Inverting Matrices: Determinants and Matrix Multiplication 

Determinants

Square matrices have determinants, which are useful in other matrix operations,
especially inversion.
a 11 a12
a 21 a22
¿
righ
¿
¿
¿

For a second-order square matrix, A, , the determinant of A,

[¿ ] ¿
¿
¿

|A|=a11∗a 22−a12∗a21

Consider the following bivariate raw data matrix:

Subject # 1 2 3 4 5
X 12 18 32 44 49
Y 1 3 2 4 5

from which the following XY variance-covariance matrix is obtained:

X Y
X 256 21.5
Y 21.5 2.5

COV XY 21 .5
r= = =0. 9
S X SY √256 √2 .5 |A|=256(2.5)−21.5(21.5)=177 .75
Think of the variance-covariance matrix as containing information about the two
variables – the more variable X and Y are, the more information you have. The total
amount of information you have is reduced, however, by any redundancy between X
and Y – that is, to the extent that you have covariance between X and Y you have less
total information. The determinant of a matrix is sometimes called its generalized
variance, the total amount of information you have about variance in the scores, after
removing the redundancy between the variables – look at how we just computed the
determinant – the product of the variances (information) less the product of the
covariances (redundancy).

Now think of the information in the X scores as being represented by the width of
a rectangle, and the information in the Y scores represented by the height of the
rectangle. The area of this rectangle is the total amount of information you have. Since

 Copyright 2011, Karl L. Wuensch - All rights reserved.

[Link]
 2

I specified that the shape was rectangular (X and Y are perpendicular to one another),
the covariance is zero and the generalized variance is simply the product of the two
variances.

Now allow X and Y to be correlated with one another. Geometrically this is

reducing the angle between X and Y from 90 degrees to a lesser value. As you reduce
the angle the area of the parallelogram is reduced – the total information you have is
less than the product of the two variances. When X and Y become perfectly correlated
(the angle is reduced to zero) the determinant had been reduced to value zero.

Consider the following bivariate raw data matrix:

Subject # 1 2 3 4 5
X 10 20 30 40 50
Y 1 2 3 4 5

from which the following XY variance-covariance matrix is obtained:

X Y
X 250 25
Y 25 2.5

COV XY 25
r= = =1
S X SY √ 250 √ 2 .5 |A|=250(2.5)−25(25)=0
Inverting a Matrix

Determinants are useful in finding the inverse of a matrix, that is, the matrix that
when multiplied by A yields the identity matrix. That is, AA1 = A1A = I

An identity matrix has 1’s on its main diagonal, 0’s elsewhere. For example,
1 0 0
0 1 0
0 0 1
¿
righ
¿
¿
¿
[ ¿] [ ¿] ¿
¿
¿

With scalars, multiplication by the inverse yields the scalar identity, 1:

1 1 a
a∗ =1. a∗ = .
a Multiplying by an inverse is equivalent to division: b b
a 22 - a 12
− a 2 1 a1 1
¿
r i gh
¿
¿
¿
2 . 5 -2 1 . 5
- 21 . 5 25 6
¿
r i gh
¿
¿

The inverse of a 2 * 2 matrix,

¿
[ ¿ ] ¿
1 1
A −
2 = ∗¿ ¿
| A | 2
¿

for our example.

 3

Multiplying a scalar by a matrix is easy - simply multiply each matrix element by the
scalar, thus,
. 014064698 - . 120956399
- . 120956399 1 . 440225035
¿
righ
¿
¿
¿
[ ¿ ] ¿
−1
A =¿ ¿
¿

Now to demonstrate that AA1 = A1A = I, but multiplying matrices is not so easy.
For a 2  2,
a b
c d
¿
r ig h
¿
¿
¿
w x
y z
¿
r ig h
¿
¿
¿
row 1∗c ol 1 row 1∗c ol 2
row 2∗c ol 1 row 2∗c ol 2
¿
r ig h
¿
¿
¿
aw + by a x + bz
c w + dy c x + dz
¿
r ig h
¿
¿
¿
[ ¿ ] ¿
¿
¿

256 21 . 5
21 . 5 2 . 5
¿
righ
¿
¿
¿
. 014064698 - . 120956399
−. 120956399 1 . 440225035
¿
righ
¿
¿
¿
1 0
0 1
¿
righ
¿
¿
¿
[ ¿ ] ¿
¿
¿

Third-Order Determinant and Matrix Multiplication

The determinant of a third-order square matrix,

a11 a12 a13
a21 a22 a23
a31 a32 a33
¿
righ
¿
¿
¿
[ ¿ ] [ ¿] ¿
¿
¿

=a11∗a22∗a 33+ a12∗a 23∗a 31+a 13∗a32∗a21−a 31∗a22∗a13−a11∗a32∗a 23−a12∗a21∗a33

Matrix multiplication for a 3 x 3
a b c
d e f
g h i
¿
rig h
¿
¿
¿
r s t
u v w
x y z
¿
rig h
¿
¿
¿
a r + b u + cx a s + b v + cy a t + b w + cz
d r + eu + fx ds + ev + fy d t + e w + fz
gr + hu + ix gs + h v + iy gt + h w + iz
¿
rig h
¿
¿
¿
[ ¿ ] [ ¿ ] ¿
¿
¿

That is,
row 1∗col 1 row 1∗col 2 row 1∗col 3
row 2∗col 1 row 2∗col 2 row 2∗col 3
row 3∗col 1 row 3∗col 2 row 3∗col 3
¿
righ
¿
¿
¿
[¿ ] [¿ ]¿
¿
¿

Isn’t this fun? Aren’t you glad that SAS will do matrix algebra for you? Copy the
little program below into the SAS editor and submit it.
 4

SAS Program
Proc IML;
reset print;
XY ={
256 21.5,
21.5 2.5};
determinant = det(XY);
inverse = inv(XY);
identity = XY*inverse;
quit;

Look at the program statements. The “reset print” statement makes SAS display
each matrix as it is created. When defining a matrix, one puts brackets about the data
points and commas at the end of each row of the matrix.

Look at the output. The first matrix is the variance-covariance matrix from this
handout. Next is the determinant of that matrix, followed by the inverted variance-
covariance matrix. The last matrix is, within rounding error, an identity matrix, obtained
by multiplying the variance-covariance matrix by its inverse.

SAS Output
XY 2 rows 2 cols (numeric)
256 21.5
21.5 2.5

DETERMINANT 1 row 1 col (numeric)

177.75

INVERSE 2 rows 2 cols (numeric)

0.0140647 -0.120956
-0.120956 1.440225

IDENTITY 2 rows 2 cols (numeric)

1 -2.22E-16
-2.08E-17 1

Copyright 2011, Karl L. Wuensch - All rights reserved.