See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3024539

H∞ Control for Nonlinear Descriptor Systems

Article  in  IEEE Transactions on Automatic Control · December 2002

DOI: 10.1109/TAC.2002.804465 · Source: IEEE Xplore

CITATIONS READS
111 249

3 authors, including:

He-Sheng Wang Chee-Fai Yung

National Taiwan Ocean University National Taiwan Ocean University
42 PUBLICATIONS   453 CITATIONS    102 PUBLICATIONS   672 CITATIONS   

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Chee-Fai Yung on 10 February 2014.

The user has requested enhancement of the downloaded file.

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 47, NO. 11, NOVEMBER 2002 1919

IV. AN ILLUSTRATIVE EXAMPLE AND CONCLUSION REFERENCES

We consider inner–outer factorization for [1] W. F. Arnold and A. J. Laub, “Generalized eigen-problem algorithms
and software for algebraic Riccati equations,” Proc. IEEE, vol. 72, pp.
s(s 0 2) 1
1746–1754, 1984.
[2] T. Chen and B. Francis, “Spectral and inner–outer factorizations of ra-
G(s) =
s(s 0 2) (s + 1)3
: (40)
tional matrices,” SIAM J. Matrix Anal. Appl., vol. 10, pp. 1–17, 1989.
[3] C. C. Chu and J. C. Doyle, “On inner–outer and spectral factorization,”
Proc. IEEE Conf. Decision and Control, 1984.
This example is from [12]. A minimal realization with identity observ-
[4] D. J. Clements and K. Glover, “Spectral factorization via Hermitian
ability gramian is found for G(s) (with four decimal points) pencil,” Linear Alg. Appl., pp. 797–846, 1989.
00 1948 01 1264 2 1535
: : :
[5] P. Van Dooren, “A generalized eigenvalue approach for solving Riccati
equations,” SIAM J. Sci. Stat. Comput., vol. 2, pp. 121–135, 1981.
A = 0 7476
: 00 1841 0 9221 : : [6] B. A. Francis, A Course in H Control Theory. Berlin, Germany:
00 7242 0 4672 02 6210
: : : Springer-Verlag, 1987, vol. 88, Lecture Notes in Control and Informa-
0 5458
: 00 4414 00 4414 : :
tion Sciences.
[7] K. Glover, “All optimal Hankel-norm approximation of linear multivari-
B = 00 3521 :
T = 00 4291 00 4291C : : : able systems and their L -error bounds,” Int. J. Control, vol. 39, pp.
0:6732 1:6190 1:6190 1115–1193, 1984.
[8] S. Hara and T. Sugie, “Inner–outer factorization for strictly proper func-
Following Step 4 of the algorithm, an orthogonal matrix U can be easily tions with j! -axis zeros,” Syst. Control Lett., vol. 16, pp. 179–185,
obtained which is listed as follows, together with the results of Step 3 1991.
[9] P. Kokotovic, H. K. Khalil, and J. O’Reilly, Singular Perturbation
of the algorithm: Methods in Control. Philadelphia, PA: SIAM, 1999.
0:3764 00 7196
: 0:5835
[10] M. Vidyasagar, Control System Synthesis: A Factorization Ap-
U = 0:9105 0:1711 00 3764
:
proach. Cambridge, MA: MIT Press, 1985.
[11] Y.-Y. Wang, S.-J. Shi, and Z.-J. Zhang, “A descriptor-system approach to
0:1711 0:6729 0:7196 singular perturbation of linear regulators,” IEEE Trans. Automat. Contr.,
vol. 33, pp. 370–373, Apr. 1988.
0:7071
V = F2 = 1:5119: [12] X. Xin and T. Mita, “Inner–outer factorization for nonsquare proper
0:7071 functions with infinite and finite j! -zeros,” Int. J. Control, vol. 71, pp.
Since F2 6= 0 and m = 1, we need compute a stabilizing solution to
145–161, 1998.

(32). However, due to the existence of zeros at the origin, the stabilizing
solution to (32) does not exist. Thus, we used the same method as in
[12] to compute a semistabilizing solution, and F1 , which are given by
0 042 12
: 00 200 72 : Control for Nonlinear Descriptor Systems
Z =
00 200 72 0 956 51
: :

1 = [ 0 014 66 00 069 86 ]
He-Sheng Wang, Chee-Fai Yung, and Fan-Ren Chang
F : : :

The state–space realization for the inner factor is obtained as

Abstract—In this note, we study the control problem for nonlinear

^
00 091 32
: 0:435 18
^
0:812 50 descriptor systems governed by a set of differential-algebraic equations
A =
0:400 54 01 908 68 :
B =
01 875 00
:
(DAEs) of the form
( ), where
_ = ( ), = ( ),
is, in general, a singular matrix. Necessary and
=

^
00 290 25
: 1:383 12
^
0:7071 sufficient conditions are derived for the existence of a controller solving the
C =
00 290 25
: 1:383 12
D = V =
0:7071
: problem. We first give various sufficient conditions for the solvability of
control problem for DAEs. Both state-feedback and output-feedback
cases are considered. Then, necessary conditions for the output feedback
Clearly, the mode corresponding to zero eigenvalue of A^ is not ob- control problem to be solvable are obtained in terms of two Hamilton–Ja-
servable. Indeed, by transforming the state–space realization into the cobi inequalities plus a weak coupling condition. Moreover, a parameteri-
transfer matrix, we find that zation of a family of output feedback controllers solving the problem is also

0 p 02
provided.
1 2 1
0:7071s 1:4146s s
G i (s) = 2 = Index Terms—Descriptor Systems, differential games, differential-alge-
1 s + 2s 1 2(s + 2) braic equations (DAEs), dissipation inequalities.
identical to the result in [12].
To conclude this note, we would like to point out that for the case I. INTRODUCTION
det(F2 ) = 0, the matrix pencil method as in [1], [4], and [5] can also
be used to compute the solution F1 . However, it is not easy to program, For the purpose of control, nonlinear descriptor systems are fre-
in terms of the accuracy of the stabilizing solution to (29). The method quently described by a set of differential-algebraic equations (DAEs)
from [11] seems to be more effective in this regard. It should also be
pointed out that for the case det(F2 ) = 0, the inner factor Gi (s), as in Manuscript received September 1, 1998; revised July 10, 2000 and July 16,
(37), represents a singular system. Interested readers are referred to [9] 2001. Recommended by Associate Editor H. Huijberts.
for a reduction of singular systems to regular systems using the singular H.-S. Wang is with the Department of Guidance and Communications
perturbation method. Technology, National Taiwan Ocean University, Keelung 202, Taiwan (e-mail:
hswang@mail.ntou.edu.tw).
C.-F. Yung is with the Department of Electrical Engineering, National Taiwan
ACKNOWLEDGMENT Ocean University, Keelung 202, Taiwan (e-mail: yung@mail.ntou.edu.tw).
F.-R. Chang is with the Department of Electrical Engineering, National
The author would like to thank the reviewer for bringing to his at- Taiwan University, Taipei, Taiwan (e-mail: frchang@ac.ee.ntu.edu.tw).
tention references [8] and [12]. Digital Object Identifier 10.1109/TAC.2002.804465

0018-9286/02$17.00 © 2002 IEEE

Authorized licensed use limited to: National Taiwan University. Downloaded on January 9, 2009 at 02:37 from IEEE Xplore. Restrictions apply.
1920 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 47, NO. 11, NOVEMBER 2002

of the form
4
where x = col(x1 ; . . . ; xn ) are local coordinates for an n-dimensional
x_ 1 =f (x1 ;
x 2 ; u) state-space manifold X . E is a constant matrix and F (0; 0) = 0. The
constant matrix E 2 IRn2n is, in general, a singular matrix with rank
(1)
0 =g (x 1 ; x 2 ; u) (2) E = r < n. Without loss of generality, we can assume that
I 0
or in a more compact form Ex_ = F (x; u); where E = and Ir 0
0 0 E = :
x = [x T xT ]T
4
= col(x1 ; . . . ; xn ) are local coordinates for an n-di-
0 0
1 2
mensional state–space manifold X . In the state–space X , dynamic state The following definition will be used (see [2]).
variables x 1 and instantaneous state variables x2 are distinguished. The Definition 1: [2] The DAE (3) is said to be of (uniform) index one
dynamics of the states x1 is directly defined by (1), while the dynamics if the constant coefficient system
of x2 is such that the system satisfies the constraint (2). In many cases, E w_ (t) 0 Fx (^x; 0)w(t) = g(t)
the algebraic constraint (2) of the full DAEs can be eliminated (usu-
ally due to the consistence of initial conditions). As a consequence, the is impulse free for all x
^ in a neighborhood of the graph of the solution,
DAEs reduce to a well-known state-variable system. Nevertheless, in where Fx denotes the Jacobian matrix @F =@x x.
some cases this kind of elimination is not possible (often due to in- The index of a DAE can be thought of as the generalization of the
consistent initial conditions), since it may result in loss of accuracy or nilpotent index [2] of a linear time-invariant descriptor system. The
loss of necessary information. A large class of physical systems can notion of index provides an easy way to guarantee the solvability of a
be modeled by this kind of DAEs. The paper by Newcomb and Dziurla given DAE. Rewrite the DAE (3) in the following form:
[6] gives many practical examples, including circuit and system design, Ir 0 x_ 1 F1 (x1 ; x 2 ; u)
robotics, neural network, etc., and presents an excellent review on non- = :
0 0 x_ 2 F2 (x1 ; x 2 ; u)
linear DAEs. Many other applications of DAEs, as well as numerical
treatment, can be found in [2]. Suppose that the aforementioned DAE is of index one. Then, from
In this note, we investigate the contractive property of DAEs, namely x2 )F2 (x1 ; x2 ; 0) is nonsingular
Definition 1, it is necessary that (@=@x
the H1 control problem. Our note is mainly divided into two parts. The around the equilibrium point x = 0. Consequently, by the implicit
first part concerns various sufficient conditions for the solvability of the function theorem, there exists a function h() so that the DAE reduces
H1 control problem. Both state feedback and output feedback cases to an ODE
are considered. We seek sufficient conditions under which a given DAE
has an L2 gain no greater than a prescribed positive number with in- x_ 1 = F1 (x1 ; h(x1 ); u)
ternal stability, and in the mean time, eliminates possible impulse dy-
namics and other singularity-induced nonlinearity of the system. We which is always solvable provided that F1 is smooth enough. This im-
will derive a family of output feedback controllers solving the H1 plies that DAE (3) is solvable.
control problem. The underlying ideas are differential games and dissi- In [9], some stability definitions and Lyapunov stability theorems for
pation inequalities. These ideas were also used by Isidori and Kang [3] nonlinear descriptor systems have been given. For the sake of brevity,
and Yung et al. [11], in which they have given the central controller and we do not reproduce those results here. Instead, we will derive an im-
a family of controllers, respectively, solving the H1 output feedback proved version of the Lyapunov stability theorem for DAE (3).
control problem for general nonlinear systems in usual state-variable Theorem 2: Consider DAE (3) with u = 0. Let Ex x(0) = Ex x0
form, i.e., the E matrix is nonsingular. be given. Suppose that there exists a C 3 function V : IRn 0! IR
The second part of this note is devoted to a converse result, namely vanishing at the points where Ex x = 0 and positive elsewhere which
the derivation of necessary conditions for solutions of local disturbance satisfies the following properties:
attenuation to exist. We obtain necessary conditions given in terms of i) (@=@x x)V = V~ T (x)E for some C 2 function V~ : IRn 0! IRn ;
the existence of nonnegative solutions to two Hamilton–Jacobi inequal- ii) V ( )F (x; 0) < 0 for all x 6= 0;
~ T x
ities, together with a weak coupling condition. A similar result has pre- iii) E T V~x = V~xT E  0, where V~x denote the Jacobian of V~ .
viously been published in [1] for (W -)input affine nonlinear systems Then, the equilibrium point x = 0 is locally asymptotically stable and
with nonsingular E matrix. In a recent monograph [7], among many the DAE is of index one.
other important contributions, van der Schaft addressed a number of Proof: We first show that the DAE has index one. Set V~ (x) =
issues related to necessary conditions for solutions of local disturbance V~1 (x1 ; x2 )
attenuation to exist (see also [3] for some related work). Our results : It is easy to show that
V~2 (x1 ; x2 )
can be thought of as a parallel extension of the results of [1] and [7] to
@2 ~ T
the DAEs case. As a matter of fact, the results in this note reduce to the V (x)F (x; 0) jx=0 = V~xT (0)Fx (0; 0)
ones given in [1] for state–space systems. @xx2
T ~x (0)
+Fx (0; 0)V < 0: (4)
This note is organized as follows. In Section II, we will review some
notions of DAEs together with some preliminary results for the theory
Condition iii) implies that (@=@x x2 )V~1 = 0; this, in turn, implies
of DAEs, including stability theory and dissipativity. Our main results
that (@=@x x2 )F2 (0; 0) is nonsingular. Consequently, by the continuity
will be summarized in Sections III and IV. We will concentrate on the
output feedback case. However, for the sake of completeness, we will of F , the DAE (3) is of index one. Now, consider the constant pair
first investigate the state feedback control. We also give a parameteri- {E , Fx (0; 0)}. From inequality (4) and condition iii), we can conclude
zaton of a family of output feedback controllers. In Section IV, we will that V~x (0) is a solution satisfying the generalized Lyapunov inequality
give a necessary condition for the H1 output feedback control problem V~xT (0)Fx (0; 0) + FxT (0; 0)V~x (0) <0
E T V~x (0) = V~xT (0)E 0:
to be solvable.

II. ELEMENTS FOR NONLINEAR DESCRIPTOR SYSTEMS Hence, by the standard result on Lyapunov stability of linear de-
scriptor systems [8], the pair {E , Fx (0; 0)} is admissible (i.e., regular,
Consider the following DAE:
asymptotically stable and impulse free). This, in turn, implies that the
Ex_ (t) = F (x; u); u2U  IRm (3) DAE (3) with u = 0 is asymptotically stable by noting that the pair

Authorized licensed use limited to: National Taiwan University. Downloaded on January 9, 2009 at 02:37 from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 47, NO. 11, NOVEMBER 2002 1921

{E , Fx (0; 0)} is a linearization of the DAE around the equilibrium if there exists a constant matrix L such that {E , A + LG} is impulse
point x = 0. Q.E.D. free), where
Remark: Because the intrinsic property of descriptor systems, the
@
x0 . Its rationale can be better
initial values must be given in the form Ex D= H
@u (x;u)=(0;0)
understood by investigating linear descriptor systems. Consider the fol-
@
lowing linear differential equation: A= F
@xx
E x_ = Ax
x(t) + Bu(t): (x;u)=(0;0)

@
G= H :
Taking Laplace transform of the previous equation yields @xx (x;u)=(0;0)

X (s) = (sE 0 A)01 [Ex

x(0) + BU (s)]:
We assume the invertibility of the pencil (sE 0 A) so that unique so-
lutions of the above equation are obtained for all Ex x(0) and U (s). In
particular, we point out that the initial conditions must be given in the
form Ex x(0). As a matter of fact, given u(t) for t  0, the knowledge
of Exx(0) is necessary and sufficient to completely determine x(t) for
t  0. It is part of the reason that the candidate Lyapunov function
2 kuk2  0, for all u 2 U ;
V (x) should be vanishing at the points where Ex x = 0 rather than
x = 0. On the other hand, for an index one descriptor system, (2) is
and the equilib-
simply an algebraic constraint. Therefore, only the part that Ex 6= 0
contributes to the energy function (see also [4], [9], and [10] for more
details).
Next, we give an extension of the LaSalle invariance principle.
Theorem 3: Consider the DAE
x_ =f1 (x; y )
0 =f2 (x; y)
where f1 , f2 are continuously differentiable functions. Suppose the
DAE is of index one. Let (x; y ) = (0; 0) be an equilibrium point for
the DAE (5a), (5b). Let V (x; y ) : D 0! IR+ = [0; 1) be a smooth
positive–definite function on a neighborhood D of (x; y ) = (0; 0),
such that V_ (x; y )  0. Let S = f(x; y ) 2 D j V_ = 0g, and suppose
that no solution can stay forever in S , other than the trivial solution.
Then, the origin is locally asymptotically stable.
Suppose that any bounded trajectory x (t) of the system
Ex_ = F (x(t); 0) satisfying H (x(t); 0) = 0 for all t  0 is
such that limt!1 x(t) = 0. Suppose also that there exists a C 3
function V : IRn 0! IR+ vanishing at the points where Ex x = 0 and
positive elsewhere which satisfies the following properties:
i) (@=@x x)V = V~ T (x)E for some C 2 function V~ : IRn 0! IRn ;
4
ii) Y0 = V~ T (x)F (x; u) + ky k2 0
iii) E T V~x = V~xT E .
Then, the DAE has an L2 gain less than or equal to
rium point x = 0 is locally asymptotically stable. Moreover, the DAE
is of index one.
Proof: The proof of the dissipative property is standard, hence
omitted. We now prove asymptotical stability and index one property.
We first show that the DAE (6) is of index one. Setting u  0 in Y0
(5a) and taking the second-order partial derivative of Y0 with respect to x
at (x; u) = (0; 0) yields
(5b)
AT V~x (0) + V~xT (0)A + GT G  0: (7)
Since {E , A, G} is impulse observable, inequality (7) along with con-
dition iii) implies that DAE (6) is of index one. To prove asymptotical
stability, observe that along any trajectory x() of the DAE with u  0
is such that
dV (x(t))
dt
 0kyk  0: 2

Proof: The proof is straightforward. Since the DAE is of index

one, as far as the (5b) is concerned, there exists a unique solution This shows that the equilibrium point x = 0 of the DAE (6) is stable.
y = g (x) such that f2 (x; g (x)) = 0, with g (0) = 0, provided by In addition, observe that any trajectory x () such that V_ (x(t)) = 0 for
the implicit function theorem. In this case, if limt!1 x(t) = 0, then all t  0 is necessarily a trajectory of Ex_ = f (x; 0) such that x (t) is
limt!1 y(t) = 0. The condition limt!1 x(t) = 0 is a direct conse- bounded and H (x(t); 0) = 0 for all t  0. Hence, by hypothesis, it is
quence of the usual LaSalle invariance principle. Q.E.D. concluded that limt!1 x (t) = 0 by using Theorem 3. Q.E.D.
Remark: In the aforementioned theorem, V_ denotes differentiation
with respect to t along the solution trajectory of (5a) and (5b). Because
III. THE H1 CONTROL PROBLEM
the descriptor system is of index one, it possesses a solution which is
impulse free. Let 6 be a nonlinear system described by the following DAE:
w 2 W  IRl ; u 2 U  IR
For the remainder of this section, we will investigate dissipative
property of a given DAE. Consider the following DAE:
Ex_ =F (x; w; u); m

z =Z (x ; w; u); z 2 Z  IRs
Ex_ =F (x; u); u 2 U  IRm ; F (0; 0) = 0
y =Y (x; w; u); y 2 Y  IRp (8)
y =H (x; u); y 2 Y  IRp ; H (0; 0) = 0 (6)
where x 2 X . Here u stands for the vector of control inputs, w is the
where x 2 X , together with a function exogenous input (disturbances to-be-rejected or signals to-be-tracked),
s : U 2 Y 0! IR y is the measured output, and finally z denotes the to-be-controlled
outputs (tracking errors, cost variables). It is assumed throughout that
called the supply rate. It is well known [7] that a usual state–space F (0; 0; 0) = 0, Z (0; 0; 0) = 0 and Y (0; 0; 0) = 0. The standard
system (i.e., E  I , the identity matrix) has L2 gain  if it is H1 control problem consists of finding, if possible, a controller 0 such
dissipative with respect to the supply rate that the resulting closed-loop system has a locally asymptotically stable
4 equilibrium point at the origin, is of index one, and has L2 gain (from w
s(u; y ) = 2 kuk2 0 ky k2 ; > 0:
to z ) less than or equal to . In the state feedback H1 control problem
This result can also be applied to DAE (6). In fact, we have the fol- we assume that y = x in (8), i.e., that the whole state is available for
lowing very important result. measurement. We suppose the following.
Theorem 4: Consider DAE (6) with Ex x(0) = Ex x0 given. Suppose A1) The matrix D12 has rank m and the matrix D11 T
D11 0 2 I is
that the matrix DT D 0 2 I is negative definite and {E , A, G} is negative definite, where D12 = (@Z=@u)(x;w;u)=(0;0;0) and D11 =
impulse observable (the triple {E , A, G} is called impulse observable (@Z=@w)(x;w;u)=(0;0;0) .

Authorized licensed use limited to: National Taiwan University. Downloaded on January 9, 2009 at 02:37 from IEEE Xplore. Restrictions apply.
1922 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 47, NO. 11, NOVEMBER 2002

A2) Any bounded trajectory x(t) of the system E x_ (t) = A4) The matrix D21 = (@Y =@w)(x;w;u)=(0;0;0) has rank p.
F (xx(t); 0; u(t)) satisfying Z(xx(t); 0; u(t)) = 0 for all t  0 is Define
such that limt!1 x(t) = 0.
r11 (xx) = 1 @ 2 H(xx; V~ T (xx); w; u)
A 0 j!E B2 2 @w2
A3) The matrix pencil has full column rank w= (x);u= (x)
C1 D12
for all ! 2 IR [ f1g, where A = (@F =@x x)(x;w;u)=(0;0;0) , B2 = r12 (xx) = 1 @ H(xx; V~ T (xx); w; u)
2

(@F =@u)(x;w;u)=(0;0;0) , and C1 = (@Z=@xx)(x;w;u)=(0;0;0) . 2 @u@w w= (x);u= (x)

Two preliminary lemmas will be needed in the sequel.
r21 (xx) = 1 @ H(xx; V~ T (xx); w; u)
2
Lemma 5: Consider the DAE (8). Assume that assumptions 2 @w@u w= (x);u= (x)
A1)–A3) are satisfied. Suppose the following hypothesis also holds.
H1: There exists a smooth real-valued function V (xx), locally de- r22 (xx) = 1 @ H(xx; V~ T (xx); w; u)
2

fined on a neighborhood of the equilibrium point x = 0 in X , which 2 @u2 w= (x);u= (x)

x = 0 and positive elsewhere such
is vanishing at the points where Ex and set
(1 0 1 )r11 (xx)
that the function
r12 (xx)
Y1 (xx) = H(xx; V~ (xx); 1 (xx); 2 (xx)) R(xx) =
r21 (xx) (1 + 2 )r22 (xx)
is negative semidefinite near x = 0, where the function H : IRn 2
where 1 and 2 are any real numbers satisfying 0 < 1 < 1 and
IRn 2 IRl 2 IRm 0! IR is defined on a neighborhood of (xx; p; w; u) =
(0; 0; 0; 0) as 2 > 0, respectively. The following theorem is readily obtained.
Theorem 6: Consider (11). Suppose assumptions A1)–A4) are satis-
H(xx; p; w; u) = pT F (xx; w; u) + kZ(xx; w; u)k2 0 2 kwk2 (9) fied. Suppose hypothesis H1 of Lemma 5 holds. Suppose the following
(@V =@xx) = V~ T (xx)E defined as shown in Theorem 4, 1 (xx) = hypothesis also holds.
w3 (xx; V~ (xx)) and 2 (xx) = u3 (xx; V~ (xx)), and w3 (xx; p) and u3 (xx; p) H2: There exists a smooth real-valued function Q(x x), locally de-
are defined on a neighborhood of (x x; p) = (0; 0) satisfying fined on a neighborhood of x = 0, which is vanishing at the points
@H (xx; p; w3 (xx; p); u3 (xx; p)) =0 where Ex x = 0 and positive elsewhere such that the function
@w ~ x); w(x
Y2 (xx) = K(xx; Q(x ~ x); y^(xx; Q(x
^ x; Q(x ~ x))); y^(xx; Q(x
~ x)))
@H (xx; p; w3 (xx; p); u3 (xx; p)) =0 is negative definite near x = 0, and its Hessian matrix is nonsingular
@u at x = 0. Here Q ~ : IRn ! IRn is a smooth function defined by
with w3 (0; 0) = 0 and u3 (0; 0) = 0.
x) solves the H1 state feedback (@Q=@xx) = Q~ T E with E T Q~x = Q~x E , the function K : IRn 2
Then, the feedback law u = 2 (x
IRn 2 IRl 2 IRp ! IR is defined on a neighborhood of the origin as
control problem. 555
Proof: The result of the lemma is a direct consequence of The- K(xx; p; w; y) = pT F (xx; w + 1 (xx); 0) 0 yT Y (xx; w + 1 (xx); 0)
orem 4. The details are thus omitted. w T w
Next, consider the case in which the state x of the DAE (8) is not
+
02 (xx) R(xx) 02 (xx)
available for direct measurement. Motivated by the work of Isidori and ^ x; p; y), respectively y^(xx; p), defined on a neigh-
and the function w(x
Kang [3] and Yung et al. [11], we consider a dynamic controller of the borhood of (0,0,0), respectively (0,0), satisfies
form
@K(xx; p; w; y) = 0 w(0;
^ 0; 0) = 0
E^ _ =F (; 1 (); 2 ()) + G()(y 0 Y (; 1 (); 2 ())) @w w=w^ (x;p;y)
u =2 () (10) respectively
where  = col(1 ; . . . ; n ) are local coordinates for the state-space @K(xx; p; w(x
^ x; p; y); y) = 0 y^(0; 0) = 0:
manifold Xc of the controller 0. The matrix G(), called the output @y y=^y(x;p)
injection gain, is to be determined. Substitute the controller (10) in (8) Then, if
to obtain the corresponding closed-loop system as
E ex_ e = F e (xxe ; w) z = Z e (xxe ; w) = Z(xx; w; ()) (11) ~ x)G(xx) = y^T (xx; Q(x
Q(x ~ x)) (13)
E 0 has a smooth solution G(xx) near x = 0, the nonlinear H1 output
where E e =
0 E^
, shown in the expressions at the bottom of the
feedback control problem is solved by the output feedback controller
page. Again, we try to render the closed-loop system locally dissipative
with respect to the supply rate 2 kwk2 0 kz k2 . Clearly, it suffices
^ = E.
(10) with E 555
to show that there exists a smooth nonnegative function U(x xe ) with Proof: Since the result of the theorem is a special case of that
(@U=@xxe ) = U~ T E e and E eT U~x = U~xT E e such that given in Theorem 8, we omit the proof here for brevity.

U~T F e (xxe ; w) + kZ(xxe ; w)k2 0 2 kwk2  0; for all w (12) A. Parameterization of Output Feedback Controllers
and such that the closed-loop system is locally asymptotically stable Recently, Yung et al. [11] have derived a set of parameterized so-
and is of index one. To state the main result of this section, a further lutions to the H1 control problem for general nonlinear systems in
assumption is needed. state-variable form. They have considered both output feedback and

x
xe = and

F (xx; w; 2 ())
F e (xxe ; w) = :
F (; 1 (); 2 ()) + G()(y 0 Y (; 1 (); 2 ())

Authorized licensed use limited to: National Taiwan University. Downloaded on January 9, 2009 at 02:37 from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 47, NO. 11, NOVEMBER 2002 1923

state feedback cases. Indeed, we can extend the technique developed Y3 (xa) = J (xa ; M ~ (xa ); w
~ (xa ; M
~ (xa ))) vanishes at the points
in [11] to give a family of H controllers for nonlinear differential-al-
1 where x a = col(Ex x; Exx; 0) = col(0; 0; 0) and is negative elsewhere.
gebraic systems. Then, the family of controllers (14) with E ^ = E solves the H 1
Motivated by the work of [11], we consider the family of controllers output feedback control problem.
described by the following DAEs: Proof: Set U (xa ) = V (x) + M (xa ). It follows that
E^ _ =F (; 1 ( ); 2 ( ) + c()) (@U=@x xa ) = U~ T (xa )Ea with EaT U~x (xa ) = U~xT (xa )Ea , where
+ G( )(y 0 Y (; 1 ( ); 2 ( ) + c( )))
V~ x 0 0
U~x = 0 0 0 +M
~x :
+g ^1 ( )c( ) + g^2 ( )d( )
0 0 0
E _ =a(; y 0 Y (; 1 ( ); 2 ( ) + c()))
Q
With the equations Y1 (x) and Y3 (x) in hand, we have the following
u =2 ( ) + c() (14) Hamiltonian equation by using Taylor series expansion:
where  and  are defined on some neighborhoods of the origins in X c dU +kZ (x; w;  ( ) + c())k2 0 2 kwk2
and IR , respectively. G() satisfies (13). a(; ) and c() are smooth dt
q 2

functions with a(0; 0) = 0 and c(0) = 0. g^1 (), g^2 () and d() are C k
w 0 2 (x) T

=Y1 (x) + Y3 (x ) +
functions (k  1). E is a constant matrix, and, in general, is singular.
Q 2 ( ) + c() 0 2 (x)
a

The functions a(; ), c(), g^1 (), g^2 (), d(), and the matrix E are
1 1 r110 (x) 02 r022 (x) 2 ()w+0c(2)(0x)2 (x)
Q

to-be-determined variables such that the closed-loop system (8)–(14)

is dissipative with respect to the supply rate 2 kwk2 0 kz k2 , and is
+ kw 0 w ~ (x ))k2~
~ (x ; M
T
locally asymptotically stable with index one. a ( ) a R x

Observe first that the DAEs describing the closed-loop system w 0 1 ( )

x 3
+o
(8)–(14) can be put in the form 2 ( ) + c() 0 2 (x)
E x_ =F (x ; w)
a a a a

z =Z (x; w; 2 ( ) + c())
+o kw 0 w~(x ; M~ (x ))k
a
T
a
3
(16)

4 E0 E0^ 00
where
x 4 col(x ; ;  ), E = 41 @ 2 J (x ; M~ (x ); w) T
where a = a , and the equa- R~ (xa) =
a a

0 0 E 2 @w2
4
Q =~( ~ ( )) w w x ;M x

tion at the bottom of the page holds. In that equation, F~ (;  ) = and the notation kv k ~ stands for v Rv . It is easy to verify that R
2 ~ T ~ (0) =
F (; 1 (x); 2 ( ) 0 G()Y (;  ();  () + c()):
+ c()) 1 2 (1 0 1 )(D11 D11 0 I ). Since Y1 (x) and Y3 (x ) are nonpositive,
T 2
R

2 IR 2 IR ! IR
a
Consider a Hamiltonian function J : IR 2n+q 2n+q r
(16) implies that
defined as follows: dU kZ (x; w;  () + c())k 0 kwk  0 (17)
+
2 2 2
J (x ; p ; w) =p F (x ; w)
a a
T
a a a dt 2

w 0 2 (x) T which, in turn, implies that the closed-loop system has an L -gain less 2
+
2 ( ) + c() 0 2 (x) than or equal to . Set w = 0, rearrange terms, and use (16) to get
dU
2 R(x) 2 ()w+0c(2)(0x)2 (x) : 0 kZ (x; 0;  () + c())k + Y (x) + Y (x ) 2
=
(15) dt 2 2 3 a

0 (x) 2 r (x) 0
T
1 11
 ( )+ c() 0  (x) 0 r (x)
It is easy to check that +
2 0 2 2 22
@ J (x ; p ; w)
2
0  )(D D 0 I ) 1  () +0c(()x0)  (x) + kw~(x ; M~ (x ))k
a a T 2
= 2(1
@w2 (x ;p ;w)=(0;0;0)
1 11 11 2
a
T
a
2
~ (x
R )
2 2
which is negative definite by A1). Then, by the implicit function the-
orem, there exists a unique smooth function w ~ (xa ; pa ), defined on a +o
0 (x) 1
3

neighborhood of the origin, satisfying  ( ) + c() 0  (x)

2 2

@J (x ; p ; w) + o kw ~ (x ))k
~ (x ; M
T 3
a a
=0 w~ (0; 0) = 0: a a

@w w =w
~ (x ;p )
which is negative semidefinite near xa = 0 by hypothesis. This shows
Lemma 7: Consider (8) and (14). Suppose assumptions A1)–A4) that the closed-loop system is stable locally around the equilibrium
are satisfied. Suppose hypotheses H1 of Lemma 5 and H2 of Theorem point. We claim that the DAE (14) has index one. To see this, observe
6 hold. Furthermore, suppose that the following hypothesis also holds. that any trajectory satisfying (dU=dt)(x(t);  (t);  (t)) = 0 for all
H3: There exists a smooth real-valued function M (xa ), locally t  0 is necessarily a trajectory of
defined on a neighborhood of the origin in IR2n+q , which van-
ishes at the points where x a = col(Ex x; Exx; 0) = col(0; 0; 0), Ex_ (t) = F (x; 0; 2 ( ) + c()) (18)
is positive elsewhere, satisfies (@M (xa )=@x xa ) = M~ T (xa )Ea such that x(t) is bounded and Z (x ; 0; 2 ( ) + c( )) = 0 for all
with Mx (xa )Ea = Ea Mx (xa ), and is such that the function
~ T T ~
t  0. This shows that the previous DAE has index one. Moreover,

F (x; w; 2 ( ) + c())
F a
4
= F~ (; ) + G( )Y (x; w; 2 ( ) + c()) + g^1 ( )c() + g^2 ( )d() :
a(; Y (x; w; 2 ( ) + c()) 0 Y (; 1 ( ); 2 ( ) + c()))

Authorized licensed use limited to: National Taiwan University. Downloaded on January 9, 2009 at 02:37 from IEEE Xplore. Restrictions apply.
1924 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 47, NO. 11, NOVEMBER 2002

hypotheses H1 and H3 along with assumption A1) imply that the tra- for all (x,  , w ) in a neighborhood of (0, 0, 0). Consider the case that
jectory satisfying (dU=dt)(x x(t); (t); (t)) = 0 for all t  0 is  6= 0 and U~ (xx; ) 6= 0. Since  6= 0, we have 8 6= 0. Hence,
necessarily a trajectory such that x (t) = (t) and (t) = 0 for all from (22), we have inf max W (x x; ; w) = 01, because inequality
t  0. Setting x(t) = (t)  0 and w(t) = 0 in (15), we have 8; 2 w
(22) contains a term linearly in 8. Next, consider the case that  6= 0
(M~  )T a(; Y (0; 0; c()) 0 Y (0; 0; c())) < 0; for all  6= 0, where but U~ (xx; ) = 0. Suppose that U~ is nonsingular for every (x ,  )
M~ T (xxa ) = [(M~ x )T (M~  )T (M~  )T ]. This shows that the DAE satisfying U ~ (xx; ) = 0: Then, by the implicit function theorem, the
EQ _ = a(; Y (0; 0; c()) 0 Y (0; 0; c())) (19) previous identity has a differentiable solution  = `(x x) with `(0) = 0.
has index one and is asymptotically stable. Hence, by hypothesis H2 The previous statement is needed in the subsequent proof. We take it
as a standing assumption.
and the fact that DAE (18) and (19) have index one, we can conclude
A5) U ~ (xx; ) = 0 if and only if  = `(xx) for some smooth function
` with `(0) = 0. Furthermore, U~ (xx; )j=`(x) is nonsingular.
that the closed-loop system (8)–(14) has index one. Asymptotical sta-
bility then easily follows by Theorem 4. Q.E.D.
The previous lemma gives a general form of the output feedback con- Setting  = `(x x) in (12) yields
trollers. However, it does not explicitly specify how we can choose the V~ T F (xx; w; 2(`(xx))
free system parameters EQ , a(; ) and c() in order to meet the hy- +kZ(xx; w; 2(`(xx))k2 0 2 kwk2  0; 8w: (23)
pothesis in Lemma 7. In the sequel, we give a way to meet the condition
in Lemma 7, and in the mean time, to reduce the number of indepen- This shows that inf max W (x x; ; w) = Y1 (xx); where  = `(xx).
0; 2() w
dent variables. Consider the following DAE: Hence, the state feedback law u = 2(`(xx)) solves the state feedback
EQ _ = a(; ): (20) H1 control problem for 6. This shows that V is a solution of Y1 . A
If there exists a smooth function L(), locally defined on a neighbor- further necessary condition is obtained by restricting to the class of
hood of  = 0, which vanishes at the points where EQ  = 0, is posi- controller 0 which produces zero control input u. Consider the Hamil-
tive elsewhere, satisfies (@L()=@) = L ~ ()T EQ with EQT L~  () = tonian function K : IRn 2 IRn 2 IRl 2 IRp ! IR defined as
~L ()T EQ , and is such that L~ T ()a(; ) < 0, then we can conclude K (xx; p; w; y) = pT F (xx; w; 0) 0 yT Y (xx; w; 0)
from Theorem 2 that DAE (20) is locally asymptotically stable and has +kZ(xx; w; 0)k2 0 2 kwk2 : (24)
index one. Henceforth, if some further hypotheses are imposed in the
above inequality, the condition in Lemma 7 can be met. This is sum- It is easy to verify that
marized in the following theorem. @ 2 K (xx; p; w; y) T
= 2(D11 D11 0 2 I):
Theorem 8: Consider (8) and (14). Suppose assumptions A1)–A4) @w2 (x;p;w;y)=(0;0;0;0)
are satisfied. Suppose hypotheses H1 of Lemma 5 and H2 of Theorem
6 hold. Suppose also that the following hypothesis holds. This shows that there exists a smooth function w(x^ x; p; y) defined in a
H4: There exists a smooth function L(), defined as previously neighborhood of (0,0,0) such that
shown, such that the function @K (xx; p; w; y) = 0 w(0;
^ 0; 0) = 0:
~T w T w @w w=^
w(x;p;y )
Y4 (; w) = L ()a(; Y (0; w; 0)) + R(0)
c() c()
at w = w+ (), viewed as a function of  , is negative definite near
Furthermore, it is also easy to check that

 = 0, and its Hessian matrix is nonsingular at  = 0. The func- @ 2 K (xx; p; w(x

^ x; p; y); y) =
tion w+ () is defined on a neighborhood of  = 0, which satisfies @y2 (x;p;y)=(0;0;0)
(@Y4 (; w)=@w)w=w () = 0 with w+ (0) = 0 (This function exists, 1 ( 2 I 0 DT D )01 D DT : (25)
for R(0) is nonsingular). 2 11 11 21 21
Then, if g^1 () and g^2 () satisfy 3
Thus, there exists a smooth function y (x; p) defined in a neighbor-
~ x)^g1 (xx) = 2 T (xx; 0; 0)r12 (xx) 0 2(1 + 2 )T2 (xx)r22 (xx)
Q(x hood of (0,0) such that
and @K (x; p; w(x ^ x; p; y); y) = 0 y3 (0; 0) = 0:
@y y =y (x;p)
~ x)^g2 (xx) = aT (0; Y (xx; 1 (xx) + (xx; 0; 0); 0))
Q(x
Set w3 (xx; p) = w(x ^ x; p; y3 (xx; p)). Then, we have
respectively, where (x x; ; )=~ ~ x 0 ) 0 Q(x
w(xxa ;[ Q(x ~ x 0 )L~ () ]),
4 K (xx; p; w; y)  K (xx; p; w(x ^ x; p; y); y) (26)
the family of controllers (14) with d() = L() solves the H1 output
~
feedback control problem. 555
4 Q(xx 0 ) +
for all (x , p, w , y ) in a neighborhood of the origin and
Proof: It is straightforward to verify that M(x xa ) = K (xx; p; w(x^ x; p; y); y)
L() satisfies the hypothesis of Lemma 7.  K (x; p; w3 (xx; p); y3 (xx; p)); y3 (xx; p)) (27)
IV. CONVERSE RESULT—A NECESSARY CONDITION for all (x , p, y ) in a neighborhood of the origin. We will show that it is
Suppose that the H1 control problem is solved by the output feed-
necessary
back controller 0 which has the following representation: K (xx; P~ (xx); w3 (x; P~ (xx)); y3 (xx; P~ (xx)))  0 (28)
E^ _ =8(; y) for some storage function P (x ~
x) with (@P =@xx) = P E . This is sum-
T

u =2() (21) marized in the following statement.

Theorem 9: Consider system (8) and suppose assumptions A1)–A5)
and let U be a smooth function satisfying
hold. Suppose that the H1 control problem is solved by the output
F (xx; w; 2())
W (xx; ; w) = U~x U~ feedback controller (21). Suppose that there exists a smooth real-valued
8(; Y (xx; w; 2()) function U(x x; ), which vanishes at the points where E exe = 0 and
+kZ(xx; w; 2())k2 0 2 kwk2  0 (22) ~x = U~xT E e , and satisfies (21) for
is positive elsewhere with E eT U

Authorized licensed use limited to: National Taiwan University. Downloaded on January 9, 2009 at 02:37 from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 47, NO. 11, NOVEMBER 2002 1925

all (x,  , w ) in a neighborhood of (0, 0, 0). Then, the Hamilton–Jacobi respectively. Next, observe that
inequalities
Y1 (x)  0 and K (x; P~ (x); w3 (x; P~ (x)); y 3 (x; P~ (x)))  0 @2Q
(0) = @2P
(0) 0 @2V
(0)
x
@x 2 @xx 2 x2
@x
have solutions V (x) and, respectively, P (x) (with (@P =@x x) = 2 2 2
P~ T E ) given by V (x) = U (x; `(x))  0 and, respectively, = @@xxU2 (0; 0) 0 @@xxU2 (0; 0) + @@xxU (0; 0)`x (0)
P (x) = U (x; 0)  0. Furthermore, Q(x) = P (x) 0 V (x)  0. 555
4
2 2
Proof: It is obvious that V (x) is a solution satisfying Y1 (x)  0 = 0 @@xxU (0; 0)`x (0) = `xT (0) @@U2 (0; 0)`x (0)  0:
from our previous observation. It is claimed that P (x) is a solution of
inequality (28). To see this, setting  = 0 in (12) yields
(34)

P~ T F (x; w; 0) + U~  (x; 0)8(0; Y (x; w; 0)) The last inequality holds by assumption A5). This concludes that
Q(x)  0 by noting that Q(x) has the following Taylor series
+kZ (x; w; 0)k2 0 2 kwk2  0: (29) expansion:

Let U~  (x; 0)8(x; y) = 5T (x; y)y , where 5(x; y) is a vector @Q 1 @ 2 Q (0)x + h:o:t:  0:
Q(x) = Q(0) + (0)x + xT
of smooth functions. This can always be done because the func-
tion 8(0; y ) vanishes at y = 0. Use 5T (x; y )y and choose
x
@x 2 @xx2
w=w ^(x; P~ ; y) in (12) to obtain It is nonnegative around the origin because it vanishes at the origin
together with its first-order derivative, and its second-order derivative
K (x; P~ ; w
^(x; P~ ; y);
5(x; Y (x; w^(x; P~ ; y))))  0: (30) is positive by (34). This completes the proof. Q.E.D.

Observe that the Hessian matrix of y 05(x ; Y (x; w^(x; P~ ; y))) is non- REFERENCES
singular [from assumption A4) and (25)]. Hence, by the implicit func-
tion theorem, there exists a unique solution, denoted by y^(x), satisfying
[1] J. A. Ball, J. W. Helton, and M. L. Walker, “ control for nonlinear

^(x; P~ ; y^(x))) = 0, y^(0) = 0. Set y = y^(x) in

y^(x ) 0 5(x ; Y (x; w
systems with output feedback,” IEEE Trans. Automat. Contr., vol. 38,
pp. 546–559, Apr. 1993.
(30) to obtain [2] K. E. Brenan, S. L. Campbell, and L. R. Petzold, Numerical
Solution of Initial-Value Problems in Differential-Algebraic Equa-
K (x; P~ (x); w
^(x; P~ (x); y^(x)); y^(x))  0: tions. Philadelphia, PA: SIAM, 1996.
[3] A. Isidori and W. Kang, “ control via measurement feedback for
general nonlinear systems,” IEEE Trans. Automat. Contr., vol. 40, pp.
This shows that 466–472, Mar. 1995.
[4] E. Jonckheere, “Variational calculus for descriptor problems,” IEEE
K (x; P~ (x); w3 (x; P~ (x)); y 3 (x ; P~ (x)))  0 Trans. Automat. Contr., vol. 28, pp. 491–495, May 1988.
[5] W. M. Lu and J. C. Doyle, “ control of nonlinear system via output
feedback: Controllers parameterization,” IEEE Trans. Automat. Contr.,
from (27). In order to complete the proof, we have to show that Q(x) =
4
vol. 39, pp. 2517–2521, Dec. 1994.
P (x) 0 V (x)  0. Note that the function U (x;  ) has the following [6] R. W. Newcomb and B. Dziurla, “Some circuits and systems applications
Taylor series expansion: of semistate theory,” Circuits, Syst., Signal Processing, vol. 8, no. 3, pp.
235–260, 1989.
[7] A. J. van der Schaft, -Gain and Passivity Techniques in Nonlinear
@U
U (x;  ) =U (0; 0) +
x
@x
(0; 0)x + @U
@
(0; 0) Control. Berlin, Germany: Springer-Verlag, 1996, vol. 218, Lecture
Notes in Control and Information Sciences.
x (0; 0) x@ (0; 0)
T 1 @ U 1 @ U [8] H. S. Wang, C. F. Yung, and F. R. Chang, “Bounded real Lemma and
+ x 2 @x
T
2 @x x control for descriptor systems,” in Proc. Inst. Elect. Eng., vol. 145,
1 @ U
2 @xx@ (0; 0) 2 @ (0; 0)
1 @ U  1998, pp. 316–322.
+ h:o:t: (31)
[9] H. Wu and K. Mizukami, “Stability and robust stabilization of nonlinear
descriptor systems with uncertainties,” in Proc. 33rd Conf. Decision and
Control, Lake Buena Vista, FL, 1994, pp. 2772–2777.
where “h.o.t.” means high order terms. Let  = `(x), then we have the [10] H. Xu and K. Mizukami, “Hamilton-Jacobi equation for descriptor sys-
following two Taylor series expansions: tems,” Syst. Control Lett., vol. 21, pp. 321–327, 1993.
[11] C. F. Yung, J. L. Wu, and T. T. Lee, “ control for more general non-
linear systems,” IEEE Trans. Automat. Contr., vol. 43, pp. 1724–1727,
@2U @2U @2U
x
@x 2
(x; `(x)) = 2 (0; 0) +
x
@x x@
@x
(0; 0)`x (x) Dec. 1998.

+ (h:o:t:)xx j=`(x) (32)

@2U @2U T @2U
@@xx
(x; `(x)) =
x@
@x
(0; 0) +
@ 2
(0; 0)`x (x)
+ (h:o:t:)xx j=`(x) : (33)

Note that (@ 2 U =@x

x2 )(x; `(x)) = (@ 2 V =@x
x2 )(x), and (@ 2 U =@@x
x)
(x; `(x)) = 0. Set x = 0 in (32) and (33), respectively, to get
@2U @2V 2 2

x2
@x
(0; ` (0)) = x2
@x
(0) = @@xxU2 (0; 0) + @x
@ U
x@
(0; 0)`x (0)
and
2 2 T 2
@ U
0 = @@x
x
(0; 0) = @@xxU@ (0; 0) + @@U2 (0; 0)`x (0)

Authorized
View publication stats licensed use limited to: National Taiwan University. Downloaded on January 9, 2009 at 02:37 from IEEE Xplore. Restrictions apply.