Transcript
Page 1: [IEEE 2010 Third International Workshop on Advanced Computational Intelligence (IWACI) - Suzhou, China (2010.08.25-2010.08.27)] Third International Workshop on Advanced Computational

1

Exponential Stability of Singular Impulsive Systems withTime-varying Delays

Zhiguo Yang and Zhichun Yang

Abstractβ€” In this article, a model of singular systems involv-ing time-varying delays and impulses is considered. By estab-lishing a singular delay differential inequality with impulsiveinitial condition and using the property of 𝑀 -matrix, some newsufficient conditions ensuring the global exponential stability ofthe zero solution of singular systems are obtained. The resultscan extend and improve those of the earlier publications. Anexample is given to illustrate the theory.

I. INTRODUCTION

S INGULAR systems (known as semistate systems, dif-ferential algebraic systems, generalized state-space sys-

tems, etc.) have been of interest in the literature since theyhave many important applications in, for example, circuitsystems, robotics, aircraft modelling, social, biological, andmultisector economic systems, dynamics of thermal nuclearreactors singular perturbation systems, and so on. Manyinteresting results in dealing with singular systems have beenreported [1][2]. Furthermore, singular systems with delayshave been extensively studied in the past years due to thefact that delayed singular systems can even accurately depictthe evolutionary processes [3][4][5].

However, besides delay effects, impulsive effects likewiseexist in a wide variety of evolutionary processes in whichstates are changed abruptly at certain moments of time,involving such fields as medicine and biology, economics,mechanics, electronics and telecommunications, etc.. Aspointed out by [6], many sudden and sharp changes occurinstantaneously in singular systems, in the form of impulses.Therefore, it is very important, and indeed necessary, to studysingular impulsive systems.

The stability analysis plays an important role in the studyof singular system theory. Various stability properties ofsingular systems have been investigated in papers [4][5][6].But stability investigation to singular systems with impulsesand delays have not yet been fully developed [7]. Especially,there are few papers dealing with the exponential stability ofsingular impulsive systems with time-varying delays. Thisprompted us to discuss this problem.

Zhiguo Yang is with the College of Mathematics and SoftwareScience, Sichuan Normal University, Chengdu, 610068, China (email:[email protected]). Zhichun Yang is with the Department of Mathe-matics, Chongqing Normal University, Chongqing 400047, China (email:[email protected]).

This work was supported by National Natural Science Foundation ofChina under the grant No. 10926033, 10971147 and 10971240, A ProjectSupported by Scientific Reserch Fund of SiChuan Provincial EducationDepartment (08zb026), Key Research Project of Sichuan Normal University,Natural Science Foundation of Chongqing under Grant CSTC2008BB2364.

More specifically, in this paper, we will obtain some newsufficient conditions ensuring the global exponential stabilityof the zero solution of a singular impulsive system with time-varying delays by establishing a singular delay differentialinequality with impulsive initial condition and using theproperty of 𝑀 -matrix. The results extend and improve thoseof the earlier publications.

II. MODEL AND PRELIMINARIES

To begin with, we introduce some notations and recallsome basic definitions. Let 𝒩 Ξ”

= {1, 2, . . . , 𝑛}. For 𝐴,𝐡 βˆˆπ‘…π‘›Γ—π‘› or 𝐴,𝐡 ∈ 𝑅𝑛, 𝐴 β‰₯ 𝐡(𝐴 ≀ 𝐡,𝐴 > 𝐡,𝐴 < 𝐡)means that each pair of corresponding elements of 𝐴 and 𝐡satisfies the inequality β€œβ‰₯(≀, >,<)”.𝐢[𝑋,π‘Œ ] denotes the space of continuous mappings from

the topological space 𝑋 to the topological space π‘Œ .𝑃𝐢[𝐽,𝑅𝑛]

Ξ”= {πœ“ : 𝐽 β†’ 𝑅𝑛 ∣ πœ“(𝑑+) = πœ“(𝑑) and πœ“(π‘‘βˆ’)

exists for 𝑑 ∈ 𝐽 , πœ“(π‘‘βˆ’) = πœ“(𝑑) for all but at most countablepoints 𝑠 ∈ 𝐽}, where 𝐽 βŠ‚ 𝑅 is an interval, πœ“(𝑑+) and πœ“(π‘‘βˆ’)denote the right-hand and left-hand limits of the functionπœ“(𝑑), respectively.

For π‘₯ = (π‘₯1, . . . , π‘₯𝑛)𝑇 ∈ 𝑅𝑛, 𝐴 = (π‘Žπ‘–π‘—)𝑛×𝑛 ∈ 𝑅𝑛×𝑛,

πœ‘(𝑑) = (πœ‘1(𝑑), . . . , πœ‘π‘›(𝑑))𝑇 ∈ 𝑃𝐢[𝐽,𝑅𝑛], we define

[π‘₯]+ = (∣π‘₯1∣, . . . , ∣π‘₯π‘›βˆ£)𝑇 , [𝐴]+ = (βˆ£π‘Žπ‘–π‘— ∣)𝑛×𝑛,

[πœ‘(𝑑)]𝜏 = ([πœ‘1(𝑑)]𝜏 , . . . , [πœ‘π‘›(𝑑)]𝜏 )𝑇 , [πœ‘(𝑑)]+𝜏 = [[πœ‘(𝑑)]+]𝜏 ,

where [πœ‘π‘–(𝑑)]𝜏 = supβˆ’πœβ‰€π‘ β‰€0{πœ‘π‘–(𝑑 + 𝑠)}, 𝑖 ∈ 𝒩 , 𝜏 is apositive constant.

In this paper, we consider the following singular impulsivesystem with time-varying delays

𝑒𝑖𝐷π‘₯𝑖(𝑑) = βˆ’π‘π‘–π‘₯𝑖(𝑑)𝐷𝑣𝑖(𝑑)

+π‘›βˆ‘

𝑗=1

π‘Žπ‘–π‘—π‘“π‘—(π‘₯𝑗(𝑑))𝐷𝑒𝑗(𝑑)

+

π‘›βˆ‘π‘—=1

𝑏𝑖𝑗𝑔𝑗(π‘₯𝑗(π‘‘βˆ’ πœπ‘–π‘—(𝑑)))𝐷𝑀𝑗(𝑑),

𝑖 ∈ 𝒩 , (1)

with the initial condition

π‘₯𝑖(𝑑0 + 𝑠) = πœ™π‘–(𝑠), βˆ’πœ ≀ 𝑠 ≀ 0, 𝑖 ∈ 𝒩 , (2)

where the matrix 𝐸 = diag{𝑒1, . . . , 𝑒𝑛} β‰₯ 0 may besingular. The delays functions πœπ‘–π‘—(𝑑) are continuous for 𝑑 ∈ 𝑅and πœπ‘–π‘—(𝑑) ∈ [0, 𝜏 ], 𝑖, 𝑗 ∈ 𝒩 , 𝜏 is a positive constant.𝑓𝑗(β‹…), 𝑔𝑗(β‹…) ∈ 𝐢[𝑅,𝑅], 𝑗 ∈ 𝒩 . The initial condition πœ™ =(πœ™1(𝑠), . . . , πœ™π‘›(𝑠))

𝑇 ∈ 𝑃𝐢[[βˆ’πœ, 0], 𝑅𝑛]. 𝐷π‘₯𝑖(𝑑), 𝐷𝑣𝑖(𝑑),

630

Third International Workshop on Advanced Computational Intelligence August 25-27, 2010 - Suzhou, Jiangsu, China

978-1-4244-6337-4/10/$26.00 @2010 IEEE

Page 2: [IEEE 2010 Third International Workshop on Advanced Computational Intelligence (IWACI) - Suzhou, China (2010.08.25-2010.08.27)] Third International Workshop on Advanced Computational

2

𝐷𝑒𝑗(𝑑) and 𝐷𝑀𝑗(𝑑) denote the distributional derivatives [8]of the functions π‘₯𝑖(𝑑), 𝑣𝑖(𝑑), 𝑒𝑗(𝑑) and 𝑀𝑗(𝑑) respectively.Without loss of generality, we may assume that

𝐷𝑣𝑖(𝑑) = 1 +

βˆžβˆ‘π‘˜=1

π›Όπ‘–π‘˜π›Ώ(π‘‘βˆ’ π‘‘π‘˜),

𝐷𝑒𝑖(𝑑) = 1 +βˆžβˆ‘π‘˜=1

π›Ύπ‘–π‘˜π›Ώ(π‘‘βˆ’ π‘‘π‘˜),

𝐷𝑀𝑖(𝑑) = 1 +

βˆžβˆ‘π‘˜=1

π›½π‘–π‘˜π›Ώ(π‘‘βˆ’ π‘‘π‘˜), 𝑖 ∈ 𝒩 ,

where 𝛿(β‹…) is the Dirac impulse function. π›Όπ‘–π‘˜, π›Ύπ‘–π‘˜ and π›½π‘–π‘˜are constants. The impulsive moments π‘‘π‘˜ ( π‘˜ = 1, 2, . . . )satisfy 𝑑1 < 𝑑2 < . . . and limπ‘˜β†’+∞ π‘‘π‘˜ = +∞.

Throughout this paper, we assume that for any πœ™ βˆˆπ‘ƒπΆ[[βˆ’πœ, 0], 𝑅𝑛], the system (1) has at least one solutiondenoted by π‘₯(𝑑, 𝑑0, πœ™) or π‘₯(𝑑). Moreover, we assume that𝑓𝑗(0) = 𝑔𝑗(0) = 0, 𝑗 ∈ 𝒩 for the stability purpose of thispaper. Then the system (1) admits the zero solution π‘₯(𝑑) ≑ 0.

Definition 1: The zero solution of the singular system(1) is said to be globally exponentially stable if there existconstants 𝛼 > 0 and πœ… > 0 such that for any solutionπ‘₯(𝑑, 𝑑0, πœ™) with the initial condition πœ™ ∈ 𝑃𝐢[[βˆ’πœ, 0], 𝑅𝑛],

βˆ₯π‘₯(𝑑, 𝑑0, πœ™)βˆ₯ ≀ πœ…βˆ₯πœ™βˆ₯πœπ‘’βˆ’π›Ό(π‘‘βˆ’π‘‘0), 𝑑 β‰₯ 𝑑0, (3)

where βˆ₯π‘₯(𝑑, 𝑑0, πœ™)βˆ₯ = max1≀𝑖≀𝑛{∣π‘₯𝑖(𝑑, 𝑑0, πœ™)∣}, βˆ₯πœ‘βˆ₯𝜏 =max1≀𝑖≀𝑛{supβˆ’πœβ‰€π‘ β‰€0 βˆ£πœ‘π‘–(𝑠)∣}.

For an 𝑀 -matrix 𝑆 [9], we define

Ω𝑀 (𝑆)Ξ”= {𝑧 ∈ 𝑅𝑛 ∣ 𝑆𝑧 > 0, 𝑧 > 0}.

Lemma 1: [9] For an 𝑀 -matrix 𝑆, Ω𝑀 (𝑆) is nonemptyand satisfies,

π‘˜1𝑧1 + π‘˜2𝑧2 ∈ Ω𝑀 (𝑆),

for any π‘˜1, π‘˜2 > 0, 𝑧1, 𝑧2 ∈ Ω𝑀 (𝑆).

III. SINGULAR DELAY DIFFERENTIAL INEQUALITY

For the singular impulsive system (1), we need to estimateevery part on [π‘‘π‘˜, π‘‘π‘˜+1) with its initial function on [π‘‘π‘˜βˆ’πœ, π‘‘π‘˜]for π‘˜ = 1, 2, . . . . It is therefore difficult to obtain the estimate(3). To overcome these difficulties we establish the followingsingular delay differential inequality with impulsive initialcondition.

Theorem 1: Let 𝑃 = (𝑝𝑖𝑗)𝑛×𝑛 and 𝑝𝑖𝑗 β‰₯ 0 for 𝑖 βˆ•= 𝑗,𝑄 = (π‘žπ‘–π‘—)𝑛×𝑛 β‰₯ 0 and 𝑆 = βˆ’(𝑃 + 𝑄) be an 𝑀 -matrix. For 𝑏 ∈ (𝑑0,+∞), let 𝑒(𝑑) = (𝑒1(𝑑), . . . , 𝑒𝑛(𝑑))

𝑇 ∈𝐢[[𝑑0, 𝑏), 𝑅

𝑛] be a solution of the following delay differen-tial inequality with the initial condition 𝑒(𝑠) ∈ 𝑃𝐢[[𝑑0 βˆ’πœ, 𝑑0], 𝑅

𝑛],

𝐾𝐷+𝑒(𝑑) ≀ β„Ž(𝑑)(𝑃𝑒(𝑑) +𝑄[𝑒(𝑑)]𝜏 ), 𝑑 ∈ (𝑑0, 𝑏), (4)

where 𝐾 = diag{π‘˜1, . . . , π‘˜π‘›} β‰₯ 0,𝐷+𝑒(𝑑) is the upper rightderivative of 𝑒(𝑑). β„Ž(𝑑) > 0 satisfies sup𝑑β‰₯𝑑0

∫ 𝑑

π‘‘βˆ’πœβ„Ž(𝑠)𝑑𝑠 ≀

𝐻 <∞. Then

𝑒(𝑑) ≀ π‘§π‘’βˆ’πœ†

∫ 𝑑𝑑0

β„Ž(𝑠)𝑑𝑠, 𝑑 ∈ [𝑑0, 𝑏), (5)

provided that the initial condition satisfies

𝑒(𝑠) ≀ π‘§π‘’βˆ’πœ†

∫ 𝑠𝑑0

β„Ž(𝑠)𝑑𝑠, 𝑑0 βˆ’ 𝜏 ≀ 𝑠 ≀ 𝑑0, (6)

where 𝑧 = (𝑧1, . . . , 𝑧𝑛)𝑇 ∈ Ω𝑀 (𝑆) and the positive constant

πœ† satisfies the following inequality

[πœ†πΎ + 𝑃 +π‘„π‘’πœ†π» ]𝑧 < 0. (7)

Proof: Since 𝑆 is an 𝑀 -matrix, there exists a vector 𝑧 βˆˆΞ©π‘€ (𝑆) such that

𝑆𝑧 > 0 or [𝑃 +𝑄]𝑧 < 0.

By using continuity, we obtain that there must exist a positiveconstant πœ† satisfying the inequality (7), that is,

π‘›βˆ‘π‘—=1

[𝑝𝑖𝑗 + π‘žπ‘–π‘—π‘’πœ†π» ]𝑧𝑗 < βˆ’πœ†π‘˜π‘–π‘§π‘–, 𝑖 ∈ 𝒩 . (8)

We at first shall prove that for any positive constant πœ–

𝑒𝑖(𝑑) ≀ (1 + πœ–)π‘§π‘–π‘’βˆ’πœ†

∫ 𝑑𝑑0

β„Ž(𝑠)𝑑𝑠 Ξ”= 𝑦𝑖(𝑑),

𝑑 ∈ [𝑑0, 𝑏), 𝑖 ∈ 𝒩 . (9)

If inequality (9) is not true, by (6), there must be constant𝑑1 ∈ (𝑑0, 𝑏) and some integer π‘š ∈ 𝒩 such that

π‘’π‘š(𝑑1) = π‘¦π‘š(𝑑1), 𝐷+π‘’π‘š(𝑑1) β‰₯ π‘¦β€²π‘š(𝑑1), (10)

𝑒𝑖(𝑑) ≀ 𝑦𝑖(𝑑), 𝑑 ∈ [𝑑0 βˆ’ 𝜏, 𝑑1], 𝑖 ∈ 𝒩 . (11)

By using (4), (8), (9), (10), (11) and 𝑝𝑖𝑗 β‰₯ 0 (𝑖 βˆ•= 𝑗), 𝑄 β‰₯0, we obtain that

π‘˜π‘šπ·+π‘’π‘š(𝑑1) ≀ β„Ž(𝑑1)

π‘›βˆ‘π‘—=1

[π‘π‘šπ‘—π‘’π‘—(𝑑1) + π‘žπ‘šπ‘— [𝑒𝑗(𝑑1)]𝜏 ]

≀ β„Ž(𝑑1)

π‘›βˆ‘π‘—=1

[π‘π‘šπ‘—(1 + πœ–)π‘§π‘—π‘’βˆ’πœ†

∫ 𝑑1𝑑0

β„Ž(𝑠)𝑑𝑠

+π‘žπ‘šπ‘—(1 + πœ–)π‘§π‘—π‘’βˆ’πœ†

∫ 𝑑1βˆ’πœπ‘‘0

β„Ž(𝑠)𝑑𝑠]

≀ β„Ž(𝑑1)(1 + πœ–)π‘’βˆ’πœ†

∫ 𝑑1𝑑0

β„Ž(𝑠)𝑑𝑠

Γ—π‘›βˆ‘

𝑗=1

[π‘π‘šπ‘— + π‘žπ‘šπ‘—π‘’πœ†π» ]𝑧𝑗

< βˆ’πœ†π‘˜π‘šπ‘§π‘šβ„Ž(𝑑1)(1 + πœ–)π‘’βˆ’πœ†βˆ« 𝑑1𝑑0

β„Ž(𝑠)𝑑𝑠. (12)

If π‘˜π‘š > 0, the inequality (12) yields that

𝐷+π‘’π‘š(𝑑1) < π‘¦β€²π‘š(𝑑1),

which contradicts the inequality in (10). If π‘˜π‘š = 0, theinequality (12) yields that 0 < 0. It is a contradiction.

Thus (9) holds. Therefore, letting πœ–β†’ 0, we have

𝑒𝑖(𝑑) ≀ π‘§π‘–π‘’βˆ’πœ†

∫ 𝑑𝑑0

β„Ž(𝑠)𝑑𝑠, 𝑑 ∈ [𝑑0, 𝑏), 𝑖 ∈ 𝒩 .

The proof is completed.

631

Page 3: [IEEE 2010 Third International Workshop on Advanced Computational Intelligence (IWACI) - Suzhou, China (2010.08.25-2010.08.27)] Third International Workshop on Advanced Computational

3

IV. GLOBAL EXPONENTIAL STABILIT

In this section, we will obtain the sufficient conditionensuring global exponential stability of the zero solution ofthe singular system (1) by using the property of 𝑀 -Matrixand employing Theorem 1. Here, we firstly introduce thefollowing assumptions.(𝐴1) There exist nonnegative constants �̄�𝑗 and 𝑣𝑗 such

that continuous functions 𝑓𝑗(β‹…) and 𝑔𝑗(β‹…) satisfy

βˆ£π‘“π‘—(𝑦)∣ ≀ �̄�𝑗 βˆ£π‘¦βˆ£, βˆ£π‘”π‘—(𝑦)∣ ≀ 𝑣𝑗 βˆ£π‘¦βˆ£, 𝑗 ∈ 𝒩 , 𝑦 ∈ 𝑅.

(𝐴2) Let 𝑆 = βˆ’(𝑃 +𝑄) be an 𝑀 -matrix, where

𝑃 = βˆ’πΆ + [𝐴]+π‘ˆ, 𝑄 = [𝐡]+𝑉, 𝐡 = (𝑏𝑖𝑗)𝑛×𝑛,

𝐢 = diag{𝑐1, . . . , 𝑐𝑛} > 0, 𝐴 = (π‘Žπ‘–π‘—)𝑛×𝑛,

π‘ˆ = diag{οΏ½Μ„οΏ½1, . . . , �̄�𝑛}, 𝑉 = diag{𝑣1, . . . , 𝑣𝑛}.

(𝐴3) Let π‘†π‘˜ = [𝐸+πΆπ›Όπ‘˜]+βˆ’ [𝐴]+π‘ˆ [π›Ύπ‘˜]

+, π‘†βˆ’1π‘˜ exist and

π‘†βˆ’1π‘˜ β‰₯ 0, π‘˜ = 1, 2, . . . , where

𝐸 = diag{𝑒1, . . . , 𝑒𝑛} β‰₯ 0,

π›Όπ‘˜ = diag{𝛼1π‘˜, . . . , π›Όπ‘›π‘˜},

π›Ύπ‘˜ = diag{𝛾1π‘˜, . . . , π›Ύπ‘›π‘˜}.

(𝐴4) There exists a positive constant 𝜎 such that

lnπœŽπ‘˜π‘‘π‘˜ βˆ’ π‘‘π‘˜βˆ’1

≀ 𝜎 < πœ†, π‘˜ = 1, 2, . . . . (13)

Where the positive constant πœ† satisfies

[πœ†πΈ + 𝑃 +π‘„π‘’πœ†πœ ]π‘§βˆ— < 0, (14)

for a given π‘§βˆ— ∈ Ω𝑀 (𝑆) and πœŽπ‘˜ β‰₯ 1 satisfies

πœŽπ‘˜π‘§βˆ— β‰₯ (π‘†βˆ’1

π‘˜ 𝐸 + π‘†βˆ’1π‘˜ [𝐡]+𝑉 [π›½π‘˜]

+π‘’πœ†πœ )π‘§βˆ—,π‘˜ = 1, 2, . . . , (15)

where π›½π‘˜ = diag{𝛽1π‘˜, . . . , π›½π‘›π‘˜}.Theorem 2: Assume that (𝐴1) - (𝐴4) hold. Then the zero

solution of the singular system (1) is globally exponentiallystable and the exponential convergence rate is equal to πœ†βˆ’πœŽ.

Proof: By the property of the Dirac impulse function 𝛿(β‹…),we see that the derivative π‘₯′𝑖(𝑑) of π‘₯𝑖(𝑑) exists on (π‘‘π‘˜βˆ’1, π‘‘π‘˜).Thus the system (1) becomes

𝑒𝑖π‘₯′𝑖(𝑑) = βˆ’π‘π‘–π‘₯𝑖(𝑑) +

π‘›βˆ‘π‘—=1

π‘Žπ‘–π‘—π‘“π‘—(π‘₯𝑗(𝑑))

+π‘›βˆ‘

𝑗=1

𝑏𝑖𝑗𝑔𝑗(π‘₯𝑗(π‘‘βˆ’ πœπ‘–π‘—(𝑑))),

𝑖 ∈ 𝒩 , 𝑑 ∈ (π‘‘π‘˜βˆ’1, π‘‘π‘˜). (16)

From (16) and (𝐴1), it is easy to obtain

𝑒𝑖𝐷+∣π‘₯𝑖(𝑑)∣ ≀ βˆ’π‘π‘–βˆ£π‘₯𝑖(𝑑)∣+

π‘›βˆ‘π‘—=1

βˆ£π‘Žπ‘–π‘— βˆ£βˆ£π‘“π‘—(π‘₯𝑗(𝑑))∣

+

π‘›βˆ‘π‘—=1

βˆ£π‘π‘–π‘— βˆ£βˆ£π‘”π‘—(π‘₯𝑗(π‘‘βˆ’ πœπ‘–π‘—(𝑑)))∣

≀ βˆ’π‘π‘–βˆ£π‘₯𝑖(𝑑)∣+π‘›βˆ‘

𝑗=1

βˆ£π‘Žπ‘–π‘— βˆ£οΏ½Μ„οΏ½π‘— ∣π‘₯𝑗(𝑑)∣

+

π‘›βˆ‘π‘—=1

βˆ£π‘π‘–π‘— βˆ£π‘£π‘— ∣π‘₯𝑗(π‘‘βˆ’ πœπ‘–π‘—(𝑑))∣,

𝑖 ∈ 𝒩 , 𝑑 ∈ (π‘‘π‘˜βˆ’1, π‘‘π‘˜).

That is,

𝐸𝐷+[π‘₯(𝑑)]+ ≀ 𝑃 [π‘₯(𝑑)]+ +𝑄[π‘₯(𝑑)]+𝜏 ,

𝑑 ∈ (π‘‘π‘˜βˆ’1, π‘‘π‘˜), π‘˜ = 1, 2, . . . . (17)

Since 𝑆 is an 𝑀 -matrix, from Lemma 1 we may choosea vector π‘§βˆ— = (𝑧1, . . . , 𝑧𝑛)

𝑇 ∈ Ω𝑀 (𝑆) and π‘§βˆ— β‰₯ 𝐼𝑛 =(1, . . . , 1)𝑇 ∈ 𝑅𝑛 such that

π‘†π‘§βˆ— > 0 or [𝑃 +𝑄]π‘§βˆ— < 0.

By using continuity, we obtain that there must exist a positiveconstant πœ† satisfying the inequality (14).

For the initial condition π‘₯(𝑑0 + 𝑠) = πœ™(𝑠), 𝑠 ∈ [βˆ’πœ, 0],where πœ™ ∈ 𝑃𝐢[[βˆ’πœ, 0], 𝑅𝑛] and 𝑑0 ∈ 𝑅 (without loss ofgenerality, we assume 𝑑0 < 𝑑1), we can get

[π‘₯(𝑑)]+ ≀ π‘§βˆ—βˆ₯πœ™βˆ₯πœπ‘’βˆ’πœ†(π‘‘βˆ’π‘‘0), 𝑑0 βˆ’ 𝜏 ≀ 𝑑 ≀ 𝑑0. (18)

Then, all conditions of Theorem 1 are satisfied by (17),(18), Condition (𝐴2), β„Ž(𝑑) ≑ 1 and 𝐻 = 𝜏 . So

[π‘₯(𝑑)]+ ≀ π‘§βˆ—βˆ₯πœ™βˆ₯πœπ‘’βˆ’πœ†(π‘‘βˆ’π‘‘0), 𝑑0 ≀ 𝑑 < 𝑑1. (19)

Suppose that for all π‘š = 1, . . . , π‘˜, the inequalities

[π‘₯(𝑑)]+ ≀ 𝜎0 . . . πœŽπ‘šβˆ’1π‘§βˆ—βˆ₯πœ™βˆ₯πœπ‘’βˆ’πœ†(π‘‘βˆ’π‘‘0),

π‘‘π‘šβˆ’1 ≀ 𝑑 < π‘‘π‘š, (20)

hold, where 𝜎0 = 1.On the other hand, (1) implies that

𝑒𝑖[π‘₯𝑖(π‘‘π‘˜)βˆ’ π‘₯𝑖(π‘‘π‘˜ βˆ’ πœ–)]=

∫ π‘‘π‘˜

π‘‘π‘˜βˆ’πœ–

[βˆ’π‘π‘–π‘₯𝑖(𝑑)𝐷𝑣𝑖(𝑠)

+

π‘›βˆ‘π‘—=1

π‘Žπ‘–π‘—π‘“π‘—(π‘₯𝑗(𝑑))𝐷𝑗𝑒(𝑠)

+

π‘›βˆ‘π‘—=1

𝑏𝑖𝑗𝑔𝑗(π‘₯𝑗(π‘‘βˆ’ πœπ‘–π‘—(𝑑)))𝐷𝑀𝑗(𝑠)], 𝑖 ∈ 𝒩 , (21)

632

Page 4: [IEEE 2010 Third International Workshop on Advanced Computational Intelligence (IWACI) - Suzhou, China (2010.08.25-2010.08.27)] Third International Workshop on Advanced Computational

4

where πœ– > 0 is sufficiently small, as πœ–β†’ 0+, which yields

𝑒𝑖[π‘₯𝑖(π‘‘π‘˜)βˆ’ π‘₯𝑖(π‘‘βˆ’π‘˜ )]

= βˆ’π‘π‘–π‘₯𝑖(π‘‘π‘˜)π›Όπ‘–π‘˜ +

π‘›βˆ‘π‘—=1

π‘Žπ‘–π‘—π‘“π‘—(π‘₯𝑗(π‘‘π‘˜))π›Ύπ‘—π‘˜

+π‘›βˆ‘

𝑗=1

𝑏𝑖𝑗𝑔𝑗(π‘₯𝑗(π‘‘π‘˜ βˆ’ πœπ‘–π‘—(π‘‘π‘˜)))π›½π‘—π‘˜, 𝑖 ∈ 𝒩 . (22)

This, together with (𝐴1), yields that

∣(𝑒𝑖 + π‘π‘–π›Όπ‘–π‘˜)π‘₯𝑖(π‘‘π‘˜)∣

= βˆ£π‘’π‘–π‘₯𝑖(π‘‘βˆ’π‘˜ ) +π‘›βˆ‘

𝑗=1

π‘Žπ‘–π‘—π‘“π‘—(π‘₯𝑗(π‘‘π‘˜))π›Ύπ‘—π‘˜

+

π‘›βˆ‘π‘—=1

𝑏𝑖𝑗𝑔𝑗(π‘₯𝑗(π‘‘π‘˜ βˆ’ πœπ‘–π‘—(π‘‘π‘˜)))π›½π‘—π‘˜βˆ£

≀ π‘’π‘–βˆ£π‘₯(π‘‘βˆ’π‘˜ )∣+π‘›βˆ‘

𝑗=1

βˆ£π‘Žπ‘–π‘— βˆ£οΏ½Μ„οΏ½π‘— ∣π‘₯𝑗(π‘‘π‘˜)βˆ£βˆ£π›Ύπ‘—π‘˜βˆ£

+

π‘›βˆ‘π‘—=1

βˆ£π‘π‘–π‘— βˆ£π‘£π‘— ∣π‘₯𝑗(π‘‘π‘˜ βˆ’ πœπ‘–π‘—(π‘‘π‘˜))βˆ£βˆ£π›½π‘—π‘˜βˆ£

≀ π‘’π‘–βˆ£π‘₯(π‘‘βˆ’π‘˜ )∣+π‘›βˆ‘

𝑗=1

βˆ£π‘Žπ‘–π‘— βˆ£οΏ½Μ„οΏ½π‘— ∣π‘₯𝑗(π‘‘π‘˜)βˆ£βˆ£π›Ύπ‘—π‘˜βˆ£

+π‘›βˆ‘

𝑗=1

βˆ£π‘π‘–π‘— βˆ£π‘£π‘— [∣π‘₯𝑗(π‘‘π‘˜)∣]𝜏 βˆ£π›½π‘—π‘˜βˆ£, 𝑖 ∈ 𝒩 . (23)

That is,

[𝐸 + πΆπ›Όπ‘˜]+[π‘₯(π‘‘π‘˜)]

+

≀ 𝐸[π‘₯(π‘‘βˆ’π‘˜ )]+ + [𝐴]+π‘ˆ [π›Ύπ‘˜]

+[π‘₯(π‘‘π‘˜)]+

+[𝐡]+𝑉 [π›½π‘˜]+[π‘₯(π‘‘π‘˜)]

+𝜏 . (24)

Then, we have

([𝐸 + πΆπ›Όπ‘˜]+ βˆ’ [𝐴]+π‘ˆ [π›Ύπ‘˜]

+)[π‘₯(π‘‘π‘˜)]+

≀ 𝐸[π‘₯(π‘‘βˆ’π‘˜ )]+ + [𝐡]+𝑉 [π›½π‘˜]

+[π‘₯(π‘‘π‘˜)]+𝜏 (25)

yielding, together with (𝐴3), that

[π‘₯(π‘‘π‘˜)]+ ≀ π‘†βˆ’1

π‘˜ 𝐸[π‘₯(π‘‘βˆ’π‘˜ )]+ + π‘†βˆ’1

π‘˜ [𝐡]+𝑉 [π›½π‘˜]+[π‘₯(π‘‘π‘˜)]

+𝜏 . (26)

Then, from (15), (20) and (26)

[π‘₯(π‘‘π‘˜)]+ ≀ (π‘†βˆ’1

π‘˜ 𝐸 + π‘†βˆ’1π‘˜ [𝐡]+𝑉 [π›½π‘˜]

+π‘’πœ†πœ )

Γ—π‘§βˆ— 𝜎0 . . . πœŽπ‘˜βˆ’1βˆ₯πœ™βˆ₯πœπ‘’βˆ’πœ†(π‘‘π‘˜βˆ’π‘‘0)

≀ π‘§βˆ— 𝜎0 . . . πœŽπ‘˜βˆ’1πœŽπ‘˜βˆ₯πœ™βˆ₯πœπ‘’βˆ’πœ†(π‘‘π‘˜βˆ’π‘‘0). (27)

This, together with (20) and πœŽπ‘– β‰₯ 1, 𝑖 = 1, 2, . . . , leads to

[π‘₯(𝑑)]+ ≀ 𝜎0 . . . πœŽπ‘˜βˆ’1πœŽπ‘˜π‘§βˆ—βˆ₯πœ™βˆ₯πœπ‘’βˆ’πœ†(π‘‘βˆ’π‘‘0)

= 𝜎0 . . . πœŽπ‘˜βˆ’1πœŽπ‘˜βˆ₯πœ™βˆ₯πœπ‘’βˆ’πœ†(π‘‘π‘˜βˆ’π‘‘0)π‘§βˆ—π‘’βˆ’πœ†(π‘‘βˆ’π‘‘π‘˜),

𝑑 ∈ [π‘‘π‘˜ βˆ’ 𝜏, π‘‘π‘˜]. (28)

By Lemma 1 again, the vector

𝜎0 . . . πœŽπ‘˜βˆ’1πœŽπ‘˜βˆ₯πœ™βˆ₯πœπ‘’βˆ’πœ†(π‘‘π‘˜βˆ’π‘‘0)π‘§βˆ— ∈ Ω𝑀 (𝑆).

Then, all conditions of Theorem 1 are satisfied again by (17),(28), Condition (𝐴2), β„Ž(𝑑) ≑ 1 and 𝐻 = 𝜏 . So

[π‘₯(𝑑)]+ ≀ 𝜎0 . . . πœŽπ‘˜βˆ’1πœŽπ‘˜βˆ₯πœ™βˆ₯πœπ‘’βˆ’πœ†(π‘‘π‘˜βˆ’π‘‘0)π‘§βˆ—π‘’βˆ’πœ†(π‘‘βˆ’π‘‘π‘˜)

= 𝜎0 . . . πœŽπ‘˜βˆ’1πœŽπ‘˜π‘§βˆ—βˆ₯πœ™βˆ₯πœπ‘’βˆ’πœ†(π‘‘βˆ’π‘‘0),

π‘‘π‘˜ ≀ 𝑑 < π‘‘π‘˜+1. (29)

By the mathematical induction, we can conclude that

[π‘₯(𝑑)]+ ≀ 𝜎0 . . . πœŽπ‘˜βˆ’1π‘§βˆ—βˆ₯πœ™βˆ₯πœπ‘’βˆ’πœ†(π‘‘βˆ’π‘‘0),

π‘‘π‘˜βˆ’1 ≀ 𝑑 < π‘‘π‘˜, π‘˜ = 1, 2, . . . . (30)

Noticing that πœŽπ‘˜ ≀ π‘’πœŽ(π‘‘π‘˜βˆ’π‘‘π‘˜βˆ’1) by (13), we can use (30) toconclude that

[π‘₯(𝑑)]+ ≀ π‘’πœŽ(𝑑1βˆ’π‘‘0) . . . π‘’πœŽ(π‘‘π‘˜βˆ’1βˆ’π‘‘π‘˜βˆ’2)π‘§βˆ—βˆ₯πœ™βˆ₯πœπ‘’βˆ’πœŽ(π‘‘βˆ’π‘‘0)

≀ π‘§βˆ—βˆ₯πœ™βˆ₯πœπ‘’πœŽ(π‘‘βˆ’π‘‘0)π‘’βˆ’πœ†(π‘‘βˆ’π‘‘0)

= π‘§βˆ—βˆ₯πœ™βˆ₯πœπ‘’βˆ’(πœ†βˆ’πœŽ)(π‘‘βˆ’π‘‘0),

𝑑 ∈ [π‘‘π‘˜βˆ’1, π‘‘π‘˜), π‘˜ = 1, 2, . . . .

So,

[π‘₯(𝑑)]+ ≀ π‘§βˆ—βˆ₯πœ™βˆ₯πœπ‘’βˆ’(πœ†βˆ’πœŽ)(π‘‘βˆ’π‘‘0), 𝑑 β‰₯ 𝑑0.

This implies that the conclusion of the theorem holds.Remark 1: Condition (𝐴3) must hold when π›Όπ‘˜ > 0 and

π›Ύπ‘˜ is sufficiently small. In fact, the positive diagonal matrix[𝐸 + πΆπ›Όπ‘˜]

+ is an 𝑀 -matrix when π›Όπ‘˜ > 0. So, π‘†π‘˜ = [𝐸 +πΆπ›Όπ‘˜]

+ βˆ’ [𝐴]+π‘ˆ [π›Ύπ‘˜]+ is also an 𝑀 -matrix for sufficiently

small π›Ύπ‘˜. Then, π‘†βˆ’1π‘˜ exists and π‘†βˆ’1

π‘˜ β‰₯ 0.Remark 2: When πœπ‘–π‘—(𝑑) ≑ 𝜏 and π›Ύπ‘˜ = π›½π‘˜, some global

exponential stability criteria for the singular impulsive system(1) have been established in [7]. However, from assumption1) in Theorem 2 and Corollary 4.1 of [7], the impulsivecoefficient max1β‰€π‘˜<∞, 1≀𝑖≀𝑛{π›½π‘–π‘˜} must be bounded whenmax1β‰€π‘˜<∞, 1≀𝑖≀𝑛{π›Όπ‘–π‘˜} is bounded. Here, Theorem 2 abovedoes not require the boundedness of impulsive coefficient anddrops the additional assumption that π‘‘π‘˜ βˆ’ π‘‘π‘˜βˆ’1 β‰₯ π›Ώπœ, 𝛿 > 1in [7].

If 𝐸 is a 𝑛 Γ— 𝑛 unit matrix, the system (1) becomes theimpulsive delay differential system without singularity,

𝐷π‘₯𝑖(𝑑) = βˆ’π‘π‘–π‘₯𝑖(𝑑)𝐷𝑣𝑖(𝑑)

+π‘›βˆ‘

𝑗=1

π‘Žπ‘–π‘—π‘“π‘—(π‘₯𝑗(𝑑))𝐷𝑒𝑗(𝑑)

+π‘›βˆ‘

𝑗=1

𝑏𝑖𝑗𝑔𝑗(π‘₯𝑗(π‘‘βˆ’ πœπ‘–π‘—(𝑑)))𝐷𝑀𝑗(𝑑),

𝑖 ∈ 𝒩 . (31)

For the system (31), we have the following corollary byTheorem 2.

Corollary 1: Assume that (𝐴1) - (𝐴4) hold. Then the zerosolution of the system (31) is globally exponentially stableand the exponential convergence rate is equal to πœ†βˆ’ 𝜎.

Remark 3: When coefficient matrix 𝐡 ≑ 0, the system(31) has been investigated in [10] where function 𝑓𝑗(β‹…) isdifferentiable and 𝑓 ′𝑗(β‹…) is invertible and bounded, 𝑗 ∈ 𝒩 .And Guan and Chen also have discussed the system (31)

633

Page 5: [IEEE 2010 Third International Workshop on Advanced Computational Intelligence (IWACI) - Suzhou, China (2010.08.25-2010.08.27)] Third International Workshop on Advanced Computational

5

with πœπ‘–π‘—(𝑑) ≑ 𝜏 [11]. In [11], 𝑓𝑗(β‹…) and 𝑔𝑗(β‹…) must satisfy𝑧𝑓𝑗(𝑧) β‰₯ 0 and 𝑧𝑔𝑗(𝑧) β‰₯ 0, 𝑧 ∈ 𝑅, 𝑗 ∈ 𝒩 . However, inCorollary 1 above, these restrictive conditions are removed.

V. ILLUSTRATIVE EXAMPLE

The following illustrative example will demonstrate theeffectiveness of our results.

Example 1: Consider the following impulsive singularsystem with time-varying delaysβŽ§βŽ¨βŽ©π·π‘¦1(𝑑) = βˆ’6𝑦1(𝑑)𝐷𝑣1(𝑑) + 3βˆ£π‘¦1(π‘‘βˆ’ 𝜏11(𝑑))∣

×𝐷𝑀1(𝑑)βˆ’ 2𝑦2(π‘‘βˆ’ 𝜏12(𝑑))𝐷𝑀2(𝑑),0 = βˆ’5𝑦2(𝑑)𝐷𝑣2(𝑑)βˆ’ βˆ£π‘¦1(π‘‘βˆ’ 𝜏21(𝑑))βˆ£π·π‘€1(𝑑)

+3 cos(𝑦2(π‘‘βˆ’ 𝜏22(𝑑)))𝐷𝑀2(𝑑),

(32)

where πœπ‘–π‘—(𝑑) = ∣ sin((𝑖+ 𝑗)𝑑)∣ ≀ 1Ξ”= 𝜏 and

𝐷𝑣𝑖(𝑑) = 1+βˆžβˆ‘π‘˜=1

𝛼𝛿(π‘‘βˆ’ π‘‘π‘˜), 𝐷𝑀𝑖(𝑑) = 1+βˆžβˆ‘π‘˜=1

𝛽𝛿(π‘‘βˆ’ π‘‘π‘˜),

for 𝑖, 𝑗 = 1, 2, where 𝛿(β‹…) is the Dirac impulse function andthe impulsive moments π‘‘π‘˜ ( π‘˜ = 1, 2, . . . ) satisfy: 𝑑1 = 0.3,𝑑1 < 𝑑2 < . . . and limπ‘˜β†’+∞ π‘‘π‘˜ = +∞.

Thus the parameters of Conditions (𝐴1) - (𝐴4) are asfollows:

𝐸 =

(1 00 0

), 𝑉 =

(1 00 1

),

𝐢 =

(6 00 5

), 𝐴 =

(0 00 0

),

π›Όπ‘˜ = diag{𝛼, 𝛼}, π›½π‘˜ = diag{𝛽, 𝛽},

𝐡 =

(3 βˆ’2βˆ’1 3

), 𝑄 = [𝐡]+𝑉 =

(3 21 3

),

𝑃 = βˆ’πΆ + [𝐴]+π‘ˆ = βˆ’πΆ =

( βˆ’6 00 βˆ’5

),

𝑆 = βˆ’(𝑃 +𝑄) =

(3 βˆ’2βˆ’1 2

),

π‘†π‘˜ = [𝐸 + πΆπ›Όπ‘˜]+ βˆ’ [𝐴]+π‘ˆ [π›Ύπ‘˜]

+

=

( ∣1 + 6π›Όβˆ£ 00 5βˆ£π›Όβˆ£

), (33)

which yield that 𝑆 is an 𝑀 -matrix, π‘†βˆ’1π‘˜ exists when 𝛼 βˆ•= 0

and 𝛼 βˆ•= βˆ’ 16 . Furthermore,

Ω𝑀 (𝑆) = {(𝑧1, 𝑧2)𝑇 > 0 ∣ 23𝑧2 < 𝑧1 < 2𝑧2},

π‘†βˆ’1π‘˜ =

(1

∣1+6π›Όβˆ£ 0

0 15βˆ£π›Όβˆ£

).

Let π‘§βˆ— = (1, 1)𝑇 ∈ Ω𝑀 (𝑆) and πœ† = 0.1 which satisfies theinequality (πœ†πΈ + 𝑃 + π‘„π‘’πœ†πœ )π‘§βˆ— < 0. So πœŽπ‘˜ can satisfy theinequality

πœŽπ‘˜π‘§βˆ— β‰₯ (π‘†βˆ’1

π‘˜ 𝐸 + π‘†βˆ’1π‘˜ [𝐡]+𝑉 [π›½π‘˜]

+π‘’πœ†πœ )π‘§βˆ—

= (π‘†βˆ’1π‘˜ 𝐸 + π‘†βˆ’1

π‘˜ [𝐡]+βˆ£π›½βˆ£π‘’0.1)π‘§βˆ—,π‘˜ = 1, 2, . . . , (34)

provided that

πœŽπ‘˜ β‰₯ max{1 + 5βˆ£π›½βˆ£π‘’0.1∣1 + 6π›Όβˆ£ ,

4βˆ£π›½βˆ£π‘’0.15βˆ£π›Όβˆ£ },

π‘˜ = 1, 2, . . . . (35)

Case 1: Let 𝛼 = βˆ’1, 𝛽 = 𝑒0.2π‘˜, πœŽπ‘˜ = 𝑒0.5π‘˜ and π‘‘π‘˜ βˆ’π‘‘π‘˜βˆ’1 = 6π‘˜, then πœŽπ‘˜ satisfies the inequality (35) and πœŽπ‘˜ β‰₯ 1,

lnπœŽπ‘˜π‘‘π‘˜ βˆ’ π‘‘π‘˜βˆ’1

=ln 𝑒0.5π‘˜

6π‘˜β‰€ 0.084 = 𝜎 < πœ†,

π‘˜ = 1, 2, . . . .

Clearly, all conditions of Theorem 2 are satisfied. So the zerosolution of the singular system (32) is globally exponentiallystable and the exponential convergence rate is equal to 0.016.

Case 2: Let 𝛼 = 𝑒0.1, 𝛽 = 1, πœŽπ‘˜ = 1 and π‘‘π‘˜ βˆ’ π‘‘π‘˜βˆ’1 =πœ‡π‘˜ ∈ (0, 𝜏), then πœŽπ‘˜ satisfies the inequality (35) and πœŽπ‘˜ β‰₯ 1,

lnπœŽπ‘˜π‘‘π‘˜ βˆ’ π‘‘π‘˜βˆ’1

= 0 = 𝜎 < πœ†, π‘˜ = 1, 2, . . . .

So by Theorem 2, the zero solution of the singular system(32) is globally exponentially stable and the exponentialconvergence rate is equal to 0.1.

Remark 4: Even if πœπ‘–π‘—(𝑑) ≑ 𝜏 , the conditions in Theorem4.1 and Corollary 4.1 of [7] are not satisfied since in Case 1π›½π‘–π‘˜ = 𝑒0.2π‘˜ β†’ ∞, π‘˜ β†’ ∞, while max1β‰€π‘˜<∞, 1≀𝑖≀2{π›Όπ‘–π‘˜}is bounded, and in Case 2 πœ‡π‘˜ ∈ (0, 𝜏) which does not satisfythe assumption that π‘‘π‘˜ βˆ’ π‘‘π‘˜βˆ’1 β‰₯ π›Ώπœ, 𝛿 > 1 in [7].

REFERENCES

[1] J.D. Aplevich, Implicit Linear Systems. Springer-Verlag, New York, US,1991.

[2] S.L. Campbell, Singular Systems of Differential Equations. Pitman, NewYork, US, 1982.

[3] Y. Li and Y. Liu, β€œBifurcation on stability of singular systems withdelay”, International Journal of Systems Science, vol. 30, no. 6, pp.643-649, 1999.

[4] S. Xu, P.V. Dooren, R. Stefan and J. Lam, β€œRobust stability andstabilization for singular systems with state delay and parameter un-certainty”, IEEE Transactions on Automatic Control, vol. 47, pp. 1122-1128, 2002.

[5] S. Xu, J. Lam and C. Yang, β€œRobust 𝐻∞ control for uncertain singularsystems with state delay”, International Journal Robust NonlinearControl, vol. 13 pp. 1213-1223, 2003.

[6] Z. Guan, J. Yao and D.J. Hill, β€œRobust 𝐻∞ control of singularimpulsive systems with uncertain perturbations”, IEEE Transactions onCircuits and Systems - II: Express Briefs, vol. 52, pp. 293-298, 2005.

[7] Z. Guan, C.W. Chan, Andrew Y. T. Leung and G. Chen, β€œRobust sta-bilization of singular-impulsive-delayed systems with nonlinear pertur-bations”, IEEE Transactions on Circuits and Systems - I: FoundmentalTheory and Application, vol. 48, pp. 1011-1019, 2001.

[8] S.G. Deo and S.G. Pandit, Differential Systems Involving Impulses.Springer-Verlag, New York, US, 1982.

[9] Z. Yang, D. Xu and L. Xiang, β€œExponential p-stability of impulsivestochastic differential equations with delays”, Phys Lett A , vol. 359,pp. 129-137, 2006.

[10] Z. Guan, J. Lam and G. Chen, β€œOn impulsive autoassociative neuralnetworks”, Neural Networks , vol. 13, pp. 63-69, 2000.

[11] Z. Guan and G. Chen, β€œOn delayed impulsive Hopfield neural net-works”, Neural Networks, vol. 12, pp. 273-280, 1999.

634


Top Related