Preprints of the 18th IFAC World Congress

Milano (Italy) August 28 - September 2, 2011

Robust stability analysis and stabilization of fractional order polytopic systems

Christophe Farges, Jocelyn Sabatier, Mathieu Moze
University of Bordeaux, IMS Laboratory (CRONE Team), CNRS UMR 5218
351 cours de la Libration, 33405 Talence, France
(e-mail: author@ims-bordeaux.fr)

Abstract: This paper addresses the problem of robust pseudo state feedback stabilization of
commensurate fractional order polytopic systems (FOS). In the proposed approach, Linear Matrix
Inequalities (LMI) formalism is used to check if the pseudo state matrix eigenvalues belong to the FOS
whatever the value of the uncertain parameters. The paper focuses particularly on the case 0 < Q < 1 as
the stability region is non convex and associated LMI condition is not as straightforward to obtain as in
the case 1 < Q < 2. The quadratic stabilisation problem involving a single matrix in order to prove
stability of the closed loop system is first addressed. Additional variables are then introduced in order to
decouple system matrices from the ones proving stability of the closed loop system. This decoupling
allows using parameter dependant stability matrices and leads to less conservative results as attested by a
numerical example.
Keywords: Fractional order systems; Linear Matrix Inequalities; Robust control; State feedback;
Polytopic systems


provide sufficient conditions. Note also that these results are

dedicated to SISO systems.

1 INTRODUCTION
As for linear time invariant integer order systems, it is now
well known that stability of a linear fractional order system
depends on the location of the systems poles in the complex
plane. However, pole location analysis remains a difficult
task in the general case. For commensurate fractional order
systems, powerful criterions have been proposed. The most
well known is the Matignons stability theorem (Matignon,
1996). It permits to check the system stability through the
location in the complex plane of the dynamic matrix
eigenvalues of the state space like system representation.
Matignons theorem is in fact the starting point of several
results in the field. Some of them are described in the survey
(Petras, 2009). Matignons theorem is the starting point of the
Linear Matrix Inequalities (LMI) stability conditions recently
proposed by the authors (Farges et al, 2010). Those
conditions are used to synthetize a stabilizing pseudo state
feedback whatever the system fractional order Q in the set
]0,2[.
Although much progress has been made in the field of
fractional system stability, linear time invariant fractional
systems robust stability remains an open problem. Among the
existing results and only for interval fractional systems, the
stability issue was discussed in (Petras et al, 2004), (Tan et al,
2009) and (Chen et al, 2006). As commented in (Chen et al,
2006) and (Ahn et al, 2007), the result is rather conservative.
To reduce the conservatism, in (Ahn et al, 2007), a new
robust stability checking method was proposed for interval
uncertain systems, where Lyapunov inequality is utilized for
finding the maximum eigenvalue of a Hermitian matrix.
However the results presented in (Ahn et al, 2007), only

Copyright by the
International Federation of Automatic Control (IFAC)

In this paper, the robust stability and stabilization problems

of linear time invariant fractional order linear systems with
convex polytopic uncertainties is studied. The paper
particularly focuses on the case 0 < Q < 1 as the stability
region is non convex and associated LMI stability condition
is not as straightforward to obtain as in the case 1 < Q < 2. In
a first approach, stability condition in (Farges et al, 2010), is
directly used to propose a quadratic stability analysis
condition that involves a single variable. This condition is
used to derive a quadratic pseudo state feedback synthesis
method. Additional variables are then introduced in order to
decouple system matrices from the ones proving system
stability. This decoupling allows using parameter dependant
stability matrices and lead to less conservative results both
for analysis and synthesis purposes as attested by a numerical
example.
Notations: The transpose of a matrix A is denoted A, its
conjugate A and its conjugate transpose A*. For Hermitian
matrices, > () denotes the Lwner partial order, i.e. A > B
iff A B is (semi) positive definite.
2 PRELIMINARIES AND PROBLEM STATEMENT
In this paper are considered Linear Time Invariant (LTI)
commensurate FOS. In this section, preliminary results are
stated in the certain case for an LTI FOS admitting a pseudo
state space representation of the form

10800

DQ xt 

y t 

A B xt 
C D u t 

(1)

Preprints of the 18th IFAC World Congress

Milano (Italy) August 28 - September 2, 2011

where xt  R n is the pseudo state vector, u t  R m is the

input vector, y t  R p is the output vector, Q is the
fractional order of the system and A, B, C and D are constant
matrices. DQ is the fractional differentiation operator of order
Q (presented results are valid whatever definition is used:
Riemann-Liouville (Miller and Ross, 1993), (Caputo, 1967)
or others (Samko et al, 1993)). Transfer matrix is

1
H s  C sQ I  A B  D and impulse response matrix is

condition to the case 0 < Q < 1 is far from trivial because the
location of eigenvalues in this region corresponds to unstable
integer order systems. Moreover, region of the complex plane
defined by (3) is not convex as shown in figure 1. However
this problem has been solved in (Farges et al, 2010) in which
the following result was proposed.
Theorem 2: Fractional system (1) of order 0 < Q < 1 is BIBO
stable iff X

X * C nxn ! 0 s.t.

rX  rX ' A' ArX  r X   0

ht  L 1 ^H s `

(4)

j 1Q 

Definition 1: (Matignon, 1996) A linear fractional order

system defined by its impulse time response h is bounded-

where r

input bounded-output (BIBO) stable iff u Lf R  , R m ,

Using this result, the pseudo state feedback stabilization

problem has been solved in (Farges et al, 2010).

h u Lf R  , R p .

LTI integer order systems stability can be checked via the

location of the eigenvalues of the pseudo state matrix A in the
complex plane. This result was extended to LTI
commensurate fractional order systems of order 0 < Q < 1 by
D. Matignon.
Theorem 1 (Matignon, 1996): System (1), with minimal
triplet (A, B, C) and 0 < Q < 1, is BIBO stable if and only if
Arg eig  A ! Q

Theorem 3: Fractional system (1) of order 0 < Q < 1 is BIBO

stabilizable by pseudo state feedback control law

X * C nxn ! 0 and Y R mxn ! 0

u Kx  y r iff X
s.t.

rX  rX ' A' ArX  r X   Y ' B' BY  0

(5)

where r

j 1Q 

. A stabilizing controller gain is then:

K

(2)

2
This result remains valid when 1 < Q < 2 as proved in
(Sabatier et al, 2008). Stability domain is thus defined as
follows:

(3)
z C : Arg z  ! Q
2

The corresponding stability regions of the complex plane are

represented by figure 1 (grey regions).

Y rX  r X

1

(6)

3. FEEDBACK STABILIZATION OF POLYTOPIC

FRACTIONAL ORDER SYSTEMS
3.1. Problem statement

Ds

Let the polytopic fractional order system described by:

DQ xt 

y t 

AO  BO  xt 
C O  DO  u t 

xt 

M O 
u t 

(7)

where O is a vector of parametric uncertainties. The

parameter-dependent system matrix M(O) belongs to the
convex polytope M with N vertices defined by
M

and M i
Fig. 1. Stability domain of fractional systems (gray region)
Remark 1 Throughout the paper, triplet (A, B, C) is always
supposed to be minimal.
Testing if the eigenvalues of matrix A belong to a region of
the left half plane defined by (3) with 1 < Q < 2 is a wellknown problem in LMI control theory because it corresponds
to a performance requirement on the damping ratio of the
system. A solution of this problem is provided by the LMI
region framework (Boyd et al, 1994). Extending this LMI

co^M 1 ,  , M N ` M O 

Ai
C
i

Bi
where /
Di

Oi M i : O / (8)
i 1

N
O R : O t 0, Oi
i 1

1 .

This paper is devoted to giving constructive conditions for

pseudo state feedback control laws of the form u = Kx+yr,
where K is a constant matrix gain and yr is the reference
signal, robustly stabilizing the closed loop system:
DQ xt  AO   BO K BO  xt 

y t  C O   DO K DO  y r t 

.(9)
A cl K , O  B cl K , O  xt 
xt 
cl

M K , O 
cl

cl
y r t 
C K , O  D K , O  y r t 

10801

Preprints of the 18th IFAC World Congress

Milano (Italy) August 28 - September 2, 2011

The closed loop system matrices M cl O  belong to the

polytope Mcl defined by:
M cl

cl
co M 1cl , , M N

`
(10)

Ai  Bi K
C  D K
i
i

Bi
Di

Aicl
cl
C i

Bicl
.
Dicl

The next sections present two results on robust control of

MIMO fractional order systems. The first one is a
straightforward extension of theorem 3 to handle uncertain
polytopic fractional systems (7). In the second one,
elimination lemma is used to derive a less conservative
condition.

3.2 Quadratic stabilization

Theorem 4: Fractional polytopic system (7) of order

0 < Q < 1 is robustly BIBO stable if  X = X*Cnn , X>0 s.t.
i ^1..N ` :

rX  r X ' Ai ' Ai rX  r X   0

Theorem 6: Fractional polytopic system (7) of order 0 < Q <

1 is robustly BIBO stable if there exist N matrices Xi =
Xi*Cnn , Xi>0 and a matrix GC2nn s.t. i ^1..N ` :

where r

(12)

According to theorem 2, the closed-loop system is robustly

stable and this completes the proof.
Please note that, in equation (12), a single matrix X proves
the system stability. For integer order systems, this
corresponds to use a parameter independent Lyapunov
function and is known as quadratic stability (Yan and Lam,
2001).
Theorem 4 is now used to derive the following synthesis
theorem.
Theorem 5: Fractional polytopic system (7) of order 0 < Q <
1 is robustly BIBO stabilizable by pseudo state feedback
control law u = Kx+yr if  X = X*Cnn , X>0 and YRmn
s.t. i ^1..N ` :

rX  r X ' Ai ' Ai rX  r X   Y ' Bi ' BiY  0

e

j 1Q 

Y rX  r X

1

(0)

AO 

G 'G>A' O   I @  0
I

(16)

with:
X O  iN 1 Oi X i .
(17)
Applying elimination lemma (Skelton et al, 1998) to the last
inequality leads to:

0
rX O   r X O  ' 1
AO @

 0 (18)
0
rX O   r X O 
A' O 
which is exactly (4) with parameter dependant matrices and
thus, according to theorem 2, proves the robust stability of
the system.

>I n

As for theorem 4, theorem 6 provides a sufficient condition

for stability but decoupling between stability matrix and
dynamic matrix in (16) allows to use the parameter
dependant stability matrix X O  defined by (17). As proved
in (Paucelle, 2000), stability condition of theorem 6 is always
less conservative than the one of theorem 4.
Based on theorem 6, the following result allows to design a
pseudo state feedback control law while stability of the
closed loop system is attested by a parameter dependant
stability matrix.

(14)

Theorem 7: Fractional polytopic system (7) of order 0 < Q <

1 is robustly BIBO stabilizable by pseudo state feedback
control law u = Kx+yr if there exist N matrices Xi = Xi*Cnn
, Xi>0, FRnn and KtRmn s.t. i ^1..N ` :

. A stabilizing controller gain is then:

rX O   r X O '

(13)

where r

rX  r X ' AO ' AO rX  r X   0 .

( 0)

rX O   r X O 

(11)

Proof: Suppose X is a solution to LMI (11). The computation

of the combinations over the N vertices allows to write:

(0)
rX i  r X i ' Ai

 G 'G>Ai '  I @  0 (15)

( 0)
rX i  r X i
 I
Proof: Suppose a solution (Xi, G) to (15). Computation of
convex combination over the N vertices allows to write for all
uncertainties:

S
2

variables

Using a single matrix X to attest stability for the whole set of

uncertainty is known to be overly conservative. However, the
coupling between stability matrix X and dynamic matrix
AO  in (12) prevents from directly using a parameter
dependent stability matrix X cl O  . The following result
allows us to overcome this problem.

From theorem 2 and 3, the following result can be obtained.

j 1Q 

of

3.3 Polytopic stabilization

where
M icl

Applying
the
lineraizing
change
Y K rX  r X leads to inequality (13).

Proof : Theorem 4 states that closed loop system is robustly

stable if  X = X*>0 and K s.t. i ^1..N ` :

rX  r X '  Ai  Bi K ' Ai  Bi K rX  r X   0
10802

Preprints of the 18th IFAC World Congress

Milano (Italy) August 28 - September 2, 2011

rX i  r X i '  Ai F  Bi K t >A

0'  Im @
F
(0)

A
 0 > Ai F  Bi K t '  F '@  0
 I m
(19)
A stabilizing controller gain is then:

K t F 1

(20)

Proof: Suppose a solution (Xi, F, Kt) to (19). Then the

controller definition implies that K t KF , which allows to
write:

(0)

rX i  r X i

In order to preserve convexity, matrix A0 appearing in

theorem 7 cannot be a variable but must be chosen a priori
such that it is stable. Indeed, elimination lemma shows that
existence of matrices Xi = Xi*>0, F and A0 verifying (19) is
equivalent to existence of matrices Xi = Xi*>0 and A0
verifying the following two inequalities:

rX i  r X i ' 1  0 ,

Ai '

(23)

4 NUMERICAL EXAMPLE
The proposed numerical application is a fractional version of
an example proposed in (De Souza and Trofino, 2000).
Studied system is described by representation (7) where:
2
 3  D
3
1.5  J

1
E

-2

-3
-3

(25)

-2.5

-2

-1.5

-1

-0.5
Real part

0.5

1.5

Fig. 2. Matrix A eigenvalues locus () and stability domain

limits (__)
The two theorems proposed in section 3 are applied in order
to compute a stabilizing pseudo state feedback for model (2526).
First, theorem 4 is used to compute a stabilizing controller of
gain K. Parser Yalmip (Lofberg, 2004) and LMI solver
SDPT3 are used to get matrices X and Y solutions of the
semidefinite problem associated with LMI condition (11).
For such a problem, a solution exists for values of D max up to
quad
D max

0
rX i  r X i ' 1
A0 @
 0 . (24)
0
rX

r
X
i
i
A0 '
Please note that, contrary to the analysis case, synthesis result
of theorem 7 cannot be proved to be always less conservative
than the one of theorem 5 based on the use of quadratic
stability condition. However, improvement can be significant
on some given examples, as shown in next section.

(26)

-1

0
Ai @
rX i  r X i

J d 1.5

The goal is now to compute a pseudo state feedback control

law of the form u = Kx+yr that robustly stabilises the system.

0'  I@

I
A
 0 F ' > Ai  Bi K '  I @  0
 I
(21)

(0)

E d 0.5

Fractional order Q is chosen equal to 0.7. Eigenvalues of

matrix A as parameters D, E and J vary in the intervals
defined by relation (26) are represented figure 2 for
D max 0.7 . That figure demonstrates that the system can be
stable or unstable depending on the values D, E and
Juncertain parameters values.

rX i  r X i '  Ai  Bi K F >A

rX i  r X i '
( 0)

(0)
rX i  r X i

.
(22)
Ai  Bi K

G 'G> Ai  Bi K '  I @  0
I
According to theorem 6, this last inequality proves that the
closed loop system is robustly stable.

>I

D d D max

A0
 I F ' gives

Defining G

>I

with

Imaginary part

(0)

rX i  r X i

0.57 .

Then, this solver has been used to get matrices Xi, i ^1..N ` ,
F and Kt associated with LMI condition (19) of theorem 7
with the matrix A0 chosen equal to:
A0

 2 0
0  2

(27)

A0
A solution exists for values of D max up to D max
0.74 and
corresponding gain K is obtained using equation (14):

> 15.56

33.29@ .

(28)

This represents an improvement of about 30%.

Eigenvalues of matrix A +BK (closed loop state matrix) as
parameters D, E and J vary in the intervals defined by relation
(26) with D max 0.74 are represented figure 3. This figure

10803

Preprints of the 18th IFAC World Congress

Milano (Italy) August 28 - September 2, 2011

demonstrates that the closed loop system is robustly BIBO

stable.

0
O r 1
0 O .
r2

(30)

Figure 5 represents the D max values obtained with

50 d Or1 d 0 and 0 d O r2 d 2 . As previously, this figure

10
8

A2

0
shows that the maximal value of D max ( D max
obtained for A0 eigenvalues O r1 19.31 and O r2

4
2
Imaginary part

A02

0.88 ) is
19.31 .

As shown in table 1, the result provided by theorem 7,

substantially increases the size of the uncertain domain for
which a controller can be computed. The best obtained D max
is 0.88 and corresponds to an improvement of 54% over the
quadratic case.

0
-2
-4
-6
-8
-10
-35

-30

-25

-20

-15
Real part

-10

-5

0.9
0.8
0.7

Fig. 3. Matrice A+BK eigenvalues locus () and stability

domain limits (__)

0.6

D max 0.5
0.4

The degrees of freedom offered by matrix A0 are now used to

find a controller K for higher values of parameter D max .
Matrix A0 is first chosen equal to:

0.3
0.2
0.1
0

O r
O
i

A01

 Oi
.
O r

(29)
-50

Figure 4 represent the D max values obtained with

50 d Or d 0 and 0 d Oi d 2 . This figure highlights the
existence of an infinity of matrices A0 that permit to obtain a
A1

0
solution and provides a value of D max equal to D max

for

Or = -15.86 and Oi

0.86

Or1
-100

-100

-90

-80

-70

-60

-50

-40

-30

-20

-10

Or 2

Fig. 5. Research of D max for various values of Or1 and Or2

0.
1

D quad max

D A0 max

D A0 max

D A0 max

0.57

0.74

0.86

0.88

Table 1. Values of D max obtained

5 CONCLUSION
1
0.8

D max0.6
0.4

0
1

0.2
2

0
-80

-70

-60

3
-50

Or

-40

-30

-20

4
-10

10

Oi

Fig. 4. Research of D max for various values of O r and Oi

Matrix A0 is now chosen equal to:

In this paper, a solution is proposed for the robust stability

and stabilization problems of fractional order linear systems
subjected to convex polytopic uncertainties. Presented results
are derived from the LMI stability analysis and synthesis
conditions recently proposed by the authors for the certain
case (Farges et al, 2010).
A first analysis result involving a single matrix in order to
prove stability of the system is first presented. Additional
variables are then introduced in order to decouple system
matrices from the ones proving stability of the closed loop
system. This decoupling allows using parameter dependant
stability matrices and obtained LMI stability analysis
condition is always less conservative than the one involving a
single stability matrix.

10804

Preprints of the 18th IFAC World Congress

Milano (Italy) August 28 - September 2, 2011

Both methods are extended to handle the state feedback

synthesis problem. Although, synthesis result based on the
use of parameter dependant matrices cannot be proved to be
always less conservative than the quadratic one, significant
improvement is obtained on a numerical example.
As shown in the numerical example, this last condition offers
some degree of freedom. Some parameters have to a priori set
and this choice has an influence on the quality of the obtained
result. Authors are currently working on a systematic method
to fix those parameters.
REFERENCES

dpendant des paramtres, PhD thesis dissertation, Toulouse

III University.
Peaucelle D. and Arzelier D. (2000), New LMI-Based
Conditions for Robust H2 Performance Analysis. Dans
American Control Conference, pages 317321, Chicago
Illinois.
Petras I., Chen Y., Vinagre B. M., Podlubny I. (2004),
Stability of linear time invariant systems with interval
fractional orders and interval coefficients, Proceedings of the
International Conference on Computation Cybernetics
(ICCC04), Vienna, Austria, 30th August-1st September, pp. 14.

Ahn H. S., Chen Y. Q. and I. Podlubny (2007), Robust

stability test of a class of linear time-invariant interval
fractional-order system using Lyapunov inequality, Appl.
Math. Comput., vol. 187, no. 1, pp. 27-34

Petras, I., Stability of fractional order systems with rational

orders: A survey, Fractional Calculus and Applied Analysis,
vol. 12, no. 3 (2009), 269-298.

Boyd S., El Ghaoui L., Feron E., and Balakrishnan V.(1994)

Linear Matrix Inequalities in System and Control Theory.
SIAM Studies in Applied Mathematics, Philadelphia.

Sabatier J., Moze M. and Farges C. (2008), On stability of

fractional order systems. In Third IFAC workshop on
fractional differentiation and its applications FDA08,
Ankara, Turkey.

Caputo M. (1967), Linear models of dissipation whose q is

almost frequency independent - II. Geophysical Journal
International, 13(5):529539.

Sabatier J. and C. Farges (2010), On stability of fractional

order systems. In Fourth IFAC workshop on fractional
differentiation and its applications FDA10, Badajoz, Spain

Chen Y. Q., Ahn H. S. and Podlubny I. (2006), Robust

stability check of fractional order linear time invariant
systems with interval uncertainties, Signal Processing, vol.
86 pp. 26112618.

Samko S. G., Kilbas A. A., and Marichev O. I (1993),

Fractional Integrals and Derivatives: Theory and
Applications. Gordon and Breach Science Publishers.

De Souza C. E. and Trofino A. (2000) An LMI approach to

stabilization of linear discrete-time periodic systems. Int. J.
Control, 73:696709.

Skelton R. E., Iwazaki T. and Grigoriadis K. (1998), A

unified approach to linear control design. Taylor and Francis
series in Systems and Control.

Farges C., Moze M., Sabatier, J. (2010), Pseudo state

feedback stabilization of commensurate fractional order
systems,
Automatica,
In
Press,
http://www.sciencedirect.com/science/article/B6V2150KMVV7-4/2/0b3f2aa00c9ad522b63188facb01fa4d

Tan N., Ozguven O. F., and Ozyetkin M. M. (2009), Robust

stability analysis of fractional order interval polynomials,
ISA Transactions, vol. 48, no. 2, pp. 166-172

Lofberg J. (2004). Yalmip: A toolbox for modeling and

optimizationvin MATLAB. In Proceedings of the CACSD
Conference, Taipei, Taiwan

Yan W. Y. and Lam J. (2001), On quadratic stability of

systems with structured uncertainty, IEEE Trans. Autom.
Control, vol. 46, no. 11, pp.17991805

Matignon D. (1996), Stability results on fractional differential

equations with applications to control processing. In
Computational Engineering in Systems Applications, volume
2, pages 963968, Lille, France.
Miller K. S. and Ross B. (1993), An Introduction to the
Fractional Calculus and Fractional Differential Equations.
John Wiley & Sons
Moze M., Sabatier J. and Oustaloup A. (2005), LMI tools for
stability analysis of fractional systems. In 20th ASME
International Design Engineering Technical Conferences &
Computers and Information in Engineering Technical
Conference (ASME IDETC/CIE05), Long Beach, California,
USA.
Paucelle D. (2000), Formulation gnrique de problmes en
analyse et commande robuste par les fonctions de Lyapunov
10805