# Session Keynote Lyapunov Function Candidates for Descriptor Systems Problems and Solutions

Session Keynote: Lyapunov Function Candidates for Descriptor Systems: Problems and Solutions
Vladimir B. Baji? c Centre for Engineering Research, Technikon Natal, P .O.Box 953, Durban 4000, Republic of South Africa e-mail: bajic.v@umfolozi.ntech.ac.za Keywords: Lyapunov’s direct method, Lyapunov functions, descriptor (singular) systems, stability, qualitative analysis ABSTRACT This paper presents in a systematic way problems encountered in the construction of Lyapunov function candidates for descriptor systems. The solutions to some of the major difficulties in the application of Lyapunov’s direct method to descriptor systems are presented. Some new results regarding the extension of Lyapunov’s direct method tied to the construction of the Lyapunov functions are given. Also, the algebraic necessary and sufficient conditions for some specific properties of motions of descriptor systems are developed. INTRODUCTION AND BACKGROUND It is a tradition to consider the equations describing internal system dynamics of continuous time systems in the normal form of the so-called state equations (1) (% ' s E|c %c  5 Uc o and % 5 Uc where ( ' _*_| is the ordinary time derivative operator, and where | 5 U, 5 U denote the time, the input vector and the state vector, respectively. Systems governed by the model (1) we usually call the state variable systems. However, there are physical systems for which the state models do not exist. It was recognized that the more natural models for these cases go beyond the classical state-variable description (1). These models are given by (2) s E|c %c (%c  ' f 5 U? where R 9' ? may be allowed. Systems with such models we denote as implicit. In the special case the models (2) contain the canonical form of models linear in (%, and they are of the form  E|c %c (% ' s E|c %c  (3) It should be pointed out that the matrix  can even be a rectangular one. Systems governed by (3) are known as descriptor, as well as singular, semi-state, generalized state-space or differential-algebraic systems. STABILITY ANALYSIS OF DESCRIPTOR SYSTEMS The surveys of some of the results concerning both continuous and discrete descriptor and general implicit systems can be found, for example, in the books [1]-[3], [6]-[7] and [10] and in the special issues of the journal Circuits, Systems and Signal Processing [8]-[9]. The systematic introduction and presentation of part of the results relating to the general application of Lyapunov’s direct method (LDM) for the analysis of DS is given in [2], [3]. For some other results on stability of DS, particularly with regard to mechanical DS, see [12]-[16]. It has been shown in [2], [4], that specific structure of DS may lead to several problems regarding the construction of Lyapunov function candidates (LFCs) for the intended qualitative analysis. These problems do not appear in the application of the LDM to systems in the normal forms (1). This paper aims to contribute to the general methodology relating to the construction of the LFCs for descriptor systems. We present in a systematic way problems

encountered in the construction of LFCs for continuous-time DS, as well as the solutions to some of the major difficulties in the application of the LDM. Some new results regarding the extension of the LDM tied to the construction of LFCs are given. PROBLEM STATEMENT AND SIGNIFICANCE: LDM AND THE ROLE OF LFC There are essentially two problems that relate to the preliminary construction of the LFC:  the problems that appear in the process of evaluation of the total time derivative (TTD) of an LFC along the motions of (2) and (3).  the structure and properties of the matrix  in (3) The application of the LDM requires the selection of a concrete LFC. When the exact analytical form of an LFC is known and when the necessary and sufficient conditions that ensure the appropriate qualitative concept of motion are expressed directly in terms of system and LFC parameters, then we say that the construction of the LFC is given. Unfortunately, in the general case the main drawback of the LDM is the conceptual nonexistence of a systematic procedure for the practical construction of an LFC for the qualitative concept of interest (one of possible general solutions for this is proposed in this paper). There are only a few results of the general nature on the construction of the LFC for DS (see references in [2]). So far there are no global answers on how to select an LFC for DS. Thus any general hint that concerns the construction of an LFC is of particular importance. The aim of this presentation is to provide some suitable solutions of this problem in a very specific way by the evaluation of the derivative of an LFC along the solutions of DS. This poses practical problems in finding a TTD of an LFC for DS, and, also, directly restricts the classes of useful LFCs. To make this more evident let the matrix  and the functions s and in (3) be sufficiently smooth and such that the systems considered possess the continuous and differentiable in | solutions in some domain. For simplicity, let T E|c % be an arbitrary function of | and %, which is continuously differentiable, and let it be an LFC for systems (2) or (3). In any qualitative analysis of the solutions of systems considered by means of the LDM, specific properties of the LFC are required. In what follows the TTD of T along the solutions of any of the systems (1) to (3) we will denote by (P T In all cases, the calculation of the TTD (P T d|c E|o of the LFC T , along the system motion , has to be found. In general, this is not possible in a direct way for systems (2) or (3), unless T E|c % depends on % in a specific manner. As is well known, (P T d|c E|o can be found without the knowledge of solutions if the system analyzed is in the normal form (1) [17, p. 12]. Using the same argument, the (P T d|c E|o of the LFC T E|c % along the motion is given by   YT d|c E|o YT d|c E|o e (P T d|c E|o ' n ( E| (4) Y| Y may be computed along the solutions of (2) or (3) if ( is obtained from these equations and substituted in (4). It is obvious that this can be done directly for systems in the normal form (1), in which case along the solutions of (1) we have  e YT E|c % YT E|c % (P T E|c % ' n sE|c %c  (5) Y| Y% However, systems (2) or (3) need not be solvable in (%. Thus, in the cases when (2) or (3) are inherently singular (non-solvable in (%), (4) cannot be used directly for the evaluation of (P T d|c E|o. This is one of the main difficulties relating to successful construction of the LFC for DS. In this exposition will provide some general solutions to this problem. LFC FOR IMPLICIT SYSTEMS: ALGEBRAIC NECESSARY AND SUFFICIENT CONDITION We consider a system P governed by implicit differential equations (2). Let W ' i| 5 U G | 5o|r c |s dj, |r c |s 5 U, |r |s . The solutions of (2) are functions G W W Uc 6 E|c |f c %f  :\$ E|c |f c %f  5

Uc c where |f and %f are regarded as the initial moment and the initial value of a solution c respectively. The notation E|c |f c %f  represents the value of the solution of P at the moment |c that at the moment |f had the value %f  We consider the case when (2) has only unique solutions that are continuous functions of their arguments and differentiable in |. Also, we allow that the E|f c %f  5 W V8 E|f c |s c where V8 is the set of the consistent initial values %f at the moment |f such that from each %f 5 V8 E|f c |s  at least one solution that exists on WE|f  ' d|f c |s d Wc is generated. Further, we will consider properties of solutions of P only in the domain G ' W V8 E|r c |s  6 E|c %. Obviously, G is invariant w.r.t. solutions c i.e. if E|f c %f  5 Gc then E|c E|c |f c %f  5 G for any |  |f c | 5 W. We will provide the algebraic necessary and sufficient conditions for a function T to be a Lyapunov function for (2), by which the following useful qualitative concepts can be verified. The concepts considered are expressed by S d|c %E|c |f c %f o   d|c |f c SE|f c %f o c SE|f c %f  '  d|f c |f c SE|f c %f o c (6)

where S and  are scalar real functions. When all solutions of P in the domain G satisfy (6) we say that P has the property  in G We can give different meaning to the functions S and  Theorem 1. Let an auxiliary function T G W Uc 6 E|c % :\$ T E|c % 5 Uc T E|c % 5 EW Uc c be chosen and let C G W U \$ U define a scalar differential equation ( ' CE|c  that ;E|f c f  5 W U has unique solutions E|c |f c f  that are continuously differentiable in |. Let the Eulerian derivative (P T along the solutions of the model of (2) be given by (P T E|c % ' }E|c % If (2) has unique solutions c then for every solution that exists on WE|f  the property  given by (6) will hold if and only if }E|c %  Cd|c T E|c %o in WE|f  V8 E|r c |s  The last conditions are also necessary and sufficient that T E|c %  SE|c % be the Lyapunov function for EPc    This result is based on [5] and is quite opposite to Gruji? ’s method [11]. It gives the algebraic necc essary and sufficient conditions for the verification of the examined property as well as the direct construction of T Some partial results of this approach have been already successfully applied to the problems of nonlinear uncertain and fuzzy logic systems. GENERAL PROCEDURE FOR ELIMINATION OF (% FROM THE TTD OF AN LFC Consider the model (3) and assume that possibly all entries of the matrix  are dependent on | and %. The construction of an LFC in such a situation is a complicated task. The method that follows is a general method for the calculation of the TTD of an LFC for DS (3) with nonlinear and possibly rectangular matrix  . Let T be a scalar LFC, T G U Uc 6 E|c % :\$ T E|c % 5 U and let T be differentiable in the ordinary sense. If the matrix equation  e YT E|c % . ' .7c where . ' (7) Y% has any solution 7, then the formula for the TTD of the LFC T along the solutions of DS (3) is  e YT E|c % YT E|c % n 7s E|c %c  (P T E|c % ' Y| Y% This procedure is general and can be used directly for stability problem analysis as given in the following results [2]. Theorem 2. Let W be a continuous open time interval, W  Uc and let l  U. Let T G W l 6 E|c % 5 Un c T 5 EW lc and let T be a differentiable function of its arguments. Assume that > G W\$ U is an integrable function. If there is any solution 7 of the matrix equation (7), and if ;E|c % 5 W l   YT E|c % YT E|c % e n 7s E|c %c   > E|Tc (8) Y| Y% then any solution of (3) which starts in l at | ' |f , satisfies as long as it remains in l the

estimate T d|c E|o  T d|f c E|f o i T To analyze systems (3) with inputs

] |
|f

 >Er_r  (9)

we considered the following.

Theorem 3. Let W be a continuous open time interval, W Uc and let l  Uc  Assume that the function s in (3) is of the form s E|c %c  ' E|c %% n (E|c % , and that  is a positive definite symmetric matrix. Let 7 be any solution of the matrix equation . ' .7 in W lc where . ' YE %*Y%e  Let 8 ' e n Y*Y| n .7 and ' ( n .7( Let 8f be any positive definite constant matrix and let a solution of (3) passes through l Assume that > G W\$U is an integrable function, and m E|c %m Rc m E|c %m v. Let K' % 5 Uc G m%m  Rvb3? E8f  . If the T matrix ~f ' 8 n 8f >E| is negative semi definite for all values of its arguments, the estimate (9) holds with T ' %e  %c as long as E| 5 K _ lc providing %f 5 K _ l, and if K _ l is not empty. If E|c %  f, (9) holds in l  NONDIFFERENTIABLE LFC There are some classes of physical systems for which the ’’natural’’ LFCs are not very regular and may lead to nondifferentiable LFC. The extension of the LDM to the nondifferentiable LFC was provided n for systems in the normal form by Y oshizawa [18, pp.3-4]. Let (P T denote the upper right-hand Dini n derivative of an LFC T E|c %, for the system (1). Calculation of (P T requires a knowledge of the son lutions of (1). However, the formula for the calculation of the (P T , which requires no knowledge of the system solution cis given in [18]. This formula is valid under the mild assumptions that T E|c % should be continuous in the domain of interest and should locally satisfy the Lipschitz condition n with respect to % It states that (P T d|c %E|o G ' *4?<fn d 3 ET d| n c % n s E|c %o T E|c %o c where s is the function on the right side of the equality sign in (1), and where it is assumed that the input variable does not appear in s . Unfortunately, this result cannot be directly applied to systems (2) or (3). To overcome this, we introduce a new variable  5 Ur , and consider the problem in terms of E|c  instead of E|c % Let W be an open time interval of interest and let K% be an open set, K%  Ur  Denote the LFC T as T E|c  G W K% \$ Uc T 5 EW K% c T 5Lip% EW K% c where Lip% EW K%  denotes the class of functions that locally satisfy the Lipschitz condition with respect to  when E|c  5 W K % . Let (2) has solutions %E| on W . At an arbitrary moment | 5 W the values of all solutions %E| of (2) which exist on W determine the points in the system descriptor space U c which form the set V8 E|. Define also the set V% E| ' i 5 Ur G  ' C E|c % c % 5 V8 E|j c ;| 5 W . Then the following hold [2]. Theorem 4. Let W be an open continuous time interval and let the function C G W Uc \$ Ur define the variable  ' CE|c % and is such that: (a) for a given function G W Uc , the TTD ( d|c %c E|o along the solutions of the (2) is known, and is a bounded function when E|c  belongs to a compact set in W Ur ; (b) when E|c % continuously change in the set W V8 E| then E|c  continuously change in the set W V% E|. Let K % be an open set, V% E|  K %  U r c ;| 5 Wc and let and LFC T G W K% 6 E|c  \$ T E|c  5 Uc T 5 EW K% c T 5Lip% EW K%  If (2) has continuous solutions %E| on W , such that E| ' C d|c %E|o satisfies E|c E| 5 W V% E|c then the upper right-hand Dini derivative of T , along the solutions %E| at the moment | 5 Wc is given by n (P T d|c E|o ' *4 t T E 3? ET E| n c  n ( T E|c  G \$ fn    TIME DISCONTINUOUS LFC Consider the situation where  and s in (3) are time-discontinuous for a fixed %. This is motivated, for example, by the usage of time-discontinuous components like switching capacitors and resistive switches in singular electrical circuits. Our interest will be in treating the case when  and s are continuous in % for a fixed |, and piecewise continuous in | for a fixed %, while for a fixed % the matrix  is differentiable for all | where both  and s are continuous. We will develop sufficient conditions for the upper bounds on the system response. From these bounds the conditions for the

asymptotic behavior of response will be derived. We will utilize the LDM and a time-discontinuous Lyapunov function. Consider the singular model (3) of the form  d|c %E|o(%E| ' s d|c %E|oc %c s E|c % 5 Uc c  E|c % 5 Ucfc (10) c be fixed and denote by W7 the set consisting of all points of W where where | 5 W  U. Let % 5 U ? V either  or s are discontinuous in |. We will assume that the set W? ' iW7 G % 5 Uc j is such that ? W? _ E where E  W and where E is bounded, has finite number of elements | i.e. W? ' i| j. Denote by W? the set WqW? . Without loss of generality we will assume that the elements of W? are ordered, i.e. ; : , |& : | . We will assume that all discontinuities in | for  and s are of the first kind, and that  is continuously differentiable on A? for a fixed % 5 Uc . We assume that  and s are such as to ensure the existence and uniqueness of solutions of (10) which are continuous on W and continuously differentiable on W? . Such solutions will satisfy (10) for all | 5 W? . The set of such solutions we denote by 7? . Let W? E|f  ' W? _ i| 5 W G |  |f j for |f 5 W. In what follows Un ' i@ 5 U G @  fj and P? denotes the complement of the set P. We can state the following results based on [4] Theorem 5. Consider the model (10) for which V? 9' >. Let W be a continuous open time interval, W  U, and let l  Uc . Assume that there exist a function T G W l 6 E|c % :\$ T E|c % 5 Un , such that: (i) T is continuous in E|c % on W? l, T E| c % ' T E|n c % for all E| c % 5 W? l, and T has  only discontinuities in | of the first kind; (ii) YT E|c %*Y| exists and is continuous in E|c % on W? l; (iii) dYT E|c %*Y%oe exists and is continuous in E|c % on W? l; (iv) f  qE|4E%  T E|c %  kE|4E% where k G W \$ U, q G W \$ U, kE| : f, qE| : f, and where k and q are continuous functions on A? and bounded for finite | 5 W? , and 4 G l 6 % :\$ 4E% 5 Un , where 4 is continuous on W? and such that qE|4E% : f, ;E|c % 5 P? , where P ' iE|c % 5 W l G T E|c % ' fj. Let > G W :\$ U be a function continuous on W. If there is any solution 7 of the matrix equation . ' .7 , where . ' dYT E|c %*Y%oe , such that ;E|c % 5 W? l equation (8) holds, then for any solution 5 V? of (10), which starts from %f 5 l at | ' |f 5 A , the estimate 4 3 ] |  8E| \ kE|n   D T d|c E|c |f c %f o  ST E|f c %f  i T >Er_r c S ' C (11) qE|3  |f 
'& E|f

holds as long as remains in l, where  is the subscript of the first moment |& of discontinuity of  or s in W _ d|f c n4d.  Theorem 6 Let l  Uc and in addition to the other conditions of Theorem 3 let there exist : fc 5 U, such that E;|W 5 W for which EW _ d|W c |W n d 9' > we have 4 3 ] |  8E| \ kE|n   D C i T >Er_r c  Z ?c ;| 5 W _ d|W c |W n d qE|3  |W  '& E|  where  is the subscript of the first moment |& of discontinuity of  or s in d|W c |W n d . Let ? ' |f n , where is a nonnegative integer Then for any solution 5 V? of (10), which starts from . %f 5 l at | ' |f 5 A , the estimate 3 4 ] |  8E| n \ kE|   T d|c E|c |f c %f o  Z ? ST E|f c %f  i T >Er_r c ;| 5 W _ d ? c ?n? dc S ' C 3 D qE| ? holds. Moreover if W ' d@c n4dc @ 5 Uc and qE|  6 : f on W, then *4|<n" 4d E|c |f c %f o ' f , 
'& E ?
W

CONCLUSIONS This paper gives a systematic approach to the elimination of some of the crucial problems in the construction of the LFC for time-continuous descriptor systems. The results obtained can be used as a basis for further development of stability and boundedness problems of descriptor systems. REFERENCES [1] J. D. Aplevich, Implicit Linear Systems, Springer-V erlag, Berlin, 1991. [2] V B. Baji? , Lyapunov’ Direct Method in the Analysis of Singular Systems and Networks, Shades . c s Technical Publications, Hillcrest, RSA, 1992. [3] V B. Baji? , Generic concepts of system behavior and the subsidiary parametric function method, . c Sacan, Natal, RSA, 1992. [4] V B. Baji? , Lyapunov’ direct method for singular systems with time-discontinuous models, Pro. c s ceedings of the IEEE-SMC and IMACS Multiconference CESA ’96, Symposium on Modelling, Analysis and Simulation, Vol.1 of 2, pp.68-70, Lille, France, July 9-12, 1996. [5] V B. Baji? , Methodology for direct construction of Lyapunov functions, to appear in Proc. of the . c 11th International Conference on Mathematical and Computer Modelling & Scientific Computing, March 31-April 3, Washington, D.C., USA, 1997. [6] S. L. Campbell, Singular Systems of Differential Equations, Pitman, Marshfield, Mass., USA, 1980. [7] S. L. Campbell, Singular Systems of Differential Equations II, Pitman, Marshfield, Mass., USA, 1982. [8] Circuits, Systems and Signal Processing, Special Issue on Semistate Systems, Vol. 5, No. 1, 1986. [9] Circuits, Systems and Signal Processing, Special Issue: Recent Advances in Singular Systems, Vol. 8, No. 3, 1989. [10] L. Dai, Singular Control Systems, Springer-V erlag, Berlin, 1989. [11] L. T. Gruji? , Solutions to Lyapunov stability problem of sets: Nonlinear systems with differenc tiable motions, International Journal of Mathematics and Mathematical Sciences, Vol. 17, No. 1, pp. 103-112, 1994. [12] P C. Müller, On Stability of Descriptor Systems. In: M. Frik (ed.): Nonlinear Problems in . Dynamical Systems – Theory and Applications, Universit?t–GH Duisburg, pp. 116-126, 1991. [13] P C. Müller, Stability of Linear Mechanical Systems with Holonomic Constraints. Applied Me. chanics Review, Vol. 46, No. 11, Part 2, pp. S160-S164, 1993. [14] P C. Müller, On Stability of Mechanical Descriptor Systems. In: R. Bogacz, K. Popp (Eds.): . Dynamical Problems in Mechanical Systems, Proc. 3rd Polish-German, Workshop, Institute of Fundamental Technological Research, Polish Academy of Sciences, Warsaw, pp. 55-66, 1994. [15] P C. Müller, Stabilit?t mechanischer Systeme mit holonomen Bindungen. Z. Angew. Math. . Mech., Vol. 75, S93-S94, 1995. [16] P C. Müller, Stability of Nonlinear Descriptor Systems. ICIAM/GAMM 95, Z. Angew. Math. . Mech., Vol. 76, Supplement 4, pp. 9-12, 1996. [17] N. Rouche, P Habets, M. Laloy, Stability Theory by Lyapunov’ Direct Method, New Y . s ork: Sptinger-V erlag, 1977. [18] T. Y oshizawa, Stability theory by Lyapunov’ second method, The Mathematical Society of Japan, s Tokyo, 1966.