Computation of Structural Decomposition for Linear Singular Systems MINGHUA HE and BEN M. CHEN Department of Electrical and Computer Engineering The National University of Singapore 10 Kent Ridge Crescents, Singapore 117576 REPUBLIC OF SINGAPORE Abstract: We present in this paper computation algorithms for the structural decomposition of general linear multivariable singular systems. Such kind of decomposition has a distinct feature of capturing and displaying all the structural properties, such as the finite and infinite zero structures, invertibility structures and redundant dynamics of the given system. The computation will make it a powerful and convenient tool in solving control problems for singular systems. Keywords: Computation; MATLAB programs; singular systems; structure decomposition; structure properties. 1. Introduction Consider the following linear singular system Ex Ax Bu : y Cx Du x n , y p , u m , (1) where rank ( E ) n . It is also alternatively called a descriptor system, an implicit system or a generalized linear system in literature. Linear singular system has attracted researchers for more than three decades since many real systems, such as electrical power systems, can be naturally modeled as singular systems. Among the issues discussed for linear singular system [7], system structure and equivalence is an essential topic. Campbell [1] presented an effective structural decomposition method and got the corresponding equivalent system, while Verghese et al. [11] defined a strong system equivalence using a trivial augmentation and deflation technique. On the other hand, structural invariants also received much attention in literature. Further, Misra et al. [10] and Liu et al. [9] have presented their algorithms to compute the invariant structural indices of singular systems. More recently, He and Chen [5] and He et al [6] have developed a structural decomposition method for single-input single-output and multivariable linear singular systems respectively. Such a structural decomposition can not only give the invariant structural indices but also explicitly display the structural properties, such as the finite and infinite zero dynamics, invertibility structures and redundant dynamics of the given systems. And it is expected to be a powerful tool in solving system and control problems as its counterpart in nonsingular linear system [3]. This paper focuses on the computation algorithms of the structural decomposition. First, to make this paper more self-contained, we briefly describe the structural decomposition theorem in Section 2. And in Section 3, the main MATLAB computation programs are presented while several numerical examples will be included in Section 4. Finally, a conclusion will be drawn in Section 5. 2. Structural Decomposition Theorem and Its Properties We first summarize the structural decomposition of general multivariable singular systems in compact matrix form. And its essential properties will also be given in brief in this section. Theorem 2.1 Consider the general multivariable linear singular system in (1). Then, there exist nonsingular state and output transformations s nn and o p p , and a nonsingular transformation e nn , as well as an mm input transformation i (s) , whose inverse has all its elements being some polynomials of s (i.e., its inverse contains various differentiation operators), which together give a structural decomposition of and display explicitly its structural properties. This structural decomposition can be described in the following equation form, xz xe y0 u0 (2) x x s ~ x s a , y o ~ y yb , u i ( s )u~ uc , xb y u x d d c x d and C 0z 0 C 0a C 0b C 0c C 0d ~ ~ ~ 1 C o Cs C v C k C dz 0 0 0 0 Cd Ck , 0 C bz 0 0 C b 0 xd1 yd1 ud1 xd 2 yd 2 ud 2 xd , y , u , d d x dm ydm udm d d d I m0 ~ ~ 1 D o Di ( s) D v Dk ( s) 0 0 xz 0, xe Be0u0 Becuc Bedud , xa Aaa xa B0a y0 Lad yd Lab yb , xb Abb xb B0b y0 Lbd yd , yb Cb xb , where (3) B0 0 0 B0a ~ ~ ~ ~ Ak ~ x Bk ( s) u~ 0, Ck ~ x Dk ( s) u~ 0. and for each i 1, 2, , md , x di Aqi x di Li 0 y 0 Lid y d Bqi [u di md M ia xa M ib x b M ic x c M ij xdj ], y d C d xd . The structural decomposition can also be expressed in the following compact form. 0 0 0 0 0 0 0 0 0 I na 0 0 0 0 I nb 0 0 0 0 I nc 0 0 0 0 0 0 0 , 0 0 I nd 0 0 0 0 0 0 Aaa 0 Bc M ca Bd M da 0 0 Property 2.1 The given system in (1) is stabilizable if and only if Acon , Bcon is stabilizable, and it is detectable if and only if Aobs , Bobs is detectable, where A Acon : aa 0 Lab Cb B , Bcon : 0a Abb B0b Lad , Lbd (8) and A Aobs : aa Bc M ca ~ ~ A e1 As Av B0 C 0 Ak 0 I ne (7) The equation form of this theorem and detail proof can be found in He, Chen and Lin [6]. Here, we briefly introduce the essential properties of this structural decomposition. (4) j 1 I nz 0 0 0 0 0 (6) and y0 C0a xa C0b xb C0c xc C0d xd u0 J nz E ez E az ~ E e1 Es E v E bz E cz E dz B0b B0c B0d , C 0 0 0 C 0a C 0b C 0c C 0d , xc Acc xc B0c y0 Lcd yd Lcb yb Bc M ca xa Bcuc y di C qi xi , 0 0 ~ 0 0 Dk ( s), 0 0 0 C , Bcon : 0a Acc M da C0c M dc . (9) 0 0 Lab C b 0 Abb 0 Lcb C b Acc Bd M db Bd M dc 0 0 Lad C d ~ B0 C 0 Ak , Lbd C d Lcd C d Add 0 0 0 B Bde Bce 0e 0 ~ B0a 0 ~ ~ 1 B e Bi ( s ) B v B k ( s ) B k ( s ), 0 B0b 0 B0c 0 Bc B0d Bd 0 Property 2.2 The invariant zeros of the given system are the eigenvalues of Aaa . The normal rank of is equal to m0 md . Here md is the dimension of u d . Property 2.3 The given system has m0 infinite zero of order 0. And its infinite zero structure (of order greater than 0) is given by (10) S () q1 , q2 , , qmd , that is, for each i 1, 2, , md , has an infinite zero of order qi , respectively. (5) Property 2.4 The given system is right invertible if and only if x b and hence y b are non- existent, is left invertible if and only if xc and hence uc are non-existent, and is invertible if and only if both x b and xc are non-existent. The properties show that our structural decomposition can explicitly display the structure properties of the given singular system, and hence it is expected to be a powerful tool in solving singular system and control problems. 3. MATLAB Computation Programs for the Structural Decomposition singular. The decomposition can be characterized as the following transformations, 0 A1 0 I PEQ n1 , PAQ , 0 I n 0 N 2 B PB 1 , CQ C1 C2 , B2 where N is a nilpotent matrix. (11) ctr_cf.m The function transform a matrix pair A, B into its control canonical form as follows, As mentioned before, a detailed constructive decomposition algorithm will not be given here due to the limit pages. And in this section, we will give brief descriptions of the main functions for the computation. The computation programs introduced are all in MATLAB codes. A Acc B (12) T 1 AT c , T 1 B c , 0 0 Ac where Ac , Bc is completely controllable while Ac , 0 is totally uncontrollable. SD.m bdc_cf.m This is the main function, that is, structural decomposition function for general linear singular systems. The function transforms the given singular system ( E , A, B, C , D ) into its structural This function decomposes a complete controllable pair Ac , Bc into a special block controllability canonical form [2], in which every submatrix block corresponds to a distinct input channel. The decomposition process can be described as follows, ~ ~ ~ ~ ~ decomposition form ( E , A, B , C , D ) , which can explicitly display all the structural properties, such as the finite and infinite zero structures, invertibility structures and even redundant dynamics of the given system. sys_hat.m This function separates two kinds of redundant states from the original system. One kind of redundant states x z are static and identical zero all the time, whereas the other redundant states xe are linear combination of appropriate order of system input's derivatives. Such states are associated with the so called impulse modes, which are introduced by the derivatives of the system input. pre_decom.m This one is to perform a fast-slow decomposition (see e.g., [4] for details) for the given singular system. With two constant transform matrices P and Q , it transforms the given singular system into two subsystems, one is nonsingular and the other is J1 0 1 R Ac R 0 0 B1 0 R 1 Bc 0 0 0 0 J2 0 , 0 0 0 Jk B12 B1k B1l B2 B2 k B2 l 0 0 0 Bk Bkl (13) where J i , i 1, 2, , k are Jordan blocks with zero eigenvalue and 0 0 Bi , 1 kronecker.m Bij . 0 (14) The function transform the given system's system matrix P (s ) to its Kronecker Canonical Form with two constant transform matrices M and N , P~ ( s ) M P ( s ) N A sE B M N D C diag sI J f I sJ L1 L p LT1 LTq (15) Here every block of the diagonal entries in P~ ( s ) 1 0 E 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0, A 0 0 0 1 1 0 1 1 0 B 1 , C 2 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 1 2 1 1, D 0, (16) is associated the distinct structure index. SCB.m This is the function of structural decomposition for linear nonsingular system. The function was developed by Lin and Chen [8], and it decomposes a given linear system A, B, C, D and explicitly displays its structural properties. The function SD.m is its natural extension to linear singular systems. r_jordan.m This function transforms a real matrix H to its Jordan canonical form. The functions introduced here are only some main procedures, and there are still many other functions needed in our computation. But due to the limit of page, we can not introduce every function here. However, this omission will not affect our illustration of computation process in the following section. 4. Some Numerical Examples To illustrate the computation of our structural decomposition algorithm, we present in this section two numerical examples, one is of single-input and single-out linear singular system while the other is of multi-input and multi-output case. Let us first look at the following single-input and single-output system, the statement Ge , Gs , Gi , Go , E v , Av , Bv , Cv , Dv , nz , ne , na , nb , nc , nd SD E , A, B, C , D , returns 0 0 G e 0 1 0 0 0 Gs 0 1 0 Go 1 , 1 0.3333 0.7071 0 0.6667 0 1 0 0 0 0 0 0 0.6667 0.7071 0 0.3333 0.7071 1 0 0 0 0.6667 0.7071 0 0 0 0 0.6667 0 Gi ( s ) 0 1 0 , 0 0 0 0 0 0 1 , 1 , s2 (17) (18) (19) and 0 0 0 1 0 0 Ev 6 0 1 4 . 2426 0 0 5 0 0 0 1 0 0 1 0 Av 0 0 1 0 0 0 0 0 0.6667 0 0 0 0 0 0 , Cv 1 1 0 0 1 0 0 0 0 0 3 , Bv 0 1.4142 0 2 0 0 0 1, (20) 0 1 0 , 0 1 Dv 0, nz 1, ne 1, na 2 , nb 0 , nc 0 , nd 1. And finally this decomposition result can be verified by the following operation. The statement Pv , M , N Kronecker E, A, B, C, D returns 1 s 0 0 s 0 0 Pv 0 0 0 0 0 0 0 0 0 0 0 0 1 s 0 0 1 0 0 0 1 0 0 0 0 0 0 . 0 s 1 (21) From Pv , the Kronecker Canonical Form of the given system, we can see clearly that the structure indices are the same as our computation results. Now we look at the following multi-input multioutput linear singular system, 1 0 0 0 0 0 E 0 0 0 1 0 0 0 1 1 1 1 1 1 0 B0 0 1 0 1 2 1 0 0 0 0 1 0 0 0 0 , A I7 0 0 1 1 1 1 0 1 0 1 0 0 1 1 1 0 1 0 0 1 0 , 1 , D 0 0 0 1 1 1 1 0 0 0 0 0 1 C . 0 1 1 0 2 1 2 0 1 0 0 0 0 0 0 0 1 0 1 (22) Ge , Gs , Gi , Go , E v , Av , Bv , Cv , Dv , nz , ne , na , nb , nc , nd SD E , A, B, C , D . 0 0 1 G e 0 0 1 1 1 1 0 1 1 0 0 0 0.7746 0 1 0.2582 0.3162 0 0 0 0 0.5164 0.3162 0 0.2582 0.3162 2 0.2582 0.3162 0 0.2582 0.3162 0 0 0 1 0 0 0 0 0.6 0 0.4 , 0.4 0.6 0.4 0 0 1 0 0 0 1 0 0 0 1 0 0 0 0.7746 0 0 0 0.2582 0.3162 0.2582 0.3162 0.5164 0.3162 0.2582 0.3162 0.5164 0.3162 0 0 0 0 1 0 1 0 0 0.6 0.6 0.4 0.4 0.4 (23) s 1 s2 s 1 -1 Gi ( s ) s2 1 1 , 0.8944 s 0.4472 0.8944 1.3416 1 0 Go . 0 1 (24) And Then its structural decomposition form is in the results of the following statement, And the results is 0 0 1 Gs 0 0 1 1 0 0 0 E v 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 Av 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 , 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1.4142 , 1.5215 0.3333 1.8648 0.7172 0.7071 0 0 1 0.7071 0.3873 0.8333 0.3333 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0.2 0.8944 1 0 0 Bv 0 0 0 , 0 0 1.0541 0 0 1 0 1 0 0 0 0 Dv , 0 0 0 0 0 0 0 0 1 0 Cv . 0 0 0 0 0 0 1 (25) And the corresponding structure indices are nz 1, ne 2 , na 1, nb 0 , nc 1, nd 2 . (26) Again, this result can be verified by the following computation, Pv , M , N Kronecker E, A, B, C, D , and its computation result is: 1 s 0 0 s 0 0 0 0 Pv 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 s 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 1 1 0 2 1 2 0 2 0 0 1 1 1 1 1 2 1 4 1 3 0 0 0 1 0 0 1 M 0 2 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 1 0 1 1 0 1 0 1 0 0 0 1 0 0 1 1 1 0 0 0 1 0 0 0 0 0 0 0 1 1 2 2 1 0 0 0 2 0 1 1 0 N 1 3 2 1 1 0 0 1 0 0 1 0 1 0 1 0 0 0 0 1 s 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 , 0 0 0 1 1 0 1 0 (27) 0 , 1 0 0 0 0 0 0 0 0 0 1 1 2 0 0 0 0 0 0 1 0 0 1 0 1 0 0 0 1 0 1 0 0 0 . 1 1 3 1 0 0 1 0 1 1 0 0 0 0 0 0 1 0 0 0 1 0 0 1 1 1 1 0 1 0 Thus, with these two numerical examples, we illustrate the computation process of our structural decomposition algorithm. It can be seen that the MATLAB functions are effective in computing the structural decomposition. 5. Conclusions We have presented in this paper MATLAB computation functions for the structural decomposition technique for general linear singular systems. The structural decomposition has a distinct feature of explicitly capturing and displaying the structure properties, which make it a powerful tool in solving system and control problems as its counterpart in nonsingular systems. The numerical examples showed that our computation programs are effective in giving a singular system’s structural decomposition form. They thus enhance the structural decomposition’s role as a powerful tool in solving practical problems. References: [1] S. L. Campbell, Singular System of Differential Equations II, Pitman, New York, 1982. [2] B. M. Chen, Robust and H Control, Springer, London, 2000. [3] B. M. Chen, A. Saberi, P. Sannuti and Y. Shamash, Construction and parameterization of all static and dynamic H 2 -optimal state feedback solutions, optimal fixed modes and fixed decoupling zeros, IEEE Transactions on Automatic Control, Vol. 38, 1993, pp.248-261. [4] L. Dai, Singular control systems, SpringerVerlag, Berlin, 1989. [5] M. He and B. M. Chen, Structural decomposition of linear singular systems: The single-input and single-output case, Systems and Control Letters, Vol. 47, No. 4, 2002, pp.325-332. [6] M. He, B. M. Chen and Z. Lin, Structural decomposition of general multivariable linear singular systems, Submitted to publish. [7] F. L. Lewis, A survey of linear singular systems, Circuits, Systems, and Signal Processing, Vol. 5, No. 1, 1986, pp.3-36. [8] Z. Lin and B. M. Chen, Linear systems and control toolbox, Technical Report, Department of Electrical and computer engineering, University of Virginia, USA, 2000 [9] X. Liu, B. M. Chen and Z. Lin, Computation of structural invariants of singular linear systems, Proceedings of the 2002 Information, Decision and Control Symposium, Adelaide, Australia, 2002, pp.35-40. [10] P. Misra, P. V. Dooren and A. Varga, Computation of structural invariants of generalized state-space systems, Automatica, Vol 30, 1994, pp. 1921-1936. [11] G. C. Verghese, B. C. Levy and T. Kailath, A generalized state-space for singular systems, IEEE Transactions on Automatic Control, Vol. 26, No. 4, 1981, pp.811-831.