WHAT IS A MATRIX?

A MATRIX is RECTANGULAR ARRAY OF REAL NUMBERS ARRANGED

IN m ROWS AND n COLUMNS.

1 9 8 5 This is a 4 x 4 matrix.
C= -2 3 -7 6
-8 -5 4 0
7 -6 2 -1

Each number appearing in the array is said to be an ELEMENT, or COMPONENT,

of the matrix. Elements of a matrix are designated using a lowercase form of the same
letter used to symbolize the matrix itself. These letters are subscripted, as aij, to give
the row and column location of the element within the array. The first subscript
always refers to the row location of the element; the second subscript always refers to
its column location. Thus, component aij is the component located at the intersection
of the ith row and the jth column.
The number of rows, m, and the number of columns, n, of the array give its ORDER,
or its DIMENSIONS, m x n (read “m by n”). The abbreviated notation

Amxn or [aij] (mxn)

is often used to represent the matrix and specify its order.

Example 1 The following are examples of matrices.

1 7 This is a 3x2 matrix.

A= 5 3 Element a12 = 7
4 -2 Element a21 = 5

B= 1 2 -1 This is a 2x3 matrix.

3 5 6 Element b13 = -1
Element b23 =6

1 9 8 5 This is a 4 x 4 matrix.
C= -2 3 -7 6 Element c33 = 4
-8 -5 4 0 Element c44 =-1
7 -6 2 -1 Element c31 =-8

Matrices provide a most convenient vehicle for organizing and storing

large quantities of data. Because the basic idea is to organize the data,
we cannot overemphasuze the importance of the location of each
number within the matrix. It is not simply a matter of putting numbers
into rows and columns; each row-column location within each matrix
carries with it a special interpretation.

1
 SPECIAL TYPES OF MATRICES

Matrices that consist of just one row or just one column may be referred to as
VECTORS. More specifically, any m x 1 matrix may be called a COLUMN
VECTOR and any 1 x n matrix may be called a ROW VECTOR.

Example
V = [-1 -6 2]

is a 1 x 3 matrix or a three-component ROW VECTOR.

1
W= 0
3
2

is a 4 x 1 matrix or a four-component COLUMN VECTOR. []

A matrix that has the same number of rows as it has columns is called
a SQUARE MATRIX or an nth-ORDER MATRIX.
A square matrix in which all of the primary diagonal entries are ones
and all of the offdiagonal entries are zeros is called an IDENTITY
MATRIX. The identity matrix, symbolized In, is used in matrix
algebra in much the same way as the number one is used in regular
algebra.

Example 5 Each of the following is an IDENTITY MATRIX.

1 0 1 0 0 1 0 0 0
I2 = 0 1 I3 = 0 1 0 I4 = 0 1 0 0
0 0 1 0 0 1 0
0 0 0 1

Note that each identity matrix is a square matrix. []

A ZERO MATRIX is a matrix that has zero for every entry. The zero
matrix is generally denoted 0m x n and is used in matrix algebra in much
the same way that the number zero is used in regular algebra.
The TRANSPOSE of an m x n matrix A, denoted At, is the n x m
matrix whose rows are the columns in A (in the same order) and whose
columns are the rows in A (in the same order).

Example
1 2 3 1 4
If A = 4 5 6 then At = 2 5
3 6

Note that aijt = aji.

The transpose of a row vector is a column vector and the transpose of a
column vector is a row vector.
Two matrices, A and B, are said to be EQUAL only if they are of the
same dimenisons and if each element in A is identical to its

2
 corresponding element in B; that is, if and only if aij =bij for every pair
of subscripts I and j. If A = B, or if AB, then B  A.

MATRIX ADDITION

Two matrices of the same dimensions are said to be CONFORMABLE FOR

ADDITION. The addition is performed by adding corresponding elements from the
two matrices and entering the result in the same row-column position of a new matrix.

If A and B are two matrices, each of size m xn, then the SUM of A and B is the m x n
matrix C whose elements are
Cij = aij + bij for i = 1,2,…, m
j = 1,2,…, n (3.1)

Example (a) 1 2 + 7 14 = (1 + 7) (2 + 14) = 8 16

-1 0 0 3 (-1 + 0) (0 + 3) -1 3

(b) [5 8 2] + [3 0 1] = [(5 + 3) (8 + 0) (2 + 1)] = [8 8 3]

© 6 8
3 4 + 7 11 12 is not defined because the two matrices
-9 1 0 3 7 are not of the same dimension []

Example Theater-on-the-Square is presenting a special production of Faust.

Ticket sales for the afternoon matinee were as follows: 331 orchestra
seats, 427 loge seats, and 642 balcony seats. For the night
performance, 282 orchestra seat tickets, 438 loge seat tickets, and 834
balxony seat by type, for the two performances.
We set up two vectors as follows:
M = [331 427 642] and N = [282 438 834]
Then, total ticket sales, by type, is given by
S = M + N = [613 865 1,476]

Laws of Matrix Addition The operation of adding two matrices that are
conformable for addintion has these basic ptoperies.
1. A + B = B + A (The commutative law
of matrix addition)
2. (A + B) + C = A + (B + C) (The associativ law of
matrix addition)

MULTIPLICATION BY A CONSTANT

A matrix can be multiplied by a constant by multiplying each composnent in the

matrix by the constant. The result is a new matrix of the same dimensions as the
original matrix.

If K is any real number and A is an m x n matrix, then the product KA is defined to be

the matrix whose components are given by K times the corresponding component of A;
that is,
KA = [kaij](m x n) (3.2)

3
Example (a) If V = [1 2 9], then 5V = [(5 x 1) (5 x 2) (5 x 9)]
= [5 10 45]

(b) If
9 -3
B= 6 1 then
12 4

(1 / 3) x 9 (1 / 3) x –3 3 -1
(1 / 3)B = (1 / 3) x 6 (1 / 3) x 1 = 2 1/3
(1 / 3) x 12 (1 / 3) x 4 4 4/3
[]

MULTIPLICATION OF MATRICES

Let us begin by considering the multiplication of two vectors, one a row vector and
the other a column vector, but each with the same number of components.

The product of an n-component row vector A and an n-component column vector B is

the number C, defined as
C = AB = [a1 b1 + a2 b2 + … + an bn] (3.3)

Example Given
4
A = [1 2 3] and B= 5
6

the product AB = C is
4
C = AB = [1 2 3] . 5 = [(1 x 4) + (2 x 5) + (3 x 6)]
6

= [4 + 10 + 18] = [32] []

Note these important things about the procedure. A ROW IS

MULTIPLIED IN TO A COLUMN. In an orderly progression, the
first element in the in the row is multiplied times the first element in
the column, the second element in the row times the second element in
the column, and so on until the nth row element is multiplied times the
nth column element. These products ate then summed to obtain the
single number that is the product of the two vectors. Clearly, the
procedure can be executed only if the row vector contains the same
number of components as the column vector.

Example Vanassa purchased 6 cases of soft drinks, 18 pounds of hot dogs, 12

packages of hot=dog buns, 3 jars of mustard, and 4 jars of relish. Soft
drinks cost $5 a case, hot dogs cost $2.25 a pound, buns cost $0.90 a

4
 package, mustard $1.90 a jar, and relish $1.75 a jar. Use vector
multiplication to determine the total amount Vanassa spent.
We set up a price vector and a quantity vector (one must be a row
vector and the other a column vector), as
6
18
C = PQ = [5 2.25 0.90 1.90 1.75] 12 = $94.00
3
4
[]
Example Given
4 5
A = [1 2 3] and B= 6 7
8 9

the product C = AB is computed

4 5
C = AB = [1 2 3]. 6 [1 2 3]. 7
8 9

= [(1 x 4 + 2 x 6 + 3 x 8) (1 x 5 + 2 x 7 + 3 x 9)]
= [40 46]
Now let us multiply two m x n matrices. If A and B are two matrices,
the product AB is defined if and only if the number of columns in A is
equal to the number of rows in B. If this requirement is met, A is said
to be CONFORMABLE to B FOR MULTIPLICATION. The
matrix resulting from the multiplication has dimensions equivalent to
the number of rows in A and the number of columns in B. Thus, for
example, a 2 x 3 matrix can be multiplied by a 3 x 4 matrix and the
product matrix will be of dimensions 2 x 4. However, multiplication of
a 3 x 4 matrix by a 2 x 2 matrix is not defined.
The following schematic illustrates the dimensional requirements for
the multiplication C = AB.
A(m x p) B(p x n) = C(m x n)

Must be the same

Dimensions of product matrix

Example A firm manufacturing office furniture finds that it has the following
variable costs, in dollars

Desk Chair Table Cabinet

50 20 15 25 Material
V= 30 15 12 15 Labor
30 15 8 20 Overhead

5
 That is, the manufacture of each desk requires materials costing $50,
labor costing $30, and overhead of $30. Each chaur requires material
costing $20, and so on.
An order for 5 desks, 6 chairs, 4 tables, and 12 cabinets has just been
received. This information is organized in a column vector, as

5 Desks
Q= 6 Chairs
4 Tables
12 Cabinets

The total material, total labor, and total overhead cost associated with
filling this order can be computed as follows:

5
50 20 15 25 6 730 material
C = VQ = 30 15 12 15 4 = 468 Labor
30 15 8 20 12 512 Overhead

Note that the computations made to determine that total material cost is
$730 were as follows:
$50 material per desk x 5 desks = $250
$20 material per chair x 6 chairs = 120
$15 material per table x 4 tables = 60
$25 material per cabinet x 12 cabinets = 300
$730
Similar computations yielded the total labor cost of $468 and the total
overhead cost of $512. []

Example parts explosion problem A manufacturer of microcomputers

produces three final models, I, II, and III, which are made from three
subassemblies, A, B, and C. Furthermore, each sybassembly is made
from parts 1, 2, 3, and 4.

Parts SUBASSEMBLIES MODEL

(1, 2, 3, 4) (A,B,C) (I, II, III)

Model I requires two subassemblies A and two subassemblies C.

Model II requires one each of subassemblies A and C two
subassemblies B. Model III requires four subassemblies B and one
subassembly C. This information can be organized in matrix format, as
A B C
2 0 3 I
S= 1 2 1 II
0 4 1 III

6
 In a similar manner, the number of each different type of part required
for each subassembly can be organized in matrix format. This
information is as follows:

1 2 3 4
1 2 0 1 A
P= 3 1 4 0 B
0 3 1 2 C

The matrix SP, obtained by multiplying the subassembly matrix S

times the parts matrix P, gives the number of units of each part
required in the production of one unit of each of the final products.

1 2 3 4
2 0 3 1 2 0 1 2 13 3 8 I
SP = 1 2 1 x 3 1 4 0 = 7 7 9 3 II
0 4 1 0 3 1 2 12 7 17 2 III

Notice that in multiplying row 1 of S rimes column 1 of P, these

computations are made: [(number of units of subassembly A in one
unit of product I) x (number of units of part 1 in one unit of
subassembly A)] + [(number of units of subassembly B in one unit of
product I) x (number of units of part 1 in one unit of subassembly B)]
+ [(number of units of subassembly C in one unit of product I) x
(number of units of part 1 in one unit of subassembly C)] = number of
units of part 1 in eaqch unit of product 1. In matrix multiplications give
meaningful result in the context of the problem situation.
Other numbers in the SP matrix have a similar meaning. Thusm from
the first row of this matrix we read that each unit of Model I requires 2
units of part 1, 13 units of part 2, 3 units of part 3, and 8 units of part 4.
The entries on the second row of the matrix give the number of units of
each type of part required in the manufacture of one unit of Model II.
The third row gives the number of units of each part required by one
unit of Model III.
Now, the firm has just received an order for six units of Model I, three
units of Model II, and four units of Model III and needs to know the
number of units of each type of part required to fill the order. An order
quantity vector can be estblished as
I II III
Q = [6 3 4]
By multiplying Q times SP, we obtain the desired information, as
follows:
1 2 3 4
I II III 2 13 3 8 I 1 2 3 4
Q(SP) = [6 3 4] 7 7 9 3 II = [81 127 113 65]
12 7 17 2 III

7
 We see, thus, that 81 units of part 1, 127 units of part 2, 113 units of
part 3, and 65 units of part 4 will be required to manufacture the items
included in the order.
The latest price catalog shows that each unit of part 1 costs $4.50, each
unit of part 2 costs $9.00, each unit of part 3 costs $7.25, and each unit
of part 4 costs $12.00. With this information we can compute the total
parts cost for the order. Costs per unit are set up in a vector, as

4.50
C= 9.00
7.25
12.00

Then, total cost is determined by

4.50
T = (QSP)C = [81 127 113 65]. 9.00 = [3106.75]
7.25
12.00

Parts needed to fill the order will have a total cost of $3,106.75. []

INVERSE OF A MATRIX

The material of this section refers to square matrices; that is, matrices which have the
same number of rows as columns. A square matrix with n rows and n columns is said
to be of order n.

UNIT MATRIX

A UNIT (OR IDENTITY) matrix is a square matrix whose diagonal elements from
upper left to lower right are each 1, and whose other elements are all zeros.

Thus, 1 0 0
I= 0 1 0
0 0 1

is a unit matrix of order 3. The unit matrix of order 2 is

I= 1 0
0 1

The most important property of the unit matrix is illustrated by the statements

AI = A and IA = A

That is, the product of any given matrix and the unit matrix is the given matrix itself.
In other words, the unit matrix behaves like the number 1 in ordinary arithmetic
where we say

8
 (a) (1) = (1) (a) = a

The reader may verify the unity property of I by carrying out the following
multiplication in which the first written matrix, I , when multiplied by the second
matrix, yields the second matrix as the product.

1 0 0 2 3 4 2 3 4
0 1 0 -1 -2 0 = -1 -2 0
0 0 1 5 2 -3 5 2 -3

Exercise: Given

A= a b
c d

Verify that AI = A and that IA = A.

ROW OPERATIONS

There are THREE types of row operations:

1. Multiplying (dividing) a row by a non zero constant,

Example: Given the matrix

A= 1 2
3 9

We may multiply the first row by -2 to give another matrix

D= -2 -4
3 9

2. Multiplying (dividing) a row by a non zero constant and adding (subtracting)

the result to (from) another row.

Example: Consider the following two matrices:

E= 1 2 ; F= 1 2
3 9 5 13

Matrix F is obtained from matrix E by multiplying the first row of matrix E by 2 and
adding the result to the second row.

3. Interchanging of any two rows.

9
 Examples: Consider the following two matrices:

A= 2 3 ; B= 4 5
4 5 2 3
Matrix B is obtained from matrix A by Interchanging of first and second rows.

Exercise: What row operation performed on

A= 2 5
6 13
will lead to another matrix which will have a 0 in place of the 6, and what will the
new matrix be?

Answer: Multiply the first row of the given matrix by -3 and add to the second row,
obtaining.
B= 2 5
0 -2

If we have a statement of matrix equality, such as

AB = C

and we perform the same row operations on A and C (not on B and C), we are led to
new statements of equality. For example,

1 2 2 -4 = 0 6
0 3 -1 5 -3 15

is a true statement. If we multiply the first row of the leftmost matrix by 3 and the first
row of the rightmost matrix by 3, we obtain

3 6 2 -4 = 0 18
0 3 -1 5 -3 15

which the reader may verify is also a true statement,. Or, starting over, if we take the
leftmost and the rightmost matrices and add twice the first row to the second row, we
obtain

1 2 2 -4 = 0 6
2 7 -1 5 -3 27

which computation will prove to be a true statement.

10
Exercise: Starting with the original matrices in the foregoing, select the leftmost and
rightmost matrices and multiply the second row by -2 and add to the first. Write the
new statement and verify that it is an equality.

COMPUTATION OF INVERSE OF A MATRIX

The inverse of a matrix A is another matrix, written as A-1, such that the product of
the two is the unit matrix. Thus,

AA-1 = I and A-1 A = I

The symbol A is read as “ A inverse.” The -1 superscript is not an exponent in the

usual algebraic sense. We can verify that two matrices are inverses by carrying out
the multiplication specified in the definition. Thus, the two leftmost matrices in the
following are inverses because their product is the unit matrix.

8 5 2 -5 = 1 0
3 2 -3 8 0 1

Exercise: Verify that the following matrices are inverses of each other

A= 13 3 B= 1 -3
4 1 -4 13

Answer: Multiplication of the two matrices leads to the unit matrix

A= 1 0
0 1
We shall see later that not every matrix has an inverse. However, if the inverse of a
does exist, it is unique: that is, a matrix which has an inverse has exactly one inverse.
The uniqueness of the inverse may not seem obvious, so, let us see what would
happen if in addition to A-1, A had another inverse, call it B. Then, according to the
definition of the inverse, the product AB would have to be I; that is,
AB = I
Now if we multiply both sides of this expression by A-1, we have

A-1 AB = A-1 I

Noting that on the left A-1A can be replaced by I, and then IB can be replaced by B,
and on the right A-1 I can be replaced by A-1, we have,

B = A-1
So, the supposedly different inverse, B, turns out to be the original inverse, A-1, we
shall make use of the uniqueness of the inverse a little later when we will ask what the
expression (?) must be to make the following a true statement.

11
 I = (?) A

The answer will be that (?) must be A-1.

We are now ready to attack the main objective of this section of our study, which is to
compute the inverse of a given matrix. After we have learned the computational
procedures, we shall discover that they are directly applicable to the solution of n by n
systems of linear equations.

Briefly, the inverse of a given matrix can be found by writing the given matrix at the
left, and the corresponding unit matrix next to it, at the right.

Then select and carry out row operations which will convert the given matrix in to the
unit matrix, and apply the dame operations to the matrix at the right. When the left
(given becomes the unit matrix, the matrix on the right will be the desired inverse. To
illustrate, let us find the inverse of the matrix.

C= 3 2
1 1

We start by writing the given matrix to the left and the unit matrix to the right; thus

3 2 1 0
1 1 0 1

To change the left matrix into the unit matrix will require several steps (which we
perform on both matrices). We start by getting a 1 in the upper left corner. Divided
the first row above by 3 to obtain.

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

Next, get a o in the lower left corner by multiplying the first row above by -1 and
adding to the second row to obtain

1 2/3 1/3 0
1 1/3 -1/3 1

multiply the second row, just above, by 3 to obtain

1 2/3 1/3 0
0 1 -1 3
Finally, multiply the second row, just above, by -2/3 and add to the first to obtain

12
 1 0 1 -2
0 1 -1 3

We have found the inverse of

C= 3 2
1 1

it is the matrix at the right; namely,

C -1 = 1 -2
-1 3

Basic procedure recommended:

1. Working one column at time and using the appropriate row operations, first
get the 1 into the correct location (with respect to the identify matrix) with in that
column.

The 1 can be obtained in the cell of the matrix by multiplying by the reciprocal of the
number currently in that cell. ( Row operation type 1).
2. Then use this 1 to obtain zeros in all other locations with in that column. Only
after once column is in the correct format should you move to the next column.
The zero can be obtained in a cell by using an appropriate multiple of the row with the
1 in that column. (Row Operation type 2).
3. Depending on the objective and the situation we may also interchange tow rows.
(Row operation type 3).

Exercise: Apply the procedure and find the inverse of the matrix.
C= 7 3
2 1

Answer: The inverse is the matrix

A -1 = 1 -3
-2 7

We next illustrate the computation of the inverse of a matrix of order 3, the left matrix
in the following:

2 3 1 1 0 0
1 4 2 0 1 0
5 6 4 0 0 1

13
Divided the first row above by 2 to obtain

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

1 4 2 0 1 0
5 6 4 0 0 1

Multiply the first row above by -1 and add to the second row to obtain

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

0 5/2 3/2 -1/2 1 0
5 6 4 0 0 1

Multiply the first row above by -5 and add to the third row to obtain

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

0 5/2 3/2 -1/2 1 0
0 -3/2 3/2 -5/2 0 1

The tactics we have followed are these: first, get a 1 in the first column, first row, then
use combinations of this row with each of the other rows to get zeros in the first
columns of these rows. We now repeat these tactics, starting by getting a 1 in the
second column of the second row, then using this row to get zeros in the second
columns of the other rows.

Divide the second row by 5/2 to obtain the new second row
0 1 3/5 -1/5 2/5 0

If we multiply this new second row by -3/2 and add to the first row, then multiply the
new second row by 3/2 and add to the third row, we obtain the following new
matrices.

1 0 -2/5 4/5 -3/5 0

0 1 3/5 -1/5 2/5 0
0 0 12/5 -14/5 3/5 1

Finally, we repeat the tactics by getting a 1 in the third row of the third column and
then using this row to get zeros in the third column of the other rows. The new third
row is
0 0 1 -7/6 1/4 5/12

If we multiply this new third row by 2/5 and add to the first row, then multiply the
new third row by -3/5 and add to the second row, we find

14
 1 0 0 1/3 -1/2 1/6
0 1 0 1/2 1/4 -1/4
0 0 1 -7/6 1/4 5/12

which has the matrix I to the left and the desired inverse matrix to the right.

Exercise: Multiply the original matrix of the last illustration by the inverse and verify
that the product is I.

Inversion of large matrices is a task best left for electronic computers. However, we
should practice the procedure long enough to understand it very well. Eventhough the
arithmetic is tedious, the basic methodology, is not complicated. To see why the
method works, consider the true statement,

A = IA

suppose we carry out row operations on the left of the equal sign to change the left
matrix to I, and maintain the equality by applying the same operations to the first on
the right of the equal sign (which starts out as I ). The end result will be I on the left
and something times A on the right; thus,
I = (?) A

We showed earlier that the inverse of a matrix is unique. Hence, there is only one
appropriate entry for (?) in the forgoing. It is A-1. It follows that if we write a given
matrix A with I to its right, then change the left to I by row operations which are
applied also to the right, the end result will be A-1 on the right.

Not every matrix has an inverse. For example, consider the matrix

1 1
2 2

If we set this matrix up with the unit matrix to the right, we have

1 1 1 0
2 2 0 1

The indicated row operation is to multiply the first row by -2 and add to the second
row. We find
1 1 1 0
0 0 -2 1

No row operations for the left matrix can be found which will provide a 1 in the lower
left corner and a o in the upper right. The given matrix has no inverse.

15
Before turning to the problem set, we note that step(s) may be required to set up and
maintain the inversion problem in the form required by tactics discussed thus far.
Consider the problem of inverting

0 1
2 3

Writing this matrix with the unit matrix, we have,

0 1 1 0
2 3 0 1

We may put this into the desired from with a 1 at the upper left by multiplying the
second row by 1/2 and adding to the first. We have,

1 5/2 1 1/2
2 3 0 1

We can also interchange the two rows and multiply the new first row by 1/2 (or divide
it by 2) to obtain

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

The usual tactics may now be applied. See No. 1 of the following problem set.

SOLVING SYSTEMS OF EQUATIONS USING MATRIX METHODS

The elimination of variable procedure is generally satisfactory for solving systems of

two equations in two variables. However, for systems containing more than two
equations and involving more than two variables, this procedure is not as efficient as
we would like; so we find ourselves turning to another solution technique.
The alternative solution procedure we shall use employs a MATRIX FORMAT to
organize the data.
The two linear equations of the system
a11x 1 + a12x2 = b1
a21x1 + a 22x2 = b 2
can be represented by the MATRIX EQUATION AX = B where
A = a11 a12
a21 a 22

is called the COEFFICIENT MATRIX,

X = x1
x2

16
 is the SOLUTION VECTOR, or VECTOR OF UNKNOWNS,
B= b1
b2

is called the RIGHT-HAND-SIDE VECTOR, or VECTOR OF

CONSTANTS.
The vector X = n1 represents a solution of the system if
n2 both equaltions are satisfied when we
substitute x1 = n1 and x2 = n2.

The Gaussian Method One “matrix” procedure that can be used to solve
systems of linear equations is known as the GAUSSIAN METHOD. This procedure
was develoed by the mathematician Karl F. Gauss (1777-1855).
We begin with the augmented matrix representation [A| B] of the original system of
equations. Then we systematically reduce the augmented matrix to equivalent
augmented matrices in simpler form. We continue until we have reached the simplest
possible augmented matrix representation, one from which the solution of the sysem
can be obtained by inspection. Indeed, we seek to transform the augmented matrix [A|
B] into the augmented matrix [I|X], where I is the idemtity matrix and X is the solution
vector.

The operations that may be performed on the augmented matrices to convert them to
simpler form are the ELEMENTARY ROW OPERATIONS (EROS). The
elementary row operation may consist of the following:

Operations permitted on the rows of a matrix are called ELEMENTARY

ROW OPERATIONS. They are as follows:

Type I. Any pair of rows in a matrix may be interchanged.

(EXCHANGE operation)
Type II. A row can be multiplied by any nonzero real number.
(MULTIPLY operation)
Type III. A multiple of any row can be added to any other row. (ADD A-
MULTIPLE operation)

Again, it is recommended that we follow the same step-by-step procedure for

obtaining the identity matrix format that we used when calculating an inverse. That is,
working one column at a time, obtain a one in the correct cell using ERO Type II,
Then, obtain zeros in all other cells of the column using ERO Type III and multiples
of the row which has the one.

Example Using the Gaussian procedure, find the solution to the system of
equations
2x + y = 60
x + 3y = 105
We set up the augmented matrix tableau
2 1 60
1 3 105
To obtain a 1 in the row one-column one cell, we multiply row one by
½, to obtain

17
 1 ½ 30 R1  (½)R1
1 3 105
Now, a zero is required in the row two-column one cell. Employing
ERO Type III, we subtract row one from row two, to obtain
1 ½ 30
0 ½ 75 R2  R2 _ R1

Column one is now in the identity matrix format, and we turn our
attention to column two. To obtain a 1 in the row two-column two cell,
we multiply row two by , as
1 ½ 30
0 1 30 R2 (½) R2

Finally, to obtain a 0 in the row one-column two cell, we subtract ½

times row two from row one, as
1 0 15 R1 R1_(½) R2
0 1 30

The solution vector is

X = 15
30
which is interpreted as x = 15, y = 30.

USING THE INVERSE TO SOLVE A SYSTEM OF EQUATIONS

Given an n  n system of linear equations, represented in matrix

notation as AX = B, we may be able to find the solution vector using
the inverse of A. If the coefficient matrix A is invertible, the system
has a unique solution which is given by
X = A-1B
The result stems from the fact that, if A has an inverse, both sides of
the matrix equation can be multiplied by that inverse, as
AX = B
A-1AX = A-1B
(A-1A)X = A-1B
IX = A-1B
X = A-1B
Example x1 + 3x2 = 16
2x1 + 2x2 = 16
The inverse of the coefficient matrix
A= 1 3
2 2

is found by using elementary row operations to covert the augmented

matrix [A  I] into the format [I  A-1]. This inverse is

18
 A-1 = -½ ¾
½ -¼

This inverse can be used to determine the solution to the system.

Accordingly the solution is given as follows:

-½ ¾ . 16 = x1
½ -¼ 16 x2

(-½  16) + (¾  16) = x1

(½  16) + (-¼  16) x2

x1 = 4
x2 4

When the information becomes that 18 for the right hand side value for
the first equation and 16 as it is for the second one, the solution can
easily be determined by introducing tha new change to the solution
procedure as given below.

-½ ¾ . 18 = x1
½ -¼ 16 x2

x1 = 3
x2 = 5

The inverse could be used in this manner to determine new solution.

Markov Chain Analysis

Markov analysis is a method of analyzing the current behavior of some variables in

an effort to predict the future behavior of that same behavior. This procedure was
developed by the Russian mathematician Andrei A. Markov early in this century. He
first used it to describe and predict the behavior of particles of gas in a closed
container. As a management tool markov analysis can be applied to a wide variety of
decision situations. Perhaps its widest use is in examining and predicting the behavior of
consumers in terms of their brand loyalty and switch over from one brand to another. A
more recent application of this technique has been to the study of accounts
receivable behavior, that is, to the study of consumers as they change from " current
account" through "30 days overdue" to "30 to 60 days overdue" and then to " Bad
debt". In each of this applications management is interested in predicting what the
future will bring (number of bad debts, for example, in the account receivable
application) by analyzing what the current behavior is (propensity of customers to
move from one current account to various past due categories).

Markov analysis operates under certain assumptions. In order for a system to be

considered Markov, it must satisfy the following assumptions:

19
  It will operate or exist for a number of periods.
 The states are both mutually exclusive and collectively exhaustive.
 System changes between states from period to period should be described
by transition probabilities, which remain constant.
 The probability of the system being in a given state in a particular period
depends only on its state in the preceding period and the transition
probabilities
 The system is a closed one i.e. there will be no arrival or exits from the
system.

The markov system is useful both for short-term decisions and long-term decisions.
These decisions are important to make effective resource allocation. There are three
methods to deal with the problem, decision tree and matrix for short-term behavior
and algebraic method for long-term behavior.

State of transition matrix

A system whose behavior can be described as a markov process can be summarized

by a set of conditional, or transition probabilities, which indicate the tendency of the
system to change from one period to another period to the next. Assume a small
town, which has a population of 1000 all drinking either the Coca Cola brand or
Pepsi Cola. The following table shows switch of customers from one brand to
another.

Brand Clients at the Gains Losses Clients at the

beginning of the period end of the
period
Coca 400 120 40 480
cola
Pepsi 600 40 120 520
Cola

What Pepsi loses is what Coca gains and vice versa. Now to develop the transition
probabilities, it is important to calculate the probability of customer retention, gain
and loses.

Retention
Coca Cola customers retained, 400-40=360 the probability is then 360/400= 0.9
Pepsi Cola customers retained, 600-120=480 the probability is then 480/600= 0.8

Gain and Lose

Coca Cola loses 40 customers; the probability is then 40/400=0.1

Coca cola gains what Pepsi loses. The probability of lose for Pepsi is then 120/600=
0.2

The transition matrix can now be constructed. The loss probabilities are inserted
across the row and the gains across the column. The sum of the probabilities in each
row must be one because it contains both probabilities that are retention and lose.
For the above example the state transition probabilities are summarized in the
following state transition matrix.

Transition matrix
To
From Coca Cola Pepsi Cola
Coca Cola 0.9 0.1

20
 Pepsi Cola 0.2 0.8

The transition matrix shows that Coca cola will retain 90% of its customers and gain
20% of Pepsi cola and Pepsi cola retains 80% of its customers and gains 10% of
Coca cola customers. No customer can drink both brands at a given period and
there is no other brand in the market that is the probabilities are mutually exclusive
and collectively exhaustive. In addition to this the switch occurs every month. With
the given month the probability is assumed the same

Decision tree analysis

Assume the above example, to determine the market share of each brand in the
coming two months we can use the decision tree and calculate the probability as
follows

Coca Cola
0.81
0.9
Coca cola
0.9 0.1 Pepsi
Cola 0.09

Coca Cola Coca

Cola 0.02
0.2
0.1
Pepsi Cola

0.8 Pepsi Cola 0.08

Pepsi Cola
0.64
0.8
Pepsi cola
0.8 0.2 Coca
Cola 0.16

Pepsi Cola Pepsi Cola 0.02

0.1
0.2
Coca Cola

0.9 Coca Cola 0.18

The decision tree shows that from customers of coca cola in period 0, 90% will
remain there while 10% shift to the other brand Pepsi Cola in period 1. Like wise,
80% of Pepsi Cola customers remain loyal but 20% switch to Coca Cola. In period 2,
the joint probability should be computed to determine the probability of the market
share for each brand.

Joint probability
Coca Cola Pepsi Cola
Coca Cola 0.9x0.9+0.1x0.2=0.83 0.9x0.1+0.1x08=0.17
Pepsi Cola 0.8x0.2+0.2x0.9=0.34 0.8x0.8+0.2x0.1=0.66

21
As the table shows at the period 1 will have 90% of its customers and 20% of
customers of Pepsi Cola. 400x0.9+600x0.2, which is equal to 480 and Pepsi cola will
have 80% of its customers at period 0 plus 10% of customers of Coca Cola i.e.
0.8x600+0.1x400 which is equal to 520. In period 2 Coca Cola will have 83% of its
customers and 34% of customers of Pepsi cola at period 0. Pepsi Cola will have 66%
of its original customers and 17% of customers of Coca Cola.

Coca Cola's share of market is equal to 83%x400+34%x600 that is equal to 536.

Pepsi Cola's share of market is equal to 66%x600+17%x400 that is equal to 464.

The system will continue like this until it will reach an equilibrium condition at
some point, which will be discussed later.

Markov analysis offers such a tool for marketing analysis. The decision maker can
able to draw more accurate conclusions about the marketing position, both at
present and in future.

The decision tree is helpful in that it portrays the system behavior visually. However,
as the number of periods increases, it will be difficult to compute through this
method. When such is the case, determining by the use of matrix algebra is an
alternative.

Markov chain analysis through Matrix algebra

In the transition probabilities matrix the row represents the probability of retention
and losses of customers and the column represents the retention of customers and
the gain of customers. Look at the following example.

EX-1 Assume that only three Department stores exist in the city of Addis Ababa
namely Department store A hereinafter called simply A, Department store B
hereinafter called simply B and Department store C hereinafter simply called C. At
the beginning of June 1 2001, lets assume that 80% of A's customers are loyal and
10% goes to B and the remaining 10% to C. Like wise 90% of B's customers are loyal
and 7% will goes to B and 3% to C. 75 % of C's customers are loyal and 15% will
switch to A and 10% to B.

The probability of retention and loses are written across the row and the probability
of retention and gain down the column.

A B C
A 0.8 0.1 0.1
B 0.07 0.9 0.03
C 0.15 0.10 0.75

If we assume that all the customers have been shopping at one store at the start,
then we develop the current proportion, which is referred as a probability vector.

Starting from Current Matrix

A [ 1 0 0 ]

B [ 0 1 0 ]

C [ 0 0 1 ]

After finding the current probability vector we can easily compute the next by
multiplying the probability vector with the transition matrix. If all the customers
were at A at the beginning of the period, the next probability is computed as follows:
A B C

22
 [1 0 0] 0.8 0.1 0.1
0.07 0.9 0.03
0.15 0.1 0.75

We have 1 by 3 and 3 by 3 therefore both are conformable for multiplication. And

the resultant matrix is a 1 by 3 matrix.
1x0.8+0x0.07+0x015, 1x0.1+0x0.9+0x0.1, 1x0.1+0x0.03+0x0.75
0.8 0.1 0.1
The next probability is determined by using the new probability vector it looks as the
following
A B C
[0.8 0.1 0.1] 0.8 0.1 0.1
0.07 0.9 0.03
0.15 0.1 0.75
(0.8x0.8)+ (0.1x0.07)+ (0.1x015), (0.8x0.1)+ (0.1x0.9) + (0.1x0.1), (0.8x0.1)+
(0.1x0.03)+ (0.1x0.75)= 0.662 0.18 0.158

Now we have the probability vector [0.662 0.18 0.158]

It will continue like this up to some point and the shift will be stagnant. Such a
point is called point of equilibrium or long-term state. In general we do have the
following formula to predict the nth state of the system.

Markov Chain Formula

nth state of a Markov Chain.

Vij (n) = Vij (n-1) x p, or V ij (n) = Vij (0) x (P) n.

Or
Vij (n) = Vij (0) x (P) n.
Where: P = transition matrix
Vij (n) = Vector for period n.
Vij (n-1) = vector for period n-1
Steady state or equilibrium situations
It is reasonable to assume that in the future a state of equilibrium might be
approached regarding market shares; i.e., the exchange of customers under
equilibrium would be at a point where exchange of customers is to freeze. Of course,
equilibrium could result only if no action is taken that alters the transition
probabilities.

In our previous example, we have seen how customers switch from one Department
store to another from time to time. Taking the same example we can see that how
equilibrium will be reached.

A B C
A 0.8 0.1 0.1

B 0.07 0.9 0.03

C 0.15 0.1 0.75

The probability of A, B and C is calculated as

Eq-1 A=0.8A+0.07B+0.15C
Eq-2 B=0.1A+0.90B+0.10C
Eq-3 C=0.1A+0.03B+0.75C

23
 Because the states are mutually exclusive and collectively exhaustive the sum is 1.
Hence, A+B+C=1

To calculate for the probabilities we have to ignore one of the equations as we have
three unknowns and four equations. For convenience it is better to eliminate one of
the first three, as the fourth equation will help us calculate the probabilities
conveniently.
From equation four the value of C is equal to 1 minus the probability of A and B i.e.
C= 1-A-B
By inserting the value of C in to equation 1 and 2 we can solve for the probabilities

A= 0.8A+ 0.07B+0.15(1-A-B) which is equal to

A= 0.8A+0.07B+0.15-0.15A-0.15B and this is equivalent to
0.35A+0.08B=0.15 similarly we calculate for B and it comes
0.2B=0.1 which gives us 0.5 using the value of B the value of A is
0.35A+0.08x0.5=0.15 which is equal to 0.314, the value of C is, therefore,
1-0.5-0.314 = 0.186
Hence the steady state probability of A, B, and C is 0.314, 0.5 and 0.186
respectively.

Exercises
1. A division of the ministry of public health has conducted a sample survey on the
public attitudes towards the use of condoms. From the results of the survey the
department concluded that currently only 20% of the population uses condoms and
every month 10% of non-users become users, where as 5% of users discontinue
using.
Required
1. Write the current transition matrices.
2. What will be the percentage of users from total population just after two
months?
3. What will be the proportion of the non users and users in the long run?
Answer:
Let. U - Stands for users, and N- stands for nonuser
1. Initial state VUN (0) = 0.2 0.8

To the next month

From one Users (U) Non Users (N)
month Users (U) .95 .05
Non Users (N) .10 .90

2. V (2) UN = (.3295 0.6705)

3. VU VN = 0.67 0.33

2. A city has two suburbs: suburb x and suburb y. Over the past several years, the city
has experienced a population shift from the city to the suburbs, as shown in the
table below.
To the other place
From one City(C) Suburb X (X) Suburb Y (Y)
place City (C) .85 .07 .08
Suburb X (X) .01 .96 .03
Suburb Y (Y) .01 .02 .97

24
 In 20X0, the city had a population of 120,000, suburb X had a population of 80,000,
and suburb Y had a population of 50,000. Assuming that the population in the
metropolitan area remains constant at 250,000 people,
a. How many people will live in each of the three areas in 20X2? Answer: In 20X2
89,275, 91,200, and 69,525 people will live in the city, suburb X, and suburb Y
respectively.
b. How many people will live in each of the three areas in the long run? Answer: In
the long run 15,625, 96,350, and 138, 025 people will live in the city, suburb X,
and suburb Y respectively.

3. A population of 100,000 consumers makes the following purchases during a

particular week: 20,000 consumers make the following purchases during a
particular week: 20,000 purchases Brand A, 35,000 Brand purchase B and 45,000
purchase neither brand. From a market study, it in estimated that of those who
purchase Brand A, 80% will purchase it again next week, 15% will purchase Brand
B next week, and 5% will purchase neither brand. Of those who purchase B, 85%
will purchase it again next week, 12% will purchase brand A next week, and 3% will
purchase neither band. Of those who purchased neither brand, 20% will purchase
as A next week, 15% will purchase Brand B next week, and 65% will purchase
neither band next week. If this purchasing pattern continues, will the market
stabilize? What will the stable distribution be? Yes YA VB VC= (.4 .5 .1)

4. In a certain college class, 70% of the students who receive an “A” on the current
examination will receive an “A” on the next examination. Moreover, 10% of the
students who do not receive an “A” on the current examination will receive an “A” on
the next examination. Assuming tat this pattern continues, what is the stable
matrix?
VA VA1 = (.25 .75)

5. A vigorous television advertising campaign is conducted during the football season to

promote a well-known brand X sharing cream. For each of several weeks, a survey is
made and it is found that each week 80% of those using brand X continue to use it
and 20% switch. It is also found that those not using brand x, 20% switch to brand
X while the other 80% continue using another band.

a. Write the transition matrix, assuming the transition percentages continue hold
for succeeding weeks.
Answer:
X X1
X .8 .2
X1 .2 .8

b. If 20% of the people are using brand X at the start of the advertising campaign,
what percentage will be using brand X one week later? Two weeks later?
Answer: V (1) XX1 = (.32 .68), V (2) XX1 = (.392 .608)
c. What portion of the market will be using brand X area the end of the season,
assuming the transition matrix remains the same? (Find the Steady-state
matrix) Answer: VX VX1 = (.5 .5)

25