Matrix Norm

We will denote K a field of either real or complex numbers.

Let K m x n denote the vector space of all matrices of size m × n (m rows and n columns) with
entries in the field K .

=
Let A (=
aij ), i 1;=
m, j 1; n be matrix.

=
The absolute value (modulus) of this matrix =
is | A | (| aij |), so | −A| | A| .

If A and B matrices have the same order, so

1. | A ⋅ B |≤| A | ⋅ | B |
2. | A + B |≤| A | + | B |
3. | α A |=
| α | ⋅ | A |, α ∈ R

A matrix norm is a norm on the vector space K m x n . Thus, the matrix norm is a function
||||: K m x n  R that must satisfy the following properties:

For all scalars α ∈ K and for all matrices A, B ∈ K mxn ,

|| A ||≥ 0 (being positive-valued), || A ||= 0 iff A=0 (being definite)

1. || α A ||≤| α | ⋅ || A ||, α ∈ R (being absolutely homogeneous)

2. || A + B ||≤|| A || + || B || (being sub-additive or satisfying the triangle inequality)

Additionally, in the case of square matrices (thus, m = n), some (but not all) matrix norms satisfy
the following condition, which is related to the fact that matrices are more than just vectors:

3. || A ⋅ B ||≤|| A || ⋅ || B ||
Norm is called canonical if in addition we have these two conditions:
4. | aij |≤|| A ||

5. If | A |≤| B |, then || A ||≤|| B || .

Further for the matrix we will use these three norms:

Aram YESAYAN, PhD, Associate Professor

1
 𝑚𝑎𝑥
1. �|𝐴|� = 1≤𝑖≤𝑚(|𝑎𝑖1 | + |𝑎𝑖2 | + ⋯ + |𝑎𝑖𝑛 |), (𝑚 − 𝑛𝑜𝑟𝑚)
𝑚
𝑚𝑎𝑥
2. �|𝐴|� = 1≤𝑗≤𝑛 (|𝑎1𝑗 | + |𝑎2𝑗 | + ⋯ + |𝑎𝑚𝑗 |), (𝑙 − 𝑛𝑜𝑟𝑚)
𝑙

3. ||𝐴||𝑘 = �∑𝑚 𝑛
𝑖=1 ∑𝑗=1 𝑎𝑖𝑗 (𝑘 − 𝑛𝑜𝑟𝑚)
2

3 −1 0 1
Example 𝐴 = � �
4 2 −3 2

�|𝐴|�𝑚 = max(3 + 1 + 0 + 1; 4 + 2 + 3 + 2)

�|𝐴|�𝑙 = max(7; 3; 3; 3)

||𝐴||𝑘 = �32 + 12 + ⋯ 22

Rotation and Reflection

In linear algebra, two vectors in an inner product space are orthonormal if they
are orthogonal (or perpendicular along a line) unit vectors. A set of vectors form
an orthonormal set if all vectors in the set are mutually orthogonal and all of unit length. An
orthonormal set which forms a basis is called an orthonormal basis.

An orthogonal matrix, or orthonormal matrix, is a real square matrix whose columns and
rows are orthonormal vectors.

One way to express this is

Q=
T
=
Q QQ T
I
where QT is the transpose of Q and I is the identity matrix.

This leads to the equivalent characterization: a matrix Q is orthogonal if its transpose is equal
to its inverse:

QT = Q −1 where Q −1
is the inverse of Q.

Rotation in mathematics is a concept originating in geometry. Any rotation is a motion of a

certain space that preserves at least one point. It can describe, for example, the motion of a rigid

Aram YESAYAN, PhD, Associate Professor

2
body around a fixed point. A rotation is different from other types of motions: translations,
which have no fixed points, and (hyperplane) reflections, each of them having an entire (n − 1)-
dimensional flat of fixed points in a n-dimensional space. A clockwise rotation is a negative
magnitude so a counterclockwise turn has a positive magnitude.

Mathematically, a rotation is a map. All rotations about a fixed point form

a group under composition called the rotation group (of a particular space). But
in mechanics and, more generally, in physics, this concept is frequently understood as
a coordinate transformation (importantly, a transformation of an orthonormal basis), because for
any motion of a body there is an inverse transformation which if applied to the frame of
reference results in the body being at the same coordinates. For example, in two dimensions
rotating a body clockwise about a point keeping the axes fixed is equivalent to rotating the axes
counterclockwise about the same point while the body is kept fixed. These two types of rotation
are called active and passive transformations.

In two dimensions, to carry out a rotation using a matrix, the point (x, y) to be rotated
counterclockwise is written as a column vector, and then multiplied by a rotation
matrix calculated from the angle θ:

 x′   cos θ − sin θ  x 
 ′ =   
  
y sin θ cos θ  y 

Proof:

Aram YESAYAN, PhD, Associate Professor

3
 x′ r cos(α +=
= θ ) r (cos α cos θ − sin α sin θ )
y′ r sin(α + =
= θ ) r (sin α cos θ + cos α sin θ )
So

=x′ x cos θ − y sin θ

=y′ x sin θ + y cos θ
Thus, it is proved.

We denote the rotation matrix by

 cos θ − sin θ 
Rθ =  
 sin θ cos θ 

Rotation of axes

In mathematics, a rotation of axes in two dimensions is a mapping from an xy-Cartesian

coordinate system to an x'y'-Cartesian coordinate system in which the origin is kept fixed and
the x' and y' axes are obtained by rotating the x and y axes counterclockwise through an angle A
point P has coordinates (x, y) with respect to the original system and coordinates (x', y') with
respect to the new system.

So applying trigonometry as above we get

Aram YESAYAN, PhD, Associate Professor

4
 x′   cos θ sin θ  x  x −1  x 
=
 ′  =   R=−θ   Rθ  
 y   − sin θ cos θ  y   y  y

Reflection

In Geometry, a reflection is known as a flip. A reflection is a mirror image of the shape. An

image will reflect through a line, known as the line of reflection. A figure is said to be a
reflection of the other figure, then every point in a figure and each corresponding point in
another figure are at equidistant from the line of reflection. The reflected image should have the
same shape and size, but the image faces in the opposite direction. In reflection, translation may
also take place because of its changes in the position. Here, the original image is called pre-
image, and its reflection is called image.

In a plane (or, respectively, 3-dimensional) geometry, to find the reflection of a point drop
a perpendicular from the point to the line (plane) used for reflection, and extend it the same
distance on the other side. To find the reflection of a figure, reflect each point in the figure.

Reflection over the straight line y=ax+b

At first, let us to reflect the point (x1, y1) across the OX axis

Aram YESAYAN, PhD, Associate Professor

5
Therefore, we will find out that

 x1′   x1   1 0
 ′ X=
r   , where X r  
 y1   1
y  0 − 1 
If we rotate the straight line clockwise by θ, we will get parallel line to the OX axis going
through (0, b).

Therefore, it is logic to move down everything by b.

Thus

 x   0
 + .
 y   −b 
After that, we rotate it clockwise by θ

 x   0
R−θ    +   
  y   −b  
Further, we apply X r reflection matrix

Aram YESAYAN, PhD, Associate Professor

6
  x   0
X r R−θ    +   
  y   −b  
Now we need move everything back that were originally. We rotate everything back

 x   0
Rθ X r R−θ    +   
  y   −b  
Remember that we moved everything down by b, so we have to move up

0  x   0
  + Rθ X r R−θ    +   
b   y   −b  
Finally, we get

 x′   0   x   0
 ′ =   + Rθ X r R−θ    +    ,
 y  b   y   −b  
Doing matrix algebra and using trigonometry we obtain

 x′   x cos 2θ + ( y − b) sin 2θ 
 ′ =  
  
y x sin 2θ − ( y − b ) cos 2θ + b 
In particular case when b=0

 x′  x
 ′  = Re f (θ )   ,
y   y
 cos 2θ sin 2θ 
where Re f (θ ) =  .
 sin 2θ − cos 2θ 

Aram YESAYAN, PhD, Associate Professor

7
 Gauss Algorithm

Gauss Algorithm for the determinant calculation:

Using Gauss method we can bring the matrix A to the matrix B with ones on the main
diagonal, so

1
Det B = 1= (1) (2) ( n −1)
.Det A
a a a ...a
11 22 33 nn
Example

Gauss Algorithm for finding the inverse of the matrix:

2 1
Find the inverse of A =  
1 2

 2 1   x11 x12   1 0 
 ⋅ x =
x22   0 1 
 1 2   21

Aram YESAYAN, PhD, Associate Professor

8
So

 2 −1 
 x11 x12   3 3 
 = .
 x21 x22   −1 2 
 
 3 3 

Approximate Solution of System of Linear Equations

Sequential Approximation (Iteration) Method:

We are given the system of the linear equations:

a11 x1 + a12 x2 + ... + a1n xn = b1

a x + a x + ... + a x =
 21 1 22 2 2n n b2

....................................
an1 x1 + an 2 x2 + ... + ann xn = bn (1)

Suppose that aii ≠ 0 fot all i

So by dividing the i-th equation by aii ≠ 0 , we get

Aram YESAYAN, PhD, Associate Professor

9
  x1 = β1 + α12 x2 + ... + α1n xn
 x = β + α x + ... + α x
 2 2 21 1 2n n

....................................
 xn = β n + α n1 x1 + ... + α nn −1 xn −1 (2)

The last system is easy for applying the iteration method. We do these notations:

 0 α12 ... α1n   β1   x1 

     
 α 21 0...α 2 n  =β2   x2 
=α = ;β ;X
 ..............   ...   ... 
     
 α n1 α n 2 ...0   βn   xn  .

So

X= β + α X (3).

We solve it using the sequential approximation method.

We take the free column as an initial solution of the system (2):

 x1(0)   β1 
 (0)   
 x2   β 2 
 ...  =  ...  (4)
 
 x (0)   β n 
 n 

So using (3) we get the first approximation:

X (1)= β + α X (0)

Further

β + α X (1) ... X ( k +1) =

X (2) = β + α X (k )

If the sequence of column-vectors

X (0) , X (1) ,..., X ( k ) ,...

has a limit, so it will be the approximate solution of the system (2) (so of the system (1)).

Aram YESAYAN, PhD, Associate Professor

10
Example: To solve this system to within accuracy of 10-4.

4 x1 + 0.24 x2 − 0.08 x3 =
8

0.09 x1 + 3 x2 − 0.15 x3 =
9
0.04 x − 0.08 x + 4 x =
 1 2 3 20

So we get

 x1 =
2 − 0.06 x2 + 0.02 x3

 x2 =
3 − 0.03 x1 + 0.05 x3
x =
 3 5 − 0.01x1 + 0.02 x2

 0 − 0.06 0.02   2  x1 
     
α=
 −0.03 0 0.05  ; β =
3 ; X =
 x2 
 −0.01    x 
 0.02 0  5  3

We take

 2
 
X (0)
=  3  and we get
5
 

x1(1) = 2 − 0.06 ⋅ 3 + 0.02 ⋅ 5 = 1.92

x2(1) = 3.19
x3(1) = 5.04
We continue and in the next step we will get
x1(2) = 1.9094
x2(2) = 3.1944
x3(2) = 5.0446

Aram YESAYAN, PhD, Associate Professor

11
In the next step we get
x1(3) ≈ 1.90923
x2(3) ≈ 3.19495
x3(3) ≈ 5.04479

So

x1 ≈ 1.9092
x2 ≈ 3.1949
x3 ≈ 5.0448

Gauss-Seidel Method:

This method is a modification of the iteration method. In this method for the calculation of
(k+1)-th approximation of xi we take account the (k+1)-th approximations of x1 , x2 ,..., xi −1 .

Suppose we got the following system:

 x1 = β1 + α11 x1 + α12 x2 + ... + α1n xn

x =
 2 β 2 + α 21 x1 + α 22 x2 ... + α 2 n xn

....................................
 xn = β n + α n1 x1 + α n 2 x2 ... + α nn xn
(5)

By selecting randomly the initial approximations

x1(0) , x2(0) ,..., xn(0) and putting in (5), we get

x1(1) = β1 + α11 x1(0) + α12 x2(0) + ... + α1n xn(0)

Putting in the second equation x1(1) in place of x1 we get

x2(1) = β1 + α 21 x1(1) + α 22 x2(0) + ... + α 2 n xn(0) .

By analogy we get

x3(1) = β1 + α 31 x1(1) + α 32 x2(1) + α 33 x3(0) + ... + α 3n xn(0)

Aram YESAYAN, PhD, Associate Professor
12
Finally we get

xn(1) = β1 + α n1 x1(1) + α n 2 x2(1) + α n 3 x3(1) + ... + α nn −1 xn(1)−1 + α nn xn(0)

Therefore if we have the k-th approximations, we can get the (k+1)-th approximations:

x1( k +1) = β1 + α11 x1( k ) + α12 x2( k ) + ... + α1n xn( k )

x2( k +1) = β1 + α 21 x1( k +1) + α 22 x2( k ) + ... + α 2 n xn( k )

xn( k +1) = β1 + α n1 x1( k +1) + α n 2 x2( k +1) + ... + α nn −1 xn( k−1+1) + α nn xn( k ) . (6)

Example:

10 x1 + x2 + x3 = 12

2 x1 + 10 x2 + x3 = 13
2 x + 2 x + 10 x =
 1 2 3 14

 x1 =
1.2 − 0.1x2 − 0.1x3

 x2 =1.3 − 0.2 x1 − 0.1x3
x =
 3 1.4 − 0.2 x1 − 0.2 x2

We take

1.2 
 
X (0)
=  0  and we get
 
0 

x1(2) = 1.2 − 0.1 ⋅1.06 − 0.1 ⋅ 0.948 = 0.9992

x2(2) = 1.3 − 0.2 ⋅ 0.9992 − 0.1 ⋅ 0.948 = 1.0054
x3(2) = 1.4 − 0.2 ⋅ 0.9992 − 0.2 ⋅1.0054 = 0.991

Etc…

Aram YESAYAN, PhD, Associate Professor

13
x1(4) ≈ 1
x2(4) ≈ 1
x3(4) ≈ 1

Theorem : If || α ||< 1, then the following estimation of error take place for the k − th approximation :
|| α ||k
|| X − X ( k ) ||≤ || X (1) − X (0) || . (7)
1− || α ||

In particular case if we have

X=
(0)
β then X =
(1)
αβ + β .

|| α ||k || α ||k +1
So || X − X ( k ) ||≤ || αβ ||≤ || β || . (8)
1− || α || 1− || α ||

Example:

To prove that this system converges for each initial approximation.

To find the number of the approximations for the solution to within an accuracy of 10-4.

10 x1 − x2 − 2 x3 − 3 x4 = 0
 x + 10 x − x + 2 x =
 1 2 3 4 5

2 x1 + 3 x2 + 20 x3 − x4 = −10
3 x1 + 2 x2 + x3 + 20 x4 = 15

 x1 = 0.1 x2 − 0.2 x3 + 0.3 x4

 x =0.5 − 0.1x + 0.1x − 0.2 x
 2 1 3 4

 x3 = −0.5 − 0.1x1 − 0.15 x2 + 0.05 x4
 x4 =0.75 − 0.15 x1 − 0.1x2 − 0.05 x3

 0 0.1 − 0.2 0.3 

 
α  −0.1 0 0.1 − 0.2 
, where || α ||l < 1
 −0.1 − 0.15 0 0.05 
 
 −0.15 − 0.1 − 0.05 0 

So the sequence of the approximations converges for each initial approximation.

Aram YESAYAN, PhD, Associate Professor

14
We take

0 
 
 0.5 
X = β=
(0)
, where || β ||=l 1.75.
 −0.5 
 
 0.75 

So from (8) we get

|| α ||k || α ||k +1 0.55k +1 ⋅1.75

|| X − X ( k ) ||≤ || αβ ||≤ = || β || < 10−4.
1− || α || 1− || α || 0.45

Consequently k>16.7, we take k=17.

Determination of Largest Eigenvalue in Absolute Value

=
Suppose that the square matrix A (=
aij ), i 1;=
n, j 1; n has only one maximal eigenvalue in

absolute value, i.e. | λ1 |>| λ2 |≥| λ3 |≥ ... ≥| λn | .

Determination algorithm:

1. To choose the initial vector

Y (0) ,

2. To form the next iterations:

Y (1) = AY (0) ,
= =
Y (2) AY (1)
A2Y (0) ,
....
( k +1)
Y= =
AY (k )
A( k +1)Y (0) ,
....

Aram YESAYAN, PhD, Associate Professor

15
 3. To choose two successive vectors from the sequence of the vectors:
=Y ( k ) A=Y and Y ( k +1) Ak +1Y (0) .
k (0)

yi ( k +1) yi ( k +1)
=So λ1 lim or λ1 ≈ ,
k →∞ yi ( k ) yi ( k )

where the numbers

yi( k +1) and yi( k )

are the coordinates of the vectors

Y ( k +1) and Y ( k ) .

In some cases for the eigenvalue approximation we take the

( k +1)
and Y ( k ) .
average of the ratio of the corresponding coordinates of the vectors Y

Us an eigenvector we can take

Y ( k +1) .

Aram YESAYAN, PhD, Associate Professor

16
 Determination of Smallest Eigenvalue in Absolute Value

=
Suppose that the square matrix A (=
aij ), i 1;=
n, j 1; n has only one minimal eigenvalue in

absolute value, i.e. | λ1 |≥| λ2 |≥| λ3 |≥ ... | λn −1 |>| λn |> 0.

−1
In that case | λn |>| λn −1 |≥| λn − 2 |≥ ... ≥| λ1 |> 0 will be the eigenvalues of A and we apply
−1 −1 −1 −1

the algorithm of the determination of the largest eigenvalue (

λn −1 ) in absolute value.

Determination of Next Eigenvalue

=
Suppose that the eigenvalues of the matrix A (=
aij ), i 1;=
n, j 1; n are numerated in this form

| λ1 |>| λ2 |>| λ3 |≥ ... ≥| λn | .

So for finding λ2 we will use λ difference by using λ1 .

∆ λ AmY = Am +1Y − λ AmY or ∆ λ Y ( m ) = Y ( m +1) − λY ( m ) ,

So ∆ λ1Y ( m ) = Y ( m +1) − λ1Y ( m ) ,
Y ( m ) − λ1Y ( m −1) .
∆ λ1Y ( m −1) =

The second eigenvalue is calculated by the following formula:

∆ λ1 yi ( m ) yi ( m +1) − λ1 yi ( m )
λ2 ≈ =
∆ λ1 yi ( m −1) yi ( m ) − λ1 yi ( m −1)

In some cases the approximation of ʎ2 is taken as the average of the quotient of the

corresponding coordinates of the vectors ∆ λ Y and ∆ λ Y

( m −1)
. and ∆ λ Y
(m) (m)
1 1 can be taken as a
1

second eigenvector.

The value of
λ1 is found by the approximate method, so for the calculation of λ2 the number of

the iterations must be less than the number of the iterations for λ1 , because if the number of the
iterations increases the degree of inaccuracies increases too.

Aram YESAYAN, PhD, Associate Professor

17