Вы находитесь на странице: 1из 16

Partitioned Matrix

Dr. Jeff Chak-Fu WONG


Department of Mathematics
Chinese University of Hong Kong

jwongmath. uhk.edu.hk

MAT581SS
Mathematics for Logistics
Produced by Jeff Chak-Fu WONG

M ATRIX O PERATIONS : PAR TITIONING

M ATRIX O PERATIONS : PARTITIONING

Sometimes is it useful to divide a matrix into submatrices,


especially if there is some logical reason to distinguish some
rows and columns from others.
This is known as partitioning the matrix; the submatrices are
usually separated by dashed or dotted lines.
E.g., we might create four submatrices from a 4 4 matrix A

a11

a21

A=
a31

a41

M ATRIX O PERATIONS : PARTITIONING

a12
a22

a13
a23

a14
a24

a32

a33

a34

a42

a43

a44

A11
=

A21

A12

A22

If two matrices are partitioned so that submatrices are


conformable for multiplication, then products of partitioned
matrices can be found as products of their submatrices.
E.g., suppose that in conjunction with the 4 4 matrix A above
we have a 4 3 matrix B

b11 b12 b13


b21 b22 b23

B11 B12
B=
=
b31 b32 b33
B21 B22

b41 b42 b43


Here
B11 and B21 are 2 2 matrices and
B12 and B22 are 2 1 matrices (column vectors).

M ATRIX O PERATIONS : PARTITIONING

Then the product AB, which will be a 4 3 matrix, can be found


as products of the submatrices in A and B, because they have
been formed so that all submatrices are conformable both for
the multiplications and also for the summations in which they
appear.
Here

AB =

A11
A21

A12
A22

B11
B21

A11 B11 + A12 B21


A21 B11 + A22 B21

B12
B22

A11 B12 + A12 B22

A21 B12 + A22 B22

(1)

and one can check that all conformability requirements are


met.

M ATRIX O PERATIONS : PARTITIONING

T HE I NVERSE OF A PAR TITIONED M ATRIX

HE I NVERSE OF A

PARTITIONED M ATRIX

The elements of an inverse matrix can also be found in blocks,


using partitioning.
Consider again a 4 4 matrix A partitioned into four 2 2
submatrices

A=

A11

A21

A12
A22

The partitioning has been carried out so that the two


on-diagonal submatrices, A11 and A22 , are square (both 2 2
matrices in this example).

HE I NVERSE OF A

PARTITIONED M ATRIX

Denote A1 by R (for reciprocal), and assume that it is


partitioned in the same way, into four 2 2 submatrices:

R11 R12

R=
R21 R22
Then the defining relation for the inverse, AA1 = AR = I4 , is

R
A
A12
R12
11
11
= I4
(2)
A21 A22
R21 R22

HE I NVERSE OF A

PARTITIONED M ATRIX

The multiplication on the left can be carried out following (1) for
the product of any pair of conformably partitioned matrices;
the 4 4 identity matrix can similarly be partitioned into

I4 =
0

HE I NVERSE OF A

PARTITIONED M ATRIX

0
1

0
0


0
I2
=
0
0

I2

A11
A21

A12

A22

R11
R21

R12
R22

I2
0

I2

(2)

Equation (2) represents the following four matrix equations


(i) A11 R11 + A12 R21 = I2

(ii) A21 R11 + A22 R21 = 0

(3)

(i) A11 R12 + A12 R22 = 0

(ii) A21 R12 + A22 R22 = I2

(4)

One should check that all products and terms in (3) and (4)
meet the conformability requirements for multiplication,
addition, and equality.

HE I NVERSE OF A

PARTITIONED M ATRIX

10

The elements in the four A submatrices are known.


The elements in the four R submatrices are what we want to
determine.
Both A22 1 and [A11 A12 (A22 )1 A21 ]1 exist.

HE I NVERSE OF A

PARTITIONED M ATRIX

11

(i) A11 R11 + A12 R21 = I2

(ii) A21 R11 + A22 R21 = 0

(3)

Assume that the inverse for A22 exists and can be found.
Then, form (3)(ii), we obtain
R21 = (A22 )1 (A21 R11 )

(5)

Substituting (5) into (3)(i) gives


A11 R11 A12 (A22 )1 A21 R11 = [A11 A12 (A22 )1 A21 ]R11 = I2
and therefore
R11 = [A11 A12 (A22 )1 A21 ]1

HE I NVERSE OF A

PARTITIONED M ATRIX

(6)

12

(i) A11 R12 + A12 R22 = 0

(ii) A21 R12 + A22 R22 = I2

(4)

From (4)(ii), multiplying by (A22 )1 again


(A22 )1 A21 R12 + R22 = (A22 )1
and so
R22 = (A22 )1 (A22 )1 A21 R12 = (A22 )1 [I2 A21 R12 ] (7)
Substituting (7) into (4)(i), we have
A11 R12 + A12 [(A22 )1 (A22 )1 A21 R12 ] = 0
or
[A11 A12 (A22 )1 A21 ]R12 = A12 (A22 )1
but, using the result in (6),i.e., R11 = [A11 A12 (A22 )1 A21 ]1 ,
this becomes
R12 = R11 A12 (A22 )1
HE I NVERSE OF A

PARTITIONED M ATRIX

(8)
13

Putting all the pieces together, we obtain

[A11 A12 (A22 )1 A21 ]1


R11 A12 (A22 )1
1

A =R=
(A22 )1 (A21 R11 )
(A22 )1 [I A21 R12 ]

HE I NVERSE OF A

PARTITIONED M ATRIX

14

Therefore, from (5) through (8), we see that it is possible to find


all the elements of the inverse matrix R, provided we solve
equations in the correct order.
First we must find (A22 )1 .
Then
1. find R11 using (6), i.e.,
R11 = [A11 A12 (A22 )1 A21 ]1 .
2. use the result for R11 in (5) and (8) to find the elements of the
"off-diagonal" blocks R21 and R12 , i.e.,
R21 = (A22 )1 (A21 R11 ) and R12 = R11 A12 (A22 )1
3. finally, use R12 in (7) to find R22 , i.e.,
R22 = (A22 )1 [I2 A21 R12 ].
HE I NVERSE OF A

PARTITIONED M ATRIX

15

Finding A1 in this way requires finding inverses to two smaller


matrices:
A22 , which is 2 2 in this example, and
[A11 A12 (A22 )1 A21 ], which is also 2 2
and performing several matrix multiplications.
In addition, for A matrices with particular structures, the solution
via (5) through (8) may be especially simple.

HE I NVERSE OF A

PARTITIONED M ATRIX

16

Вам также может понравиться