RANK FACTORIZATION AND

MOORE-PENROSE INVERSE

and Predrag S. Stanimirovic

Gradimir V. Milovanovic
Abstract. In this paper we develop a few representations of the MoorePenrose inverse, based on full-rank factorizations of matrices. These representations we divide into the two dierent classes: methods which arise
from the known block decompositions and determinantal representation. In
particular cases we obtain several known results.

1. Introduction
The set of m n complex matrices of rank r is denoted by Cmn
= {X
r
C
: rank(X) = r}. With A|r and A|r we denote the rst r columns of
A and the rst r rows of A, respectively. The identity matrix of the order k
is denoted by Ik , and O denotes the zero block of an appropriate dimensions.
mn

We use the following useful expression for the Moore-Penrose generalized

inverse A , based on the full-rank factorization A = P Q of A [1-2]:
A = Q P = Q (QQ )1 (P P )1 P = Q (P AQ )1 P .
We restate main known block decompositions [7], [16-18]. For a given
there exist regular matrices R, G, permutation matrices
matrix A Cmn
r
E, F and unitary matrices U , V , such that:
1991 Mathematics Subject Classication. 15A09, 15A24.
Typeset by AMS-TEX

G. V. Milovanovic and P. S. Stanimirovic


Ir O
= N1 ,
(T1 ) RAG =
(T2 )
O O


I K
(T3 ) RAF = r
= N3 ,
(T4 )
O O


Ir O
= N1 ,
(T5 ) U AG =
(T6 )
O O


B O
(T7 ) U AV =
= N2 ,
(T8 )
O O


B O
= N6 ,
(T9 ) EAV =
K O


A11 A12
= N7 , where
(T10 ) EAF =
A21 A22


B
RAG =
O

I
EAG = r
K

I
RAV = r
O

B
U AF =
O


O
= N2 ,
O

O
= N4 ,
O

O
= N1 ,
O

K
= N5 ,
O

rank(A11 )=rank(A).

(T11 ) Transformation of similarity for square matrices [11]:





Ir K
T1 T2
1
1
1
F R =
.
RAR = RAF F R =
O O
O O
The block form (T10 ) can be expressed in two dierent ways:


A11 T
A11
, where the multipliers S and T satisfy
(T10a ) EAF =
SA11 SA11 T
T = A1
S = A21 A1
11 A12 ,
11 (see [8]);
 


A11
A11 A12
A12
=
(see [9]).
(T10b ) EAF =
A21 A22
A21 A21 A1
11 A12
Block representations of the Moore-Penrose inverse is investigated in [8],
[11], [1519]. In [16], [18] the results are obtained by solving the equations
(1)(4). In [15], [8] the corresponding representations are obtained using the
block decompositions (T10a ) and (T10b ) and implied full-rank factorizations.
Also, in [19] is introduced block representation of the Moore-Penrose inverse, based on A = A T A , where T A AA {1}.
Block decomposition (T11 ) is investigated in [11], but only for square
matrices and the group inverse.
The notion determinantal representation of the Moore-Penrose inverse of
A means representation of elements of A in terms of minors of A. Determinantal representation of the Moore-Penrose inverse is examined in [12],
[46], [1214]. For the sake of completeness, we restate here several notations and the main result. For an m n matrix A let = {1 , . . . , r }
and = {1 , . . . , r } be subsets of {1, . . . , m} and {1, . . . , n}, respectively.

RANK FACTORIZATION AND MOORE-PENROSE INVERSE

 ... 
1
r
Then A 1 ... r = | A
| denotes the minor of A determined by the rows
indexed by and the columns indexed by , and A
represents the corresponding submatrix. Also, the algebraic complement of A
is dened by





1 ... p1 p+1 ... r

1 ... p1 i p+1 ... r
p+q
| A
A 1 ... q1 q+1 ... r .
| = Aij 1 ... q1 j q+1 ... r = (1)
aij
Adjoint matrix of a square matrix B is denoted by adj(B), and its determinant by |B|.
Determinantal representation of full rank matrices is introuced in [1], and
for full-rank matrices in [46]. In [1214] is introduced an elegant derivation
for determinantal representation of the Moore-Penrose inverse, using a fullrank factorization and known results for full-rank matrices. Main result of
these papers is:
Proposition 1.1. The (i, j)th element of the Moore-Penrose inverse G =
(gij ) of A Cmn
is given by
r
 

A 
| A
|

a
ji
:j ; :i

gij =
.

| A | | A |
,

In this paper we investigate two dierent representations of the MoorePenrose inverse. The rst class of representations is a continuation of the
papers [8] and [15]. In other words, from the presented block factorizations of
matrices nd corresponding full-rank decompositions A = P Q, and then apply A = Q (P AQ )1 P . In the second representation, A is represented
in terms of minors of the matrix A. In this paper we describe an elegant
proof of the well-known determinantal representation of the Moore-Penrose
inverse. Main advantages of described block representations are their simply
derivation and computation and possibility of natural generalization. Determinantal representation of the Moore-Penrose inverse can be implemented
only for small dimensions of matrices (n 10).
2. Block representation
Theorem 2.1. The Moore-Penrose inverse of a given matrix A Cmn
r
can be represented as follows, where block representations (Gi ) correspond to
the block decompositions (Ti ), i {1, . . . , 9, 10a, 10b, 11}:
 1|r   1 1  1|r 

R
R
A G |r
,
(G1 ) A = G1 |r

G. V. Milovanovic and P. S. Stanimirovic


 1|r   1 1  1|r 
R
R
A = G1 |r
B A G |r
B ,





1 


Ir
Ir
1|r
1|r
AF
,
R
(G3 ) A = F
R
K
K



1
[ Ir , K ] E,
[ Ir , K ] EA G1 |r
(G4 ) A = G1 |r

1


(G5 ) A = G1 |r
U|r A G1 |r
U|r ,


1 

|r
|r
(G6 ) A = V |r R1
AV |r
,
R1

1
B U|r ,
(G7 ) A = V |r B U|r AV |r
 

 1
B
B
U|r AF
(G8 ) A = F
U|r ,
K
K

1
(G9 ) A = V |r [ B , K ] EAV |r
[ B , K ] E,




1
Ir
Ir

A11 [ Ir , S ] E
(G10a ) A = F
A11 [ Ir , S ] EAF
T
T


Ir

1
=F
[ Ir , S ] E,
(Ir + T T )1 A1
11 (Ir + S S)
T
 

 1
A11
A

(A11) [A11 , A21 ]EAF 11

(G10b ) A = F
(A11)1[A11 , A21 ]E
A12
A12
 
A
1
1
(A11 A11 +A12 A12) A11 (A11 A11 +A21 A21) [A11 , A21 ]E,
= F 11
A12




1


|r
|r
Ir
Ir


1

(G11 ) A = R
R1 T1 .
T1 AR
R
1
1
T1 T2
T1 T2
(G2 )

Proof. (G1 )

Starting from (T1 ), we obtain



 
O
1 Ir
1
1 Ir
A=R
G =R
[ Ir , O ] G1 ,
O O
O

which implies
P =R
Now, we get


1

Ir
O

|r

= R1 ,

Q = [ Ir , O ] G1 = G1 |r .

 1|r   1 1  1|r 

A = Q (P AQ )1 P = G1 |r
A G |r
.
R
R

RANK FACTORIZATION AND MOORE-PENROSE INVERSE

The other block decompositions can be obtained in a similar way.

(G5 )

Block decomposition (T5 ) implies



 
O
Ir
1
Ir
A=U
G =U
[ Ir , O ] G1 ,
O O
O

which means

P =U
(G7 )

Ir
O

|r

= U ,

Q = [ Ir , O ] G1 = G1 |r .

It is easy to see that (T7 ) implies



 
B O
B
A = U
V = U
[ Ir , O ] V .
O O
O

Thus,


P =U


|r
B
= U B,
O

Q = [ Ir , O ] V = V |r ,

which means
P = B U|r ,

Q = V |r .

(G10a )

From (T10a ) we obtain



 
A11 T
A11

Ir
A=E
F =E
A11 [ Ir , T ] F ,
SA11 SA11 T
S

which implies, for example, the following full rank factorization of A:

 
Ir
P =E
A11 , Q = [ Ir , T ] F .
S
Now,

I ,
A = F r
T

A11

[ Ir , S ] EAF

Ir ,
T

1

The proof can be completed using

 
I
EAF = r A11 [ Ir , T ] .
S

A11 [ Ir , S ] E.

Remarks 2.1. (i) A convenient method for nding the matrices S, T and A1
11 ,
required in (T10a ) was introduced in [8], and it was based on the following
extended Gauss-Jordan transformation:

G. V. Milovanovic and P. S. Stanimirovic

A11
A21

A12
A22



I
I T

O
O O


A1
11
.
S

(ii) In [3]
full-rank factorization of A, derived
 it was used the following

A11
A12
from A =
:
A21 A21 A1
11 A12


A11
P =
,
Q = [ Ir , A1
11 A12 ] .
A21
3. Determinantal representation
In the following denition are generalized concepts of determinant, algebraic complement, adjoint matrix and determinantal representation of generalized inverses. (see also [14].)
Denition 3.1. Let A be m n matrix of rank r.
(i) The generalized determinant of A, denoted by Nr (A), is equal to
  
Nr (A) =
A | A | ,
,

(ii) Generalized algebraic complement of A corresponding to aij is

 

A 
Aij =
| A
|.

aji
:j;:i
(iii) Generalized adjoint matrix of A, denoted by adj (A) is the matrix
whose elements are Aij .
For full-rank matrix A the following results can be proved:
Lemma 3.1. [14] If A is an m n matrix of full-rank, then:

| AA | , r = m
(i) Nr (A) =
| R A | , r = n.

(A adj(AA ))ij , r = m

(ii) Aij =
( adj(A A)A )ij , r = n.

A (AA )1 , r = m
(iii) A =
(A A)1 A , r = n.

A adj(AA ), r = m

(iv) adj (A) =

adj(A A)A , r = n.
Main properties of the generalized adjoint matrix , generalized algebraic
complement and generalized determinant are investigated in [14].

RANK FACTORIZATION AND MOORE-PENROSE INVERSE

Lemma 3.2. [14] If A = P Q is a full-rank factorization of an m n matrix

A of rank r, then
(i) adj (Q) adj (P ) = adj (A);
(ii) Nr (Q) Nr (P ) = Nr (P ) Nr (Q) = Nr (A);
From Lemma 3.1 and Lemma 3.2 we obtain an elegant proof for the
determinantal representation of the Moore-Penrose inverse.
Theorem 3.1. Let A be an m n matrix of rank r, and A = P Q be its fullrank factorization. The Moore-Penrose inverse of A posseses the followig
determinantal representation:
 

A 
| A
|

a
ji
:j ; :i


aij =
.

| A | | A |
,

Proof. Using A = Q P [3] and the results of Lemma 3.1 and Lemma 3.2,
we obtain:
Q adj(QQ ) adj(P P )P
A = Q P = Q (QQ )1 (P P )1 P =
=
| QQ |
| P P |
adj (Q) adj (P )
adj (A)
=
=
.

Nr (Q)Nr (P )
Nr (A)
4. Examples
Example 4.1. Block decomposition (T1 ) can be obtained by applying transformation (T3 ) two times:


I K
R1 AF1 = r
= N3 ,
O O


Ir O
T
= N1 .
R2 N3 F2 =
O O
Then, the regular matrices R, G can be computed as follows:
N1 = N1T = F2T N3 R2T

1
For the matrix A = 11
1

= F2T R1 AF1 R2T R = F2T R1 ,

1
3 5
7
0 2 0
4
1 1 5 15 we obtain
2
4 10 18

G = F1 R2T .

G. V. Milovanovic and P. S. Stanimirovic

0
1
1
1
R1 = 1
2
2 1

1
0
1
0
R2 = 2
1
0
5
4 11
From R = R1 , G = R2T , we get

1 1
|2
0
R1 = 1
1 1 ,
1 2
Using formula (G1 ), we obtain

169
6720

128

781
A =
13440

128
197
13440

0
0
1
0
0
0
1
0
0

0
0, F = I ,
1
5
0
1

0 0
0 0
0 0 , F2 = I4 .
1 0
0 1

G1 |2 =


0 2 0
4 .
1
1 5 11

1
0

67
2240

233
6720

137
6720

1
128

1
128

1
128

330
4480

1037
13440

5
128

5
128

653
13440 .

5
128

151
4480

709
13440

59
13440

1
0
1
2
1
1
0
1
0 1
1
3

Example 4.2. For the matrix A =

0
1 1 3 we obtain
1 1
0
1
1
0 1 2

1 1




1 0
1 , T = 1 2 [5].
A1
S = 1
11 = 1 1 ,
1 3
0 1
1
0
Using (G10a ) we obtain
5
3
34 17
4
13

102
51

A =
5
7
102
102
1
1
34
17

1
34

1
34

3
17

5
102

5
102

13
102

1
51

1
51

5
102

3
34

3
34

1
34

5
34

4
51

.
7
102
1
17

RANK FACTORIZATION AND MOORE-PENROSE INVERSE

Example 4.3. For the

1
0
1
1

R = 1
1
2 1

0
0
1
0

4 1
1
2
2
0 1 we obtain
matrix A = 2
6 3
1
3
10
4 2 5

0

 

0 , T = 1 0 , T = 1 2 , F = I .
1
2
4
1 3
01
0
1

Then, the following results can be obtained:



4 6





|r
Ir
1
0 1
1
3

=
R1 T1 = 1
0
1 1 1 , R
1 1 .
T11 T2
2 3
Finally, using (G11 ), we get

8
81

47

162
A =
7
54
4
81

10
81

2
81

79
162

16
81

11
54

2
27

5
81

1
81

2
27

5
54

.
1
18
1
27

References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]

Arghiriade, E. e Dragomir, A., Une nouvelle denition de linverse generalisee dune

matrice, Rendiconti dei Lincei, serir XIII Sc. s. mat. e nat. 35 (1963), 158165.
Bapat, R.B.; Bhaskara, K.P.S. and Manjunatha Prasad, K., Generalized inverses
over integral domains, Linear Algebra Appl. 140 (1990), 181196.
Ben-Israel, A. and Grevile, T.N.E., Generalized inverses: Theory and applications,
Wiley-Interscience, New York, 1974.
Gabriel, R., Das Verallgemeinerte Inverse einer Matrix u
ber einem beliebigen K
orper - analytish betrachtet, J. Rewie Ansew Math. 244(V) (1968), 8393.
Gabriel, R., Extinderea complementilor algebrici generalizati la matrici oarecare,
Studii si Cercetari Matematice 17 -Nr. 10 (1965), 15661581.
Gabriel, R., Das allgemeine element des Verallgemeinerte Inverse von Moore-Penrose, Rev. Roum. Math. Pures Appl. 37 No 6 (1982), 689702.
Lancaster, P. and Tismenetsky, M., The Theory of Matrices with Applications,
Academic Press, 1985.
Noble, P., A methods for computing the generalized inverse of a matrix, SIAM J.
Numer. Anal. 3 (1966), 582584.
Penrose, R., A generalized inverse for matrices, Proc. Cambridge Philos. Soc. 51
(1955), 406413.
Penrose R., On best approximate solution of linear matrix equations, Proc. Cambridge Philos. Soc. 52 (1956), 1719.

10
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]

G. V. Milovanovic and P. S. Stanimirovic

Robert, P., On the Group inverse of a linear transformation, J. Math. Anal. Appl.
22 (1968), 658669.
Stanimirovic, P. and Stankovic, M., Generalized algebraic complement and MoorePenrose inverse, Filomat 8 (1994), 5764.
Stanimirovic, P. and Stankovic, M., Determinantal representation of weighted Moore-Penrose inverse, Matematicki Vesnik 46 (1994), 4150.
Stanimirovic, P. and Stankovic, M., Determinantal representation of generalized
inverses over integral domains, Filomat 9:3 (1995), 931940.
Tewarson, R.P., A direct method for generalized matrix inversion, SIAM J. Numer.
Anal. 4 No 4 (1967), 499507.
Zielke, G., A survey of generalized matrix inverse, Banach Center Publication 13
(1984), 499526.
Zielke, G., Die Au
osung beliebiger linearer algebraisher Gleichungssysteme durch
Blockzerlegung, Beitr
age zur Numerischen Mathematik 8 (1980), 181199.
Zielke, G., Motivation und Darstellung von verallgemeinerten Matrixinversen, Beitr
age zur Numerischen Mathematik 7 (1979), 177218.
Zlobec, S., An explicit form of the Moore-Penrose inverse of an arbitrary complex
matrix, SIAM Rev. 12 (1970), 132134.

Department of Mathematics, Faculty of Electronic Engineering, P.O.Box

73, Beogradska 14, 18000 Ni
s
University of Ni
s, Faculty of Philosophy, Departement of Mathematics,

Cirila
i Metodija 2, 18000 Ni
s