IEEE TRANSACTIONS ONVOL.

AUTOMATIC
CONTROL. AC-32, NO. 5 , MAY 1987 423

TABLE I Pole Assignment in a Specified Disk

-__.__~
-~~. ___- ___.____
_____~~.__ _ . . _ _ . _ _ _ _ _ _ _ _ _ . ~

A B N~ x , o0 D, D: r, J, K . FURUTA AND S . B. KIM

-. _ _ . _ _ _ _ _ _ ~ - - _ _ _ _ _ _ _ - ~

0 1 0 0 0 1 0 0 0 1 1 1 Abstract-The problem of assigning all poles of a closed-loopsystem in

1 0 0 0
0 0 1 0 1 0 0 1 0 a specified disk by state feedback is considered for both continuous and
0
0 t o 1 1 1 1 I O 1
1 0 - 1 - 1 - 1 0discrete systems. A state feedback control law is determided by using a
-~ ~~ ~ ~ ~ _ _ _ _ discrete- Riccati equation. This kind of pole assignment problem is named
D-pole assignment, and its relation lo the optimal control problem and its
robustness properties are discussed. The gain and phase margins for all
TABLE I1
__ .~
~
-~ ~

~~
~ _ _ _ _ ~ . _ _ _ _ _ _ ~ _
__ _ _ _ ~
~~

~~ -___ closed-loop
~ polesto stay inside thespecified disk D are determinedfor the
proposed control.
N(rJ ) DF J 1 F
__ ________
I. INTRODUCTION
1 0 1 2 -6 0 2213 4 1013 ~

-3 -4 -1 Pole assignment, using state feedback, is one of the most frequently

-2 -3 1 -5 27 0 -8513 -16 3713' employed methods used to design control systems. Much research
concerning pole assignment has been done (1).
In designing control systems based on pole assignment, it may be
posable pairs and Lemma 2.1, one can characterize the admissible Jordan satisfactory in practice that the closed-loop system has all poles in a
pairs that can be assigned via state feedback. The method developed has desired region. But in this type of design method, the variation of the pole
advantages over other methods in several respects. First, when assigning locations due to plant parameter perturbations has not been studied.
multiple defective eigenvalues it avoids solving for matrices Z(h) and Bogachev et al. (2) proposed a method to give a state feedback law to
V ( h )that satisfies the relation [(XI - A ) Z ( h ) + BV(h) = I].Second, assign all poles of the closed-loop system in a specified region, in the Iefi
(3.9) is parametric in terms of (r,J). From (3.9) one can compute F for half of the complex plane for a continuous system or in the unit disk with
different r and J . and then compare several F s for different design the center at the origin for a discrete system. Their method is based on
objectives. Third, the proposed method gives more insight to the theory of conformal mapping and the control law is only determined by using the
pole placement from the matrix polynomial point of view. However. there
iteration of a complicated process. Mori er al. (6) proposed to use the
may be difficulty in calculating the transfer function (3.2).
discrete Riccati equation to determine the control law of the continuous
It is very well known [2]-[4] that there is more freedom offered by state system. But they did not consider properties of the control system, such as
feedback than assigning the eigenvalues in the multiinput case. This
its relation to the optimal control and robustness that all poles remain in
freedom lies in the particular choice of the matrix r, this matrix assigns
the specified disk in the presence of plant parameter perturbations.
the eigenvectors of the closed-loop state matrix, Since the matrix N(rJ )
This technical note considers a similarproblem to assign all poles of the
consists of all eigenvectors and possibly generalized eigenvectors, then
closed-loop system in a specified disk for both continuous systems and
the extra freedom offered can be used to achieve someother design
discrete systems by using the Riccati equation. In this proposed algorithm,
objectives 191-1121by properly choosing r within the conditions of the state feedback law is determined by using the solution of a discrete
theorem (3.1 ) .
Riccati equation which can be computed directly. This pole assignment is
ACKNOWLEDGMENT named D-pole assignment, and the gain and phase margins for all poles
staying in the specified disk D can be evaluated.
The author wishes to thank the reviewers for their valuable comments
and for bringing [l] to my attention. He also acknowledges the support of II. D-POLE ASSIGNMEKT PROBLEM .4ND ITS ALGORITHM
UPhl .
In this section, both the continuous system
REFERENCES

W. M. Wonham. "On pole assignment in multi-input, controllable linear ~=Ax+Bu (la)

systems," IEEE Trans. Auromat. Contr.. vol. AC-12. pp. 660-665. Dec. 1967.
V. M. Popob. "Invariant description o f linear. time-invariant controllable y =c x (Ib)
systems." SIAM. J . Contr.. vol. 10, no. 2, May 1972.
B. C. Moore. "On the flexibility offered by state feedback in multivariable
systems beyond closed loop eigenvalue assignment." IEEE Trans. Automat. and the discrete system
Conrr.. vol. AC-21, pp. 689-692. Oct. 1976.
G . Klien and B. C. Moore. "Eigen\,alue generalized eigenvector assignment with Xk+, = A & + Bu* (2a)
state feedback." IEEE Trans. Automat. Contr.. vol. AC-22, pp. 10.1-141, Feb.
1977.
I. Gohberg. P. Lancaster, and L. Rodman, h4urrir Po/ynornia/s. New York: Y c = CXk (2b)
Academic, 1982.
hi. E. Warren and A. E. Fxkberg. Jr.. "On the dimension of controllability
subspaces: A characterization via polynomial matrices and Kronecker invariants." are considered, where u is an m-dimensional input vector, x is an n-
S1il.W. J . Contr., vol. 13. no. 2 , Feb. 1975. dimensional state vector, and y is a p-dimensional output vector, and A ,
C. T. Chen. Introduction Io Linear System Theory. New York: Holt.
Rinehart, and Winston, 1970.
B, C are constant matrices of appropriate dimensions. It is also assumed
T. M. Bakri, "Parametric approach to eigenstmcture assignment and robustness that the pair ( A , B ) is controllable (or reachable).
anal>sis," Ph.D. dissertation. Colorado State Univ.. 1984. The problem to be considered is to determine the state feedback
J . Kautsky. X. K. Nichols. and P. Van Dooren. "Robust poleassignment in linear
state feedback." Int. J . Contr.. vol. 41, no. 5, pp. 1139-1155.
P. Petkov. X. D. Christor. and M . Konstantinos. "A computational algorithm for u=Fx (3)
pole assignmsnt of linear single-input systems." IEEE Trans. Automat. Contr..
vol. AC--39, pp. 1015-10.18. Nov. 1981.
- , "A computational algorithm for pole assignment of linear multinput Manuscript received April 7, 1986: revised April 25. 1986. October 6. 1986. and
systems." IEEE Trans. Automaf. Conrr.. vol. AC-31. no. I I . pp. IW-1047, January 5. 1987.
1986. The authorsare with the Department of Control Engineering. Tokyo Institute of
G. Mminis and C. Paise. .'An algorithm for pole assignment of time invariant Technology. Tokyo. Japan.
linear systems," Inr. J . Conrr.. vol. 35. no. 2 . pp. 341-354. IEEE Log Number 8613532.

0018-9286/87/0500~23S01.OO 0 1987 IEEE

424 IEEE TRANSACTIONS
ON AUTOMATIC
COhTROL. VOL. AC-32. NO. 5 , MAY 1987

or
=F x ~ (3) '

such that all poles of the closed-loop system, i.e., the roots of

det ( S I - ( A + BF))= 0 (4)

are assigned in the disk D with the center at a + j 0 and the radius r which
may be shown in Fig. l(a) for the continuous system and in Fig. l(b) for
the discrete system.
The assignment of all poles of the closed-loop system in the specified
disk D shown in Fig. 1 is defined as D-pole assignment problem in this
note. To solve this problem, we present the conditions such that all (a) (b)
eigenvalues of a matrix should be located in a specified disk as shown in Fig. 1. (a) Disk D in the left half of the complex plane. (b)Disk D in the unlt disk with
Fig. 1. Note that the following conditions are satisfied for both continuous the center at the origin.
and discrete systems.
Lemma I: Consider the matrix equation
become zero.Therefore, it proves that there exists a positive definite
-aA*P-aPA+A*PA+(a2-r2)P= -0 (5) solution P such as Iimk+= PI- = P which satisfies ( 5 ) .
Remark: We may see from the abovelemma that a different expression
where Q is an arbitrarily positive definite matrix, and * denotes the for (5) can be given
conjugate transpose of amatrix, and cy, r are scalars.Then,the
eigenvalues of matrix A are located within the specified disk as shown in (A-aZ)' (A-aZ)
P-- p= --
Q
(9
Fig. 1 if and only if there exists a positive definite solution P satisfying r r r2
(5).
Proof: Let A. v be an eigenvalue and right eigenvector of A , then and that the eigenvalues of the matrix ( A - a Z ) / r are located within a
unit disk with its center at the origin if and only if there exists a positive
A u = X u , u*A*=Xu* (6) definite solution P that satisfies (5). In the previous lemma. we considered
only the case such that Q is positive definite. Also. in the case that Q =
and substituting these expressions into ( 5 ) yields HTH, the lemma is satisfied for the positive semidefinite Q as far as the
pair ( A , H ) is observable (7).
{ - a ( X + h ) + IXIz+a2-r2]u*Pu=-o*Qu. (7) Consider the state feedback
Let h = x + j y and using it in (7). we obtain u=Fx (3)

{ - 2 o ~ u + x ~ + y ~ + a ~ - r ~ } u-v*Qu
*Pv= (8) then, from Lemma 1, we can have the conditions that the eigenvalues of
the closed-loop mamx A + 5 F should be located within a specified disk
or D.
Theorem I : Consider the following matrix equation:
{ ( x - a ) 2 + y 2 - r z ] u * P u =-u*Qv. (8) '
-a(A+BF)*P-aP(A+BF)+(A+BF)*P(A+BF)
Since Q is positive definite, the positivity of P yields
+ ( a z - r 2 ) P =- 0 (13)
(x-a)'+y2-r2<0 (9)
where Q is positive definite.Thenthe eigenvalues of ( A + 5F) are
which means the condition that all eigenvalues of the matrix A should be within a specified disk D as shown in Fig. 1 if and only if there exists a
located in a specified disk with radius r a n d center at a + j0 (Fig. 1). positive definite solution P satisfying (13).
Conversely, let the eigenvalues of matrix A be inside a specified disk of As a method to choose a state feedback law that satisfies our problems,
Fig. 1. that is. the eigenvalues of ( A - d ) / r are within the unit disk we present the following theorem by using the discrete Riccati equation.
with its center at the origin. Consider the following matrix: Theorem 2: The state feedback law
u= -(rZR+BTPB)-'BTP(A-aZ)x (14)
k l
Pk= 7 ( ( A- cyI)*)'Q(A- aZ)'. (10)
assigns all the closed-loop poles of a continuous system (1) or discrete
i=o
system (2) in the disk D shown in Fig. 1. where P is a positive definite
Since Q > 0, it is shown that the right term of (IO) is positive definite for symmetric solution of the Riccati equation
all i. From ( 5 ) and (lo), we can obtain
(A-an'. ( A-a]) ( A - a1)
ps- +HTH--
J

1 1 r r r
-(A-aI)*Pk(A-aI)-Pk= -7 Q
rz r
. P B ( r Z R + B J P B ) - i B (TAP-y~ 2 )
I (15)

R is an arbitrarily positive definite matrix, and H i s a matrix such that the

and pair ( A , H ) is observable.
Proof: Substituting (14) into (Is), w'e can rewrite (15) as the
following:

-a(A+BF)'PPaP(A+BF)+(A+BF)'P(A+BF)+(a2-r2)P
By hypothesis of the matrix A , the following is yielded as k -+ 03
= -r2(FTRF+H'H) (15)'
(llA -aZll/r)k+l+O where

which means that the right-hand side of (12) and the second term of (1 1) F= - ( r ' R + B J P B ) - ' B J P ( A-aZ).
IEEE TRANSACTIONS OK AUTOMATIC
CONTROL, VOL. AC-32. NO. 5, MAY 1987 325

In ( l 5 ) ' , let where P given by (15) is the solution PCof the following matrix equation:

Q= r2(FTRF+ HTH) (A-BRc-'BTPc)TP,+Pc(A-BRc-'BTPc)

and using the above Lemma 1, its remarks and Theorem 1, it proves that 1
=- { ( A - B R , ' B T P , ) T P ~ ( A - B R ~ - ' B T P , ) + P , B R c I r 2 R R ~ - ' B T P ,
the poles of the closed-loop system, Le., the eigenvalues of the matrix A 01
+ BF are located within the specified disk D . +r2HrH+(a2-r2)Pc}. (22)
Remark: We may also prove Theorem 2 from the following. For the
discrete system The algorithm to find the positive definite solution of (15). which is
similar to the one in (4), is given as follows. Let X, (i = 1, 2, --,n ) be an
eigenvalue of the matrix (18), and the corresponding eigenvector be v,.
Then the following is yielded:
it is well known that the following control law
+ B ( r 2 R + B T P B ) - ' B 7AP ,- Q r ) u,=O (23)

and from (15) we can obtain

is the optimal control such that the quadratic criterion function (A-ffOT
PUi = ~ PX,u,+HTHui.

Defining ui as
I -Lpvi
is minimized and all the closed-loop poles, Le., the eigenvalues of the
matrix (23) and (24) may be rewritten as

are located within the unit disk with the center at the origin (3). It is easily
found that the eigenvalues of the matrix
[ -H7H
'1 [:,]
I
=X, [ Z Br2R-'B7

A-B(r2R+BTPB)-'BTP(A-aT) (18)'
By solving the above generalized eigenvalue eigenvector problem for
eigenvalues to be less than 1, P may be given by
are located in the specified disk D.
P = [ u , ,u2, ---,U,l[Ul, v2, ----,u,l-'. (26)
Thus. the theorem means that the state feedback control law (14)
stabilizes the closed-loop system with poles located in the specified disk.
111. D-STABILITYMARGIN
Remark: For the continuous system, let the control given by (14)
denote The advantages of the D-pole assignment presented in the previous
section are not only that fewer parameters are used in the specification for
u=Fx
the control law in comparing to assignment of all poles, but also that the
then it may be the optimal control minimizing the criterion function uncertainty of the plant model: which may lead to poles outside of the disk
D,can be treated similartothe stability margin. To present the gain
J= sr (llxll&+ l l 4 l ~ Jdt (19)
margin and the phase margin for all poles staying in a specified disk D,
instead of the control law (14), the following control, as shown in Fig. 2,
is considered to be supplied to the plant.
for certain Qc and R,. Therefore, F = -R,BTPc, where PC is the
u = -K(r2R+BTPB)-IBTP(A-aI)x (27)
positive definite symmetric solution of
P KFx (27)'
ATP,+P,A+Qc-PcBR~lBrPc=O (20)
where
which yields
K=diagk 2{ k, l , -----,
k,"}. (28)
( A + B F ) T P , + P c ( A + B F ) + F T R c F -0
= -c.
Then we are led to the following lemma which means that the closed-loop
system has the robustness for parameter perturbation as shown in Fig. 2 in
On the other hand, (15) may be rewritten as view of the above statement.
Lemma 3: Suppose that there is a positive definite solution satisfying
(15) with R = diag { T I , TI, -------.
F,,,}. Then the gain margin for D-pole
assignment is given by

+ F ' r 2 R F + r 2 H T H + ( a 2 - r 2 ) P ] + F T R , F(15)' I 1
-<k,$-
1 +a, - 1 -a,
where a < 0. Those two equations imply that the weighting matrix Qc of where = 2, ____,
pn
(19) is given by

1
Qr= - ;
{(A+BF)TP(A+BF)+F7r2RF+r2HTH
where X , is a maximum eigenvalue of the matrix BTPB.
+ ( a Z - r 2 ) P } - F T R c F(21) Proo$ As the general gain margin of the discrete optimal control,
426 IEEE TRAKSACTIOh'S OK AUTO!vlATIC CONTROL. VOL. AC-32. S O . 5 . MAY 1987

1 X

II n I

Fig. 2. Closed-loop system employingthe pole assignment feedback law in the presence
of unknown gain { k , } .

(29) is already given by Safonov ( 5 ) . But here a different proof can be

given as follows: (15) can be rewritten as Fig. 3. Variation of closed-loop poles for perturbation k : r = 2, o = -6, Q= I , R = I .

( A - L Y I + B K F()A
~-aI+BKF) (A-aI)T
P= +HTH+-
r ' r For the disk D with radius r = 2 and center at - 6 + j 0 , the D-pole
. PB(r2R+BTPB)-'{(r2R+BTPB)K+KT(r'R+BTPB) assignment has been done for the several cases.
I ) Case 1: R = 0.1, Q = I
(A - CXZ)
-KTBTPBK-(r2R+BTPB)}(rZR+BTPB)-'BTP-

In the above equation (31), as far as the following:

. (31) F= -
[
0.140
28.56
6.313
-28.56

X,(A + BF) : { -5.344?jO.O16. -5.451

0.024
10.69
1 ED
(r'R+BTPB)K+KT(r'R+BTPB)-KTBTPBK-(r'R+BTPB)zO
where X,(A + B F ) denotes eigenvalues of ( A + BF).
(32) 2) Case2: R = I , Q = diag { 100, 100, 100)

is satisfied, we may see that the closed-loop system can be found D-pole
assigned. Let the maximum eigenvalue of the matrix BTPB be X, then
the left-hand side of (32) may be rewritten as
2.433
31.14
4.4556
-31.14

X , ( A + B F ) : (-5.344,-5.827,-5.891
0.406
11.17 1 E D.
(r'R+h,I)K+KT(r?R+X,I)-X,KTK-(rZR+X,I)
+ ( K - I ) T ( X , , , I - B T P B ) ( K - I ) . (33) 3) Case 3: R = diag { 1 0 0 , loo}, Q = I

1
Since in (33) the following inequality:
6.427 2.49~
(K-I)T(X,,I-BTPB)(K-I)zO 28.45 -28.45 10.67

is satisfied, the rest of the terms are need to be considered. From R = hi(A + BF) : { - 5.333 kj2.078 x - 5.43} E D.
diag { Fl, f ~ ---
, F m } , and (33), it is known that the k,'s satisfying the
followjing: 4) Case 4: R = diag { 1000, lOOO}, Q = I

or
2(r'i,+X,,,)k,-X,kT-(r'i,+Xm)~0 (34) F= -
[
1.48x 10-4
28.44
6.43
-28.44
2.49~
10.67
1
Xi(A+BF): { - 5 . 3 3 + j 6 . 5 9 ~ 1 0 - ' , - 5 . 4 3 ) E D.
(34) '
Thus. it follows that the proposed algorithm gives D-pole assignment
give the inequality (32). since A,,, > 0 from the positive definity of P. and the pole locations within the specified disk D are dependent on the
Therefore, it follows that k, satisfying (34) is given by (29). matrices Q, R of the Riccati equation.
Remark: We may see that in (32) ai is going to approach to zero as r + In the case of the system (36) with perturbation K such as Fig. 2, to
0, which means that the gain margin for D-pole assignment becomes illustrate the variation of the closed-loop poles. as an example we take
smaller as r + 0.This case means that all poles of the closed-loop system Case 1) of the above D-pole assignment which has D-stability margin as
are assigned on a same point, so its robustness may be quit weak. 0.93 5 k, 5 I .08 and Q; 5 1.47". i = 1, 2. Fig. 3 shows the variations
Thus. in the above lemma. it is shown that the gain margin for D-pole of the closed-loop system poles for parameter perturbation K = k l where
assignment can be evaluated by (29). Similarly. w e may also see that the k is a scalar. As we see in Fig. 3 . it may be not ensured that the closed-
phase margin for D-pole assignment is obtained as loop poles are located \vithin a specified disk for perturbation beyond the
gain margin obtained by D-pole assignment.

V. CONCLUSIONS

IV. EXAMPLES This technical note presents an algorithm to assign all poles of the
closed-loop system in the disk D with the center on the real axis of the
To illustrate D-pole assignment, we take an example as the following: complex plane and arbitrary radius. Instead of locating all the poles of the
closed-loop system. the parameters of the desired disk are just needed to
be specified. and the pole locations within the specified disk D are only
dependent on the matrixes Q, R of the discrete type Riccati equation.
(36) The proposed pole assignment is said D-pole assignment, and its gain
and phase margin for the D-pole assignment are presented.
 IEEE TRANSACTIOKS ON AUTOMATIC CONTROL, VOL. AC-32. NO.5 . MAY 1987 427

REFERENCES
F. Fallside. Ed.. Control System Design by Pole-ZeroAssignment. New
n. SYSTEMS AND DEHNITIONS
York: Academic.1977.
A. V. Bogachev. V . V . Grigorev. V . N . Drozdov.and A . N. Korovyakov,
We consider a h e a r system described by the state equations
"Analytic design of controls from root indicators," Automat., Remote Contr.,
vol.40, no. 8, pp. 1118-1123. 1979. X(t)=Ax(t)+Bu(t)+Dw(t);
B. C. Kuo, Digital Confrol Systems. New York: Holt, finehart, snd Winston,
1980.
T. Pappas, T. A. J. h u b . and N. R. Sandell. "On the numerical solution of the
discrete-time algebraic Riccati equation," IEEE Trans. Automat. Contr., vol.
AC-25, no. 4. pp. 631-641, 1980.
M . G. Safonov, Stability and Robustness of Multivariable Feedback Sys-

tems. Cambridge. M A : M.I.T. Press, 1980.

where x(t) E B nis the state, ~ ( tE) am
is the control, w(t) E Rp is the
Y. Mori and Y . Shimemura, "On a design method for feedback control law to disturbance, y ( t ) E F I r is the measured output, and z ( t ) E W is the
locate the eigenvalues in a specified region," SICE (in Japanese), vol. 16. no. 3. controlled output.
Short paper, pp. 462463. 1980. The standard H" optimization problem is concerned with constructing
S. Barnrtl. Pol?'norniuls and Linear Control Systems. New York:Marcel
a dynamic feedback compensator u = F(s)yto minimize the H mnorm of
Dekker.1980.
the transfer function from w to z; e.g., see [2]. In this note, we consider
the special case in which the full state canbe measured; i.e., C = I. This
motivates us to restrict attention to constant state feedback controllers of
the form I( = Fx. The problem of extending our results to the case in
which the full state cannot be measured is the topic of current research.
One idea being pursued is the use of an observer to reconstruct the state
Disturbance Attenuation and H" Optimization: variables; see [3].
A Design Method Based on the The first problem considered in this note is that of reducing the effect of
Algebraic Riccati Equation the disturbances to a prespecified level. Hence, we introduce the
following definition.
Definition 2. I: Let theconstant y > 0 be given. The system (E)is said
IAN R. PETERSEN to be stabilizable with disturbance attenuation y if there exists a state
feedback matrix F E R n x msuch that the following conditions are
Absrruct-This note presents a method for designing a state feedback satisfied.
control law to reduce the effect of disturbances on the oufput of a given 1) The matrix A = A + BF is a stability matrix. That is, all of the
linear system. The problem under consideration involves attenuating the eigenvalues of A lie in the open left-half plane.
disturbancesto a prespecified level. The construction of thestate 2) The transfer function matrix
feedback control law requires the solution of a certain algebraic Riccati
equation. By applying the procedure with successively smaller values of
the prespecified disturbance level, the result also provides an approach to
the H" optimization problem for the state feedback case. satisfies the bound'

I. INTRODUCTION

When designing a control system, one often begins with a plant which for all E 12.That is, the H" norm of H(s) is less than or equal to y.
(J

is subject to external disturbances. A common design objective is to We now present a condition which can be used to test whether a given
reduce the effect of these disturbances to an acceptable level. For system is stabilizable with disturbance attenuation y. In order to test this
example. in the "disturbance decoupling problem" described in [l], state condition, the designer must specify two positive-definite matrices Q and
feedback control is used to ensure that the disturbances are completely R. It willbe shown in the sequel that the question of disturbance
decoupled from the output. However, for some systems, it may be attenuation is independent of the choice of Q and A .
impossible to reduce the effect of the disturbances below a certain Condition I: Let Q E B n X n and R E B m x m be given positive-definite
threshold value. Hence, the disturbance decoupling problem would have matrices and let the constant y > 0 be prespecified. Then the system (E)
no solution in this case. In this note. we present a procedure for designing is said to satisfy Condition 1 (with attenuation constant y) if there
a stabilizing state feedback control which will reduce the effect of the exists an E > 0 such that the Riccati equation
disturbances to a prespecified level.
The results of this note can also be applied to a class of H aoptimization 1 1 1
A ' P + P A - - PBR-'B'P+-PDD'P+-E'E+EQ=O (2.1)
problems in which the effect of the disturbances is minimized in a certain E Y Y
sense.Indeed, by applying our procedure with successively smaller
values of the prespecified disturbance level, we can construct a state has a positive-definite solution P.
feedback control which is arbitrarily close to the H" optimum. Thus, the Remark: Riccati equation (2.1) is of a type which also arises in linear
results of this note are closely related to recent results on the N" quadratic differential games. Furthermore, the process of decreasing E
optimization problem; see [2]. However, previous results onthe H" corresponds to a linear quadratic differential game with cheap control for
optimization problem have led to quite complicated design procedures. In the minimizing player; see [4]. Thus, we will see that there is a connection
contrast, the design procedure proposed in this note is a simple scheme between the H" optimization problem and linear quadratic differential
basedonthe algebraic Riccati equation. It follows that the algebraic games with cheap control.
Riccati equation which was previously associated with LQG optimization Theorem 2. I: Suppose that there existpositive-definite matrices Q
also plays a role in H" optimization. E W X n , R E L P and " ~a ~ constant e > 0 such that Riccati equation
(2.1) has a positive-definite solution. Then,givenany positive-
ManuscriptreceivedOctober 21. 1985;revised h'o\,ember 10, 1986. Paperrecom- definite matrices Q E P x n andl? E R m x mthere , existsa constant E*
mended by Past Associate Editor. S . P. Bhattacharyya.
The author is with theDepartment of Electrical and Electronic Engineering, University
of NewSouthWales.AustralianDefenceForceAcademy.Campbell,ACT 2600, ' Given two matrices N,1 i 4 E P " " , the notation N 2 M refers to the fact that the
Australia. -
matrix N M is positive semi-definite. Similarly, the notationN > M refers to the fact
IEEE Log Number 8613845. that the matrix h'- M is positivedefinite.

0018-9286/87/0500-07$01.00 0 1987 IEEE