Lecture 1 :

Diagonalization Revisited
ENSIA, December 2022
 Contents

• Complex Eigenvalues
• Triangularization
• Cayley-Hamilton Theorem
• Minimal Polynomial
• Matrix Exponential
• Systems of Differential Equations
 Eigenvalues and Eigenvectors

Throughout this lecture, 𝐾 will denote ℝ or ℂ, and 𝑉 = 𝐾 𝑛 .

Definition 1
Let 𝐴 ∈ ℳ𝑛 𝐾 . If there exist 𝜆 ∈ 𝐾 and a nonzero column vector
𝑥 ∈ 𝐾 𝑛 satisfying
𝐴𝑥 = 𝜆𝑥,
we say that 𝜆 is an eigenvalue of 𝐴 and 𝑥 is an eigenvector of 𝐴
corresponding to 𝜆.
 Characteristic Polynomial
Definition 2
The characteristic polynomial of the matrix 𝐴 is defined by
𝑃𝐴 𝑋 = det 𝐴 − 𝑋𝐼𝑛 .

The characteristic polynomial allows us to find all the eigenvalues of a

given matrix as the following result shows.

Proposition 1
A scalar 𝜆 ∈ 𝐾 is an eigenvalue of 𝐴 if and only if 𝑃𝐴 𝜆 = 0.
 Example

Example 1
Consider the matrix
0 1
𝐴= .
−1 0
We have

−𝑋 1
𝑃𝐴 𝑋 = det 𝐴 − 𝑋𝐼2 = = 𝑋 2 + 1.
−1 −𝑋

Then, the eigenvalues of 𝐴 are 𝜆1 = 𝑖 and 𝜆2 = −𝑖.

 Eigenspace
Proposition 2
Let 𝜆 be an eigenvalue of 𝐴. Then the set
𝑉𝜆 = 𝑣 ∈ 𝑉 ∶ 𝐴𝑣 = 𝜆𝑣
is a subspace of 𝑉.

Definition 3
The subspace 𝑉𝜆 is called eigenspace corresponding to 𝜆.
 Example
Example 2
Consider again the matrix of Example 1, then
𝑉𝜆1 = 𝑉𝑖 = 𝑣 ∈ ℂ2 : 𝐴𝑣 = 𝑖𝑣 .
𝑥
Set 𝑣 = 𝑦 , then we have
0 1 𝑥 𝑥 𝑦 = 𝑖𝑥
𝐴𝑣 = 𝑖𝑣 ⇔ 𝑦 =𝑖 𝑦 ⇔ ⇔ 𝑦 = 𝑖𝑥.
−1 0 −𝑥 = 𝑖𝑦
2 1
Therefore 𝑉𝑖 is the subspace of ℂ spanned by , and dim 𝑉𝑖 = 1.
𝑖
 Complex Eigenvalues of Real Matrices
Proposition 3
Let 𝐴 ∈ ℳ𝑛 ℝ and suppose that 𝜆 is a non-real complex eigenvalue of
𝐴 with multiplicity 𝑚. Then we have
• The conjugate 𝜆 is an eigenvalue of 𝐴 with multiplicity 𝑚.
• A vector 𝑣 is an eigenvector corresponding to 𝜆 if and only if its
conjugate 𝑣 is an eigenvector corresponding to 𝜆.
• The corresponding eigenspaces 𝑉𝜆 and 𝑉𝜆 have the same dimension.
More precisely, 𝑣1 , … , 𝑣𝑘 is a basis of 𝑉𝜆 if and only if 𝑣1 , … , 𝑣𝑘 is a
basis of 𝑉𝜆 .
 Example
Example 3
Consider once again the matrix of Example 1, then, from Proposition 3,
we have
• 𝜆2 = −𝑖 is an eigenvalue of 𝐴 with multiplicity 1.
1
•𝑣= is an eigenvector of 𝐴.
−𝑖
2 1
• 𝑉−𝑖 is the subspace of ℂ spanned by , and dim 𝑉−𝑖 = 1.
−𝑖
 Diagonalization
Definition 4
A square matrix 𝐴 ∈ ℳ𝑛 𝐾 is said to be diagonalizable if it is similar to a
diagonal matrix, that is, there exists an invertible matrix 𝑃 ∈ ℳ𝑛 𝐾 such
that 𝑃−1 𝐴𝑃 is a diagonal matrix.
Remark 1
Let 𝑓 be the endomorphism of 𝑉 represented by the matrix 𝐴. Then « 𝐴 is
diagonalizable » means that there exists a basis ℬ of 𝑉 consisting of
eigenvectors such that ℳ(𝑓, ℬ) is diagonal. Moreover, we can take 𝑃 as
matrix whose columns are eigenvectors 𝑣1 , ⋯ , 𝑣𝑛 corresponding respectively
to eigenvalues 𝜆1 , ⋯ , 𝜆𝑛 and we have
𝑃−1 𝐴𝑃 = diag 𝜆1 , ⋯ , 𝜆𝑛 .
 Diagonalization : A Sufficient Condition

Theorem 1
If a matrix 𝐴 ∈ ℳ𝑛 𝐾 has 𝑛 distinct eigenvalues, then it is
diagonalizable.
 Diagonalization : A Necessary and Sufficient
Condition
Theorem 2
Let 𝜆1 , ⋯ , 𝜆𝑘 the distinct eigenvalues of a matrix 𝐴 ∈ ℳ𝑛 𝐾 , and
denote by 𝑚1 , ⋯ , 𝑚𝑘 their respectives multiplicities as roots of the
characteristic polynomial of 𝐴. Then 𝐴 is diagonalizable if and only if
dim 𝑉𝜆𝑖 = 𝑚𝑖
for all 𝑖 ∈ 1, ⋯ , 𝑘 .
 Example
Example 4
We have seen that the matrix
0 1
𝐴=
−1 0
has two distinct eigenvalues 𝜆1 = 𝑖 and 𝜆2 = −𝑖. Then, by Theorem 1, 𝐴 is diagonalizable.
1 1
We have also seen that 𝑣1 = is an eigenvector corresponding to 𝜆1 , and 𝑣2 = is
𝑖 −𝑖
an eigenvector corresponding to 𝜆2 . Thus, we can take
1 1
𝑃= ,
𝑖 −𝑖
and we obtain
𝑖 −𝑖 −1 0 1 1 1 𝑖 0
𝑃−1 𝐴𝑃 = = .
2 −𝑖 1 −1 0 𝑖 −𝑖 0 −𝑖
 Diagonalizable Endomorphism

Definition 5
An endomorphism 𝑓 of 𝑉 is said to be diagonalizable if there exists a
basis ℬ of 𝑉 such that the corresponding matrix ℳ 𝑓, ℬ is diagonal.
 A Characterization of Diagonalizability
Theorem 3
Let 𝑓 be an endomorphism of 𝑉 and let 𝜆1 , ⋯ , 𝜆𝑘 be the distinct
eigenvalues of 𝑓. Then the following are equivalent :
1) 𝑓 is diagonalizable ;
2) 𝑉 has a basis consisting of eigenvectors of 𝑓 ;
3) 𝑉 = ⨁𝑘𝑗=1 𝑉𝜆𝑗 ;
𝑘
4) 𝑗=1 dim 𝑉𝜆𝑗 = dim 𝑉.
 Triangularization
Definition 6
An endomorphism 𝑓 of 𝑉 is said to be triangularizable if there exists a
basis ℬ of 𝑉 such that the corresponding matrix ℳ 𝑓, ℬ is (upper)
triangular.
Theorem 4
An endomorphism 𝑓 of 𝑉 is triangularizable if and only if its
characteristic polynomial 𝑃𝑓 (𝑋) can be written as a product of linear
factors, that is
𝑃𝑓 𝑋 = −1 𝑛 𝑋 − 𝜆1 ⋯ 𝑋 − 𝜆𝑛 ,
where 𝜆1 , ⋯ , 𝜆𝑛 are (not necessarily distinct) elements of 𝐾.
 Example

Example 5
The matrix
0 1
𝐴=
−1 0
is triangularizable over ℂ but not over ℝ.
 Polynomials of Matrices and Polynomials of
Endomorphisms
𝑚 𝑗
Given 𝑃 𝑋 = 𝑎
𝑗=0 𝑗 𝑋 ∈ 𝐾 𝑋 , 𝐴 ∈ ℳ𝑛 𝐾 and an endomorphism 𝑓
of 𝑉, we write
𝑚

𝑃 𝐴 ≔ 𝑎𝑗 𝐴 𝑗
𝑗=0
and
𝑚

𝑃 𝑓 ≔ 𝑎𝑗 𝑓 𝑗 ,
𝑗=0
where 𝐴0 = 𝐼𝑛 , 𝑓 0 = 𝐼𝑑𝑉 , and 𝑓 𝑗 = 𝑓 ∘ ⋯ ∘ 𝑓 (𝑗 times).
 Properties of Polynomials of Endomorphisms
Let 𝑃 𝑋 , 𝑄(𝑋) ∈ 𝐾[𝑋] and let 𝑓 be an endomorphism of 𝑉, then we
have
𝑃 𝑋 .𝑄 𝑋 𝑓 = 𝑃 𝑓 ∘ 𝑄 𝑓 .
Although the composition is not commutative in general, in our case,
thanks to the commutativity of the product of polynomials, we have

𝑃 𝑓 ∘𝑄 𝑓 =𝑄 𝑓 ∘𝑃 𝑓 .
More generally, for 𝑚 polynomials we have
(𝑃1 𝑋 ⋯ 𝑃𝑚 (𝑋))(𝑓) = 𝑃1 (𝑓) ∘ ⋯ ∘ 𝑃𝑚 (𝑓).
 Eigenvalues and Roots of Polynomials

Theorem 5
Let 𝑉 be a finite dimensional vector space over 𝐾 and let 𝑓 be an
endomorphism of 𝑉. Let 𝑃(𝑋) ∈ 𝐾[𝑋] a nonzero polynomial verifying
𝑃 𝑓 = 0.
Then every eigenvalue of 𝑓 is a root of 𝑃(𝑋).
 Characteristic Polynomial of Endomorphism

Definition 7
Let 𝑉 be a finite dimensional vector space over 𝐾 and let 𝑓 be an
endomorphism of 𝑉. Let ℬ be a basis of 𝑉. Then the characteristic
polynomial of 𝑓 is given by
𝑃𝑓 𝑋 = det(ℳ 𝑓, ℬ − 𝑋𝐼𝑛 ) .
 The Cayley-Hamilton Theorem
Theorem 6 (Cayley-Hamilton Theorem)
Let 𝑉 be a finite dimensional vector space over 𝐾 and let 𝑓 be an
endomorphism of 𝑉. Then we have
𝑃𝑓 𝑓 = 0.
In matrix form :
Let 𝐴 ∈ ℳ𝑛 𝐾 , then we have
𝑃𝐴 𝐴 = 0.
 Example
Example 6
The matrix
0 1
𝐴=
−1 0

has characteristic polynomial 𝑃𝐴 𝑋 = 𝑋 2 + 1. We have

0 1 0 1 1 0 −1 0 1 0
𝑃𝐴 𝐴 = + = + = 0.
−1 0 −1 0 0 1 0 −1 0 1
 Minimal Polynomial

The Cayley-Hamilton Theorem allows us to define the following.

Definition 6
The minimal polynomial of an endomorphism 𝑓 is the monic
polynomial 𝑚𝑓 (𝑋) of least degree such that 𝑚𝑓 𝑓 = 0.
 Minimal Polynomial Properties
Proposition 4
Let 𝑓 be an endomorphism of 𝑉.
1) If 𝜆 is an eigenvalue of 𝑓, then 𝑚𝑓 𝜆 = 0.
2) Let 𝑃(𝑋) ∈ 𝐾 𝑋 such that 𝑃 𝑓 = 0. Then
𝑚𝑓 𝑋 |𝑃(𝑋).
In particular, 𝑚𝑓 𝑋 |𝑃𝑓 𝑋 , so we have

deg 𝑚𝑓 𝑋 ≤ dim 𝑉.
 Determining the Minimal Polynomial
Suppose that the characteristic polynomial of 𝑓 has the form
𝑃𝑓 𝑋 = (−1)𝑛 (𝑋 − 𝜆1 )𝛼1 … (𝑋 − 𝜆𝑘 )𝛼𝑘 ,
where 𝜆1 , ⋯ , 𝜆𝑘 are distinct elements of 𝐾, then the minimal
polynomial of 𝑓 has the form
𝑚𝑓 𝑋 = (𝑋 − 𝜆1 )𝛽1 … (𝑋 − 𝜆𝑘 )𝛽𝑘 ,
where 1 ≤ 𝛽𝑖 ≤ 𝛼𝑖 for all 𝑖 ∈ 1, ⋯ , 𝑘 .
To determine the minimal polynomial, we check if
𝑋 − 𝜆1 𝑐1 … 𝑋 − 𝜆𝑘 𝑐𝑘 𝑓 = 0 (1)
by giving increasing values to the 𝑐𝑖 until we get the equation (1).
 Example
Example 7
The characteristic polynomial of the matrix

1 1 0
𝐴= 0 1 0
0 0 1
is given by 𝑃𝐴 𝑋 = 1 − 𝑋 3 . We have

𝑋 − 1 𝐴 = 𝐴 − 𝐼3 ≠ 0,
2
0 1 0
𝑋−1 2 𝐴 = (𝐴 − 𝐼3 )2 = 0 0 = 0.
0
0 0 0

So the minimal polynomial of 𝑓 is 𝑚𝐴 𝑋 = 𝑋 − 1 2 .

 Decomposition Theorem
Theorem 7
Let 𝑓 be an endomorphism of 𝑉 and let 𝑃 𝑋 ∈ 𝐾 𝑋 . Suppose that
𝑃 𝑋 = 𝑃1 𝑋 𝑃2 𝑋 ⋯ 𝑃𝑘 𝑋 ,
where the polynomials 𝑃1 𝑋 , 𝑃2 𝑋 , ⋯ , 𝑃𝑘 𝑋 are pairwise coprime. Then we
have
ker 𝑃 𝑓 = ker 𝑃1 (𝑓)⨁ ker 𝑃2 (𝑓)⨁ ⋯ ⨁ker 𝑃𝑘 (𝑓).
 A New Characterization of the Diagonalizability

Theorem 8
An endomorphism 𝑓 is diagonalizable if and only if its minimal polynomial
is the product of distinct linear factors.
 Example
Example 8
The matrix of Example 7 is not diagonalizable.
The characteristic polynomial of the matrix

1 0 −2
𝐴= 0 1 1
0 0 2
2
is given by 𝑃𝐴 𝑋 = 1 − 𝑋 2 − 𝑋 . We have

0 0 −2 −1 0 −2
𝑋 − 1 𝑋 − 2 𝐴 = 𝐴 − 𝐼3 𝐴 − 2𝐼3 = 0 0 1 0 −1 1 = 0.
0 0 1 0 0 0

Then the minimal polynomial of 𝑓 is 𝑚𝐴 𝑋 = 𝑋 − 1 𝑋 − 2 , and so 𝐴 is diagonalizable.