Appl. Math. and Comp. Sci, 1998, Vola, No.2, 269-286 STABILITY AND OPTIMAL CONTROL OF NONLINEAR DESCRIPTOR SYSTEMS: A SURVEY Peren ©. MOLLER* Ta recent years the analysis and synthesis of contol systems in deseiptor form have bor establishod. This general description of dyzamical rystems fe ini portant for many application i mechanics and mechatronic, in electrical and ‘ectronie engineering, sod im chemi engineering aa well ‘This consibtion presents a survey of sults a the sibs analysis and onthe optimal contol ‘Sesiga of such systems. Lyapunov's abit theory i geueralinad for desetiptor systems based on the stay Ubary with respect to 8 prt of variables. Simi- las the ealeus of warations and Pentyogin's maximum principle ae checked for a posable application in descriptor ayseans. Here, the notion of causality plays ax important coe in whether or not Pontryagis maximum prizeple ean be appli. 1. Introduction In the last decade signiticane progres in dynamic systems described by difeential- algebraic equations (DAE) has been observed. In control theory descriptor (or sing- Jat) contol systems have beon investigated (Dai, 1989) and elective codes have been veloped in numerical mathematics to simulate such systems (Brenan et al, 1989, Fuhrer, 1988; Griepentrog and Mirz, 1986; Hairer et al, 1989; Simeon, 1904). But still there are many unsclved problems related to the analysis design and simulation ‘of descriptor systems. Recent progress in parameter identifeation of finest mechanical ‘descriptor systems was outlined in (Schnidt, 1994). New methods for the analysis and ‘control design of linear mechanical desciptor systems can be found in (Schiyphaus, 1905). A thorough investigation of observers for general linear descriptor systems ‘vas conducted in (Hou, 1995). Ako the stablity behaviour of such systems has been discussed (Bajié, 1992; Miller, 1993) and frst attompts to deal with the stability problems of nonlinear mechanical descriptor systems can be found (Miller, 1994). I (Miler, 1998: 1994) more references on related investigations are given. Following (Gaer, 19960), in this contribution some new results on the stability of nonlinear ‘descriptor systems are presented Similarly, the calelus of variations and Pontryagin’s maximum principle are checked fora possible application in descriptor systems. The main problem ofthe op- ‘imal contol design consists in the causality or non-causlity of the descriptor syste ‘niversiy of Woppatal, Safety Control Baginesing, D-£2097 Wuppertal, Germany, ‘tal ular leaps Je270 P.O. Maller (Mille, 19970). Cauaaity plays an important role in whether or not Pontryagin’s ‘maximum principle can be applied. Causality and non-causality distinguish between the cases where the descriptor system is welusively governed by the conteol input or by its time-derivatives additionally. In the unusual case of non-causal systems a {quite diferent problem of optimal control design has to be considered (Miller, 1997; 19063) 2. Problem Statement Controlled time-invariant fnite-dimensional descriptor systems ean be described in semiexplicit form by = f(e.220) co) 0 = f@.22.0) @ where zi, 4 = 1,2 are n-dimensional vctors and ny +z =n. For the follow jing disession of stability and of optimal control design we assume that the vector functions are sufficiently smooth, Additiorally, the uncontrolled system may have an equilibrium point 220, 2 =0, f,(6,0,0) £210,0,0) © ‘A special class of diferential-algebrais equations (1), (2) consists of mechanical descriptor systems with halonomic constaints described by Lagrange's equation of fst Kind M(z)3 R23) = FY(2)A+ Tahu ® f(2)4+Ru @ We confine ourselves to this case, but mnholonomie constraints could be consid- ‘red too. The f-dimensional vector of displacements is denoted by = and w is the ‘dimensional veetor of control inputs. The p-dimensional vector A represents con- straint forces due to p holonomic constraints (5) (inthe ealculus of variations they faze Lagrange multpiers). The matix of inertia M(=) is assumed to be symmetric, bounded and (unifrmely) positive define. The vector function (2,2) includes Coriolis and centrifugal forces and uncontolled applied forces aswell, is the input matrix of suitable dimension, The px matrix F(2) ts the Jacobian matrix of the Constraint (3) ry = 2 o tis assumed that the constraints are ind>pendent. Ta the case of Hinear constraints this assumption is guaranteed by the concition rank, » oStability and optimal control of nonlinear Jescriptor systems: A survey an ‘The matrix 2 represents a possible constant control input matrix in the constraints ‘The form (1), (2) of the mechanical descriptor system (4), (5) is obtained by defining ==[1]: mea © m= 2h m=p a) ‘The problems of stability and optimal control of nonlinear descriptor systems consist on the one hand in the discussion of the stability ofthe equilibrium point (3) of the system (1), (2) and, on the other hand, in the control design for (1), (2) with respect to the performance eiterion 30 [soso oo where 0 is ast of bounded or unbounded control functions w(t). Again it is assumed that the function J is sufciently smooth with respect to its arguments, 3. Representations of the System ‘To study descriptor systems carefully i i advantageous to have in mind diferent forms of representation of the dynamical system (1), (2) which are related either to the generation of difeential equation fr the 2y-vortordiferentiating the algebraic ‘equations (2) as lng. as necessary orto thealimination of redundant coordinates gen- trating the state space differential equations. In both the cases the key for that isthe index & of the system (1), (2) which s roughly speaking the nunber of iferentiations ‘of the algebraic equations (2) to get the unlerlying set of ordinary differential eu tions (ODE). For simplicity, we assume that all algebraic equations have a uniform index, i. that they all have the same indes. Then we have in the case of index &h aE) ska 2) toa (Bete) a = alee) aa) were 10) =£n0) 4160+ £0) a a) ) = Bn bavi = FW, Bent Lysent) (15)me RG Miller ‘The operator (14) is defined by the operators (16) where the fist two are some Lie derivatives and the last one is only appled to. Lo by difeentatng the contol fnput fonction with respect to time, The vector fnetions LA(f,), f= Os-.4k—2 depend on 2y but not on sy. "The function Z!-"(f,) depends on @s such that the Seloed Jacobian mats is epulat and the diferent equation (18) can be derived Tei(f,) = 0 is frst integral of (13). Additionally, the functions 12(f,) depend ‘everalyon the ie derivatives ofthe ental input ws ust... © “Therefore the descriptor syste (2), (2) can also be represented by the diferent equations (1) and (13) of a maniuld decribed bythe invariants (12). ‘Additionally, in pinpe es possite wo use the ivarants (12) 0 eliminate na redundant variables and to end up with ase of my (I) ordinary differential fesquatins in the stat space. If 2s) rank) op =(k-1) 8) wh) which we may assume, the invariants L-(f;) = 0, j = 0,...,k~2 can be solved ‘with respect to some redundant variables 2, depending on some essential variables Been (esa) an with he ing [E]-e= 4 where Pisa regular permutation matric. AS regards the dimensions, we have dimzy, = (k= Ina, dimaye =m ~ (B= 1)ma 09) ‘The lst invariant Z4-¥(J) = 0 results in 822 Ba (21 tyes} = ae (tiesto yt”) 0) ‘where (17) was used in (20). By these elminations wo end with au Sieltieth oo y¥®-M) [So ]-recmn=[ feces |e ‘The related state space ODE of the DAB system (1), (2) is given by the essential part of (22): ne = Iie (rate)‘Stability and optimal control of nonlinear descriptor systems: A survey 2a In principle, the control system under consideration can be represented in three descriptions, either in the form of a descriptor system by DAE’s (I), (2) of in the form of ODE's (1), (18) on the maniflé (12), or in the form of the state space ‘ODE's (22). ‘The description (1), (2) the most natural one sine it applies usual ‘modelling procedures. The description (22) isthe most favourable one, but in many Applications with nonlinear dynamics itis almost impossible to obtain it, Finally, ‘the desripton (1), (18), (12) isa connecting ink helpfully used in analytical and numerical investigations, ‘The three types of descriptions have to be supplemented by consistent inital conditions satisfying the invariants (12), re othe rations (17), (20): { 210 = Bie (hort) fn = (trots?) au asbirary had (23) en 4, Causality The diferent representations of the contol system in the preceding section show that the system behaviour may depend not only on the control input u but also on its time-derivatives t,%,..,ul, Although the DAB description (1), (2) shows explicily only the input ‘w, there may be hidden effects related Lo time- derivatives ti,---yulasit is shown by the representations (1), (12), (18) or (22). This situation ‘ety different from the common state space discussions. Any control design method has to take care ofthis unconvenient problem, Therefore Is necessary to clarify this ‘unusual situation. Similarly to a definition by Dai (1980) for discrete-time systems, the notion of “causality” is introduced. Definition 1. The descriptor system (1), (2) is called causal If the solution [2x(0),22(0)] does not depend on w(@),...,ul*"(0) but only on u(t) ‘This definition does not have to consider u'®? as it may be expected by the notation of (18), because Z*-1(f,) = 0 is aft integral of (13) depending only on {Ws .os,u*-¥(0- Te should be mentioned that descriptor systems of index k= 1 are aiways causal. Non-causal epttame mar only appear for b> 2 [Now the question arises how causality can be checked. For linear descriptor systems characteriza by regular mateix pencils @necesiary and sucent condition is avilable if the system is epresented in its Weiersrase-Kronecker eanoneal form (Dai, 1989). But this solution does not help usin the ease of nonlinear descriptor systems (1), (2). Therefore, in (Miller, 1997; 1096a; 1007) a new criterion was Introduced by assuming suffent smoothress ofthe vector functions, Based on the description (1), (12), (13) the flowing torem can be proved24 PC. Maller ‘Theorem 1, (Miller, 19976) Given a uiform index, the descriptor system (1), (2) or (1), (18), (18) is causal if a (tts) In this case the invariants (12) are of the form Da) = Uy (Fa) Us) ‘They do not depend on the time-derivativs of the control input. Accordingly all the Functions in (20)-(24) do not depend on the time-drivatives ofthe contral input, but only on u(t) yk=2 (5) ess) (26) wt E fagaleitaa) =0 (en ‘To illustrate this result, a linear descriptor system in sem-explicit form is con- sidered: fy = Ann + Ann + Bye (28) 0 Anay + Anta + Box (29) [eis assumed that the index & holds uniformly in all algebraic equations (29), Ary regular 0) An =0, An Aa regular an An =0, An Ayan =0, 7=0.08-3, AnAiz? Ave ropae (oa) “Then the descriptor system (28), (29) i casa fan only 1 alnays (3) =2: Boo B29: B= 0, AnALBr=0, j= 0..1k-3 (35) Causality will be later an essential point to settle whether Pontryagin’s maximum principle can be applied as usual or not 5. Stability Sal, General Theory For standard state-space systems the notin of stability” i well defined (even if there is not only one definition of stability but a great number of definitions related to‘Stability and optimal control of nonlinear descriptor systems: A survey 5 ferent requirements of applications) anda elated stability theory is well-established (Ffahn, 1967). For example, let ws recall cly the notions of stability in the sense of Lyapunov, asymptotic stability, absolute stability, andthe well known stability theory of Lyapunov. ‘When compared with the large amoustof researc related to stability problems of state-space systems almost nothing har been investigated regarding the stability Dehaviour of descriptor systems. Fist general attempls were presented by Baie (Bajie, 1986; 1987; 1988a; 1988b; 1992; Bajié and Milie, 1987; Baie et el, 1980) and by a fow other authors (Doleal, 1987; Griepentrog and Maz, 1086; Hill and ‘Marcel, 1990). In the case of linear timeinvariant desriptor systems stability was defined by th cigenvalus of a related matrix pencil (Dal, 1989). Some recent results for nonlinear deseriptar systems were reported in. (Miller, 1993; 1904; 19960). One essential dificult of a suitable stability defniton isthe problem in which (eub-]space stability has tobe defined. To avoid impulte solutions of (1, (2), the initial conditions 221(to) = yo, @a{to) = #20 must be consistent with the invediants (23). By these algebraic constraints consistent solutions belong to a consistent manifold which is a subspace ofthe generalized state space (x, 22). By restricting the stability problem to the consistent manifold, the stability f an equilbrinm point (8) can be defined in the sense of Lyapunov confining the peturbed motion to the consistent manifold, Similarly, asymptote stability is defined with respect to this manifold, In tis rstricted sense, stability result for linear time-invariant descriptor sys- tems were reported in (Miller, 1993) based on the discussion of a generalized Lya- ‘puno matrix equation generalizing the rests of (Owens and Debeljkovie, 1985). For linear mechanical descriptor systems the well-known theorem of Thomson, Tait and ‘Chetaev (Miller, 1977) was generalized. Additionally, in (Miller, 194) the results of (Maller, 1993) were brought forward to nonlinear mechanical descriptor systems using ‘the Hamiltonian as a Lyapunoy function. In (Miller, 1996b) an extension for general nonlinear descriptor systems was considered. Summarizing these results, we ean de- sive the Lyapunov-like stability theorem, In the following, we discuss the control free stability problom for u = 0 using the notation of (26). Introducing 21+ Dadsa(210) (36) such that dy = f(2,02,0) = 7, (#y,%,0) and assuming a Lyapunov function fe) = wlan) +S BFF 0)f (01.0) on E ih te property PT lan 2 0) = EF ,0,0)4 5 FF en OV(@) (8) for suitable functions (23), we have the following result26 PC. Miller ‘Theorem 2. (Mille, 19060) If thre ex constants ai, = Os.-yh~B, 8:20, Gresk8, and. function vale) swe that (21) >0 forall 40, ¥(0)=0 (09 condion (38) sash, ond ste) = OLB 7 C00) 425 HPO l00) $0 (8) forall; #0 ina neighbourhood of x =O, then the epuibrium point (3) is stable, 1f, additionally, Krasovski’s condition (Burbasin and Krasousii, 1952) we) s0 7 ale is filled, then the equilibrium point (8) is asymptotically stable ‘A characteristic of this result is that the Lyapunov function and its time- derivative depend only on a part of the variables (21) and do not depend on 2 ‘This fact is stimulating to apply the thenry of stability with respect to a part of ‘variables (Miller, 1982; Oriraner and Runiantsey, 1972) to the stability problem of descriptor systems. Comparing or problem with the dainition andthe results of par- tial stability, essentially we have ta look for stability with expect tothe vasables while the behaviour of 21», 22 is not of explicit interest. But the practical problem Js that usually 2j¢ is not explicitly known. Therefore, in applications the variables 22, which embed ye are considered. Nevertheless, the properties of deiniteness will be essentially related to aye by the following statement. a) Lemma 1. If thee exist tuo functions w(2s) and 1 Fa(2150) lle sllFaa-a6@2,0) 11] > 0 (a) such that (ns) = tons) +5] Foy (e150) lhl Fas afers0) |) > 0 (49) fs postive definite, then corner EDee [actin] tuhere (ae) 1 positive definite with reget t9 216 uelen) >0 (44) ‘Therefore, by using the invariants (20) as frst integrals, the construction of sultable Lyapunov functions i simplied and definiteness propertos can essentially be directed to consistent solution manfild, [Now we ae able to transfer sme of tae retults of partial stability (Miller, 1982; Oziraner and Rumiantsev, 1972) to the stability problem of noalinear descriptor sy tems.Stability and optimal control of nonlinear descriptor systems: A survey 7 ‘Thoorem 3. (Miller, 19965) If thee exist functions vp and of Lemma I such that atan2n) = SE f.(es.22 9) 50 oo) then the equiv point (9) is tae wih respect to tye = 0 I atonal, fest) $ wlan) <0 6) olds for a certain positive definite functicn wa(a), then aie = 0 is asymptotically stable, If (46) 18 replaced by (e122) wales) $0 « for a positive definite function w(x), then asymptotic stability is guarantor! if ‘Krasousi’s condition We) 20 if my 20 8) 4s eddtonally met. ‘Theorem 3 shows how the thoory of stabil’ with respect to a part of variables can successfully be applied to the stability problem of descriptor systems. 5.2. Thomson-Tail-Chetaev Theorem Tm the case of the linear time-invariant holonomic mechanical system M3(t) + (D+G)3(t) + Ket) = FTA) (9) F(t) =0, (60) where the matrices D = D?, G = -GT, K = KT are respectively related to damping, gyroscopic and stifiaess forces, the result of Theorem 2 can be simplified, Without the constrains (60) the stability behaviour of the unconstrained system (49) (22 0) was investigated in detail in (Miller, 1977). A vory famous stability result 's the thoorem of Thomson and Tait whic has been proved by Chetaev by applying Lyapunov's theory. A generalization with respect to pervasive damping instead of complete damping was presented in (Miller, 1977). Now itis possible o generalize ‘these stability results to constrained mechanical systems (49), (50), 4. (Miller, 1999; 1904). "Theorem 4. (Miller, 1993) Ifthe system (49), (0) satisfies the conditions M=M">0, DiqFTF>0, K+aF'P>O oy for some wasaters ay 20 and a 2 0, then the mechanical system ts asymptotically stable ifthe condition Me s(D4G)a+K —FT zanke F o |=ser @ U-F Pps 0 fs met for all 8€ ©.28 PC. Maller "The condition (52) is an explicit form of Krasovski’s requirement (41). ‘The matrix F represents the Moore-Penroseinverse matrix of F "Theorem d shows explicitly, by the constants a and a, the stabilization eect of the holonomic constraints. ‘The unconstrained system (F = 0) may be unstable bbut the constraints may lead to a stable constrained system. ‘Then the instabilities tare located inthe space of the constrained modes. ‘The theorem represents & proper nation of the well-known stability theorem of Thomson, it, and Chetaev including also the effect of pervasive damping due to (52). ‘The results of Theorems 2 and 8 can also be applied to nonlinear constrained mechanical systems (4), (5) by using the Hamiltonian as a Lyapunov function. The Stability dheorems were eported in (Mille, 1994), 6. Optimal Control 6.1, General Theory ‘The purpose ofthis section is to derive conditions for the design of optimal control of ‘descriptor systems, If we look forthe system desertion (1), (12), (18) instead of (1), (@), then an obvious problem appears. Generally, the system behaviour may depend ot only on the control input a, but also on its time-derivatives ti t,--.,ul®), This is an inconvenient problem statement, so we have defined the property of causality in Section 4 to distinguish between the cases where time-derivatives appear or not ‘Therefore, in the following we have to distinguish between causal and non-causal descriptor systems (1), (2). In the fist case we can apply Pontryagin’s maximum principle without any problem whereas inthe other we have to discuss a more com- plicated optimization problem. Accordingto the discusions in (Mille, 19973; 1906a) ‘the following results are presented for the aptimal control design ofthe desriptorsys- team (1), (2) with expect to the performance criterion (11), ‘Theorem 5. (Miller, 1996a; 1997) For causal descriptor systems (1), (2) the Pon: ‘tryagin mazimam principle con be applicé as usual. Necessarily thee are nontrivial adjoint vectors Az and z such that with the Hamittonian H=Mfit+Mfe-fo (53) the adjoint diferentil-lgebraie epations an = (54) (63) ‘are satisfied and the optimal control fulfils mae = Hoge (56)‘Stability and optimal control of nonlinear descriptor systems: A survey 219 or simplicity, the boundary conditions ofthe adjoin variables are not considered ‘explicitly, but they have to be appropriately chosen. ‘The optimal control problem looks quite different for non-causal descriptor sys- tems because of the infuence ofthe time-derivatives of the control input. Actually, the following procedure is recommended. Having regard to th...,u"™") the rep resentation (1), (12), (13) of the descriptor system is preferred. By introducing the &au Gat, 6 Gaul, vault (67) sn extended tate vetor a=[ta gd ~ al" re) fs dained which satin the ordinary dient equation Silaiyz2,€)) Faery 2, 68a). £9) te & (59) & and the algebraic oqutions Ufa) & faj(@rs Barby Say- 184) =, FO. R-T (60) ‘The original control constraints w € U appear now as state constraints € ¢ U. ‘Additionally, the question arses whether the control problem is stated properly. For ‘ reasonable problem statement, additions constraints on €j = ty.--.€4 = uw!) and particularly on v = ul!) may be elicent. Nom, the corresponding Hamilton function is introduced: MANDAN hve t ly eeLieeete 1) or the extended system (59) and the Hamilton fonction (61) the procedures of the calculus of variations or Pontryagin’s maximum principle can be applied leading to ‘an optimal control design of non-causal desriptor ystems. But agai the constraints €U of the extended states must be talon into account, which highly complicates the optimization procedue. Note that the algebraic constraints (@) do not have to be included in (61) ex plictly because they are taken into account by the second set of diferentil equa tons, Le. by (13). But the inital conditions of the system (89) have to be chasen consistently, i. they have to satisfy the algebraic constraints (60) according to the roquirements (23)280 PO. Miller 6.2. Mlustrative Examples In the following, three small academic examples are discussed to illustrate the ques- tions of causality and optimization, ‘The frst two examples ae essentially related to the distinction of causal and non-cavsal mechanical descriptor systems 6.2.1. Causal Mechanical Descriptor System ‘The aystem considered consists of two spring-mass-oscllators where the masses are connected by a rigid bar (Fig. 1). Additionally, the fst osilator is controlled. Lagrange’ equations of frst kind (when neglecting the static forces due to the equi- librium postion =.) and regazding only the dynamic bebaviout) are as follows: mij tan = A+ (2) sms +2 = -2 (6) asn=o (69) Fig. 1. Two spring-mas oscidators connected by arg bar ‘The description (1), (2) is obtained by aa ft &], 2=2] (65) (66) 1 follows from (22) and (38) that the system has index 3 and that it is causal ‘There is a more general result behind. Holonomic mechanical descriptor sys tems (4), (5) have uniform index k = 3 and they ate causal iff = 0, ie if the‘Stability and optimal cntrol of nonlinear descriptor systems: A survey 281 constraints (5) are not controlled. Ifa least one constraint is conteolled, ie it de- pends explicitly on u, then the system (4), (5) is non causal ‘The optimal control design is performed according to Theorem 8. In (Miller, 1997a) the linar-quadratic optimal regulator problem was discussed. In (Mille 1996a) the sime-optimal contro probler: was solved. In both the cases well-known, results were obtained, but based on a DAB-approach and not on a classical state space approach which arises ffom the ordinary differential equation (om ema) +(e, Fender (on) where 2, is chosen asa generalized coondinate 6.2.2. Non-Causal Mechanical Deseriptor System ‘The second example differs from the fist one by Introducing the cantol ta the con- straint such that we have the typical problem of «controled mechanism mi tan =) (68) mis bean = ©) acatu=0 () ‘The description (1), (2) obtained from (65) and as va fzannse m S,,-2 ‘ma? na ‘The condition (85) is violated (By # 0) and therefore this system is non-causal. This fact can be shown explicitly if we write down Lagrango’s equation of second Kind with respect to the generalized coordinate 2, and the equation determining the constraint force (im, +m) (a ten = ou mai 7 an mime (2-2) 2 -3} as ‘The representation (72) shows that introduction of extended state varlables (67) is necessary to handle propery the optimization procedure. In the same way the solution 222 = [\] depend on @ such thatthe system (68)-(70) ie non-causal. In (Miller, 1997a; 199d) some discussions ofthe optimal consol design are presented.232 PO Miller 6.2.8. Non-Causal Academie Example In order to illustrate the difcultis and the surprising results of the optimal control design of non-eausal descriptor systems, ve consider the following simple academic ‘example: yaa tbe 4) yom thu (73) ot bat (76) "The index ig k= 3 and the system is causal for by = 0, by = 0. ‘The underlying covdinaty difeental equations on ® maniflé are written down sezoeding to (2), (12), (13): fy = hy ail = ™ onthe (78) yeaa the (79) with the invariants my bby t ball bs (eo) tat bavt by = 0 oo) aa tbau=0 ©) ‘The invariants sepresent the explicit solution of the system (74)-(76). According +0 (80), (81) the solutions of 2, 22 depend on cand & ifthe eoefiients bs, bs do not vanish ‘Looking for an optimal control with -espect to the quadratic peeformance crite 3 (oat 03+ 92k +r) at 9 minima @) where w is unbounded, and r > 0; 4) 0) to the optimal feedback control, the caleuls of variations lend [lear bseras + onze + boas2a — hams + boa] (84) 1 FF abibs While a simple proportional feedback y braves (5)Stability and optimal control of nonlinear descriptor systems: A survey 2s Js designed forthe causal system, forthe non-causal one a proportional and a (twice) ‘ime-derivative feedback (84) aries. The rated closed-loop contol system isa stan dad system of fourth order (bs # 0, a >): avs ~ baboons ~ blbsen ~ bian)eu + babotota + (r+ dubata + Ban) b cs Arbitrary values of 239, #10, 209, 49 canbe chosen a8 initial conditions It is easy to prove that the charactarstc polynomial of the closed-loop sys- ‘tem (86) is b-quadratic such that the system is unstable, This is a consequence of an regulary stated optinzation problem, The performance criterion (83) considers ‘the square ofthe control but itis not related to the time derivatives ofthe input, To regularize the problem, the weighting of has to be introduced corresponding to the solution (80). Then a stable optimal feedtack contol can be expected ‘According to (67) the extended variables ont Bn (so) 4 e=t, (87) axe intron. The eiteron (89) med ita [fF oat eit okt nat ena +i) (9 Since the solution (80}-(82) ofthe desriptor system (74)-(76) i known, the resulting optimization problem reads as follows, For the dynaimicl system G=@ G0 (69) the performance criterion T= BL {oles + as bl? + oats +b tablet tthe trv }ae o«) has to be minimized with respect to v. This isa classical linear-quadratic optimal regulator problem which can be solved by the usual Riceati approach, After some lengthy ealeuations we have ve aap (lmbte + Pil6 + abd + Pade} oy vith Fan = ~anbiby + ylebatalQoR + ra) + abi ibid + bars) (92) Pa = white + lat ath tr 2Paaia or) (03)284 _ PC. Miler Having in mind (87), the result (91) represents an asymptotically stable diferetial ‘equation of second order with respect to az Therefore this control lads to an asymp- totically stable behaviour of the descriptor system (F4)-(76) for consistent initial conditions 1 should be mentioned that the solution of the linear quadratic optimal con- trol problem can be discussed by intodcing a generalized Riceati equation, But this only makes sense in the ease of caus systems, because the feedback control is then a proportional feedback ofthe desciptorvariables. But even in this case the teneralized Ricati equation has to be handled very carefully. A detailed discussion ‘vas presented in (Hou, 1995: Schipphans, 1996) also including numerical aspects, For non-causal systems the Riczati equation approach des nat meet the problem, because the typical time-derivative feedback partis not represented by this approach directly. Only when the extension (57)-(C) is performed, there may exis a certain chance to define a Ricca equation appreach. But there is stil a problem, because the extended state space system (59) is aot completely controllable because of the Invariants (60). Therefore this approach is a completely open problem which has to be discussed in future. Summarizing, this simple example ckarly shows the dificultis of the optimal control design of non-causal descriptor sysiems. To solve such problems eareflly, the extended state vector (58) according tothe time derivatives (57) has to be introduced ‘hut also the performance erterion has to be re-considered by introducing penalty functions of timederivatives of the contra! input to regularize the optimal contral problem, 7. Conclusions ‘The description of nonlinear dynamical systems by the so-called deseriptor systems becomes more and more popular. Thereare tools for the analysis and control de- sgn of tuch systems are needed. In this contribution some results on the stability Analysis and optimal control design have been reported. The stability analysis is e=- tentally based on Lyapunov's stability theory with respect to apart of the variables, In the case of linear mechanical desriptcr systems, the famous result of Thomson, ‘Trt, and Chetaev has been generalized. For the optimal control design it was nec: ‘essary to introduce the notion of causality and to distinguish between causal and ‘on-causal descriptor systems. In the fist case the calculus of variations and Pon- ‘ryagin's maximom principle can be applied as usual, but in the second case one has to take care and itis recommended to solw the problem on the bass of the so-called underlying ordinary differential equation ana manifold. Two examples of constrained \-mass oscillators illustrate the occurence of eausal and non-caisal mechanical descriptor systems. The third academic example shows the severe difculties of the optimal control design of non-causal descriptor systems. Surprising results appeared demonstrating that many problems are still open. Tn particular, efficent numerical ‘methods are missing, especially for non-ccusal systems.Stability and optimal control ef nonlinear descriptor systems: A survey 285, References Bayle VB. (1986): Partial exponential stabi of semi-state syitoms. — Int. J. Contr, Vol, No., pp 1389-1804, Bajie VB. (1987): yqpamoe function eandilats for semstte systems, — Int. J. Cont, Vlas, Noo, pp 2171-2181 Bajié V.B. (19883): Byuation of prtured motions and stability of state and sems-state systems, — In. J. Conte, Vola, No pp 1810-1860, Baie VIB. (19856)> Generic staibly and loundedness of semistte system — IMA J. ‘Mah. Contr. Inform, Vol, pp. 109-11. Basie V.B. (1992): Lyapunor’s Diet Met in the Analysis of Singular Sytems ond ‘Networks, — Hilerest, Natal, RSA: Shades Techical Publications Biie ViB. and Mile MM. (1987): Batendel stailly of maton of semistate systems. — Int. J. Conte, Vol46, No, pp 2189-2197 Bajie VB, Zivkové MM, and Zivkovit, JS, (1999): Nomncariies in semistte systems ‘and cponentia stability of response. — Int. J. Conte, Val, No, pp1559-1861 Barbas B.A. and Keasovkli NI. (1052) On the stability of motion in the large. — Dallady Akad Nauk SSSR, Vol, pp 483-856, (in Rosia). Brenan K.E., Campbell $1, and Petaold L. (1980): Numerical Solution of Initial Ve ‘Probiems in DierentatAlgbrase Equctions. — New York: North-Holland, ai L. (1080): Singular Control Systems, —LNCIS, Vol 18, erlin-Heidelberg: Springer Verlag DolezalV. (1987): Some practical stability enitesa for smistte equations. — Circuit Syst Sigal Process, Val, No, pp 335-348 uluer ©, (1988): Differentabalghmische Gleichungseyteme én mechanischen Mehrbir. persstemen: Thor, numeriihe Anatse und Anwendungen, Ph.D. Thess, Teh sche Universite Munchen, Gropontrog B, and Matz R (1086): Difeesil Algebraic Bgutions and Their Numerical ‘reviment.— Leipaig: Teubner Verag Hahn W. (1967): Stability of Motion. — Besin-Heidelborg: Springe-Verlag Hainer E, Labick C. and Rocha M, (189): The numerical soltion of derental- algebraic ‘systems by Range-utt methods. — LNM, VoL1408, Berin-Heidlberg Springer Verlag Wi D3. and Marels LNCY. (1900). Stbitiy theory for dierentab-algeraie sets with application to power systems, Ins Robust Costol of Linear Systems and. Noa mr Control (MA Kaadhont, TH Vm Schoppen, ACM. Rawy Ede) — Bal Birkhicwr, pp.407-405, Hou M. (1985): Descriptor systems: Observers and fault diagnosis. — Fortschr-Bet. VD, Foie 8, No.4, Disieldaré VDL-Vereg Malle .C, (1977): Stbiitt und Matrien. — Resin Heidelberg: Springer-Verlag, ‘Miller P.C. (1982): Zum Problem der portillenanympatschon Stabe, — Festschrift Prof. Dr. rr. nat. Kurt Magnus, dtu B Mr Mecha, Techsische Universitit Manchen, pp 297-268,286 PC. Miller Maller PC. (1995): Stability of linear mechaicl systems with holonomic constraints. — ‘Appl Meeb. Rev, VoL46, No.l, part 2, pp. $i00-Si64 Maller P.C, (194): On etait of mecheie’ descriptor systems, In: Dynamical Problems in Mechanical Systems (R. Bogacs, K.Pepp, Eds). Proc ed Poli German Wore shop, Institate of Fundamental Technoligcal Reach, Polch Academy of Sciences, Ware, pp 55-66 Milles P.C. (1996): Stability and optimal contra of constraint nonlinear mechanical sy ‘ome. — Pron. Int. Syemp_ Analysis and Synehoof Nonlinear Dynatalal Systm a ‘Mechanics, Riga Techneal Univers, pp 102-18. Mller PC. (10966): Stability of nonlinear deeriptr stems, —Z. Angew. Math. Mech, ‘Vl-6, Supplement 4, pp 2-12. (Maller PC. (19073): Optimal contra of mactnicl deseriporaystems, — Proc: TOTAM Symp. Interaction Botwoon Dyaamits and Conta a Advanced Mechanical Systems, Dordreeht: Kluwer Academic Pub, pp 47-258. Milles PC (907%): Kausale und nichtausle Destriploreyteme, — 2, Angew Math Mech, VoL77, Supplement 1, pp 5231-3252 (Owens DH. and Debeljevié DL. (1985): Consistency and Lyapunos sabity of linear eseniptor systems: a general analysis. IMA J. Math. Cnt. Inform, Vol, pp.139~ 1, Oriraner A'S. and Rumiantsey V.V. (972): The method of Lyapunov functions in the ‘tailty problem for motion with respect to «part of he variables, ~~ App. Math, ‘Mech, Val 36, pp.St-962, Schmidt. (904): Payomaterschéteung bei Melrhérpersystomen mit Zeangsbedingungen “Fortschr Ber. VDI, Reike 8, No.3, Disseldoré VDL. Vesag Salpphaus R. (1995): Reelungstecinsehe Analyse und Synthese von Merkorpersystemen 'n Desriptorform. — Frtshr-Ber. VI, Rebe 8, NosT8, Disseldrt: VDI-Verlag. Sincon B. (1901): Numerisehe Jntgration machanischer Medskirpersysteme:_ Projisie ‘rende Desriptorfarmen, Aiorithmen und Rechenpragrammne. ~ Pisce: Ber, VDI, Reihe 20, No180, Diseldost, VDI Vera