Bocconi University

APPLIED MATHEMATICS
Exercises
Mauro D’Amico
Jacopo De Tullio
Dovid Fein
Guido Osimo
Fabio Tonoli

BIEM - BIEF
Academic Year 2023/2024
Version: March 13, 2024
Contents

1 Eigenvalues and eigenvectors 5

1.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 Quadratic forms 15
2.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3 Unconstrained optimization 21
3.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

4 Implicit functions and constrained optimization 27

4.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.2 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

5 The Riemann integral 31

5.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.2 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

6 Integral calculus 39
6.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
6.2 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

7 Improper integrals 47
7.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
7.2 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

8 The Stieltjes integral 53

8.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
8.2 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

9 Probability measures 59
9.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
9.2 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

10 Random variables 65
10.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
10.2 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

3
4 CONTENTS

11 Distribution and density functions 73

11.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
11.2 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

12 Elementary financial calculus 81

12.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
12.2 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

13 Asset pricing 87
13.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
13.2 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
Chapter 1

Eigenvalues and eigenvectors

1.1 Exercises
1.1. Consider the square matrix  
1 −1
A=
−1 1
and the pairs (λi , xi ), where i = 1, 2,
     
t −s
(λ1 , x1 ) = 0, (λ2 , x2 ) = 2,
t s

where t, s are non-zero real numbers. Using the definition, verify that (λi , xi ), where i = 1, 2,
are the eigenpairs associated with A.
1.2. Let (λ, x), with λ ∈ R and with non-zero x ∈ Rn be an eigenpair of a symmetric matrix A
of order n. Prove that ∀c ∈ R, c ̸= 0, (λ, cx) is an eigenpair of A.
1.3. Let A be the square matrix of order 2 where
     
1 2 1 2 1 1
(1) A = (2) A = (3) A =
2 1 2 −2 1 −1
(a) Find the spectrum σ(A) of A.
(b) Find the eigenvectors associated with the eigenvalues of A.
(c) Verify that the eigenvectors associated with distinct eigenvalues are orthogonal.
1.4. Let a, b, c ∈ R and A be the square matrix of order 2 where
 
a c
A=
c b
(a) Verify that for every a, b, c ∈ R, A has two eigenvalues λ1 , λ2 ∈ R.
(b) Suppose that det(A) = 0. Verify that A has at least one zero eigenvalue and determine
when the multiplicity of the zero eigenvalue is 2.
(c) Now suppose that det(A) = ab. Verify that
λ1 = min{a, b} ≤ max{a, b} = λ2
and determine when λ1 = λ2 .

5
6 CHAPTER 1. EIGENVALUES AND EIGENVECTORS

1.5. Let A = aIn and B = bIn with a, b ∈ R and In the identity matrix of order n.

(a) Verify that λ = 1 is an eigenvalue of In .

(b) Verify that A admits the eigenvalue µ = a.
(c) Verify that B 2 = BB admits the eigenvalue ν = b2 .
(d) Verify that AB 2 = ABB admits the eigenvalue ξ = ab2 .

1.6. Let A be the square matrix of order 2 where

     
1 3 2 −1 −2 1
(1) A = (2) A = (3) A =
3 1 −1 0 1 2

(a) Determine the eigenvalues and eigenvectors of A.

(b) Show that tr(A) = 2s=1 λs and det(A) = 2s=1 λs .
P Q

(c) Find the eigenspaces associated with the eigenvalues of A.

1.7. Let A be the square matrix of order 3 where

     √ 
2 1 0 1 1 0 √ 1 2 0
(1) A = 1 2 0 (2) A = 1 1 0 (3) A =  2 0 0
0 0 2 0 0 1 0 0 1

(a) Determine the eigenvalues and eigenvectors of A.

(b) Show that tr(A) = 3s=1 λs and det(A) = 3s=1 λs .
P Q

(c) Find the eigenspaces associated with the eigenvalues of A.

1.8. Let A be the square matrix of order 3 where

     
1 0 1 −1 0 1 2 −1 1
(1) A = 0 1 1 (2) A =  0 1 0 (3) A = −1 1 0
1 1 0 1 0 1 1 0 1

(a) Determine the characteristic polynomial pA (t) associated with A.

(b) Calculate det(A) and verify that pA (0) = (−1)3 det(A).
(c) Determine the eigenvalues of A.

1.9. Let A be a symmetric matrix of order n and let λ be a real number.

(a) Prove that if λ is an eigenvalue of A, then λ2 is an eigenvalue of A2 .

√
(b) Show that if µ is an eigenvalue of A2 then µ is not necessarily an eigenvalue of A.

1.10. Let A be a square matrix of order 2 where

 √ 
1 2
A= √
2 0

(a) Determine the eigenvalues and eigenvectors of A.

(b) Determine the orthonormal matrix B of the eigenvectors of A.
(c) Verify that B T AB = Λ, where Λ is the matrix of eigenvalues of A.
1.2. SOLUTIONS 7

1.11. Let A be the square matrix of order 3 where

 
1 1 0
A = 1 1 0
0 0 2

(a) Determine the eigenvalues and eigenvectors of A.

(b) Determine the orthonormal matrix B of the eigenvectors of A.
(c) Verify that B T AB = Λ, where Λ is the matrix of eigenvalues of A.

1.12. Let A be the square matrix of order 3 where

 √ 
2 0 2
A = √0 2
 0
2 0 3

(a) Determine the eigenvalues and eigenvectors of A.

(b) Determine the orthonormal matrix B of the eigenvectors of A.
(c) Verify that B T AB = Λ, where Λ is the matrix of eigenvalues of A.

1.2 Solutions
1.1. For the first pair, (λ1 , x1 ) we have
      
1 1 −1 t t−t 0 t
Ax = = = =0 = λ1 x1
−1 1 t −t + t 0 t

Thus, (λ1 , x1 ) is an eigenpair associated with matrix A. In a similar fashion, it can be

verified that the second pair, (λ2 , x2 ) is also an eigenpair associated with matrix A.

1.2. We assume, by hypothesis, that (λ, x) is an eigenpair of A, implying that Ax = λx. Letting
c be a non-zero real number, we obtain

A(cx) = c(Ax) = c(λx) = λ(cx)

and see that (λ, cx) is also an eigenpair of A.

1.3. (1) (a) Let λ ∈ R. The determinant of the matrix A − λI2 (I2 = the identity matrix of
order two) is
1−λ 2
det(A − λI2 ) = = (1 − λ)2 − 4
2 1−λ
From the equation (1 − λ)2 − 4 = 0 we find the eigenvalues of A, which are the distinct
values λ1 = −1 and λ2 = 3. The spectrum of A is therefore σ(A) = {−1, 2}.
(b) The eigenvectors associated with the eigenvalue λ = −1, denoted by x1 , are the (non-
zero) solutions to the linear system (A − (−1)I2 )x1 = 0. [For simplicity, we denote the
components of the vector x1 with x1 , x2 instead of x11 , x12 .] We have:
( (  
2x1 + 2x2 = 0 x1 = −t 1 −t
⇒ ⇒ x = , t ∈ R − {0}
2x1 + 2x2 = 0 x2 = t t
8 CHAPTER 1. EIGENVALUES AND EIGENVECTORS

The eigenvectors associated with the eigenvalue λ = 3, denoted by x2 , are the (non-zero)
solutions to the linear system (A − 3I2 )x2 = 0. [Again, for simplicity we denote the
components of the vector x2 with x1 , x2 instead of x21 , x22 .] We have:
( (  
−2x1 + 2x2 = 0 x1 = s 2 s
⇒ ⇒ x = , s ∈ R − {0}
2x1 − 2x2 = 0 x2 = s s

(c) We must show that for every non-zero t, s ∈ R, we have x1 · x2 = 0. Given that
2
X
∀t, s ∈ R − {0}, x1 · x2 = x1s x2s = (−t) · s + t · s = −ts + ts = 0
s=1

we conclude that the eigenvectors associated with distinct eigenvalues, x1 and x2 , are
orthogonal.
(2) (a) σ(A) = {−3, 2}
(b) We obtain
     
1 −t/2 2 2s
(λ1 , x ) = −3, e (λ2 , x ) = 2,
t s
with non-zero t, s ∈ R.
(c) Verify that ∀t, s ∈ R non-zero, x1 · x2 = 0.
√ √
(3) (a) σ(A) = {− 2, 2}
(b) We obtain
√  √ 
√ (1 − 2)t √ (1 + 2)s
   
(λ1 , x1 ) = − 2, e (λ2 , x2 ) = 2,
t s
with non-zero t, s ∈ R
(c) Verify that ∀t, s ∈ R non-zero, x1 · x2 = 0.
1.4. (a) Let λ ∈ R. The determinant of A − λI2 is
 
a−λ c
det = (a − λ)(b − λ) − c2 = λ2 − (a + b)λ + ab − c2 .
c b−λ
which is a second degree polynomial in λ with discriminant ∆ given by
∆ = (a + b)2 − 4(ab − c2 ) = a2 + b2 + 2ab − 4ab + 4c2 = (a − b)2 + 4c2
It is ∆ ≥ 0 for every choice of a, b, c ∈ R. Therefore A has two eigenvalues λ1 , λ2 ∈ R
(equal or distinct) for every choice of a, b, c ∈ R.
(b) When det(A) = 0, we have that ab − c2 = 0 and the determinant of A − λI2 simplifies
to the equation
det(A − λI2 ) = λ(λ − (a + b)) = 0
from which we find that λ1 = 0 and λ2 = a + b. A will therefore have a zero eigenvalue
with multiplicity 2 for the infinitely many values of a, b ∈ R such that b = −a.
(c) Assuming that det(A) = ab (equivalently, A is a diagonal matrix since c = 0), we have
that
a + b ∓ |a − b|
det(A − λI2 ) = λ2 − (a + b)λ + ab = 0 ⇒ λ1,2 =
2
thus λ1 = min{a, b} and λ2 = max{a, b} with λ1 ≤ λ2 . If a = b then λ1 = λ2 , and A
admits the eigenvalue λ = a with multiplicity 2.
1.2. SOLUTIONS 9

1.5. (a) We can proceed in two different ways. The condition that λ = 1 is an eigenvalue of
In and x is the corresponding non-zero eigenvector is expressed in the writing In x = 1x,
which is certainly true since In x = x and 1x = x.
Alternatively, we can observe that
det(In − λ · In ) = det[(1 − λ)In ] = (1 − λ)n det(In ) = (1 − λ)n
which tells us that λ = 1 is an eigenvalue of In with multiplicity n.
We can proceed in a similar fashion for the other points.
1.6. (1) (a) The eigenvalues of A are λ1 = −2, λ2 = 4. The eigenvectors associated with λ1 = −2
are  
1 1 −t
(A − λ1 I2 )x = 0 ⇒ x = , t ∈ R ̸= 0
t
while the eigenvectors associated with λ2 = 4 are
 
2 2 s
(A − λ2 I2 )x = 0 ⇒ x = , s ∈ R ̸= 0
s
(b) We observe that
2
X 2
Y
tr(A) = 2 = 4 − 2 = λs and det(A) = −8 = (−2) 4 = λs
s=1 s=1

(c) Let α, β ∈ R. The eigenspaces associated with λ1 = −2 and λ2 = 4, denoted by W−2

and W4 are
W−2 = {α(−1, 1) : α ∈ R} and W4 = {β(1, 1) : β ∈ R}
each with dimension 1 (from a geometric standpoint, they are lines passing through the
origin).
√ 
√ √
√  
1 2−1 2 −(1 + 2)
(2) (a) λ1 = 1 − 2, x = t, λ2 = 1 + 2, x = s, with non-zero
1 1
t, s ∈ R.
(b) We have that
2 2
√ √ X √ √ Y
tr(A) = 2 = (1− 2)+(1+ 2) = λs and det(A) = −1 = (1− 2) (1− 2) = λs
s=1 s=1

(c) Let α, β ∈ R. We obtain

√ √
W1−√2 = {α( 2 − 1, 1) : α ∈ R} and W1+√2 = {β(−1 − 2, 1) : β ∈ R}
 √
√ √
 √ 
− 5−2 5−2
(3)(a) λ1 = − 5, x1 = t, λ2 = 5, x2 = s, with non-zero t, s ∈ R.
1 1
(b) We have that
2 2
√ √ X √ √ Y
tr(A) = 0 = − 5 + 5 = λs and det(A) = −5 = − 5 5 = λs
s=1 s=1

(c) Let α, β ∈ R. We obtain

√ √
W−√5 = {α(− 5 − 2, 1) : α ∈ R} and W√5 = {β( 5 − 2, 1) : β ∈ R}
10 CHAPTER 1. EIGENVALUES AND EIGENVECTORS

1.7. (a) Let λ ∈ R. The determinant of the matrix A − λI3 (I3 is the identity matrix of order
three) is
2−λ 1 0
det(A − λI3 ) = 1 2−λ 0 = (2 − λ)[(2 − λ)2 − 1]
0 0 2−λ
From the equation (2 − λ)[(2 − λ)2 − 1]=0 we find the eigenvalues of A, which are the
distinct values λ1 = 1, λ2 = 2 and λ3 = 3. So, the spectrum of A is σ(A) = {1, 2, 3}.
The eigenvectors associated with the eigenvalue λ1 = 1, denoted by x1 , are the non-zero
solutions to the linear system (A − 1I3 )x1 = 0, thus
 
x1 + x2 = 0 x1 = t
 
  t
x1 + x2 = 0 ⇒ x2 = −t ⇒ x1 = −t , t ∈ R − {0}
0
 
x3 = 0 x3 = 0
 

The eigenvectors associated with the eigenvalue λ2 = 2, denoted by x2 , are the non-zero
solutions to the linear system (A − 2I3 )x2 = 0, thus

x2 = 0
 
 0
2
x1 = 0 ⇒ x = 0 , s ∈ R − {0}

s

x3 = s


The eigenvectors associated with the eigenvalue λ3 = 3, denoted by x3 , are the non-zero
solutions to the linear system (A − 3I3 )x3 = 0, thus
 
−x1 + x2 = 0 x1 = k
 
  k
1
x1 − x2 = 0 ⇒ x2 = k ⇒ x = k  , k ∈ R − {0}

0
 
−x3 = 0 x3 = 0
 

(b) We have that

3
X 3
Y
tr(A) = 6 = 1 + 2 + 3 = λs and det(A) = 6 = 1 · 2 · 3 = λs
s=1 s=1

(c) Let α, β, γ ∈ R. We obtain

W1 = {α(1, −1, 0) : α ∈ R}; W2 = {β(0, 0, 1) : β ∈ R}; W3 = {γ(1, 1, 0) : γ ∈ R}

all of dimension 1.
(2) (a) λ1 = 0, λ2 = 1, λ3 = 2. The corresponding eigenvectors, denoted respectively by
x1 , x2 , x3 , are      
1 0 1
x1 = −1 t x2 = 0 s x3 = 1 k
0 s 0
with non-zero t, s, k ∈ R.
(b) We have that
3
X 3
Y
tr(A) = 3 = 0 + 1 + 2 = λs and det(A) = 0 = 0 · 1 · 2 = λs
s=1 s=1
1.2. SOLUTIONS 11

(c) Let α, β, γ ∈ R. We obtain

W0 = {α(1, −1, 0) : α ∈ R}; W1 = {β(0, 0, 1) : β ∈ R}; W2 = {γ(1, 1, 0) : γ ∈ R}
all of dimension 1.
(3) (a) λ1 = −1, λ2 = 1, λ3 = 2. The corresponding eigenvectors, denoted respectively by
x1 , x2 , x3 , are
     
√ 1 0 √ 1
x1 = − 2 t x2 = 0 s x3 = 1/ 2 k
0 s 0
with non-zero t, s, k ∈ R.
(b) We have that
3
X 3
Y
tr(A) = 2 = −1 + 1 + 2 = λs and det(A) = −1 = (−1) · 1 · 2 = λs
s=1 s=1

(c) Let α, β, γ ∈ R. We obtain

√ √
W0 = {α(1, − 2, 0) : α ∈ R}; W1 = {β(0, 0, 1) : β ∈ R}; W2 = {γ(1, 1/ 2, 0) : γ ∈ R}
all of dimension 1.

1.8. (a) Let t ∈ R. The function pA : R → R defined by pA (t) = det(tI3 − A) is called the
characteristic polynomial of A. We have that
 
t−1 0 −1    
t−1 1 0 t−1
det(tI3 − A) = det  0 t−1 1 = (t − 1) det
 − det
1 t −1 1
−1 1 t
from which we obtain
pA (t) = (t − 1) [t(t − 1) − 1] − (t − 1) = (t − 1)(t2 − t − 2)
(b) The determinant of A is −2. We observe that
pA (0) = 2 = (−1)(−2) = (−1)3 (−2) = (−1)3 det(A)
(c) Let λ ∈ R. The real number λ is an eigenvalue of A if and only if λ is a zero of the
characteristic polynomial of A, that is, pA (λ) = 0. (Note that we are guaranteed the
existence of at least one real zero by the zero-value theorem. This is due to the continuity of
the function PA (t) and the fact that for t → −∞ and t → +∞ the function has asymptotic
behaviour of opposite signs.) Finding the values of λ that are the solutions to pA (t) = 0,
i.e., the values λ for which
(λ − 1)(λ2 − λ − 2) = 0
yields the following: λ = −1, λ = 1 and λ = 2.
(2) (a) pA (t) = (t − 1)(t2 − 2).
(b) Proceed√as in (1) (b). √
(c) λ1 = − 2, λ1 = 1 and λ1 = 2.
(3) (a) pA (t) = t(t − 1)(t − 3).
(b) Proceed as in (1) (b).
(c) λ1 = 0, λ1 = 1 and λ1 = 3.
12 CHAPTER 1. EIGENVALUES AND EIGENVECTORS

1.9. (a) We assume, by hypothesis, that λ is an eigenvalue of A, so we know there exists a

non-zero vector x ∈ Rn such that Ax = λx (x is an eigenvector of A associated with λ).
Using the associative property, we have
A2 x = A(Ax) = A(λx) = λ(Ax) = λ(λx) = λ2 x
and we conclude that λ2 is an eigenvalue of A2 .
(b) Considering A = −I, we have that A2 = I. Let x ∈ Rn be a non-zero vector. We have
that
Ax = (−I)x = (−1)x and A2 x = Ix = 1x
so the only eigenvalue associated with matrix A is λ = −1, with multiplicity n; and the
only eigenvalue associated with matrix A2 is µ = 1, again with multiplicity n. Observing
√
that λ = −1 ̸= 1 = µ, we have the desired result.
1.10. (a) The determinant of the matrix A − λI2 is
det(A − λI2 ) = −λ(1 − λ) − 2 = λ2 − λ − 2
so det(A − λI2 ) = 0 if and only if λ1 = −1 and λ2 = 2 (A admits distinct eigenvalues), and
we conclude that the spectrum of A is σ(A) = {−1, 2}. The eigenvectors associated with
λ1 = −1, denoted by x1 , are the non-zero solutions to the equation (A − (−1)I2 )x1 = 0,
which are  √   √ 
1 −t/ 2 −1/ 2
x = = t t ∈ R − {0}
t 1
while the eigenvectors associated with λ2 = 2, denoted by x2 , are the non-zero solutions to
the equation (A − 2I2 )x2 = 0, which are
√  √ 
1 2s 2
x = = s s ∈ R − {0}
s 1

Observe that x1 · x2 = 0 for every non-zero t, s ∈ R.

(b) We set t = s = 1 and call S the subset of R2 defined by
√ √
S = {x1 , x2 } = {(−1/ 2, 1), ( 2, 1)} ⊆ R2
p √
S is an orthogonal basis of R2 . Given that ||x1 || = 3/2 and ||x2 || = 3, we normalize
the eigenvectors x1 and x2 , obtaining
x1 1 √
√ √ √

1
=p −1/ 2 1 = −1/ 3 2/ 3
||x || 3/2
and
x2 1 √
√ √ √

2
=√ 2 1 = 2/ 3 1/ 3
||x || 3
So, the subset of R2 , denoted by S ∗ , described as follows
√ √ √
√ √ √

S ∗ = {x1 /||x1 ||, x2 /||x2 ||} = { −1/ 3 2/ 3 , 2/ 3 1/ 3 } ⊆ R2
is an orthonormal basis of R2 (i.e., it is an eigenbasis of R2 ). The orthonormal matrix B,
constructed from the orthonormal eigenvectors of A, is given by
" # " √ √ √ #
x1 /||x1 || −1/ 3 2/ 3
B= 2 2
= √ √ √
x /||x || 2/ 3 1/ 3
1.2. SOLUTIONS 13

(c) The product B T AB is

" √ √ √ #  √  " √ √ √ # 
−1/ 3 2/ 3 −1/ 3 2/ 3

T √1 2 −1 0
B AB = √ √ √ √ √ √ =
2/ 3 1/ 3 2 0 2/ 3 1/ 3 0 2

which is equal to the diagonal matrix Λ whose main diagonal entries are the eigenvalues of
A.

1.11. (a) The determinant of the matrix A − λI3 is

det(A − λI3 ) = (2 − λ)[1 − λ)2 − 1] = −λ(2 − λ)2

so det(A − λI3 ) = 0 if and only if λ1 = 0, with m(0) = 1, and λ1 = 2, with m(2) = 2, and
we conclude that the spectrum of A is σ(A) = {0, 2}. The eigenvectors associated with
λ1 = 0, denoted by x1 , are the non-zero solutions to the equation (A − 0I3 )x1 = 0, which
are    
t 1
x1 = −t = −1 t t ∈ R − {0}
0 0
while the eigenvectors associated with λ2 = 2, denoted by x2 , are the non-zero solutions to
the equation (A − 2I3 )x2 = 0, which are
     
s 1 0
x2 =  s = 1 s + 0 k s, k ∈ R − {0}
k 0 1

The eigenspaces W0 and W2 are

W0 = {α(1, −1, 0) : α ∈ R} W2 = {β(1, 1, 0) + γ(0, 0, 1) : β, γ ∈ R}

while the normalized eigenspaces W0∗ and W2∗ are

√ √ √ √
W0∗ = {α(1/ 2, −1/ 2, 0) : α ∈ R} W2∗ = {β(1/ 2, 1/ 2, 0) + γ(0, 0, 1) : β, γ ∈ R}

(b) Let S be the subset of R3 defined by

√ √ √ √
S = {(1/ 2, −1/ 2, 0), (1/ 2, 1/ 2, 0), (0, 0, 1)} ⊆ R3

S is an orthonormal basis and an eigenbasis of R3 . The orthonormal matrix B, formed by

the orthormal eigenvectors of A, is
 √ √ 
1/ 2 1/ 2 0
√ √
B = −1/ 2 1/ 2 0
 

0 0 1

(c) The product B T AB is

 √ √   √ √  
1/ 2 −1/ 2 0 1 1 0 1/ 2 1/ 2 0 0 0 0

 √ √ √ √
B T AB = 1/ 2 1/ 2 0 1 1 0 −1/ 2 1/ 2 0 = 0 2 0
  

0 0 1 0 0 2 0 0 1 0 0 2

which is the diagonal matrix Λ whose main diagonal entries are the eigenvalues of A.
14 CHAPTER 1. EIGENVALUES AND EIGENVECTORS

1.12. (a) The eigenvalues of A are λ1 = 1, λ2 = 2 and λ3 = 4 (A admits distinct eigenvalues). The
eigenvectors associated with the eigenvalues of A, denoted by x1 , x2 and x3 , respectively,
are  √     √ 
− 2 0 1/ 2
x1 =  0  t x2 = 1 s x3 =  0  k
1 0 1
for every non-zero t, s, k ∈ R.
(b) The orthonormal matrix B, formed by the orthonormal eigenvectors of A, is
 √ √ √ 
− 2/ 3 0 1/ 3
B= 0 1 0
 
√ √ √
1/ 3 0 2/ 3

(c) The product B T AB is

 √ √ √  √   √ √ √  
− 2/ 3 0 1/ 3  2 0 2 − 2/ 3 0 1/ 3 1 0 0


B T AB =  0 1 0 √0 2 0   0 1 0 = 0 2 0
   
√ √ √ √ √ √
1/ 3 0 2/ 3 2 0 3 1/ 3 0 2/ 3 0 0 4

which is equal to the diagonal matrix Λ whose main diagonal entries are the eigenvalues of
A.
Chapter 2

Quadratic forms

2.1 Exercises
2.1. Consider the quadratic form f in R2 defined by f (x) = x · Ax with
       
3 1 1 2 3 −1 −1 −2
(1) A = (2) A = (3) A = (4) A =
1 1 2 1 −1 2 −2 −4

(a) Write the quadratic form f as the sum of homogeneous second degree monomials.
(b) Using the definition, determine the sign of the quadratic form f .

2.2. Consider the quadratic form f in R3 defined by f (x) = x · Ax with

     
1 1 0 1 1 0 2 1 2
(1) A = 1 2 0 (2) A = 1 1 0 (3) A = 1 1 0
0 0 3 0 0 3 2 0 1

(a) Write the quadratic form f as the sum of homogeneous second degree monomials.
(b) Using the definition, determine the sign of the quadratic form f .

2.3. Let a ∈ R. Consider the quadratic form in R2 given by

(1) f (x1 , x2 ) = x21 + ax22 (2) f (x1 , x2 ) = x21 − ax22 (3) f (x1 , x2 ) = x21 + ax22 + 2x1 x2

Determine the sign of f as a function of a.

2.4. Let f , g be two quadratic forms in Rn .

(a) Prove the following implication: If f and g are positive-definite, then f + g is positive-
definite.
(b) Prove the following implication: If f is positive definite and g is positive-semidefinite,
then f + g is postive-definite.
(c) Show, using an appropriate counterexample, that the converse of the implication
stated in part (b) is false.

15
16 CHAPTER 2. QUADRATIC FORMS

2.5. Let f be a quadratic form in Rn and k ∈ R.

(a) Prove that f is negative-definite if and only if −f is positive-definite.

(b) Prove that f is positive-definite if and only if there exists k > 0 such that (kf ) is
positive definite.
(c) Prove that if f is positive-definite, then f is positive-semidefinite. Show, using an
appropriate counterexample, that the converse is false.

2.6. Let A be a square matrix of order 2 where

 √       
√ 2 2 4 2 1 2 −1 2
(1) A = (2) A = (3) A = (4) A =
2 2 2 1 2 1 2 −1

Using the eigenvalues of A, determine the sign of the quadratic form associated with matrix
A.

2.7. Let a ∈ R and let A be a square matrix of order 2 where

 
1 a
A=
a 1

Using the eigenvalues of A, determine the sign of the quadratic form associate with matrix
A as a varies.

2.8. Let A be a square matrix of order 3 where

 
2 1 0
A = 1 2 0
0 0 5

(a) Find the eigenvalues of A and determine the sign of f (x) = x · Ax.
(b) Using the Sylvester-Jacobi criterion, determine the sign of f (x) = x · Ax.

2.9. Let A be a square matrix of order 3 where

 
1 1 1
A = 1 1 1
1 1 1

(a) Find the eigenvalues of A and determine the sign of f (x) = x · Ax.
(b) Using the Sylvester-Jacobi criterion, determine the sign of f (x) = x · Ax.

2.10. Let a ∈ R and let A be a square matrix of order 3 where

     
a 1 0 a 1 0 1 a 0
(1) A = 1 1 1 (2) A = 1 a 1 (3) A = a 1 a
0 1 a 0 1 a 0 a 1

Using the Sylvester-Jacobi criterion, find the values a such that the quadratic form
f (x) = x · Ax is positive-definite.
2.2. SOLUTIONS 17

2.2 Solutions
2.1. (1) (a) Let x ∈ R2 . Evaluating the inner product of x and Ax we obtain
     
3 1 x1 3x1 + x2
f (x) = (x1 , x2 ) · = (x1 , x2 ) ·
1 1 x2 x1 + x2
= x1 (3x1 + x2 ) + x2 (x1 + x2 ) = 3x21 + 2x1 x2 + x22

so f is a sum of second degree monomials.

(b) Observe that f can be written as

f (x) = 2x21 + x21 + 2x1 x2 + x22 = 2x21 + (x1 + x2 )2

so for every x1 , x2 ∈ R ̸= 0 we have that f (x) > 0. Thus, f is positive-definite in R2 .

(2) (a) Let x ∈ R2 . We obtain

f (x) = x21 + 4x1 x2 + x22

(b) f is indefinite in R2 .
(3) (a) Let x ∈ R2 . We obtain

f (x) = 3x21 − 2x1 x2 + 2x22

(b) f is positive-definite in R2 .
(4) (a) Let x ∈ R2 . We obtain

f (x) = −x21 − 4x1 x2 − 4x22

(b) f is negative-semidefinite in R2 .

2.2. (1) (a) Let x ∈ R3 . Evaluating the inner product of x and Ax we obtain
     
1 1 0 x1 x1 + x2
f (x) = (x1 , x2 , x3 ) · 1 2 0 x2  = (x1 , x2 , x3 ) · x1 + 2x2 
0 0 3 x3 3x3
= x1 (x1 + x2 ) + x2 (x1 + 2x2 ) + 3x23 = x21 + 2x1 x2 + 2x22 + 3x23

so f is a sum of second degree monomials.

(b) The quadratic form f can be written as

f (x) = x21 + 2x1 x2 + 2x22 + 3x23 = (x1 + x2 )2 + x22 + 3x23

so for every x1 , x2 , x3 ∈ R ̸= 0 we have that f (x) > 0. Thus, f is positive-definite in R3 .

(2) (a) Let x ∈ R3 . We obtain

f (x) = (x1 + x2 )2 + 3x23

(b) Observe that ∀x ∈ R3 , f (x) ≥ 0 and there exists at least one non-zero vector in R3 , for
example, x∗ = (1, −1, 0), such that f (x) = 0. We conclude that the quadratic form f is
positive-semidefinite in R3 .
18 CHAPTER 2. QUADRATIC FORMS

(3) (a) Let x ∈ R3 . We obtain

f (x) = (x1 + x2 )2 + (x1 + 2x2 )2 − 3x23

(b) Consider the two distinct vectors x′ = (1, 0, 0) and x′′ = (0, 0, 1). They represent two
vectors in R3 for which the quadratic form obtains different sign, i.e., f (x′ ) = 2 > 0 while
f (x′′ ) = −3 < 0. We conclude that the quadratic form f is indefinite in R3 .

2.3. (1) Let a be any real number. If a > 0, then the quadratic form can be re-written as
√
f (x) = x21 + ax22 = x21 + ( a)2 x22

So, f (x) > 0 for every x1 , x2 ∈ R ̸= 0. Thus, f is positive-definite in R2 . Conversely,

suppose that f is positive-definite in R2 (i.e., f (x) > 0 for every x ∈ R2 ̸= 0) and suppose,
by contradiciton, that a ≤ 0. This immediately leads to a contradiction with the starting
hypothesis and we conclude that a > 0. Therefore, f is positive-definite if and only if a > 0.
Following the same line of reasoning, we find that f is positive semi-definite if and only if
a = 0. Finally, it can be confirmed that f is indefinite if and only if a < 0.
(2) f is positive-definite (positive-semidefinite/indefinite) if and only if a < 0 (a = 0/a > 0).
(3) f is positive-definite (positive-semidefinite/indefinite) if and only if a > 1 (a = 1/a < 1).

2.4. (a) By hypothesis, f, g are positive-definite: so f (x) = x · Ax > 0 and g(x) = x · Bx > 0 for
every x ∈ Rn ̸= 0. By the properties of the inner product and by the definition of the sum
of two definite functions in Rn , we have that (f + g)(x) = x · (A + B)x. Let x ∈ Rn ̸= 0.
Given that
(f + g)(x) = x · (A + B)x = x · Ax + x · Bx > 0 + 0 = 0
we conclude that the quadratic form f + g is positive-definite in Rn .
(b) Use the same strategy as in part (a).
(c) We must show that if f + g is positive-definite in Rn , it is not guaranteed that f will
be positive-definite and g positive-semidefinite in Rn . We consider the quadratic form
(f + g)(x) = 2x21 + 2x1 x2 + 2x22 , which is positive-definite in R2 . Given that

(f + g)(x) = 2x21 + 2x22 + 2x1 x2 , with f (x) = 2x21 + x22 and g(x) = 2x1 x2

where f is positive-definite in R2 and g is indefinitein R2 . We have thus found an acceptable

counterexample.

2.5. (a) Let −f be positive-definite. Thus, (−f )(x) > 0 for every non-zero x ∈ Rn . Given that
f (x) = −(−f (x)) < 0 for every x ∈ Rn , we conclude that f is negative-definite.
Let f be negative-definite. Thus, f (x) < 0 for every non-zero x ∈ Rn . Since (−f )(x) =
−f (x) > 0 for every x ∈ Rn , we deduce that −f is positive-definite.
(b) Use the same strategy as in part (a).
(c) Assume that f is positive-definite. Thus, ∀x ∈ Rn ̸= 0, f (x) > 0. Given that
f (x) > 0 ≥ 0, we have that f (x) ≥ 0 for every x ∈ Rn =
̸ 0. Thus, f is positive-semidefinite.
For the converse of the implication it suffices to observe that by hypothesis there exists
at least one non-zero x ∈ Rn , such that f (x) = 0: as a consequence the thesis is false.
For a concrete example, consider the quadratic form f (x) = x21 in R2 , which is positive-
semidefinite, and observe that we have f (0, α) = 0 for every α ∈ R.
2.2. SOLUTIONS 19

2.6. (1) From the equation

√ (2 − λ)2 − 2√= 0 we find the eigenvalues of A, which are√the distinct
√
values λ1 = 2 − 2 and λ2 = 2 + 2. So, the spectrum of A is σ(A) = {2 − 2, 2 + 2}.
Given that the eigenvalues of A are strictly positive, we conclude that the quadratic form
associated with matrix A is positive-definite.
(2) Positive-semidefinite.
(3) Indefinite.
(4) Indefinite.

2.7. The eigenvalues of A are λ1 = 1 − |a| and λ2 = 1 + |a|. Observe that λ2 > 0 for every real
value of a. We distinguish between the following cases: (1) if |a| > 1, then λ1 < 0 < λ2
and the quadratic form is indefinite; (2) if |a| = 1, then λ1 = 0 < λ2 and the quadratic
form is positive-semidefinite; (3) if |a| < 1, then 0 < λ1 ≤ λ2 and the quadratic form is
positive-definite.

2.8. (a) It can be shown that the spectrum of A is σ(A) = {1, 3, 5}. Thus, the quadratic form
f associate with matrix A is positive-definite. (All of the eigenvalues of A are strictly
positive.)
(b) The NW principal minors of A are

det(A1 ) = 2 > 0 det(A2 ) = 3 > 0 det(A3 ) = 15 > 0

Thus, we can conclude that f is positive definite.

2.9. (a) It can be shown that the spectrum of A is σ(A) = {0, 3} with m(0) = 2. Thus, the
quadratic form f associated with matrix A is positive-semidefinite. (All of the eigenvalues
of A are positive.)
(b) Studying the NW principal minors of A is inconclusive. However, given that the
minors of A (principal and non-principal minors!) are all positive, we conclude that f is
positive-semidefinite.

2.10. (a) The NW principal minors of A are

det(A1 ) = a det(A2 ) = a − 1 det(A3 ) = a(a − 2)

which are strictly positive if and only if a > 2. Thus, the quadratic form f is positive-definite
if and only
√ if a > 2.
(b) a > √ 2. √
(c) −1/ 2 < a < 1/ 2.
Chapter 3

Unconstrained optimization

3.1 Exercises
3.1. Let f : A ⊆ R2 → R with

1 x1
(a) f (x) = x31 x2 + 2x22 (b) f (x) = + (c) f (x) = ln(x21 + 2x22 + 1)
x1 x2

Calculate the gradient operator of f on (−1, 1).

3.2. Let f : R2 → R with

f (x1 , x2 ) = x1 ex1 x2

(a) Calculate the gradient and the Hessian matrix of f for a generic point x ∈ R2 .
(b) Calculate the NW principal minors of ∇2 f (1, 0).

3.3. Let f : R3 → R with

f (x) = x21 + 2x1 x2 + x22 + 2(x3 − 1)2

(a) Calculate the gradient and determine the stationary points of f .

(b) Calculate the NW principal minors of the Hessian matrix of f .

3.4. Let f : A ⊆ R2 → R with

f (x) = ln x1 + ln x2 − x21 − x22

(a) Verify that f is strictly concave in A.

(b) Calculate the second-order Taylor expansion of f centered at (1, 1).

3.5. Let f : R2 → R with

f (x) = x1 ex1 +x2

(a) Calculate the gradient and the Hessian matrix of f for a generic point x ∈ R2 .
(b) Calculate the second-order Taylor expansion of f centered at (0, 0).

21
22 CHAPTER 3. UNCONSTRAINED OPTIMIZATION

3.6. Consider the following optimization problem

2 2
max 1 + e−x1 −x2 sub x ∈ R2


(a) Show that it is a coercive optimization problem.

(b) Solve the problem.

3.7. Consider the following optimization problem

1
max 2 ln x1 − x21 − 2 − 16x2 + 15 sub x ∈ (0, +∞) × (0, +∞) = (0, +∞)2
 
x2

(a) Show that it is a concave optimization problem.

(b) Solve the problem.

3.8. Let f : R2 → R with

f (x) = x21 − 2x22 x1 + 2x22
Determine, if they exist, the points of local extrema of f .

3.9. Let f : A ⊆ R2 → R with

1
f (x) = ln x1 + − ln x2 + 2x2
x1
Determine, if they exist, the points of local extrema of f .

3.10. Let f : R3 → R with

(a) f (x) = x31 − x22 − x23 − 3x1 + 2x3 (b) f (x) = x31 + 6x1 x3 − x22 + 3x23 − 9x1

Determine, if they exist, the points of local extrema of f .

3.2 Solutions
3.1. (a) The function f is defined on R2 and f ∈ C 1 (R2 ) (f is continuously derivable on R2 – a
sufficient condition for the differentiability of f on R2 ). The first partial derivatives of f
are
fx′ 1 (x) = 3x21 x2 fx′ 2 (x) = x31 + 4x2
The gradient operator of f is the vector-valued function ∇f : R2 → R2 that associates the
gradient ∇f (x) to each x ∈ R2 . In this case, for each vector (x1 , x2 ) ∈ R2 we have

∇f (x1 , x2 ) = 3x21 x2 x31 + 4x2

 

The gradient operator is defined at the point (−1, 1) and by substituting we obtain
∇f (−1, 1) = (3, 3)
(b) The function f is defined on the subset of the Cartesian plane, A = (R−{0})×(R−{0}).
The gradient operator of f is defined at the point (−1, 1) ∈ A and we obtain ∇f (−1, 1) =
(0, 1).
(c) ∇f (−1, 1) = (−1/2, 1).
3.2. SOLUTIONS 23

3.2. The function is defined on R2 and f ∈ C 2 (R2 ).

(a) The gradient and the Hessian matrix of f for x ∈ R2 are

(2x2 + x1 x22 )ex1 x2 (2x1 + x21 x2 )ex1 x2

 
x x 2 x x 2
 
∇f (x) = (1 + x1 x2 )e 1 2 x1 e 1 2 , ∇ f (x) =
(2x1 + x21 x2 )ex1 x2 x31 ex1 x2

(b) At the point (1, 0), the Hessian matrix of f is

 
0 2
∇f (1, 0) =
2 1

so the NW principal minors of the Hessian matrix at the point (1, 0) are 0 and −4.

3.3. The function f is defined on R3 and f ∈ C 2 (R3 ).

(a) The gradient of f is
 
∇f (x) = 2x1 + 2x2 2x1 + 2x2 4(x3 − 1)

and from the equation ∇f (x) = 0 we find that the stationary points of f are all points
belonging to S = {(k, −k, 1)}, with k ∈ R (an infinite set of points). We observe that all
points which do not belong to S are not stationary points of f and consequently cannot be
points of local extrema of f .
(b) The Hessian matrix of f for x ∈ R3 is
 
2 2 0
∇2 f (x) = 2 2 0
0 0 4

and the NW principal minors of ∇2 f (x) are 2, 0 and 0.

3.4. The function f is defined on A = (0, +∞) × (0, +∞) = (0, +∞)2 and f ∈ C 2 (A).
(a) The gradient and the Hessian matrix of f are
" 1 #
1 1
− x2 − 2 0
2

∇f (x) = x1 − 2x1 x2 − 2x2 , ∇ f (x) = 1
0 − x12 − 2
2

By analyzing the NW principal minors of f for x ∈ A we conclude that ∇2 f is negative-

definite on A and thus f is strictly concave in A.
(b) The second-order Taylor expansion at the point x0 = (1, 1) is
1
f (x) = f (x0 ) + ∇f (x0 ) (x − x0 ) + (x − x0 )T ∇2 f (x) (x − x0 ) + o(||x − x0 ||2 ), as x → x0
2
The value, the gradient and the Hessian matrix of f evaluated at x0 are
 
  2 −3 0
f (1, 1) = −2 ∇f (1, 1) = −1 −1 ∇ f (1, 1) =
0 −3
and by substituting we have
3 3
f (x) = −2 − (x1 − 1) − (x2 − 1) − (x1 − 1)2 − (x2 − 1)2 + o(||x − x0 ||2 ), as x → x0
2 2
24 CHAPTER 3. UNCONSTRAINED OPTIMIZATION

3.5. The function f is defined on R2 and f ∈ C 2 (R2 ).

(a) The gradient and the Hessian matrix of f for a generic point x ∈ R2 are

(2 + x1 )ex1 +x2 (1 + x1 )ex1 +x2

 
x +x x +x 2
 
∇f (x) = (1 + x1 )e 1 2 x1 e 1 2 , ∇ f (x) =
(1 + x1 )ex1 +x2 x1 ex1 +x2

(b) The second-order Taylor expansion at the point x0 = (0, 0) is

1
f (x) = f (x0 ) + ∇f (x0 ) x + xT ∇2 f (x) x + o(||x||2 ), as x → 0
2
The value, the gradient and the Hessian matrix of f evaluated at x0 are
 
  2 2 1
f (0, 0) = 0 ∇f (0, 0) = 1 0 ∇ f (0, 0) =
1 0

and by substituting we have

f (x) = x1 + x21 + x1 x2 + o(||x||2 ), as x → 0

3.6. The function f is defined on R2 and f ∈ C 2 (R2 ).

(a) The objective function is coercive on R2 (as it is the composition of a strictly increasing
function and the function g(t) = −x21 − x22 , which is coercive on R2 ). Thus, the given
problem is an unconstrained optimization problem (as there are no constraints on the
variables), differentiable and coercive. Therefore, Tonelli’s theorem guarantees the existence
of a global maximum for f on R2 .
(b) The gradient of f is
h i
2 2 2 2
∇f (x) = −2x1 e−x1 −x2 −2x2 e−x1 −x2

and S = {(0, 0} is the set of stationary points of f . We can conclude that arg max f (x) =
{(0, 0)} and max f (x) = {2}.

3.7. The function f is defined on A = (0, +∞) × (0, +∞) and f ∈ C 2 (A).
(a) The gradient and the Hessian matrix of f for a generic point x ∈ A are

−2/x21 − 2
 
 3
 2 0
∇f (x) = 2/x1 − 2x1 2/x2 − 16 , ∇ f (x) =
0 −6/x42

and we can observe that the Hessian matrix is negative-definite for every x ∈ A. Thus, f
is concave in A. We conclude that the given problem is an unconstrained optimization
problem, differentiable and concave (as the objective function is concave).
(b) The gradient of f is equal to 0 only at the point (1, 1/2) ∈ A, so we conclude that
arg max f (x) = {(1, 1/2)} and max f (x) = {2}.

3.8. The function is defined on R2 and f ∈ C 2 (R2 ). The gradient of f is

∇f (x) = 2x1 − 2x22 4x2 − 4x1 x2

 
3.2. SOLUTIONS 25

From the equation ∇f (x) = 0 we find the set of stationary points of f ,

S = {(0, 0), (1, −1), (1, 1)}

The Hessian matrix of f is

 
2 2 −4x2
∇ f (x) =
−4x2 4 − 4x1

which, evaluated at the points in S yields

     
2 0 2 4 2 −4
∇2 f (0, 0) = ∇2 f (1, −1) = ∇2 f (1, 1) =
0 4 4 0 −4 0

Therefore,

• At (0, 0) the Hessian matrix is positive-definite and (0, 0) is a point of local minimum
for f with the value f (0, 0) = 0.
• At (1, −1), the Hessian matrix is indefinite and (1, −1) is not a point of local extremum
for f (we call (1, −1) a saddle point of f ).
• At (1, 1) the Hessian matrix is indefinite and (1, 1) is again a saddle point of f .

3.9. The function f is defined on A = (0, +∞)2 and f ∈ C 2 (A). We obtain that the function f
admits the point of local minimum (1, 1/2), with the value 2 + ln 2.

3.10. (a) The function f is defined on R3 and f ∈ C 2 (R3 ). The function admits the following
set of stationary points: S = {(−1, 0, 1), (1, 0, 1)}. We find that the function f admits the
point of local maximum (−1, 0, 1), with value −1, while the other stationary point is not a
point of local extremum for f .
(b) The function f is defined on R3 and f ∈ C 2 (R3 ). The function admits the following set
of stationary points: S = {(−1, 0, 1), (3, 0, −3)}. We find that the two stationary points
are not points of local extremum for f .
Chapter 4

Implicit functions and constrained

optimization

4.1 Exercises
4.1. Let g be a real-valued function of 2 variables with g(x, y) = 2x − 3 − y. Verify that ∀k ∈ R
there exists a function f : R → R, such that g(x, f (x)) = k for all x ∈ R.

4.2. Let g be a real-valued function of 2 variables with g(x, y) = x4 + y. Verify that there exists
a function f : R → R, such that g(x, f (x)) = 0 for all x ∈ R.

4.3. Let g be a real-valued function of 2 variables with g(x, y) = ex − 1 + e−y . Verify that there
exists a function f : (−∞, 1) → R, such that g(x, f (x)) = e − 1 for all x ∈ (−∞, 1).

4.4. Consider g : R2 → R with

g(x1 , x2 ) = 2x1 ex2 − x2 ex1 − 2, g(1, 0) = 0

(a) Show that there exists a unique function f : B(1) → V (0) such that x2 = f (x1 ) for
every (x1 , x2 ) ∈ B(1) × V (0).
(b) Calculate f ′ (1) and write the first-order Taylor expansion of f centered at x1 = 1.

4.5. Consider g : R2 → R with

g(x1 , x2 ) = 2x31 − x2 − ex2 −1 , g(1, 1) = 0

(a) Show that there exists a unique function f : B(1) → V (1) such that x2 = f (x1 ) for
every (x1 , x2 ) ∈ B(1) × V (1).
(b) Calculate f ′ (1) and write the first-order Taylor expansion of f centered at x1 = 1.

4.6. Consider g : R2 → R with

g(x1 , x2 ) = x31 + x1 x22 − 2x32 , g(1, 1) = 0

(a) Show that there exists a unique function f : B(1) → V (1) such that x2 = f (x1 ) for
every (x1 , x2 ) ∈ B(1) × V (1).

27
28 CHAPTER 4. IMPLICIT FUNCTIONS AND CONSTRAINED OPTIMIZATION

(b) Calculate f ′ (1) and write the first-order Taylor expansion of f centered at x1 = 1.

4.7. Consider the following constrained optimization problem

max x2 − x21 sub x21 + x22 = 4

 
x

Find, if they exist, the points of constrained maximization for the given problem.

4.8. Consider the following constrained optimization problem

max ln x1 − x22 sub x21 + x22 = 1

 
x

Find, if they exist, the points of constrained maximization for the given problem.

4.9. Consider the following constrained optimization problem

max 5 − x21 − x22 sub x2 + x21 = 2

 
x

Find, if they exist, the points of constrained maximization for the given problem.

4.10. Consider the following constrained optimization problem

max x21 − x22 sub x21 + x22 = 1

 
x

Find, if they exist, the points of constrained maximization for the given problem.

4.2 Solutions
4.1. Let A = B = R, with A × B = R2 (the domain of g) and k ∈ R. The function g is
continuous with respect to y. Given that

∀x ∈ R, inf (2x − 3 − y) = −∞ < k < +∞ = sup(2x − 3 − y)

y∈R y∈R

there exists a function f : R → R such that g(x, f (x)) = k for all x ∈ R. In fact, the
function f : R → R is given by f (x) = 2x − 3 − k.

4.2. Let A = B = R, with A × B = R2 (the domain of g) and k ∈ R. The function g is

continuous with respect to y. Given that

∀x ∈ R, inf (x4 + y) = −∞ < 0 < +∞ = sup(x4 + y)

y∈R y∈R

there exists a function f : R → R such that g(x, f (x)) = 0 for all x ∈ R. In fact, the
function f : R → R is given by f (x) = −x4 .

4.3. Let A = (−∞, 1), B = R, with A × B = (−∞, 1) × R (the domain of g). The function g is
continuous with respect to y. Given that

∀x ∈ (−∞, 1), inf (ex − 1 + e−y ) = ex − 1 < e − 1 < +∞ = sup(ex − 1 + e−y )

y∈R y∈R

there exists a function f : (−∞, 1) → R such that g(x, f (x)) = e − 1 for all x ∈ (−∞, 1).
4.2. SOLUTIONS 29

4.4. (a) Given that g ∈ C 2 (R2 ), with

∂g(x1 , x2 ) ∂g(1, 0)
= 2x1 ex2 − ex1 , = 2 − e ̸= 0
∂x2 ∂x2
by the implicit function theorem there exists a unique real-valued function f : B(1) → V (0)
(i.e., the function is defined on a neighborhood of the point 1 and takes values in a
neighborhood of the point 0), which is continuously derivable, such that

∀x1 ∈ B(1), 2x1 ef (x1 ) − f (x1 )ex1 − 2 = 0

(b) Observe that the partial derivative of g with respect to x1 is gx′ 1 (x1 , x2 ) = 2ex2 − x2 ex1 .
Consequently, we have
2ex2 − x2 ex1
∀(x1 , x2 ) ∈ g −1 (0) ∩ (B(1) × V (0)), f ′ (x1 ) = −
2x1 ex2 − ex1
with f ′ (1) = 2/(e − 2) and the first-order Taylor expansion of f centered at x1 = 1 is

f (x1 ) = 2(e − 2)−1 (x1 − 1) + o(x1 − 1), as x1 → 1

4.5. (a) By the implicit function theorem there exists a unique real-valued function f : B(1) →
V (1), which is continuously derivable, such that

∀x1 ∈ B(1), 2x31 − f (x1 ) − ef (x1 )−1 = 0

(b) We have f ′ (1) = 3 and

f (x1 ) = 1 + 3(x1 − 1) + o(x1 − 1), as x1 → 1

4.6. (a) By the implicit function theorem there exists a unique real-valued function f : B(1) →
V (1), which is continuously derivable, such that

∀x1 ∈ B(1), x31 + x1 (f (x1 ))2 − 2(f (x1 ))3 = 0

(b) We have f ′ (1) = 1 and

f (x1 ) = x1 + o(x1 − 1), as x1 → 1

4.7. Let f, g : R2 → R, with f (x) = x2 − x21 , g(x) = x21 + x22 , b = 4. The functions f, g ∈ C 1 (R2 ),
so D = R2 . We have

C − D = {x ∈ R2 : x2 + x22 = 4} − R2 = ∅

and D0 = {x ∈ D : ∇g(x) = 0} = {(0, 0)}, by which C ∩ D0 = ∅ (i.e., there are no singular

points that satisfy the constraint C). We now determine the set S of regular points. We
consider the Lagrangean function L : R3 → R with

L(x, λ) = f (x) + λ(b − g(x)) = x2 − x21 + λ(4 − x21 − x22 )

30 CHAPTER 4. IMPLICIT FUNCTIONS AND CONSTRAINED OPTIMIZATION

The gradient of L is

∇L(x) = −2x1 − 2x1 λ 1 − 2λx2 4 − x21 − x22

 

and from the stationary point condition ∇L(x) = 0 we find

√ √
R = {(0, −2, −1/4), (0, 2, 1/4), (− 15/2, −1/2, −1), ( 15/2, −1/2, −1)}

Omitting the values of the multiplier λ we have:

√ √
S = {(0, −2), (0, 2), (− 15/2, −1/2), ( 15/2, −1/2)}

The local solutions to the constrained optimization problem, if they exist, belong to the set

S ∪ (C ∩ D0 ) ∪ (C − D) = S ∪ ∅ ∪ ∅ = S ⊆ C

We observe that C = {x ∈ R2 : x21 + x22 = 4} is a compact (closed and bounded) subset of

R2 and f is continuous on C, so we can apply Weierstrass’s theorem to f on C. Calculating
the value of f for every point in S we obtain
√ √
f (S) = {f (0, −2), f (0, 2), f (− 15/2, −1/2), f ( 15/2, −1/2)} = {−2, 2, −17/4, −17/4}

so the point (0, 2) is a point of constrained maximum of f in C, with the constrained

maximum value of 2.

4.8. We have R = {(1, 0, 1/2)} and S = {(1, 0)}. The point (1, 0) is a point of constrained
maximum of f in C, with a constrained maximum value of 0.

4.9. The function f is continuous and coercive on C. We have

p p
R = {(− 3/2, 1/2, −1), ( 3/2, 1/2, −1), (0, 2, −4)}

and p p
S = {(− 3/2, 1/2), ( 3/2, 1/2), (0, 2)}
p p
The points (− 3/2, 1/2) and ( 3/2, 1/2) are points of constrained maximum of f in C,
both with a constrained maximum value of 13/4.

4.10. The function f is continuous and C is a compact subset of R2 . We have

R = {(0, −1, −1), (0, 1, −1), (−1, 0, 1), (1, 0, 1)}

and
S = {(0, −1), (0, 1), (−1, 0), (1, 0)}
The points (1, 0) and (−1, 0) are points of constrained maximum of f in C, both with a
constrained maximum value of 0.
Chapter 5

The Riemann integral

5.1 Exercises
5.1. Let f (x) = x for x ∈ [0, 1] and let πn be a subdivision of [0, 1] into n equal parts.

(a) Calculate I(f, πn ) and S(f, πn ).

(b) Show that when n → +∞,

1
lim I(f, πn ) = lim S(f, πn ) = .
2

(c) Using the definition, is f Riemann-integrable?

1
5.2. Let f (x) = x for x ∈ [1, 2] and let πn be a subdivision of [1, 2] into n equal parts.

(a) Calculate I(f, π4 ) and S(f, π4 ).

(b) Find I(f, πn ) and S(f, πn ) in terms of n.
(c) Show that when n → +∞,

lim I(f, πn ) = lim S(f, πn ).

(d) Is f Riemann-integrable on [1, 2]?

3x+1
5.3. Let f (x) = 2x+1 for x ∈ [0, 2] and let πn be a subdivision of [0, 2] into n equal parts.

(a) Find the points of the subdivision πn .

(b) Calculate S(f, πn ) − I(f, πn ).
(c) Show that when n → +∞, lim S(f, πn ) − I(f, πn ) = 0. Hence, verify that f is
Riemann-integrable on [0, 2].
√
5.4. Let f (x) = x for x ∈ [1, 4] and let πn be a subdivision of [1, 4] into n equal parts.

(a) Find the points of the subdivision πn .

(b) Calculate S(f, πn ) − I(f, πn ).
(c) Show that when n → +∞, lim S(f, πn ) − I(f, πn ) = 0. Hence, verify that f is
Riemann-integrable on [1, 4].

31
32 CHAPTER 5. THE RIEMANN INTEGRAL

5.5. Let f : [−k, k] → R be a function that is integrable on [−k, k]. Using the definition of
integrable functions, prove each of the following statements.
(a) If π is a subdivision of [−k, k], then there exists a refinement of π, π̃, that contains
0 and is symmetric, in the sense that x ∈ π̃ ⇐⇒ −x ∈ π̃; this allows us to write
π̃ = π ′ ∪ −π ′ where π ′ is a subdivision of [0, k] and −π ′ = {−x : x ∈ π} .
Rk
(b) If f is an odd function, then −k f (x)dx = 0.
Rk Rk
(c) If f is an even function, then −k f (x)dx = 2 0 f (x)dx.
Rb
5.6. Let f, g : [a, b] → R be continuous on [a, b]. Prove that a f (x)g(x)dx cannot be found
Rb Rb
given just the integrals a f (x)dx and a g(x)dx. Hint: use two simple functions f, g, where
Rb
g depends on a parameter k such that a f (x)g(x)dx varies as a function of k.
5.7. Let f : [a, b] → R be continuous on [a, b].
(a) Prove, without using the mean value theorem of integral calculus, that if f is a positive
function and its integral is equal to zero, that f must be equal to the zero function.
(b) Prove that the theorem of part (a) does not hold if f is not positive.
5.8. (a) Give the definition of integral mean value.
(b) Denote with χA the indicator function of a set A ⊆ R:
(
1 x∈A
χA (x) =
0 x∈/A
Consider the function f : [0, 4] → R defined as:
f = χ[0,1) + 2χ[1,2) + 3χ[2,3) + 4χ[3,4]
Calculate its integral mean value λ.
(c) Does there exist a value c ∈ [0, 4] such that f (c) = λ?
5.9. (a) Let (
x x ∈ [0, 1)
f (x) =
x − 1 x ∈ [1, 2]
R2
Calculate 0 f (x)dx and the integral mean value of f without using antiderivatives.
Rn
(b) Again without using antiderivatives, calculate 0 f (x)dx and the integral mean value
of the ’sawtooth’ function,
(
x−k x ∈ [k, k + 1) ∀k = 0, 1, 2, . . . , n − 2
f (x) =
x − (n − 1) x ∈ [n − 1, n]

5.10. Let f, g : [a, b] → R be two integrable functions.

Rb
(a) Express a (f (x) + λg(x))2 dx as a polynomial in λ, P (λ).
(b) Confirm that P (λ) is a polynomial of degree 2 and that P (λ) ≥ 0 for every λ.
(c) Using the necessary condition of the discriminant ∆ of P (λ), show that
Z b 2 Z b Z b
f (x)g(x)dx ≤ f 2 (x)dx g 2 (x)dx.
a a a
5.2. SOLUTIONS 33

5.2 Solutions
i i+1

5.1. (a) Let i = 0, 1, ..., n − 1. In the interval n, n , we have that

i i+1
inf f (x) = , sup f (x) = .
i i+1
x∈[ n , n ] n x∈[ ni , i+1 n
n ]

So, we obtain

n−1 n−1
X i1 1 X 1 (n − 1)n 1n−1
I(f, πn ) = = 2 i= 2 = ,
nn n n 2 2 n
i=0 i=0

n−1 n n
X i+11 X i 1 1 X 1 n(n + 1) 1n+1
S(f, πn ) = = = 2 i= 2 = .
n n nn n n 2 2 n
i=0 i=1 i=1

(b) For n → +∞, we have that

1 n−1 1
lim I(f, πn ) = lim = ,
2 n 2

1 n+1 1
lim S(f, πn ) = lim = .
2 n 2
(c) We have
Z 1
1
f (x)dx = sup I(f, π) ≥
0 π∈Π 2

and
Z 1
1
f (x)dx = inf S(f, π) ≤ ,
0 π∈Π 2
so it holds that
Z 1 Z 1
1
f (x)dx = f (x)dx = .
0 0 2

Therefore, f is integrable on [0, 1] according to the definition.

5.2. (a) Let i = 0, 1, 2, 3. In the interval 1 + 4i , 1 + i+1

 
4 we have

4 4
inf f (x) = , sup f (x) = .
x∈[1+ 4i ,1+ i+1 ] 5+i x∈[1+ 4i ,1+ i+1 4+i
4 4 ]

So, we obtain
 
1 4 4 4 4 1 1 1 1
I(f, πn ) = + + + = + + + ,
4 5 6 7 8 5 6 7 8
 
1 4 4 4 4 1 1 1 1
S(f, πn ) == + + + = + + + .
4 4 5 6 7 4 5 6 7
34 CHAPTER 5. THE RIEMANN INTEGRAL

(b) Now let i = 0, 1, ..., n − 1. In the interval 1 + ni , 1 + i+1

 
n we have
n n
inf f (x) = , sup f (x) = .
x∈[1+ ni ,1+ i+1
n ]
n+i+1 x∈[1+ ni ,1+ i+1 n+i
n ]

So, we obtain
n−1 n−1 n
X n 1 X 1 X 1
I(f, πn ) = = = ,
n+i+1n n+i+1 n+i
i=0 i=0 i=1

n−1 n−1
X n 1 X 1
S(f, πn ) = = .
n+in n+i
i=0 i=0

(c) To show that when n → +∞,

lim I(f, πn ) = lim S(f, πn ),

we show that  
lim S(f, πn ) − I(f, πn ) = 0.
In fact:
n−1 n
X 1 X 1 1 1 1
S(f, πn ) − I(f, πn ) = − = − = → 0.
n+i n+i n 2n 2n
i=0 i=1

(d) We use a criterion of integrability. We have seen that the subdivision into n equal
1
parts satisfies S(f, πn ) − I(f, πn ) = 2n → 0. Thus, for every ϵ > 0, there exists a
subdivision π such that S(f, π) − I(f, π) < ϵ. It suffices to consider a subdivision into
n equal parts with n > 1/2ϵ. Therefore f is Riemann-integrable on [1, 2].

5.3. (a) The points of the subdivision of [0, 2] into n equal parts are

2 4 2i 2(n − 1) 2n
0, , , . . . , , . . . , , = 2.
n n n n n

(b) The function f (x) = 3x+12x+1 is increasing on [0, 2]. Consequently, in the interval
h i
2i 2(i+1)
n, n we have that
   
2i 2(i + 1)
inf i f (x) =f , sup f (x) = f .
x∈
h
2i 2(i+1)
, n n h
2i 2(i+1)
i n
n x∈ n
, n

It follow that
 
1 1 7 1 2
S(f, πn ) − I(f, πn ) = f (2) − f (0) = −1 = → 0.
n n 5 n 5n

(c) We have seen that the subdivision into n equal parts satisfies S(f, πn ) − I(f, πn ) =
2
5n → 0. So, for every ϵ > 0, there exists a subdivision π such that S(f, π) − I(f, π) < ϵ.
It suffices to consider a subdivision into n equal parts with n > 2/5ϵ. Therefore, f is
Riemann-integrable on [0, 2].
5.2. SOLUTIONS 35

5.4. (a) The points of the subdivision of [1, 4] into n equal parts are

3 6 3i 3(n − 1) 3n
1, 1 + ,1 + ,...,1 + ,...,1 + ,1 + = 4.
n n n n n
√
(b) The function f (x) = x for x ∈ [1, 4] is monotonically increasing, so
r r
3i 1 + 3(i + 1)
inf i f (x) = 1+ , sup f (x) = .
h
x∈ 1+ 3i ,1+
3(i+1) n h
3i 1+3(i+1)
i n
n n x∈ 1+ n , n

It follows that
√ 1 √ 1 2 1 1
S(f, πn ) − I(f, πn ) = 4 − 1 = − = .
n n n n n

(c) Here again, the subdivision into n equal parts satisfies S(f, πn ) − I(f, πn ) = n1 → 0.
So, for every ϵ > 0, there exists a subdivision π such that S(f, π) − I(f, π) < ϵ. It
suffices to consider a subdivision into n equal parts with n > 1/ϵ. Therefore, f is
Riemann-integrable on [1, 4].
Note: The procedure outlined in the above examples can be generalized for every monotonic
function, and leads to the proof of the integrability of monotonic functions.

5.5. (a) Given π we consider π̃ = π ∪ {0} ∪ −π: it contains the points of π, 0, and the points
of π with opposite signs. Clearly π ⊆ π̃.
It remains to define a subdivision π ′ of [0, k] such that π̃ = π ′ ∪ −π ′ . Clearly, we take
π ′ = π̃ ∩ [0, k].
(b) Let f be an odd function. Since f is integrable on [−k, k], then for every ϵ there exists
a subdivision π of [−k, k] such that S(f, π) − I(f, π) < ϵ. We take the corresponding
π̃: this is a refinement of π, so it holds that S(f, π̃) − I(f, π̃) < ϵ.
But f is an odd function – taking the corresponding subdivision π ′ of [0, k], we have

I(f, π̃) = I(f, π ′ ) + I(f, −π ′ ) = I(f, π ′ ) − S(f, π ′ ),

S(f, π̃) = S(f, π ′ ) + S(f, −π ′ ) = S(f, π ′ ) − I(f, π ′ ).

and thus
ϵ > S(f, π̃) − I(f, π̃) = 2S(f, π ′ ) − 2I(f, π ′ ),
from which S(f, π ′ ) − I(f, π ′ ) < 2ϵ and f is integrable on [0, k].
We now pass to the integral on [−k, k]. As we have seen above, since π̃ is symmetric
and f is odd, we have that S(f, −π ′ ) = −I(f, π ′ ), and thus
ϵ
S(f, π̃) = S(f, π ′ ) + S(f, −π ′ ) = S(f, π ′ ) − I(f, π ′ ) < .
2
Analogously,
ϵ
I(f, π̃) = I(f, π ′ ) + I(f, −π ′ ) = I(f, π ′ ) − S(f, π ′ ) > − .
2
So for every ϵ > 0 there exists a partition π̃ such that
ϵ ϵ
− < I(f, π̃) ≤ S(f, π̃) < ,
2 2
36 CHAPTER 5. THE RIEMANN INTEGRAL

Rk
which implies that −k f (x)dx = 0.
Alternatively, we can reason using the upper and lower bounds over the partitions π ′
of [0, k]. From

S(f, π̃) = S(f, π ′ ) + S(f, −π ′ ) = S(f, π ′ ) − I(f, π ′ )

we obtain

inf S(f, π̃) = inf′ S(f, π ′ ) + inf′ S(f, −π ′ ) = inf′ S(f, π ′ ) − sup I(f, π ′ )
π̃ π π π π′

and therefore Z k Z k Z k
f (x)dx = f (x)dx − f (x)dx = 0.
−k 0 0

(c) Let f be an even function. We proceed as above. Since f is even, we have that
I(f, −π) = I(f, π) and S(f, −π) = S(f, π). So, I(f, π̃) = 2I(f, π ′ ) and S(f, π̃) =
Rk
2S(f, π ′ ). Therefore, f is integrable on [0, k] and we have that −k f (x)dx =
Rk
2 0 f (x)dx.

5.6. We recall that the set indicator function for A ⊆ R is the function:
(
1 x∈A
χA (x) =
0 x∈ /A

Now consider f, g : [0, 10] → R such that f = χ[0,1] and g = χ[k,k+1] with k ∈ [0, 9]. We
have Z 10 Z 10
f (x)dx = g(x)dx = 1,
0 0

but also f (x)g(x) = χ[0,1]∩[k,k+1] , from which


Z 10 1
 k=0
f (x)g(x)dx = 1 − k k ∈ (0, 1)
0 
0 k ∈ [1, 9]


5.7. (a) Suppose, by contradiction, that there exists an x0 ∈ [a, b] such that f (x0 ) ̸= 0.
Then f (x0 ) > 0. Since f is continuous, for ϵ = f (x0 )/2 there exists an interval
I = (x0 − δ, x0 + δ) where f (x) ∈ Bϵ (f (x0 )) and therefore f (x) > f (x0 )/2 > 0 for
every x ∈ I. But then
Z b Z x0 −δ Z x0 +δ Z b
f (x)dx = f (x)dx + f (x)dx + f (x)dx ≥ 0 + 2δϵ + 0 > 0,
a a x0 −δ x0 +δ

which is absurd.
(b) It sufficed to consider an odd function
R1 f : [−1, 1] → R, for example, f (x) = sin(x).
As we have seen in exercise 5.4, −1 sin(x)dx = 0, but sin(x) is not the zero function.

5.8. (a) See the textbook.

5.2. SOLUTIONS 37

(b) Using the linearity of the Riemann integral with respect to the integrand, we have:
Z 4 Z 4 Z 4 Z 4 Z 4
f (x)dx = χ[0,1) dx+2 χ[1,2) dx+3 χ[2,3) dx+4 χ[3,4] dx = 1+2+3+4 = 10
0 0 0 0 0

So the integral mean value of f on [0, 4] is

10 5
λ= = .
4 2
(c) The function f only takes integer values, therefore there cannot exist a value c ∈ [0, 4]
such that f (c) = λ.
R1
5.9. (a) We have already seen that 0 xdx = 12 . By the additive property of the integral with
respect to the interval of integration we have
Z 2 Z 1 Z 2 Z 1
1 1 1
f (x)dx = xdx + (x − 1)dx = + xdx = + = 1.
0 0 1 2 0 2 2

So the integral mean value is λ = 21 .

(b) Analogously to what we have seen above, we have
Z n n−1
X Z k+1 n−1
XZ 1 n−1
X 1 n
f (x)dx = (x − k)dx = xdx = = .
0 k 0 2 2
k=0 k=0 k=0

Thus, the integral mean value is always λ = 12 .

5.10. Let f, g : [a, b] → R be two integrable functions.

(a) By the property of the linearity of the integral

Z b Z b
2
f 2 (x) + 2λf (x)g(x) + λ2 g 2 (x) dx


(f (x) + λg(x)) dx =
a a
Z b Z b Z b
2 2
= f (x)dx + 2λ f (x)g(x)dx + λ g 2 (x)dx
a a a

The polynomial P (λ) is

Z b Z b Z b
P (λ) = f 2 (x)dx + 2λ f (x)g(x)dx + λ2 g 2 (x)dx,
a a a

(b) The polynomial P (λ) given in part (a) is a second degree polynomial in λ. Fur-
ther, P (λ) ≥ 0 for every λ since it is the integral of a positive function, i.e.,
Rb 2
a (f (x) + λg(x)) dx.
(c) The reduced discriminant ∆/4 of the polynomial is
Z b 2 Z b Z b
∆
= f (x)g(x)dx − f 2 (x)dx g 2 (x)dx.
4 a a a

Since P (λ) ≥ 0 for every λ , it must be true that ∆/4 ≤ 0, that is

Z b 2 Z b Z b
f (x)g(x)dx − f 2 (x)dx g 2 (x)dx ≤ 0,
a a a
38 CHAPTER 5. THE RIEMANN INTEGRAL

from which 2
Z b Z b Z b
f (x)g(x)dx ≤ f 2 (x)dx g 2 (x)dx.
a a a
The above is called the Cauchy-Schwarz inequality, and it generalizes the one we have
already seen concerning the scalar product and the norm of two vectors in Rn .
Chapter 6

Integral calculus

6.1 Exercises
6.1. Calculate the following integrals:
Z 4
(a) (x − 1)4 dx
2
Z 3√
(b) x + 1dx
0
Z 6 1
(c) (x + 3)− 2 dx
1

6.2. Calculate the following integrals:

Z 2
(a) (x2 − 2)2 dx
0
Z 1
(b) (1 + x)(1 + 2x)dx
−1
Z 3
1
(c) dx
1 x+1
Z 3
3x
(d) dx
1 x+1

6.3. For a function F : R → R we are given that F ′ (x) = x + 1 and F (0) = 2.

(a) Calculate F (4).

(b) Calculate the analytic expression of the function F .

6.4. Given the relation: Z x

4 2
x +x = f (t)dt.
0

(a) Calculate f (2).

(b) Calculate the analytic expression of the function f .

39
40 CHAPTER 6. INTEGRAL CALCULUS

6.5. (a) Let Z x

2
F (x) = e−t dt.
0
Calculate F ′ (x).
(b) Let
Z 2x+3
2
F (x) = e−t dt.
0
Calculate F ′ (x).

6.6. Integrating by parts, calculate:

Z
(a) sin2 x dx
Z
(b) ex sin x dx

6.7. Integrating by parts, calculate:

Z
(a) x sin 3x dx
Z
(b) xe2x dx
√
Z
(c) x ln x dx
Z
(c) (ln x)2 dx

6.8. Calculate the following integrals of rational functions:

Z
x+3
(a) 2
dx on the interval (2, +∞)
x −x−2
x2
Z
(b) dx on the interval (1, 2)
x2 − 3x + 2
Z
x+1
(c) 2
dx on the interval (−∞, 2)
x − 4x + 4
Z
1
(d) dx
a2 + x2
6.9. Using an appropriate substitution, calculate:
Z
arctan x
(a) dx
1 + x2
Z
1
(b) dx on the interval (1, +∞)
x ln x
Z
1
(c) dx
ex + e−x
Z −1
e x
(d) dx on the interval (0, +∞)
x2
6.2. SOLUTIONS 41

6.10. Using an appropriate substitution, calculate:

Z
x
(a) √ dx
Z x+1
1
(b) 2x x
dx
Z e +e
√
(c) x ex dx

6.2 Solutions
1
6.1. (a) An antiderivative of (x − 1)4 is (x − 1)5 , so
5
Z 4
1 4 1  5  243 − 1 242
(x − 1)4 dx = (x − 1)5 2 = 3 −1 = = .
2 5 5 5 5
√ 2 3
(b) An antiderivative of x + 1 is
(x + 1) 2 , so
3
√
Z 3
2h 3 3
i 2 h 3 i4 2  3  14
x + 1dx = (x + 1) 2 = x2 = 2 −1 = .
0 3 0 3 1 3 3
1 1
(c) An antiderivative of (x + 3)− 2 is 2(x + 3) 2 , so
Z 6
1 6
1
h i h 1 i9
(x + 3)− 2 dx = 2 (x + 3) 2 = 2 x 2 = 2 [3 − 2] = 2.
1 1 4

6.2. We have:
Z 2 Z 2  5 2 
4x3

2 2 4 2 x 32 32
(a) (x − 2) dx = (x − 4x + 4)dx = − + 4x = − +8 −0=
0 0 5 3 0 5 3
64 56
− +8= .
15 15
Z 1 Z 1  1  
3 2 3 2
(b) (1 + x)(1 + 2x)dx = (1 + 3x + 2x2 )dx = x + x2 + x3 = 1+ + −
−1 −1 2 3 −1 2 3
 
3 2 4 10
−1 + − =2+ = .
2 3 3 3
1
(c) For x > −1, an antiderivative of = (x + 1)−1 is ln(x + 1).
Z 3 x+1
1
So, dx = [ln(x + 1)]31 = ln 4 − ln 2 = 2 ln 2 − ln 2 = ln 2.
1 x + 1
3x 3(x + 1) − 3 1
(d) We have = =3−3 .
Z 3 x + 1 x + 1 x + 1
3x
So, dx = [3x − 3 ln(x + 1)]31 = 9 − 3 ln 4 − 3 + 3 ln 2 = 6 − 3(ln 4 − ln 2) =
1 x+1
6 − 3 ln 2.
6.3. (a) We have
4 4 4
t2
Z Z 
′
F (4) − F (0) = F (t)dt = (t + 1)dt = + t = (8 + 4) − 0 = 12.
0 0 2 0
So, F (4) = 12 + F (0) = 14.
42 CHAPTER 6. INTEGRAL CALCULUS

(b) We have
x x x
t2 x2
Z Z 
F (x) − F (0) = F ′ (t)dt = (t + 1)dt = +t = + x.
0 0 2 0 2

x2
So, F (x) = + x + 2.
2
6.4. (a) By the second fundamental
Z x theorem of integral calculus, we know that the integral
function F (x) = f (t)dt is equal to x4 + x2 , and that its derivative with respect to
0
x is f (x). So, f (x) = F ′ (x) = 4x3 + 2x and f (2) = 36.
(b) The analytic expression of the function f is f (x) = 4x3 + 2x.
Z x
2 2
6.5. (a) Given that F (x) = e−t dt, we have that F ′ (x) = e−x .
0
Z x
2 2
(b) Let G(x) = e−t dt. By the preceding part, we have that G′ (x) = e−x . Now, we
0
have that F (x) = G(2x + 3), so
2
F ′ (x) = 2G′ (2x + 3) = 2e−(2x+3) .

6.6. (a) We set f ′ (x) = g(x) ′

Z = sin x. Then f (x) = − cosZx and g (x) = cos x and the integration
by parts formula f ′ (x)g(x)dx = f (x)g(x) − f (x)g ′ (x)dx gives us

Z Z Z
2 2
1 − sin2 x dx


sin x dx = − cos x sin x + cos x dx = − cos x sin x +

Thus, Z Z
sin2 x dx = − cos x sin x + x − sin2 x dx

and we obtain Z
2 sin2 x dx = − cos x sin x + x + c,

yielding Z
1 1
sin2 x dx = − cos x sin x + x + c
2 2
(b) Using integration by parts,
Z Z Z
ex sin x dx = ex sin x − ex cos x dx = ex sin x − ex cos x − ex sin x dx,

and we obtain Z
1
ex sin x dx = ex (sin x − cos x) + c.
2

6.7. (a) Z Z
1 1 1 1
x sin 3x dx = − x cos 3x + cos 3x dx = − x cos 3x + sin 3x + c
3 3 3 9
6.2. SOLUTIONS 43

(b) Z Z
2x 1 1 1
xe dx = xe2x − e2x dx = e2x (2x − 1) + c
2 2 4
(c)
√
Z Z
2 3 2 1 2 3 4 3
x ln x dx = x 2 ln x − x 2 dx = x 2 ln x − x 2 + c
3 3 3 9
(d) Using integration by parts,
Z Z Z Z
2 2 2 1 2
(ln x) dx = 1 · (ln x) dx = x(ln x) − x · 2 ln x dx = x(ln x) − 2 ln x dx.
x
Now, repeating the procedure for the integral of ln x,
Z Z
1
ln x dx = x ln x − x dx = x ln x − x + c,
x
Thus, Z
(ln x)2 dx = x(ln x)2 − 2x ln x + 2x + c

6.8. (a) Observe that the denominator x2 − x − 2 can be factored into two distinct linear
terms, (x + 1)(x − 2). So we are searching for two real numbers A, B such that
x+3 A B
= + .
x2 −x−2 x−2 x+1
The equation becomes
x+3 A(x + 1) + B(x − 2) (A + B)x + (A − 2B)
= = .
x2 −x−2 (x − 2)(x + 1) (x − 2)(x + 1)
By first equating the numerators and then equating the corresponding coefficients
we obtain A + B = 1 and A − 2B = 3, from which B = 1 − A and 3A − 2 = 3, thus
A = 53 and B = − 23 . We obtain
Z Z Z
x+3 5 1 2 1 5 2
dx = dx − dx = ln(x − 2) − ln(x + 1) + c
x2 − x − 2 3 x−2 3 x+1 3 3
(b) In this case the degree of the numerator is greater than or equal to the degree of the
denominator. We write x2 = (x2 − 3x + 2) + (3x − 2) and thus
x2 3x − 2 3x − 2
Z Z Z Z
2
dx = 1 dx + 2
dx = x + 2
dx.
x − 3x + 2 x − 3x + 2 x − 3x + 2
We can now proceed as in part (a): x2 − 3x + 2 = (x − 1)(x − 2) and we are therefore
searching for A, B ∈ R such that
3x − 2 A B
= + .
− 3x + 2x2 x−1 x−2
We obtain A = −1 and B = 4. Thus,
3x − 2
Z Z Z
1 1
2
dx = − dx + 4 dx = − ln(x − 1) + 4 ln(2 − x) + c,
x − 3x + 2 x−1 x−2
1
since on the interval (1, 2) an antiderivative of x−2 is ln(2 − x) and not ln(x − 2).
To recap, on the interval (1, 2) we have
x2
Z
dx = x − ln(x − 1) + 4 ln(2 − x) + c
x2 − 3x + 2
44 CHAPTER 6. INTEGRAL CALCULUS

(c) Since x2 − 4x + 4 = (x − 2)2 , we have a repeated linear factor and we write

x+1 A B
= + ,
x2 − 4x + 4 x − 2 (x − 2)2
obtaining A = 1 and B = 3. Thus,
Z Z Z
x+1 1 1 3
2
dx = dx + 3 2
dx = ln(2 − x) − +c
x − 4x + 4 x−2 (x − 2) x−2
1
since on the interval (−∞, 2) an antiderivative of x−2 is ln(2 − x) and not ln(x − 2).
(d) Z Z
1 1 1
dx = 2 dx.
2
a +x 2 a 1 + ( xa )2
x
By setting y = , we can rewrite
a
Z Z
1 1 1 x
x 2 dx = a 2
dx = a arctan + c
1 + (a) 1+y a a
and therefore Z
1 1 x
dx = arctan + c
a2 + x2 a a
1
6.9. (a) We let y = arctan x and observe that y ′ = . Thus,
1 + x2
Z Z Z
arctan x ′ 1 1
dx = y y dx = y dy = y 2 + c = arctan2 x + c
1 + x2 2 2
1
(b) We are on the interval (1, +∞). We let y = ln x and observe that y ′ = .
x
Thus, Z Z ′ Z
1 y 1
dx = dx = dy = ln y + c = ln ln x + c
x ln x y y
(c) We let y = ex and observe that y ′ = ex . Thus,
Z Z Z Z
1 1 y 1
x −x
dx = 1 dx = 2
dx = 2
dy =
e +e y+y y +1 y +1

= arctan y + c = arctan ex + c
1 1 1
(d) We are on the interval (0, +∞). We let y = e− x and observe that y ′ = 2 e− x .
x
Thus,
Z −1 Z Z
e x ′ 1
dx = y dx = dy = y + c = e− x + c
x2
√
6.10. (a) Let y = x + 1 and observe that x = y 2 − 1 and x′ = 2y. Thus,
Z 2
y −1
Z Z
x 2
√ dx = 2ydy = 2 (y 2 − 1) dy = y 3 − 2y + c =
x+1 y 3
2√ √
= x + 1(x + 1) − 2 x + 1 + c
3
6.2. SOLUTIONS 45

1
(b) Let y = ex and observe that x = ln y and x′ = . Thus,
y
Z Z Z
1 1 1 1
dx = dy = dy.
e2x + ex y2 + y y y 2 (y + 1)

We now proceed as we normally do for a rational function with denominator of degree

at least 2. So, we must determine if there exist A, B, C ∈ R such that
1 A B + Cy
= +
y 2 (y + 1) y+1 y2

Combining,
1 Ay 2 + (B + Cy)(y + 1)
= .
y 2 (y + 1) y 2 (y + 1)
Equating the numerators we obtain 1 = Ay 2 + (B + Cy)(y + 1), from which it
follows immediately that A + C = 0, or C = −A. The equation then becomes
1 = Ay 2 + (B − Ay)(y + 1) = By + B − Ay, from which we obtain that B = 1 and
A = 1. Thus,
1 1 1−y
2
= +
y (y + 1) y+1 y2
and we have:
1−y
Z Z Z Z
1 1 1
dx = dy = dy + dy
e + ex
2x 2
y (y + 1) y+1 y2
Z Z Z
1 1 1
= dy + 2
dy − dy
y+1 y y
1
= ln (y + 1) − − ln y + c
y
= ln (e + 1) − e−x − x + c
x

(recall that y = ex > 0).

√ 2
(c) Let y = ex . Then x = ln (y 2 ) = 2 ln y and x′ = . We obtain:
y
√
Z Z Z
2
x ex dx = 2 ln y y dy = 4 ln y dy = 4y(ln y − 1) + c
y
√ 1 √
= 4 ex ( x − 1) + c = 2 ex (x − 2) + c
2
Note
√ that this exercise can be resolved using integration by parts, observing that
x
ex = e 2 .
Chapter 7

Improper integrals

7.1 Exercises
Z +∞
2x
7.1. Calculate dx.
0 x2 + 1
Z +∞
7.2. Calculate e−x dx.
1
Z +∞
7.3. Calculate sin x dx.
0
Z +∞
1
7.4. Calculate dx.
2 x ln2 x
Z +∞
1
7.5. Calculate dx.
0 (x + 3)2
7.6. Calculate the following improper integrals:
Z +∞
2x
(a) dx
1 (1 + x2 )2
Z +∞
2x
(b) 2 2
dx
−∞ (1 + x )

7.7. Calculate the following improper integrals:

Z +∞
1
(a) 2 + 2x + 2
dx
0 x
Z +∞
1
(b) 2 + 2x + 2
dx
−∞ x
7.8. Calculate the following Cauchy principal values:
Z +∞
2x
(a) P V 2
dx
−∞ x + 1
Z +∞
(b) P V sin xdx
−∞

47
48 CHAPTER 7. IMPROPER INTEGRALS
Z +∞
1
(c) P V dx
−∞ x2 + 2x + 2
7.9. Determine if the following improper integrals converge:
Z +∞ 2
ln(1 + x4 ) + x 3
(a) dx
−∞ x2 + 1
Z +∞ −x
e
(b) 1 dx
2 sin x
Z +∞
cos x
(c) 4 1 − cos 1

dx
2 x x

7.10. Determine if the following improper integrals converge:

ln(1 + x12 )
Z +∞
(a) dx
2 sin x1
Z +∞
1
(b) dx
2 x lnβ x
α

7.2 Solutions
7.1. Setting y = x2 + 1, we obtain y ′ = 2x.
We have: Z Z ′ Z
2x y 1
2
dx = dx = dy = ln y + c = ln(x2 + 1) + c
x +1 y y
and therefore:
Z +∞ Z k
2x 2x k
dx = lim ln (x2 + 1) 0 = lim ln (k 2 + 1) = +∞.

2
dx = lim 2
0 x +1 k→+∞ 0 x + 1 k→+∞ k→+∞

7.2. We have:
Z +∞ Z k
−x
k 1
e−x dx = lim −e−x = lim (−e−k + e−1 ) = .

e dx = lim 1
1 k→+∞ 1 k→+∞ k→+∞ e
7.3. We have: Z +∞
sin xdx = lim [− cos x]k0 = lim (− cos k + 1).
0 k→+∞ k→+∞
Since the above limit does not exist, the integral in question is undefined.
7.4. We have:
1 k
Z +∞    
1 1 1 1 1
lim −
2 dx = k→+∞ = lim − + =0+ = .
2 x ln x ln x 2 k→+∞ ln k ln 2 ln 2 ln 2

7.5. Setting y = x + 3, we obtain y ′ = 1.

We have:
Z Z Z
1 1 1
2 dx = y y dx = y −2 dy = − + c = −
−2 ′
+ c.
(x + 3) y x+3
and therefore:
Z +∞  k  
1 1 1 1 1 1
lim −
2 dx = k→+∞ = lim − + =0+ = .
0 (x + 3) x + 3 0 k→+∞ k+3 3 3 3
7.2. SOLUTIONS 49

7.6. We have: Z
2x 1
dx = − +c
(1 + x2 )2 x2 + 1
and therefore:

(a)
Z +∞  k
2x 1 1
2 2
dx = lim − 2 = .
1 (1 + x ) k→+∞ x +1 1 2
(b)
Z +∞ Z 0 Z +∞
2x 2x 2x
dx = dx + dx
−∞ (1 + x2 )2 2
−∞ (1 + x )
2
0 (1 + x2 )2
 0  k
1 1
= lim − 2 + lim − 2
k→−∞ x + 1 k k→+∞ x +1 0
   
1 1
= lim −1 + 2 + lim − 2 +1
k→−∞ k +1 k→+∞ k +1
= −1 + 1 = 0

We note that the improper integral over R exists since the integrals on (−∞, 0] and
[0, +∞) exist and are finite.

7.7. We have: Z Z
1 1
2
dx = dx = arctan(x + 1) + c
x + 2x + 2 (x + 1)2 + 1
and therefore:

(a)
Z +∞
1
dx = lim [arctan(x + 1)]k0
0 x2 + 2x + 2 k→+∞
π π π
= lim [arctan(k + 1) − arctan(1)] = − =
k→+∞ 2 4 4

(b)
Z +∞
1
dx = lim [arctan(x + 1)]0k + lim [arctan(x + 1)]k0
−∞ x2 + 2x + 2 k→−∞ k→+∞
π
= lim [arctan(1) − arctan(k + 1)] +
k→−∞ 4
hπ π i π
= + + =π
4 2 4
7.8. We have already computed the necessary antiderivatives.
The Cauchy principal values are:

(a)
Z +∞ Z k
2x 2x  2
k
PV dx = lim dx = lim ln(x + 1) −k
−∞ x2 + 1 k→+∞ 2
−k x + 1 k→+∞

= lim ln(k 2 + 1) − ln(k 2 + 1) = lim (0) = 0

 
k→+∞ k→+∞
50 CHAPTER 7. IMPROPER INTEGRALS

(b)
Z +∞
PV sin xdx = lim [− cos x]k−k = lim [− cos(k) + cos(−k)]
−∞ k→+∞ k→+∞

= lim [− cos(k) + cos(k)] = lim (0) = 0

k→+∞ k→+∞

(c)
Z +∞
1
PV dx = lim [arctan(x + 1)]k−k
−∞ x2 + 2x + 2 k→+∞
π π
= lim arctan(k + 1) − lim arctan(k + 1) = + =π
k→+∞ k→−∞ 2 2

The first two cases illustrate that the Cauchy principal value may exist even when the
corresponding bilateral improper integral over R does not (for example, in the case of an
odd integrable function).
The third case illustrates that if the bilateral improper integral exists - i.e., the one-sided
improper integrals on (−∞, c] and [c, +∞) each exist - then the Cauchy principal value
exists and is equal to the value of the integral over R.

7.9. To determine if the following improper integrals converge we use the integrability criteria, i.e.,
the asymptotic comparison criterion, the comparison criterion and the absolute convergence
criterion.

(a) We observe that

2 2
+∞ +∞
ln(1 + x4 ) + x 3 ln(1 + x4 ) + x 3
Z Z
dx = 2 dx,
−∞ x2 + 1 0 x2 + 1

since the integrand is an even function.

At this point we analyze the integrand. As x → +∞ we have
2 2
ln(1 + x4 ) + x 3 x3 1
∼ = 4.
x2 + 1 x2 x3
We know that the improper integral of x1α is convergent on (1, +∞) for α > 1, so
by the asymptotic comparison criterion it follows that also the improper integral in
question is convergent.
(b) We observe that the integrand is positive on (2, +∞) and as x → +∞ we have

e−x e−x x
1 ∼ 1 = ex ,
sin x x

whose improper integral is convergent on (2, +∞). By the asymptotic comparison

criterion it follows that the improper integral in question is convergent.
Z +∞
cos x
(c) To analyze
dx we proceed as follows.
2 x 1 − cos x1
4

First of all, we note that the integrand is a function that changes sign, so we
must use the absolute convergence criterion: the given integral will converge if
7.2. SOLUTIONS 51

Z +∞
cos x

dx is convergent.
2 1 − cos x1
x4
We next observe that for x ∈ (2, +∞) we have that 0 < cos x1 < 1 and

cos x | cos x| 1
1

= 1

≤
.
x4 cos x − 1 x4 1 − cos x x4 1 − cos x1

Z +∞ Z +∞
cos x 1
So, by the comparison criterion,
dx converges if
dx
2 x4 1 − cos x1 2 x4 1 − cos x1
is convergent.
Finally, we consider this last integral. As x → +∞ we have
1 1 1 2
1 = 4


=
∼ 2
x 1 − 1 − 2x12 + o( x12 ) 1 1


x4 1 − cos x x4 2x2
+ o( x2 ) x

and by the asymptotic comparison criterion this last integral is convergent.

Thus, it follows that the original integral in question is also convergent.

7.10. (a) The integrand is positive on (2, +∞). As we know, as x → +∞ we have

1 1
ln(1 + x2
) x2 1
1 ∼ 1 = ,
sin x x
x

whose integral on (2, +∞) diverges to +∞. Thus, the integral in question also diverges
to +∞.
(b) We need to analyze
Z +∞
1
dx.
2 xα lnβ x

Let α > 1. For simplicity, we can consider the integral on (3, +∞), which has the
same behavior: in this case, for x > 3 it holds that ln x > 1, and it is also true that
lnβ x > 1 for every β. Consequently,
1 1
β
< α
xα ln x x
whose integral on (3, +∞) converges. It follows that the integral in question is
convergent.

Let 0 < α < 1. It holds that lnβ x = o(xk ) for every k > 0. Take a value k > 0 such
that k < 1 − α. Then xα lnβ x < xα+k at least eventually. So, for x greater than or
equal to some particular x0 ∈ (2, +∞) we will have that

1 1
> , with 0 < α + k < 1.
xα lnβ x xα+k
Since the improper integral of 1/xa diverges to +∞ for a < 1, it follows that the
integral in question also diverges to +∞.
1
For α = 0 the integrand reduces to lnβ x
, whose integral diverges to +∞ for every β.
52 CHAPTER 7. IMPROPER INTEGRALS

For α < 0 the integrand is unbounded so the integral diverges to +∞.

It remains to consider the case when α = 1. The integral becomes
Z +∞
1
dx.
2 x lnβ x

Suppose β ̸= 1. Using the substitution y = ln x, in a few steps we obtain:

Z +∞ Z +∞  k
1 1 −β 1
dx = ln x dx = lim ln−β+1 x .
2 x lnβ x 2 x k→+∞ −β + 1 2

Computing the limit, we find that if β > 1 the integral converges while if β < 1 the
integral diverges to +∞.
Finally, if β = 1, we have:
Z +∞
1
dx = lim [ln ln x]k2 = +∞.
2 x ln x k→+∞

To recap, the given integral converges when α > 1 and when α = 1, β > 1.
Chapter 8

The Stieltjes integral

8.1 Exercises
8.1. Let f (x) = 1 for x ∈ [0, 1] and let πn be a subdivision of [0, 1] into n equal parts. Consider
the integrator function g(x) = x2 .

(a) Calculate I(f, g, πn ) and S(f, g, πn ).

(b) Show that as n → +∞,

lim I(f, g, πn ) = lim S(f, g, πn ) = 1.

(c) Using the definition, Rcan you conclude that f is Stieltjes-integrable on [0, 1]? If so,
1
compute the integral 0 f dg.

8.2. Let f (x) = x for x ∈ [0, 1] and let πn be a subdivision of [0, 1] into n equal parts. Consider
the integrator function g(x) = 20x.

(a) Calculate I(f, g, πn ) and S(f, g, πn ).

(b) Show that as n → +∞,

lim I(f, g, πn ) = lim S(f, g, πn ) = 10.

(c) Using the definition, Rcan you conclude that f is Stieltjes-integrable on [0, 1]? If so,
1
compute the integral 0 f dg.

8.3. Let f (x) = x and (

20x x<1
g(x) = .
20 + 10(x − 1) x ≥ 1
Using the definition, determine if f is Stieltjes-integrable
R2 on [0, 2] with the integrator
function g. If it is, compute the integral 0 f dg. (Hint: Consider a subdivision πn that
subdivides [0, 1] into n equal parts and [1, 2] into another n equal parts and proceed as in
the previous exercises.)
Re
8.4. Determine if 1 xd(ln x) exists and if so, compute the integral.
R2 √
8.5. Determine if 1 (x2 + 4x)d( x) exists and if so, compute the integral.
R 20 x
8.6. Determine if 10 (5x + 11)d(1 − e− 10 ) exists and if so, compute the integral.

53
54 CHAPTER 8. THE STIELTJES INTEGRAL

R2 √
8.7. Determine if 0 (x2 + 4x)d( x) exists and if so, compute the integral.
R4 √
8.8. Determine if 0 (1 + x3 )d( x) exists and if so, compute the integral.
Re 1
8.9. Determine if 1 (x + ln x)d(ln x) exists and if so, compute the integral.
R4 1
8.10. Calculate 0 x+1 d(g(x)), where g(x) = [x] is the integer part of x, i.e., g(x) is the largest
integer y such that y ≤ x.
R3
8.11. Calculate 0 xd(g(x)), where

x x ≤ 1

g(x) = 5 1<x<2.

10 x ≥ 2


8.2 Solutions
i i+1

8.1. (a) Let i = 0, 1, ..., n − 1. On the interval n, n we have

inf f (x) = sup f (x) = 1.

x∈[ ni , i+1
n ] x∈[ ni , i+1
n ]

So, we obtain

n−1   2 ! n−1
2 X (i2 + 2i + 1) − i2
X i+1i
I(f, g, πn ) = S(f, πn ) = 1 − =
n n n2
i=0 i=0
n−1 n−1 n−1
! !
X 2i + 1 1 X 1 X
= = (2i + 1) = 2 i +1
n2 n2 n2
i=0 i=0 i=0
 
1 n(n − 1) n(n − 1) + 1
= 2 2 +1 = .
n 2 n2

(b) As n → +∞ we have

n(n − 1) + 1
lim I(f, g, πn ) = lim S(f, g, πn ) = lim = 1.
n2

(c) We have
Z 1 Z 1
1 ≤ sup I(f, g, π) = f dg ≤ f (x)dx = inf S(f, π) ≤ 1,
π∈Π 0 π∈Π
0

so,
Z 1 Z 1
f dg = f dg = 1.
0 0
R1
Therefore, according to the definition, f is Stieltjes-integrable on [0, 1] and 0 f dg = 1.
8.2. SOLUTIONS 55

i i+1

8.2. (a) Let i = 0, 1, ..., n − 1. On the interval n, n we have
i i+1
inf f (x) = sup f (x) = .
i i+1
x∈[ n , n ] n x∈[ ni , i+1 n
n ]

So, we obtain
n−1   n−1
X i i+1 i X i 20
I(f, g, πn ) = 20 − 20 =
n n n n n
i=0 i=0
n−1
20 X n(n − 1)
= i = 10 ,
n2 n2
i=0

n−1   n−1
X i+1 i+1 i X i + 1 20
S(f, g, πn ) = 20 − 20 =
n n n n n
i=0 i=0
n−1
20 X (n + 1)n
= (i + 1) = 10 ,
n2 n2
i=0

(b) As n → +∞ we have
n(n − 1) + 1 (n + 1)n
lim I(f, g, πn ) = 10 lim = 10 lim S(f, g, πn ) = 10 lim = 10.
n2 n2
(c) We have
Z 1 Z 1
10 ≤ sup I(f, g, π) = f dg ≤ f (x)dx = inf S(f, π) ≤ 10,
π∈Π 0 π∈Π
0
so,
Z 1 Z 1
f dg = f dg = 10.
0 0
R1
Therefore, according to the definition, f is Stieltjes-integrable on [0, 1] and 0 f dg = 10.
8.3. We consider the subdivision πn obtained by subdividing [0, 1] into n equal parts and
subdividing [1, 2] into another n equal parts.
Then we have
n−1   n−1
X  
X i i+1 i i i+1 i
I(f, g, πn ) = 20 − 20 + 1+ 20 + 10 − 20 − 10
n n n n n n
i=0 i=0
| {z } | {z }
pieces in [0,1] pieces in [1,2]
n−1 n−1 n−1
! ! !
20 X X i

10 n(n − 1) 10 X i
= i + 1+ = 10 + 1+
n2 n n n2 n n
i=0 i=0 i=0
n−1
!
n(n − 1) 10 X n(n − 1) n(n − 1)
= 10 2
+ 10 + 2 i = 10 2
+ 10 + 5 → 25.
n n n n2
i=0

A similar computation can be used to verify that S(f, g, πn ) → 25.

R2
We conclude that f is Stieltjes-integrable on [0, 2] and that 0 f dg = 25.
56 CHAPTER 8. THE STIELTJES INTEGRAL

8.4. We observe that f is bounded on [1, e], the interval of integration. Further, the integrator
function g(x) = ln x is differentiable and its derivative g ′ (x) = x1 is integrable on the same
interval. We can therefore reduce the Stieltjes integral to a Riemann integral.
Since f (x)g ′ (x) = x x1 = 1 is Riemann-integrable on [1, e], it follows that f is Stieltjes-
integrable with respect to g on [1, e] and we obtain
Z e Z e Z e
1
xd(ln x) = x dx = 1dx = e − 1.
1 1 x 1

√
8.5. We observe that f is bounded on [1, 2]. Further, the integrator function g(x) = x is
differentiable and its derivative g ′ (x) = 2√1 x is integrable on the same interval. We can
therefore reduce the Stieltjes integral to a Riemann integral.
Therefore, f is Stieltjes-integrable with respect to g on [1, 2] and we obtain
Z 2 Z 2
1 2 3
Z 2
√
Z
1 1
(x2 + 4x)d( x) = (x2 + 4x) √ dx = x 2 dx + 2 x 2 dx
1 1 2 x 2 1 1
 2  2
1 2 5 2 3 1 h 5 i2 4 h 3 i2
= x2 + 2 x2 = x2 + x2
2 5 1 3 1 5 1 3 1
1 h √ i 4 h √ i 4 8 √ 1 4
= 4 2−1 + 2 2−1 =( + ) 2− −
5 3 5 3 5 3
52 √ 23
= 2− .
15 15
x
8.6. We observe that f is bounded on [10, 20]. Further, the integrator function g(x) = 1 − e− 10
x
1 − 10
is differentiable and its derivative g ′ (x) = 10 e is integrable on the same interval. We
can therefore reduce the Stieltjes integral to a Riemann integral.
Therefore, f is Stieltjes-integrable with respect to g on [10, 20] and we obtain
Z 20 Z 20
1 20 − x 11 20 − x
Z Z
x 1 x
(5x + 11)d(1 − e− 10 ) = (5x + 11) e− 10 dx = xe 10 dx + e 10 dx
10 10 10 2 10 10 10
   Z  20
1 x
− 10

− 10x x
− 10
= x −10e + 10 e dx − 11e
2 10
x 20
h x x
i
− 10 − 10 − 10
= −5xe − 50e − 11e
10
x 20
h i
= −(5x + 61)e− 10
10
= −161e−2 + 111e−1 .
R2 √ √
8.7. In this case we cannot reduce 0 (x2 + 4x)d( x) to a Riemann integral, since x is not
differentiable at x = 0. We can, however, use Ito’s formula, which yields a result even faster
(when applicable).
In fact, the integrand can be written as
√ √ √
x2 + 4x = ( x)4 + 4( x)2 = f ′ ( x) for x ∈ [0, 2],

where f ′ (t) = t4 +4t2 is the derivative of f (t) = 15 t5 + 43 t3 , which is continuously differentiable

√
on [0, 2]. The integrator function g(x) = x is continuous on [0, 2].
8.2. SOLUTIONS 57

Thus,
√
Z 2 √
Z 2 √ √ √
Z 2
2 4 2
t4 + 4t2 dt



(x + 4x)d( x) = ( x) + 4( x) d( x) =
0 0 0
√ 2
4√ 8√ 52 √

1 5 4 3
= t + t = 2+ 2= 2,
5 3 0 5 3 15
which is consistent with the result we found in exercise 8.5.
8.8. Analogously to the preceding exercise, the integrand can be written as
√ √
1 + x3 = 1 + ( x)6 = f ′ ( x) for x ∈ [0, 2],
where f ′ (t) = 1 + t6 is the derivative of f (t) = t + 17 t7 , which is continuously differentiable
√
on [0, 4]. The integrator function g(x) = x is continuous on [0, 4].
Thus,
4 4 2 2
√ √ √
Z Z Z 
3 6 6 1
dt = t + t7



(1 + x )d( x) = 1 + ( x) d( x) = 1+t
0 0 0 7 0
1 142 2
= 2 + 128 = = 20 + .
7 7 7
8.9. Following the procedure from the previous exercises, the integrand can be written as
1 −1
+ ln x = eln [(x) ] + ln x = e− ln x + ln x = f ′ (ln x) for x ∈ [1, e],
x
where f ′ (t) = e−t + t is the derivative of f (t) = −e−t + 12 t2 , which is continuously
differentiable on [1, e]. The integrator function g(x) = ln x is continuous on [1, e].
Thus,
1 2 1
Z e Z e Z 1  
1 − ln x

−t

−t
( + ln x)d(ln x) = e + ln x d(ln x) = e + t dt = −e + t
1 x 1 0 2 0
1 1 3 1
=− + +1−0= − .
e 2 2 e
8.10. The integrator function is a step function. We first recall the procedure for computing the
integral in this case.
If f : [a, b] → R is a continuous function on [a, b] and g : [a, b] → R is a step function with
points of discontinuity {d1 , d2 , . . . , dn } ⊆ [a, b], we denote
g d− g d+ g a− = g (a) , g b+ = g (b) .





k = lim g(x), k = lim g(x),
x→d−
k x→d+
k

Then,
Z b n
X
−
f (dk ) g d+



f (x) dg (x) = k − g dk .
a k=1

In this case, on the interval [0, 4], the integer part function has discontinuities at 1,2,3,4.
So we have
Z 4
1 1 1 1 1 1 1 1 1
d(g(x)) = [1 − 0] + [2 − 1] + [3 − 2] + [4 − 3] = + + +
0 x+1 2 3 4 5 2 3 4 5
30 + 20 + 15 + 12 77
= = .
60 60
58 CHAPTER 8. THE STIELTJES INTEGRAL

8.11. Here the integrator function is neither continuous nor a step function. By the property
of additivity with respect to the interval of integration, we can split the integral into two
pieces: Z 3 Z Z 1 3
xd(g(x)) = xd(g(x)) + xd(g(x)).
0 0 1

On [0, 1] the integrator function g(x) is continuous and we have

1 1 1
x2
Z Z 
1
xd(g(x)) = xdx = =
0 0 2 0 2

while on [1, 3] the integrator function g(x) is a step function with discontinuities at 1 and
2, and we have Z 3
xd(g(x)) = 1[5 − 1] + 2[10 − 5] = 4 + 10 = 14.
1

Therefore, the entire integral is

Z 3 Z 1 Z 3
1 29
xd(g(x)) = xd(g(x)) + xd(g(x)) = + 14 = .
0 0 1 2 2
Chapter 9

Probability measures

9.1 Exercises
|A|
9.1. Let Ω = {ω1 , ω2 , ..., ωn }. Let µ be a set function µ : 2Ω → R such that µ(A) = for
|Ω|
every A ∈ 2Ω , where |A| denotes the cardinality of A, i.e., the number of states ωi ∈ A.

(a) Prove that µ(A ∩ B) ≤ µ(A ∪ B).

(b) Prove that µ is a measure.
(c) Prove that µ is a probability.

9.2. Consider a set Ω and a measure µ : 2Ω → R such that µ(A) = 0.3 and µ(B) = 0.9.

(a) Calculate the minimum and maximum possible values of µ(A ∩ B).
(b) Calculate the minimum and maximum possible values of µ(A ∪ B).
(c) If µ is a probability, can the events A and B be disjoint?

9.3. Given a set Ω and A, B ∈ 2Ω , consider the two Dirac probabilities δA , δB .

(a) Prove or disprove that δA∪B = δA + δB .

(b) Prove or disprove that δA∪B = max{δA , δB }.

2 3
9.4. Consider a set Ω and two probabilities P1 , P2 : 2Ω → [0, 1]. Prove that P1 + P2 is a
5 5
probability.

9.5. Consider the set Ω = N, the Poisson probability with parameter λ = 0.6 and the events
A = {1, 3, 5}, B = {ω ∈ Ω : ω ≥ 4} ∈ 2Ω . Calculate:

(a) P (A).
(b) P (B).

59
60 CHAPTER 9. PROBABILITY MEASURES

9.6. Consider the set Ω = N, the geometric probability with parameter p = 0.3 and the events
A = {ω ∈ Ω : 0 ≤ ω ≤ 8}, B = {ω ∈ Ω : ω > 10} ∈ 2Ω . Calculate:

(a) P (A).
(b) P (B).

9.7. Consider a game in which two players (Player A and Player B) take turns rolling a fair,
6-sided die. Player A starts the game and the first player to roll a 6 wins. Using the
geometric probability, calculate the probability that Player A wins.

9.8. Consider the following state spaces and their probabilities:

3
(a) Ω = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}, with P ({2, 3, 4} = 5 and P ({8} = 15 .
1
(b) Ω = N, with P (ω is even) = 2 and P (ω is odd) = 12 .
(c) Ω = N, with P ({0 ≤ ω ≤ 15}) = 37 , P ({20 ≤ ω ≤ 50}) = 27 , P ({70 ≤ ω ≤ 100}) = 17 ,
P ({120 ≤ ω ≤ 130}) = 71 .
1
(d) Ω = R, with P ([0, 1]) = 3 and P ((1, 8]) = 32 .

Determine which of the above are simple probabilities.

9.9. Consider Ω = N and let P be a probability on N such that:

5 2 4
P ({20, 30}) = , P ({10}) = , P ({10, 30}) =
7 7 7

(a) Calculate P ({20}), P ({30}).

(b) Determine if P is simple.
(c) If P is simple, identify its support.

9.10. Consider Ω = Z and let P be a probability on Z such that:

6 3 3 1
P ({−1, 0, 1}) = , P ({0, 5}) = , P ({−5, 0}) = , P ({0}) =
10 10 10 10

(a) Determine if P is simple.

(b) Calculate P ({10}).
(c) If P is simple, identify its support.

9.2 Solutions
|A ∩ B| |A ∪ B|
9.1. (a) Since (A ∩ B) ⊆ (A ∪ B), we have that |A ∩ B| ≤ |A ∪ B|, so ≤ and
|Ω| |Ω|
µ(A ∩ B) ≤ µ(A ∪ B).
9.2. SOLUTIONS 61

0
(b) Since |∅| = 0, we have that µ(∅) = = 0; thus, µ is grounded.
|Ω|
|A|
Since |A| ≥ 0 ∀A ∈ 2Ω , we have that µ(A) = ≥ 0; thus, µ is positive.
|Ω|
Finally, for every A, B ∈ 2Ω such that A ∩ B = ∅ we have that |A ∪ B| = |A| + |B| −
|A ∪ B| |A| + |B| |A| |B|
|A ∩ B| = |A| + |B|, so µ(A ∪ B) = = = + = µ(A) + µ(B);
|Ω| |Ω| |Ω| |Ω|
thus, µ is additive.
We conclude that µ is a measure.
|Ω|
(c) Since µ(Ω) = = 1, we have that µ is normalized; therefore, µ is a probability.
|Ω|

9.2. (a) The minimum value is obtained when µ(A ∩ B) = 0, which certainly occurs if
A ∩ B = ∅, but can also occur when A ∩ B =
̸ ∅. The maximum value is obtained when
µ(A ∩ B) = 0.3, which certainly occurs when A ⊆ B, but can also occur when A is
not contained in B. Observe that µ(A ∩ B) > 0.3 is not possible, since (A ∩ B) ⊆ A
and µ must be monotonic.
(b) From the relation µ(A ∪ B) = µ(A) + µ(B) − µ(A ∩ B) we have that the maximum
value is obtained when µ(A ∩ B) = 0 and therefore µ(A ∪ B) = 1.2. Analogously, the
minimum value is obtained when µ(A ∩ B) = 0.3 and therefore µ(A ∪ B) = 0.9.
(c) If µ is a probability, then it must be normalized and it follows that the maximum
value of the union is µ(A ∪ B) = 1. We then obtain that the minimum value of the
intersection is µ(A ∩ B) = µ(A) + µ(B) − µ(A ∪ B) = 0.2; thus, the two events cannot
be disjoint.

9.3. (a) By definition:

( (
1 if ω ∈ A 1 if ω ∈ B
δA (ω) = δB (ω) =
0 if ω ∈
̸ A 0 if ω ∈
̸ B

and (
1 if ω ∈ A ∪ B
δA∪B (ω) =
0 if ω ∈
̸ A∪B
Consider ω ̸∈ A ∪ B, so δA∪B (ω) = 0 and δA (ω) = δB (ω) = 0 since ω ̸∈ A and ω ̸∈ B.
Next, consider ω ∈ A ∪ B, so δA∪B (ω) = 1 and we examine the three cases:
– ω ∈ A and ω ̸∈ B, so δA (ω) = 1 and δB (ω) = 0, so δA∪B (ω) = δA (ω) + δB (ω)
– ω∉ A and ω ∈ B, so δA (ω) = 0 and δB (ω) = 1, so δA∪B (ω) = δA (ω) + δB (ω)
– ω ∈ A and ω ∈ B, so δA (ω) = 1 and δB (ω) = 1, so δA∪B (ω) ̸= δA (ω) + δB (ω)
Thus, it is false that δA∪B (ω) = δA (ω) + δB (ω).
(b) Consider ω ̸∈ A ∪ B, so δA∪B (ω) = 0 and δA (ω) = δB (ω) = 0 since ω ̸∈ A and ω ̸∈ B.
Next, consider ω ∈ A ∪ B, so δA∪B (ω) = 1 and we examine the three cases:
– ω ∈ A and ω ∈
̸ B, so δA (ω) = 1 and δB (ω) = 0, so δA∪B (ω) = max{δA (ω), δB (ω)}
– ω∉ A and ω ∈ B, so δA (ω) = 0 and δB (ω) = 1, so δA∪B (ω) = max{δA (ω), δB (ω)}
– ω ∈ A and ω ∈ B, so δA (ω) = 1 and δB (ω) = 1, so δA∪B (ω) = max{δA (ω), δB (ω)}
Thus, it is true that δA∪B (ω) = max{δA (ω), δB (ω)}.
62 CHAPTER 9. PROBABILITY MEASURES

9.4. We verify the properties of a probability:

 
2 3 2 3 2 3
– P1 + P2 (∅) = P1 (∅) + P2 (∅) = · 0 + · 0 = 0.
5 5 5 5 5 5
 
2 3 2 3
– P1 + P2 (A) = P1 (A) + P2 (A) ≥ 0, ∀A ∈ 2Ω .
5 5 5 5
– Let A, B ∈ 2Ω such that A ∩ B = ∅. Then:
 
2 3 2 3 2
P1 + P2 (A ∪ B) = P1 (A ∪ B) + P2 (A ∪ B) = (P1 (A) + P1 (B)))+
5 5 5 5 5
   
3 2 3 2 3
+ (P2 (A) + P2 (B))) = P1 + P2 (A) + P1 + P2 (B).
5 5 5 5 5
 
2 3 2 3 2 3
– P1 + P2 (Ω) = P1 (Ω) + P2 (Ω) = · 1 + · 1 = 1.
5 5 5 5 5 5

9.5. (a) By definition:

λn −λ
P (ω = n) = e
n!
So by additivity,

0.6 −0.6 0.6 −0.6 0.65 −0.6

P (A) = P (ω = 1)+P (ω = 3)+P (ω = 5) = e + e + e (= 0.3494)
1 3! 5!

(b) Using the property P (B) = 1 − P (B c ), we have:

P (B) = 1 − P ({0, 1, 2, 3}) = 1 − (P (ω = 0) + P (ω = 1) + P (ω = 2) + P (ω = 3)) =

0.6 −0.6 0.62 −0.6 0.63 −0.6

= 1 − e−0.6 − e − e − e (= 0.003358)
1 2 6

9.6. (a) By definition:

P (ω = n) = (1 − p)pn
So by additivity,
8 8
X X 1 − 0.39
P (A) = 0.7 · 0.3n = 0.7 · 0.3n = 0.7 · = 1 − 0.39 (= 0.99998)
1 − 0.3
n=0 n=0

(b) Using the property P (B) = 1 − P (B c ), we have:

10
X
P (B) = 1 − P ({ω ∈ Ω : 0 ≤ ω ≤ 10}) = 1 − 0.7 · 0.3n =
n=0

1 − 0.311
= 1 − 1 − 0.311 = 0.311 (= 0.00000177147)


= 1 − 0.7 ·
1 − 0.3
9.2. SOLUTIONS 63

9.7. The probability of failure is p = 56 and the probability of success is 1 − p = 16 . Let P (n) be
the probability of the first success occurring on the nth roll. The probability that Player A
wins is equal to the probability that the first success occurs on an odd-numbered roll. So,
the probability of Player A winning is:

+∞  2n
1 1 5 5 X1 5
P (1) + P (3) + ... = + · · + ... = · =
6 6 6 6 6 6
n=0
+∞  n
1 X 25 1 1 6
= = · 25 =
6 36 6 1 − 36 11
n=0

Analogously, the probability that Player B wins is equal to the probability that the first
success occurs on an even-numbered roll. So, the probability of Player B winning is:

+∞  2n+1
1 5 1 5 5 5 X1 5
P (2) + P (4) + ... = · + · · · + ... = · =
6 6 6 6 6 6 6 6
n=0
+∞  2n +∞  n
1 5 X 5 5 X 25 5 1 5
= · = = · 25 =
6 6 6 36 36 36 1 − 36 11
n=0 n=0

Note that Player B winning is the complementary event of Player A winning. So, we could
have calculated this probability using the property of the complementary event.

9.8. (a) Given that Ω is finite, P is surely simple, since P (Ω) = 1.

(b) We have that the only event with probability 1 is Ω, which is not finite. Thus, P is
not simple.
(c) Observe that the four events are pairwise disjoint. By the additive property, we find
that their union is a finite event with probability 1. Thus, P is simple.
(d) The two events are disjoint and the probability of their union is P ([0, 8]) = 1. Although
[0, 8], an interval in R, is bounded, it is not a finite set. Thus, P is not simple.

2
9.9. (a) By additivity, we have that P ({30}) = P ({10, 30}) − P ({10}) = .
7
3
Analogously, P ({20}) = P ({20, 30}) − P ({30}) = .
7
(b) By additivity, we have that P ({10, 20, 30}) = 1, so there exists a finite event with
probability 1, thus P is simple.
(c) The support of P contains precisely the elementary events with non-zero probabilities.
In this case, we have that suppP = {10, 20, 30}.

9.10. (a) By additivity, we have:

2
P ({5}) = P ({0, 5}) − P ({0}) = .
10
2
P ({−5}) = P ({−5, 0}) − P ({0}) = .
10
Since {−5}, {−1, 0, 1}, {5} are pairwise disjoint, we have:
P ({−5, −1, 0, 1, 5}) = P ({−5}) + P ({−1, 0, 1}) + P ({5}) = 1, thus P is simple.
64 CHAPTER 9. PROBABILITY MEASURES

(b) Since {10} ̸∈ {−5, −1, 0, 1, 5}, we have that P ({10}) = 0.

(c) In this case it is not possible to identify the support of P , since although we have that
P ({−1, 0, 1}) > 0 we cannot establish whether the probabilities P ({−1}), P ({0}), P ({1})
are all strictly positive. We can be sure, however, that suppP ⊆ {−5, −1, 0, 1, 5}.
Chapter 10

Random variables

10.1 Exercises
10.1. Let (Ω, P ) be a probability space with Ω = {ω1 , ω2 , ω3 , ω4 } and P : 2Ω → R such that:

P (ω1 ) = 0.4 P (ω2 ) = 0.1 P (ω3 ) = 0.3 P (ω4 ) = 0.2

and let f : Ω → R be a random variable such that:

f (ω1 ) = −100 f (ω2 ) = −50 f (ω3 ) = 100 f (ω4 ) = 200

(a) Calculate P (f (ω) = 100).

(b) Calculate P (f (ω) = 0).
(c) Calculate P (f (ω) < −50).
(d) Calculate P (−150 < f (ω) ≤ 50).
(e) Calculate P (f (ω) ≥ 0).

10.2. Let (Ω, P ) be a probability space with Ω = {1, 2, 3, 4, 5, 6} and P : 2Ω → R such that:
(
p if ω is even
P (ω) =
2p if ω is odd

and consider the random variable f : Ω → R:


−3
 if ω = 1, 2
f (ω) = 1 if ω = 3, 4, 5

4 if ω = 6


(a) Calculate the value of p.

(b) Calculate EP (f ), VP (f ) and σP (f ).
(c) Consider the random variable g : Ω → R such that:
(
−2 if ω is even
g(ω) =
2 if ω is odd

Calculate CovP (f, g).

65
66 CHAPTER 10. RANDOM VARIABLES

10.3. A company has purchased software for processing large volumes of data. However, the
software is subject to making ω ∈ N errors, with probabilities P (ω). In particular,
P (0) = 0.25, P (1) = 0.23, P (2) = 0.19, P (3) = 0.14, P (4) = 0.11, P (5) = 0.06 and
P (6) = 0.02. To correct the inaccuracies, the company must pay a fee that depends on the
number of errors. If the number of errors is less than or equal to 3, the cost incurred is
4000 Euro per error. If more than 3 errors are made, the company pays a flat rate of 20000
Euro.
(a) Calculate the probability that the software will make at most 3 errors. Calculate the
probability that the software will make at least 5 errors.
(b) Calculate the expected value of the cost the company will bear. Calculate the
probability that the fees incurred will be less than 10000 Euro.

10.4. Consider two random variables f, g : Ω → R and a simple probability P : 2Ω → R

represented by the following vectors:
f = (−10, 0, 2, 4), g = (−25, −5, 5, 10), P = (0.3, 0.15, 0.35, 0.2)
(a) Calculate EP (f ) and EP (g).
(b) Calculate VP (f ) and VP (g).
(c) Calculate EP (2f − 3g).
(d) Calculate ρP (f, g).
(e) Calculate VP (2f − 3g).

10.5. Let f, g, h : Ω → R be random variables defined on a simple probability P : 2Ω → R.

Prove that:
(a) CovP (f, f ) = VP (f ).
(b) CovP (f, k) = 0 ∀k ∈ R.
(c) CovP (f + g, h) = CovP (f, h) + CovP (g, h).

10.6. Let Ω = R and let P : 2Ω → R be a probability distributed uniformly on suppP =

{−200, −100, 0, 100, 200}. Consider the random variables:



1 if ω = −200




2 if ω = −100

3 if ω = 0
f (ω) = and g(ω) = f 2 (ω)


4 if ω = 100

5

 if ω = 200


0 otherwise
Determine if f and g are positively/negatively/not correlated.

10.7. Consider a European call option that gives you the right to purchase a stock in one year at a
price of 70 Euro. Suppose the contract has price p = 0 and that according to forecasts, the
value of the stock in one year will be Ω = {ω1 = 60, ω2 = 75, ω3 = 100} with probabilities
P = (0.35, 0.55, 0.1).
 10.2. SOLUTIONS 67

(a) Calculate the probability that you decide to exercise the option in one year.
(b) Calculate the expected value of the call option with respect to the probability P .

10.8. Given Ω = N and a simple probability such that P (10) = 0.2, P (20) = 0.25, P (30) =
0.4, P (40) = 0.15. Let f, g : Ω → R such that:
( (
1 if ω is even 1 if ω is a multiple of 5
f (ω) = and g(ω) =
−1 if ω is odd 0 otherwise

(a) Prove that f and g are equal almost everywhere.

(b) Calculate ρP (f, g).

10.9. The game of roulette is an experiment in which the set of elementary events is Ω = {n ∈
1
N : 0 ≤ n ≤ 36}, all of which have the same probability P (ω) = 36 .

(a) Player A places a bet of 1 Euro on the event ’odd’. If the ball lands on an odd number,
the player wins double their bet; otherwise, the player loses. Write the random variable
f that represents the game. Calculate EP (f ).
(b) Player B places a bet of 1 Euro on the event that the ball lands on 0. If this occurs,
the player wins 36 times their bet; otherwise, the player loses. Write the random
variable g that represents the game. Calculate EP (g).

10.10. Let Ω = {ω1 , ω2 , ω3 } and let P (ω1 ) = 0.43, P (ω2 ) = 0.25, P (ω3 ) = 0.32. Let f, g : Ω → R
be random variables defined as:

f (ω1 ) = −1 f (ω2 ) = 0 f (ω3 ) = 1

g(ω1 ) = −2 g(ω2 ) = 1 g(ω3 ) = 5

Consider the random variable h = 5f + g. Calculate VP (h).

10.2 Solutions
10.1. (a)
P (f (ω) = 100) = P (ω3 ) = 0.3
(b)
P (f (ω) = 0) = 0
(c)
P (f (ω) < −50) = P (f (ω) = −100) = P (ω1 ) = 0.4
(d)

P (−150 < f (ω) ≤ 50) = P (f (ω) = −100)+P (f (ω) = −50) = P (ω1 )+P (ω2 ) = 0.5

(e)

P (f (ω) ≥ 0) = P (f (ω) = 100) + P (f (ω) = 200) = P (ω3 ) + P (ω4 ) = 0.5

68 CHAPTER 10. RANDOM VARIABLES

10.2. We have:
(a) P (1) + P (2) + P (3) + P (4) + P (5) + P (6) = 1, that is, 2p + p + 2p + p + 2p + p = 9p = 1
so p = 19 .
(b) The expected value is:

EP (f ) = f (1) · P (1) + f (2) · P (2) + f (3) · P (3) + f (4) · P (4) + f (5) · P (5) + f (6) · P (6) =
2 1 2 1 2 1
= −3 · −3· +1· +1· +1· +4· =0
9 9 9 9 9 9
The variance is given by:

VP (f ) = EP (f 2 ) − [EP (f )]2

with
2 1 2 1 2 1 48
EP (f 2 ) = (−3)2 · + (−3)2 · + 12 · + 12 · + 12 · + 42 · = ,
9 9 9 9 9 9 9
so
48 48
VP (f ) = − 02 =
9 9
Finally, the standard deviation is:
√
p 4 3
σP (f ) = VP (f ) =
3
(c) The covariance between the two random variables is given by:

CovP (f, g) = EP (f g) − EP (f )EP (g)

The expected value of g is:

EP (g) = g(1) · P (1) + g(2) · P (2) + g(3) · P (3) + g(4) · P (4) + g(5) · P (5) + g(6) · P (6) =
2 1 2 1 2 1 2
=2· −2· +2· −2· +2· −2· =
9 9 9 9 9 9 3
The random variable f g : Ω → R, given by the product of the two random variables f
and g, is defined as follows:

f g(1) = −6, f g(2) = 6, f g(3) = 2, f g(4) = −2, f g(5) = 2, f g(6) = −8

Its expected value is thereby

2 1 2 1 2 1 8
EP (f g) = −6 · +6· +2· −2· +2· −8· =−
9 9 9 9 9 9 9
and the covariance is:
8 2 8
CovP (f, g) = − − 0 · = −
9 3 9

10.3. (a) The probability that the software makes at most 3 errors is:

P (0) + P (1) + P (2) + P (3) = 0.81

The probability that it makes at least 5 errors is:

P (5) + P (6) = 0.08

10.2. SOLUTIONS 69

(b) To calculate the expected value of the cost incurred, we introduce the random variable
f : Ω → R where: 

 0 if ω = 0

4000 if ω = 1



f (ω) = 8000 if ω = 2

12000 if ω = 3





20000 if ω ≥ 4


Its expected value is:

EP (f ) = 0·0.25+4000·0.23+8000·0.19+12000·0.14+20000·(0.11+0.06+0.02) = 7920

The probability that the fees incurred are less than 10000 Euro is:

P (f (ω) < 10000) = P (f (ω) = 0) + P (f (ω) = 4000) + P (f (ω) = 8000) = 0.67

10.4. (a)
EP (f ) = f · P = −10 · 0.3 + 0 · 0.15 + 2 · 0.35 + 4 · 0.2 = −1.5
EP (g) = g · P = −25 · 0.3 − 5 · 0.15 + 5 · 0.35 + 10 · 0.2 = −4.5
(b) We first calculate the expected values of f 2 and g 2 :

EP (f 2 ) = (−10)2 · 0.3 + 02 · 0.15 + 22 · 0.35 + 42 · 0.2 = 34.6

EP (g 2 ) = (−25)2 · 0.3 + (−5)2 · 0.15 + 52 · 0.35 + 102 · 0.2 = 220

The variances are:
VP (f ) = EP (f 2 ) − [EP (f )]2 = 32.35
VP (g) = EP (g 2 ) − [EP (g)]2 = 199.75
(c) By linearity of the expected value:

EP (2f − 3g) = 2EP (f ) − 3EP (g) = 10.5

(d) Before calculating the linear correlation coefficient, we must find the covariance
between the two random variables:

EP (f g) = (−10) · (−25) · 0.3 + 0 · (−5) · 0.15 + 2 · 4 · 0.35 + 4 · 10 · 0.2 = 85.8

CovP (f, g) = EP (f g) − EP (f )EP (g) = 79.05

Thus,
CovP (f, g) 79.05
ρP (f, g) = =√ √ = 0.98
σP (f )σP (g) 32.35 · 199.75
(e) We have that:

VP (2f − 3g) = 22 VP (f ) + (−3)2 VP (g) + 2 · 2 · (−3) · CovP (f, g) =

= 4 · 32.35 + 9 · 199.75 − 12 · 0.98 = 1915.39

70 CHAPTER 10. RANDOM VARIABLES

10.5. The covariance between two random variables is given by:

CovP (f, g) = EP (f g) − EP (f )EP (g)

(a)
CovP (f, f ) = EP (f f ) − EP (f )EP (f ) = EP (f 2 ) − [EP (f )]2 = VP (f )
(b)
CovP (f, k) = EP (kf ) − EP (f )EP (k)
By linearity of the expected value and since EP (k) = k, we obtain:

CovP (f, k) = k · EP (f ) − k · EP (f ) = 0

(c)

CovP (f +g, h) = EP ((f + g)h)−EP (f +g)EP (h) = EP (f h + gh)−EP (f +g)EP (h) =

By linearity of the expected value:

= EP (f h) + EP (gh) − (EP (f ) + EP (g)) EP (h) =

= EP (f h) + EP (gh) − EP (f )EP (h) − EP (g)EP (h) =

= EP (f h) − EP (f )EP (h) + EP (gh) − EP (g)EP (h) = CovP (f, h) + CovP (g, h)

10.6. We must calculate:

CovP (f, g) = EP (f g) − EP (f )EP (g) = EP (f 3 ) − EP (f )EP (f 2 )

We start with EP (f ), EP (f 2 ), EP (f 3 ).
1
The probability is distributed uniformly on its support so P (ωi ) = 5 ∀ωi ∈ suppP .
1 1 1 1 1
EP (f ) = 1 · +2· +3· +4· +5· =3
5 5 5 5 5
1 1 1 1 1
EP (f 2 ) = 1 · +4· +9· + 16 · + 25 · = 11
5 5 5 5 5
1 1 1 1 1
EP (f 3 ) = 1 · + 8 · + 27 · + 64 · + 125 · = 45
5 5 5 5 5
Thus:
CovP (f, g) = 45 − 3 · 11 = 12
We deduce that f and g are positively correlated.

10.7. The payoffs of the European call option are defined by a random variable f : Ω → R such
that f (ωi ) = max{ωi − 70, 0}. We obtain f = (0, 5, 30).

(a) The option will be exercised in the case of a strictly positive payoff, whose probability
is given by:

P (f > 0) = P (f = 5) + P (f = 30) = P (ω2 ) + P (ω3 ) = 0.65

(b) The expected value is:

EP (f ) = f · P = 5.75
 10.2. SOLUTIONS 71

10.8. (a) suppP = {10, 20, 30, 40} and P (ω) = 0 ∀ω ∈ / suppP .
We have that f (ω) = g(ω) ∀ω ∈ suppP and the two random variables differ only for
elementary events that have zero probabilities. Therefore, f and g are equal almost
everywhere.
(b) We first calculate the expected values of f , g and f g:
EP (f ) = f (10) · P (10) + f (20) · P (20) + f (30) · P (30) + f (40) · P (40) = 1
EP (g) = EP (f ) = 1 (since f and g are equal almost everywhere)
EP (f g) = f (10)·g(10)·P (10)+f (20)g(20)P (20)+f (30)g(30)·P (30)+f (40)g(40)·P (40) = 1
We obtain CovP (f, g) = EP (f g) − EP (f )EP (g) = 0. Thus, ρP (f, g) = 0.

10.9. (a) The random variable f associates +1 (the winnings minus the bet) when the event
odd = {1, 3, 5, ..., 35} occurs, and −1 otherwise:
(
+1 if ω ∈ {1, 3, 5, ..., 35}
f (ω) = .
−1 if ω ∈ {0, 2, 4, ..., 36}
18
We obtain EP (f ) = 37 − 19 1
37 = − 37 .
(b) The random variable g associates +35 (the winnings minus the bet) when the event
{0} occurs, and −1 otherwise:
(
+35 if ω = 0
g(ω) = .
−1 if ω ∈ {1, 2, 3, ..., 36}
35 36 1
We obtain EP (g) = 37 − 37 = − 37 .

10.10. We have:
VP (h) = 52 VP (f ) + 12 VP (g) + 2 · 5 · 1 · CovP (f, g)
We calculate EP (f ) and EP (f 2 ):
EP (f ) = −1 · 0.43 + 0 · 0.25 + 1 · 0.32 = −0.11
EP (f 2 ) = (−1)2 · 0.43 + 02 · 0.25 + 12 · 0.32 = 0.75
So the variance of f is:
VP (f ) = EP (f 2 ) − [EP (f )]2 = 0.7379
We calculate EP (g) and EP (g 2 ):
EP (g) = −2 · 0.43 + 1 · 0.25 + 5 · 0.32 = 0.99
EP (g 2 ) = (−2)2 · 0.43 + 12 · 0.25 + 52 · 0.32 = 9.97
So the variance of g is:
VP (g) = EP (g 2 ) − [EP (g)]2 = 8.9899
To calculate the covariance we must first compute the expected value of f g:
EP (f g) = 2 · 0.43 + 0 · 0.25 + 5 · 0.32 = 0.74
and we obtain
CovP (f, g) = EP (f g) − EP (f )EP (g) = 0.74 − (−0.11) · 0.99 = 0.8489
Thus:
VP (h) = 25 · 0.7379 + 8.9899 + 10 · 0.8489 = 35.9264
Chapter 11

Distribution and density functions

11.1 Exercises
11.1. Let (Ω, P ) be a probability space, f : Ω → R be a random variable and

1
 x2
 if 1 ≤ x ≤ 2
ϕ(x) = a if 2 < x ≤ 6 , with a > 0

0 otherwise


be the density function of f .

(a) Determine the value of a.

(b) Calculate the distribution function Φ(x) associated with ϕ(x).
(c) Calculate P (f > 4) and P (3 < f ≤ 4).
(d) Calculate EP (f ).

11.2. Consider the function 

0 if x < 2
Φ(x) = 4
1 − if x ≥ 2
x2

(a) Verify that Φ(x) satisfies the properties of a distribution function and find its carrier,
if it exists.
(b) Write the density function ϕ(x) defined by the function Φ(x).
(c) Calculate the expected value EP (f ) of a random variable f whose density function is
ϕ(x).

11.3. Let (Ω, P ) be a probability space and let f : Ω → R be a simple random variable.
Prove that f is essentially bounded.

73
74 CHAPTER 11. DISTRIBUTION AND DENSITY FUNCTIONS

11.4. Consider Ω = {ω1 , ω2 , ω3 , ω4 , ω5 , ω6 } and P = (0.08, 0.17, 0.25, 0.2, 0.14, 0.16). The random
variable f is defined as 


 −30 if ω = ω1




 −20 if ω = ω2

5 if ω = ω3
f (ω) =
20
 if ω = ω4





 40 if ω = ω5

100 if ω = ω6

(a) Determine if the random variable is simple.

(b) Write the distribution function Φ(x) of the random variable f and find its carrier, if it
exists.
(c) Write the density function ϕ(x) of the random variable f .
(d) Calculate the expected value of f and the expected value of 3f − 10.

11.5. Consider the function 

0
 if x < 0
Φ(x) = x3 if 0 ≤ x < 1

1 if x ≥ 1


(a) Verify that Φ(x) satisfies the properties of a distribution function and find its carrier,
if it exists.
(b) Write the density function ϕ(x) defined by the function Φ(x).
(c) Calculate the expected value of a random variable f whose density function is ϕ(x).

11.6. Given the probability space (Ω, P ), the random variable f : Ω → R, and its distribution
function 
0
 if x < 14

1
if 14 ≤ x < 23


5

Φ(x) = 12 if 23 ≤ x < 1
 3
if 1 ≤ x < 3




 4
1 if x ≥ 3


calculate the expected value and variance of the random variable f .

11.7. Consider the function

1
ϕ(x) =
π(1 + x2 )

(a) Verify that ϕ(x) satisfies the properties of a density function.

(b) Calculate the distribution function Φ(x) associated with ϕ(x) and prove that it is
continuously differentiable on R.
 11.2. SOLUTIONS 75

11.8. Consider the density function of a random variable f

(
(α − 1)x−α if x ≥ 1
ϕ(x) = , with α > 1
0 if x < 1

For which values of α > 1 does the random variable f admit a finite expected value?

11.9. Consider the distribution function of a random variable f ,


0
 if x < −2
x+2
Φ(x) = if − 2 ≤ x ≤ 4
 6
1 if x > 4


(a) Calculate P (f > 2), P (f ≤ −1), P (−5 < f ≤ 5).

(b) Find the density function ϕ(x) and the expected value of f .
(c) Determine the expected value of f , if it exists. (If possible, apply Cavalieri’s theorem.)

11.10. Consider a probability space (Ω, P ) with Ω = R and where P is the Dirac probability
measure δ√3 .
Let f : Ω → R be a random variable defined by f (ω) = ω 2 .
Calculate the distribution function Φ(x) associated with f .

11.2 Solutions
11.1. (a) We observe that the density ϕ is non-zero only on the interval [1, 6]. We confirm that
it is integrable on that same interval. Since ϕ is a density, we must have that ϕ(x) ≥ 0
R +∞
∀x ∈ R and −∞ ϕ(x)dx = 1. The first condition holds for all a > 0. The second
condition gives us:
Z 1 Z 2 Z 6 Z +∞
1
0 dx + dx + a dx + 0 dx = 1
−∞ 1 x2 2 6

1 2
 
0+ − + [ax]62 + 0 = 1
x 1
1
+ 4a = 1
2
1
a=
8
Thus,

1
 x2
 if 1 ≤ x ≤ 2
1
ϕ(x) = if 2 < x ≤ 6
8
0 otherwise

76 CHAPTER 11. DISTRIBUTION AND DENSITY FUNCTIONS

(b) We have that: Z x

Φ(x) = ϕ(t) dt
−∞
If x < 1: Z x
Φ(x) = 0 dt = 0
−∞
If 1 ≤ x ≤ 2:
Z 1 Z x  x
1 1 1
Φ(x) = 0 dt + 2
dt = 0 + − =1−
−∞ 1 t t 1 x
If 2 < x ≤ 6:
Z 1 Z 2 Z x  2  x
1 1 1 t 1 x 2 x+2
Φ(x) = 0 dt + dt + dt = 0 + − + = 1− + − =
−∞ 1 t2 2 8 t 1 8 2 2 8 8 8
If x > 6:
Z 1 Z 2 Z 6 Z +∞  2  6
1 1 1 t
Φ(x) = 0 dt + dt + dt + 0 dt = 0 + − + +0=
−∞ 1 t2 2 8 6 t 1 8 2
1 6 2
=1− + − +0=1
2 8 8
Thus: 


0 if x<1
1 − 1

if 1≤x≤2
Φ(x) = x+2 x
 8

 if 2<x≤6

1 if x>6
(c) We have:
6 1
P (f > 4) = 1 − P (f ≤ 4) = 1 − Φ(4) = 1 −
=
8 4
6 5 1
P (3 < f ≤ 4) = P (f ≤ 4) − P (f ≤ 3) = Φ(4) − Φ(3) = − =
8 8 8
(d)
Z +∞ Z +∞ Z 1 Z 2 Z 6 Z +∞
1 x
EP (f ) = xdΦ(x) = xϕ(x) dx = 0 dx+ dx+ dx+ 0 dx =
−∞ −∞ −∞ 1 x 2 8 6
2 6
 
x
0 + [ln x]21 + + 0 = ln 2 + 2
16 2

11.2. (a) Φ(x) is positive and increasing, assumes values in [0, 1] and
lim Φ(x) = 0
x→−∞

lim Φ(x) = 1
x→∞
Further, Φ is continuous on R, since limx→2− Φ(x) = limx→2+ Φ(x) = Φ(2) = 0, so
in particular Φ is continuous from the right. Therefore, Φ(x) can be considered a
distribution function of a random variable f .
Φ does not admit a carrier since there does not exist an interval [a, b] such that
Φ(x) = 0 ∀x ≤ a and Φ(x) = 1 ∀x ≥ b (the second condition is not realized).
11.2. SOLUTIONS 77

(b) Φ(x) has a point of non-derivability at x0 = 2. Differentiating Φ(x) on (2, +∞) we

obtain x83 . So we have that
(
0 if x < 2
ϕ(x) = 8
x3
if x ≥ 2

satisfies Z x
ϕ(t)dt = Φ(x)
−∞
and Z +∞ Z +∞
ϕ(t)dt = ϕ(t)dt = 1
−∞ 2
So ϕ is a density function defined by Φ.
(c) We have Z +∞ Z +∞ Z +∞
8x
xdΦ(x) = xϕ(x)dx = dx =
−∞ 2 2 x3
Z +∞
8
= dx = 4
2 x2

11.3. Since f is simple we have that Imf = {y1 , y2 , ..., yn }. Choosing m, M ∈ R such that
m < min{y1 , y2 , ..., yn } and M > max{y1 , y2 , ..., yn } then we have that P (m ≤ f ≤ M ) = 1,
that is, f is essentially bounded.

11.4. (a) The set of image values of f is {−30, −20, 5, 20, 40, 100} which is a finite set, so f is
simple.
(b) We have:


0 if x < ω1

0.08 if ω1 ≤ x < ω2





0.08 + 0.17 if ω2 ≤ x < ω3



Φ(x) = 0.08 + 0.17 + 0.25 if ω3 ≤ x < ω4

0.08 + 0.17 + 0.25 + 0.2 if ω4 ≤ x < ω5





0.08 + 0.17 + 0.25 + 0.2 + 0.14 if ω5 ≤ x < ω6





0.08 + 0.17 + 0.25 + 0.2 + 0.14 + 0.16 if x ≥ ω6


which simplifies to 

0 if x < ω1

0.08 if ω1 ≤ x < ω2





0.25 if ω2 ≤ x < ω3



Φ(x) = 0.5 if ω3 ≤ x < ω4

0.7 if ω4 ≤ x < ω5





0.84 if ω5 ≤ x < ω6





1 if x ≥ ω6


A carrier is given by the interval [a, b], with a < −ω1 and b ≥ ω6 .
78 CHAPTER 11. DISTRIBUTION AND DENSITY FUNCTIONS

(c) We have: 

0.08 if x = ω1

0.17 if x = ω2





0.25 if x = ω3



ϕ(x) = 0.2 if x = ω4

0.14 if x = ω5





0.16 if x = ω6





0 otherwise


(d)

EP (f ) = −30 · 0.08 − 20 · 0.17 + 5 · 0.25 + 20 · 0.2 + 40 · 0.14 + 100 · 0.16 = 21.05

By linearity of the expected value we have:

EP (3f − 10) = 3EP (f ) − 10 = 53.15

11.5. (a) Φ(x) is positive and increasing, assumes values in [0, 1] and

lim Φ(x) = 0
x→−∞

lim Φ(x) = 1
x→∞

Further, Φ is continuous on R, since limx→0− Φ(x) = limx→0+ Φ(x) = Φ(0) = 0 and

limx→1− Φ(x) = limx→1+ Φ(x) = Φ(1) = 1, so in particular Φ is continuous from
the right. Therefore, Φ(x) can be considered a distribution function of the random
variable f .
A carrier of Φ is an interval [a, b] with a ≤ 0 and b ≥ 1 (since Φ(x) = 0 ∀x ≤ a and
Φ(x) = 1 ∀x ≥ b).
(b) Φ(x) has two points of non-derivability at x0 = 0 and x1 = 1. Differentiating Φ(x) on
(0, 1) we obtain 3x2 . It can be verified that
(
3x2 if 0 < x < 1
ϕ(x) =
0 otherwise

satisfies Z x Z x
ϕ(t)dt = ϕ(t)dt = Φ(x)
−∞ 0

and Z +∞ Z 1
ϕ(t)dt = ϕ(t)dt = 1
−∞ 0

Thus, ϕ is a density function defined by Φ.

(c) We have
Z +∞ Z 1 Z 1
3
xdΦ(x) = xϕ(x)dx = 3x3 dx =
−∞ 0 0 4
11.2. SOLUTIONS 79

11.6. We have:
Z +∞ Z 3 4
X
xi Φ(xi ) − Φ(x−
 
EP (f ) = xdΦ(x) = xdΦ(x) = i ) =
1
−∞ 4 i=1
    −      − 
1 1 1 2 2 2
+1 Φ(1) − Φ(1− ) +3 Φ(3) − Φ(3− ) =
   
= Φ −Φ + Φ −Φ
4 4 4 3 3 3
       
1 1 2 1 1 3 1 3 5
= −0 + − +1 − +3 1− =
4 5 3 2 5 4 2 4 4
For the variance formula, we must first compute EP (f 2 ):
Z +∞ Z 3 4
X
EP (f 2 ) = x2 dΦ(x) = x2 dΦ(x) = x2i Φ(xi ) − Φ(x−
 
1
i ) =
−∞ 4 i=1
  
    
1 1 4 1 1 3 1 3 127
= −0 + − +1 − +9 1− =
16 5 9 2 5 4 2 4 48
Thus:
127 25 13
VP (f ) = EP (f 2 ) − [EP (f )]2 = − =
48 16 12

11.7. (a) We have that ϕ(x) ≥ 0 ∀x ∈ R. Further:

Z +∞ Z +∞ Z 0 Z k
1 1 1
ϕ(x)dx = 2
dx = lim 2
dx+ lim dx =
−∞ −∞ π(1 + x ) h→−∞ h π(1 + x ) k→+∞ 0 π(1 + x2 )
 0  k
1 1
lim arctan x + lim arctan x =
h→−∞ π h k→+∞ π 0
   
1 1 1 1
lim − arctan h + lim arctan k = + = 1
h→−∞ π k→+∞ π 2 2
(b) By definition we have:
Z x Z x
1
Φ(x) = ϕ(t)dt = dt =
−∞ −∞ π(1 + t2 )
Z x  x
1 1 1
= lim dt = lim arctan t = lim [arctan x − arctan h] =
h→−∞ h π(1 + t2 ) h→−∞ π h π h→−∞

1h πi 1 1
= arctan x + = arctan x +
π 2 π 2

Since ϕ is continuous on R, by the second fundamental theorem of integral calculus

we have that Φ is continuously differentiable on R.

11.8. The expected value is:

Z +∞ Z +∞ Z +∞ Z +∞
−α+1 1
EP (f ) = xdΦ(x) = xϕ(x)dx = (α − 1)x dx = (α − 1) dx
−∞ −∞ 1 1 xα−1
This is a generalized (improper) integral that converges when α − 1 > 1. So, the expected
value is finite when α > 2.
 80 CHAPTER 11. DISTRIBUTION AND DENSITY FUNCTIONS

11.9. 
0
 if x < −2
x+2
Φ(x) = if − 2 ≤ x ≤ 4
 6
1 if x > 4


(a) We have:
4 1
P (f > 2) = 1 − P (f ≤ 2) = 1 − Φ(2) = 1 − =
6 3
1
P (f ≤ −1) = Φ(−1) =
6
P (−5 < f ≤ 5) = P (f ≤ 5) − P (f ≤ −5) = Φ(5) − Φ(−5) = 1 − 0 = 1
(b) We have: (
1
6 if − 2 ≤ x ≤ 4
ϕ(x) =
0 otherwise
and thus: Z +∞
EP (f ) = xdΦ(x) =
−∞
Z +∞ Z 4  2 4
x x 16 4
= xϕ(x)dx = dx = = − =1
−∞ −2 6 12 −2 12 12
(c) Since P (−5 < f ≤ 5) = 1 the random variable f is essentially bounded. We can
therefore apply Cavalieri’s theorem – we obtain that the expected value of f is:
Z +∞ Z 0
EP (f ) = (1 − Φ(x))dx − Φ(x)dx =
0 −∞

4 Z 0 4  0
(x + 2)2 (x + 2)2
Z  
x+2 x+2
= 1− dx − dx = x − − =1
0 6 −2 6 12 0 12 −2

11.10. We know that: ( √

0 if 3 ∈
/A
P (A) = δ√3 (A) = √
1 if 3 ∈ A
and
Φ(x) = P (f ≤ x)
If x ≤ 3:
√ √
Φ(x) = P (f ≤ x) = P ({ω ∈ Ω : ω 2 < x}) = P ({ω ∈ Ω : − x < ω < x}) = 0
√ √ √
since 3 ∈
/ {ω ∈ Ω : − x < ω < x}.
If x > 3:
√ √
Φ(x) = P (f ≤ x) = P ({ω ∈ Ω : ω 2 < x}) = P ({ω ∈ Ω : − x < ω < x}) = 1
√ √ √
since 3 ∈ {ω ∈ Ω : − x < ω < x}.
Thus: (
0 if x ≤ 3
Φ(x) =
1 if x > 3
Chapter 12

Elementary financial calculus

12.1 Exercises
12.1. Let a(t) be a function that is continuous and positive ∀t ∈ [0, +∞). Show that the final
value of a principal C ≥ 0 invested from 0 to t defined by
 Z t 
M (C, t) = C 1 + a(s) ds
0

satisfies the four capitalization axioms.

12.2. Let a(t) be a function that is continuous and positive ∀t ∈ [0, +∞). Show that the final
value of a principal C ≥ 0 invested from 0 to t defined by
Rt
M (C, t) = C e 0 a(s) ds

satisfies the four capitalization axioms.

12.3. Let f be the function defined on the interval [0, +∞) as

√
f (t) = 1 + 0.12 t

Show that f defines an accumulation factor on [0, +∞).

12.4. Let f be the function defined on the interval [0, +∞) as

f (t) = 1 + 0.12 max(0, t − 1)

Show that f defines an accumulation factor on [0, +∞).

12.5. A principal of 1000 e is invested for 9 months at the simple annual interest rate of 3%.

(a) Calculate the final value at maturity.

(b) How much does the annual interest rate need to rise for the future value calculated in
part ’a’ to increase by 15%?

12.6. A principal of 1000 e is invested for 3 years at the annual compound interest rate of 4%.

81
 82 CHAPTER 12. ELEMENTARY FINANCIAL CALCULUS

(a) Calculate the final value at maturity.

(b) How much does the annual interest rate need to rise for the future value calculated in
part ’a’ to increase by 20%?

12.7. Let A be an ordinary annuity consisting of 18 semi-annual installments of 150 e at a

nominal annual interest rate of 12%.

(a) Calculate the effective annual interest rate i.

(b) Calculate the present value V0 of annuity A.
(c) Calculate the final value M of annuity A.
(d) Suppose you must substitute annuity A with an ordinary annuity B consisting of 10
constant semi-annual installments R > 0, at the same nominal annual interest rate.
Find the amount of the installments R such that the two annuities are financially
equaivalent.

12.8. Let B be an ordinary perpetuity with constant installments R > 0 and let i > 0 be the
annual compound interest rate.

(a) Verify that the present value of the perpetuity is

R
A(R, i) = = R a∞ i
i
(b) Suppose that the interest rate increases from i to i′ . Show that

A(R, i′ ) < A(R, i).

(c) Suppose that the perpetuity has constant installments of R = 150. Calculate the
present value given the annual interest rate i = 3.25%.

12.9. A bond purchased today reaches maturity in 2 years. Coupons with a value of 100 are
redeemed in 1 year and at maturity along with the reimbursement value of 1000.

(a) Construct the sequence of cash flows guaranteed by the bond.

(b) Using A > 0 to denote the nominal value of the bond, write the expression for the
DCF.
(c) What value must A have so the bond’s gross annual yield is 10%?
(d) Determine the duration of the bond.

12.10. A bond purchased today reaches maturity in 1 year and 4 months. Coupons with a value
of 30 are redeemed in 2 months, 8 months and at maturity, along with the reimbursement
value of 1000. The annual compound interest is x = 3.5%.

(a) Construct the sequence of cash flows guaranteed by the bond.

(b) Using A > 0 to denote the nominal value of the bond, write the expression for the
DCF.
(c) Calculate A.
(d) Determine the duration of the bond.
12.2. SOLUTIONS 83

12.2 Solutions
12.1. Axiom A.1 is satisfied since M depends only on C ≥ 0 and t ≥ 0. We observe that
 Z t 
M (C1 + C2 , t) = (C1 + C2 ) 1 + a(s) ds
0
 Z t   Z t 
= C1 1 + a(s)ds + C2 1 + a(s)ds = M (C1 , t) + M (C2 , t)
0 0
and we see that axiom A.2 is satisfied. For every t1 , t2 ≥ 0, we assume t1 ≤ t2 . Since
a(t) ≥ 0, we have
Z t1 Z t1 Z t2 Z t2
a(s) ds ≤ a(s) ds + a(s) ds = a(s) ds
0 0 t1 0

It follows that
 Z t1   Z t2 
M (C, t1 ) = C 1+ a(s) ds ≤C 1+ a(s) ds = M (C, t2 )
0 0
by which we observe that M is monotonically increasing with respect to the lifetime t and
consequently, axiom A.3 is satisfied. Finally, setting t = 0 we have that M (C, 0) = C and
therefore, axiom A.4 is satisfied. In conclusion, M (C, t) satisfies the four capitalization
axioms.

12.2. Axiom A.1 is satisfied since M depends only on C ≥ 0 and t ≥ 0. We observe that
Rt
M (C1 + C2 , t) = (C1 + C2 ) e 0 a(s) ds
Rt Rt
= C1 e 0 a(s) ds + C2 e 0 a(s) ds = M (C1 , t) + M (C2 , t)
and we see that axiom A.2 is satisfied. For every t1 , t2 ≥ 0, we assume t1 ≤ t2 . Since
a(t) ≥ 0, we have
Z t1 Z t1 Z t2 Z t2
a(s) ds ≤ a(s) ds + a(s) ds = a(s) ds
0 0 t1 0

It follows that
R t1 R t2
M (C, t1 ) = C e 0 a(s) ds ≤ C e 0 a(s) ds = M (C, t )
2

by which we observe that M is monotonically increasing with respect to the lifetime t and
consequently, axiom A.3 is satisfied. Finally, setting t = 0 we have that M (C, 0) = C and
therefore, axiom A.4 is satisfied. In conclusion, M (C, t) satisfies the four capitalization
axioms.

12.3 If t = 0, we have that f (0) = 1. For every t1 , t2 > 0, assuming t1 < t2 we have
√ t1 √ √
r
f (t1 ) = 1 + 0.12 t1 = 1 + 0.12 t2 < 1 + 0.12 t2 = f (t2 )
t2
by which f is strictly monotonically increasing on [0, +∞). Thus, f defines an accumulation
factor on [0, +∞).
An alternative approach is to consider that f is differentiable ∀t > 0 and f ′ (t) > 0 ∀t > 0,
so f is strictly monotonically increasing on [0, +∞).
84 CHAPTER 12. ELEMENTARY FINANCIAL CALCULUS

12.4. We can proceed as in the previous exercise. Observe that for the monotonicity condition
we must consider three different cases: (a) 0 ≤ t1 ≤ t2 < 1; (b) 0 ≤ t1 ≤ 1 ≤ t2 ; (c)
1 ≤ t1 ≤ t2 . It can then be confirmed that f is monotonically increasing (though not
strictly) on [0, +∞).
3 9


12.5. (a) 1000 1 + 100 12 = 1022.5
(b) From the stated requirement:
 
3 9
1000 1 + (1 + δ) = 1022.5(1 + 15%)
100 12
We find that δ = 23.45%.
4 3


12.6. (a) 1000 1 + 100 = 1169.86
(b) From the stated requirement:
 3
4
1000 1 + (1 + δ) = 1169.86(1 + 20%)
100
We find that δ = 8.85%.

12.7. (a) From the nominal interest rate we find that the semi-annual compound interest rate
is i2 = T AN
2 = 0.06. We then calculate the equivalent annual compound interest rate
i = (1 + i2 )2 − 1 = 0.1236
(b) V0 = 1624.14
(c) M = 4635.85
(d) R = 220.67

12.8. (a) We have

R R R R
A(R, i) = + + ··· + + + ···
1 + i (1 + i)2 (1 + i)n−1 (1 + i)n
 
R R R R
= lim + + ··· + +
n→+∞ 1 + i (1 + i)2 (1 + i)n−1 (1 + i)n
1 − (1 + i)−n R
= R lim =
n→+∞ i i
(b) If i < i′ , since we have constant installments R, then it certainly follows that:
R R
′
<
i i
Therefore, the present value of an ordinary perpetuity decreases as the annual interest rate
increases.
(c) We have Ri = 0.0325
150
= 4615.38.

12.9 (a) The bond has the following sequence of cash flows:

{(100, 1), (1100, 2)}

(b) The bond’s DCF is

100 1100
G(x) = −A + +
1 + x (1 + x)2
 12.2. SOLUTIONS 85

(c) From the condition that G(0.1) = 0, we have

100 1100
−A + + =0 ⇒
1 + x (1 + x)2
100 1100 100 1100 110 + 1100 1210
A= + 2
= + 2
= 2
= = 1000
1 + x (1 + x) 1.1 (1.1) (1.1) 1.21
(d) The duration of the bond is

100 · 1.1−1 + 1000 · 1.1−2 · 2

D(0.035) = = 1.744
1000

12.10. (a) The bond has the following sequence of cash flows:
     
2 8 14
30, , 30, , 1030,
12 12 12

(b) The bond’s DCF is

30 30 1030
G(x) = −A + 2 + 8 + 14
(1 + x) 12 (1 + x) 12 (1 + x) 12

(c) Letting A denote the nominal value of the bond, we have

30 30 1030
A= 2 + 8 + 14 = 1048.63
(1 + 0.035) 12 (1 + 0.035) 12 (1 + 0.035) 12

(d) The duration of the bond is

2 8 14
30 · 1.035− 12 12
2
+ 30 · 1.035− 12 12
8
+ 1030 · 1.035− 12 14
12
D(0.035) = = 1.1242
1048.63
Chapter 13

Asset pricing

13.1 Exercises
13.1. In the market (L, p), with p = (16, 36, 52), there are three financial assets whose payoffs,
under the two possible scenarios, are described by the following vectors:
     
10 30 40
y1 = y2 = y3 =
20 40 60
 
120
(a) Determine whether the contingent claim w = is replicable.
200
(b) Show that the market is complete.
(c) Calculate the price kernel vector.
(d) Show that the law of one price (LOP) holds.

13.2. In the market (L, p), with p = (9, 6), there are two financial assets whose payoffs, under
the three possible scenarios, are described by the following vectors:
   
8 3
y1 = 10
  y2 = 5 

16 11

(a) Determine if the market is complete.

(b) Find, if they exist, all the portfolios that guarantee a certain payoff of 50 under any
scenario and calculate their price.
(c) Show that the law of one price holds.

13.3. In the market (L, p) there are three financial assets whose payoffs, under the three possible
scenarios, are described by the following vectors:
     
−2 5 3
y1 =  0  y2 = 10 y3 = 10
3 10 13

The price vector is p = (p1 , p2 , p3 ) with p1 , p2 , p3 > 0, p1 < p2 and p3 = 2p2 .

87
88 CHAPTER 13. ASSET PRICING

(a) Determine if the market is complete.

(b) Show that the LOP does not hold in this market.
(c) Show that the portfolio x = (−1, 2, −1) generates a Type I arbitrage.

13.4. In the market (L, p) there are three securities whose annual yield depends on which scenario
transpires in a year, up or down. In particular:

• Stock 1: today’s price 20, annual yield +20% up scenario and −50% down scenario.
• Stock 2: today’s price 10, annual yield +30% up scenario and −60% down scenario.
• Stock 3: today’s price 15, annual yield +20% up scenario and −20% down scenario.

(a) Write the payoff matrix Y that describes the market and the price vector p.
(b) Determine if the market defined above is complete.

13.5. In the market (L, p) there are two assets whose payoffs are described in the following
matrix:  
2 4
Y = 3 6
4 8

(a) Show that with the price vector p = (1, 3) the market permits Type I and Type II
arbitrage opportunities.
(b) Show that with the price vector p = (1, 2) the market does not permit Type I or Type
II arbitrage opportunities.

13.6. In the market (L, p) there are two assets whose payoffs are described in the following
matrix:  
3 4
Y =
0 1
The price vector is p = (1, 2).

(a) Show that the market is complete.

(b) Calculate the prices of the portfolios that replicate the Arrow contingent claims.
(c) Determine if the market permits arbitrage opportunities.

13.7. In the complete market (L, p) the payoffs of the assets are described in the following matrix:
 
3 1 4
Y = 2 1 3
1 0 1

The vector representation of the pricing rule is π = 13 , 0, 1 .

(a) Calculate the price vector.

(b) Determine if the law of one price holds.
(c) Determine if the market permits arbitrage opportunities.
 13.1. EXERCISES 89

13.8. In the market (L, p) there are three financial assets whose payoffs, under two possible
scenarios, are described by the following vectors:
     
10 16 13
y1 = y2 = y3 =
5 7 6

The price vector is p = (8, 12, 8).

(a) Determine if the market is complete.

(b) Discuss if the LOP holds for this market.
(c) Given that the contingent claim w = (0, 0) is replicated by portfolios of the form
(a, a, −2a) with a ∈ R, find, if it exists, a Type I arbitrage.

13.9. Consider the market (L, p), with p = (20, 10α, 40) and α ∈ R, represented by the payoff
matrix  
20 40 80
Y = 20 20 60
20 60 20

(a) Identify how many financial assets are present in the market, how many scenarios can
occur, and the corresponding payoffs of each financial asset.
(b) Determine if the market is complete.
(c) Determine the values of the parameter α such that the market does not permit Type
I or Type II arbitrage opportunities.

13.10. Consider the market (L, p) with n financial assets and k possible scenarios.
Let Y be the k × n payoff matrix.
 
Y
(a) Prove that, given M = , the (k + 1) × n matrix obtained by augmenting Y with
p
the row of prices:

rank Y = rank M ⇐⇒ the law of one price holds

(b) Use the above result to determine if the LOP holds in the market described by the
payoff matrix
 
10 2 15
Y =
20 8 20

and the price vector p = (16, 4, 17).

(c) Use the above result to determine if the LOP holds in the market described by the
payoff matrix
 
6 4 2
Y =
12 4 7

and the price vector p = (9, 2, 6).

90 CHAPTER 13. ASSET PRICING

13.2 Solutions
13.1. (a) The question of whether w is replicable is equivalent to asking whether w ∈ span{y1 , y2 , y3 },
which is equivalent to the question of whether the system Y x = w admits solutions:
 
 x
10 30 40  1 
  
120
x2 =
20 40 60 200
x3
This is a linear system in which rank Y = rank(Y |w) = 2, so it admits infinitely
many solutions of the form:
 
14 + k
−2 − 2k  with k ∈ R
k

For example, when k = 1, in order to replicate the contingent claim w we buy 15 units
of the first asset, sell 4 units of the second asset and buy 1 unit of the third asset.
(b) Since rank Y = 2, we have that the vector subspace of replicable contingent claims
W = span{y1 , y2 , y3 } has dimension 2, which coincides with R2 and the market is
complete.
(c) The price kernel vector is obtained by solving the following linear system:

(π1 , π2 )Y = (p1 , p2 , p3 )
 
10 30 40
(π1 , π2 ) = (16, 36, 52)
20 40 60

10π1 + 20π2 = 16 
π1 = 0.4
30π1 + 40π2 = 36
π2 = 0.6
40π1 + 60π2 = 52


(d) Since the market is complete and the price kernel is strongly positive, i.e., (π1 , π2 ) ≫
(0, 0), there are no arbitrage opportunities and the LOP holds.

13.2. (a) The payoff matrix is of dimension 3 × 2 and we have:

 
8 3
Y = 10 5
16 11

Y has rank 2, which is less than the number of possible scenarios, so the market is
incomplete.
 
50
(b) We must verify that the contingent claim w = 50 is replicable. This equates to
50
solving the linear system:    
8 3   50
10 5  x1 = 50 ,
x2
16 11 50
13.2. SOLUTIONS 91
 
50
The systems admits solutions since rank Y = rank(Y |w), so w = 50 is replicable.
50
   
x1 10
Solving the linear system yields the replicating portfolio: = .
x2 −10
   
9 10
Its price is p · x = · = 30.
6 −10
(c) We observe that when the system Y x = w admits solutions, they are unique. So, each
replicable contingent claim has a unique replicating portfolio associated with it. Thus,
the price is unique.

13.3. (a) We consider the payoff matrix:

 
−2 5 3
Y =  0 10 10
3 10 13

Y has rank 2, so the market is not complete.

(b) We observe that the asset y3 is equal to the sum of the assets y1 and y2 . If the pricing
rule were linear, then we would have that f (y3 ) = f (y1 ) + f (y2 ), i.e., p3 = p1 + p2 .
But since p1 < p2 we have that p3 = p1 + p2 < 2p2 and the condition p3 = 2p2 would
not be satisfied. The pricing rule is not linear and it follows that the law of one price
does not hold in this market.
(c) By the theorem that states that in a market where there are no Type I arbitrage
opportunities, the LOP must hold, we deduce that in this market it is possible to find
a portfolio x that generates a Type I arbitrage, namely:
(
Yx≥0
p·x<0

Verifying that x = (−1, 2, −1) satisfies the system described above:

      
−2 5 3 −1 9 0
Y x = 0 10 10
   2 = 10 ≥ 0
   
3 10 13 −1 4 0

p · x = (p1 , p2 , p3 ) · (−1, 2, −1) = −p1 + 2p2 − p3 =

= −p1 + 2p2 − 2p2 = −p1 < 0

13.4. (a) The payoffs of the three assets under the two scenarios are:
           
20 · 1.2 24 10 · 1.3 13 15 · 1.2 18
y1 = = y2 = = y3 = =
20 · 0.5 10 10 · 0.4 4 15 · 0.8 12

So, the payoff matrix is:  

24 13 18
Y =
10 4 12
The vector of today’s prices is p = (20, 10, 15).
92 CHAPTER 13. ASSET PRICING

(b) Since the rank of matrix Y is 2, we conclude that the market is complete.

13.5. (a) A Type I arbitrage is realized when:

(
Yx≥0
p·x<0

A Type II arbitrage is realized when:

(
Yx>0
p·x≤0

Calculating the vector Y x:

   
2 4   2x1 + 4x2
x
Y x = 3 6 1 = 3x1 + 6x2 
x2
4 8 4x1 + 8x2

and the scalar p · x:

(1, 3) · (x1 , x2 ) = x1 + 3x2
We have a Type I arbitrage when:
   
2x1 + 4x2 0
3x1 + 6x2  ≥ 0 and x1 + 3x2 < 0
4x1 + 8x2 0

That is, when:

x1 ≥ −2x2 and x1 < −3x2
which are satisfied, for example, when x1 = 2 and x2 = −1.
We have a Type II arbitrage when:
   
2x1 + 4x2 0
3x1 + 6x2  > 0 and x1 + 3x2 ≤ 0
4x1 + 8x2 0

That is, when:

x1 > −2x2 and x1 ≤ −3x2
which are satisfied, for example, when x1 = 3 and x2 = −1.
(b) We have:    
2 4   2x1 + 4x2
x
Y x = 3 6 1 = 3x1 + 6x2 
x2
4 8 4x1 + 8x2
and
p · x = (1, 2) · (x1 , x2 ) = x1 + 2x2
We have a Type I arbitrage when:
   
2x1 + 4x2 0
3x1 + 6x2  ≥ 0 and x1 + 2x2 < 0
4x1 + 8x2 0
13.2. SOLUTIONS 93

That is, when:

x1 ≥ −2x2 and x1 < −2x2
which are never satisfied.
We have a Type II arbitrage when:
   
2x1 + 4x2 0
3x1 + 6x2  > 0 and x1 + 2x2 ≤ 0
4x1 + 8x2 0
That is, when:
x1 > −2x2 and x1 ≤ −2x2
which are never satisfied.

13.6. (a) Since the rank of matrix Y is 2, the dimension of the space of replicable contingent
claims is 2 and the market is complete.
(b) Since the market is complete, it is certainly true that the Arrow contingent claims are
replicable. To find the replicating portfolios, we must solve the following two systems:
   
1 0
Yx= and Y x =
0 1
 
1/3 1
The solution to the first system is the portfolio x1 = , whose price is p · x1 = .
0 3
 
−4/3
The solution to the second system is the portfolio x2 = , whose price is
1
2
p · x2 = .
3
(c) The prices of the portfolios that replicate the Arrow
contingent claims are precisely
the components of the price kernel vector, π = 31 , 23 . Since the vector π is strongly
positive, by the fundamental theorem of finance there are no arbitrage opportunities
(of either type) in the market.

13.7. (a) From the relation:

πY = p
 
  3 1 4
1
, 0, 1 2 1 3 = (p1 , p2 , p3 )
3
1 0 1
we have that 

 p1 = 2



p2 = 31




p3 = 73


(b) Since the price kernel vector is positive (π ≥ 0) there are no Type I arbitrage
opportunities in the market and therefore the law of one price holds.
(c) Since the price kernel is positive (π ≥ 0) but not strongly, there are Type II arbitrage
opportunities in the market.
94 CHAPTER 13. ASSET PRICING

13.8. (a) The payoff matrix is given by:

 
10 16 13
Y =
5 7 6

Y has rank 2, so the market is complete.

(b) We observe that y3 = 12 y1 + 12 y2 but p3 = 8, the price of the third asset, is different
from 12 p1 + 12 p2 = 10. The pricing rule is not linear and thus the LOP does not hold.
(c) Since the law of one price does not hold, there are Type I arbitrage opportunities in
the market.
The price of the portfolios of the form (a, a, −2a) is equal to (8, 12, 8) · (a, a, −2a) = 4a.
For example, if a = −1 we obtain that the price of the portfolio is 4 < 0, and we have
constructed a Type I arbitrage.

13.9. (a) In the payoff matrix Y , the number of columns represents the number of financial
assets and the number of rows represents the number of possible scenarios. In this
case, Y is a 3 × 3 matrix, so there are 3 assets and 3 possible scenarios. In particular,
the respective payoffs of each of the 3 assets are:
     
20 40 80
y1 = 20
 y2 = 20
 y3 = 60

20 60 20

Observation: The payoffs of asset 1 are all equal, independent of the scenario. We
can interpret this asset as certain, since its future payoff is fixed.
(b) Given W = span{y1 , y2 , y3 }, the market is complete when dim W = 3 (the number of
scenarios). In this case we have that dim W = rank Y = 3, so the market is complete.
(c) To ensure both no-arbitrage conditions, we can invoke the fundamental theorem of
finance and impose that the price kernel must be strongly positive, i.e., π ≫ 0. We
have:
(π1 , π2 , π3 )Y = (p1 , p2 , p3 )

 20π1 + 20π2 + 20π3 = 20
40π1 + 20π2 + 60π3 = 10α
80π1 + 60π2 + 20π3 = 40


 
 π1 + π2 + π3 = 1  π1 = 1 − π2 − π3
4π1 + 2π2 + 6π3 = α 4(1 − π2 − π3 ) + 2π2 + 6π3 = α
4π1 + 3π2 + π3 = 2 4(1 − π2 − π3 ) + 3π2 + π3 = 2
 

 
 π1 = 1 − π2 − π3  π1 = 1 − (2 − 3π3 ) − π3
−2π2 + 2π3 = α − 4 −2(2 − 3π3 ) + 2π3 = α − 4
π2 + 3π3 = 2 π2 = 2 − 3π3
 

 π1 = α4 − 1
 
π1 = −1 + 2π3
8π3 = α π2 = 2 − 3α8
π2 = 2 − 3π3 π3 = α8
 
 13.2. SOLUTIONS 95

For the price kernel to be strongly positive we must have:

α
 −1>0
 4


3α
2− >0

 α 8
>0


8
That is: 
α>4
α < 16
3
α>0


yielding 4 < α < 16

3 .
Observation: If we had instead imposed the condition of a price kernel that is only
positive (π ≥ 0, which would yield 4 ≤ α ≤ 16
3 ) we would have ensured only the Type
I no-arbitrage condition.

13.10. (a) Let w be a replicable contingent claim for which we have two distinct portfolios
x1 , x2 such
 that  Y x2 = w. Then, Y (x1 −x2 ) = 0. Weconsider
 Y x1 = w and
w w 0
M x1 = and M x2 = . Then, M (x1 − x2 ) = .
p · x1 p · x2 p · x1 − p · x2
(⇒) Suppose, by contradiction, that the LOP does not hold, i.e., p · x1 ̸= p · x2 . Then,
the last row of matrix M is linearly independent of the others and rank M > rank Y ,
which contradicts our hypothesis.
(⇐) Suppose, by hypothesis, that the LOP holds, i.e., p·x1 −p·x2 = 0. We have that if
Y (x1 − x2 ) = 0, then M (x1 − x2 ) = 0, and thus x1 − x2 ∈ ker Y since x1 − x2 ∈ ker M .
That is, we have that dim ker Y ≤ dim ker M . So, by the rank-nullity theorem, we
have n − rank Y ≤ n − rank M , thus rank Y ≥ rank M . But since Y is a submatrix
of M we have that rank Y ≤ rank M , therefore rank Y = rank M .
(b) Matrix Y has rank 2. We consider matrix
 
10 2 15
M = 20 8 20
16 4 17

whose rank is 3. Thus, the LOP does not hold in this market.
(c) Matrix Y has rank 2. We consider matrix
 
6 4 2
M = 12 4 7
9 2 6

whose rank is 2. Thus, the LOP holds in this market.