Mathematics and Statistics II - Linear Algebra

Week 6: Eigenvalues and Eigenvectors

Andrés Perea

Maastricht University

Period 2

Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 1 / 20

Eigenvalues and Eigenvectors

Consider a linear transformation T : Rn ! Rn .

Then, there is an n n matrix A such that T (v̂ ) = A v̂ for all

v̂ 2 Rn .

A vector v̂ 6= 0̂ is an eigenvector of T (or A) if there is a number

λ 2 R such that
A v̂ = λ v̂ .

In this case, the number λ is the eigenvalue that corresponds to v̂ .

Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 2 / 20

Computing Eigenvalues and Eigenvectors

Suppose that v̂ 6= 0̂ is an eigenvector of A with eigenvalue λ.

Then, A v̂ = λ v̂ = λ I v̂ , and hence (A λ I ) v̂ = 0̂.

Hence, v̂ 6= 0̂ is in the nullspace of A λ I.

This can only be if A λ I is not invertible.

Hence, det(A λ I ) = 0.

Theorem
The eigenvalues of a square matrix A are exactly the numbers λ for which
det(A λ I ) = 0.

Here, det(A λ I ) is a polynomial of degree n in λ, and it is called

the characteristic polynomial of A.

Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 3 / 20

Theorem
The eigenvalues of a square matrix A are exactly the numbers λ for which
det(A λ I ) = 0.

2 1
Example: Take A = . Compute the eigenvalues for A.
0 1

2λ 1
We have: A λ I = and hence
0 1 λ
det(A λ I ) = (2 λ)( 1 λ).

Here, det(A λ I ) is called the characteristic polynomial of A.

By setting det(A λ I ) = 0, we …nd λ1 = 2 and λ2 = 1.

Eigenvalues.

Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 4 / 20

 2 1
Example continued: For A = , the eigenvalues are λ1 = 2
0 1
and λ2 = 1.

How to compute the eigenvectors that belong to these eigenvalues?

Vector v̂ 6= 0̂ is an eigenvector with eigenvalue λ = 2 precisely when

A v̂ = 2 v̂ .

Hence,

0 1 x 0
(A 2 I ) v̂ = 0̂ () =
0 3 y 0
() y = 0.

Thus, eigenvectors with eigenvalue 2 are all vectors

x 1 1
=x . This is a subspace with basis .
0 0 0
This is the eigenspace for λ = 2.
Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 5 / 20
 2 1
Example continued: For A = , the eigenvalues are λ1 = 2
0 1
and λ2 = 1.

Compute eigenspace for λ = 1.

We must solve
3 1 x 0
(A ( 1) I ) v̂ = 0̂ () =
0 0 y 0
() 3x + y = 0.

By choosing x as the free variable, the eigenspace for λ = 1

x 1
contains all vectors =x .
3x 3
1
Hence, eigenspace for λ = 1 has basis .
3

Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 6 / 20

 2 0
Example: Take A = . Then,
0 2
2 λ 0
det(A λ I ) = det = (2 λ )2 .
0 2 λ
Only eigenvalue is λ = 2. Eigenspace for λ = 2 :

0 0 x 0
(A 2 I ) v̂ = 0̂ () = .
0 0 y 0

1 0
Eigenspace for λ = 2 has basis , .
0 1

Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 7 / 20

 2 1
Example: Take B = . Then,
0 2
2 λ 1
det(B λ I ) = det = (2 λ )2 .
0 2 λ
Only eigenvalue is λ = 2. Eigenspace for λ = 2 :

0 1 x 0
(B 2 I ) v̂ = 0̂ () =
0 0 y 0
() y = 0.

1
Eigenspace has basis .
0

Two matrices with the same characteristic polynomial may have

di¤erent eigenspaces.

Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 8 / 20

Theorem
Let A be a square matrix, and let v̂1 , ..., v̂n be eigenvectors with
eigenvalues λ1 , ..., λn .
If all these eigenvalues are di¤erent, then fv̂1 , ..., v̂n g is linearly
independent.
3 2
3 4 7
Example: Take A = 4 0 1 8 5 . Then,
0 0 2
2 3
3 λ 4 7
A λ I =4 0 1 λ 8 5.
0 0 2 λ
Eigenvalues: det(A λ I ) = (3 λ)(1 λ)(2 λ) = 0. Hence,
λ1 = 3, λ2 = 1 and λ3 = 2.

Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 9 / 20

Theorem
Let A be a square matrix, and let v̂1 , ..., v̂n be eigenvectors with
eigenvalues λ1 , ..., λn .
If all these eigenvalues are di¤erent, then fv̂1 , ..., v̂n g is linearly
independent.
2 3
3 4 7
Example: Take A = 4 0 1 8 5 . Eigenspace for λ1 = 3 :
0 0 2
2 3 2 3 2 3
0 4 7 x 0
(A 3 I ) v̂ = 0̂ () 0 4 2 8 5 4 y 5 = 0 5
4
0 0 1 z 0
() y = 0 and z = 0.
82 39
< 1 =
Eigenspace for λ1 = 3 has basis 4 0 5 .
: ;
0
Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 10 / 20
Theorem
Let A be a square matrix, and let v̂1 , ..., v̂n be eigenvectors with
eigenvalues λ1 , ..., λn .
If all these eigenvalues are di¤erent, then fv̂1 , ..., v̂n g is linearly
independent.
2 3
3 4 7
Example: Take A = 4 0 1 8 5 . Eigenspace for λ2 = 1 :
0 0 2
2 3 2 3 23
2 4 7 x 0
4
(A 1 I ) v̂ = 0̂ () 0 0 8 5 4 y 5=4 0 5
0 0 1 z 0
() z = 0 and 2x + 4y = 0.
82 39
< 2 =
Eigenspace for λ1 = 1 has basis 4 1 5 .
: ;
0
Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 11 / 20
Theorem
Let A be a square matrix, and let v̂1 , ..., v̂n be eigenvectors with
eigenvalues λ1 , ..., λn .
If all these eigenvalues are di¤erent, then fv̂1 , ..., v̂n g is linearly
independent.
2 3
3 4 7
Example: Take A = 4 0 1 8 5 . Eigenspace for λ2 = 2 :
0 0 2
2 3 2 3 2 3
1 4 7 x 0
(A 2 I ) v̂ = 0̂ () 0 4 1 8 5 4 y 5 = 0 5
4
0 0 0 z 0
() y = 8z and x = 39z.
82 39
< 39 =
Eigenspace for λ1 = 2 has basis 4 8 5 .
: ;
1
Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 12 / 20
Theorem
Let A be a square matrix, and let v̂1 , ..., v̂n be eigenvectors with
eigenvalues λ1 , ..., λn .
If all these eigenvalues are di¤erent, then fv̂1 , ..., v̂n g is linearly
independent.
2 3
3 4 7
Example: Take A = 4 0 1 8 5.
0 0 2
The eigenvalues are λ1 = 3, λ2 = 1 and λ3 = 2.

The
82 bases
39for8the
2 corresponding
39 8eigenspaces
2 39 are
< 1 = < 2 = < 39 =
4 0 5 , 4 1 5 and 4 8 5 .
: ; : ; : ;
0 0 1
Hence, three eigenvectors for λ1 = 3, λ2 = 1 and λ3 = 2 will be
linearly independent.
Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 13 / 20
Diagonalizing a Matrix
2 3
3 4 7
4
Example: Take A = 0 1 8 5 with characteristic polynomial
0 0 2
(3 λ)(1 λ)(2 λ).
The eigenvalues are λ1 = 3, λ2 = 1 and λ3 = 2.
The
82 bases39for8the
2 corresponding
39 8eigenspaces
2 39 are
< 1 = < 2 = < 39 =
4 0 5 , 4 1 5 and 4 8 5 .
: ; : ; : ;
0 0 1
82 3 2 3 2 39
< 1 2 39 =
In fact: B = 4 0 5,4 1 5,4 8 5 is a basis of
: ;
0 0 1
eigenvectors. 32
3 0 0
Hence, [T ]BB = 4 0 1 0 5 . Diagonal matrix.
0 0 2
Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 14 / 20
 In general, let A be an n n matrix with a basis of eigenvectors
B = fv̂1 , ..., v̂n g, and corresponding eigenvalues λ1 , ..., λn .
If A represents the linear transformation T , then
2 3
λ1 0 0
6 7
[T ]BB = 4 0 . . . 0 5 = D.
0 0 λn
Let C be the standard basis for Rn . Then, [T ]BB = PB 1 [T ]CC PB .
Since [T ]CC = A, we have that PB 1 A PB = D.

Theorem
Let A be an n n matrix.
(a) If A has n di¤erent eigenvalues, then there is a basis of Rn that
consists of eigenvectors of A.
(b) If there is a basis of eigenvectors, then there is an invertible matrix P
such that P 1 A P = D is a diagonal matrix.The diagonal entries in D
are the eigenvalues, and the columns in P constitute the basis of
eigenvectors.
Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 15 / 20
 Matrix A can be diagonalized if there is an invertible matrix P and a
diagonal matrix D such that P 1 A P = D .

Hence, matrix A can be diagonalized precisely when there is a basis of

eigenvectors.
2 3
2 0 0
Example: Consider the matrix A = 4 3 0 1 5 . Can A be
0 1 0
diagonalized?

Characteristic polynomial:
2 3
2 λ 0 0
det(A λI ) = det 4 3 λ 1 5 = (2 λ ) ( λ2 1) =
0 1 λ
(2 λ)(λ 1)(λ + 1).
Three di¤erent eigenvalues λ1 = 2, λ2 = 1 and λ3 = 1.

There is a basis of eigenvectors, and A can be diagonalized.

Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 16 / 20
 2 3
2 0 0
Example continued: A = 4 3 0 1 5 has eigenvalues λ1 = 2,
0 1 0
λ2 = 1 and λ3 = 1.
3 2
2 0 0
Find matrix P such that P 1 A P = D = 4 0 1 0 5.
0 0 1
82 39 82 39
< 1 = < 0 =
The three eigenspaces have bases 4 2 5 , 4 1 5 and
: ; : ;
1 1
82 39
< 0 =
4 1 5 . Verify this.
: ;
1
2 3
1 0 0
Hence, P = 4 2 1 1 5.
1 1 1
Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 17 / 20
Similar Matrices
Two n n matrices A and B are called similar if there is an invertible
matrix P such that B = P 1 A P.
Theorem
Two similar matrices A and B have the same characteristic polynomials,
and hence the same eigenvalues.

Proof: We have
1 1 1
B λI = P A P λI = P A P P λI P
1
= P (A λI ) P.
Hence, characteristic polynomial of B is
1
det(B λI ) = det(P (A λI ) P )
1
= det(P ) det(A λI ) det(P )
1
= det(A λI ) det(P ) = det(A λI ).
det(P )
Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 18 / 20
Representation of Linear Transformations
Suppose that two n n matrices A and B are similar.
Then, there is an invertible matrix P such that B = P 1 A P.
Let C be the standard basis for Rn and T : Rn ! Rn the linear
transformation with [T ]CC = A.
Since P is invertible, the columns in P constitute a basis
D = fd̂1 , ..., d̂n g for Rn .
Hence, PD = P.
Hence,
B=P 1
A P = PD 1 [T ]CC PD = [T ]D
D .

Theorem
Two n n matrices A and B are similar, if and only if, there is a linear
transformation T , and two di¤erent bases C and D for Rn , such that
A = [T ]CC and B = [T ]DD .
Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 19 / 20
 Example: Consider the linear transformation T : R3 ! R3 with
02 31 2 3 02 31 2 3 02 31 2 3
1 2 0 3 0 1
@4 0 5A = 4 0 5 , T @4 1 5A = 4 1 5 , T @4 0 5A = 4 1 5 .
0 3 0 4 1 1
82 3 2 3 2 39
< 1 1 1 =
Consider the basis B = 4 1 5 , 4 1 5 , 4 0 5 . What is [T ]BB ?
: ;
1 0 0
If C is the standard basis for R3 , then [T ]BB = PB 1 [T ]CC PB .
2 3 2 3
1 1 1 2 3 1
We have: PB = 1 1 4 0 5 and [T ]CC = 4 0 1 1 5 .
1 0 0 3 4 1
2 3 2 3 2 3
0 0 1 2 3 1 1 1 1
Hence, [T ]BB = 4 0 1 1 5 4 0 1 1 5 4 1 1 0 5=
1 1 0 3 4 1 1 0 0
2 3
2 1 3
4 0 0 3 5.
0 0 2
Andrés Perea (Maastricht University) Eigenvalues and Eigenvectors Period 2 20 / 20