EE 501 (SUPPLEMENTARY NOTES)

COMPUTATION OF THE ) ( A R AND ) ( A N

Our aim is to determine basis sets for ) ( A R and ) (A N , where A is an n m real matrix.
The analysis is similar in the complex case.

Elementary Column Operations (e.c.o.)

Let A be an n m matrix. The three elementary column operations are:
i. Multiplication of a column by a nonzero real number.
ii. Adding the product of column j by a real number to column i , where j i .
iii. Interchanging two columns.

Theorem: If B is obtained from A by a sequence of e.c.o., then A can be obtained from
B by a sequence of e.c.o.

Proof: It is sufficient to show this for each of the three e.c.o. Let [ ]
n
a a a A K
2 1
= , where
m
i
R a are columns of A.
i. Let c be a nonzero real number. Multiply
i
a by c . Now,
i
ca is the i -th column
of B . A can be obtained from B by multiplying the i -th column by c c 1
1
=

.
ii. Multiply the j -th column of A by c and add it to column i . Now,
j i
ca a + is
the i -th column of B . To obtain A back, subtract the product of the j -th
column of B (i.e.
j
a ) by c , from i -th column of B .
iii. If columns
i
a and
j
a are interchanged to obtain B , interchange the same two
columns to obtain A back.

Theorem: Any of the three e.c.o. defined above corresponds to multiplying A from right
by an n n nonsingular matrix E where E is obtained by performing the corresponding
e.c.o. on the unit matrix
n
I .
Proof: Exercise.

Examples:

(

1 0
0
22 21
12 11
22 21
12 11
22 21
12 11
c
a a
a a
a ca
a ca
a a
a a
(First e.c.o.)

(

0 1
1 0
22 21
12 11
21 22
11 12
22 21
12 11
a a
a a
a a
a a
a a
a a
(Third e.c.o.)

(

+
+

1 0
1
22 21
12 11
21 22 21
11 12 11
22 21
12 11
c
a a
a a
ca a a
ca a a
a a
a a
(Second e.c.o.)

We therefore see that if B is obtained from A by a sequence of e.c.o., then
AK E E AE B
k
= = K
2 1
where
k
E E E K K
2 1
= and 0 det K .
i
E is obtained by
performing the i -th e.c.o. on
n
I . Clearly,
1
i
E corresponds to an e.c.o. performed on
n
I
as well.

Definition: If B is obtained from A by a sequence of e.c.o., then B is said to be column
equivalent to A.

Theorem: If B is column equivalent to A, then ) ( ) ( A R B R = .
Proof: Let ) ( A R y . Then, there exists an
n
R x such that y Ax = . Since
1
= BK A ,
x BK y
1
= . Let x K x
1 ~
= . So, x B y
~
= , which means that ) (B R y . So ) ( ) ( B R A R .
Now, let ) (B R y . Then, there exists an
n
R x such that y x B = . AK B = and let
x K x =
~
. Then, ) ( A R y . So, ) ( ) ( A R B R .

Remark: By using the theorem given above, we can simplify a matrix A by e.c.o. and
obtain a matrix B which is column equivalent to A. Then we can find a basis for ) (B R ,
which also turns out to be a basis for ) ( A R .

The Algorithm to find a basis for ) ( A R :
1. Set 1 = i .
2. Find the smallest k such that
i l a
a
lk
ik
< =

all for 0
0

If such a k does not exist, go to step 4.
3. For all k j > such that
i l a
a
lj
ij
< =

all for 0 and

0

multiply column k by
ik ij
a a and add to column j .
4. If m i = , stop. Otherwise, let 1 + i i and go to step 2.

As a result of this algorithm, the matrix A is transformed into A by e.c.o. where each
column of A is either zero or if
i
k is the first non-zero entry (counting from the top) of
column i , then
j i
k k when j i .

The matrix A looks like

(
(
(
(
(
(

=
M K
K M
M
M
M
M
K M
K
K
x
x
x
A
0
0 0
0 0 0
0 0 0 0


By reordering columns of A , we can have:

n k k k
r
< < < K
2 1
1

where r is the number of non-zero columns of A . Then { }
r
a a K ,
1
constitute a basis for
) ( ) ( A R A R = . Prove this last statement as exercise.

Example:

A A =
(
(
(
(

(
(
(
(

(
(
(
(

(
(
(
(

=
0 0 4 1 1
0 0 0 1 2
0 0 0 0 1
0 0 0 0 1
5 6 4 1 1
0 0 0 1 2
0 0 0 0 1
0 0 0 0 1
4 4 4 1 1
1 2 0 1 2
0 0 0 0 1
0 0 0 0 1
7 5 6 2 1
7 4 4 1 2
3 1 2 1 1
3 1 2 1 1


So,

(
(
(
(

(
(
(
(

(
(
(
(

4
0
0
0
1
1
0
0
1
2
1
1
, , is a basis for ) ( ) ( A R A R = .

Elementary Row Operations (e.r.o.)

The three elementary row operations are:

i. Multiplication of a row by a nonzero real number.
ii. Adding the product of row j by a real number to row i , where j i .
iii. Interchanging two rows.

Theorem: If B is obtained from A by a sequence of e.r.o., then A can be obtained from
B by a sequence of e.r.o.
Proof: Exercise

Theorem: Let B be obtained from A by a sequence of e.r.o., then KA B = where
1 1
E E E K
k k
K

= is a non-singular m m matrix and each

i
E is obtained by performing
the i -th e.r.o. on the m m identity matrix
m
I .
Proof: Exercise.

Definition: If B is obtained from A by a sequence of e.r.o., then B is said to be row
equivalent to A.

Theorem: If B is row equivalent to A, then ) ( ) ( A N B N = .
Proof:
Let ) ( A N x . Then, 0 = Ax , and 0 = = Bx KAx . Hence ) (B N x .
Next, let ) (B N x . Then, 0 = Bx , and 0
1
= =

Ax Bx K . So, ) ( A N x .

The Algorithm to find a Basis for ) ( A N

I. First phase (e.r.o.)
1. Set 1 = i .
2. Find the smallest k such that
i j a
a
kj
ki
< =

all for 0
0

If such a k does not exist, go to step 4.
5. For all k j such that 0
ji
a , multiply row k by
ki ji
a a and add to row j .
6. If n i = , stop. Otherwise, let 1 + i i and go to step 2.

Note that this portion of the algorithm resembles the algorithm for finding a basis for
) ( A R . At this point, the rows are either all zeros or in the form
x x x K K 0 0 0

II. Second phase (Reordering the rows)
1. Interchange the rows so that all the zero rows lie below the nonzero rows. Define
integers
r
k k k , , , K
2 1
as follows:
i
k is the location of the first non-zero entry of
the i -th row.
2. The rows are ordered such that
r
k k k < < < K
2 1


The resulting matrix A (which is row equivalent to A) has the following properties:
i. Rows m r r , , , K 2 1 + + have only zero elements.
ii. All elements to the left of ) , (
i
k i are zero for all r i , , , K 2 1 = .
iii. All elements above ) , (
i
k i are zero for all r i , , , K 2 1 = .
Define the index set { }
r
k k k K K , ,
2 1
= , where n k k k
r
< < < K
2 1
1 .
Define also the index set { } K n L = K , ,2 1 as { }
r n
l l l L

= K , ,
2 1

So, L is the set of integers from 1 to n which are not in K . Order
i
l s such that
n l l l
r n
< < <

K
2 1
1 . In other words { } n L K K U , ,2 1 = .




III. The basis for ) ( A N
Solve the system of equations 0 = x A r n times, each time by setting successively all
components of x with indices in L equal to zero except one component which is set
equal to one. Then, solve for the components of x with indices in K .
Let
i
l
x be the solution obtained when the
i
l -th component is set equal to 1 and other
j
l s
are set to 0, i.e.

(
(
(
(
(

=
i
i
i
i
l n
l
l
l
u
u
u
x
,
,
,
M
2
1
, where 1 =
i i
l l
u
,
and 0 =
i j
l l
u
,
for j i .

and
i j
l k
u
,
s are obtained from 0 =
i
l
Ax . In this way, r n solutions { }
r n
l l l
x x x

K , ,
2 1
are
obtained. This set is a basis for ) ( ) ( A N A N = .

Example:

A A =
(
(
(
(

(
(
(
(

(
(
(
(

(
(
(
(

=
0 0 0 0
0 0 0 0
2 1 2 0
2 0 1
0 0 0 0
0 0 0 0
2 1 2 0
1 1 1 1
10 5 10 0
4 2 4 0
2 1 2 0
1 1 1 1
13 2 7 3
3 3 5 1
4 1 0 2
1 1 1 1
2
1

1
1
= k , 2
2
= k and 3
1
= l , 4
2
= l

(
(
(
(

= =
0
1
1
1
1
2
1
3
l
l
l
u
u
x x
,
,
0
1
=
l
Ax
(
(
(
(

0
1
2
1
2
1
3
x

(
(
(
(

= =
1
0
2
2
2
2
1
4
l
l
l
u
u
x x
,
,
0
2
=
l
Ax
(
(
(
(

=
1
0
1
2
4
x

(
(
(
(

(
(
(
(

1
0
1
2
0
1
2
1
2
1
, is a basis for ) ( ) ( A N A N =