Chapter 2

Reduction of endomorphisms and

matrices

In all this chapter, K denotes a commutative field and E is a K-vector space of finite
dimension n ∈ N∗ .

2.1 Characteristic polynomial

Proposition 2.1.1 Let A ∈ Mn (K) and X be an indeterminate. The determinant

PA (X) = |XIn − A|

is a monic polynomial in X with coefficients in K of degree n. We call it the characteristic

polynomial of A. Moreover,

PA (X) = X n − tr(A)X n−1 + · · · + (−1)n |A|.

Example 2.1.1

1) We have:
PIn (X) = |XIn − In | = |(X − 1)In | = (X − 1)n .
 
a b
2) If A = ∈ M2 (R), then
c d

X −a −b
PA (X) = |XI2 − A| =
−c X −d
= (X − a)(X − d) − bc
= X 2 − (a + d)X + (ad − bc)
= X 2 − tr(A)X + |A|.

22
Chapter 2 Reduction of endomorphisms and matrices

 
4 −1 −1 0
 0 3 −1 0 
 0 −1 3 0 . Then
Example 2.1.2 Let A =  

2 −1 −1 2

X −4 1 1 0
0 X −3 1 0
PA (X) = |XI4 − A| =
0 1 X −3 0
−2 1 1 X −2
X −4 1 1
= (X − 2) 0 X −3 1
0 1 X −3

X −3 1
= (X − 2)(X − 4)
1 X −3

= (X − 2)(X − 4)(X 2 − 6X + 8) = (X − 2)2 (X − 4)2 .

Proposition 2.1.2 Two similar matrices have the same characteristic polynomial.

Proof Let A, B ∈ Mn (K) be two similar matrices, then there exists P ∈ Mn (K) invertible
such that B = P −1 AP . So
XIn − B = P −1 (XIn )P − P −1 AP = P −1 (XIn − A)P.
Therefore the two matrices XIn − B and XIn − A are similar, so they have the same
determinant by the proposition 1.4.2. Hence
PB (X) = |XIn − B| = |XIn − A| = PA (X). 2

Remark 2.1.1 Let f be an endomorphism of E. We call characteristic polynomial of f ,

and we denote it by Pf (X), the characteristic polynomial of the representative matrix of f
in a basis of E over K. This definition is independent of the choice of the basis. Indeed, if b
and b0 are two bases of E over K, then the two matrices A = M (f, b) and B = M (f, b0 )
are similar by the theorem 1.4.1, and therefore PA (X) = PB (X) by the proposition 2.1.2.
As tr(A) = tr(f ) and |A| = det(f ), then
Pf (X) = X n − tr(f )X n−1 + · · · + (−1)n det(f ).

Example 2.1.3 Let f be the endomorphism of R4 defined by:

f (x1 , x2 , x3 , x4 ) = (4x1 − x2 − x3 , 3x2 − x3 , −x2 + 3x3 , 2x1 − x2 − x3 + 2x4 ).
Since M (f, C) = A is the matrix given in the example 2.1.2, then
Pf (X) = PA (X) = (X − 2)2 (X − 4)2 .

c Ali Ayad page 23

Chapter 2 Reduction of endomorphisms and matrices

Proposition 2.1.3 Let  

A B
M = ∈ Mm+n (K),
0 D
where A ∈ Mm (K), B ∈ Mm,n (K) and D ∈ Mn (K). Then
1) |M | = |A| |D|.

2) PM (X) = PA (X)PD (X).

 
0 1 4 5
 3 0 0 2 
Example 2.1.4 If M = 
 0 0 2
, then
0 
0 0 −5 4

X −1 X −2 0
PM (X) = = (X 2 − 3)(X − 2)(X − 4).
−3 X 5 X −4

Corollary 2.1.1 Let  

A1 ∗ ∗
..
M = 0 ∗  ∈ Mn (K),
 
.
0 0 Ar
where Ai ∈ Mni (K) and ni ∈ N∗ for all 1 ≤ i ≤ r.
1) |M | = |A1 | · · · |Ar |.

2) PM (X) = PA1 (X) · · · PAr (X).

2.2 Eigenvalues and eigenvectors

Definition 2.2.1 Let f be an endomorphism of E.
1) A nonzero vector x of E is said to be an eigenvector of f if there exists a scalar λ ∈ K
such that f (x) = λx.

2) A scalar λ ∈ K is said to be an eigenvalue of f if there exists a nonzero vector x of

E such that f (x) = λx. In this case, such a nonzero vector x of E is said to be an
eigenvector of f associated to the eigenvalue λ.

Example 2.2.1 Let f be the endomorphism of R3 defined by f (x1 , x2 , x3 ) =

(x1 , 2x2 , 3x3 ). As

f (e1 ) = e1 , f (e2 ) = 2e2 and f (e3 ) = 3e3 ,

then e1 , e2 and e3 are eigenvectors of f associated to the eigenvalues 1, 2 and 3 of f

respectively.

c Ali Ayad page 24

Chapter 2 Reduction of endomorphisms and matrices

Definition 2.2.2 Let f be an endomorphism of E.

1) We call spectrum of f , and we denote it by sp(f ), the set of all eigenvalues of f .
2) For every eigenvalue λ of f , put:
Eλ (f ) = ker(f − λ idE ).
Eλ (f ) is a subspace of E called the eigenspace of f associated to λ.

Remark 2.2.1 Let f be an endomorphism of E.

1) Let λ ∈ sp(f ) and x ∈ E − {0}. Then
(x is an eigenvector of f associated to λ) ⇔ f (x) = λx
⇔ (f − λ idE )(x) = 0
⇔ x ∈ Eλ (f ).
So Eλ (f ) − {0} is the set of all eigenvectors of f associated to λ. Hence
Eλ (f ) 6= {0}.

2) Let b be a basis of E over K and A = M (f, b). Let λ ∈ sp(f ). By the proposi-
tion 1.3.5,
 
M f − λ idE , b = M (f, b) − λM (idE , b) = A − λIn .

By the corollary 1.3.1,

Eλ (f ) = ker(f − λ idE ) = ker(A − λIn ).

3) If λ, β ∈ sp(f ) such that λ 6= β, then

Eλ (f ) ∩ Eβ (f ) = {0}.
In other words, an eigenvector of f is associated to one and only one eigenvalue of f .
Indeed, let x ∈ Eλ (f ) ∩ Eβ (f ), then f (x) = λx and f (x) = βx, so λx = βx, and
therefore (λ − β)x = 0. As λ − β 6= 0, then x = 0. Moreover,
Eλ (f ) + Eβ (f ) = Eλ (f ) ⊕ Eβ (f ).

Proposition 2.2.1 Let f be an endomorphism of E. A scalar λ ∈ K is an eigenvalue of f

if and only if the endomorphism f − λ idE is not bijective (i.e., not an automorphism of E).

Proof Let λ ∈ K. Then

h i
λ ∈ sp(f ) ⇔ (∃ x ∈ E) x =
6 0 and f (x) = λx
h i
⇔ (∃ x ∈ E) x =6 0 and (f − λ idE )(x) = 0
⇔ ker(f − λ idE ) 6= {0}
⇔ f − λ idE is not injective (by the proposition 1.3.2)
⇔ f − λ idE is not bijective (by the proposition 1.3.3). 2

c Ali Ayad page 25

Chapter 2 Reduction of endomorphisms and matrices

Corollary 2.2.1 Let f be an endomorphism of E. The eigenvalues of f are the roots of its
characteristic polynomial Pf (X) in K. Consequently, sp(f ) is a finite set of cardinal ≤ n.

Proof Let b be a basis of E over K and A = M (f, b), then

Pf (X) = PA (X) = |XIn − A|.

Let λ ∈ K. Then, by the proposition 1.3.5,

 
M f − λ idE , b = M (f, b) − λM (idE , b) = A − λIn .

So

λ ∈ sp(f ) ⇔ f − λ idE is not bijective (by the proposition 2.2.1)

⇔ A − λIn is not invertible (by the proposition 1.3.9)
⇔ |A − λIn | = 0 ⇔ (−1)n |λIn − A| = 0 ⇔ |λIn − A| = 0
⇔ PA (λ) = 0 ⇔ Pf (λ) = 0.

Hence the eigenvalues of f are the roots of Pf (X) in K. In the other side, as Pf (X) ∈ K[X]
is a 
polynomial
 of degree n, then Pf (X) has at most n roots in K, and therefore
card sp(f ) ≤ n. 2

Example 2.2.2 Let f be the endomorphism of R4 defined in the example 2.1.3. Since
Pf (X) = (X − 2)2 (X − 4)2 , then

sp(f ) = {2, 4}.

Let A = M (f, C) given in the example 2.1.2, then

E2 (f ) = ker(A − 2I4 ) and E4 (f ) = ker(A − 4I4 ).

Let x = (x1 , x2 , x3 , x4 ) ∈ R4 , then

x ∈ E2 (f ) ⇔ (A − 2I4 )X = 0
   
2 −1 −1 0 x1 
 2x1 − x2 − x3 = 0
 0 1 −1 0   x2  
x2 − x3 = 0
⇔ 
 0 −1 1   x3  = 0 ⇔  −x2 + x3 = 0
 
0 
2 −1 −1 0 x4 2x1 − x2 − x3 = 0

⇔ x1 = x2 = x3
⇔ x = (x3 , x3 , x3 , x4 ) = x3 (1, 1, 1, 0) + x4 (0, 0, 0, 1)
⇔ x = x3 u1 + x4 u2 ,

where u1 = (1, 1, 1, 0) and u2 = (0, 0, 0, 1). So

E2 (f ) = vect(u1 , u2 )

c Ali Ayad page 26

Chapter 2 Reduction of endomorphisms and matrices

and {u1 , u2 } is a basis of E2 (f ) over R. In the other side,

x ∈ E4 (f ) ⇔ (A − 4I4 )X = 0
   
0 −1 −1 0 x1 
 −x2 − x3 = 0
 0 −1 −1 0   x2 

−x2 − x3 = 0
⇔ 
 0 −1 −1 0  
 =0 ⇔
x3   −x2 − x3 = 0

2 −1 −1 −2 x4 2x1 − x2 − x3 − 2x4 = 0


⇔ x1 = x4 and x2 = −x3
⇔ x = (x4 , −x3 , x3 , x4 ) = x3 (0, −1, 1, 0) + x4 (1, 0, 0, 1)
⇔ x = x3 u3 + x4 u4 ,

where u3 = (0, −1, 1, 0) and u4 = (1, 0, 0, 1). So

E4 (f ) = vect(u3 , u4 )

and {u3 , u4 } is a basis of E4 (f ) over R.

Proposition 2.2.2 Let f be an endomorphism of E. Eigenvectors of f associated to

distinct eigenvalues of f are linearly independent over K.

Corollary 2.2.2 Let f be an endomorphism of E. The sum of eigenspaces of f associated

to distinct eigenvalues of f is direct. In other words, if λ1 , . . . , λr ∈ sp(f ) are pairwisely
distinct, then
Eλ1 (f ) + · · · + Eλr (f ) = Eλ1 (f ) ⊕ · · · ⊕ Eλr (f ).

Definition 2.2.3 Let f be an endomorphism of E and λ ∈ sp(f ).

1) We call algebraic multiplicity of λ, and we denote it by am(λ), the order of multiplicity
of λ as a root of Pf (X).

• If am(λ) = 1, then we say that λ is a simple eigenvalue of f .

• If am(λ) = 2 (resp. am(λ) = 3), then we say that λ is a double (resp. triple)
eigenvalue of f .

2) We call geometric multiplicity of λ, and we denote it by gm(λ), the dimension of the

eigenspace Eλ (f ) of f associated to λ.

Example 2.2.3 Let f be the endomorphism of R4 defined in the example 2.1.3. Since
Pf (X) = (X − 2)2 (X − 4)2 , then am(2) = 2 and am(4) = 2. By the example 2.2.2,

gm(2) = dim E2 (f ) = 2 and gm(4) = dim E4 (f ) = 2.

Corollary 2.2.3 Let f be an endomorphism of E. For every λ ∈ sp(f ),

1 ≤ gm(λ) ≤ am(λ).

c Ali Ayad page 27

Chapter 2 Reduction of endomorphisms and matrices

Definition 2.2.4 Let A ∈ Mn (K) and fA be the endomorphism of Kn canonically associ-

ated to A, i.e., A = M (fA , C) where C is the canonical basis of Kn .

1) A nonzero vector x of Kn is said to be an eigenvector of A if it is an eigenvector of

fA , i.e., if there exists a scalar λ ∈ K such that AX = λX where X = M (x, C).

2) A scalar λ ∈ K is said to be an eigenvalue of A if it is an eigenvalue of fA , i.e., if

there exists a nonzero vector x of Kn such that AX = λX where X = M (x, C). In
this case, such a nonzero vector x of Kn is said to be an eigenvector of A associated to
the eigenvalue λ.

3) We call spectrum of A, and we denote it by sp(A), the set of eigenvalues of A, i.e.,

sp(A) = sp(fA ).

4) For every eigenvalue λ of A, put:

Eλ (A) = ker(A − λ In ).

Eλ (A) is a subspace of Kn called the eigenspace of A associated to λ.

Proposition 2.2.3 Let A ∈ Mn (K).

1) For any λ ∈ sp(A), Eλ (A) − {0} is the set of eigenvectors of A associated to λ.

Hence
Eλ (A) 6= {0}.

2) If λ, β ∈ sp(A) such that λ 6= β, then

Eλ (A) ∩ Eβ (A) = {0}.

3) A scalar λ ∈ K is an eigenvalue of A if and only if the matrix A−λ In is not invertible.

4) The eigenvalues of A are the roots of its characteristic polynomial PA (X) in K. Con-
sequently, sp(A) is a finite set of cardinal ≤ n.

5) Two similar matrices have the same eigenvalues by using the part (4) and the fact that
they have the same characteristic polynomial (proposition 2.1.2).

6) Eigenvectors of A associated to distinct eigenvalues of A are linearly independent over

K.

7) The sum of eigenspaces of A associated to distinct eigenvalues of A is direct. In other

words, if λ1 , . . . , λr ∈ sp(A) are pairwisely distinct, then

Eλ1 (A) + · · · + Eλr (A) = Eλ1 (A) ⊕ · · · ⊕ Eλr (A).

Definition 2.2.5 Let A ∈ Mn (K) and λ ∈ sp(A).

1) We call algebraic multiplicity of λ, and we denote it by am(λ), the order of multiplicity

of λ as a root of PA (X).

c Ali Ayad page 28

Chapter 2 Reduction of endomorphisms and matrices

• If am(λ) = 1, then we say that λ is a simple eigenvalue of A.

• If am(λ) = 2 (resp. am(λ) = 3), then we say that λ is a double (resp. triple)
eigenvalue of A.

2) We call geometric multiplicity of λ, and we denote it by gm(λ), the dimension of the

eigenspace Eλ (A) of A associated to λ.

Remark 2.2.2 Let A ∈ Mn (K).

1) For any λ ∈ sp(A),

1 ≤ gm(λ) ≤ am(λ).

2) The spectrum of A depends on the field K. Hence, if A ∈ Mn (R), then A can be also
considered as a matrix over C. For this, we distinguish two spectrums of A:

• The real spectrum of A, denoted by spR (A), is the set of eigenvalues of A as a

real matrix.
• The complex spectrum of A, denoted by spC (A), is the set of eigenvalues of A as
a complex matrix.
 
0 1
For example, if A = ∈ M2 (K), then PA (X) = X 2 + 1.
−1 0

• If K = R, then spR (A) = ∅.

• If K = C, then spC (A) = {i, −i}.

2.3 Diagonalization
Definition 2.3.1 We say that an endomorphism f of E is diagonalizable if there exists a
basis b of E over K such that M (f, b) is a diagonal matrix.

Example 2.3.1

1) The identity endomorphism idE of E is diagonalizable since M (idE , b) = In is diag-

onal for any basis b of E over K.

2) Let f be the endomorphism of R3 defined by f (x1 , x2 , x3 ) = (x1 , 2x2 , 3x3 ). As

 
1 0 0
M (f, C) =  0 2 0 
0 0 3

is a diagonal matrix, then f is diagonalizable.

c Ali Ayad page 29

Chapter 2 Reduction of endomorphisms and matrices

Remark 2.3.1 Let f be an endomorphism of E. Suppose that f is diagonalizable, then there

exists a basis b of E over K such that the matrix D = M (f, b) is diagonal, so there exist
λ1 , . . . , λn ∈ K (not necessary distinct) such that

D = diag(λ1 , . . . , λn ).

So
Pf (X) = PD (X) = (X − λ1 ) · · · (X − λn ).
Hence the terms on the main diagonal of M (f, b) are the eigenvalues of f counted with their
algebraic multiplicities. In particular,
X
n = deg Pf = am(λ).
λ∈sp(f )

Proposition 2.3.1 Let f be an endomorphism of E. Then f is diagonalizable if and only

if E has a basis formed by eigenvectors of f .

Corollary 2.3.1 If an endomorphism f of E has n pairwisely distinct eigenvalues, then f

is diagonalizable.

Proof Let λ1 , . . . , λn ∈ sp(f ) to be pairwisely distinct. Let u1 , . . . , un ∈ E be

eigenvectors of f associated to the eigenvalues λ1 , . . . , λn respectively. By the proposi-
tion 2.2.2, the vectors u1 , . . . , un are linearly independent over K. As dimK E = n, then
the family b = {u1 , . . . , un } is a basis of E over K formed by eigenvectors of f . Hence f
is diagonalizable by the proposition 2.3.1. 2

Example 2.3.2 Let f be the endomorphism of R4 defined by

f (x1 , x2 , x3 , x4 ) = (x2 + 4x3 + 5x4 , 3x1 + 2x4 , 2x3 , −5x3 + 4x4 ).

The representative matrix of f in the canonical basis C of R4 is M (f, C) = M where M is

the matrix given in the example 2.1.4. As
√ √
Pf (X) = PM (X) = (X + 3)(X − 3)(X − 2)(X − 4),

then n √ √ o
sp(f ) = − 3, 3, 2, 4 ,
and therefore f has 4 pairwisely distinct eigenvalues. Hence f is diagonalizable by the
corollary 2.3.1.

Proposition 2.3.2 Let f be an endomorphism of E. Then f is diagonalizable if and only if

E is the direct sum of the eigenspaces of f , i.e.,
M
E= Eλ (f ).
λ∈sp(f )

c Ali Ayad page 30

Chapter 2 Reduction of endomorphisms and matrices

Theorem 2.3.1 Let f be an endomorphism of E. Then f is diagonalizable if and only if

the following two conditions are satisfied:
i) Pf (X) splits in K[X], i.e., Pf (X) has all its roots in K.
ii) For every λ ∈ sp(f ), we have:
gm(λ) = am(λ).

Example 2.3.3 Diagonalize an endomorphism f of E returns to determine a basis b of E

over K and a diagonal matrix D ∈ Mn (K) such that M (f, b) = D.
Let f be the endomorphism of R4 defined in the example 2.1.3. By the example 2.2.3, Pf (X)
splits in R[X] and
gm(2) = am(2) and gm(4) = am(4).
So f is diagonalizable, and therefore
R4 = E2 (f ) ⊕ E4 (f ).
Moreover, by the example 2.2.2, {u1 , u2 } is a basis of E2 (f ) over R and {u3 , u4 } is a basis
of E4 (f ) over R. So b = {u1 , u2 , u3 , u4 } is a basis of R4 over R and
 
2 0 0 0
 0 2 0 0 
M (f, b) =   0 0 4 0 .


0 0 0 4

Remark 2.3.2 The converse of the corollary 2.3.1 is not necessarily true. Indeed, the endo-
morphism f of the example 2.3.3 is diagonalizable, but f has only two distinct eigenvalues
(n = 4).

Definition 2.3.2 We say that a matrix A ∈ Mn (K) is diagonalizable if the endomorphism

fA of Kn canonically associated to A is diagonalizable.

Proposition 2.3.3 Let A ∈ Mn (K). Then A is diagonalizable if and only if A is similar

to a diagonal matrix.

Proof Let fA be the endomorphism of Kn canonically associated to A, then A = M (fA , C)

where C is the canonical basis of Kn .
N.C. Since A is diagonalizable, then fA is diagonalizable, so there exists a basis b of Kn
such that the matrix D = M (fA , b) is diagonal. As A = M (fA , C), then A is similar to
the diagonal matrix D by the theorem 1.4.1.
S.C. Let D ∈ Mn (K) be a diagonal matrix such that A is similar to D. By the
proof of the necessary condition of the theorem 1.4.1, there exists a basis B of Kn such
that D = M (fA , B). As D is diagonal, then fA is diagonalizable, and therefore A is
diagonalizable. 2

c Ali Ayad page 31

Chapter 2 Reduction of endomorphisms and matrices

Remark 2.3.3 Let A ∈ Mn (K). Suppose that A is diagonalizable, then A is similar to a

diagonal matrix D ∈ Mn (K), so there exist λ1 , . . . , λn ∈ K (not necessary distinct) such
that
D = diag(λ1 , . . . , λn ).
So, by using the proposition 2.1.2, we obtain:

PA (X) = PD (X) = (X − λ1 ) · · · (X − λn ).

Hence the terms on the main diagonal of D are the eigenvalues of A counted with their
algebraic multiplicities.

Example 2.3.4

1) Every diagonal matrix is diagonalizable by the fact that it is similar to itself.

 
2 −1 3
2) The matrix A =  0 2 5  ∈ M3 (R) is not diagonalizable. Indeed, suppose
0 0 2
that A is diagonalizable, then, by the remark 2.3.3, A is similar to a diagonal matrix
D ∈ M3 (R) whose terms of the main diagonal are the eigenvalues of A counted with
their algebraic multiplicities. As PA (X) = (X − 2)3 , then 2 is the only eigenvalue of
A with am(2) = 3. Therefore
 
2 0 0
D =  0 2 0  = 2I3 .
0 0 2

Since A and D are similar, then there exists P ∈ M3 (R) invertible such that

A = P −1 DP = P −1 (2I3 )P = 2P −1 P = 2I3 ,

which is impossible. Hence A is not diagonalizable.

Proposition 2.3.4 Let f be an endomorphism of E and A = M (f, b) where b is a basis

of E over K. Then f is diagonalizable if and only if A is diagonalizable.

Proof
N.C. Since f is diagonalizable, then there exists a basis B of E over K such that the matrix
D = M (f, B) is diagonal. As A = M (f, b), then A is similar to the diagonal matrix D
by the theorem 1.4.1. Hence A is diagonalizable by the proposition 2.3.3.
S.C. Since A is diagonalizable, then, by the proposition 2.3.3, A is similar to a diagonal matrix
D ∈ Mn (K), so there exists an invertible matrix P ∈ Mn (K) such that D = P −1 AP .
By the remark 1.2.1, there exists a basis b0 of E over K such that P is the transition matrix
from b to b0 . So
M (f, b0 ) = P −1 M (f, b)P = P −1 AP = D.
Therefore M (f, b0 ) is diagonal, and then f is diagonalizable. 2

c Ali Ayad page 32

Chapter 2 Reduction of endomorphisms and matrices

Proposition 2.3.5 Let A ∈ Mn (K).

1) A is diagonalizable if and only if Kn has a basis formed by eigenvectors of A.
2) If A has n pairwisely distinct eigenvalues, then A is diagonalizable.
3) A is diagonalizable if and only if Kn is the direct sum of the eigenspaces of A, i.e.,
M
Kn = Eλ (A).
λ∈sp(A)

4) A is diagonalizable if and only if the following two conditions are satisfied:

i) PA (X) splits in K[X], i.e., PA (X) has all its roots in K.
ii) For every λ ∈ sp(A), we have:
gm(λ) = am(λ).

Remark 2.3.4
1) The diagonalization of A depends on the field K. For example, if
 
0 1
A= ∈ M2 (K),
−1 0
then PA (X) = X 2 + 1.
• If K = R, then PA (X) does not split in R[X], and therefore A is not diagonaliz-
able (over R).
• If K = C, then PA (X) splits in C[X] and it has two distinct roots which are i and
−i. So A has two distinct eigenvalues, and as n = 2, then A is diagonalizable
(over C).
2) Diagonalize a matrix A ∈ Mn (K) returns to determine an invertible matrix
P ∈ Mn (K) and a diagonal matrix D ∈ Mn (K) such that P −1 AP = D.

Example 2.3.5 Let A be the matrix defined in the example 2.1.2. The endomorphism asso-
ciated to A is the endomorphism f of the example 2.3.3. Since f is diagonalizable, then A
is diagonalizable. Moreover, b = {u1 , u2 , u3 , u4 } is a basis of R4 over R where
u1 = (1, 1, 1, 0), u2 = (0, 0, 0, 1), u3 = (0, −1, 1, 0) and u4 = (1, 0, 0, 1).
Let  
1 0 0 1
 1 0 −1 0 
P =
 1

0 1 0 
0 1 0 1
be the transition matrix from the basis C to the basis b. As A = M (f, C), then
 
2 0 0 0
 0 2 0 0 
P −1 AP = M (f, b) =   0 0 4 0 .


0 0 0 4

c Ali Ayad page 33

Chapter 2 Reduction of endomorphisms and matrices

2.4 Trigonalization
Definition 2.4.1 We say that an endomorphism f of E is trigonalizable if there exists a
basis b of E over K such that M (f, b) is an upper triangular matrix.

Example 2.4.1
1) Every diagonalizable endomorphism is trigonalizable by the fact that every diagonal
matrix is an upper triangular matrix. The converse is not necessarily true. Indeed, let
fA be the endomorphism of R3 canonically associated to the matrix
 
2 −1 3
A =  0 2 5  ∈ M3 (R).
0 0 2

Since M (fA , C) = A is an upper triangular matrix, then fA is trigonalizable. On the

other hand, fA is not diagonalizable since A is not diagonalizable by the example 2.3.4.

2) Let f be the endomorphism of R3 defined by

f (x1 , x2 , x3 ) = (x1 − 3x2 + 7x3 , x2 − 2x3 , 3x3 ).

As  
1 −3 7
M (f, C) =  0 1 −2 
0 0 3
is an upper triangular matrix, then f is trigonalizable.

Remark 2.4.1 Let f be an endomorphism of E. Then f is trigonalizable if and only if there

exists a basis b of E over K such that M (f, b) is a lower triangular matrix. Indeed, if f is
trigonalizable, then there exists a basis b = {u1 , . . . , un } of E over K such that M (f, b) is
an upper triangular matrix, i.e., M (f, b) is of the form:
 
a11 . . . a1n
M (f, b) =  0
 .. ..  .
. . 
0 0 ann

Consider the basis b0 = {un , . . . , u1 } of E over K obtained from the basis b by reversing
the order of its elements. For every 1 ≤ j ≤ n,

f (uj ) = a1j u1 + · · · + ajj uj = ajj uj + · · · + a1j u1 .

So  
ann 0 0
0  .. ..
M (f, b ) =  .

. 0 
a1n . . . a11
is a lower triangular matrix. The converse is done in a similar way.

c Ali Ayad page 34

Chapter 2 Reduction of endomorphisms and matrices

Remark 2.4.2 Suppose that f is a trigonalizable endomorphism of E, then there exists a

basis b of E over K such that the matrix T = M (f, b) is an upper triangular matrix. Let
λ1 , . . . , λn ∈ K (not necessary distinct) be the diagonal terms of T . Then

Pf (X) = PT (X) = (X − λ1 ) · · · (X − λn ).

Hence Pf (X) splits in K[X] and the terms of the main diagonal of T are the eigenvalues of
A counted with their algebraic multiplicities. We deduce that:
n
X n
Y
tr(f ) = tr(T ) = λi and det(f ) = |T | = λi
i=1 i=1

and rk(f ) = rk(T ) is equal to the number of nonzero scalars among λ1 , . . . , λn .

Theorem 2.4.1 An endomorphism f of E is trigonalizable if and only if its characteristic

polynomial Pf (X) splits in K[X].

Remark 2.4.3 Let f be an endomorphism of E.

1) Suppose that K = C, then Pf (X) ∈ C[X]. By Alembert-Gauss theorem, Pf (X) has

all its roots in C, i.e., Pf (X) splits in C[X]. So f is trigonalizable by the theorem 2.4.1.

2) Trigonalize f returns to determine a basis b of E over K and an upper triangular

matrix T ∈ Mn (K) such that M (f, b) = T .

Definition 2.4.2 We say that a matrix A ∈ Mn (K) is trigonalizable if the endomorphism

fA of Kn canonically associated to A is trigonalizable.

Proposition 2.4.1 Let A ∈ Mn (K). Then A is trigonalizable if and only if A is similar

to an upper triangular matrix.

Proof Let fA be the endomorphism of Kn canonically associated to A, then A = M (fA , C)

where C is the canonical basis of Kn .
N.C. Since A is trigonalizable, then fA is trigonalizable, so there exists a basis b of Kn such
that the matrix T = M (fA , b) is an upper triangular matrix. As A = M (fA , C), then A
is similar to the upper triangular matrix T by the theorem 1.4.1.
S.C. Let T ∈ Mn (K) be an upper triangular matrix such that A is similar to T . By the
proof of the necessary condition of the theorem 1.4.1, there exists a basis B of Kn such
that T = M (fA , B). As T is an upper triangular matrix, then fA is trigonalizable, and
therefore A is trigonalizable. 2

Remark 2.4.4

c Ali Ayad page 35

Chapter 2 Reduction of endomorphisms and matrices

1) By taking into account the remark 2.4.1, the proposition 2.4.1 remains true if we replace
the word ”upper” by ”lower”, i.e., a matrix A ∈ Mn (K) is trigonalizable if and only
if A is similar to a lower triangular matrix. More precisely, if T ∈ Mn (K) is a lower
triangular matrix, then the matrix T 0 = P −1 T P is an upper triangular matrix, where
 
0 0 1
P =  0 . . . 0  ∈ Mn (K), with P −1 = P.
 

1 0 0

2) Let A ∈ Mn (K). Suppose that A is trigonalizable, then A is similar to an upper

triangular matrix T ∈ Mn (K). Let λ1 , . . . , λn ∈ K (not necessary distinct) be the
diagonal terms of T . By using the proposition 2.1.2, we obtain:

PA (X) = PT (X) = (X − λ1 ) · · · (X − λn ).

Hence the terms of the main diagonal of T are the eigenvalues of A counted with their
algebraic multiplicities. We deduce that:
n
X n
Y
tr(A) = tr(T ) = λi and |A| = |T | = λi
i=1 i=1

and rk(A) = rk(T ) is equal to the number of nonzero scalars among λ1 , . . . , λn .

Proposition 2.4.2 Let f be an endomorphism of E and A = M (f, b) where b is a basis

of E over K. Then f is trigonalizable if and only if A is trigonalizable.

Proof
N.C. Since f is trigonalizable, then there exists a basis B of E over K such that the matrix
T = M (f, B) is an upper triangular matrix. As A = M (f, b), then A is similar to the upper
triangular matrix T by the theorem 1.4.1. Hence A is trigonalizable by the proposition 2.4.1.
S.C. Since A is trigonalizable, then, by the proposition 2.4.1, A is similar to an upper
triangular matrix T ∈ Mn (K), so there exists an invertible matrix P ∈ Mn (K) such that
T = P −1 AP . By the remark 1.2.1, there exists a basis b0 of E over K such that P is the
transition matrix from b to b0 . So

M (f, b0 ) = P −1 M (f, b)P = P −1 AP = T.

Therefore M (f, b0 ) is an upper triangular matrix, and therefore f is trigonalizable. 2

Remark 2.4.5

1) Every triangular matrix is trigonalizable by the fact that it is similar to itself.

2) A matrix A ∈ Mn (K) is trigonalizable if and only if its characteristic polynomial

PA (X) splits in K[X].

3) Every matrix A ∈ Mn (C) is trigonalizable.

c Ali Ayad page 36

Chapter 2 Reduction of endomorphisms and matrices

4) The trigonalization of A depends on the field K. For example, if

 
0 1
A= ∈ M2 (K),
−1 0

then PA (X) = X 2 + 1.

• If K = R, then PA (X) does not split in R[X], and therefore A is not trigonalizable
(over R) by the part (3).
• If K = C, then A is trigonalizable (over C) by the part (4).

5) Trigonalize a matrix A ∈ Mn (K) returns to determine an invertible matrix

P ∈ Mn (K) and an upper triangular matrix T ∈ Mn (K) such that P −1 AP = T .

Lemma 2.4.1 Let  

A B
M = ∈ Mn (K),
C D
where

A ∈ Mp (K), B ∈ Mp,q (K), C ∈ Mq,p (K) and D ∈ Mq (K),

with p, q ∈ N∗ such that n = p + q. We suppose that:

i) A is similar to a matrix A0 with A0 = P −1 AP where P ∈ Mp (K) invertible.

ii) D is similar to a matrix D 0 with D 0 = Q−1 DQ where Q ∈ Mq (K) invertible.

Then the matrix M is similar to the matrix

 0
A B0

0
M = with M 0 = S −1 M S,
C 0 D0

where  
P 0
S= ∈ Mn (K), B 0 = P −1 BQ and C 0 = Q−1 CP
0 Q

Example 2.4.2 We will trigonalize the matrix

 
2 −1 −1
A =  2 1 −2  ∈ M3 (R).
3 −1 −2

We have:
X −2 1 1
PA (X) = |XI3 − A| = −2 X −1 2 C1 ← C1 + C3
−3 1 X +2

c Ali Ayad page 37

Chapter 2 Reduction of endomorphisms and matrices

X −1 1 1
= 0 X −1 2 R3 ← R3 − R1
X −1 1 X +2
X −1 1 1
= 0 X −1 2 = (X + 1)(X − 1)2 .
0 0 X +1

So sp(A) = {−1, 1}. As PA (X) splits in R[X], then A is trigonalizable. We have:

E−1 (A) = ker(A + I3 ) and E1 (A) = ker(A − I3 ).

Let x = (x1 , x2 , x3 ) ∈ R3 , then

x ∈ E−1 (A) ⇔ (A + I3 )X = 0
   
3 −1 −1 x1  3x1 − x2 − x3 = 0
⇔  2 2 −2   x2  = 0 ⇔ 2x1 + 2x2 − 2x3 = 0
3 −1 −1 x3 3x1 − x2 − x3 = 0


⇔ x2 = x1 and x3 = 2x1
⇔ x = (x1 , x1 , 2x1 ) = x1 (1, 1, 2)
⇔ x = x1 u1 , where u1 = (1, 1, 2).

So
E−1 (A) = vect(u1 )
and {u1 } is a basis of E−1 (A) over R. In the other side,

x ∈ E1 (A) ⇔ (A − I3 )X = 0
   
1 −1 −1 x1  x1 − x2 − x3 = 0
⇔  2 0 −2   x2  = 0 ⇔ 2x1 − 2x3 = 0
3 −1 −3 x3 3x1 − x2 − 3x3 = 0


⇔ x1 = x3 and x2 = 0
⇔ x = (x3 , 0, x3 ) = x3 (1, 0, 1)
⇔ x = x3 u2 , where u2 = (1, 0, 1).

So
E1 (A) = vect(u2 )
and {u2 } is a basis of E1 (A) over R. Remark that

gm(1) = dim E1 (A) = 1 6= 2 = am(1).

So A is not diagonalizable. Since u1 and u2 are two eigenvectors of A associated to two

distinct eigenvalues of A, then the family {u1 , u2 } is free over R, therefore we can complete
it to a basis of R3 over R. Put u3 = (1, 0, 0) and
 
1 1 1
P =  1 0 0 .
2 1 0

c Ali Ayad page 38

Chapter 2 Reduction of endomorphisms and matrices

As |P | = 1 6= 0, then P is invertible and therefore the family b = {u1 , u2 , u3 } is a basis

of R3 over R, and P is the transition matrix from the canonical basis C = {e1 , e2 , e3 } to
the basis b. In order to calculate the matrix T = P −1 AP , we distinguish the following two
methods:
1st method: As P −1 is the transition matrix from b to C, then
  
0 1 0  e1 = u3
P −1 =  0 −2 1  since e2 = u1 − 2u2 + u3 (∗)
1 1 −1 e3 = u2 − u3


So, by matrix calculations, we obtain the following upper triangular matrix:

 
−1 0 2
T = P −1 AP =  0 1 −1  .
0 0 1

2nd method: Let fA be the endomorphism of R3 canonically associated to A. As A =

M (fA , C), then
T = P −1 AP = M (fA , b).
Since u1 and u2 are two eigenvectors of fA associated to the eigenvalues −1 and 1 of A
respectively, then
fA (u1 ) = −u1 and fA (u2 ) = u2 .
In the other side,
    
2 −1 −1 1 2
fA (u3 ) = Au3 =  2 1 −2   0  =  2 
3 −1 −2 0 3
= 2e1 + 2e2 + 3e3
= 2u1 − u2 + u3 (by using the relations (∗)).

Hence  
−1 0 2
T = M (fA , b) =  0 1 −1 .
0 0 1
 
0 1 1
Example 2.4.3 We will trigonalize the matrix A =  −1 1 1  ∈ M3 (R). We have:
−1 1 2

X −1 −1
PA (X) = |XI3 − A| = 1 X −1 −1 C1 ← C1 + C3
1 −1 X −2
X −1 −1 −1
= 0 X −1 −1 R3 ← R3 − R1
X −1 −1 X −2
X −1 −1 −1
= 0 X −1 −1 = (X − 1)3 .
0 0 X −1

c Ali Ayad page 39

Chapter 2 Reduction of endomorphisms and matrices

So sp(A) = {1}. As PA (X) splits in R[X], then A is trigonalizable. Let E1 (A) =

ker(A − I3 ) and x = (x1 , x2 , x3 ) ∈ R3 , then

x ∈ E1 (A) ⇔ (A − I3 )X = 0
   
−1 1 1 x1  −x1 + x2 + x3 = 0
⇔  −1 0 1   x2  =0 ⇔ −x1 + x3 = 0
−1 1 1 x3 −x1 + x2 + x3 = 0


⇔ x1 = x3 and x2 = 0
⇔ x = (x3 , 0, x3 ) = x3 (1, 0, 1)
⇔ x = x3 u1 , where u1 = (1, 0, 1).

So
E1 (A) = vect(u1 )
and {u1 } is a basis of E1 (A) over R. Remark that

gm(1) = dim E1 (A) = 1 6= 3 = am(1).

So A is not diagonalizable. Since u1 is a nonzero vector, then the family {u1 } is free over
R, we can then complete it to a basis of R3 over R. Put u2 = (0, 1, 0) and u3 = (0, 0, 1)
and  
1 0 0
P =  0 1 0 .
1 0 1
As |P | = 1 6= 0, then P is invertible and therefore the family b = {u1 , u2 , u3 } is a basis
of R3 over R, and P is the transition matrix from the canonical basis C = {e1 , e2 , e3 } to
the basis b. As P −1 is the transition matrix from b to C, then
  
1 0 0  e1 = u1 − u3
P −1 =  0 1 0  since e2 = u2
−1 0 1 e3 = u3


So, by matrix calculations, we obtain the following upper triangular matrix:

 
1 1 1
T = P −1 AP =  0 1 1  .
0 0 1

Remark 2.4.6

1) In the example 2.4.3, the choice of u2 and u3 can sometimes make the calculation longer
and more difficult. For example, we can take u2 = (1, 1, 0) and u3 = (1, 1, 1) and
 
1 1 1
P =  0 1 1 .
1 0 1

c Ali Ayad page 40

Chapter 2 Reduction of endomorphisms and matrices

As |P | = 1 6= 0, then P is invertible and therefore the family b = {u1 , u2 , u3 } is

a basis of R3 over R, and P is the transition matrix from the canonical basis C =
{e1 , e2 , e3 } to the basis b. As P −1 is the transition matrix from b to C, then
  
1 −1 0  e1 = u1 + u2 − u3
P −1 =  1 0 −1  since e2 = −u1 + u3
−1 1 1 e3 = −u2 + u3


So, by matrix calculations, we obtain the following matrix (which is not upper triangu-
lar!):  
1 1 1
M = P −1 AP =  0 1 0  .
0 −1 1
We write M in the form
   
I1 B
1 0
M = , where B = 1 1 , C = 0 and D= .
C D −1 1

We will trigonalize the matrix D. By simple calculations, we find

PD (X) = (X − 1)2 and

E1 (D) = vect(v1 ), where v1 = (0, 1).
 
0 1
Put v2 = (1, 0) and Q = . As |Q| = −1 = 6 0, then Q is invertible and
1 0
therefore the family b0 = {v1 , v2 } is a basis of R2 over R. As
 
−1 0 1
Q = = Q,
1 0

then, put:  
0 −1 1 −1
D =Q DQ = .
0 1
Let  
  1 0 0
I1 0
S= = 0 0 1 
0 Q
0 1 0
By lemma 2.4.1, put:
I1 B 0
 
−1
T =S MS = ,
C 0 D0
where
 
0 1
0
I1−1 BQ and C 0 = Q−1 CI1 = 0.



B = = 1 1 = 1 1
1 0

So  
1 1 1
T =  0 1 −1 
0 0 1

c Ali Ayad page 41

Chapter 2 Reduction of endomorphisms and matrices

is an upper triangular matrix. Put

    
1 1 1 1 0 0 1 1 1
R = PS =  0 1 1  0 0 1  =  0 1 1 
1 0 1 0 1 0 1 1 0

Then

R−1 AR = (P S)−1 A(P S) = S −1 (P −1 AP )S = S −1 M S = T.

2) In the example 2.4.3, if we impose to prove that A is similar to the matrix

 
1 1 0
J = 0 1 1 ,
0 0 1

then, we have to choose u2 and u3 more carefully. Indeed, let fA be the endomorphism
of R3 canonically associated to A. As A = M (fA , C), then we choose u2 and u3 in a
way that the family b = {u1 , u2 , u3 } is a basis of R3 over R and that J = M (fA , b).
So u2 and u3 must satisfy:

fA (u2 ) = u1 + u2 and fA (u3 ) = u2 + u3 ,

i.e.,
(A − I3 )u2 = u1 and (A − I3 )u3 = u2 .

• Put u2 = (x1 , x2 , x3 ). Then

    
−1 1 1 x1 1
(A − I3 )u2 = u1 ⇔  −1 0 1   x2  =  0 
−1 1 1 x3 1

 −x1 + x2 + x3 = 1
⇔ −x1 + x3 = 0
−x1 + x2 + x3 = 1


⇔ x1 = x3 and x2 = 1
⇔ u2 = (x3 , 1, x3 ).

We take a value of x3 for which u1 and u2 are linearly independent over R. For
example, we take x3 = 1 and then u2 = (1, 1, 1).
• Put u3 = (x1 , x2 , x3 ). Then
    
−1 1 1 x1 1
(A − I3 )u3 = u2 ⇔  −1 0 1   x2  =  1 
−1 1 1 x3 1

 −x1 + x2 + x3 = 1
⇔ −x1 + x3 =1
−x1 + x2 + x3 = 1


⇔ x1 = x3 − 1 and x2 = 0
⇔ u3 = (x3 − 1, 0, x3 ).

c Ali Ayad page 42

Chapter 2 Reduction of endomorphisms and matrices

We take a value of x3 for which the family b = {u1 , u2 , u3 } is a basis of R3 over

R. For example, if we take x3 = 0, then u3 = (−1, 0, 0). Let
 
1 1 −1
P =  0 1 0 .
1 1 0
As |P | = 1 6= 0, then P is invertible and therefore b is a basis of R3 over R, and
P is the transition matrix from C to b.
Hence
P −1 AP = M (fA , b) = J.

2.5 Applications
2.5.1 Power of a matrix
Remark 2.5.1
 
λ1 0 0 0
 0 λ2 0 0 
1) If D =   ∈ Mn (K) is a diagonal matrix, then, for any m ∈ N∗ ,
 0 0 · 0 
0 0 0 λn
 m 
λ1 0 0 0
 0 λm 0 0 
Dm =  2 .
 0 0 · 0 
0 0 0 λm n

2) If A ∈ Mn (K) is diagonalizable, then there exist P ∈ Mn (K) invertible and a diagonal

matrix D ∈ Mn (K) such that P −1 AP = D. So A = P DP −1 , and therefore, for
any m ∈ N∗ ,
Am = (P DP −1 )m = P D m P −1 .
The calculation of D m can be done by the part (1).

Example 2.5.1 Let A be the matrix defined in the example 2.1.2. By the example 2.3.5,
P −1 AP = D, where
   
1 0 0 1 2 0 0 0
 1 0 −1 0   0 2 0 0 
P =  1
 and D= .
0 1 0   0 0 4 0 
0 1 0 1 0 0 0 4
As P −1 is the transition matrix from the basis b = {u1 , u2 , u3 , u4 } to the basis C, then

 e1 = −u 2 + u4
 
0 1 1 0 
1
 e2 = 2 (u1 + u2 − u3 − u4 )


1  −2 1 1 2 
P −1 =   since
2  0 −1 1 0 
e = 12 (u1 + u2 + u3 − u4 )

 3

2 −1 −1 0


e4 = u2

c Ali Ayad page 43

Chapter 2 Reduction of endomorphisms and matrices

For any m ∈ N∗ ,

Am = (P DP −1 )m = P D m P −1
  m 
1 0 0 1 2 0 0 0
 1 0 −1 0   0 2m 0 0 
=    P −1
 1 0 1 0  0 0 4 m 0 
0 1 0 1 0 0 0 4m
 m
4m
 
2 0 0 0 1 1 0
1 m 0 −4m 0 
 2   −2 1 1 2 

= m m

2 2 0 4 0   0 −1 1 0 
0 2m 0 4m 2 −1 −1 0
22m+1 2m − 4m 2m − 4m
 
0
1 0 2m + 4m 2m − 4m 0 
= 
m m m m
.
2  0 2 −4 2 +4 0 
22m+1 − 2m+1 2m − 4m 2m − 4m 2m+1

Definition 2.5.1 A matrix N ∈ Mn (K) is said to be nilpotent if there exists an integer

k ∈ N∗ such that N k = 0. The smallest integer k ∈ N∗ such that N k = 0 is called the
index of nilpotence of N .

Example 2.5.2 The matrix

 
0 1 1
N =  0 0 1  ∈ M3 (R)
0 0 0

is nilpotent with index of nilpotence 3 since N 6= 0, N 2 6= 0 and N 3 = 0.

Proposition 2.5.1 Every strictly upper triangular matrix, i.e., of the form:
 
0 a12 · · a1n
 0 0 a23 · a2n 
 
N = · · · · ·  ∈ Mn (K)

 · · · · an−1,n 
0 · · · 0

is nilpotent.

Remark 2.5.2

1) If T ∈ Mn (K) is an upper triangular matrix, then we write:

T = D + N,

c Ali Ayad page 44

Chapter 2 Reduction of endomorphisms and matrices

where D ∈ Mn (K) is the diagonal matrix whose diagonal terms are those of T and N
is the remaining strictly upper triangular matrix (i.e., N = T − D). If DN = N D,
then, for any m ∈ N∗ ,

T m = (D + N )m
= D m + Cm
1
D m−1 N + · · · + Cm
m−1
DN m−1 + N m .

The fact that N is nilpotent (by the proposition 2.5.1) ensures the existence of an integer
k ∈ N∗ such that N k = 0, and therefore N i = 0 for all i ≥ k.

2) If A ∈ Mn (K) is trigonalizable, then there exist P ∈ Mn (K) invertible and an upper

triangular matrix T ∈ Mn (K) such that P −1 AP = T . So A = P T P −1 , and
therefore, for any m ∈ N∗ ,

Am = (P T P −1 )m = P T m P −1 .

The calculation of T m can be done by the part (1).

Example 2.5.3 Let A be the matrix given in the example 2.4.3. We have:
     
1 1 1 1 0 0 1 0 0
T = P −1 AP =  0 1 1  , P =  0 1 0  and P −1 =  0 1 0  .
0 0 1 1 0 1 −1 0 1

The upper triangular matrix T can be written in the form T = D + N , where

 
0 1 1
D = I3 and N =  0 0 1  .
0 0 0

Since DN = N D and N 3 = 0 (by the example 2.5.2), then, for any integer m ≥ 2,

T m = D m + Cm
1
D m−1 N + Cm
2
D m−2 N 2
m(m − 1) 2
= I3 + mN + N
2
1 m m(m+1)
 
2
=  0 1 m .
0 0 1

So  
2−m2 −m m(m+1)
2
m 2
Am = (P T P −1 )m = P T m P −1 =  −m 1 m .
 
m2 +m+2
− m(m+1)
2
m 2

c Ali Ayad page 45

Chapter 2 Reduction of endomorphisms and matrices

2.6 Exercises
Exercise 2.1
Let A, B ∈ Mn (K).

1) Verify that
     
XIn − AB A In 0 In 0 XIn A
= .
0 XIn B In B In 0 XIn − BA

2) Deduce that AB and BA have the same characteristic polynomial.

Exercise 2.2
Let A ∈ Mn (K) be an invertible matrix.

1) Show that 0 is not an eigenvalue of A.

2) a) Show that if λ ∈ sp(A), then λ−1 ∈ sp(A−1 ) and

Eλ (A) = Eλ−1 (A−1 ).

In other words, an element x ∈ Kn is an eigenvector of A associated to λ if and

only if x is an eigenvector of A−1 associated to λ−1 .
b) Deduce that n o
sp(A−1 ) = λ−1 , λ ∈ sp(A) .

3) a) Prove that
(−1)n
 
n 1
PA−1 (X) = X PA .
|A| X
b) Find again the spectrum of A−1 in terms of that of A (part (2)(b)).

4) Show that if A is diagonalizable (resp. trigonalizable), then A−1 is diagonalizable

(resp. trigonalizable).

Exercise 2.3
Let f be an endomorphism of E.

1) Show that if λ ∈ K is an eigenvalue of f , then

(a) aλ is an eigenvalue of af for all a ∈ K∗ .

(b) λm is an eigenvalue of f m for all m ∈ N∗ .

2) We suppose that K = R and f 2 − 2f + idE = 0. Show that if λ ∈ R is an eigenvalue

of f , then λ = 1.

3) We suppose that f is nilpotent, i.e., there exists k ∈ N∗ such that f k = 0.

a) Show that 0 is the only eigenvalue of f .

c Ali Ayad page 46

Chapter 2 Reduction of endomorphisms and matrices

b) Deduce Pf (X) when f is trigonalizable.

c) Show that if f is diagonalizable, then f = 0.

Exercise 2.4 (Continuation of the exercise 1.1)

Let F and G be two supplementary subspaces of E. Let pF be the projection of E on F
parallel to G.

a) We suppose that F and G are nonzeros.

i) Without calculating the characteristic polynomial of pF , find the spectrum of pF .

ii) Determine the eigenspaces of pF .
iii) Determine the characteristic polynomial of pF .
iv) Show that pF is diagonalizable.

b) Show that pF is diagonalizable.

c) Show that tr(pF ) = rk(pF ) = dim F .

Exercise 2.5 (Continuation of the exercise 1.2)

Let F and G be two supplementary subspaces of E. Let sF be the symmetry with respect to
F parallel to G.

a) We suppose that 1K 6= −1K and that F and G are nonzeros.

i) Without calculating the characteristic polynomial of sF , find the spectrum of sF .

ii) Determine the eigenspaces of sF .
iii) Determine the characteristic polynomial of sF .
iv) Show that sF is diagonalizable.

b) Show that sF is diagonalizable.

c) Show, by two different methods, that tr(sF ) = 2 dim F − n.

Exercise 2.6
Let A ∈ Mn (K). The following questions are independent.

1) Give an example which proves that the sum of two eigenvectors of A is not necessary
an eigenvector of A.

2) Show that if A ∈ M2 (R) and tr(A)2 ≥ 4 |A|, then A is trigonalizable.

3) (a) Show that A and tA have the same characteristic polynomial.

(b) Deduce that if A is trigonalizable, then tA is trigonalizable.
(c) Deduce that if A is similar to an upper triangular matrix, then A is similar to a
lower triangular matrix.

c Ali Ayad page 47

Chapter 2 Reduction of endomorphisms and matrices

4) Show that if A is diagonalizable, then tA is diagonalizable.

5) Show that if A is invertible and diagonalizable, then A−1 is diagonalizable.

6) Show that two diagonalizable matrices are similar if and only if they have the same
characteristic polynomial
 
a b
7) Prove that, for all a, b, d ∈ R, the matrix A = is diagonalizable over R.
b d

Exercise 2.7
In each of the following cases, say if the matrix A is diagonalizable or not over K = R or C
and, in the case where it is diagonalizable, find an invertible matrix P and a diagonal matrix
D such that P −1 AP = D.
     
2 1 0 2 0 0 0 1 0
1) A =  0 2 0  2) A =  4 3 0  3) A =  0 0 1 
0 0 2 1 −2 −1 −1 −1 −1
 
  0 1 0 0 0
3 −1 1 −1  0 0 1 0 0 
 0 1 0 0   
4) A =   −4 2 −1 2  5) A =  0 0 0 1 0 .
  
 0 0 0 0 1 
−2 1 −1 2
1 0 0 0 0

Exercise 2.8  
1 1
Let E = M2 (R) and A = . Let f be the endomorphism of E defined by:
0 −1

f (X) = AX + XA, ∀ X ∈ E.

1) Calculate Pf (X).

2) Diagonalize f .

Exercise 2.9
In each of the following cases, say if the matrix A is trigonalizable over R or not and, in the
case where it is trigonalizable, find an invertible matrix P and an upper triangular matrix T
such that P −1 AP = T .
     
7 3 −4 2 0 1 −2 0 0
1) A =  −6 −2 5  2) A =  1 1 0  3) A =  3 1 0 
4 2 −1 −1 1 3 −1 1 3
   
3 1 0 0 1 0 0 0
 −4 −1 0 0   4 −2 −3 −3 
4) A =   7
 5) A =  .
1 2 1   0 4 4 3 
17 −6 −1 0 −2 −1 0 1

c Ali Ayad page 48

Chapter 2 Reduction of endomorphisms and matrices

Exercise 2.10
Let  
4 0 −2
A =  2 5 4  ∈ M3 (R),
4 2 m
where m is a real parameter.

1) Calculate PA (X).

2) For which values of m, the matrix A is trigonalizable ? Justify your answer.

3) For which values of m, the matrix A is diagonalizable ? Justify your answer.

4) Determine, according to the values of m, an invertible matrix P ∈ M3 (R) and a

diagonal matrix D ∈ M3 (R) such that P −1 AP = D.

5) Let fA be the endomorphism of R3 canonically associated to A. Determine, according

to the values of m, a basis b of R3 over R such that M (fA , b) is a diagonal matrix.

Exercise 2.11
Let  
−2 −1 2
A =  −15 −6 11  ∈ M3 (R).
−14 −6 11

1) Calculate PA (X).

2) Determine real numbers a and b and an invertible matrix P ∈ M3 (R) such that
 
1 a 0
P −1 AP =  0 1 b .
0 0 1

c Ali Ayad page 49