the positive semi-axes (i.e.
V (x) > 0 in D − {0} xh (0, t2 ) and xv (t1 , 0) for t1 , t2 ∈ R≥0 ). For this 2D
(3)
system, the stability notion is defined in the sequel.
V̇ (x) ≤ 0 in D
Then, x=0 is stable. Definition 2. The point xe = 0 for 2D system (8) is:
• stable in the sense of Lyapunov if for each ε > 0, there
A function V (x) which satisfies conditions in (3) is is δ = δ(ε) > 0 such that:
called Lyapunov function. As it is clear from Theorem 1, 
the existence of a Lyapunov function is a sufficient condition kx(0, t2 )k < δ
⇒ kx(t1 , t2 )k < ε, ∀t1 , t2 ≥ 0.
for stability. However, this theorem does not provide any kx(t1 , 0)k < δ
methodology for constructing such a function. Constructing (9)
Lyapunov functions by hand is limited to systems of low • unstable if it is not stable.

orders and needs a good analytic skills. If we replace In the following the direct Lyapunov method for stability is
the non-negativity conditions by sum of square(SOS) extended for 2D systems. This theorem has been previously
conditions, then not only testing the Lyapunov function proposed for discrete-time 2D systems, however to the best
conditions but also constructing the Lyapunov function can of our knowledge it has not been proposed for continuous-
be done efficiently using semidefinite programming (SDP) time 2D nonlinear systems. In the following theorem we use
[27] ,[28]. In this way the polynomial nonlinearities can functions of class K. A continuous function α : R+ → R+ is
be handled exactly and a large class of other nonlinearities said to be of class K, if α(0) = 0 and α is strictly increasing.
can be treated introducing some auxiliary variables and
constrains[30], [31]. In the sequel, For D = Rn , Theorem 1 Theorem 2. Let V h : Rn → R and V v : Rm → R be
is formulated in the SOS framework. continuous functions such that:
Proposition 1 [28]. Assume that for the system (1) there
exists a polynomial V (x) such that V (0) = 0, and: αh ( xh ) ≤ V h (xh ) ≤ βh ( xh )
(10)
αv (kxv k) ≤ V v (xv ) ≤ βv (kxv k)
V (x) − Φ(x) is SOS (4) where αh , αv , βh , βv are functions of class K. If there exist
∂V V (xh , xv ), such that:
− f (x) is SOS (5)
∂x
where for a polynomial V (x) of degree 2d , Φ(x) is defined V (xh , xv ) := V h (xh ) + V v (xv ), (11)
as:
n X d
X ∂V h (xh (t1 , t2 ))
Φ(x) = εij x2j
i (6) ≤ −γh ( xh ), (12)
i=1 j=1
∂t1
where: ∂V v (xv (t1 , t2 ))
m ≤ −γv (kxv k), (13)
X ∂t2
εij > γ, γ > 0, εij ≥ 0, (7)
j=1
where γh and γv are functions of class K, then the
equilibrium xe = 0 of system (8) is stable in the sense of
then the zero equilibrium of (1) is globally stable (i.e., the Lyapunov.
equilibrium is stable in the sense of Lyapunov).
This SOS problem can be solved using SOSTOOLS using Proof:
semidefinite programming. A detailed description of how Given ε > 0, we select rv and rh such that:
SOSTOOLS works can be found in the SOSTOOLS users
guide [29]. ε2 = rv2 + rh2 (14)

4587
, we define: using semidefinite programming. In such framework the
polynomial nonlinearities can be handled exactly and a large
B(rh , rv ) = x ∈ Rn+m : kxv k ≤ rv , xh ≤ rh

(15) class of other nonlinearities can be treated introducing some
Let α = min V (xh , xv ), then α > 0 according auxiliary variables and constrains [30], [31]. For the system
kxv k=rv ,kxh k=rh (8), we partition f into fh and fv .
to (10). If we choose β ∈ (0, α) and define:
∂xh (t1 ,t2 )
 
V (xh , xv ) ≤ β

Ωβ = x ∈ B(rh , rv ) : (16) ∂t1

fh (xh (t1 , t2 ), xv (t1 , t2 ))

= (22)
 
fv (xh (t1 , t2 ), xv (t1 , t2 ))

Then Ωβ is in the interior of B(rh , rv ). This set has the ∂xv (t1 ,t2 )
property that any trajectory starting in Ωβ at t1 = 0 or t2 = 0 ∂t2

stays in Ωβ for all t1 and t2 . This is because of (12) and where: fh (xh (t1 , t2 ), xv (t1 , t2 ))∈ Rn . The SOS-based
(13) and the fact that: direct Lyapunov method is as following:
V (xh , xv ) Proposition 2. If for system (8) there exists a polynomial
= V h (xh (t1 , t2 )) + V v (xv (t1 , t2 )) (17) V (xh , xv ) := V h (xh ) + V v (xv ) such that V (0) = 0, and:
≤ V h (xh (0, t2 )) + V v (xv (t1 , 0)) ≤ β
We know that V (xh , xv ) is continuous and V (0) = 0. V h (xh ) − Φ1 (xh ) is SOS (23)
Therefore there is τ > 0 such that:
V v (xv ) − Φ2 (xv ) is SOS (24)
xh ≤ τ

⇒ V (xh , xv ) ≤ β, (18)
kxv k ≤ τ
∂V h
therefore: −fh (xh , xv ) is SOS (25)
∂xh
B(τ, τ ) ⊂ Ωβ ⊂ B(rh , rv ), ∂V v
− v fv (xh , xv ) is SOS (26)
Consequently, we have: ∂x
where for a polynomial V (x) of degree 2d , Φk (x), k = 1, 2
(xh (0, t2 ), xv (t1 , 0)) ∈ B(τ, τ )
is defined as:
⇒ (xh (0, t2 ), xv (t1 , 0)) ∈ Ωβ n X d
(19) X
⇒ (xh (t1 , t2 ), xv (t1 , t2 )) ∈ Ωβ Φk (x) = εij x2j
i (27)
⇒ (xh (t1 , t2 ), xv (t1 , t2 )) ∈ B(rh , rv ) i=1 j=1

Therefore: where:
m
kxv (0, t2 )k < τ
 X
v
⇒ kx k < rv , x h
< rh εij > γ, γ > 0, εij ≥ 0, (28)
xh (t1 , 0) < τ
(20) j=1
√ then the zero equilibrium of (8) is stable in the sense of
⇒ kx(t1 , t2 )k < 2 2
rv + rh ≤ ε, ∀t1 , t2 ≥ 0.
Lyapunov.
2
On the other hand, using the equality: kx(t1 , t2 )k = This proposition enables us to use SOS to search for
2 2
kxv (t1 , t2 )k + xh (t1 , t2 ) , we have: a Lyapunov function. This can be done efficiently using
SOSTOOLS.
√
kxv (0, t2 )k < δ √2
 
kx(0, t2 )k < δ IV. A N I LLUSTRATIVE E XAMPLE
⇒ (21)
kx(t1 , 0)k < δ xh (t1 , 0) < δ 2 Consider a continuous-time 2D nonlinear systems of the
√ following Roesser form:
If we use (20), choosing τ = δ 2, we have: kx(t1 , t2 )k <
ε, ∀t1 , t2 ≥ 0. This proves that (8) is satisfied and hence
∂xh (t1 ,t2 )
 
the system is stable in the sense of Lyapunov.
fh (xh (t1 , t2 ), xv (t1 , t2 ))
 
∂t1
= (29)
 
fv (xh (t1 , t2 ), xv (t1 , t2 ))

∂xv (t1 ,t2 )
∂t2
A function V (x) which satisfies conditions in Theorem
2 is a Lyapunov function for continuous-time 2D nonlinear where:
−x1 5 − 2x1 − x2 3 − x3 5
 
systems. As it is clear from this theorem, the existence of
a Lyapunov function is a sufficient condition for stability. fh (xh (t1 , t2 ), xv (t1 , t2 )) =  −x1 3 − x2 3 ,
However, similar to 1D case this theorem does not provide 3
x1 − x3
any methodology for constructing such a function. In the fol-
−x5 3 − x4 − x4 5
 
lowing we use an SOS-based Lyapunov direct method. This
fv (xh (t1 , t2 ), xv (t1 , t2 )) = .
not only facilitate testing the Lyapunov function conditions −x4 − x5 3
but also helps to construct a Lyapunov function efficiently (30)

4588
 Using SOSTOOLS from Proposition 2, the following [11] P.-A. Bliman, Lyapunov equation for the stability of 2-D systems,
functions are obtained: Multidimensional Systems and Signal Processing, vol. 13, no. 2,
pp.201-222, 2002.
[12] K. Galkowski, J. Lam, S. Xu, and Z. Lin, LMI approach to state feed-
back stabilization of multidimensional systems,, International Journal
V h (xh ) = 5.4634x1 2 − (2.8419e − 006)x1 x2 + 10.9267x1 x3 of Control, vol. 76, no. 14, pp. 1428-1436, 2003.
2 2
+6.4054x2 − (1.7455e − 006)x2 x3 + 10.9267x3 [13] W. Paszke, J. Lam, K. Galkowski, S. Xu, and Z. Lin, Robust stability
(31) and stabilization of 2D discrete state-delayed systems, Systems and
V v (xv ) = 150714.2832x4 4 + 403985.6651x4 2 Control Letters, vol. 51, pp. 277-291, 2004.
(32) [14] H.R. Shaker, and M. Tahavori, Stability analysis for a class of discrete-
+201970.0774x5 4 + 1.9924x5 2 time two-dimensional nonlinear systems, Multidimensional Systems
h v and Signal Processing, Vol. 21, No. 3, pp. 293–299, 2010.
V (x , x ) is obtained from: [15] T. Kaczorek, LMI approach to stability of 2D positive systems,
Multidimensional Systems and Signal Processing, vol. 20, no. 1,
V (xh , xv ) = V v (xv ) + V h (xh ) (33) pp.39-54, 2009.
[16] D. N. Avelli, P. Rapisarda, and P. Rocha, Lyapunov stability of 2D
Therefore the xe = 0 system is stable in the sense of finite-dimensional behaviours, International Journal of Control, vol.
Lyapunov. 84, no. 4, pp. 737-745, 2011.
[17] N. Yeganefar, N. Yeganefar, M. Ghamgui, and E. Moulay, Lyapunov
theory for 2D nonlinear Roesser models: application to asymptotic
V. C ONCLUSIONS and exponential stability, IEEE Transactions on Automatic Control, In
The stability of continuous-time 2D nonlinear systems in [18] Press, DOI: 10.1109/TAC.2012.2220012.
F. L. Lewis, A review of 2-D implicit systems, Automatica, Vol. 28,
the Roesser form is addressed in this paper. The concept from No. 2, pp. 345–354, 1992.
1D Lyapunov stability theory is extended to 2D nonlinear [19] W. Marszalek, Two-dimensional state-space discrete models for hy-
perbolic partial differential equations, Appl. Math. Modelling, Vol. 8,
systems. To check the stability 2D direct Lyapunov method pp. 11–14, 1984.
is developed. The direct Lyapunov method has been recently [20] W. Marszalek, On modelling of distributed processes with two-
proposed for discrete-time 2D nonlinear systems,but to the dimensional discrete linear equations, Rozprawy Elektrotechniczne, ,
best of our knowledge this paper presents the first results [21] Vol. 33, pp. 627–640, 1987.
F. L. Lewis, Walsh function analysis of 2-D generalized continuous
of this kind on stability of continuous-time 2D nonlinear systems., IEEE Transactions on Automatic Control, Vol. 35, No. 10,
systems. Furthermore, a new technique for constructing Lya- pp. 1140–1144, 1990.
punov functions for 2D nonlinear systems is proposed. The [22] Hall, H. Khalil, Nonlinear Systems, 3rd ed. Englewood Cliffs, NJ: Prentice
2002.
proposed method is based on the sum of squares decompo- [23] R. I. Leine, The historical development of classical stability concepts:
sition, therefore it formulates the Lyapunov function search Lagrange, Poisson and Lyapunov stability, Nonlinear Dynamics, Vol.
algorithmically. In this way polynomial nonlinearities can [24] 59, No 1-2, pp 173–182, 2010.
D. Aniszewska, M. Rybaczuk, Lyapunov type stability and Lyapunov
be handled exactly and a large class of other nonlinearities exponent for exemplary multiplicative dynamical systems, Nonlinear
can be treated introducing some auxiliary variables and Dynamics, Vol. 54, No. 4, pp 345–354, 2008.
[25] Q. Ling, X. L. Jin, Y. Wang, H. F. Li, Z. L. Huang, Lyapunov function
constrains. construction for nonlinear stochastic dynamical systems, Nonlinear
Dynamics ,2013,DOI:10.1007/s11071-013-0757-3.
R EFERENCES [26] M. Kidouche, H. Habbi , On Lyapunov stability of interconnected
nonlinear systems: recursive integration methodology, Nonlinear Dy-
[1] J. Shanks, S. Treitel, and J. Justice, Stability and synthesis of two- namics,Vol. 60, No. 1-2 , pp 183/-191, 2010.
dimensional recursive filters, IEEE Trans. on Audio and Electroacous- [27] L. Vandenberghe and S. Boyd, Semidefinite programming, SIAM
tics,vol. AU-20, no. 2, pp. 115 - 128, 1972. Review, Vol. 38, No. 1, pp.4995, 1996.
[2] T. Huang, Stability of two-dimensional recursive filters, IEEE Trans [28] A. Papachristodoulou and S. Prajna, A tutorial on sum of squares
on Audio and Electroacoustics, vol. AU-20, no. 2, pp. 158 - 163, 1972. techniques for systems analysis, in proceedings of the American
[3] B. Anderson and E. Jury, Stability test for two-dimensional recursive Control Conference, vol. 4, pp. 2686–2700, 2005.
filters, IEEE Trans. on Audio and Electroacoustics, vol. AU–21, no. [29] S. Prajna, A. Papachristodoulou, P. Seiler,and P. A. Parrilo,Sum of
4, pp. 366372, 1973. squares optimization toolbox for MATLAB users guide, 2004,avail-
[4] L. Pandolfi, Exponential stability of 2-D systems, Systems Control able at:http://www.cds.caltech.edu/sostools/sostools.pdf.
Letters, vol. 4, pp. 381-385, 1984. [30] A. Papachristodoulou and S. Prajna, Analysis of non-polynomial sys-
[5] D. Goodman, Some stability properties of two dimensional linear shift tems using the sum of squares decomposition, in Positive Polynomials
invariant digital filters, IEEE Trans. on Circuits and Systems, vol. in Control, D. Henrion and A. Garulli, Eds. Berlin, Germany:Springer-
CAS-24, no. 4, pp. 201208, 1977. Verlag, pp. 23-44, 2005.
[6] R. Roesser, Discrete state-space model for linear image processing., , [31] A. Papachristodoulou and S. Prajna, On the construction of Lyapunov
IEEE Transactions on Automatic Control, vol. 20, no. 1, pp. 1 - 10, functions using the sum of squares decomposition, in proceedings of
1975. 41st IEEE Conference on Decison and Control, vol. 3, pp. 34823487,
[7] E. Fornasini and G. Marchesini, State-space realization theory of two- 2002.
dimensional filters, IEEE Transactions on Automatic Control, vol.
21,no. 4, pp. 484 - 492, 1976.
[8] E. Fornasini and G. Marchesini, Doubly-indexed dynamical systems:
State-space models and structural properties, , Mathematical Systems
Theory,vol. 12, pp. 59 - 72, 1978.
[9] T. Hinamoto, The Fornasini-Marchesini model with no over flow
oscillations and its application to 2D digital filter design, in Proc.IEEE
Int. Symp. Circuits and Systems (ISCAS), vol. 3, Portland, USA, 1989,
pp. 1680-1683.
[10] C. Du and L. Xie, Stability analysis and stabilization of uncertain two-
dimensional discrete systems: An LMI approach, IEEE Transactions
on Circuits and Systems I: Fundamental Theory and Applications, vol.
46, no. 11, pp. 1371-1374, 1999.

4589