Module 1: Matrices and Linear Algebra

Lesson 8

Diagonalization of Matrices

8.1 Introduction

Diagonalizable matrices are of particular interest in linear algebra because of their

application to computation of several matrix operations and functions easily. Not
all matrices are diagonalizable. In this lecture we learn technique to identify
matrices that are diagonalizable.

8.2 Similar Matrices

Diagonalizable matrices are defined through similar matrices. Two square matrices
A and B are said to be similar if there exists an invertible matrix P such that A = P−
1
B P or equivalently PA = BP.

3 5  2 4
Example 8.2.1: (i) Matrices A =   and B =   are similar because PA
3 1  4 2
 4 0
= PB, where P =   . Note that P is invertible as det P = 20 ≠ 0. However
1 5
2 0 2 1
matrices R =   and S =   are not similar because otherwise the
 0 2  0 2
a b
matrix P1 satisfying P1R = SP1 will be of the form   and is a non-invertible
0 0
(or singular) matrix.

In the following we shall present an important result on similar matrices.

 Diagonalization of Matrices

Theorem 8.2.1: Similar matrices have the same characteristic equation (and hence
the same eigenvalues).

Proof: Let A and B be similar matrices. We have to show that

det (A − λI) = det (B − λI). A = P− 1 B P, where P is an invertible matrix.

det (A − λI) = det (P− 1 B P – P− 1λI P) = det (P− 1 (B − λI) P)

= det (P− 1) det(B − λI) det (P) = det (B − λI).

The above theorem also gives a criteria for checking that the given matrices are
similar or not.

1 2 4 1
Example 8.2.1: Matrices A =   and B =   are not similar because
 4 3  3 2
their characteristic polynomials are λ2 − 4λ – 5 and λ2 − 6λ + 5 respectively.

8.3 Algebraic and Geometric Multiplicities

For diagonalization of matrices we need to understand the algebraic and geometric

multiplicities of eigenvalues. Let λ0 be an eigenvalue of A. The geometric
multiplicity of λ0 is the dimension of the eigenspace of λ0, that is the dimension of
the solution space of (A – λ0I) x = 0, which is also the nullity of (A – λ0I).
Whereas the algebraic multiplicity of λ0 is the largest positive integer k such that (λ
– λ0)k is a factor of the characteristic polynomial of A.
 Diagonalization of Matrices

 −3 1 −1 
 
Example 8.3.1: Consider the matrix A =  −7 5 −1  . Characteristic polynomial
 −6 6 −2 
 
of A is det (A − λI) = (λ + 2)2 (λ − 4). So − 2 is an eigenvalue of multiplicity two
and therefore algebraic multiplicity of the eigenvalue − 2 is equal to 2. One can
check that rank of (A + 2I) is equal to two hence its nullity is equal to one. So
geometric multiplicity of the eigenvalue − 2 is equal to 1. The following theorem
gives a relation between these two multiplicities.

Theorem 8.3.1: The algebraic multiplicity of an eigenvalue is not less than its
geometric multiplicity.

8.4 Diagonalizable Matrices

A square matrix is said to be diagonalizable if it is similar to a diagonal matrix. In

other words A is diagonalizable if and only if there is an invertible matrix P such
that P-1 A P is a diagonal matrix.

The following theorem gives a criteria for diagonalizable matrices.

Theorem 8.4.1: Let An×n be an square matrix with eigenvalues λ1, λ2, . . . , λk. Let
γ1, γ2, . . . , γk be the geometric multiplicity of λ1, λ2, . . . , λk respectively. Then A is
diagonalizable if and only if γ1 + γ2 + . . . + γk = n.

From theorems 8.3.1 and 8.4.1 we get the following result.

 Diagonalization of Matrices

Corollary 8.4.1: A matrix An × n is diagonalizable if and only if for every

eigenvalue λ of A, the algebraic multiplicity of λ is equal its geometric
multiplicity.

Corollary 8.4.2: If An × n has n distinct eigenvalues then A is diagonalizable.

8.5 Algorithm to Diagonalize a Matrix

Input: A square matrix An × n.

Output: A diagonal matrix similar to A.

(1) Find eigenvalues of A say λ1, λ2, . . . , λk, k ≤ n.

(2) Find geometric multiplicity γi of λi, 1 ≤ i ≤ k.

(3) If γ1 + γ2 + . . . + γk = n then continue otherwise return that A is not

diagonalizable.
λi
(4) Find basis for eigenspace of each λi. Let { x j : 1 ≤j ≤ γi } be a basis for the
eigenspace corresponding to λi, 1 ≤ i ≤ k.

(5) Take P = ( x1λ1  x γ1 λ1 x1λ2 x 2λ2  x γ 2 λ2  x1λk x 2λk  x γ k λk ) be the n × n

matrix such that each x j λi is a column vector i.e. a matrix of size n × 1.

Obviously P is invertible.

(6) P− 1 A P = diag(λ1, λ1,… ,λ1, λ2, λ2,. . . λ2 , ….λk ,λk , … λk ) is the diagonal
matrix similar to A.
 Diagonalization of Matrices

5 4 2
 
Example 8.5.1: Consider the matrix A =  4 5 2  .
 2 2 2
 

A has two eigenvalues λ1 = 10 and λ2 = 1, where algebraic multiplicities of λ1 and

λ2 are 1 and 2 respectively. Recall that the eigenspace of λ1 is

S1 = {(2α, 2α, α): α ϵ R} (here we include the zero vector also).

dim S1 = 1 = γ1, the geometric multiplicity of λ1. Eigen space of λ2 is

S2 = {(− (2α + β), α, β): α, β ϵ R} and dim S2 = 2 = γ2 .

Now γ1 + γ2 = 3 = size of the matrix A. So A is diagonalizable.

A basis for S1 is {(2, 2, 1)}. A basis for S2 is {(− 1, 1, 0), (− )} (obtained by

taking α = 1, β = 0 and then α = 0, β = 1). So

P
 1
 2 −1 −
2
 
= .
2 1 0
1 0 1 
 
 
 Diagonalization of Matrices

10 0 0 
 
−1
One checks that P A P =  0 1 0  and is similar A.
 0 0 1
 
Not all matrices are diagonalizable and we will see such an example below.

Example 8.5.2: As we have seen in Example 8.3.1 that for the matrix A =
 −3 1 −1 
 
 −7 5 −1  the eigenvalues are λ1 = − 2, λ2 = 4, λ1 is of multiplicity 2. Also the
 −6 6 −2 
 
algebraic multiplicity of λ1 is 2 and the geometric multiplicity of it is 1. Therefore
A is not a diagonalizable matrix.

8.6 Computation of Functions of Diagonalizable Matrices

In the following theorem we shall list some properties of diagonal and

diagonalizable matrices.

Theorem 8.6.1: The following are true for a diagonal or a diagonalizable matrix
D:

a 0
(I) If D =  0 b  is a diagonal matrix the kth power of D is equal to
 n x n
 ak 0
  .
0 bk n x n

(II) If A is a diagonalizable matrix with A = P− 1 D P, where D is a diagonal matrix,

then Ak = P− 1 Dk P. (For k=2 one verifies that A2 = A.A = (P− 1 D P) (P− 1 D P) = P− 1
D (PP− 1) D P = P− 1D2 P.)
 Diagonalization of Matrices

(III) If P(x) = a0 + a1x + . . . + anxn be any polynomial and A be a diagonalizable

matrix with A = M D M−1, where D is diagonal, then P(A)= M P (D) M− 1. (One can
get this by taking P(A)= M a0 I M− 1 + M a0 D M − 1 + . . . +M a0 Dn M− 1.)

 0 1
Example 8.6.1: Here, We compute A30 for A =   . This matrix is
 −2 3 

1 1  1 0
diagonalizable as A = M D M , where M = 
−1
 and D =   . Thus by
1 2  0 2
1 1  1 0  2 −1
Theorem 8.6.1(i) and (ii), A30 = M D30 M− 1 =   30  .
1 2  0 2  −1 1 
 2 − 230 230 − 1
= .
2−2 231 − 1 
31

Example 8.6.2: If P(x) = x17 – 3x5+2x2+1 then we find the value of P(A) = A17 –
3A5 + 2A2 + I, for the same matrix A in Example 8.6.2. By Theorem 8.6.1 (iii) ,
and Example 8.61, P(A)=A17 – 3A5 + 2A2 + I

 2 − 217 217 − 1  2 − 25 25 − 1   2 − 2 2 22 − 1  1 0 
=  − 3  + 2 + .
2−2 218 − 1  2 − 26 2 6 − 1   2 − 23 23 − 1   0 1 
18

 89 − 217 −88 + 217 

= .
176 − 2 −175 + 218 
18

8.7 Conclusions

Here we have seen that finding higher powers of a diagonalizable matrix or value
of any polynomial on a diagonalizable matrix can be computed easily.
 Diagonalization of Matrices

Keywords: Similar matrices, diagonalizable matrices, algebraic multiplicity,

geometric multiplicity, functions of diagonalizable matrices.

Suggested Readings:

Linear Algebra, Kenneth Hoffman and Ray Kunze, PHI Learning pvt. Ltd., New
Delhi, 2009.

Linear Algebra, A. R. Rao and P. Bhimasankaram, Hindustan Book Agency, New

Delhi, 2000.

Linear Algebra and Its Applications, Fourth Edition, Gilbert Strang, Thomson
Books/Cole, 2006.

Matrix Methods: An Introduction, Second Edition, Richard Bronson, Academic

press, 1991.