You are on page 1of 27

Linear Algebra and its Applications 322 (2001) 141167

www.elsevier.com/locate/laa

Generalized inverse matrix Pad approximation


on the basis of scalar products
Chuanqing Gu
Department of Mathematics, Shanghai University, 99 Qi Xiang Road, Shanghai 200436,
Peoples Republic of China
Received 14 November 1999; accepted 15 July 2000
Submitted by R.A. Brualdi

Abstract
A new type of generalized matrix inverse is used to define the generalized inverse matrix
Pad approximants (GMPA). GMPA is introduced on the basis of scalar product of matrices,
with the form of matrix numerator and scalar denominator. It is different from the existing
matrix Pad approximants in that it does not need multiplication of matrices in the construction
process. Some algebraic properties are discussed. The representations of GMPA are provided
with the following three forms: (i) the explicit determinantal formulas for the denominator scalar polynomials and the numerator matrix polynomials; (ii) -algorithm expression; (iii) Thiele-type continued fraction expression. The equivalence relations above three representations
are proposed. 2001 Elsevier Science Inc. All rights reserved.
AMS classification: 65D05; 41A21; CR: G.1.2
Keywords: Scalar product of matrices; Generalized inverse; Matrix Pad approximation; Algebraic properties; Determinantal formula; Thiele-type continued fraction; -Algorithm

1. Introduction
Let f (z) be a given power series with matrix coefficients, i.e.,
f (z) = c0 + c1 z + c2 z2 + + cn zn + ,
ci = (ci(uv)) Cst ,

z C,

The work is supported by the National Natural Science Foundation of China (19871054).
E-mail address: guchqing@guomai.sh.cn (C. Gu).

0024-3795/01/$ - see front matter  2001 Elsevier Science Inc. All rights reserved.
PII: S 0 0 2 4 - 3 7 9 5 ( 0 0 ) 0 0 2 3 0 - 5

(1.1)

142

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

where Cst consists of all s t matrices with their elements in the complex plan C.
A (right) matrix Pad approximant of f (z) is an expression of the form U (z)V (z)1 ,
such that
f (z)V (z) U (z) = R(z),

(1.2)

where U (z) and V (z) are matrix polynomials of degree at most m and n, respectively,
whose expansion agrees with f (z) up to and including the term zm+n . R(z) in (1.2)
is referred to as the residual of the approximant. A left matrix Pad approximant of
f (z) can be similarly defined.
The definition of a Pad approximant can be made more formal in a variety of
ways. Typically, U (z) and V (z) are s s polynomial matrices, and V (z) is further
restricted by the condition that the constant term, V (0), is invertible (cf. [5,7,27]).
Labahn and Cabay called such approximants matrix Pad fractions, which were consistent with the scalar (p = 1) case (cf. [18]). They introduced and developed the
notion of a matrix power series remainder sequence and its corresponding cofactor
sequence in [25]. An algorithm for constructing these sequences was presented. Xu
and Bultheel considered some possible definitions of matrix Pad approximants for
a power series with rectangular matrix coefficients in [24]. They had to consider
left and right approximants on account of the noncommutativity of the matrix multiplication. Depending on the normalization of the denominator, they defined type I
(constant term is the unit matrix) and type II (by conditions on the leading coefficient) approximants. A uniform approach was given by Beckermann and Labahn for
different concepts of matrix-type Pad approximants, such as descriptions of vector
and matrix Pad approximants along with generalizations of simultaneous and Hermite Pad approximants. In [3], they introduced the definition for a power Hermite
Pad approximant (PHPA) which takes right-hand (left-hand) and rectangular matrix
Pad approximant, matrix Hermite Pad approximant and matrix simultaneous Pad
approxmant as its special case (see [3, Example 2.12.4]).
Various vector rational interpolants were introduced by Graves-Morris (cf. [20
22]). The problem here is to approximate a number of functions by rational functions
with a common denominator. Graves-Morris and Jenkins [22] at first presented an
axiomatic approach which uniquely define vector Pad approximants and established their algebraic structure without reference to matrix-valued C-fractions. It
is important for vector-valued rational approximation problems. However, in contrast to our GMPA approach, Graves-Morris and Roberts extended their approach
from vector Pad approximants to matrix Pad approximants by exploiting an isomorphism between vectors and matrices by means of Clifford algebra representation. By using modified Euclidean and Kronecker algorithms, they established
the interconnections between vector Pad approximation and matrix Pad approximation in [23]. In this respect, Antoulas [1] gave a recursive method of updating the partial realization to include further Markov parameters, so making a major generalization of the generalized Euclidean algorithm. Bultheel and Van Barel [8] formulated a generalization of the Euclidean algorithm to treat the case of

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

143

matrix Pad approximation which reduces to Kroneckers algorithm in the scalar


case.
Motivated by Graves-Morriss Thiele-type vector-valued rational interpolants [20]
and Graves-Morris and Jenkins axiomatic approach to vector-valued rational interpolants [22], we discussed matrix-valued rational interpolants in Thiele-type form
and the determinantal formula form in [10] and bivariate matrix-valued rational
interpolants in Thiele-type form in [11]. We defined the reciprocal of a matrix as
the generalized inverse of the matrix (2.4), which is shown to be successful in matrix
continued fraction interpolation (cf. [1012]), but we do not analyse the basis of the
definition and do not discuss matrix-valued Pad approximation problems, which is
distinct from interpolation in definitions and algorithms.
In this paper, we present an axiomatic definition to the matrix Pad approximants (GMPA), which is an extension and improvement of [22] in the case of
matrices and analyse the connection between the definition and scalar product of
matirces. Some algebraic properties are discussed. The expression of GMPA is of
the form of matrix numerator and scalar denominator. We obtain three efficient algorithms, where the determinantal formulas and Thiele-type formulas are distinct
from interpolation form [10]. As for -algorithm, the interpolation problems [10]
do not deal with it. In this way, the representations of GMPA are provided with
the following three forms: (i) the explicit determinantal formulas for the denominator scalar polynomials and the numerator matrix polynomials (Section 4); (ii)
-algorithm expression (Section 5); (iii) the expressin of convergents of Thiele-type
continued fractions (Section 6). The equivalence relations above three representations are proposed. Uniqueness is discussed in detail in Section 3. Given results
show that arbitrary two GMPA of the same type may only differ by a scalar polynominal factor. The result also holds for vector-valued Pad approximants [22]. An
existence theorem is given in the case of determinantal formulas in Section 4, but
holds for other forms. A simple proof of Wynns identity for GMPA is given in
Section 5.
As compared to the existing matrix Pad approximants (cf. [3,5,78,2325,27]),
GMPA do not need multiplication of matrices in the construction process, and hence,
we do not have to define left-handed and right-handed approximants. It may be useful
in the noncommutativity problems of the matrix multiplication. Second, the existence condition of GMPA is relaxed if only Q(0) of scalar denominator polynomial is
not equal to zero. So, it can be applied to singular matrices. Third, the construction of
GMPA can be simplified in the computation because it only computes the reciprocals
of matrices instead of usual matrix inverse in the case of -algorithm expression and
Thiele-type continued fraction expression. On the other hand, the method of GMPA
possesses the degree constraint (2.8) and divisibility constratint (2.9), which are due
to the construction process of GMPA. The construction constraints imply that our
method does not construct matrix Pad approximants of type [m/n] when n is odd.
Thus, it may not be as effective as the existing matrix Pad approximation in some
cases.

144

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

2. Definition
E , bE Cd . The scalar product of
Let aE = (a1 , a2 , . . . , ad ), bE = (bP
1 , b2 , . . . , bd ), a
d
vector aE and bE is given by aE bE = i=1 ai bi . The following definition is a natural
extension from vector to matrix, which is different from trA of matrix A in the case
of complex matrix.
Definition 2.1. Let A = (aij ), B = (bij ), A, B Cst . The scalar product of matrices A and B is defined by
AB =

t
s X
X

aij bij ,

(2.1)

i=1 j =1

where Euclidean norm


1/2

t
s X
X
|aij |2 .
kAk =

(2.2)

i=1 j =1

From (2.1) and (2.2), it is held


t
s X
X

|aij |2 = kAk2 ,
AA =

(2.3)

i=1 j =1

where A denotes the complex conjugate of A.


Lemma 2.2 is given by Definition 2.1.
Lemma 2.2. Let A, B, D Cst , b C. Then hold:
(i) A B = B A;
(ii) (A + B) D = A D + B D;
(iii) (bA) B = b(A B);
(iv) A A > 0, A = 0 if and only if A A = 0.
On the basis of (2.2) and (2.3), the generalized inverse of matrix A is defined as
A
1
=
, A=
/ 0, A Cst
(2.4)
A1
r =
A
kAk2
and
A
1
=
, A=
/ 0, A Rst ,
(2.5)
A1
r =
A
kAk2
where the generalized inverse A1
r denotes the reciprocal of matrix A.
/ 0 and b R, b =
/ 0. Then hold:
Lemma 2.3. Let A, B Cst , A, B =
1
b
=
A = bC.
A
C

(2.6)

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

145

Proof. As both A, B =
/ 0, it is easy to derive b/A = 1/C from A = bC. By the
left-hand side of (2.6), we get from Definition 2.1 that bA /kAk2 = C /kCk2 . Then
A =

kAk2
C
bkCk2

or

A=

kAk2
C.
bkCk2

Thus,
kAk2 = A A =

kAk4 kCk2
.
b2 kCk4

It follows that b 2 = kAk2 /kCk2 , and so A = bC holds. 


/ 0 and b R, b =
/ 0. Then hold:
Lemma 2.4. Let A Cst , A =
1
(i) (A1
r )r = A;
1 1
(ii) (bA)1
r = b Ar .
1
Proof. The result of (ii) is evident. We only prove (i). In fact, suppose (A1
r )r =
1
1
1/Ar = b/B. By (2.6), we have B = bAr = b/A. Using (2.6) again, A = b/B
holds. 

Lemma 2.4 shows that we need not compute each inverse in the construction
process of GMPA in the -algorithm form and Thiele-type continued fraction form.
Definition 2.5. A matrix-valued polynomial N(x) = (a uv (x)) Cst is said to be
of degree m and denoted by N{N(x)} = m, if N{a uv (x)} 6 m for u = 1, 2, . . . , s,
v = 1, 2, . . . , t and N{a uv (x)} = m for some u, v(1 6 u 6 s, 1 6 v 6 t).
Definition 2.6. A GMPA of type [n/2k] for the given power series (1.1) is the
rational function
R(z) = P (z)/Q(z)

(2.7)

defined that P (z) is a matrix polynomial and Q(z) is a real scalar polynomial satisfying:
(i)

N{P (z)} 6 n, N{Q(z)} = 2k,


2

(ii)

Q(z)|kP (z)k ,

(iii)

Q(z)f (z) P (z) = O(zn+1 ),

(2.8)
(2.9)
(2.10)

where P (z) = (p(uv) (z)) Cst , the norm kP (z)k of P (z) is as in (2.2).
/ 0 and B(z) be a real
Lemma 2.7. Let A(z) Cst be a matrix polynomial, A(z) =
scalar polynomial. If using (2.4) to rational function
B(z)A (z)
P (z)
B(z)
=
,
=
2
A(z)
Q(z)
kA(z)k

146

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

then Q(z) is a real polynomial, and holds:


(i) Q(z)|kP (z)k2 ;
(ii) N{Q(z)} = 2k, k N.
Proof. As
kP (z)k2 = B 2 (z)kA (z)k2 = B 2 (z)Q(z),

Q(z) = kA(z)k2 ,

the conclusion is evident. 


Note that if P 0 (z)/Q0 (z) is the irreducible form of P (z)/Q(z) in Lemma 2.7,
may not satisfy divisibility (2.9). Lemma 2.7 explains that the generalized inverse for matrices (2.4) gives rise to the scalar denominator of rational fraction
GMPA which divides the square of the norm of numerator. It is shown that Definition
2.6 for GMPA depends on the properties of scalar product of matrices. Example 2.8
illustrates that (2.4) is efficient in matrix continued fraction interpolation problems
as compared with usual matrix inverse.
Suppose the interpolation point set U = {zi = xi , i = 0, 1, . . . , n; xi R}. By
making use of (2.4), we recrusively defined the nth convergent of Thiele-type continued fractions in [10,12]:
x xn1
x x0
Rn(0) (x) = B0 (x0 ) +
B1 (x0 x1 ) + + Bn (x0 x1 xn )
with
P 0 (z)/Q0 (z)

B0 (xi ) = A(xi ), i = 0, 1, . . . , n,
B1 (x0 x1 ) = (x1 x0 )/(B0 (x1 ) B0 (x0 )),

(2.11)

Bl (x0 x1 xl ) = (xl xl1 )/(Bl1 (x0 xl2 xl )


Bl1 (x0 xl1 )), l > 2.
Example 2.8. Using algorithm (2.11), find a rational interpolant R2 (z) = P (z)/
Q(z) for the data






2
0
1 0
0 i
, A(z1 ) =
, A(z2 ) =
A(z0 ) =
0 i
1 i
1 0
at points z0 = 1, z1 = 0, z2 = 1.
Solution 1. Using the generalized inverse (2.4), we get that


z
z+1
2
0



+ 
R2 (z)=
0 i
1 1
0 + 1 17 12i
1 2i
2i
6
11 5

 2
1
12z(z + 1)i
5z 12z + 7
=
11z2 + 6z + 7 5z2 + 12z + 7 (13z2 + 6z + 7)i
and find that R2 (zi ) = A(zi ), i = 0, 1, 2.

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

147

Solution 2. Using usual inverse A1 = adj A/det A, we get that



2
R2 (z)=
0


z+1
z
0
 

+
i
1
0 +
5 4i
i/2 2i
1
2i


2
1
4z(z + 1)i
7z + 4z + 3
= 2
z + 6z + 3 (z + 1)(3 + z) (z2 2z + 3)i

and also find that R2 (zi ) = A(zi ), i = 0, 1, 2.


But for higher order matrices the method of Solution 1 is superior to the method
of Solution 2 in matrix continued fraction interpolation problems.

3. Uniqueness and algebraic properties


We discuss uniqueness and some algebraic properties of GMPA in this section.
Lemma 3.1. Let (Pi (z), Qi (z)) be two different GMPA of type [n/2k], i = 1, 2.
Then hold
P1 (z)Q2 (z) = P2 (z)Q1 (z).

(3.1)

Proof. From Definition 2.6, we get that


U (z) = P1 (z)Q2 (z) P2 (z)Q1 (z) = (Q2 (z) Q1 (z))O(zn+1 ),

(3.2)

N{U } 6 n + 2k,

(3.3)

then
order{U } > n + 1.

From (2.9) it follows that


kU (z)k2 =kP1 (z)k2 Q22 (z) + kP2 (z)k2 Q21 (z)

Q1 (z)Q2 (z)(P1 (z) P2 (z) + P1 (z) P2 (z))


=Q1 (z)Q2 (z)Q(z)

for some real scalar polynomial Q(z). From (3.2) and (3.3), we find that
N{Q} 6 2n,

order{Q} > 2n + 2,

which are contradictory unless U (z) = 0. 


Let
Mn,2k = {(P , Q): P /Q is the GMPA of type [n/2k]},
Nn,2k = {(P , Q): P /Q satisfies (2.10) of GMPA}

148

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

and
u = min{N{P }: (P , Q) Mn,2k or (P , Q) Nn,2k },
v = min{N{Q}: (P , Q) Mn,2k or (P , Q) Nn,2k }.
Note that (P , Q) Nn,2k means that Q|kP k2 and degree conditions (2.8) may
not hold as compared with (P , Q) Mn,2k .
Lemma 3.2.
Nn,2k so that
Mn,2k or (P , Q)
(i) There exists unique (P , Q)
= v.
N{P } = u, N{Q}

(3.4)

(ii) For any (P , Q) Mn,2k , there exists a scalar polynomial (z) so that

P (z) = (z)P (z),


Q(z) = (z)Q(z).

(3.5)

Proof. By definition of u and v, it is known that there exists (Pi (z), Qi (z)) Mn,2k,
or (Pi (z), Qi (z)) Nn,2k , i = 1, 2, so that
N{P1 } = u,

N{Q1 } > v,

N{P2 } > u,

N{Q2 } = v.

From Lemma 3.1, it follows that


N{P1 } + N{Q2 } = N{P2 } + N{Q1 }.
Then, N{P2 } = u, N{Q1 } = v. So (i) holds.
For any (P , Q) Mn,2k there exists some scalar polynomial i (i = 1, 2) so that
/ 0),
(3.6)
P = 1 P + 1 , N{1 } 6 N{P }, (P =

Q = 2 Q + 2 , N{2 } < N{Q}.


From P Q = P Q by Lemma 3.1, it is derived from (3.6) and (3.7) that

(1 2 )P Q = 2 P 1 Q.

(3.7)
(3.8)

By coefficient comparison from both terms of (3.8), we obtained that


1 = 2 = .
Thus
1 = P P ,

2 = Q Q.

(3.9)

Using (2.10) in (3.9), it follows from


P ) = O(zn+1 )
2 f 1 = (Qf P ) (Qf
that (1 , 2 ) Mn,2k or (1 , 2 ) Nn,2k . By the definition of u and v, we find that
1 = 2 = 0. So (3.5) holds. 
Lemma 3.2 implies that for any (P , Q) Mn,2k , they may only differ by a sca Therefore, R(z) = P (z)/Q(z) is
lar polynomial factor between (P , Q) and (P , Q).
unique in the sense.

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

149

Theorem 3.3 (Uniqueness). If R(z) = P (z)/Q(z) is a GMPA of type [n/2k], then it


is unique.
Let (P , Q) Mn,2k be as [n/2k]f for the given power series (1.1).
Example 3.4. Let
f (z) = ezA =

1
0

 
0
0
+
1
0



1
0
z+
2
0


1 2
z +
2

(3.10)

Find [2/2]f , where




0 1
A=
.
0 2
Solution. By the determinantal formulas of GMPA (see (4.1) and (4.2) in Section 4),
we get [2/2]f = P2 (z)/Q2 (z), where


z+1
z
= P (z),
(3.11)
P2 (z) = 25(z + 1)
0
1z

Q2 (z) = 25(z + 1)2 = Q(z),

= 25(z + 1).

(3.12)

N2,2 . In fact, Q(z)


= z + 1,
Note that (P2 (z), Q2 (z)) M2,2 , but (P (z), Q(z))
2
2
2

kP (z)k = 3z + 2, Q(z)|kP (z)k does not hold.


Theorem 3.5. Let R(z) = P (z)/Q(z) be a [n/2k]f and if the coefficients in the
power series (1.1) satisfy
ci = ciT ,

i = 0, 1, 2 . . . ,

(3.13)

where T denotes the transposition of a matrix, then R(z) = R T (z).


Proof. From Definition (2.6), Q(z)f (z) P (z) = O(zn+1 ), then
Q(z)f T (z) P T (z) = O(zn+1 ).
By (3.13), we can derive that

f T (z)

(3.14)

= f (z). Thus, it is known from (3.14) that

Q(z)f (z) P T (z) = O(zn+1 ).


The divisibility condition, that Q(z)|kP T (z)k2 is easily verified because of Q(z)|
kP (z)k2 . The degree condition is evident. So we obtain [n/2k]f (z) = R(z) =
P T (z)/Q(z). By Theorem 3.3, R T (z) = R(z) holds. 
/ 0, where
Theorem 3.6. Let [n/2k]f (z) = P (z)/Q(z) and g(z) = fr1 (z), f (0) =
z R for (1.1), P (0) =
/ 0, and fr1 (z) is defined as (2.4). Then
[n + 2k/2n]g (z) = Q(z)P (z)/kP (z)k2 .

(3.15)

150

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

Proof. By Definition 2.6, Q(z)f (z) P (z) = O(zn+1 ). According to the usual
multiplication of matrices, we have
Q(z)P (z)f (z) P (z)P (z) = O(zn+1 ).
From the condition f (0) =
/ 0, it is known that
that

(3.16)

fr1 (z)

exists. From (3.16), we get

P (z)P (z)fr1 (z) Q(z)P (z) = O(zn+2k+1 ).


That is
fr1 (z) Q(z)P (z)/kP (z)k2 = O(zn+2k+1 ).
So we have
kP (z)k2 g(z) Q(z)P (z) = O(zn+2k+1 ).
The divisibility condition kP (z)k2 |kQ(z)P (z)k2 is easily verified. As for the degree, we find that
N{Q(z)P (z)} 6 n + 2k, N{kP (z)k2 } = 2n.

/ 0. Find
Example 3.7. Let g(z) = f (z)1
r in the power series (3.10). Then f (0) =
[4/4]g .
Solution. By (3.11) and (3.12), we get that
Q(z)g = kP (z)k2 = 625(z + 1)2 (3z2 + 1),


z+1
z
,
P (z)g = Q(z)P (z) = 625(z + 1)3
0
1z
where P (z)g , Q(z)g satisfy:
(i) Q(z)g g(z) P (z)g = O(z5 ),
(ii) N{P (z)g } = 4, N{Q(z)g } = 4,
(iii) kP (z)g k2 = 625(z + 1)3 Q(z)g , Q(z)g |kP (z)g k2 .
So [4/4]g = P (z)g /Q(z)g .
Theorem 3.8. Let f (z) be given by (1.1), z R and
"
#
m1
X
m
i
ci z = cm + cm+1 z + cm+2 z2 +
f (z)
g(z) = z
i=0

If
m > 1,

n m > 2k 1,

then [n m/2k]g (z) exists and


(
[n m/2k]g (z) = z

[n/2k]f (z)

(3.17)
m1
X
i=0

)
ci z

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

151

Proof. Let [n/2k]f (z) = P (z)/Q(z). By Definition 2.6,


Q(z)f (z) P (z) = O(zn+1 ).

(3.18)

By virtue of (3.17) and (3.18), we derive that


Q(z)

m1
X

ci zi P (z) =Q(z)f (z) P (z) Q(z)


=O(z

) + O(z ) = O(z ).
m

Define
P1 (z) = P (z) Q(z)

m1
X

ci z i

i=m
m

i=0
n+1

!
ci z

zm .

i=0

Then by (3.17),
N{P1 (z)} 6 n m.
From (3.18), we obtain that
Q(z) f (z)

m1
X

ci z i

!
P (z) Q(z)

i=0

m1
X

!
ci z i

= O(zn+1 ).

(3.19)

i=0

Multiplying zm to the both sides of (3.19), we find that


Q(z)g(z) P1 (z) = O(znm+1 ).
By virtue of

!2
m1

X
ci zi Q(z)
kP1 (z)k2 = kP1 (z)k2 + Q2 (z)

i=0
!
#)
"
m1
m1
X
X
i

i
ci z + P (z)
ci z
z2m
P (z)
i=0

i=0

and Q(z)|kP (z)k2 , we obtain Q(z)|kP1 (z)k2 . Hence [n m/2k]g (z) exists and
(
)
m1
X
m
i
ci z .
[n/2k]f (z)

[n m/2k]g (z) = z
i=0

4. Determinantal formulas of GMPA


Theorem 4.1. Let R(z) = P (z)/Q(z) be a GMPA of type [n/2k] for the given power series (1.1). Then hold:

152

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

L01
0
L21
..
.

L02
L12
0
..
.

..
.

L0,2k1
L1,2k1
L2,2k1
..
.

L2k1,1

L2k1,2

z2k1

z2k2

0
L10
L20
..
.

Q(z) = det

L2k1,0
z2k

L0,2k
L1,2k
L2,2k
..
.

(4.1)

L2k1,2k
1

and
P (z)

L01
0
L21
..
.

L02
L12
0
..
.

..
.

Ln1,1
1
P
ci zi+n1

Ln1,2
2
P
ci zi+n2

i=0

i=0

0
L10
L20
..
.

= det

Ln1,0

c0 z n

L0,n1
L1,n1
L2,n1
..
.
0
n1
P

ci zi+1

i=0

L0,n
L1,n
L2,n
..
.

Ln1,n

P
ci z i
i=0

(4.2)
where
Lij =

jX
i1

cl+i+n2k+1 cjl+n2k

l=0

t
s X
X
u=1 v=1

Lij = Lij ,
where cl =

(uv)
(cl )

jX
i1

(uv)
(uv)
cl+i+n2k+1 cj l+n2k ,

j >i

(4.3)

l=0

j < i,

(4.4)

is the complex conjugate matrix of cl .

The proof of (4.1) was given by Chuanqing [13], which is an extension of that
of Graves-Morris and Jenkins [22], from the vector case to the matrix case, but the
proof of (4.2) is at first given.
Proof. (i) n = 2k. We express (P (z), Q(z)) as
Q(z) = Q0 + Q1 z + + Q2k z2k ,

(4.5)

P (z) = P0 + P1 z + + Pn zn , Pi Cst

(4.6)

and define the Maclaurin section of f (z) by


Gn (z) = [f (z)]n0 .

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

153

Then, Definition 2.6 implies that


P (z) = [Q(z)f (z)]n0 = [Gn (z)Q(z)]n0 .
Definitions (2.8) and (2.9) imply that
2n 2k at most, and so that
i2n+1
h
= 0.
kP (z)k2 /Q(z)

kP (z)k2 /Q(z)

(4.7)
is a polynomial of degree

(4.8)

2n2k+1

Define a scalar polynomial F (z) as follows:


F (z)=(P (z) Gn (z)Q(z)) (P (z) Gn (z)Q(z))
=kP (z)k2 + Q2 (z)kGn (z)k2
Q(z)(P (z) Gn (z) + Gn (z) P (z)).

(4.9)

According to Definitions (2.9) and (2.10), we derive that F (z)/Q(z) = O(z2n+2 ),


which leads to

2n+1
F (z)/Q(z) 2n2k+1 = 0.
From (4.9) and (4.8), we obtain that
i2n+1
h
P (z) Gn (z) Gn (z) P (z) + Q(z)kGn (z)k2

2n2k+1

= 0.

(4.10)

By (4.7) we find that (4.10) represents 2k + 1(n = 2k) linear equations for Q0 , Q1 ,
. . . , Q2k of (4.5), which may be expressed as
2k
X

Lij Q2kj = 0,

i = 0, 1, . . . , 2k 1,

(4.11)

j =0
2k
X

L2k,j Q2kj = 0,

(4.12)

j =0

where the coefficients of Q2kj in (4.11) are Lij , as given by (4.3) and (4.4).
Eqs. (4.11) and (4.5) form a system of 2k + 1 non-homogeneous equations for
Q0 , Q1 , . . . , Q2k , as expressed by

L02
L0,2k1
L0,2k
0
L01

Q2k
L10
0
L

L
L
12
1,2k1
1,2k
Q2k1

L21
0
L2,2k1
L2,2k
L20
Q2k2

.
.
..
..
..
..
..
..
..
.

.
.
.
.

1
L2k1,0 L2k1,1 L2k1,2
0
L2k1,2k

Q0
z2k
z2k1
z2k2

z
1

154

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

..
.
.

Q(z)

(4.13)

Solving (4.13), we obtain (4.1).


From (4.6), (4.7) and n = 2k, we derive that

n
X
P (z) =c0 Q0 + (c1 Q0 + c0 Q1 )z + +
cnj Qj zn
n
X

!
ci zi Q0 +

i=0

n
X

j =0

!
ci zi+1 Q1

i=0

+ + (c0 z

n1

+ c1 zn )Qn1 + (c0 zn )Qn .

(4.14)

Using (4.13) in (4.14), we obtain (4.2) in the case of n = 2k.


(ii) n 6 2k.
Define
Di = 0,

i = 0, 1, . . . , 2k n 1;

Di = ci2k+n ,

i = 2k n, 2k n + 1, . . . , 2k.

By means of the coefficients {Di | i = 0, 1, . . . , 2k}, we construct Q(z) as in (4.1)


and P [2k/2k] (z) of the type [2k/2k] as in (4.2). The numerator polynomial is defined
by
P (z) = zn2k P [2k/2k] (z).

(4.15)

Then, P (z)/Q(z) is the [n/2k]f required.


(iii) n > 2k. Let

X
ci zi+2kn .
f(z) =
i=n2k

By means of the coefficient {cn2k+i }, i = 0, 1, . . . , 2k, we construct Q(z) as in


(4.1), and P[2k/2k] (z) of type [2k/2k] as in (4.2). The numerator polynomial is defined by
P (z) = zn2k P[2k/2k] (z) + Q(z)

n2k1
X
i=0

Then, P (z)/Q(z) is the [n/2k]f required. 

ci z i .

(4.16)

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

Example 4.2 [26]. Let


1
1 0
f (z) = 0 1 + 1
1
0 0

1
0
0 z + 0
1
0

155

0
0 z 2 .
0

Find [2/2]f .
Solution. By (4.1) and (4.2), we get

0 3 4
Q(z) = det 3 0 2 = 3(2z2 4z + 3),
z2 z 1

0
3
4

0
2
P (z) =det 3
2
2
2
c0 z + c1 z
c0 + c1 z + c2 z
c0 z

2
0
z z+3
=3 z(3 4z) 2z2 4z + 3 ,
z(3 z)
0
where (P (z), Q(z)) M2,2 satisfy:
(i) N{P } = 2, N{Q} = 2,
(ii) kP k2 = 27(3z2 + 2)Q, Q|kP k2 ,
(iii) Q(z)f (z) P (z) = O(z3 ).
To calculate higher-order determinantal formulas (4.1) and (4.2), we introduce
Cayleys theorem.
Lemma 4.3 (see [19]). Let A be a square matrix of even dimension. Let R, C denote the anti-symmetric matrices formed by altering only the rth row, column of A,
respectively. Then
det A = Pf R Pf C,
where Pf R denotes the Pfaffian formula of R.
According to (4.17), we obtain the following result in [14] (also see [19]).
Theorem 4.4. Define Pfaffian formulas, respectively,

z2k
L12 L13 L1,2k+1

L23 L2,2k+1 z2k1

..
..
..
,
1 (z) = det
.
.
.

z
L2k,2k+1

(4.17)

156

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

L12

2 (z) = det

L13

L1,2k+1

2k
P

zi

2k1
P

i+1
ci z

i=0
.
..

1
P

ci zi+2k1

i=0
2k
c0 z
ci

i=0

L23

L2,2k+1

..

..
.

L2k,2k+1

Then hold:
(i)

1 (z) = QPf (z), Q(z) = QPf (z)QPf (0);

(4.18)

(ii)

2 (z) = P Pf (z), P (z) = P Pf (z)P Pf (0),

(4.19)

where Q(z) is as in (4.1) and P (z) is as in (4.2).


Example 4.5. Find [4/4]f = P (z)/Q(z) for the given power series (1.1).
Solution. Making use of Pfaffian formulas (4.18) for the denominator polynomial of
GMPA and Pfaffian formulas (4.19) for the numerator polynomial of GMPA, we get
that, respectively,

2H12 2H13 + G2 2H14 + 2H23


0
G1

G1
0
G2
2H23
2H24 + G3

2H
G2
0
G3
2H34
Q(z)=det

12

0
G4

2H13 + G2 2H23 G3
z4
z3
z2
z
1
=QPf (z)QPf (0),
where Gi = kci k2 , Hij = ci cj ,

G1 2H12 2H13 + G2

G2
2H23

G3
QPf (z)= det

= det

b
e

c
f
h

d
g
j
k

z4
z3

z2

z
1

2H14 + 2H24
2H24 + G3
2H34
G4

z4
z3

z2

1
1

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

157

=(ah bf + ce) + (de + j a gb)z + (ak gc + df )z2


+ (j c bk hd)z3 + (gh + ek jf )z4 ,
Q (0)=(ah bf + ce)
Pf

and
P (z)

2H12
2H13 + G2 2H14 + 2H23
0
G1

G1
0
G2
2H23
2H24 + G3

2H12
G
0
G
2H34
2
3

= det

2H13 G2
2H
G
0
G
23
3
4

2
3
4

P
P
P
c0 z 4
c0 z 3 + c1 z 4
ci zi+2
ci zi+1
ci z i

i=0

i=0

i=0

=P Pf (z)P Pf (0),

(4.20)

where
P Pf (z)=(ah bf + ce)(c0 + c1 z + c2 z2 + c3 z3 + c4 z4 )
+ (de + j a gb)(c0 z + c1 z2 + c2 z3 + c3 z4 )
+ (ak gc + df )(c0 z2 + c1 z3 + c2 z4 )
+ (j c bk hd)(c0 z3 + c1 z4 ) + (gh + ek jf )c0 z4 ,
Pf
P (0)=(ah bf + ce)c0 .
Notice that P (z) = P Pf (z)P Pf (0) using usual multiplication operation of matrices
in (4.20).
Let
H (0, 2k, 2k 1)

L01
L00
L10
0

L
L21
20
=
.
..
.
.
.
L2k1,0 L2k1,1

L02
L11
0
..
.
L2k1,2

..
.

L0,2k1
L1,2k1
L2,2k1
..
.
0

L0,2k
L1,2k

L2,2k
.

..

L2k1,2k

Theorem 4.6 (Existence). Let [n/2k]f = P (z)/Q(z), where P (z) and Q(z) be given by (4.1) and (4.2), respectively, and n = 2k. Then [n/2k]f exists if and only
if
Q(0) = det H (0, 2k 1, 2k 1) =
/ 0.
Proof. By the construction of Q(z), we derive from (4.13) that

158

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

Q2k
Q2k1

H (0, 2k, 2k 1) ... = 0,

Q1
Q0
then

Q2k

Q2k1

H (0, 2k 1, 2k 1) . = Q0

..

Q1

L0,2k
L1,2k
..
.

(4.21)

L2k1,2k1

/ 0, it means that non-homogeneous


If Q(0) = Q0 = detH (0, 2k 1, 2k 1) =
equations (4.21) exist as a unique solution Q0 , Q1 , . . . , Q2k for Q(z). From (4.14),
it also means that (4.21) exists as a unique solution P0 , P1 , . . . , Pn for P (z) in the
case of n = 2k. Hence, [n/2k]f = P (z)/Q(z) exists.
Let [n/2k]f = P (z)/Q(z) exist, then it implies from (4.21) that
rank H (0, 2k 1, 2k 1) = rank H (0, 2k, 2k 1).

(4.22)

If Q(0) = det H (0, 2k 1, 2k 1) = 0, it holds from (4.22) that rank H (0, 2k,
2k 1) < 2k. Hence, it follows from (4.1) that Q(z) 0, which is contradictory to
definition (2.7) of GMPA. 
Example 4.7 [26]. Let

 
1 0
1/2
w(z) =
+
0 1
0



0
1/4 2
z +
1/4
0


0
z3 .
1/8

Note that [2/2]w = P2 (z)/Q2 (z) does not exist because




0 0
Q2 (0) = det
= 0,
0 0
but [3/2]w = P3 (z)/Q3 (z) exists because


0
5/16
Q3 (0) = det
= 25/256 =
/ 0,
5/16
0
where Q3 (z) is constructed by the coefficients {c1 , c2 , c3 }.

5. -Algorithm expression of GMPA


By making use of the generalized inverse (2.4), we define the matrix valued algorithm by
(j )

1 = 0,

j = 0, 1, 2 . . . ,

(5.1)

C. Gu / Linear Algebra and its Applications 322 (2001) 141167


(j )

0 =

j
X

j = 0, 1, 2, . . . ,

ci z i ,

i=0
(j )
(j +1)
k+1 = k1

159

(j +1)

+ (k

k )1
r ,
(j )

(5.2)
j, k > 0

(5.3)

for the given power series (1.1).


By usual construction, it involves the two-dimensional array called the -table
(see [4]).
Theorem 5.1 (Identification theorem). The matrix-valued -algorithm, as expressed
by (5.1)(5.3) with the generalized inverse (2.4), construct GMPA, is identified by
(j )

2k = [j + 2k/2k]f ,

j, k > 0.

(5.4)

Proof. For zeroth column k = 0, the proof is obvious. From (5.1)(5.3) and (2.4) it
is derived that
(j )

(j +1)

1 = (0

j +1
0 )1
= H0 (z)/zj +1 ,
r = 1/cj +1 z
j

(j )

(5.5)

where H0 (z) = cj+1 /kcj +1 k2 , N{H0 } = 0.


For the second column, by (5.3) it is obtained that
(j )

(j )

2 =

(j )

j
X

1
Ci zi + Cj +1 zj +1 + zj +2 /(Cj1
+2 Cj +1 z).

(5.6)

i=0

It holds from (5.6) that


(j )

2 = O(zj )(z ).

(5.7)

1
2
Let g(z) = Cj1
+2 Cj +1 z. Then N{kg(z)k } = 2 and (5.6) becomes
(j +1)

(j )

2 = (kg(z)k2 0

+ zj +2 g (z))/kg(z)k2 = P2 (z)/Q2 (z),

(5.8)

where Q2 (z) = kg(z)k2 . It is known from (5.6) and (5.7) that


N{P2 } = j + 2,

N{Q2 } = 2.

From
(j +1) 2

kP2 k2 = Q22 k0
we get Q2 (z)|kP2

(z)k2 .

(j +1)

k + Q2 z2(j +1) + Q2 (0

(j +1)

G + 0

G),

Hence, by above discussion, we have shown that

(j )

2 = P2 (z)/Q2 (z) = [j + 2/2]f .


Make the following inductive hypotheses, each true for j = 0, 1, 2, . . . , k > 2:
(i)

2k1 = O(zj 1 ), 2k = O(zj )(z ),

(ii)

2k1 = H2k2 (z)/zj +2k1 ,

(j )

(j )

(j )

(j )

(5.9)
(5.10)

160

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

where
(j )

(j )

H2k2 (z) Cst , N{H2k2} 6 2k 2,


(j )

(iii) 2k = [j + 2k/2k]f .

(5.11)

To prove the theorem, it is necessary to show that (5.9)(5.11) hold with k + 1.


As (i) stands, it is obtained from (5.3) that
(j +1) 1
)r )

2k+1 =O(zj 2 ) + O((2k


(j )

= O(zj 2 ) + O(zj 1 )

=O(zj 1 ) (z ),

(5.12)

(j )
(j )
2k+2 =O(zj +1 ) O((2k+1 )1
r )

= O(z )(z ),
j

so (5.9) is proved for k + 1. As it stands, (i) and (ii) are reduced fractions, suppose
that
(j +1)

2k

(j )

= S(z)/T (z),

2k = U (z)/V (z).

(5.13)

Define the matrix polynomial F (z) by


F (z) = S(z)V (z) U (z)T (z) = zj +2k+1 Fc (z),
we can prove that V (z)T (z)|kFc
N{S} 6 j + 2k + 1,

(z)k2

(5.14)

[13]. Note that

N{T } = 2k,

N{U } 6 j + 2k,

N{V } = 2k.

(5.15)

From (5.14) and (5.15), we get N{Fc } 6 2k, N{V T } = 4k. Then, it is derived that
M(z) = V (z)T (z)/Fc (z) = V (z)T (z)Fc (z)/kFc (z)k2
is a matrix polynomial and NM > 2k. It follows that
(j +1)

(2k

2k )1
r =1/(S(z)/T (z) U (z)/V (z))
(j )

=V (z)T (z)/Fc (z)zj +2k+1


=M(z)/zj +2k+1 .

(5.16)

(j )

Substituting (5.16) into 2k+1 , we get from (5.16) that


(j )

(j )

(j +1)

2k+1 =2k1 + (2k

2k )1
r
(j )

(j +1)

=H2k2 (z)/zj +2k1 + M(z)/zj +2k+1


=H2k (z)/zj +2k+1 ,
(j )

(j )

(5.17)

(j )

where H2k Cst , N{H2k } 6 2k.


Hence (5.10) is proved for k + 1. By means of (5.17) and (5.3), we have

.
(j )
(j +1)
(j +1)
(j )
+ zj +2k+2
H2k (z) zH2k (z) .
2k+2 = 2k

(5.18)

C. Gu / Linear Algebra and its Applications 322 (2001) 141167


(j +1)

Let G(z) = H2k

161

(j )

(z) zH2k (z). Then

N{G} 6 2k + 1.

(5.19)
(j +1)

Let x = be any of 2k zero of T (z) for 2k


duced from
(j +1)

S( )/T ( ) = 2k

(j +2)

= S(t)/T (z) in (5.13), it is de-

(j +2)

(j +1)

( ) = 2k2 ( ) + (2k1 ( ) 2k1 ( ))1


r

that
(j +2)

(j +1)

2k1 ( ) = 2k1 ( ).

(5.20)

Similarly, we can deduce that


(j +1)

(j )

(j +1)

2k+1 ( ) = 2k1 ( ),

(j +2)

2k+1 ( ) = 2k1 ( ).

Then, it follows from (5.20) that


(j +1)

(j )

2k+1 ( ) = 2k+1 ( ).

(5.21)

Substituting (5.21) into (5.18), we have



  (j +1)
(j )
(j )
2k+2 ( )=S( )/ T ( ) + 1 2k+1 ( ) 2k+1( )
=S( )/T ( ) + (j +2k+2) /G( ).

(5.22)

Note that T (0) =


/ 0 by the Existence Theorem 4.6, that is to say, =
/ 0. Hence, it
is known that both terms of the right-hand side of (5.22) have poles at z = . Above
discussion shows that if z = is arbitrary zero of T (z), then holds G( ) = 0. From
(5.15) and (5.19), suppose that
G(z) = T (z)G0 (z),

G0 (z) Cst

and by (5.19), it holds


N{G0 } 6 1.

(5.23)

Then, (5.18) becomes


2k+2 =S(z)/T (z) + zj +2k+2 /G(z)
(j )

=(T (z)kG0 (z)k2 S(z) + zj +2k+2 T (z)G0 (z))/T 2 (z)kG0 (z)k2


=P2k+2 (z)/Q2k+2 (z),

(5.24)

where Q2k+2 = T 2 (z)kG0 (z)k2 . From (5.24),


kP2k+2 k2 =kSk2 kG0 k2 Q2k+2 + z2(j +2k+2)Q2k+2
+ Q2k+2 zj +2k+2 (S G0 + S G0 ),
we get
Q2k+2 (z)|kP2k+2 k2 .

(5.25)

162

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

The reduced form of (5.24) is


(j )

2k+2 =P (z)/Q(z) = (S(z)kG0 (z)k2


+ zj +2k+2 G (z))/T (z)kG0 (z)k2 ,

(5.26)

where Q(z) = T (z)kG0 (z)k2 . It follows that


N{Q} = 2k + 2.

(5.27)

By means of (5.12), it is shown from (5.27) that


(j )

N{P } = N{2k+2 } + N{Q} = j + 2k + 2.

(5.28)

From (5.25)(5.28), it is proved that


(j )

2k+2 = P (z)/Q(z) = [j + 2k + 2/2k + 2]f .

Example 5.2. Let f (z) = c0 + c1 z + c2 z2 be the same as Example 4.2. Find [2/2]f
by -algorithm.
Solution. By (5.1)(5.3) and (2.4),
0(0) = c0 ,
1(0)

0(1) = c0 + c1 z, 0(2) = c0 + c1 z + c2 z2 ,

1 0
1 0
1
1
(1)
1 0 ,
=
1 = 2 0 0 ,
3z 1 0
2z 1 0

2(0)

z2 z + 3
1
z(3 4z)
= 2
2z 4z + 3
z(3 z)

0
2z2 4z + 3 = [2/2]f .
0

Now suppose
(j +1)

N = 2k2 ,

(j 1)

W = 2k

(j )

C = 2k ,

(j +1)

E = 2k

(j 1)

S = 2k+2

as bonafide entries in the -table, and


(j )

nw = 2k1,

(j +1)

ne = 2k1 ,

(j +1)

sw = 2k+1 ,

(j )

se = 2k+1 ,

which occur in column of odd index. According to the - algorithm (4.1)(4.3), they
are located as follows:
N
nw
W

ne
C

sw

E
se

(5.29)

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

163

Lemma 5.3. The identity


1
1
1
(5.30)
(N C)1
r + (S C)r = (E C)r + (W C)r
holds provided that the quantities involved are well defined by using (5.1)(5.3) and
the generalized inverse (2.4).

Proof. As
(nw ne) + (se sw) = (se ne) + (nw sw)

(5.31)

it follows from (5.1)(5.3) that (5.31) holds, then (5.30) holds by (5.29). 
By virtue of Lemma 5.3 and Theorem 5.1, we obtain the following result.
Theorem 5.4. Wynn identify of GMPA
([j + 2k 1/2k]f [j + 2k/2k]f )1
r
+ ([j + 2k + 1/2k]f [j + 2k/2k]f )1
r
= ([j + 2k 1/2k 2]f [j + 2k/2k]f )1
r
+ ([j + 2k + 1/2k + 2]f [j + 2k/2k]f )1
r
holds provided that the GMPAs involved are well defined by using (5.1)(5.3) and
the generalized inverse (2.4).
6. Continued fraction expression of GMPA
By making use of the generalized inverse (2.4), we construct Thiele-type matrixvalued continued fraction as
z
z
z
+
H (z) = B0 +
B1
B2 + + Bn +
with matrix elements Bi Cst , z R. The nth convergent of H (z) is defined by
z
z
z
+
(6.1)
Rn (z) = B0 +
B1
B2 + + Bn
and it is evaluated by backward recursion.
As the result of [16], recursive coefficient algorithm of (6.1) is given as follows
for the given power series (1.1).
Coefficient algorithm:
A0 (z) = f (z),
A1 (z) =
k > 2,

B0 = B0 (0) = A0 (0),
1
(Df (z))r = [Df (z)] /kDf (z)k2 , B1 = B1 (0) = A1 (0),

2
Bk = k(DAk1 (z))1
r |z=0 = k(DAk1 (z)) /kDAk1 (z)k |z=0 ,

Ak (z) = Bk (z) + Ak2 (z),


where D denotes derivation with respect to z.

164

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

Theorem 6.1 (Identification theorem). If zero divisors are not encountered in the
construction of Rn (z) in (6.1) by using coefficient algorithm with the generalized
inverse (2.4), then hold:

[2k/2k]f ,
Rn (z) = [n/2k]f =
[2k + 1/2k]f ,

n = 2k, k = 0, 1, 2, . . . ,
n = 2k + 1, k = 0, 1, 2, . . .

(6.2)

Proof. The proof is recursive. For n = 2k, k = 0, it holds


R0 (z) = B0 = B0 /1 = P0 (z)/Q0 (z) = [0/0]f .
For n = 2k + 1, k = 0, we get that
R1 (z) = B0 +

zB1
B0 kB1 k2 + zB1
z
P1 (z)
,
= B0 +
=
=
2
2
B1
kB1 k
kB1 k
Q1 (z)

(6.3)

where N{P1 } = 1, N{Q1 } = 0, Q1 |kP1 k2 .


Then
R1 (z) = [1/0]f .

(6.4)

For n = 2k, k = 1,
R2 (z) = B0 +

z
z
z
,
= B0 +
B1 + B2
S1 (z)

(6.5)

1 (z). By the proof of (6.4), we have


where S1 (z) = B1 + z/B2 = P1 (z)/Q
N{P1 } = 1,

N{Q 1 } = 0,

Q 1 |kP1 k2 .

Let kP1 k2 = Q 1 g1 , then N{g1 } = 2. We get that


1 P
zQ
B0 g1 + zP1
z
P2 (z)
1
= B0 +
,
=
=
S1 (z)
g1
Q2 (z)
kP1 k2
where Q2 (z) = g1 . Then N{P2 } = 2, N{Q2 } = 2. From
R2 (z) = B0 +

(6.6)

kP2 k2 = kB0 k2 g12 + z2 kP1 k2 + zg1 (B0 P1 + B0 P1 ),


it is held Q2 |kP2 k2 . Hence R2 (z) = [2/2]f .
For n 1 = 2k or n 1 = 2k + 1, k = 0, 1, 2, . . . let
Rn (z)=Pn1 (z)/Qn1 (z) = [n 1/2k]f
(
[2k/2k]f
n = 2k, k = 0, 1, 2, . . .
=
[2k + 1/2k]f n = 2k + 1, k = 0, 1, 2, . . .
For n = 2k, k = 0, 1, 2, . . . let
z
z
z
z
,
+
= B0 +
Rn (z) = B0 +
+

+
B1
B2
Bn
Sn (z)
where

(6.7)

(6.8)

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

z
z
z
Pn1 (z)
.
+
=
B2
B3 + + Bn
Q n1 (z)
Using the inductive hypothesis (6.7), we obtain that
n1 } = 2k, Q
n1 |kPn1 k2 .
N{Pn1 } = n 1 = 2k, N{Q
Sn (z) = B1 +

165

(6.9)

(6.10)

From (6.10), let kPn1 = Q n1 gn1 . Then N{gn1 } = 2k.


Substituting (6.9) into (6.8), we get that
k2

n1 P
zQ
z
n1
= B0 +
Sn (z)
kPn1 k2

B0 gn1 + zPn1
Pn (z)
,
=
=
gn1
Qn (z)

Rn (z)=B0 +

(6.11)

where Qn (z) = gn1 . Then, we obtain from (6.10) that


N{Pn } = n,

N{Qn } = 2k.

(6.12)

For
2

+ z2 kPn1
k2 + zgn1 (B0 Pn1 + B0 Pn1
)
kPn k2 = kB0 k2 gn1

we get
Qn |kPn k2 .

(6.13)

Hence it follows from (6.11)(6.13) that


Rn (z) = Pn (z)/Qn (z) = [n/2k]f = [2k/2k]f ,

n = 2k, k = 0, 1, 2, . . .

As for n = 2k + 1, k = 0, 1, 2, . . . , the proof is similar to the case of n = 2k. 


It is known from Theorem 6.1 that each convergent Rn (z) in (6.1) is a corresponding [n/2k]f and is of the sequences [0/0]f , [1/0]f , [2/2]f , [3/2]f , [4/4]f ,
[5/4]f , . . .
Example 6.2. Let f (z) = C0 + C1 z + C2 z2 be the same as Example 4.2. Find
[2/2]f by the convergent of continued fraction (6.1).
Solution. By coefficient algorithm of (6.1) and using (2.4), we get

1 0
z
z

R2 (z) = 0 1 +
1 0 +
1 0
0 0
1
1
4 0
1 0
3 1 0
2 1 0
2

z z+3
0
1
z(3 4z) 2z2 4z + 3 = [2/2]f ,
= 2
2z 4z + 3
z(3 z)
0

166

where

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

1 0
A0 (z) = f (z), B0 = B0 (0) = A0 (0) = 0 1 ,
0 0

1 + 2z 0
1
1
0 ,
A1 (z) = (Df (z))1
r =
8z2 + 8z + 3 1 + 2z 0

1 0
1
B1 = B1 (0) = A1 (0) = 1 0 ,
3 1 0
B2 = B2 (0) = 2(DA1 (z))1
r |z=0

2(8z2 + 8z + 1)
(8z2 + 8z + 3)2
1

=
8(2z + 1)
4 [(8z2 + 8z + 1) + 8(2z + 1)2 ]
2(8z2 + 8z + 1)

1 0
1
4 0 .
=
2 1 0


0
0
0 z=0

We take notice that [2/2]f in Example 4.2 is the same as that of in Example 5.2
and in Example 6.2. It illustrates that the generalized inverse (2.4), which is on the
basis of the scalar product of matrices (2.1), is successful in the matrix-valued rational approximation and interpolation. On the other hand, Lemma 2.4 shows that we
need not compute each inverse in the construction process of GMPA in some case.

Acknowledgement
The author would like to thank the referees for their corrections and many valuable suggestions.

References
[1] A.C. Antoulas, On recursiveness and related topics in linear systems, IEEE Trans. Automat. Control.
31 (1986) 11211135.
[2] A.C. Antoulas, Rational interpolation and the Euclidean algorithm, Linear Algebra Appl. 108 (1988)
157171.
[3] B. Beckermann, G. Labahn, A uniform approach for the fast computation matrix-type Pad approximants, SIAM J. Matrix Anal. Appl. 15 (1994) 804823.
[4] G.A. Baker, P.R. Graves-Morris, Pad Approximants, Part I & II, Addison-Wesley, Reading, MA,
1981.
[5] N.K. Bose, S. Basu, Theory and recursive computation of 1-D matrix Pad approximants, IEEE
Trans. Circuit Syst. CAS 27 (1980) 322325.

C. Gu / Linear Algebra and its Applications 322 (2001) 141167

167

[6] A. Bultheel, M. Van Barel, Pad techniques for model reduction in linear system theory: a survey, J.
Comput. Appl. Math. 14 (1986) 401438.
[7] A. Bultheel, Recursive algorithms for the matrix Pad table, Math. Comput. 35 (1980) 875892.
[8] A. Bultheel, M. Van Barel, A matrix Euclidean algorithm and the matrix minimal Pad approximation problem, in: C. Brezinski (Ed.), Continued Fractions and Pad approximants, North-Holland,
Amsterdam, 1990, pp. 1151.
[9] S. Cabay, G. Meleshko, A weakly stable algorithm for the Pad approximants and the inversion of
Hankel matrices, SIAM J. Matrix Anal. Appl. 14 (1993) 735765.
[10] G. Chuanqing, Thiele-type and Largrange-type generalized inverse rational interpolation for rectangular complex matrices, Linear Algebra Appl. 295 (1999) 730.
[11] G. Chuanqing, Bivariate Thiele-type matrix valued rational interpolants, J. Comput. Appl. Math. 80
(1997) 7182.
[12] G. Chuanqing, C. Zhibing, Matrix valued rational interpolants and its error formula, Math. Numer.
Sinica. 17 (1995) 7377.
[13] G. Chuanqing, Generalized inverse matrix valued Pad approximants, Numer. Sincia. 19 (1997)
1928.
[14] G. Chuanqing, Pfaffian formula for generalized inverse matrix Pad approximation and application,
J. Numer. Meth. Comput. Appl. 19 (1998) 283289.
[15] G. Chuanqing, Multivariate generalized inverse vector-valued rational interpolants, J. Comput. Appl.
Math. 84 (1997) 137146.
[16] G. Chuanqing, Z. Gongqin, Matrix valued rational approximation, J. Math. Res. Exposition 16
(1996) 301306.
[17] A. Draux, Bibliography-Index, Report AN0-145, Univ. de Sciences et Techniques de Lille, November 1984.
[18] W.B. Gragg, The Pad table and its relation to certain algorithms of numerical anaysis, SIAM Rev.
14 (1972) 161.
[19] P.R. Graves-Morris, G.A. Baker Jr., C.F. Woodcock, Cayleys theorem and its application in the
theory of vector Pad approximants, J. Comput. Appl. Math. 66 (1996) 255265.
[20] P.R. Graves-Morris, Vector valued rational interpolants I, Numer. Math. 41 (1983) 331334.
[21] P.R. Graves-Morris, Vector valued rational II, IMA J. Numer. Anal. 4 (1984) 209224.
[22] P.R. Graves-Morris, C.D. Jenkins, Vector valued rational interpolants III, Constr. Approx. 2 (1986)
263289.
[23] P.R. Graves-Morris, D.E. Roberts, From matrix to vector Pad approximants, J. Comput. Appl. Math.
51 (1994) 205236.
[24] X. Guoliang, A. Bultheel, Matrix Pad approximation: definitions and properties, Linear Algebra
Appl. 137&138 (1990) 67136.
[25] G. Labahn, S. Cabay, Matrix Pad fractions and their computation, SIAM J.Comput. 4 (1989) 639
657.
[26] C. Pestano-Gabino, C. Gonzalez-Concepcion, Rationality minimality and uniqueness of representation of matrix formal power series, J. Comput. Appl. Math. 94 (1998) 2338.
[27] Y. Starkand, Explicit formulars for matrix-valued Pad approximants, J. Comput. Appl. Math. 5
(1979) 6365.