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.
 Proof
From the fondamental Theorem of Algebra, 𝑃𝑓 is a product of linear
factors, then there exists a basis ℬ = 𝑣1 , ⋯ , 𝑣𝑛 of 𝑉 such that
𝑓 is triangularizable.
𝑃𝑓 (𝑋) = 𝜆1 − X ⋯ 𝜆𝑛 − X
𝑃𝑓 (𝑓) = 𝜆1 𝐼𝑑𝑉 − 𝑓 ∘ ⋯ ∘ 𝜆𝑛 𝐼𝑑𝑉 − 𝑓
Set 𝑔𝑖 = 𝜆1 𝐼𝑑𝑉 − 𝑓 ∘ ⋯ ∘ 𝜆𝑖 𝐼𝑑𝑉 − 𝑓 , (𝑖 = 1, ⋯ , 𝑛).
In particular 𝑔𝑛 = 𝑃𝑓 𝑓 and we want to prove that 𝑔𝑛 = 0.
We will prove by induction on 𝑖 ∈ 1, ⋯ , 𝑛 that
℘ (𝑖) ∶ 𝑔𝑖 𝑣1 = 𝑔𝑖 𝑣2 = ⋯ = 𝑔𝑖 𝑣𝑖 = 0 for all 𝑖 ∈ 1, ⋯ , 𝑛 .
 Proof
For i=1, it is true.
Let 𝑖 ∈ 1, ⋯ , 𝑛 − 1 . We suppose that ℘ (𝑖) is true and we prove
that ℘ 𝑖 + 1 is true.
We have
𝑔𝑖+1 = 𝑔𝑖 ∘ 𝜆𝑖+1 𝐼𝑑𝑉 − 𝑓 = 𝜆𝑖+1 𝐼𝑑𝑉 − 𝑓 ∘ 𝑔𝑖 ,
then 𝑔𝑖+1 𝑣1 = 𝑔𝑖+1 𝑣2 = ⋯ = 𝑔𝑖+1 𝑣𝑖 = 0
since ℘ 𝑖 is true.
We have also 𝑔𝑖+1 𝑣𝑖+1 = 𝑔𝑖 𝜆𝑖+1 𝑣𝑖+1 − 𝑓(𝑣𝑖+1 ) = 0.
Therefore ℘ 𝑖 is true for all 𝑖 ∈ 1, ⋯ , 𝑛 , thus 𝑔𝑛 = 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 7
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
2
𝑋−1 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.
 Proof
Assume that all the roots of 𝑚𝑓 𝑋 are distinct. Write
𝑚𝑓 𝑋 = (𝑋 − 𝜆1) … (𝑋 − 𝜆𝑝 ) .
(𝑋 − 𝜆𝑖 ) and (𝑋 − 𝜆𝑗 ) are coprime for all 𝑖 ≠ 𝑗.
By the decomposition theorem, we have
ker 𝑚𝑓 𝑓 = ker 𝑓 − 𝜆1 𝐼𝑑𝑉 ⨁ ⋯ ⨁ker 𝑓 − 𝜆𝑝𝐼𝑑𝑉 .
=𝑉𝜆1 ⨁ ⋯ ⨁𝑉𝜆𝑝
Since 𝑚𝑓 𝑓 = 0, then ker 𝑚𝑓 𝑓 = 𝑉, so 𝑉 = 𝑉𝜆1 ⨁ ⋯ ⨁𝑉𝜆𝑝 .
By theorem 3, 𝑓 is diagonalisable.
 Proof
Suppose 𝑓 diagonalisable. Thus there is a basis 𝐵 of 𝑉 such that
𝜆1 0 0 ⋯ 0 𝜆1 0 0 ⋯ 0
0 𝜆1 0 ⋱ ⋮ 0 𝜆1 0 ⋱ ⋮
𝑀 𝑓, 𝐵 = ⋮ 0 ⋱ ⋱ 0 = ⋮ 0 ⋱ ⋱ 0
⋮ ⋮ ⋱ 𝜆𝑝 0 ⋮ ⋮ ⋱ 𝜆𝑝 0
0 0 ⋯ 0 𝜆𝑝 0 0 ⋯ 0 𝜆𝑝

𝜆1𝐼𝑘1 0
= ⋱
0 𝜆𝑝𝐼𝑘𝑝
Where 𝜆1, 𝜆2, ⋯ , 𝜆𝑘 ∈ 𝐾 are all distinct eigenvalues of 𝑓.
 Proof
For all 𝑖 ∈ 1, ⋯ , 𝑝
(𝜆1−𝜆𝑖 )𝐼𝑘1 0
𝑀 𝑓, 𝐵 − 𝜆𝑖 𝐼𝑛 = 0 .
0 (𝜆𝑝−𝜆𝑖 )𝐼𝑘𝑝
Then 𝑀 𝑓, 𝐵 − 𝜆1 𝐼𝑛 ⋯ 𝑀 𝑓, 𝐵 − 𝜆𝑝𝐼𝑛 = 0.
So 𝑀 𝑓, 𝐵 is a root of the polynomial
𝑄 𝑋 = (𝑋 − 𝜆1 ) ⋯ (𝑋 − 𝜆𝑝 ).
By proposition 7, 𝑚𝑓 𝑋 𝑄 𝑋 so deg 𝑚𝑓 𝑋 ≤ deg 𝑄(𝑋).
By proposition 7, 𝑚𝑓 𝜆𝑖 = 0 for all 𝑖 ∈ 1, ⋯ , 𝑝 .
Since the 𝜆𝑖 are all distinct then 𝑄(𝑋) ∕ 𝑚𝑓 𝑋 .
 Proof
Since 𝑄 𝑋 and 𝑚𝑓 𝑋 are monic then 𝑚𝑓 𝑋 = 𝑄 𝑋 .
This proves that 𝑚𝑓 𝑋 has only simple roots.
 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.

 Jordan Block
Definition 7
A Jordan block is a square matrix of order 𝑝 of the form

𝜆 1 0 ⋯ 0
0 𝜆 1 ⋱ ⋮
𝐽 𝜆 = ⋮ 0 ⋱ ⋱ 0 ,
⋮ ⋮ ⋱ 𝜆 1
0 0 ⋯ 0 𝜆

where 𝜆 ∈ 𝐾 and 𝑝 ∈ ℕ∗ .
 Jordan Block Properties
Proposition 8
We have
1)
0 ⋯ 0 1
𝑝−1
𝐽 𝜆 − 𝜆𝐼𝑝 = 0 ⋯ 0 0
⋮ ⋮ ⋮ ⋮
0 0 ⋯ 0
𝑝
and 𝐽 𝜆 − 𝜆𝐼𝑝 = 0.
2) 𝑃𝐽 𝜆 𝑋 = 𝜆 − 𝑋 𝑝 = −1 𝑝
𝑋 − 𝜆 𝑝.
3) 𝑚𝐽 𝜆 𝑋 = 𝑋 − 𝜆 𝑝 .
 Jordan Matrix
Definition 8
A Jordan matrix is a block diagonal matrix of the form

𝐽1 𝜆1 ⋯ (0)
⋮ ⋱ ⋮
0 ⋯ 𝐽𝑝 𝜆𝑝
where 𝐽𝑖 𝜆𝑖 is a Jordan block for 𝑖 ∈ 1, ⋯ , 𝑝 .
Remark 2
The Jordan blocks can have distinct orders, and some 𝜆𝑖 can be equal.
 Example of Jordan Matrix
Example 9
Here is a Jordan matrix
−3 1
0 −3
−3 1 0 (0)
0 −3 1
𝐴= 0 0 −3
2
5 1 0
(0) 0 5 1
0 0 5
 Example of Jordan Matrix
Example 9
The matrix 𝐴 has
• A Jordan block 2 × 2 corresponding to the eigenvalue −3.
• A Jordan block 3 × 3 corresponding to the same eigenvalue −3.
• A Jordan block 1 × 1 corresponding to the eigenvalue 2.
• A Jordan block 3 × 3 corresponding to the eigenvalue 5.
 Existence of Jordan Form
Theorem 9
Let 𝐴 ∈ ℳ𝑛 𝐾 , and suppose that the characteristic polynomial of 𝐴 is
a product of linear factors. Then 𝐴 is similar to a Jordan matrix, that is,
there exists an invertible matrix 𝑃 ∈ ℳ𝑛 𝐾 such that

𝐽1 ⋯ (0)
𝑃−1 𝐴𝑃 = ⋮ ⋱ ⋮ ,
0 ⋯ 𝐽𝑝
where the 𝐽𝑖 , 1 ≤ 𝑖 ≤ 𝑝, are Jordan blocks.
 Existence of Jordan Form

We can rephrase Theorem 9 as follows.

Theorem 9 bis
Let 𝑓 be an endomorphism of 𝑉and suppose that the characteristic polynomial of 𝑓
is a product of linear factors. Then there exists a basis ℬ of 𝑉 such that the matrix
of 𝑓 with respect to that basis is a Jordan matrix, that is,

𝐽1 ⋯ (0)
𝑀(𝑓, ℬ) = ⋮ ⋱ ⋮ ,
0 ⋯ 𝐽𝑝
where the 𝐽𝑖 , 1 ≤ 𝑖 ≤ 𝑝, are Jordan blocks.
 Some Properties
Proposition 9
• The 𝜆𝑖 are the eigenvalues of 𝐴.
• The Jordan form of 𝐴 is unique up to reordering the Jordan blocks.
• The number of blocks corresponding to 𝜆 is equal to dim 𝑉𝜆 .
• The sum of the orders of the Jordan blocks corresponding to 𝜆 is the
multiplicity of 𝜆 as a root of the characteristic polynomial of 𝐴.
• The order of the greatest Jordan block corresponding to 𝜆 is the
multiplicity of 𝜆 as a root of the minimal polynomial of 𝐴.
 Algorithm Determining the Jordan Form
The following algorithm allows us to determine a Jordan form for each
square matrix with entries in ℂ.
In what follows, 𝐴 is an 𝑛 × 𝑛 matrix with entries in ℂ, 𝜆 is an
eigenvalue of 𝐴, and 𝑁 = 𝐴 − 𝜆𝐼𝑛 .
For convenience, we denote a matrix and the endomorphism
associated to it by the same notation.
 Step 1

Step 1
We determine the subspaces of ℂ𝑛
ker 𝑁 ⊊ ker 𝑁 2 ⊊ ⋯ ⊊ ker 𝑁 𝑚
where 𝑚 is the least positive integer so that ker 𝑁 𝑚 has maximal
dimension among the subspaces ker 𝑁 𝑖 , 𝑖 ≥ 1.
Remark
If 𝑘 is the multiplicity of the eigenvalue 𝜆 then 𝑚 ≤ 𝑘.
 Step 2 (1)
Step 2
(1) We complete a basis of ker 𝑁 𝑚−1 to obtain a basis of ker 𝑁 𝑚 .
Denote by 𝑣1 , 𝑣2 , ⋯ that completion.

𝑘𝑒𝑟 𝑁 𝑚 = 𝑘𝑒𝑟 𝑁 𝑚−1 ⨁ < 𝑣1 , 𝑣2 , ⋯ >

Then 𝑁𝑣1 , 𝑁𝑣2 , ⋯ is an ordered set of L.I. vectors of ker 𝑁 𝑚−1 , and
𝑁𝑣1 , 𝑁𝑣2 , ⋯ ∩ ker 𝑁 𝑚−2 = 0 .
 Step 2 (2)
Step 2
(2) We complete a basis of
ker 𝑁 𝑚−2 ⨁ 𝑁𝑣1 , 𝑁𝑣2 , ⋯
to obtain a basis of ker 𝑁 𝑚−1 .
Denote by 𝑤1 , 𝑤2 , ⋯ that completion.

ker 𝑁 𝑚−1 = ker 𝑁 𝑚−2 ⨁ 𝑁𝑣1 , 𝑁𝑣2 , ⋯ ⨁⟨𝑤1 , 𝑤2 , ⋯ ⟩

Then 𝑁 2 𝑣1 , 𝑁 2 𝑣2 , ⋯ , 𝑁𝑤1 , 𝑁𝑤2 , ⋯ is an ordered set of L.I. vectors of

ker 𝑁 𝑚−2 , and
𝑁 2 𝑣1 , 𝑁 2 𝑣2 , ⋯ , 𝑁𝑤1 , 𝑁𝑤2 , ⋯ ∩ ker 𝑁 𝑚−3 = 0 .
 Step 2 (3)
Step 2
(3) We complete a basis of
ker 𝑁 𝑚−3 ⨁ 𝑁 2 𝑣1 , 𝑁 2 𝑣2 , ⋯ , 𝑁𝑤1 , 𝑁𝑤2 , ⋯
to obtain a basis of ker 𝑁 𝑚−2 .
Denote by 𝑥1 , 𝑥2 , ⋯ that completion.

ker𝑁 𝑚−2 = ker 𝑁 𝑚−3 ⨁ 𝑁 2 𝑣1 , 𝑁 2 𝑣2 , ⋯ , 𝑁𝑤1 , 𝑁𝑤2 , ⋯ ⨁ 𝑥1 , 𝑥2 , ⋯

Then 𝑁 3 𝑣1 , 𝑁 3 𝑣2 , ⋯ , 𝑁 2 𝑤1 , 𝑁 2 𝑤2 , ⋯ , 𝑁𝑥1 , 𝑁𝑥2 , ⋯ is an ordered set

of L.I. vectors of ker 𝑁 𝑚−3 , and
𝑁 3 𝑣1 , 𝑁 3 𝑣2 , ⋯ , 𝑁 2 𝑤1 , 𝑁 2 𝑤2 , ⋯ , 𝑁𝑥1 , 𝑁𝑥2 , ⋯ ∩ ker 𝑁 𝑚−4 = 0 .
 Step 2 (⋮)

Step 2

⋮
⋮
⋮
 Step 2 (s-1)
Step 2
(s-1) We complete a basis of
ker 𝑁⨁ 𝑁 𝑚−2 𝑣1 , 𝑁 𝑚−2 𝑣2 , ⋯ , 𝑁 𝑚−3 𝑤1 , 𝑁 𝑚−3 𝑤2 , ⋯
to obtain a basis of ker 𝑁 2 .
Denote by 𝑦1 , 𝑦2 , ⋯ that completion.

ker 𝑁 2 = ker 𝑁⨁ 𝑁 𝑚−2 𝑣1 , 𝑁 𝑚−2 𝑣2 , ⋯ , 𝑁 𝑚−3 𝑤1 , 𝑁 𝑚−3 𝑤2 , ⋯ ⨁ 𝑦1 , 𝑦2 , ⋯

Then 𝑁 𝑚−1 𝑣1 , 𝑁 𝑚−1 𝑣2 , ⋯ , 𝑁 𝑚−2 𝑤1 , 𝑁 𝑚−2 𝑤2 , ⋯ , 𝑁𝑦1 , 𝑁𝑦2 , ⋯ is an

ordered set of L.I. vectors of ker 𝑁 .
 Step 2 (s)
Step 2
(s) At this last intermediate step, we complete the ordered set of L.I. vectors of ker 𝑁
obtained in the previous step :
𝑁 𝑚−1𝑣1, 𝑁 𝑚−1𝑣2, ⋯ , 𝑁 𝑚−2𝑤1, 𝑁 𝑚−2𝑤2, ⋯ , 𝑁𝑦1, 𝑁𝑦2, ⋯
to obtain a basis of ker 𝑁 .
Denote by 𝑧1, 𝑧2, ⋯ that completion.

ker 𝑁 = 𝑁 𝑚−1𝑣1, 𝑁 𝑚−1𝑣2, ⋯ , 𝑁 𝑚−2𝑤1, 𝑁 𝑚−2𝑤2, ⋯ , 𝑁𝑦1, 𝑁𝑦2, ⋯ ⨁ 𝑧1, 𝑧2, ⋯

So
𝑁 𝑚−1𝑣1, 𝑁 𝑚−1𝑣2, ⋯ , 𝑁 𝑚−2𝑤1, 𝑁 𝑚−2𝑤2, ⋯ , 𝑁𝑦1, 𝑁𝑦2, ⋯ , 𝑧1, 𝑧2, ⋯
is a basis of ker 𝑁 .
 Step 3
Step 3
By adjoining to that basis of ker 𝑁 all the bases of the complementary
subspaces of ker 𝑁 𝑖 in ker 𝑁𝑚𝑖+1 1 ≤ 𝑖 ≤ 𝑚 − 1 obtained in the previous
steps, we get a basis of ker 𝑁 , but we must arrange its vectors as follows :
𝑁 𝑚−1𝑣1, 𝑁 𝑚−2𝑣1, ⋯ , 𝑣1, 𝑁 𝑚−1𝑣2, 𝑁 𝑚−2𝑣2, ⋯ , 𝑣2, ⋯ , ⋯
𝑁 𝑚−2𝑤1, 𝑁 𝑚−3𝑤1, ⋯ , 𝑤1, 𝑁 𝑚−2𝑤2, 𝑁 𝑚−3𝑤2, ⋯ , 𝑤2, ⋯ , ⋯
⋮ .
𝑁𝑦1, 𝑦1, 𝑁𝑦2, 𝑦2, ⋯
𝑧1, 𝑧2, ⋯
This is a basis giving all Jordan blocks corresponding to the eigenvalue 𝜆.
 Step 4

Step 4

Finally, a Jordan basis for 𝐴 is simply the union of all the bases
corresponding to all the distinct eigenvalues of 𝐴.
 Examples
2 2 3
1) A = 1 3 3 ;
−1 −2 −2

−1 −1 0
2) A = 0 −1 −2 ;
0 0 −1

2 −1 0 1
0 3 −1 0
3) 𝐴 = ;
0 1 1 0
0 −1 0 3
 Examples
2 −4 2 2
4) 𝐴 = −2 0 1 3 ;
−2 −2 3 3
−2 −6 3 7

0 1 0 0
5) 𝐴 = 0 0 0 2 ;
0 0 0 0
0 0 0 0

1 0 0 1 1
1 2 −1 1 1
6) 𝐴 = 1 1 0 1 1 .
0 0 0 2 1
0 0 0 0 2
 Example

Example 10
Find the Jordan matrix corresponding to the matrix

4 3 −2
𝐴 = −3 −1 3 .
2 3 0
 Example

Example 11
Find the Jordan matrix corresponding to the matrix

5 0 4 −2 −3
−2 3 −3 2 4
𝐴= 0 0 3 0 0 .
0 0 0 3 1
1 0 2 −1 1