MATH1012 Mathematical Theory and

Methods

Week 5

1 / 44
So far
▶ Systems of linear equations
▶ Vector spaces R2 , R3 , . . ., Rn
▶ Subspaces have bases and a dimension
▶ Matrix addition and multiplication
▶ row/column/null space
▶ Rank-nullity theorem
▶ Matrix inverses
▶ Determinants
▶ Linear transformations
▶ Range, Kernel
▶ Composition of transformations

Now we start Chapter 5.

2 / 44
Change of basis

Let B and C be two bases for the same m-dimensional subspace

V of Rn (often n = m and V = Rn ).

What is the link between the coordinates with respect to each

basis?

(v)B ←→ (v)C

We want transition matrix PCB such that (v)C = PCB (v)B

3 / 44
Change of basis

Recall from Chapter 4 that

[f ]CB (v)B = (f (v))C

The identity mapping

I : Rn → Rn such that Iv = v

is a linear transformation.

So [I]CB takes in a B-coordinate vector and produces a

C-coordinate vector.

Common to use the name PCB for the change-of-basis matrix.

4 / 44
Change of basis matrix

If B = {u1 , . . . , un }, then the i’th column of PCB contains the

C-coordinates of ui .

 
 
 
PCB =



 

5 / 44
Example

Suppose that

B = {(1, 1, 0), (1, 0, 0), (0, 0, −1)}

C = {(0, 0, 1), (1, 0, 1), (0, 1, 0)}

Then  

PCB =  

6 / 44
Example

Suppose that

B = {(1, 1, 0), (1, 2, 3), (0, 0, 5)}

S = {(1, 0, 0), (0, 1, 0), (0, 0, 1)}

 

PSB =  

7 / 44
What is PBC ?

PCB (v)B = (v)C and PBC (v)C = (v)B and so

PCB PBC (v)C = (v)C

So PBC is the inverse of PCB .

8 / 44
Another basis

You may be given the standard matrix of a linear

transformation f and then asked to find the matrix with
respect to bases B and C.

We claim that
[f ]CB = PCS [f ]SS PSB

PCS [f ]SS PSB (v)B = PCS [f ]SS (v)S

= PCS (f (v))SS
= (f (v))C

9 / 44
A better basis?

Let f : R2 → R2 be given by f (x, y) = (y − 3x, 2x − 2y)

 
−3 1
[f ]SS =
2 −2

If B = {(−1, 1), (1, 2)} then what is [f ]BB ?

[f ]BB = PBS [f ]SS PSB

10 / 44
Transition matrices

(−1, 1) = −1(1, 0) + 1(0, 1)

(1, 2) = 1(1, 0) + 2(0, 1)

so  
−1 1
PSB =
1 2

Then
 
−1
− 32 1

−1 1 3
PBS = = 
1 2 1 1
3 3

11 / 44
And [f ]BB ?

[f ]BB = PBS [f ]SS PSB

" #" #" #

− 23 1
3 −3 1 −1 1
[f ]BB = 1 1
3 3 2 −2 1 2
" #
−4 0
=
0 −1

12 / 44
Why is this useful?

In linear algebra applications, matrices often represent a

transition from one state to another.

v0 , v1 = Av0 , v2 = Av1 , v3 = Av2 . . .

What is the long term behaviour of the system?

13 / 44
So far
▶ Systems of linear equations
▶ Vector spaces R2 , R3 , . . ., Rn
▶ Subspaces have bases and a dimension
▶ Matrix addition and multiplication
▶ row/column/null space
▶ Rank-nullity theorem
▶ Matrix inverses
▶ Determinants
▶ Linear transformations
▶ Range, Kernel
▶ Composition of transformations
▶ Transition matrices for coordinates and linear
transformations
Now we start Chapter 6.
14 / 44
Eigenvectors
If A is an n × n matrix, then a non-zero vector v ∈ Rn is called
an eigenvector of A with eigenvalue λ ∈ R if

Av = λv

Example  
1 2
A=
2 1
 
1
A =
1

 
1
A =
−1

15 / 44
Finding eigenvalues?

Is 2 an eigenvalue of A?

Solve     
1 2 x x
=2
2 1 y y

16 / 44
More systematic

If
Av = λv
then
(A − λI)v = 0

When does this system have non-zero solutions?

17 / 44
Determinant condition
The matrix
     
1 2 1 0 1−λ 2
−λ =
2 1 0 1 2 1−λ

has determinant
(1 − λ)2 − 4
When is this zero?

18 / 44
The eigenvalue 3

Solve     
1 2 x x
=3
2 1 y y
that is the homogeneous system with matrix

19 / 44
The eigenvalue −1

Solve     
1 2 x x
= −1
2 1 y y
that is the homogeneous system with matrix

20 / 44
High powers

Suppose that A is a square matrix such that Av = λv. Then

AAv =

and
A3 v =
and so on.

21 / 44
Why do we care?
What is the value of  
500 1
A ?
3

22 / 44
Characteristic polynomial

If A is an n × n matrix, then

det(A − λI)

is a polynomial of degree n in λ, called the characteristic

polynomial of A.

The equation det(A − λI) = 0 is the characteristic equation.

Its solutions are the eigenvalues of A.

23 / 44
Subspaces
For any number λ ∈ R, the solutions to

Av = λv

form a subspace.

24 / 44
Eigenspaces

When λ is an eigenvalue, the subspace has dimension at least 1

and is called an eigenspace, and denoted Eλ .

25 / 44
Standard task

Here is an n × n matrix A, what are its eigenspaces? Describe

them by giving a basis for each.

▶ Calculate det(A − λI)

▶ Find the roots λ1 , λ2 , . . ., λk (k ⩽ n)
▶ Solve (A − λ1 I)v = 0 and find a basis for Eλ1
▶ Solve (A − λ2 I)v = 0 and find a basis for Eλ2
▶ ...
▶ Solve (A − λk I)v = 0 and find a basis for for Eλk

26 / 44
Example

The matrix  
1 3
A=
4 2
has characteristic polynomial λ2 − 3λ − 10. Find a basis for
each of its eigenspaces.

27 / 44
Example 1

Consider the matrix

 
0 1
A=
−1 0

Give a geometric description of this linear transformation.

What are its eigenvalues and eigenspaces?

Rotations have no eigenspaces

28 / 44
Eigenvalues

The equation det(A − λI) = 0 is a polynomial equation of

degree n and so it has exactly n solutions

λ1 , λ2 , . . . , λn

(Eigenvalues might be repeated and might be complex numbers.)

29 / 44
Properties of eigenvalues

▶ The sum of the eigenvalues is equal to the trace of the

matrix

λ1 + λ2 + · · · + λn = A11 + A22 + · · · + Ann

▶ The product of the eigenvalues is equal to the determinant

of the matrix

λ1 × λ2 × · · · × λn = det A

▶ The matrix A is invertible if and only if 0 is not an

eigenvalue of A.

30 / 44
Properties of eigenvalues II

▶ the eigenvalues of AT are λ1 , λ2 , . . ., λn .

▶ the eigenvalues of kA (where k ∈ R) are kλ1 , kλ2 , . . ., kλn .

▶ the eigenvalues of Ak (where k ∈ Z) are λk1 , λk2 , . . ., λkn .

▶ Cayley-Hamilton Theorem: A satisfies its own

characteristic equation.

31 / 44
Example 2

Consider the matrix  

1 2
A=
0 1
which has characteristic polynomial

1−λ 2
= (λ − 1)2
0 1−λ

What are its eigenvalues and eigenspaces?

32 / 44
Multiplicity of eigenvalue

The algebraic multiplicity of a particular eigenvalue k is m if

det(A − λI) = · · · (k − λ)m · · ·

The geometric multiplicity of a particular eigenvalue k is the

dimension of the eigenspace Ek .

33 / 44
Back to Example 2

Consider the matrix  

1 2
A=
0 1

Algebraic multiplicity=

Geometric multiplicity=

34 / 44
Algebraic vs Geometric

For any eigenvalue:

1 ⩽ geometric multiplicity ⩽ algebraic multiplicity

▶ The algebra tells us the maximum possible multiplicity

▶ The geometry tells us the actual multiplicity

35 / 44
Diagonalisable

A linear transformation f : Rn → Rn is diagonalisable if there is

a basis B for Rn such that [f ]BB is a diagonal matrix.

A matrix A is called diagonalisable if the corresponding linear

transformation is diagonalisable.

This happens exactly if we can find a basis for Rn consisting of

eigenvectors for A.

36 / 44
Diagonalisability test

A matrix A is diagonalisable if and only if

▶ It has all real eigenvalues (no complex)
▶ Every eigenvalue has the maximum possible geometric
multiplicity

In other words, each eigenvalue has geometric multiplicity equal

to its algebraic multiplicity.

Special case: if the matrix has n distinct eigenvalues, then it is

diagonalisable.

37 / 44
Example

What are the eigenvalues and eigenvectors of

 
1 1 1
A =  0 1 1 ?
0 1 1

Find characteristic polynomial:

1−λ 1 1
det(A − λI) = 0 1−λ 1
0 1 1−λ
= λ(λ − 1)(λ − 2)

38 / 44
Working

Solve three systems of linear equations:

▶ Ax = 0x has solutions S = {t(0, 1, −1) : t ∈ R}
▶ Ax = 1x has solutions S = {t(1, 0, 0) : t ∈ R}
▶ Ax = 2x has solutions S = {t(2, 1, 1) : t ∈ R}

So B = {(0, 1, −1), (1, 0, 0), (2, 1, 1)} is a basis for Rn of

eigenvectors.

39 / 44
Working: Change of basis

The matrix  
0 1 2
P = PSB = 1 0 1 
−1 0 1
and  
0 0 0
P −1 AP =  0 1 0 
0 0 2

40 / 44
Summary

To diagonalise a matrix A
▶ Find the eigenvalues λ1 , λ2 , . . ., λk and their multiplicities
If some λ has geometric multiplicity too low then fail.
▶ Find a basis for each eigenspace Eλ1 , Eλ2 , . . . , Eλk .
▶ Put the basis vectors into the columns of P
Note that P is a square matrix.

Then P −1 AP = D where D is the diagonal matrix of

eigenvalues.

41 / 44
High powers

What is A5 , if A is diagonalisable?

A5 = (P DP −1 )(P DP −1 )(P DP −1 )(P DP −1 )(P DP −1 )

= P D(P −1 P )D(P −1 P )D(P −1 P )D(P −1 P )DP −1
= P DIDIDIDIDP −1
= P D5 P −1

If
λ51 0 0
   
λ1 0 0
D =  0 λ2 0  D5 =  0 λ52 0 
0 0 λ3 0 0 λ53

42 / 44
Special Case

A matrix is symmetric if A = AT .

If A is a symmetric matrix, then

▶ Each eigenvalue of A is real
▶ Each eigenvalue has geometric multiplicity equal to
algebraic multiplicity
▶ Eigenvectors from distinct eigenspaces are orthogonal

In other words, symmetric matrices are diagonalisable.

43 / 44
Applications

There are lots of applications of eigenvalues/eigenspaces to

▶ engineering (stress tensors)
▶ physics (mechanics)
▶ study of graphs (Google Page Rank)
▶ data analysis
▶ computer graphics (image compression)
This concludes our study of Linear Algebra.

44 / 44