Transcript Matrix Algebra Primer - Louisiana Tech University

A Little Necessary Matrix
Algebra for Doctoral Studies in
Business Administration
James J. Cochran
Department of Computer
Information Systems & Analysis
Louisiana Tech University
[email protected]
Matrix Algebra
Matrix algebra is a means of efficiently
expressing large numbers of calculations
to be made upon ordered sets of
numbers
Often referred to as Linear Algebra
Why use it?
Matrix algebra is used primarily to
facilitate mathematical expression.
Many equations would be completely
intractable if scalar mathematics had to
be used. It is also important to note
that the scalar algebra is under there
somewhere.
Definitions - Scalars
scalar - a single value (i.e., a number)
Definitions - Vectors
Vector - a single row or column of
numbers
 Each individual entry is called an
element
 denoted with bold small letters
 row vector
a  1 2 3 4
 column vector
1 
2
a   
3
 4 
Definitions - Matrices
A matrix is a rectangular array of
numbers (called elements) arranged in
orderly rows and columns
a
A   11
a21
a12
a22
a13 
a32 
Subscripts denote row (i=1,…,n) and
column (j=1,…,m) location of an
element
Definitions - Matrices
Matrices are denoted with bold Capital
letters
All matrices (and vectors) have an order
or dimensions - that is the number of
rows x the number of columns. Thus A
is referred to as a two by three matrix.
 Often a matrix A of dimension n x m is
denoted Anxm
 Often a vector a of dimension n (or m)
is denoted An (or Am)
Definitions - Matrices
Null matrix – a matrix for which all
elements are zero, i.e., aij = 0  i,j
Square Matrix – a matrix for which the
number of rows equals the number of
columns (n = m)
Symmetric Matrix – a matrix for which aij
= aji  i,j
Definitions - Matrices
Diagonal Elements – Elements of a
Square Matrix for which the row and
column locations are equal, i.e., aij  i = j
Upper Triangular Matrix – a matrix for
which all elements below the diagonal
are zero, i.e., aij = 0  i,j  i < j
Lower Triangular Matrix – a matrix for
which all elements above the diagonal
are zero, i.e., aij = 0  i,j  i > j
Matrix Equality
Thus two matrices are equal iff (if and
only if) all of their elements are
identical
Note that statistical data sets are
matrices (usually with observations in
the rows and variables in the columns)
Variable 1 Variable 2
Observation 1
a11
a12
Observation 2
a21
a22
Observation n
an1
an2
Variable m
a1m
a2m
anm
Basic Matrix Operations
Transpositions
Sums and Differences
Products
Inversions
The Transpose of a Matrix
The transpose A’ of a matrix A is the
matrix such that the ith row of A is the
jth column of A’, i.e., B is the transpose
of A iff bij = aji  i,j
This is equivalent to fixing the upper
left and lower right corners then
rotating the matrix 180 degrees
Transpose of a Matrix
An Example
If we have
then
i.e.,
1 4 
A  2 5 
3 6 
A '  1 2 3 
 4 5 6 
1 4 
A  2 5  A ' 
3 6 
1 2 3 
 4 5 6 
More on the
Transpose of a Matrix
(A’)’ = A (think about it!)
If A = A', then A is symmetric
Sums and Differences
of Matrices
Two matrices may be added
(subtracted) iff they are the same order
Simply add (subtract) elements from
corresponding locations
where
 a11 a12  b11 b12   c11 c12 
a21 a22  + b21 b22  = c 21 c 22 
a31 a32  b31 b32  c31 c32 

 
 

a11 + b11 = c11 , a12 + b12 = c12 ,
a21 + b21 = c 21 , a22 + b22 = c 22 ,
a31 + b31 = c 31 , a32 + b32 = c 32
Sums and Differences
An Example
If we have
1 2 
A  3 4 
5 6 
and
7 10 
B = 8 11
9 12
then we can calculate C = A + B by
1 2  7 10  8 12
C  A + B = 3 4  + 8 11 = 11 15
5 6  9 12 14 18
Sums and Differences
An Example
Similarly, if we have
1 2 
A  3 4 
5 6 
and
7 10 
B = 8 11
9 12
then we can calculate C = A - B by
1 2  7 10  -6 -8
C  A - B = 3 4  - 8 11 = -5 -7 
5 6  9 12 -4 -6 
Some Properties of
Matrix Addition/Subtraction
Note that
 The transpose of a sum = sum of
transposes  (A+B+C)’ = A’+B’+C’
 A+B = B+A (i.e., matrix addition is
commutative)
 Matrix addition can be extended beyond
two matrices
 matrix addition is associative, i.e.,
A+(B+C) = (A+B)+C
Products of Scalars
and Matrices
To multiply a scalar times a matrix,
simply multiply each element of the
matrix by the scalar quantity
 a11
b
a21
a12  ba11 ba12 
=
a22  ba21 ba22 
Products of Scalars & Matrices
An Example
If we have
1 2 
A  3 4 
5 6 
and
b = 3.5
then we can calculate bA by
1 2   3.5 7.0 
bA  3.5 3 4  = 10.5 14.0 
5 6  17.5 21.0
Note that bA = Ab if b is a scalar
Some Properties of
Scalar x Matrix Multiplication
Note that
 if b is a scalar then bA = Ab (i.e., scalar
x matrix multiplication is commutative)
 Scalar x Matrix multiplication can be
extended beyond two scalars
 Scalar x Matrix multiplication is
associative, i.e., ab(C) = a(bC)
 Scalar x Matrix multiplication leads to
removal of a common factor, i.e., if
ba11 ba12 
 a11 a12 
C  ba21 ba22  then C  bA where A = a21 a22 
ba31 ba32 
a31 a32 




Products of Matrices
We write the multiplication of two
matrices A and B as AB
This is referred to either as
 pre-multiplying B by A
or
 post-multiplying A by B
So for matrix multiplication AB, A is
referred to as the premultiplier and B is
referred to as the postmultiplier
Products of Matrices
In order to multiply matrices, they must
be conformable (the number of columns
in the premultiplier must equal the
number of rows in postmultiplier)
Note that
 an (m x n) x (n x p) = (m x p)
 an (m x n) x (p x n) = cannot be done
 a (1 x n) x (n x 1) = a scalar (1 x 1)
Products of Matrices
If we have A(3x2) and B(2x3) then
 a11 a12 a13  b11 b12   c11 c12 
AB  a21 a22 a23  x b21 b22  = c21 c22   C
a31 a32 a33  b31 b32  c31 c32 
 


 
where
c11 = a11b11 + a12b21 + a13b31
c12 = a11b12 + a12b22 + a13b32
c 21 = a21b11 + a22b21 + a23b31
c 22 = a21b12 + a22b22 + a23b32
c 31 = a31b11 + a32b21 + a33b31
c 32 = a31b12 + a32b22 + a33b32
Products of Matrices
If we have A(3x2) and B(2x3) then
b11 b12   a11 a12 a13 
BA  b21 b22  x a21 a22 a23  = undefined
b31 b32  a31 a32 a33 

 

i.e., matrix multiplication is not
commutative (why?)
Matrix Multiplication
An Example
If we have
then
1 4 7
1 4 
A  2 5 8 and B = 2 5 
3 6 9
3 6 
1 4 7 1 4   c11 c12  30 66 
AB  2 5 8 x 2 5  = c21 c 22  = 36 81
3 6 9 3 6  c31 c32  42 96
where
c11 = a11b11 + a12b21 + a13b31 = 1 1  + 4  2  + 7  3  = 30
c12 = a11b12 + a12b22 + a13b32 = 1  4  + 4 5  + 7  6  = 66
c 21 = a21b11 + a22b21 + a23b31 = 2 1 + 5  2  + 8  3  = 36
c 22 = a21b12 + a22b22 + a23b32 = 2  4  + 5 5  + 8  6  = 81
c 31 = a31b11 + a32b21 + a33b31 = 3 1  + 6  2  + 9  3  = 42
c 32 = a31b12 + a32b22 + a33b32 = 3  4  + 6  5  + 9  6  = 96
Some Properties of
Matrix Multiplication
Note that
 Even if conformable, AB does not
necessarily equal BA (i.e., matrix
multiplication is not commutative)
 Matrix multiplication can be extended
beyond two matrices
 matrix multiplication is associative, i.e.,
A(BC) = (AB)C
Some Properties of
Matrix Multiplication
Also note that
 The transpose of a product is equal to the
product of the transposes in reverse order
 (ABC)’ = C’B’A’
 If AA’ = A then A' is idempotent (and A' =
A)
Special Uses for
Matrix Multiplication
Sum Row Elements of a Matrix
 Premultiply a matrix A by a conformable
row vector of 1s – If
1 4 7 
A  2 5 8 
3 6 9
then premultiplication by
1  1 1 1
will yield the column totals for A, i.e.
1 4 7
A1  1 1 1 2 5 8 = 6 15 24 
3 6 9
Special Uses for
Matrix Multiplication
Sum Column Elements of a Matrix
 Postmultiply a matrix A by a
conformable column vector of 1s – If
1 4 7 
A  2 5 8 
3 6 9
then postmultiplication by
1
1  1
1
will yield the column totals for A, i.e.
1 4 7  1 12
A1  2 5 8 1 = 15
3 6 9 1 18
Special Uses for
Matrix Multiplication
The Dot (or Inner) Product of two Vectors
 Premultiplication of a column vector a by
conformable row vector b yields a single
value called the dot product or inner
product - If
a  3 4 6 
then ab gives us
and
5 
b  2
8
5
ab  3 4 6 2 = 3 5  + 4 2  + 6  8  = 71
8
which is the sum of products of elements
in similar positions for the two vectors
Special Uses for
Matrix Multiplication
The Outer Product of two Vectors
 Postmultiplication of a column vector a by
conformable row vector b yields a matrix
containing the products of each pair of
elements from the two matrices (called
the outer product) - If
a  3 4 6 
then ba gives us
and
5 
b  2
8
5
15 20 30 
ba  2 3 4 6  =  6 8 12 
8
24 32 48
Special Uses for
Matrix Multiplication
Sum the Squared Elements of a Vector
 Premultiply a column vector a by its
transpose – If
5 
a  2 
8
then premultiplication by a row vector a’
a'  5 2 8 
will yield the sum of the squared values of
elements for a, i.e.
5
a'a  5 2 8 2 = 52 + 22 + 82 = 93
8
Special Uses for
Matrix Multiplication
 Postmultiply a row vector a by its
transpose – If
a  7 10 1
then postmultiplication by a column
vector a’
7
a'  10 
 1 
will yield the sum of the squared values
of elements for a, i.e.
7
aa'  7 10 1 10  = 72 +102 +12 =150
 1 
Special Uses for
Matrix Multiplication
Determining if two vectors are Orthogonal
– Two conformable vectors a and b are
orthogonal iff
a’b = 0
Example: Suppose we have
a  7 10 1
then
and
b  1 0.5 2
 1 
ab'    7 10 -1 0.5 = -7 1 +10  0.5  - 1 -2 = 0
 2 
Special Uses for
Matrix Multiplication
Representing Systems of Simultaneous
Equations – Suppose we have the
following system of simultaneous
equations:
px1 + qx2 + rx3 = M
dx1 + ex2 + fx3 = N
If we let
 x1 

p
q
r

A
, x =  x 2  , and b = M
d e f 
N
 x3 
 
then we can represent the system (in
matrix notation) as Ax = b (why?)
Special Uses for
Matrix Multiplication
Linear Independence – any subset of
columns (or rows) of a matrix A are
said to be linearly independent if no
column (row) in the subset can be
expressed as a linear combination of
other columns (rows) in the subset.
If such a combination exists, then the
columns (rows) are said to be linearly
dependent.
Special Uses for
Matrix Multiplication
The Rank of a matrix is defined to be the
number of linearly independent columns
(or rows) of the matrix.
Nonsingular (Full Rank) Matrix – Any
matrix that has no linear dependencies
among its columns (rows). For a square
matrix A this implies that Ax = 0 iff x =
0.
Singular (Not of Full Rank) Matrix – Any
matrix that has at least one linear
dependency among its columns (rows).
Special Uses for
Matrix Multiplication
Example - The following matrix A
1 2 3 
A  3 4 9 
6 5 12
is singular (not of full rank) because the
third column is equal to three times the
first column.
This result implies there is either i) no
unique solution or ii) no existing
solution to the system of equations Ax
= 0 (why?).
Special Uses for
Matrix Multiplication
Example - The following matrix A
1 2 5 
A  3 4 11
6 5 16 
is singular (not of full rank) because the
third column is equal to the first column
plus two times the second column.
Note that the number of linearly
independent rows in a matrix will always
equal the number of linearly independent
columns in the matrix.
Geometry
of Vectors
Vectors have geometric properties of
length and direction – for a vector
x
x   x1 
 2 
we have
2
x2
length of x  L x 
x  x1 x2 
x1
1
x12  x 22 

x12  x 22

x'x
Why?
 x p2
Geometry
of Vectors
Recall the Pythagorean Theorem: in any
right triangle, the lengths of the
hypotenuse and the other two sides are
related by the simple formula.
2
length of hypotenuse  L h 
side a
1
side b
L2a  L2b

x12  x 22

x'x
Geometry
of Vectors
Vector addition – for the vectors
we have
x
y
x   x1  , y   y1 
 2 
 2 
 x1  y1 
 x 2  y 2 
2
y  y1 y2 
x  x1 x2 
q
1
x  y
x + y   x 1  y1 
 2
2

Geometry
of Vectors
Scalar multiplication changes only the
vector length – for the vector
we have
2
x
x   x1 
 2 
length of cx  L cx 
cx  c x1 x2 

c 2  x12  x 22 
c2  x12  x 22 
 c x'x
1
 x p2 
Geometry
of Vectors
Vector multiplication have angles
between them – for the vectors
we have
x
y
x   x1  , y   y1 
 2 
 2 
 xy 
xy
q  arccos 
 cos  q  

 L L 
L xL y
 x y
y  y1 y2 
x  x1 x2 
2
q
1
xy
x'x y'y
A Little Trigonometry Review
The Unit Circle
2
Cos q = 0
-1  Cos q  0
0  Cos q  1
Cos q = -1
q
-1  Cos q  0
Cos q = 1
1
0  Cos q  1
Cos q = 0
A Little Trigonometry Review
Suppose we rotate x any y so x lies on
2
axis 1:
y  y1 y2 
x  x1 x2 
qxy
1
A Little Trigonometry Review
What does this imply about rxy?
2
Cos q = 0
-1  Cos q  0
0  Cos q  1
qxy
Cos q = -1
Cos q = 1
1
-1  Cos q  0
0  Cos q  1
Cos q = 0
Geometry
of Vectors
What is the correlation between the
2
vectors x and y?
1.0 
-0.5


x  0.6 , y  -0.3
 


Plotting in
the column
space
gives us
x
1
y
Geometry
of Vectors
Rotating so x lies on axis 1 makes it
2
easier to see:
Qxy=1800
1
Geometry
of Vectors
What is the correlation between the
vectors x and y?
1.0
-0.5
x  0.6 , y  -0.3
 


xy
cosq 

L xL y


xy
x'x y'y
1.0  -0.5  + 0.6  -0.3 
1.0 1.0   0.6  0.6  -0.5  -0.5   -0.3  -0.3 
-0.68
= -1.00
1.36 0.34
Geometry
of Vectors
Of course, we can see this by plotting
the these values in the x,y (row) space:
1.0
-0.5
x  0.6 , y  -0.3
 


Y
X
0.6, -0.3
1.0, -0.5
Geometry
of Vectors
What is the correlation between the
vectors x and y?
1.0
-0.50 
x    ,y  
0.4 
 1.25 
xy
cosq 

L xL y


xy
x'x y'y
1.0  -0.50  + 0.4 1.25 
1.0 1.0   0.4  0.4  -0.50  -0.50   1.25  1.25 
0.00
= 0.00
1.16 1.8125
Geometry
of Vectors
Plotting in the column space gives us
-0.50 
y  
 1.25 
2
Qxy=900
1.0
x   
0.4 
1
Geometry
of Vectors
Rotating so x lies on axis 1 makes it
2
easier to see:
Qxy=900
1
Geometry
of Vectors
 The space of all real m-tuples, with
scalar multiplication and vector addition
as we have defined, is called vector
space.
 The vector
y  a1x1  a2x2 
 akxk
is a linear combination of the vectors x1,
x2,…, xk.
 The set of all linear combinations of the
vectors x1, x2,…, xk is called their linear
span.
Geometry
of Vectors
Here is the column space plot for some
vectors x1 and x2:
3
-0.50
x2   0.50
 1.50
1.00
x1  0.40
0.20
1
2
Geometry
of Vectors
Here is the linear span for some vectors
x1 and x2:
3
1
2
Geometry
of Vectors
 A set of vectors x1, x2,…, xk is said to be
linearly dependent if there exist k
numbers a1, a2,…, ak, at least one of
which is nonzero, such that
a1x1  a2x2 
 akxk  0
Otherwise the set of vectors x1, x2,…, xk
is said to be linearly independent
Geometry
of Vectors
Are the vectors
1.0 
-0.5


x  0.6 , y  -0.3
 


linearly independent?
Take a1 = 0.5 and a2 = 1.0. Then we have
1.0
-0.5
0.0
a1x  a2y  0.5 0.6  1.0 -0.3  0.0
 


 
The vectors x and y are dependent.
Geometry
of Vectors
Geometrically x and y look like this:
2
1.0
x  0.6 
 
1
-0.5
y  -0.3


Geometry
of Vectors
Rotating so x lies on axis 1 makes it
2
easier to see:
Qxy=1800
1
Geometry
of Vectors
Are the vectors
1.0
-0.50 
x    ,y  
0.4 
 1.25 
linearly independent?
There are no real values a1, a2 such that
1.0 
-0.50 
0.0 



a1x  a2y  a1
 a2

0.4 
 1.25 
0.0 
so the vectors x and y are independent.
Geometry
of Vectors
Geometrically x and y look like this:
-0.50 
y  
 1.25 
2
Qxy=900
1.0
x   
0.4 
1
Geometry
of Vectors
Rotating so x lies on axis 1 makes it
2
easier to see:
Qxy=900
1
Geometry
of Vectors
Here x and y are called perpendicular (or
orthogonal) – this is written x  y.
Some properties of orthogonal vectors:
 x’y = 0  x  y
 z is perpendicular to every vector iff z = 0
 If z is perpendicular to each vector x1, x2,
…, xk, then z is perpendicular their linear
span.
Geometry
of Vectors
Here vectors x1 and x2 (plotted in the
column space) are orthogonal
3
-0.50
x2   0.50
 1.50
1.00
x1  0.40
0.20
1
2
Geometry
of Vectors
Recall that the linear span for vectors x1
and x2 is:
3
1
2
Geometry
of Vectors
Vector z looks like this:
3
 0.10
z  -0.32
 0.14
 0.10
z  -0.32
 0.14
1
2
Geometry
of Vectors
The vector z is perpendicular to the linear
span for vectors x1 and x2:
3
 0.10
z  -0.32
 0.14
Check each of the
dot products!
1
2
Geometry
of Vectors
Here vectors x1, x2, and z from our
previous problem are orthogonal
3
-0.50
x2   0.50
 1.50
 0.10
z  -0.32
 0.14
1.00
x1  0.40
0.20
2
x1 and z are
perpendicular
1
Geometry
of Vectors
Here vectors x1, x2, and z from our
previous problem are orthogonal
3
-0.50
x2   0.50
 1.50
 0.10
z  -0.32
 0.14
1.00
x1  0.40
0.20
2
x2 and z are
perpendicular
1
Geometry
of Vectors
Here vectors x1, x2, and z from our
previous problem are orthogonal
3
-0.50
x2   0.50
 1.50
x1 and x2 are
perpendicular
 0.10
z  -0.32
 0.14
1.00
x1  0.40
0.20
2
1
Note we could rotate
x1, x2, and z until they
lied on our three axes!
Geometry
of Vectors
The projection (or shadow) of a vector x
on a vector y is given by:
x'y
y
2
Ly
If y has unit length (i.e., Ly = 1), the
projection (or shadow) of a vector x on a
vector y simplifies to (x’y)y
Geometry
of Vectors
For the vectors
0.6
0.8
x    ,y   
1.0 
0.3
the projection (or shadow) of x on y is:
0.8 
0.6
1.0

 0.3
x'y
y =
2
0.8 
Ly

0.8 0.3 0.3
0.8 

= 1.0685
=
0.3
0.8  =  0.78  0.8 

 0.3
0.3
0.73

 
0.8548 
0.3205 
Geometry
of Vectors
Geometrically the projection of x on y
2
looks like this:
projection of x on y
0.6
x   
1.0 
0.8
y   
0.3
Perpendicular wrt y
1
Geometry
of Vectors
Rotating so y lies on axis 1 makes it
2
easier to see:
1
Geometry
of Vectors
Note that we write the length of the
projection of x on y like this:
x'y
Ly
= Lx
x'y
 L x cosq xy
L xL y
For our previous example, the length of
the length of the projection of x on y is:
0.8 
0.6
1.0

 0.3
x'y
=
 0.912921
Ly

0.8 0.3 0.8
0.3
The Gram-Schmidt
(Orthogonalization) Process
For linearly independent vectors x1, x2,…,
xk, there exist mutually perpendicular
vectors u1, u2,…, uk with the same linear
span. These may be constructed by setting:
u1  x1
u2  x 2
x 2' u1
 '
u1
u1u1
uk  x k
x 2' u1
 '
u1 x1 
u1u1
x k' uk -1
 '
uk -1
uk -1uk -1
The Gram-Schmidt
(Orthogonalization) Process
We can normalize (convert to vectors z
of unit length) the vectors u by setting
zj 
uj
u'ju j
Finally, note that we can project a vector
xk onto the linear span of vectors x1,
x2,…, xk-1:
k -1

j=1
x'k z j

The Gram-Schmidt
(Orthogonalization) Process
Here are vectors x1, x2, and z from our
previous problem:
3
-0.50
x2   0.50
 1.50
 0.10
z  -0.32
 0.14
1.00
x1  0.40
0.20
1
2
The Gram-Schmidt
(Orthogonalization) Process
Let’s construct mutually perpendicular
vectors u1, u2, u3 with the same linear
span – we’ll arbitrarily select the first axis
as u1:
0.00
u1  0.00
1.00
The Gram-Schmidt
(Orthogonalization) Process
Now we construct a vector u2 perpendicular
with vector u1 (and in the linear span of x1,
x2, z):
u2
 1.50
0.00
1.50
  0.50  -0.50 0.00  0.50
-0.50
1.00
0.00
0.00


0.00
1.00
0.00
0.00
 1.50
1.00
  0.50 
0.00
-0.50
-0.50 0.50 1.50 0.00
1.00
0.00
0.00
1.00
The Gram-Schmidt
(Orthogonalization) Process
Finally, we construct a vector u3
perpendicular with vectors u1 and u2 (and
in the linear span of x1, x2, z):
The Gram-Schmidt
(Orthogonalization) Process
 0.14
0.00
1.50
 0.1025
 -0.32  0.10 0.00  0.25 0.50  -0.3325
 0.10
1.00
0.00
 0.0000
1.50
0.00 0.50 1.50 0.50
0.00

1.50
0.10 -0.32 0.14 0.50
0.00
0.00
1.00 0.00 0.00 0.00


 0.14
1.00
u1' u1
u2' u2
3
2
1
1
u  z 
ux 
u  -0.32 
1
2
0.00
z' u
z' u
 0.10
0.10 -0.32 0.14 0.00
1.00
1.50
0.50
0.00
0.00
0.00
1.00
The Gram-Schmidt
(Orthogonalization) Process
Here are our orthogonal vectors u1, u2,
and u3:
3
0.00
u2  0.50
1.50
 0.0000
u3  -0.3325
 0.1025
1.00
u1  0.00
0.00
2
1
The Gram-Schmidt
(Orthogonalization) Process
If we normalized our vectors u1, u2, and
u3, we get:
z1 
u1
u1' u1


1.0000
0.0000
0.0000
1.0000
1.0000 0.0000 0.0000 0.0000
0.0000
1.0
1.0
1.0000
1.0000
0.0000  0.0000
0.0000
0.0000
The Gram-Schmidt
(Orthogonalization) Process
and:
z2 
u2
u2' u2


0.0000
0.5000
1.5000
0.0000
0.0000 0.5000 1.5000 0.5000
1.5000
1.0
2.5
0.0000
0.0000
0.5000  0.3162
1.5000
0.9487
The Gram-Schmidt
(Orthogonalization) Process
and:
z3 
u3
u'3u3


 0.0000
-0.3325
 0.1025
 0.0000
0.0000 -0.3325 0.1025 -0.3325
 0.1025
1.0
0.1211
 0.0000
 0.0000
-0.3325  -0.9556
 0.1025
 0.2946
The Gram-Schmidt
(Orthogonalization) Process
The normalized vectors z1, z2, and z3
look like this:
3
0.0000
z2  0.3162
0.9487
 0.0000
z3  -0.9556
 0.2946
1.00
z1  0.00
0.00
2
1
Special Matrices
There are a number of special matrices.
These include
 Diagonal Matrices
 Identity Matrices
 Null Matrices
 Commutative Matrices
 Anti-Commutative Matrices
 Periodic Matrices
 Idempotent Matices
 Nilpodent Matrices
 Orthogonal Matrices
Diagonal Matrices
A diagonal matrix is a square matrix
that has values on the diagonal with all
off-diagonal entities being zero.
a11
0
0
0

0
a22
0
0
0
0
a33
0
0 
0 
0 
a44 
Identity Matrices
An identity matrix is a diagonal matrix
where the diagonal elements all equal 1
1
I  0
0
0
0
1
0
0
0
0
1
0
0
0
0
1 
When used as a premultiplier or
postmultiplier of any conformable
matrix A, the Identity Matrix will return
the original matrix A, i.e.,
IA = AI = A
Why?
Null Matrices
A square matrix whose elements all
equal 0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0 
Usually arises as the difference between
two equal square matrices, i.e.,
a–b=0a=b
Commutative Matrices
Any two square matrices A and B such
that AB = BA are said to commute.
Note that it is easy to show that any
square matrix A commutes with both
itself and with a conformable identity
matrix I.
Anti-Commutative Matrices
Any two square matrices A and B such
that AB = -BA are said to anticommute.
Periodic Matrices
Any matrix A such that Ak+1 = A is said
to be of period k.
Of course any matrix that commutes
with itself of period k for any integer
value of k (why?).
Idempotent Matrices
Any matrix A such that A2 = A is said to
be of idempotent.
Thus an idempotent matrix commutes
with itself if of period k for any integer
value of k.
Nilpotent Matrices
Any matrix A such that Ap = 0 where p
is a positive integer is said to be of
nilpotent.
Note that if p is the least positive
integer such that Ap = 0, then A is said
to be nilpotent of index p.
Orthogonal Matrices
Any square matrix A with rows
(considered as vectors) are mutually
perpendicular and have unit lengths,
i.e.,
A’A = I
Note that A is orthogonal iff A-1 = A’.
The Determinant of a Matrix
The determinant of a matrix A is
commonly denoted by |A| or det A.
Determinants exist only for square
matrices.
They are a matrix characteristic (that
can be somewhat tedious to compute).
The Determinant
for a 2x2 Matrix
If we have a matrix A such that
then
a
A   11
a21
a12 
a22 
A = a11a22 - a12a21
For example, the determinant of
is
A  1 2 
3 4 
A = 1 2  = a11a22 - a12a21  1 4  -  2  3  = -2
3 4 
Determinants for 2x2 matrices are easy!
The Determinant
for a 3x3 Matrix
If we have a matrix A such that
 a11
A  a21
a31

a12
a22
a32
a13 
a23 
a33 
Then the determinant is
det A = A = a11
a22
a32
a23
a
- a12 21
a33
a31
a23
a
+ a13 21
a31
a33
a22
a32
which can be expanded and rewritten as
det A = A = a11a22a33 - a11a23a32 + a12a23a31
- a12a21a33 + a13a21a32 - a13a22a31
(Why?)
The Determinant
for a 3x3 Matrix
If we rewrite the determinants for each of
the 2x2 submatrices in
det A = A = a11
as
a22
a32
a23
a
- a12 21
a33
a31
a23
a
+ a13 21
a31
a33
a22
a32
a23
=a22 a33 - a23 a32 ,
a33
a21
a31
a23
=a21a33 - a23 a31 , and
a33
a21
a31
a22
=a21a32 - a22 a31
a32
a22
a32
by substitution we have
A = a11a22a33 - a11a23a32 + a12a23a31 - a12a21a33 + a13a21a32 - a13a22a31
The Determinant
for a 3x3 Matrix
Note that if we have a matrix A such that
 a11
A  a21
a31

a12
a22
a32
a13 
a23 
a33 
Then |A| can also be written as
det A = A = a11
a22
a32
a23
a
- a12 21
a33
a31
det A = A = -a21
a12
a32
a13
a
+ a22 11
a33
a31
det A = A = a31
a12
a22
a13
a
a13
a
a12
- a32 11
+ a33 11
a23
a21 a23
a21 a22
or
or
a23
a
+ a13 21
a31
a33
a22
a32
a23
a
- a23 11
a31
a33
a12
a32
The Determinant
for a 3x3 Matrix
To do so first create a matrix of the
same dimensions as A consisting only
of alternating signs (+,-,+,…)
+ - +
- + -
+ - +
The Determinant
for a 3x3 Matrix
Then expand on any row or column (i.e.,
multiply each element in the selected
row/column by the corresponding sign,
then multiply each of these results by the
determinant of the submatrix that results
from elimination of the row and column
to which the element belongs
For example, let’s expand on the second
column
a
a
a 
11
A  a21
a31

12
a22
a32
13
a23 
a33 
The Determinant
for a 3x3 Matrix
The three elements on which our
expansion is based will be a12, a22, and
a32. The corresponding signs are -, +, -.
+ - +
- + -
+ - +
The Determinant
for a 3x3 Matrix
So for the first term of our expansion we
will multiply -a12 by the determinant of
the matrix formed when row 1 and
column 2 are eliminated from A (called
the minor and often denoted Arc where r
and c are the deleted rows and columns):
 a11 a12
A  a21 a22
a31 a32

which gives us
a13 
a21 a23 

a23 so A12  
a31 a33 


a33 
-a12
a21
a31
a23
a33
This product is called a cofactor.
The Determinant
for a 3x3 Matrix
For the second term of our expansion
we will multiply a22 by the determinant
of the matrix formed when row 2 and
column 2 are eliminated from A:
 a11 a12
A  a21 a22
a31 a32

a13 
a
a13
a23  so A22  11
a31 a33
a33 
which gives us
a22
a11
a31
a13
a33
The Determinant
for a 3x3 Matrix
Finally, for the third term of our expansion
we will multiply -a32 by the determinant of
the matrix formed when row 3 and
column 2 are eliminated from A:
 a11 a12
A  a21 a22
a31 a32

a13 
a
a13
a23  so A22  11
a21 a23
a33 
which gives us
-a32
a11
a21
a13
a23
The Determinant
for a 3x3 Matrix
Putting this all together yields
det A = A = -a12
a21
a31
a23
a
+ a22 11
a33
a31
a13
a
- a32 11
a33
a21
a13
a23
So there are nine distinct ways to
calculate the determinant of a 3x3 matrix!
These can be expressed as
m
det A = A =  aij  -1
j=1
i+j
n
Aij =  aij  -1
i=1
i+j
Aij
Note that this is referred to as the method
of cofactors and can be used to find the
determinant of any square matrix.
The Determinant for a 3x3
Matrix – An Example
Suppose we have the following matrix A:
1 2 3 
A = 2 5 4 
1 -3 -2 
Using row 1 (i.e., i=1), the determinant
is:
m
det A = A =  a1j  -1
j=1
1+j
A1j  1(2)  2(8)  3(11)  15
Note that this is the same result we would
achieve using any other row or column!
Some Properties
of Determinants
Determinants have several mathematical
properties useful in matrix manipulations:
 |A|=|A'|
 If each element of a row (or column) of A is
0, then |A|= 0
 If every value in a row is multiplied by k,
then |A| = k|A|
 If two rows (or columns) are interchanged
the sign, but not value, of |A| changes
 If two rows (or columns) of A are identical,
|A| = 0
Some Properties
of Determinants
 |A| remains unchanged if each element of a
row is multiplied by a constant and added to
any other row
 If A is nonsingular, then |A|=1/|A-1|, i.e.,
|A||A-1|=1
 |AB|= |A||B| (i.e., the determinant of a
product = product of the determinants)
 For any scalar c, |cA| = ck|A| where k is the
order of A
 Determinant of a diagonal matrix is simply
the product of the diagonal elements
Why are
Determinants Important?
Consider the small system of equations:
a11x1 + a12x2 = b1
a21x1 + a22x2 = b2
Which can be represented by:
Ax = b
where A =  a11 a12  , x =  x1  , and b = b1 
a21 a22 
x 2 
b2 
Why are
Determinants Important?
If we were to solve this system of
equations simultaneously for x2 we would
have:
a21(a11x1 + a12x2 = b1)
-a11(a21x1 + a22x2 = b2)
Which yields (through cancellation &
rearranging):
a21a11x1 + a21a12x2 - a11a21x1 - a11a22x2 =
a21b1 - a11b2
Why are
Determinants Important?
or (a11a2 - a21a12)x2 = a11b2 - a21b1
which implies
a11b2  a21b1
x2=
a11a22 - a12a21
Notice that the denominator is:
A = a11a22 - a12a21
Thus iff |A|= 0 there is either i) no
unique solution or ii) no existing solution
to the system of equations Ax = b!
Why are
Determinants Important?
This result holds true:
 if we solve the system for x1 as well; or
 for a square matrix A of any order.
Thus we can use determinants in
conjunction with the A matrix (coefficient
matrix in a system of simultaneous
equations) to see if the system has a
unique solution.
Traces of Matrices
The trace of a square matrix A is the sum
of the diagonal elements
Denoted tr(A)
We have
tr  A  =  a =  a
n
i=1
m
ii
j=1
For example, the trace of
is
n
A  1 2 
3 4 
tr  A  =  aii = 1 + 4 = 5
i=1
jj
Some Properties
of Traces
Traces have several mathematical
properties useful in matrix
manipulations:




For any scalar c, tr(cA) = c[tr(A)]
tr(A  B) = tr(A)  tr(B)
tr(AB) = tr(BA)
tr(B-1AB) = tr(A)
n
m
 tr  AA'  =  aij2
i=1 j=1
The Inverse of a Matrix
The inverse of a matrix A is commonly
denoted by A-1 or inv A.
The inverse of an n x n matrix A is the
matrix A-1 such that AA-1 = I = A-1A
The matrix inverse is analogous to a
scalar reciprocal
A matrix which has an inverse is called
nonsingular
The Inverse of a Matrix
For some n x n matrix A an inverse
matrix A-1 may not exist.
A matrix which does not have an
inverse is singular.
An inverse of n x n matrix A exists iff
|A|0
Inverse by
Simultaneous Equations
Pre or postmultiply your square matrix
A by a dummy matrix of the same
dimensions, i.e.,
AA 1
 a11 a12
 a21 a22
a31 a32

a13  a b c 
a23  d e f 
a33  g h i 
Set the result equal to an identity
matrix of the same dimensions as your
square matrix A, i.e.,
AA 1
 a11 a12
 a21 a22
a31 a32

a13  a b c  1 0 0 
a23  d e f   0 1 0 
a33  g h i  0 0 1 
Inverse by
Simultaneous Equations
Recognize that the resulting expression
implies a set of n2 simultaneous
equations that must be satisfied if A-1
exists:
a11(a) + a12(d) + a13(g) = 1,
a11(c) + a12(f) + a13(i) = 0;
a11(b) + a12(e) + a13(h) = 0,
a21(a) + a22(d) + a23(g) = 0,
a21(c) + a22(f) + a23(i) = 0;
a21(b) + a22(e) + a23(h) = 1,
a31(a) + a32(d) + a33(g) = 0,
a31(c) + a32(f) + a33(i) = 1.
a31(b) + a32(e) + a33(h) = 0,
or
or
Solving this set of n2 equations
simultaneously yields A-1.
Inverse by Simultaneous
Equations – An Example
If we have
1 2 3 
A = 2 5 4 
1 -3 -2
Then the postmultiplied matrix would be
AA
1
1 2 3  a b c 
 2 5 4  d e f 
1 -3 -2 g h i 
We now set this equal to a 3x3
identity matrix
1 2 3  a b c  1 0 0 
2 5 4  d e f   0 1 0 
1 -3 -2 g h i  0 0 1 
Inverse by Simultaneous
Equations – An Example
Recognize that the resulting expression
implies the following n2 simultaneous
equations:
1a + 2d + 3g = 1, 1b + 2e + 3h = 0, 1c + 2f + 3i = 0;
or
2a + 5d + 4g = 0, 2b + 5e + 4h = 1, 2c + 5f + 4i = 0;
or
1a - 3d - 2g = 0, 1b - 3e - 2h = 0, 1c - 3f - 2i = 1.
This system can be satisfied iff A-1
exists.
Inverse by Simultaneous
Equations – An Example
Solving the set of n2 equations
simultaneously yields:
a = -2/15, b = 1/3, c = 7/3,
d = -8/15, e = 1/3, f = -2/3’
g = 11/15, h =-1/3, i = -1/15
so we have that A-1
 -2
1
7 
3
15 
 15
-1
1
-2 
A =  -8
1115 -13 -115 
 15
3
15 
Inverse by Simultaneous
Equations – An Example
ALWAYS check your answer.
How? Use the fact that AA-1= A-1A =I
and do a little matrix multiplication!
 -2
1
7 
3
15  1 0 0 
1 2 3   15
-1
1
-2   0 1 0   I
AA = 2 5 4   -8
3x3
15
3
15
1 -3 -2  11
-1
-1  0 0 1 
 15
3
15 
So we have found A-1!
Inverse by the
Gauss-Jordan Algorithm
Augment your matrix A with an identity
matrix of the same dimensions, i.e.,
 a11
A | I  a21
a
 31
a12
a22
a32
a13 1 0 0 
a23 0 1 0 
a33 0 0 1 
Now we use valid Row Operations
necessary to convert A to I (and so A|I
to I|A-1)
Inverse by the
Gauss-Jordan Algorithm
Valid Row Operations on A|I
 You may interchange rows
 You may multiply a row by a scalar
 You may replace a row with the sum of that
row and another row multiplied by a scalar
(which is often negative)
Every operation performed on A must
be performed on I
Use valid Row Operations on A|I to
convert A to I (and so A|I to I|A-1)
Inverse by the Gauss-Jordan
Algorithm – An Example
If we have
1 2 3 
A = 2 5 4 
1 -3 -2
Then the augmented matrix A|I is
1 2 3 1 0 0 
A | I  2 5 4 0 1 0 
1 -3 -2 0 0 1 


We now wish to use valid row
operations to convert the A side of
this augmented matrix to I
Inverse by the Gauss-Jordan
Algorithm – An Example
Step 1: Subtract 2·Row 1 from Row 2
2 5

4 0
- 2 1 2
3 1
1 0 
0 0 
0 1 -2 -2 -1 0 


And substitute the result for Row 2 in A|I
1 2 3 1 0 0 
0 1 -2 -2 1 0 
1 -3 -2 0 0 1 


Inverse by the Gauss-Jordan
Algorithm – An Example
Step 2: Subtract Row 3 from Row 1
1

2
31 0
- 1 -3 -2 0 0
0

5
0 
1
5 1 0 -1
Divide the result by 5 and substitute for
Row 3 in the matrix derived in the
previous step 1 2 3 1 0 0 


0
1
-2
-2
1
0


0
1
1
1
-1
0

5
5 
Inverse by the Gauss-Jordan
Algorithm – An Example
Step 3: Subtract Row 2 from Row 3
0 1

1 1
5
 0 1 -2 - 2
0 0

3 11
0 -1 
5
1
0 
-1 -1 
5
5
Divide the result by 3 and substitute for
Row 3 in the matrix derived in the

previous step 1 2 3 1
0
0


0
1
-2
-2
1
0


0
0
1
11
-1
-1

15
3
15 
Inverse by the Gauss-Jordan
Algorithm – An Example
Step 4: Subtract 2·Row 2 from Row 1
1 2

3 1
 2 0 1 -2 - 2
1 0

0 0 
1 0 
7 5 -2 0 
Substitute the result for Row 1 in the
matrix derived in the previous step


1
0
7
5
-2
0


0
1
-2
-2
1
0


0
0
1
11
-1
-1

15
3
15 
Inverse by the Gauss-Jordan
Algorithm – An Example
Step 5: Subtract 7·Row 3 from Row 1
1 0

7 0 0

1 0

0 
7 5
-2
1 11
15
-1
-1 
3
15 
0 -2
1
7
15
3

15 
Substitute the result for Row 1 in the
matrix derived in the previous step
-2
1
7 

1
0
0
15
3
15 

1
0 
0 1 -2 -2
0 0 1 1115 -1 3 -115


Inverse by the Gauss-Jordan
Algorithm – An Example
Step 6: Add 2·Row 3 to Row 2
0 1 -2 - 2

2 0 0

0 1

1 11
15
-1
0 -8
1
15
0 
1
-1 
3
15 
3
-2

15 
Substitute the result for Row 2 in the
matrix derived in the previous step
-2
1
7 

3
15 
1 0 0 15
0 1 0 -8 15 1 3 -215
0 0 1 11
-1
-1 

15
3
15
Inverse by the Gauss-Jordan
Algorithm – An Example
Now that the left side of the augmented
matrix is an identity matrix I, the right
side of the augmented matrix is the
inverse of the matrix A (A-1), i.e.,
A 1
1
7 
 -2
3
15 
 15
1
-2 
  -8
15
3
15
11
-1
-1 
 15
3
15 
Inverse by the Gauss-Jordan
Algorithm – An Example
To check our work, let’s see if our result
yields AA-1 = I:
AA 1
1
7 
 -2
3
15  1 0 0
1 2 3   15
1
-2   0 1 0
 2 5 4   -8
1 -3 -2 1115 -13 -115 0 0 1 
 15
3
15 
So our work checks out!
Inverse by Determinants
Replace each element aij in a matrix A
with an element calculated as follows:
 Find the determinant of the submatrix
that results when the ith row and jth
column are eliminated from A (i.e., Aij)
 Attach the sign that you identified in the
Method of Cofactors
 Divide by the determinant of A
 After all elements have been replaced,
transpose the resulting matrix
Inverse by Determinants –
An Example
Again suppose we have some matrix A:
1 2 3 
A  2 5 4 
1 -3 -2
We have calculated the determinant of A
to be –15, so we replace element 1,1 with
11
1


 A11


  1 2 15   215
A

Similarly, we replace element 1,2 with
1 2
1


 A12


  1 8 15   815
A

Inverse by Determinants –
An Example
After using this approach to replace
each of the nine elements of A, The
eventual result will be
1
7 
 -2
3
15 
 15
 -8 15 1 3 -215 
11
-1
-1 
 15
3
15 
which is A-1!
Eigenvalues and Eigenvectors
For a square matrix A, let I be a
conformable identity matrix. Then the
scalars satisfying the polynomial equation
|A - lI| = 0 are called the eigenvalues
(or characteristic roots) of A.
The equation |A - lI| = 0 is called the
characteristic equation or the
determinantal equation.
Eigenvalues and Eigenvectors
For example, if we have a matrix A:
A  2 4 
 4 -4 
then
A  lI   2 4   l 1 0 
 4 -4 
0 1 
4  2  l 4  l  16  0
 2-l


4
-4 - l 
or
l 2  2l  24  0
which implies there are two roots or
eigenvalues -- l=-6 and l=4.
Eigenvalues and Eigenvectors
For a matrix A with eigenvectors l, a
nonzero vector x such that Ax = lx is
called an eigenvector (or characteristic
vector) of A associated with l.
Eigenvalues and Eigenvectors
For example, if we have a matrix A:
A  2 4 
 4 -4 
with eigenvalues l = -6 and l = 4, the
eigenvector of A associated with l = -6
is
 x1 
 2 4   x1 
Ax  lx 
 6  
 4 -4   x 2 
x 2 
 2x1  4x 2  6x1  8x1  4x 2  0
and
 4x1  4x 2  6x 2  4x1  2x 2  0
Fixing x1=1 yields a solution for x2 of –2.
Eigenvalues and Eigenvectors
Note that eigenvectors are usually
normalized so they have unit length, i.e.,
e
x
x'x
For our previous example we have:
e
x
x'x

 1
-2 
 1  1 
-2  
5


 -2

5

1

1 -2 

5 
-2 
Thus our arbitrary choice to fix x1=1 has
no impact on the eigenvector associated
with l = -6.
Eigenvalues and Eigenvectors
For matrix A and eigenvalue l = 4, we
have
x 
x 
Ax  lx   2 4   1   4  1 
 4 -4   x 2 
x 2 
 2x1  4x 2  4x1  2x1  4x 2  0
and
 4x1  4x 2  4x 2  4x1  8x 2  0
We again arbitrarily fix x1=1, which now
yields a solution for x2 of 1/2.
Eigenvalues and Eigenvectors
Normalization to unit length yields
e
x
x'x

 1
1 
 2
 1  1
1  1   2 
5
  2   2  
 1

5
5

1


1 1   1 
5 
4
2
2

 2
Again our arbitrary choice to fix x1=1 has
no impact on the eigenvector associated
with l = 4.
Quadratic Forms
A Quadratic From is a function
Q(x) = x’Ax
in k variables x1,…,xk where
 x1 
x 2 
x 
xk 
 
and A is a k x k symmetric matrix.
Quadratic Forms
Note that a quadratic form has only
squared terms and crossproducts, and so
can be written
k
k
Q  x    aij x i x j
Suppose we have
then
i1 j1
x 
x   1  and A  1 4 
0 2
x 2 
Q(x) = x'Ax = x12 + 4x1x 2 - 2x 22
Spectral Decomposition and
Quadratic Forms
Any k x k symmetric matrix can be
expressed in terms of its k eigenvalueeigenvector pairs (li, ei) as
k
A   l ie ie
i1
'
i
This is referred to as the spectral
decomposition of A.
Spectral Decomposition and
Quadratic Forms
For our previous example on eigenvalues
and eigenvectors we showed that
A  2 4 
 4 4 
has eigenvalues l1 = -6 and l2 = -4,
with corresponding (normalized)
eigenvectors
 1 
2 




5
5
e1 
, e2 
,
 -2

1 


5 
5 
Spectral Decomposition and
Quadratic Forms
Can we reconstruct A?
k
A   l ie ie
i1
'
i
 1 
2 
1

2




5
5
-2
 6
+4


 -2

 1  
5
5
5


5 
5 
 1 -2 
4 2 
 6  5 5   4  5 5    2 4   A
-2 5 4 5 
 2 5 1 5   4 4 




1

5 
Spectral Decomposition and
Quadratic Forms
Spectral decomposition can be used to
develop/illustrate many statistical results/
concepts. We start with a few basic
concepts:
- Nonnegative Definite Matrix – when any
k x k matrix A such that
0  x’Ax  x’ =[x1, x2, …, xk]
the matrix A and the quadratic form are
said to be nonnegative definite.
Spectral Decomposition and
Quadratic Forms
- Positive Definite Matrix – when any k x k
matrix A such that
0 < x’Ax  x’ =[x1, x2, …, xk]  [0, 0, …, 0]
the matrix A and the quadratic form are
said to be positive definite.
Spectral Decomposition and
Quadratic Forms
Example - Show that the following
quadratic form is positive definite:
2
1
2
2
6x + 4x - 4 2x1x 2
We first rewrite the quadratic form in
matrix notation:
Q(x) =  x1
 6
  x1 
-2
2
x 2  
= x'Ax



4  x 2 
-2 2
Spectral Decomposition and
Quadratic Forms
Now identify the eigenvalues of the
resulting matrix A (they are l1 = 2 and
l2 = 8).
 6

1 0 
-2
2
A  lI  

l

0 1 
-2
2
4





 6 - l -2 2   6  l  4  l   -2 2 -2 2  0
-2 2 4 - l
or
l 2  10l  16   l  2  l  8   0
Spectral Decomposition and
Quadratic Forms
Next, using spectral decomposition we
can write:
k
A   lieiei'  l1e1e1'  l 2e2e2'  2e1e1'  8e2e2'
i1
where again, the vectors ei are the
normalized and orthogonal eigenvectors
associated with the eigenvalues l1 = 2
and l2 = 8.
Spectral Decomposition and
Quadratic Forms
Sidebar - Note again that we can recreate
the original matrix A from the spectral
decomposition:
k
A   l ie ie i'
i1

 2



 2



 2 
3  1
3 2
2  +8 
-1 
-1   3
3 
2  
3
3 

3 
3 
1
 2
2 
 2   6
3
2 2   A
3   8 3
3
2
 2
  2 2 4 
2 
1
3
3 
3
3 

1
Spectral Decomposition and
Quadratic Forms
Because l1 and l2 are scalars,
premultiplication and postmultiplication
by x’ and x, respectively, yield:
'
'
'
1 1
'
'
2
2
1
2
2
x Ax  2x e e x  8x e2e x  2y + 8y  0
where
'
'
1
'
'
2
y1  x e1  e x1 and y 2  x e2  e x 2
At this point it is obvious that x’Ax is at
least nonnegative definite!
Spectral Decomposition and
Quadratic Forms
We now show that x’Ax is positive
definite, i.e.
x ' Ax  2y12 + 8y 22  0
From our definitions of y1 and y2 we have
'

 y1 
e1   x1 

or
y

Ex
'


 y 2  e2   x 2 
Spectral Decomposition and
Quadratic Forms
Since E is an orthogonal matrix, E’ exists.
Thus,
'
x Ey
But 0  x = E’y implies y  0 .
At this point it is obvious that x’Ax is
positive definite!
Spectral Decomposition and
Quadratic Forms
This suggests rules for determining if a
k x k symmetric matrix A (or
equivalently, its quadratic form x’Ax) is
nonegative definite or positive definite:
- A is a nonegative definite matrix iff li 
0, i = 1,…,rank(A)
- A is a positive definite matrix iff li > 0, i
= 1,…,rank(A)
Measuring Distance
Euclidean (straight line) distance – The
Euclidean distance between two points
x and y (whose coordinates are
represented by the elements of the
corresponding vectors) in p-space is
given by
d(x , y) 
x
 y1  +
2
1

+ xp  yp

2
Measuring Distance
For a previous example
3
0.0000
z2  0.3162
0.9487
 0.0000
z3  -0.9556
 0.2946
1.00
z1  0.00
0.00
1
2
the Euclidean (straight line) distances
are
Measuring Distance
d(z1 , z2 ) 
d(z1 , z3 ) 
d(z2 , z3 ) 
1.00  0.00 + 0.00  0.3162 + 0.00  0.9487
1.00  0.00 + 0.00  0.9556 + 0.00  0.2946
2
2
2
 1.414
2
2
2
 1.414
0.00  0.00 + 0.3162  0.9556 + 0.9487  0.2946  1.430
2
2
2
3
0.0000
z2  0.3162
0.9487
1.430
 0.0000
z3  -0.9556
 0.2946
1.414
1.414
2
1.00
z1  0.00
0.00
1
Measuring Distance
Notice that the lengths of the vectors
are their distances from the origin:
d(0, P) 
x

2
1
 0 +
2
1
x +
+x

+ xp  0

2
2
p
This is yet another place where the
Pythagorean Theorem rears its head!
Measuring Distance
Notice also that if we connect all points
equidistant from some given point z, the
result is a hypersphere with its center at
z and area of pr2:
2
In p=2 dimensions
this yields a circle
z
r
1
Measuring Distance
In p = 2 dimensions, we actually talk
about area. In p  3 dimensions, we talk
about volume - which is 4/3pr3 for this
n n2
problem or, more generally
r p
3
In p=3 dimensions
we have a sphere
z

n
 1  
2

r
1
2
Measuring Distance
Problem – What if the coordinates of a
point x (i.e., the elements of vector x)
are random variables with differing
variances?
Suppose
- we have n pairs of measurements on
two variables X1 and X2, each having a
mean of zero
- X1 is more variable than X2
- X1 and X2 vary independently
Measuring Distance
A scatter diagram of these data might
look like this:
Which point really lies
further from the origin
in statistical terms
(i.e., which point is
less likely to have
occurred randomly)?
2
1
Euclidean distance does not account for
differences in variation of X1 and X2!
Measuring Distance
Notice that a circle does not efficiently
inscribe the data:
2
r2
The area of the
ellipse is pr1r2.
r1
1
An ellipse does so much more efficiently!
Measuring Distance
How do we take the relative dispersions
on the two axes into consideration?
2
1
We standardize each value of Xi by
dividing by its standard deviation.
Measuring Distance
Note that the problem can extend
beyond two dimensions.
3
The area of the
ellipsoid is
(4/3)pr1r2r3 or
more generally
1
n
n2
r
p

i1

n
 1  
2

2
Measuring Distance
If we are looking at distances from the
origin D(0,P), we could divide coordinate
i by its sample standard deviation sii:
*
i
x =
xi
s ii
Measuring Distance
The resulting measure is called Statistical
Distance or Mahalanobis Distance:
d(0, P) 
x 
*
1
2
 x
  1
 s11

x1 has a relative
1
weight of k 1 
s11

2
1
x
+
s11
 
+
+ x
2
2
 x
+ p
 s pp


 +


+
*
p
x
2
p
s pp




2
x p has a relative
1
weight of k p 
s pp
Measuring Distance
Note that if we plot all points a constant
squared distance c2 from the origin:
c s 22
2
c s11
c s11
1
The area of
this ellipse is

p c s11
c
s 22

c s 22
all points that satisfy
2
2
x
x
d2  0, P   1 + + p =c 2
s11
s pp
Measuring Distance
What if the scatter diagram of these
data looked like this:
2
1
X1 and X2 now have an obvious positive
correlation!
Measuring Distance
We can plot a rotated coordinate system
~
~
on axes x1 and x2:
2
Q
1
This suggests that we calculate distance
~
~
based on the rotated axes x1 and x2.
Measuring Distance
The relation between the original
coordinates (x1, x2) and the rotated
~
~
coordinates (x1, x2) is provided by:
x1=x1 cos  q   x 2 sin  q 
x 2 =-x1 sin  q   x 2 cos  q 
Measuring Distance
Now we can write the distance from P =
~ ~
(x1, x2) to the origin in terms of the
original coordinates x1 and x2 of P as
d(0, P)  a11 x12 + 2a12 x1 x 2 + a22 x 22
where
a11 
cos 2  q 
cos 2  q  s11  2 sin  q  cos  q  s12  sin2  q  s 22

sin2  q 
cos 2  q  s 22  2 sin  q  cos  q  s12  sin2  q  s11
Measuring Distance
a22 
sin2  q 
cos  q  s11  2 sin  q  cos  q  s12  sin  q  s 22
2
2
cos  q 
2

cos 2  q  s 22  2 sin  q  cos  q  s12  sin2  q  s11
and
a12 
cos  q  sin  q 
cos  q  s11  2 sin  q  cos  q  s12  sin  q  s 22
2

2
sin  q  cos  q 
cos 2  q  s 22  2 sin  q  cos  q  s12  sin2  q  s11
Measuring Distance
Note that the distance from P = (x1, x2)
to the origin for uncorrelated coordinates
x1 and x2 is
d(0, P)  a11 x12 + 2a12 x1 x 2 + a22 x 22
for weights
1
aij 
s ij
Measuring Distance
What if we wish to measure distance
from some fixed point Q = (y1, y2)?
2
Q=(y1, y2)
1
_
_
In this diagram, Q = (y1, y2) = (x1, x2) is
called the centroid of the data.
Measuring Distance
The distance from any point p to some
fixed point Q = (y1, y2) is
2
P=(x1, x2)
Q=(y1, y2)
Q
1
d(P, Q)  a11  x1  y1  + 2a12  x1  y1  x 2  y 2  + a22  x 2  y 2 
2
2
Measuring Distance
Suppose we have the following ten
bivariate observations (coordinate sets
of (x1, x2)): Obs # x
x
1
2
3
4
5
6
7
8
9
10
xi
1
2
-3
-2
-1
-2
-3
-1
-3
-3
0
-2
-2.0
3
6
4
4
6
6
2
5
7
7
5.0
Measuring Distance
The plot of these points would look like
this:
2
Centroid (-2, 5)
1
The data suggest a positive correlation
between x1, and x2.
Measuring Distance
The inscribing ellipse (and major and
minor axes) look like this:
2
Q450
1
Measuring Distance
The rotational weights are:
a11 
 
 45  1.11  2 sin  45  cos  45  0.8   sin 45  2.89 
cos 2 450
cos 2

0
 
0
0
sin2  q 
   
2
0
 
cos 2 450  2.89   2 sin 45 0 cos 45 0  0.8   sin2 45 0 1.11 
 0.6429
Measuring Distance
and:
a22 
 
 45  1.11  2 sin  45  cos  45  0.8   sin  45  2.89 
sin2 450
cos 2

0
 
0
0
cos 2  q 
   
2
0
 
cos 2 450  2.89   2 sin 45 0 cos 45 0  0.8   sin2 45 0 1.11 
 0.6429
Measuring Distance
and:
a12





cos  45  1.11  2 sin  45  cos  45  0.8   sin  45  2.89 
cos  q  sin  45 

cos  45   2.89   2 sin  45  cos  45  0.8   sin  45  1.11 
sin 450 cos 450
2
0
0
0
2
0
2
0
0
2
0
 0.3571
0
0
Measuring Distance
So the distances of the observed points
from their centroid Q = (-2.0, 5.0) are:
Obs #
x1
x2
x1
x2
~
Euclidean
D(P,Q)
1
2
3
4
5
6
7
8
9
10
xi
-3
-2
-1
-2
-3
-1
-3
-3
0
-2
-2.0
3
6
4
4
6
6
2
5
7
7
5.0
0.0000
2.8284
2.1213
1.4142
2.1213
3.5355
-0.7071
1.4142
4.9497
3.5355
4.2426
5.6569
3.5355
4.2426
6.3640
4.9497
3.5355
5.6569
4.9497
6.3640
2.2361
1.0000
1.4142
1.0000
1.4142
1.4142
3.1623
1.0000
2.8284
2.0000
~
Mahalanobis
D(P,Q)
1.3363
0.8018
1.4142
0.8018
1.4142
0.7559
2.0702
0.8018
1.5119
1.6036
Measuring Distance
Mahalonobis distance can easily be
generalized to p dimensions:
d(P, Q) 
p
 a x
i1
ii
j-1
i
p

 y i  + 2 aij  x i  y i  x j  y j
2
i1 j2
and all points satisfying
p
j-1
p


2
a
x

y
+
2
a
x

y
x

y

c
 ii  i i   ij  i i  j j
i1
2
i1 j2
form a hyperellipsoid with centroid Q.

Measuring Distance
Now let’s backtrack – the Mahalonobis
distance of a random p dimensional
point P from the origin is given by
d(0, P) 
p
j-1
p
2
a
x
 ii i + 2 aij x ix j
i1
i1 j 2
p
j-1
so we can say
p
d2 (0, P)   aii x i2 + 2 aij x i x j
i1
i1 j2
provided that d2 > 0  x  0.
Measuring Distance
Recognizing that aij = aji, i  j, i =
1,…,p, j = 1,…,p, we have
0<d2 (0, P)   x1 x 2
for x  0.
 a11 a12
a21 a22
x p  
a a
 p1 p2
a1p   x1 
a2p   x 2 
    x'Ax
app   x p 
Measuring Distance
Thus, p x p symmetric matrix A is positive
definite, i.e., distance is determined from
a positive definite quadratic form x’Ax!
We can also conclude from this result that
a positive definite quadratic form can be
interpreted as a squared distance!
Finally, if the square of the distance from
point x to the origin is given by x’Ax,
then the square of the distance from point
x to some arbitrary fixed point m is given
by (x-m)’A (x-m).
Measuring Distance
Expressing distance as the square root
of a positive definite quadratic form
yields an interesting geometric
interpretation based on the eigenvalues
and eigenvectors of A.
For example, in p = 2 two dimensions
all points
x 
x 1
 x 2 
that are constant distance c from the
origin must satisfy
x ' Ax  a11x12 + 2a12 x1x 2 + a22 x 22  c
Measuring Distance
By the spectral decomposition we have
k
A   lieie  l1e1e  l 2e2e
'
i
i1
'
1
'
2
so by substitution we now have

'
'
x Ax  l1 x e1

2

'
 l 2 x e2

2
 c2
and A is positive definite, so l1 > 0 and
l2 >0, which means
2

'
c  l1 x e1
is an ellipse.

2

'
 l 2 x e2

2
Measuring Distance
Finally, a little algebra can be used to
show that
x
c
l1
e1
satisfies
2
 c

x ' Ax  l1 
e1' e1   c 2
 l1



Measuring Distance
Similarly, a little algebra can be used to
show that
x
c
l2
e2
satisfies
2
 c

x ' Ax  l 2 
e '2e 2   c 2
 l2



Measuring Distance
So the points at a distance c lie on an
ellipse whose axes are given by the
eigenvectors of A with lengths
proportional to the reciprocals of the
square roots of the corresponding
eigenvalues (with constant of
proportionality c) x c
2
e2
This generalizes
to p dimensions
l1
c
l2
e1
x1
Square Root Matrices
Because spectral decomposition allows
us to express the inverse of a square
matrix in terms of its eigenvalues and
eigenvectors, it enables us to
conveniently create a square root
matrix.
Let A be a p x p positive definite matrix
with the spectral decomposition
k
A   l ieiei'
i1
Square Root Matrices
Also let P be a matrix whose columns
are the normalized eigenvectors e1, e2,
…, ep of A, i.e.,
Then
P  e 2
e p 
e2
k
A   l ieiei'  P ' P
i1
where P’P = PP’ = I and
l1
0

0

0
l2
0
0
0
0

l p 
Square Root Matrices
Now since
(P-1P’)PP’=PP’(P-1P’)=PP’=I
we have
1
A  P P   eiei'
i1 l i
-1
Next let
 l1
1

2   0

 0

1
'
0
l2
0
k
0 

0 

lp 

Square Root Matrices
The matrix
1
2
k
P P '   li eiei'  A
i1
is called the square root of A.
1
2
Square Root Matrices
The square root of A has the following
properties:

 A


1
2
1
2
'
1

  A 2

1
2
 A A A
-1
2
1
2
-1
2
-1
2
1
2
 A A A A
 A A
A
1
-1
2
I
where A
-1
2

 A


1
2



-1
Square Root Matrices
Next let -1 denote the matrix matrix
whose columns are the normalized
eigenvectors e1, e2, …, ep of A, i.e.,
Then
P  e 2
e p 
e2
k
A   l ieiei'  P ' P
i1
where P’P = PP’ = I and
l1
0

0

0
l2
0
0
0
0

l p 
Singular Value Decomposition
We can extend the operations of spectral
decomposition for a rectangular matrix by
using the eigenvalues and eigenvectors
from the square matrix A’A or (AA’).
Suppose A is an m x k real matrix. There
exists an m x m orthogonal matrix U and
a k x k orthogonal matrix V such that
singular values of A
A = UV’
where  has ith diagonal element li  0
for i = 1, 2,…, min (m,k) and 0 for all
other elements.
Singular Value Decomposition
Singular Value decomposition can also be
expressed as a matrix expansion that
depends on the rank r of A.
There exist r
- positive constants l1, l2,…, lr,
- orthogonal m x 1 unit vectors u1, u2, …,
ur,
- orthogonal k x 1 unit vectors v1, v2, …, vr,
such that
k
A   liui v i'  Ur  r Vr'
i1
Singular Value Decomposition
k
A   liui v  Ur  r V
i1
'
i
'
r
where
- Ur = [u1, u2, …, ur]
- Vr = [v1, v2, …, vr]
-  is an r x r with diagonal entries l1,
l2,…, lr and off diagonal entries 0
Singular Value Decomposition
We can show that AA’ has eigenvalueeigenvector pairs (li, ui), so
2
i
AA'ui  l ui   iui
with
li>0, i=1,...,r and li=0, i=r+1,...,m (for m>k)
then
1
i
vi  l A'ui
Singular Value Decomposition
Alternatively, we can show that A’A has
eigenvalue-eigenvector pairs (li, vi), so
2
i
A'Avi  l vi   i vi
with
li>0, i=1,...,r and li =0, i=r+1,...,k (for k>m)
then
1
i
ui  l Avi   i Avi
Singular Value Decomposition
Suppose we have a rectangular matrix
then

2

1

1

A
1 2 2 
 2 1

2

1

1


6

2

'
1 2 
AA 
1 2 2  1 2 2 9 


Singular Value Decomposition
and AA’ has eigenvalues of 1=5 and
2=10 with corresponding normalized
eigenvectors
 2 
 -1 




5
5
u1 
, u2 
 1

 2



5 
5 
Singular Value Decomposition
Similarly for our rectangular matrix
then

2

1

1

A
1 2 2 
 2 1
5 0 0 

2

1

1

'
A A  1 2
 0 5 5
1 2 1 2 2  0 5 5
Singular Value Decomposition
and A’A also has eigenvalues of 1=0,
2=5 and 3=10 with corresponding
normalized eigenvectors




 0 
 0 
1 

v 1   1  , v 2  0  , v 3   1


2
2
0 
1 
-1 


2
2
Singular Value Decomposition
Now taking
l1 =
1 = 0, l 2 =
 2 = 5, l 3 =
 3 = 10
we find that the singular value
decomposition of A is
 2 
5  1 0 0 
A  2 1 1  5 

1 2 2 
 1


5 
 -1 



5
1
 10
0
 2
 
2

5 
1

2 
Singular Value Decomposition
The singular value decomposition is
closely connected to the approximation of
a rectangular matrix by a lower-dimension
matrix [Eckart and Young, 1936]:
First note that if a m x k matrix A is
approximated by B of same dimension
but lower rank,
m
k
  a
i1 j1
ij
- bij

2
'

= tr  A - B  A - B  


Singular Value Decomposition
Eckart and Young used this fact to show
that, for a m x k real matrix A with m  k
and singular value decomposition UV’,
s
B   l iui v
i1
'
i
is the rank-s least squares approximation
to A (where s < k = rank (A)).
Singular Value Decomposition
This matrix B minimizes
m
k
  a
i1 j1
ij
- bij

2
'

= tr  A - B  A - B  


over all m x k matrices of rank no greater
than s!
It can also be shown that the error of this
approximation is k
2
l
 i
i s 1
Random Vectors and Matrices
Random Vector – vector whose individual
elements are random variables
Random Matrix – matrix whose individual
elements are random variables
Random Vectors and Matrices
The expected value of a random vector or
matrix is the matrix containing the
expected values of the individual
elements, i.e.,
E  x 11  E  x 12 
 
  
E  x  E  x 
21
 22 
E  X     


E  x n1  E  x n2 
E  x 1p  

E  x 2p  



E  x np  

Random Vectors and Matrices
where
 
  x ijpij x ij
 all x ij



E  x ij    
 x ij fij x ij dx ij



 
Random Vectors and Matrices
Note that for random matrices X and Y
of the same dimension, conformable
matrices of constants A and B, and
scalar c



E(cX) = cE(X)
E(X+Y) = E(X) + E(Y)
E(AXB)= AE(X)B
Random Vectors and Matrices
Mean Vector – random vector whose
elements are the means of the
corresponding random variables, i.e.,
  x ipi  x i 
 all xi

E  x i   mi   
 x i fi  x i  dx i



Random Vectors and Matrices
In matrix notation we can write the
mean vector as
E  x1    m 

  1
E  x 2   m 2 
E  X   
 m

  
E  x   mp 
  p    
Random Vectors and Matrices
For the bivariate probability distribution
X1
X2
-4
1
3
p1(x 1)
1
2
0.15
0.10
0.20
0.25
0.05
0.25
0.40
0.60
p2(x 2)
0.25
0.45
0.30
1.00
the mean vector is
  x1p1  x 1  
E  x1   m   all x

1
1
 
m  E  X   

E  x 2   m 2    x 2p 2  x 2  
 all x 2


0.4(1)  0.6(2)
 1.60 



0.1(4)  0.25(1)  0.25(3)  0.35
Random Vectors and Matrices
Covariance Matrix – random symmetric
vector whose diagonal elements are
variances of the corresponding random
variables, i.e.,
2

x i  m i  pi  x i 


 all x i
2

2


E  x i  m i   i   


  x  m 2 f  x  dx
 i i i i i


Random Vectors and Matrices
and whose off-diagonal elements are
covariances of the corresponding
random variable pairs, i.e.,
    x i  mi  x k  mk  pik  x i , x k 
 all xi all xk



E  x i  mi  x k  mk   ik    

x i  mi  x k  mk  fik  x i , x k  dx idx k



 
notice that if we this expression, when i
= k, returns the variance, i.e.,
ii  i2
Random Vectors and Matrices
In matrix notation we can write the
covariance matrix as
   x1  m1  



x

m



'
2
2 
E  X - m  X - m    E  
  x 1  m1 






  x p  mp 

x
2
 m2 


 x  m  2 
E
E  x1  m1  x 2  m 2  
1

 1


 x  m  2 
E  x 2  m 2  x1  m1  
E
2

 2

 


E  x  m  x  m   E  x  m  x  m  
p
1
1
p
2
2

 p

  p








x p  mp  








E  x1  m1  x p  mp  


  11 12
E  x 2  m 2  x p  mp   21 22

 
 
 
 p1 p2
2



E x p  mp





1p 

2p 



pp 
Random Vectors and Matrices
For the bivariate probability distribution
we used earlier
X1
X2
-4
1
3
p1(x 1)
1
2
0.15
0.10
0.20
0.25
0.05
0.25
0.40
0.60
p2(x 2)
0.25
0.45
0.30
1.00
the covariance matrix is

 x  m  2 
 x 1  m1  x 2  m 2   
E
E
1
1



'


E  X - m  X - m    



2


E  x 2  m 2  x 1  m1  

E  x 2  m2 



 

2

x

m
p1  x 1 


 x1  m1  x 2  m2  p12  x1 , x 2 



1
1

 11 12 
all x1
all x1 all x 2





2


    x 2  m 2  x1  m1  p 21  x 2 , x 1 

22 
 21
  x 2  m2  p2  x 2 
 all x 2 all x1

all x 2
Random Vectors and Matrices
which can be computed as
2

x

m


 1 1 p1  x1 

'
all x1
E  X - m  X - m    

 
   x 2  m2  x1  m1  p21  x 2 , x1 
 all x 2 all x1
  x
all x1 all x 2


2
2

1  1.6   0.4    2  1.6  0.6 



  4  0.35 1  1.6  0.15    4  0.35  2  1.6  0.1

  1  0.35 1  1.6  0.2   1  0.35  2  1.6  0.25 

   3  0.35 1  1.6  0.05    3  0.35  2  1.6  0.25 
0.24 0.39  11 12 



0.39 7.0275  21 22 
 m1  x 2  m 2  p12  x1 , x 2  


2

 x 2  m2  p2  x 2 


all x 2
1
1  1.6  4  0.35  0.15   1  1.6 1  0.35 0.2 
 1  1.6  3  0.35  0.05    2  1.6  4  0.35  0.1
  2  1.6 1  0.35  0.25    2  1.6  3  0.35  0.25 








2
2
2
 4  0.35  0.25   1  0.35  0.45    3  0.35   0.3


Random Vectors and Matrices
Thus the population represented by the
bivariate probability distribution
X1
X2
-4
1
3
p1(x 1)
1
2
0.15
0.10
0.20
0.25
0.05
0.25
0.40
0.60
p2(x 2)
0.25
0.45
0.30
1.00
Have population mean vector and
variance-covariance matrix
1.60 
m
 and  
0.35
0.24 0.39 


0.39
7.0275


Random Vectors and Matrices
We can use the population variancecovariance matrix S to calculate the
population correlation matrix r.
Individual population correlation
coefficients are defined as
rik 
 ik
 ii  kk
Random Vectors and Matrices
In matrix notation we can write the
correlation matrix as

11

 11 11

21

r   22 11



p1

 
 pp 11
12
11 22
22
22 22
p2
pp 22
1p


11 pp 
 r11 r12
2p
 r
rik
ik


22 pp 
 
 
 rp1 rp2
pp

pp pp 

r1p 

r2p 


rpp 
Random Vectors and Matrices
We can easily show that
  V1 2rV1 2
where
V
12
 11

 0



 0
which implies

r V
12
0 

0 



pp 

0
 22
0
  V 

12

Random Vectors and Matrices
For the bivariate probability distribution
we used earlier
X1
X2
-4
1
3
p1(x 1)
1
2
0.15
0.10
0.20
0.25
0.05
0.25
0.40
0.60
p2(x 2)
0.25
0.45
0.30
1.00
the variance matrix is
V
12
 11

 0

0   0.24

22   0
 0.48989
0 

 

0
2.65094
7.0275  

0
Random Vectors and Matrices
so the population correlation matrix is

r V
12
  V 

12

2.04124
0
 0.24 0.39  2.04124
0





0
0.37739
0.39
7.0275
0
0.37739




1.00 0.30 


0.30 1.00 
Random Vectors and Matrices
We often deal with variables that
naturally fall into groups. In the simplest
case, we have two groups of size q and
p – q of variables.
Under such circumstances, it may be
convenient to partition the matrices and
vectors.
Random Vectors and Matrices
Here we have a mean vector and
variance-covariance matrix:
 11
 m1 

 

 



 q1
 mq   m 

   
m  ___    ___  ,   

 m     
q1
m
 q1,1
   

 

m 
 p1
 p 
1p
1,q1
qq
q,q1
q1,q
q1,q1
pq
p,q1
1p 


qp   11
 

q1,p   21


pp 
12 


 22 
Random Vectors and Matrices
Rules for Mean Vectors and Covariance
Matrices for Linear Combinations of
Random Variables: The linear
combination of real constants c and
random variables X
 x1 
c'X  c1
c2
has mean vector
 
p
x
2
c p       c j X j
  j1
 
 x p 
E  c'X   c'E  X   c'm
Random Vectors and Matrices
The linear combination of real constants
c and random variables X
c'X  c1
c2
 x1 
 
p
x
2
c p       c j X j
  j1
 
 x p 
also has variance
Var  c'X   c'  c
Random Vectors and Matrices
Suppose, for example, we have random
vector
 X1 
X= 
X2 
and we want to find the mean vector
and covariance matrix for the linear
combinations:
Z1 = X1 - X 2 ,
Z 2 = X1 + X 2
 Z1  1 1  X1 
i.e., Z=    
    CX
Z 2  1 1   X 2 
Random Vectors and Matrices
We have mean vector
1 1 m1  m1  m 2 
m Z = E Z  = Cm X  

   
1 1  m 2  m1  m 2 
and covariance matrix
Note what happens
if 11=22 – the off
diagonal terms
vanish (i.e., the
sum and difference
of two random
variables with
identical variances
are uncorrelated)
1 1 11
 Z = Cov Z  = C  X C  

1 1  21
'
11  212  22

 11  22
11  22


11  212  22 
12 

22 
Matrix Inequalities
and Maximization
These results will be useful in later
applications:
- Cauchy-Schwartz Inequality – Let b and
d be any two p x 1 vectors. Then
(b’d)2  (b’b)(d’d)
with equality iff b=cd or (or d=cb) for
some constant c.
Matrix Inequalities
and Maximization
- Extended Cauchy-Schwartz Inequality –
Let b and d be any two p x 1 vectors and
B be a p x p positive definite matrix.
Then
(b’d)2  (b’Bb)(d’B-1d)
with equality iff b=cB-1d or (or d=cB -1d)
for some constant c.
Matrix Inequalities
and Maximization
- Maximization Lemma – let d be a given
p x 1 vector and B be a p x p positive
definite matrix. Then for an arbitrary
nonzero vector x
max
x 0
 x ' d
2
x ' Bx
 d ' B 1d
with the maximum attained when x =
cB-1d for any constant c.
Matrix Inequalities
and Maximization
- Maximization of Quadratic Forms for Points
on the Unit Sphere – let B be a p x p
positive definite matrix with eigenvalues l1
 l2   lp and associated eigenvectors
e1, e2, ,ep. Then
x ' Bx
 l1 (attained when x=e1 )
max
x 0
x'x
x ' Bx
 l p (attained when x=e p )
min
x 0 x ' x
Matrix Inequalities
and Maximization
and
x ' Bx
 lk+1 (attained when x  ek+1 , k  1,2,
max
x e1 , ek x ' x
,p-1)
Summary
There are MANY other Matrix Algebra
results that will be important to our study
of statistical modeling. This site will be
updated occasionally to reflect these
results as they become necessary.