S.

Treil : Linear Algebra Done Wrong

Solutions to exercises

4.1 Main definitions

1. True or false:

(a) Every linear operator in an n-dimensional vector space has n distinct eigenvalues;

Solution :
False. By Property 1 , the number of distinct eigenvalues of an n × n matrix, which
is a linear operator, has at most n distinct eigenvalues. Counting multiplicities, the
number of distinct eigenvalues can be less than n.

(b) If a matrix has one eigenvector, it has infinitely many eigenvectors;

Solution :
True. This eigenvector belongs to an eigenspace, which is a vector space closed under
addition and scalar multiplication. This means that it has a basis of eigenvectors that
can generates infinitely many eigenvectors. By Property 2 , if x is an eigenvector, mx
is an eigenvector for any m ̸= 0.

(c) There exists a square real matrix with no real eigenvalues;

Solution :
 
0 −1
True. Let A = with real number entries.
1 0
 
0 − λ −1
We have det(A − λI) = det = 0, and the characteristic polynomial is
1 0−λ
√
λ2 + 1 = 0 ⇒ λ = ± −1 = ±i. So A only has complex eigenvalues.

(d) There exists a square matrix with no (complex) eigenvectors;

Solution :
False. By the fundamental theorem of algebra, every polynomial (of degree at least 1)
has a complex root. Let A be a n × n matrix. For any n, the characteristic polynomial
will be of degree at least 1. Since R ⊂ C, this implies that any n × n matrix has at least
one complex eigenvalue with an eigenspace containing at most n complex eigenvectors.

(e) Similar matrices always have the same eigenvalues;

Solution :
True. By Property 3 , they have the same determinant and, consequently, the same
characteristic polynomial whose roots are the eigenvalues.

(f) Similar matrices always have the same eigenvectors;

1
 Solution :
False. They have the same eigenvalues and, consequently, the same number of inde-
pendent eigenvectors. However, the basis for these eigenvectors may differ.
Let A and B be similar such that A = SBS −1 . Thus, we have Ax = λx, but

SBS −1 x = λx
S −1 SBS −1 x = S −1 λx
B(S −1 x) = λ(S −1 x)

Clearly, x and (S −1 x) may differ for the same λ.

(g) A non-zero sum of two eigenvectors of a matrix A is always an eigenvector;

Solution :
False. Since A may have more than one eigenvalue, A may have different eigenspaces
with their own basis of eigenvectors.

(h) A non-zero sum of two eigenvectors of a matrix A corresponding to the same eigenvalue
λ is always an eigenvector.

Solution :
True. For each eigenvalue, there is an eigenspace, which is a vector space closed under
addition. Thus, this eigenspace has a basis of eigenvectors and every vector that is in
the span of this basis is an eigenvector.

2. Find characteristic polynomials, eigenvalues and eigenvectors of the following matrices:

 
    1 3 3
4 −5 2 1 −3 −5 −3
, ,
2 −3 −1 4
3 3 1
Solution :
 
4 −5
Let A = . We have:
2 −3

4−λ −5
det(A − λI) =
2 −3 − λ
= (4 − λ)(−3 − λ) − (2)(−5)
= −12 − 4λ + 3λ + λ2 + 10
= λ2 − λ − 2
= (λ − 2)(λ + 1) (The characteristic polynomial of A)
Thus, det(A − λI) = 0 for λ1 = 2 and λ2 = −1, which are the eigenvalues of A.
Now, we look for the eigenvectors of A.
        
4−2 −5 2 −5 2 −5 x1 0
For λ = 2, we have: = and =
2 −3 − 2 2 −5 2 −5 x2 0

2
This leads to the system:
2x1 − 5x2 = 0 , and the eigenvector is (5, 2)T , which means E2 = span({(5, 2)T }).
        
4 − (−1) −5 5 −5 5 −5 x1 0
For λ = −1, we have: = and =
2 −3 − (−1) 2 −2 2 −2 x2 0
This leads to the system:
5x1 − 5x2 = 0
, and the eigenvector is (1, 1)T , which means E−1 = span({(1, 1)T }).
2x1 − 2x2 = 0

 
2 1
Let B = . We have:
−1 4
2−λ 1
det(B − λI) =
−1 4 − λ
= (2 − λ)(4 − λ) − (−1)(1)
= 8 − 2λ − 4λ + λ2 + 1
= λ2 − 6λ + 9
= (λ − 3)2 (The characteristic polynomial of B)
Thus, det(B − λI) = 0 for λ = 3 with multiplicity 2, which is the eigenvalue of B.
Now, we look for the eigenvectors of B.
        
2−3 1 −1 1 −1 1 x1 0
For λ = 2, we have: = and =
−1 4 − 3 −1 1 −1 1 x2 0
This leads to the system:
−x1 + x2 = 0 , and the eigenvector of is (1, 1)T , which means E3 = span({(1, 1)T }).
 
1 3 3
Let C = −3 −5 −3. Expanding the determinant along the first row, we have:
3 3 1
1−λ 3 3
det(C − λI) = −3 −5 − λ −3
3 3 1−λ
−5 − λ −3
= (−1)1+1 (1 − λ)
3 1−λ
−3 −3 −3 −5 − λ
+ (−1)1+2 (3) + (−1)1+3 (3)
3 1−λ 3 3
= (1 − λ)(−5 − λ)(1 − λ) + 9 − 3[(−3)(1 − λ) + 9] + (3)[(−9) − 3(−5 − λ)]
= (1 − λ)[(−5 − λ)(1 − λ) + 9] − 3[3λ + 6] + 3(6 + 3λ)
= (1 − λ)[(−5 − λ)(1 − λ) + 9]
= (1 − λ)[λ2 + 4λ + 4]
= (1 − λ)(2 + λ)2

3
 Thus, det(C − λI) = 0 for λ1 = 1 and λ2 = −2 with multiplicity 2. Now, we look for the
eigenvectors of C.
   
1−1 3 3 0 3 3
For λ1 = 1, we have  −3 −5 − 1 −3  = −3 −6 −3
3 3 1−1 3 3 0
    
0 3 3 x1 0
and −3 −6 −3 x2  = 0
3 3 0 x3 0

This leads to the system:

3x2 + 3x3 = 0
−3x1 − 6x2 − 3x3 = 0
3x1 + 3x2 =0
We have x1 = −x2 and x2 = −x3 , so the eigenvector is (1, −1, 1)T , which means E1 =
span({(1, −1, 1)T })..
   
1 − (−2) 3 3 3 3 3
For λ2 = −2, we have  −3 −5 − (−2) −3  = −3 −3 −3
3 3 1 − (−2) 3 3 3
    
3 3 3 x1 0
and −3 −3 −3 x2  = 0
3 3 3 x3 0

This leads to the system:

3x1 + 3x2 + 3x3 = 0

−3x1 − 3x2 − 3x3 = 0
3x1 + 3x2 + 3x3 = 0
We have x1 = −x2 and x2 = −x3 , so the eigenvector is (1, −1, 1)T .

We have x1 = −x2 − x3 , so the eigenvectors are (−1, 1, 0)T and (−1, 0, 1)T , which means
E−2 = span({(−1, 1, 0)T , (−1, 0, 1)}).

3. Compute eigenvalues and eigenvectors of the rotation matrix

 
cos α − sin α
.
sin α cos α

Note, that the eigenvalues (and eigenvectors) do not need to be real.

4. Compute characteristic polynomials and eigenvalues of the following matrices:

     
1 2 5 67 2 1 0 2 4 0 0 0
0 2 3 6 0 π 43 2  1 3 0 0
, ,
2 4 e 0,
   
0 0 −2 5  0 0 16 1 
0 0 0 3 0 0 0 54 3 3 1 1

4
  
4 0 0 0
1 0 0 0
 .
2 4 0 0
3 3 1 1
Do not expand the characteristic polynomials, leave them as products.

Solution :
   
1 2 5 67 1−λ 2 5 67
0 2 3 6 and (A − λI) =  0
 2−λ 3 6 
Let A =  
0 0 −2 5   0 0 −2 − λ 5 
0 0 0 3 0 0 0 3−λ

Since A is an upper triangular matrix, det(A − λI) = (1 − λ)(2 − λ)(−2 − λ)(3 − λ), which
corresponds to the characteristic polynomial, and we can take the eigenvalues directly from
the diagonal entries of A : λ1 = 1, λ2 = 2, λ3 = −2 and λ4 = 3.
   
2 1 0 2 2−λ 1 0 2
0
 and (A − λI) =  0
π 43 2   π−λ 43 2 
Let B =  
0 0 16 1   0 0 16 − λ 1 
0 0 0 54 0 0 0 54 − λ

Since B is an upper triangular matrix, det(B − λI) = (2 − λ)(π − λ)(16 − λ)(54 − λ), which
corresponds to the characteristic polynomial, and we can take the eigenvalues directly from
the diagonal entries of B : λ1 = 2, λ2 = π, λ3 = 16 and λ4 = 54.
   
4 0 0 0 4−λ 0 0 0
1 3 0  and (A − λI) =  1
0  3−λ 0 0 
Let C =  
2 4 e 0  2 4 e−λ 0 
3 3 1 1 3 3 1 1−λ

Since C is a lower triangular matrix, det(C − λI) = (4 − λ)(3 − λ)(e − λ)(1 − λ), which
corresponds to the characteristic polynomial, and we can take the eigenvalues directly from
the diagonal entries of C : λ1 = 4, λ2 = 3, λ3 = e and λ4 = 1.
   
4 0 0 0 4−λ 0 0 0
1 0 0 0
 and (A − λI) =  1
 0−λ 0 0 
Let D =  
2 4 0 0  2 4 0−λ 0 
3 3 1 1 3 3 1 1−λ

Since D is a lower triangular matrix, det(D−λI) = (4−λ)(0−λ)(0−λ)(1−λ) = (4−λ)λ2 (1−

λ, which corresponds to the characteristic polynomial, and we can take the eigenvalues
directly from the diagonal entries of C : λ1 = 4, λ2 = 0 (with multiplicity 2), and λ3 = 1.

5.

6.

7.

5
 8.

9.

10. Prove that determinant of a matrix A is the product of its eigenvalues (counting multiplicities).
Hint: first show that det(A − λI) = (λ1 − λ)(λ2 − λ)...(λn − λ), where λ1 , λ2 , ..., λn are
eigenvalues (counting multiplicities). Then compare the free terms (terms without λ) or plug
in λ = 0 to get the conclusion.

Solution :
For a matrix An×n , the determinant det(A − λI) is the characteristic polynomial of degree
n of the variable λ. By the fundamental theorem of algebra, any polynomial (of degree at
least 1) has a complex root, and any polynomial of degree n has exactly n roots. Thus, by
Property 1 , the characteristic polynomial of A has n roots that correspond to n complex
eigenvalues λ1 , λ2 , ..., λn , counting multiplicities. This means that (λi −λ)m with the highest
multiplicity m divides the polynomial, so det(A − λI) can be factorized as det(A − λI) =
(λ1 − λ)(λ2 − λ)(λ3 − λ)...(λn − λ).

Indeed, let a11 , a22 , ..., ann be the n terms of the diagonal of A. By using the cofactor
expansion of the determinant along the first row, where det(C̃ij ) is the determinant of the
matrix obtained by removing row i and j from (A − λI), we get:

det(A − λI) = (−1)1+1 (a11 − λ) · det(C̃11 ) + (−1)1+2 a12 · det(C̃12 )+

... + (−1)1+n a1n · det(C̃1n )
= (a11 − λ) · det(C̃11 ) + q1 (λ)

where q1 (λ) is a polynomial containing the sum of the remaining off-diagonal terms in the
cofactor expansion along row 1. Now, expanding the determinant of C̃11 from the first
diagonal entry of the second row, we get:

det(A − λI) = (a11 − λ)(a22 − λ) · det(C̃22 ) + q1 (λ) + q2 (λ)

Repeating the process recursively, we obtain:

det(A − λI) = (a11 − λ)(a22 − λ)...(ann − λ) + q1 (λ) + q2 (λ) + .... + qn (λ)

Finally, by expanding the product completely and grouping similar terms, we obtain a
characteristic polynomial of degree n that can be factorized such that: det(A − λI) =
(λ1 − λ)(λ2 − λ)(λ3 − λ)...(λn − λ) with all λi ∈ C.

With λ = 0, det(A − λI) = det(A) = λ1 λ2 λ3 ...λn , which is the product of the complex
eigenvalues of A, counting multiplicities.

11. Prove that the trace of a matrix equals the sum of eigenvalues in three steps. First, compute
the coefficient of λn−1 in the right side of the equality

det(A − λI) = (λ1 − λ)(λ2 − λ)...(λn − λ).

6
Then show that det(A − λI) can be represented as

det(A − λI) = (a1,1 − λ)(λ2,2 − λ)...(λn,n − λ) + q(λ).

where q(λ) is polynomial of degree at most n − 2. And finally, comparing the coefficients of
λn−1 get the conclusion.

Solution :
First, consider the binomial expansion of (λ1 − λ)...(λn − λ) = (−1)n λn + (−1)n−1 (λ1 + ... +
λn )λn−1 + p(λ) where p is a polynomial of degree at most n − 2.
Now, let a11 , a22 , ..., ann be the n terms of the diagonal of A. By using the cofactor expansion
of the determinant along the first row, where det(C̃ij ) is the determinant of the matrix
obtained by removing row i and j from (A − λI), we get:
det(A − λI) = (−1)1+1 (a11 − λ) · det(C̃11 ) + (−1)1+2 a12 · det(C̃12 )+
... + (−1)1+n a1n · det(C̃1n )
= (a11 − λ) · det(C̃11 ) + q1 (λ)

where q1 (λ) is a polynomial containing the sum of the remaining off-diagonal terms in the
cofactor expansion along row 1. Now, expanding the determinant of C̃11 from the first
diagonal entry of the second row, we get:
det(A − λI) = (a11 − λ)(a22 − λ) · det(C̃22 ) + q1 (λ) + q2 (λ)
Repeating the process recursively, we obtain:
det(A − λI) = (a11 − λ)...(ann − λ) + q1 (λ) + q2 (λ) + .... + qn (λ)
Xn
= (a11 − λ)...(ann − λ) + qi (λ)
i
= (a11 − λ)...(ann − λ) + q(λ)
If we pick up any (aij −λ) term in ni qi (λ) = q(λ) from the off-diagonal, there is no (aii −λ)
P
and (ajj − λ) terms, which only appears on the diagonal. If we pick up the λn−1 term from
the diagonal, qn (λ) is a polynomial of degree at most n − 2. Thus, we will only consider the
term λn−1 from the (a11 − λ)(a22 − λ)...(ann − λ) part of the determinant.
From the last exercise, we know that det(A − λI) = (λ1 − λ)...(λn − λ). Thus:
(λ1 − λ)...(λn − λ) = (a11 − λ)...(ann − λ) + q(λ)
n n n−1
(−1) λ + (−1) (λ1 + ... + λn )λn−1 + p(λ) = (a11 − λ)...(ann − λ) + q(λ)
= (−1)n λn
+ (−1)n−1 (a11 + ... + ann )λn−1
+ r(λ)
(with r(λ) of degree at most n − 1)

Now, considering the coefficient of λn−1 on both side, we have :

(−1)n−1 (λ1 + ... + λn )λn−1 = (−1)n−1 (a11 + ... + ann )λn−1

Thus, λ1 + ... + λn = a11 + ... + ann = trace (A).