MAA 201 2022-2023

Reduction of endomorphisms
Exercises - Week VIII - Diagonisable matrices and characteristic
polynomial
In the following, E will denote a finite dimensional vector space over a field K.
We let L(E) be the set of endomorphisms of E. When λ is an eigenvalue of u ∈
L(E) we let Eλ be the associated eigenspace Eλ = Eλ (u) = {x ∈ E, u(x) = λx}.

Exercise 1. √ 
2 √2 1
Show that the matrix  0 2 √3  is not diagonalisable in M3 (C).
0 0 2
√
Solution: The characteristic polynomial is √(X − 2)3 , therefore 2 is the unique eigenvalue. If
the matrix was diagonalisable, it should be 2Id.

Exercise 2.
Let E be a finite-dimensional K-vector space and f ∈ L(E) an endomorphism of
E. Let χf denote the characteristic polynomial of f . For each eigenvalue λ of f ,
prove the inequality
dim Eλ 6 order of λ as a root of χf .
Give an example of an endomorphism for which the above inequality is strict.
Solution: Set n = dimK E and let nλ = dimK Eλ be the dimension of the eigenspace Eλ . By
choosing a basis (ei )i=1,··· ,nλ of Eλ and by completing it into a basis (ei )i=1,··· ,n of E, one has a
direct sum decomposition E = Eλ ⊕ F , where F is the subspace of E spanned by enλ +1 , · · · , en .
With respect to this basis, the matrix associated with f can be written as
 
λInλ A2
A=
0 A1
with A1 ∈ Mat(n−nλ )×(n−nλ ) (K) and A2 ∈ Matnλ ×(n−nλ ) (K). Hence its characteristic polyno-
mial is
χf = χA = (λ − X)nλ · χA1 ,
which implies that nλ 6 order of the root λ in χf . Note that there is equality if and only if λ is
not an eigenvalue of A1 .
Set E = K 2 = Ke1 ⊕ Ke2 . The endomorphism f ∈ L(E) defined by
f (e1 ) = e1 f (e2 ) = e1 + e2 .
gives the required example. Since the associated matrix is ( 10 11 ), we have χf = (X − 1)2 . Hence
1 is an eigenvalue of f , but we have dim Ker(f − IdE ) < 2. Indeed, if dim Ker(f − IdE ) = 2, then
we would have Ker(f − IdE ) = E, i.e. we would have f = IdE .
Exercise 3.
Set E = R3 = Re1 ⊕ Re2 ⊕ Re3 and consider the endomorphisms f ∈ L(E) given
by the following matrices with respect to the basis {e1 , e2 , e3 } :
     
0 2 −1 0 3 2 1 0 0
A=  3 −2 0 ,  B = −2 5 2 ,
  C= 0  1 0 .
−2 2 1 2 −3 0 1 −1 2
For each of them, compute the characteristic polynomial, determine the eigenvalues
and the eigenspaces, and decide whether they are diagonalisable or not. If they are,
find a basis with respect to which the endomorphism is represented by a diagonal
matrix.
Solution: Matrix A. The characteristic polynomial of A is
−X 2 −1
χA = det(A − X · I3 ) = 3 −2 − X 0
−2 2 1−X
3 −2 − X −X 2
= (−1)1+3 (−1) + (−1)3+3 (1 − X)
expand along third column −2 2 3 −2 − X
= −(2 − 2X) + (1 − X)(X 2 + 2X − 6) = (1 − X)(X 2 + 2X − 8)
= (1 − X)(2 − X)(4 + X).
Since χA is a product of three linear factors, the matrix A is diagonalisable with eigenvalues 1,2
and −4. Thus we have to find the eigenvectors for these eigenvalues.
Let us calculate E1 : for v = t(x, y, z) ∈ R3 the condition Av = v gives the linear system

 −x + 2y − z = 0
3x − 3y = 0 ⇔ x = y = z
−2x + 2y = 0


Hence E1 = R · t(1, 1, 1). Similarly one finds that E2 = R · t(4, 3, −2) and E−4 = R · t(2, −3, 2).
In conclusion we find that A = P DP −1 with
   
1 0 0 1 4 2
D = 0 2 0  , P = 1 3 −3 .
0 0 −4 1 −2 2
Matrix B :
The characteristic polynomial of B is
−X 3 2 −X 3 2
χB = det(B − X · I3 ) = −2 5 − X 2 = 0 2−X 2−X
r3 +r2 →r2
2 −3 −X 2 −3 −X
2 − X 2 − X 3 2
= (−1)1+1 (−X) + (−1)3+1 · 2 ·
expand along first column −3 −X 2−X 2−X
= −X(2 − X)(3 − X) + 2(2 − X) = (2 − X)(X 2 − 3X + 2)
= (1 − X)(2 − X)2 .
The eigenvalues of B are thus 1 and 2. Let us calculate E1 : for v = t(x, y, z) ∈ R3 , the condition
Bv = v gives the following linear system

 −x + 3y + 2z = 0 
x+z = 0
−2x + 4y + 2z = 0 ⇔
y+z = 0
2x − 3y − z = 0


Page 2
Hence E1 = R · t(1, 1, −1). Let us calculate now the eigenspace E2 which is more complicated
since 2 is a double root of the characteristic polynomial. We need to show that E2 is of dimension
2. Vectors v = t(x, y, z) ∈ E2 satisfies the following linear system given by Bv = 2v :

 −2x + 3y + 2z = 0
−2x + 3y + 2z = 0 ⇔ 2x = 3y + 2z
2x − 3y − 2z = 0


Hence E2 = R · t(3, 2, 0) ⊕ R · t(1, 0, 1). Therefore B is indeed diagonalisable.

Thus we find B = P DP −1 with
   
1 0 0 1 3 1
D = 0 2 0 , P =  1 2 0 .
0 0 2 −1 0 1

Matrix C :
The characteristic polynomial of C is

1−X 0 0
χC = det(C − X · I3 ) = 0 1−X 0 = (1 − X)2 (2 − X).
expand along first rowr1
1 −1 2−X

Thus the situation is the same as in the previous question. By the same method one finds that :
Cv = v ⇔ x + z = 0, hence E1 = R · t(1, 1, 0) ⊕ R · t(0, 1, 1) ; and Cv = 2v ⇔ x = y = 0, hence
E2 = t(0, 0, 1). Thus C = P DP −1 with
   
1 0 0 1 0 0
D = 0 1 0 , P = 1 1 0 .
0 0 2 0 1 1

Exercise 4.
Let  
0 1
A= ∈ M2 (C).
−2 3
1. Diagonalise A and calculate An for all integers n > 0.
2. Let (un )n∈N be the sequence of complex numbers defined by the recursive
formula
un+2 = 3un+1 − 2un
with initial terms u0 = 4 and u1 = 7. Using the first question, compute the
general term un .

Solution:
1. Let us determine the diagonalisability and calculate the diagonalisation of A :

Page 3
 — Step 1 : Calculate the characteristic polynomial of A. We have

−X 1
χA = det(A − X · I2 ) = = X(X − 3) + 2 = (1 − X)(2 − X).
−2 3−X

Hence A has two eigenvalues 1 and 2, thus A is diagonalisable.

— Step 2 : Calculate the eigenspaces of A.
— The eigenspace for the eigenvalue 1 : for v = t(x, y) ∈ C2 the condition Av = v
gives : 
−x + y = 0
⇔ x = y.
−2x + 2y = 0
Hence E1 = C · t(1, 1).
— The eigenspace for the eigenvalue 2 : for v = t(x, y) ∈ C2 the condition Av = 2v
gives : 
−2x + y = 0
⇔ y = 2x.
−2x + y = 0
Hence E2 = C · t(1, 2).
— Step 3 : By the calculation above, we find that A = P DP −1 with
   
1 0 1 1
D= and P = .
0 2 1 2

In other word, the endomorphism of C2 represented by the matrix A under the stan-
dard basis is diagonalised under the basis ( t(1, 1), t(1, 2)).
In order to calculate An , let us first calculate P −1 : since detP = 1 we have
 
2 −1
P −1 = .
−1 1

Then we have :
      
1 1 1 0 2 −1 1 1 2 −1
An = (P DP −1 )n = P Dn P −1 = =
1 2 0 2n −1 1 1 2 −2n 2n
2 − 2n 2n − 1
 
= .
2 − 2n+1 2n+1 − 1

2. Set  
un
Un = ,
un+1
then we have Un+1 = AUn . By the initial condition we have
 
4
U0 = ,
7

hence by (a) we get

2 − 2n 2n − 1 1 + 3 · 2n
    
n 4
Un = A U0 = = ,
2 − 2n+1 2n+1 − 1 7 1 + 3 · 2n+1

which implies that un = 1 + 3 · 2n .

Page 4
Exercise 5.
We recall the rank of u ∈ L(E) is the dimension of the image u(E) of u. Prove
that an endomorphism of rank 1 is diagonalisable if and only if its trace is not
zero.

Solution: We work equivalently with matrices. A matrix A of rank 1 is of the form xy with x
and y being respectively a column and a line vector. Then (xy)2 = x(yx)y. But yx is the trace
of A so that A2 − tr(A)A = 0. We know that A is diagonalisable iff its minimal polynomial split
over K with simple roots. Therefore in our setting, A is diagonalisable iff tr(A) = 0.

Exercise 6.
Let u ∈ L(E) diagonalisable.
1. Show that v ∈ L(E) satisfies u ◦ v = v ◦ u if and only if v(Eλ (u)) ⊂ Eλ (u)
for any eigenvalue λ of u.
2. Let v ∈ L(E) diagonalisable with u ◦ v = v ◦ u. Show that there is a basis B
of E such that MatB (u) and MatB (v) are both diagonal.
3. Let C(u) = {v ∈ L(E), u ◦ v = v ◦ u}. Show that C(u) is a vector space over
K. Compute its dimension.

Exercise 7.
Let u ∈ L(E). We consider the endomorphism Φ : L(E) → L(E), v 7→ u ◦ v.
1. Show that any eigenvalue of u is an eigenvalue of φ.
2. Let λ be an eigenvalue of u. Check that
[v ∈ Eλ (φ)] ⇐⇒ [v(E) ⊂ Eλ (u)].
3. Assume u is diagonalisable. Show that φ is diagonalisable.

Solution:
1. Fix x 6= 0 and λ ∈ K with u(x) = λx. Complete x to a basis B of E and consider
v = vx ∈ L(E) with v(x) = x and v(y) = 0 for x 6= y ∈ B. Then Φ(vx ) = u ◦ vx = λvx .
The converse also holds true : any eigenvalue of Φ is an eigenvalue of u.
2. v ∈ Eλ (Φ) ⇐⇒ u ◦ v = λv ⇐⇒ v(E) ⊂ Eλ (u).
Pu is diagonalisable, then E = ⊕λ Eλ (u). Then any v ∈ L(E) may be written
3. If Pas v =
λ pλ ◦ v with p λ being the "projecteur" defined as p λ (x) = x λ when x = λ xλ ∈
⊕λ Eλ (u). Then vλ := u ◦ pλ lies in L(E) and satisfies vλ (E) ⊂ Eλ (u), thus belongs to
Eλ (Φ). Therefore L(E) = Vect{Eλ (Φ), λ}, but the eigenspaces form always a direct sum.
Finally L(E) = ⊕λ Eλ (Φ) and φ is diagonalisable.
Note that the converse is also true. If Φ is diagonalisable, there is a basis of Φ-eigenvectors
of L(E). Then, by item 2, any v in L(E) satisfies v(E) ⊂ ⊕λ Eλ (u). We conclude by taking
v = IdE .

Page 5