20.

Diagonalisation
Definition 20.1. Let A and B be two square n × n matrices. We say
that A and B are similar if there is an invertible square n × n matrix
P such that A = P BP −1 .
We say that A is diagonalisable if A is similar to a diagonal matrix
D.
Suppose that
A = P BP −1 .
Then
A2 = A · A
= (P BP −1 )(P BP −1 )
= P B(P −1 P )BP −1
= P BBP −1
= P B 2 P −1 .
More generally we have:
Lemma 20.2. Suppose that A and B are two n × n square matrices
and that P is an invertible matrix such that
A = P BP −1 .
Then
An = P B n P −1 .
Proof. We prove this by induction on n. It is true for n = 1 by as-
sumption. Suppose that
An = P B n P −1 ,
for some n > 0. Then
An+1 = A · An
= (P BP −1 )(P B n P −1 )
= P B(P −1 P )B n P −1
= P BB n P −1
= P B n+1 P −1
as required. Thus the result holds by induction on n. 
(20.2) gives us a practical way to compute the powers of a diagonal-
isable matrix A.
1
Theorem 20.3. Let A be an n × n matrix.
Then A is diagonalisable if and only if we can find a basis ~v1 , ~v2 , . . . , ~vn
of eigenvectors for Rn . In this case,
A = P DP −1 ,
where P is the matrix whose columns are the eigenvectors ~v1 , ~v2 , . . . , ~vn
and D is the diagonal matrix whose diagonal entries are the correspond-
ing eigenvalues λ1 , λ2 , . . . , λn .
Proof. Suppose that A = P DP −1 , where the columns of P are ~v1 , ~v2 , . . . , ~vn
and D is a diagonal matrix with entries λ1 , λ2 , . . . , λn . We have
A~vi = (P DP −1 )(P~ei )
= (P D)(P −1 P )~ei
= P (D~ei )
= P (λi~ei )
= λi (P~ei )
= λi~vi .
Therefore ~vi is an eigenvector with eigenvalue λi . The vectors ~v1 , ~v2 , . . . , ~vn
are a basis of Rn as P is invertible.
Now for the other direction. Suppose that ~v1 , ~v2 , . . . , ~vn are a basis
of eigenvectors. Let P be the matrix whose columns are the vectors
~v1 , ~v2 , . . . , ~vn . Then P is an invertible matrix. Let D = P −1 AP . Then
D~ei = (P −1 AP )~ei
= P −1 A~vi
= P −1 λi~vi
= λi P −1~vi
= λi~ei .
So D is the matrix whose ith row is the vector λi~ei . But then D is a
diagonal matrix with entries λ1 , λ2 , . . . , λn on the main diagonal. We
have
D = P −1 AP.
Multiplying both sides by P on the left, we get
P D = AP.
Finally multiplying both sides on the right by P −1 we get
A = P DP −1 . 
Let’s illustrate (20.3) by going back to the examples in §19.
2
Example 20.4. Is it possible to diagonalise
 
−8 5
A= ?
−10 7
If the answer is yes, then diagonalise A.
We already saw that
~v1 = (1, 2)
is an eigenvector with eigenvalue 2, and
~v2 = (1, 1)
is an eigenvector with eigenvalue −3.
~v1 and ~v2 are independent (either by inspection or because 2 6= −3).
Thus A is diagonalisable.
Let  
1 1
P =
2 1
the matrix whose columns are the eigenvectors. Then
   
−1 1 −1 −1 1
P = −1 = .
−2 1 2 −1
Let  
2 0
D=
0 −3
the diagonal matrix whose entries on the diagonal are the eigenvalues.
Let’s compute P DP −1 :
        
1 1 2 0 −1 1 1 1 −2 2 −8 5
= = ,
2 1 0 −3 2 −1 2 1 −6 3 −10 7
as expected.
Now we can compute any power of A easily:
An = P Dn P −1 .
We compute
 n   n  
−8 5 1 1 2 0 −1 1
=
−10 7 2 1 0 (−3)n 2 −1
  
1 1 −2n 2n
=
2 1 2(−3)n −(−3)n
 
−2n + 2(−3)n 2n − (−3)n
= .
−2n+1 + 2(−3)n 2n+1 − (−3)n
3
Example 20.5. Is it possible to diagonalise
 
1 2 1
A =  6 −1 0 ?
−1 −2 −1
If the answer is yes, then diagonalise A.
We already saw that
~v1 = (−1, −6, 13)
is an eigenvector with eigenvalue 0,
~v2 = (−1, 2, 1)
is an eigenvector with eigenvalue −4, and
~v3 = (−2, −3, 2)
is an eigenvector with eigenvalue 3.
~v1 , ~v2 and ~v3 are independent, since their eigenvalues 0, −4 and 3
are distinct. Therefore they are a basis and so A is diagonalisable.
Let  
−1 −1 −2
P = −6 2 −3
13 1 2
the matrix whose columns are the eigenvectors. Then, with the aid of
a computer,
 
7 0 7
1 
P −1 = −27 24 9 .
84 −32 −12 −8

Let  
0 0 0
D = 0 −4 0
0 0 3
the diagonal matrix whose entries on the diagonal are the eigenvalues.
Now we can compute any power of A easily:
An = P Dn P −1 .
We compute
   
−1 −1 −2 0 0 0 7 0 7
1 
−6 2 −3 0 (−4)n 0  −27 24 9 .
84 13 1 2 0 0 3 n
−32 −12 −8
4
Example 20.6. Is it possible to diagonalise
 
1 1
A=
0 1
If the answer is yes, then diagonalise A.
We compute the eigenvalues of A:
1−λ 1
= (1 − λ)2 .
0 1−λ
So the only eigenvalue is λ = 1. We want to compute the null space of
A − I2 :  
0 1
.
0 0
y is a basic variable and x is a free variable. y = 0. Thus ~e1 is an
eigenvector with eigenvalue 1. A is not diagonalisable, we cannot find
a basis of eigenvalues.
Example 20.7. Is it possible to diagonalise
 
−1 0 1
A =  3 0 −3 .
1 0 −1
If the answer is yes, then diagonalise A.
The characteristic polynomial is:
−1 − λ 0 1
−λ −3 3 −λ
3 −λ −3 = −(1 + λ) +
0 −1 − λ 1 0
1 0 −1 − λ
= −(1 + λ)2 λ + λ
= −λ2 (λ + 2).
Thus the eigenvalues are 0 and −2.
We want to calculate the nullspace of A. We apply Gaussian elimi-
nation:
     
−1 0 1 1 0 −1 1 0 −1
 3 0 −3 → 3 0 −3 → 0 0 0  .
1 0 −1 1 0 −1 0 0 0
x is a basic variable, y and z are free variables.
x−z =0 so that x = z.
The general solution is
(x, y, z) = (z, y, z) = y(0, 1, 0) + z(1, 0, 1).
5
A basis for the nullspace is given by (0, 1, 0) and (1, 0, 1).
~v1 = (0, 1, 0) and ~v2 = (1, 0, 1)
are independent eigenvectors with eigenvalue 0.
We want to calculate the nullspace of A + 2I3 .
 
1 0 1
A + 2I3 = 3 2 −3 .
1 0 1
We apply Gaussian elimination:
     
1 0 1 1 0 1 1 0 −1
3 2 −3 → 0 2 −6 → 0 1 −3 .
1 0 1 0 0 0 0 0 0
x and y are basic variables, z is a free variable.
y − 3z = 0 so that y = 3z.
Therefore
x+z =0 so that x = −z.
The general solution is
(x, y, z) = (−z, 3z, z) = z(−1, 3, 1).
~v3 = (−1, 3, 1)
is an eigenvector with eigenvalue −2. The vectors ~v1 , ~v2 and ~v3 are
independent, thus A is diagonalisable.
Let  
0 1 −1
P = 1 0 3 
0 1 1
the matrix whose columns are the eigenvectors. Then, with the aid of
a computer,  
3 1 −3
1
P −1 =  1 0 1  .
2 −1 0 1
Let  
0 0 0
D = 0 0 0 
0 0 −2
the diagonal matrix whose entries on the diagonal are the eigenvalues.
Now we can compute any power of A easily:
An = P Dn P −1 .
6
We compute
   
0 1 −1 0 0 0 3 1 −3
1
1 0 3  0 0 0  1 0 1 .
2 0 1 1 0 0 (−2)n −1 0 1