2016febreal Analysis Problems
2016febreal Analysis Problems
Problems
CLAUDIA TIMOFTE
REAL ANALYSIS
Problems
To my students
Preface
Claudia Timofte
7
Contents
1 Real Numbers 11
1.1 Elements of Topology of the Real Line. Elementary Inequalities . . . 11
1.2 Number Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2 Multidimensional Spaces 29
2.1 Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2 Elements of Topology . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.3 Sequences in Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . 39
2.4 Inner Product Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.5 Normed Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
9
10 CONTENTS
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
Chapter 1
Real Numbers
11
12 REAL ANALYSIS
Definition 1.1 For any real number a, the absolute value or the modulus of a is
defined as follows:
{
a, if a ≥ 0,
|a| =
−a, if a < 0.
As can be seen from the above definition, the absolute value of a is always positive.
All the above sets are bounded intervals. We can also consider the following
unbounded intervals: (a, ∞) = {x ∈ R | x > a}; [a, ∞) = {x ∈ R | x ≥ a};
(−∞, b) = {x ∈ R | x < b}; (−∞, b] = {x ∈ R | x ≤ b}; (−∞, ∞) = R.
Let us mention that if we take a, b ∈ R, with a = b, we obtain degenerate
intervals. For instance, (a, a) = ∅ (the empty set) and [a, a] = {a} (a singleton).
Definition 1.6 A set A ⊆ R is said to be an open set if for any x ∈ A there exists
r > 0 such that B(x, r) ⊆ A.
Definition 1.11 (The Extended Real Number Line R) Let −∞ and ∞ be two dif-
ferent elements which are not real numbers. They are called points at infinity or
improper numbers. Let R = R ∪ {−∞} ∪ {∞}. We extend to R the usual order
relation defined on R by putting −∞ < x < ∞, for any x ∈ R, and −∞ < ∞. The
set R, endowed with the above order relationship, is called the extended real line.
Definition 1.13 For some pairs of elements from R, we extend the usual operations
of addition and multiplication in R as follows:
1) a + ∞ = ∞ + a = ∞, ∀a ∈ R, a ̸= −∞;
3) a · ∞ = ∞ · a = ∞, ∀a > 0;
5) b · ∞ = ∞ · b = −∞, ∀b < 0,
−∞ < a < ∞.
The element −∞ is called minus infinity or negative infinity, while the element ∞
is called plus infinity or positive infinity.
14 REAL ANALYSIS
The extended real number system is also called the closed real line and it is
denoted by R = [−∞, +∞]. Intervals in R are sets I ⊆ R of the form (a, b), [a, b], (a, b]
or [a, b), where a, b ∈ R and a < b. So, R = [−∞, +∞].
If a < b, a, b ∈ R, then
(a, b), (a, ∞), (a, ∞], (−∞, a), [−∞, a), R, R,
Exercise 1.16 Show that the absolute value has the following properties:
√
1) |a| = a2 ;
2) |a| = 0 if and only if a = 0;
3) |a + b| ≤ |a| + |b|;
4) | − a| = |a|;
5) ||a| − |b|| ≤ |a − b|, for any a, b ∈ R.
∩
k
Exercise 1.17 Let A1 , A2 , . . . , Ak ∈ R be open sets. Prove that Ai is an open
i=1
set.
Exercise 1.18 Let {Ai }i∈I be an arbitrary family of open sets in R. Prove that
∪
Ai is also an open set.
i∈I
Exercise 1.19 Prove that any open interval (a, b) is an open set.
Exercise 1.20 Show that any closed interval [a, b] is a closed set.
∪
k
Exercise 1.21 Let B1 , B2 , . . . , Bk ⊆ R be closed sets. Prove that Bi is also a
i=1
closed set.
Exercise 1.22 Let {Bi }i∈I be an arbitrary family of closed sets in R. Prove that
∩
Bi is also a closed set.
i∈I
REAL NUMBERS 15
Exercise 1.25 Prove that for any nonnegative real numbers x1 , x2 , . . . , xk , we have
x1 + x2 + · · · + xk √
≥ k x1 · x2 · · · xk
k
and the equality holds true if and only if x1 = x2 = · · · = xk . In other words, the
arithmetic mean is greater than or equal to the geometric mean and they are equal
if and only if all the numbers xi are equal.
The above inequalities can be generalized to power means and weighted power
means.
Hint. Use the fact that the function f : [−1, ∞) → R, defined by f (x) = (1 + x)b −
bx − 1, is monotone and attains a minimum equal to zero at the point x = 0.
Exercise 1.28 Prove that for any a ∈ R, with a ≥ −1 and any n ∈ N∗ , we have
(1 + a)k ≥ 1 + k a.
Exercise 1.29 Prove that for any a1 , a2 , . . . , ak ∈ R, with ai ≥ −1 for any i and
having all the same sign, we have
(1 + a1 ) (1 + a2 ) · · · (1 + ak ) ≥ 1 + a1 + a2 + · · · + ak .
Exercise 1.31 Prove that for any a1 , a2 , . . . , ak ∈ R∗+ , with ai > 0 and such that
a1 · a2 · · · ak = 1, we have
(1 + a1 ) (1 + a2 ) · · · (1 + ak ) ≥ 2k .
Exercise 1.32 Prove that for any x > 0 and any k ∈ N, we have
xk 1
≤ .
1 + x + + · · · + x2k
x2 2k + 1
√
2k+1 1 + x + x2 + · · · + x2k
xk = 1 · x · x2k ≤ .
2k + 1
Solution. The first inequality can be immediately proven by induction. The second
one is an easy consequence of the geometric-arithmetic mean inequality.
√ √ k+1
k≤ k! ≤ ∀k ∈ N∗ .
k
,
2
bc ac ab
+ + ≥a+b+c
a b c
and
1 1 1 1 1 1
+ + ≥√ +√ +√ .
a b c bc ca ab
Definition 1.37 Let A be a nonempty set and (ak )k∈N be a given sequence in A.
A subsequence of (ak )k∈N is a sequence of the form (akp )p∈N , where (kp ) ⊆ N is a
strictly increasing sequence, for any p ∈ N.
Definition 1.38 A sequence (ak )k∈N of real numbers is called bounded if there exists
a real number M ≥ 0 such that |ak | ≤ M, ∀k ∈ N.
Example 1.39 For instance, the sequence ak = (−1)k is bounded, since |ak | ≤ 1,
for all k ∈ N.
Remark 1.40 If there exist two real numbers m and M such that m ≤ ak ≤ M , for
any k ∈ N, then the sequence (ak ) is said to be bounded from above and from below.
M is called the upper bound and m the lower bound. A sequence (ak is bounded if
and only if (ak ) has an upper and a lower bound.
Definition 1.41 A sequence (ak ) is said to be unbounded if for any positive number
M , there exists kM ∈ N such that |akM | > M .
Definition 1.43 A sequence (ak )k∈N of real numbers is called convergent if there
exists a ∈ R such that for any neighborhood V of a, there exists nV ∈ N such that
ak ∈ V , for any k ≥ kV .
The element a ∈ R is called the limit of the sequence (ak )k∈N and we shall symbolize
this as follows:
a = lim ak .
k→∞
REAL NUMBERS 19
Proposition 1.44 A convergent sequence of real numbers has only one limit.
Proposition 1.45 Let (ak )k∈N be a sequence of real numbers. The sequence (ak )k∈N
is convergent to a ∈ R if and only if for any positive number ε, there exists kε ∈ N
such that
|ak − a| < ε, ∀k ≥ kε .
Proposition 1.46 Let (ak )k∈N and (bk )k∈N be two given sequences of real numbers
with |ak | ≤ bk , for any k ∈ N. If we assume that bk → 0, then ak → 0. As a
consequence, if (ak )k∈N , (bk )k∈N and (ck )k∈N are three given sequences of real numbers
with ak ≤ bk ≤ ck , for any k ∈ N, and if we assume that (ak )k∈N and (ck )k∈N converge
to a certain limit l ∈ R, then (bk )k∈N is convergent to l, as well.
Remark 1.47 If (ak )k∈N and (bk )k∈N are convergent sequences of real numbers,
then:
lim (ak + bk ) = lim ak + lim bk ,
k→∞ k→∞ k→∞
Some authors use another terminology; more precisely, they call a sequence (ak )k∈N
of real numbers nonincreasing if ak ≥ ak+1 , ∀k ∈ N. A sequence (ak )k∈N of real
numbers is called nondecreasing if ak ≤ ak+1 , ∀k ∈ N.
20 REAL ANALYSIS
Theorem 1.57 Let (ak )k∈N be a sequence of real numbers. The sequence (ak )k∈N is
convergent if and only if it is a fundamental one.
ak ≥ M, ∀k ≥ kM .
ak ≤ M, ∀k ≥ kM .
bk = inf aj , k ≥ 1.
j≥k
ck = sup aj , k ≥ 1,
j≥k
The element l ∈ R defined by (1.1) is called the limit inferior of the sequence
(ak )k and it is denoted by lim inf ak or by lim ak .
k→∞ k→∞
The element L ∈ R defined by (1.2) is called the limit superior of the sequence
(ak )k and it is denoted by lim sup ak or by lim ak .
k→∞ k→∞
Hence, the inferior and, respectively, superior limits of the sequence (ak )k≥1 are:
and
lim sup ak = inf sup aj .
k→∞ k≥1 j≥k
Obviously,
−∞ ≤ lim inf ak ≤ lim sup ak ≤ +∞
k→∞ k→∞
Also, it is easy to see that a sequence of real numbers (ak )k≥1 is convergent in R
if and only if
lim inf ak ∈ R,
k→∞
lim sup ak ∈ R
k→∞
and
lim inf ak = lim sup ak .
k→∞ k→∞
Let us notice that if (ak )k≥1 is a bounded sequence of real numbers, then its limit
superior can be viewed as being the largest limit of any subsequence of (ak )k≥1 . Of
course, a similar statement is true for the limit inferior. In fact, the limit superior of
an arbitrary sequence can be viewed as being the largest subsequential limit of all
its convergent subsequences, including those possible subsequences tending to +∞
or −∞. Similarly, the limit inferior is the smallest subsequential limit. This remark
is better described in the following theorem:
Theorem 1.60 Let (ak )k be an arbitrary sequence. Let l and, respectively, L be its
limit inferior and, respectively, superior. Let (akp ) be an arbitrary subsequence of
(ak ) such that lim akp exists, it is +∞ or −∞. Then,
p→∞
l ≤ lim akp ≤ L.
p→∞
L = lim akp
p→∞
l = lim a ′ kp .
p→∞
REAL NUMBERS 23
ak+1 √ √ ak+1
lim ≤ lim k ak ≤ lim k ak ≤ lim .
k→∞ ak k→∞ k→∞ k→∞ ak
ak+1
Moreover, if there exists lim , then
k→∞ ak
ak+1 √
lim = lim k ak .
k→∞ ak k→∞
k2 + 1
ak = , k≥1
k
is strictly increasing.
Solution. Since
k2 + 1 1
=k+ ,
k k
we obtain
1
ak+1 − ak = 1 − > 0.
k(k + 1)
Therefore, ak+1 > ak , i.e. our sequence is strictly increasing.
k
ak = , k≥1
2k
Solution. Since
k+1
ak+1 = ak < ak ,
2k
for k > 1, it follows that the sequence (ak )k is strictly decreasing. Since, obviously,
(ak )k is bounded from below by 0, it follows that (ak )k is convergent. Let a =
1
lim ak . Then, we get a = a, which means that a = 0.
k→∞ 2
24 REAL ANALYSIS
√
Exercise 1.64 Let a1 = 2 and
√ √
ak+1 = 2+ ak , ∀k ≥ 1.
∑
k
ak = j aj , ∀k ≥ 1.
j=1
a
Prove that (ak )k is convergent and has the limit equal to .
(1 − a)2
Exercise 1.66 Let
a1 = 1,
a = k (a
k k−1 + 1).
Show that ( )
∏
k
1
lim 1+ = e ( Euler’s number).
k→∞
j=1
aj
Exercise 1.67 Let (ak )k≥1 be a convergent sequence having the limit equal to a.
Prove the existence and find the value of the following limit:
1∑ k
l = lim xi .
k→ k i=1
Prove a similar result for the geometric mean of the first k terms of a sequence of
strictly positive real numbers.
REAL NUMBERS 25
Exercise 1.68 Let (ak )k≥1 be given by the following recurrence formula:
a1 = a2 = 1,
1
ak+1 = , ∀k ≥ 2.
1
ak−1 +
ak
Exercise 1.69 Let (ak )k≥1 be defined by the following recurrence formula:
a0 = 0,
1( )
ak+1 = a + a2k , ∀k ≥ 0,
2
where a ∈ [0, 1]. Prove that this sequence is convergent and find its limit.
ln k!
a) ak = , p ≥ 2;
kp
∑
k
1
b) ak = (−1)i−1 .
i=1
i
∑
k
1
ak = − ln k ( Euler’s sequence ).
i=1
i
Solution. We prove that the sequence (ak )k is monotone and bounded. To this end,
using Lagrange’s formula for the function ln : [k, k + 1] → R, for k > 0, it follows
that there exists ξ ∈ (k, k + 1) such that ln (k + 1) − ln k = 1/ξ. Then, since
1 1 1
< < ,
k+1 ξ k
we get
1 1
< ln (k + 1) − ln k < .
k+1 k
26 REAL ANALYSIS
1 1 1
ak = 1+ + · · ·+ −ln k > (ln 2−ln 1)+(ln 2−ln 1)+· · ·+(ln(k + 1) − ln k)−ln k =
2 3 k
ln(k + 1) − ln k > 0.
Thus, the sequence (ak )k is bounded from below, and, hence, convergent.
is convergent.
Solution. It is not difficult to see that (ak )k is a Cauchy sequence in R. Hence, (ak )k
is convergent.
k!
ak = , k ≥ 1,
(2k + 1)!!
for k > 1. Thus, it follows that the sequence (ak )k is strictly decreasing. Since,
obviously, (ak )k is bounded from below by 0, it follows that (ak )k is convergent. Let
1
a = lim ak . Then, we get a = a, which means that a = 0.
k→∞ 2
5k 3 + 4k + 2
Exercise 1.76 Let ak = , for k ∈ N. Prove that ak → 5/4 as k → ∞.
4k 3 + 7k 2 + 5
√
Exercise 1.77 Find lim ( k 2 + 4k − k).
k→∞
( 2
)
sin 4 + π k
Exercise 1.78 Let ak = , for each k ≥ 1. Show that ak → 0 as
k3
k → ∞.
Solution. The sequence ak is bounded and increasing, and, hence, convergent. Let
us denote by l its limit. Passing to the limit in the recurrence formula, we get l = 2.
√
Exercise 1.80 Let a1 = −10 and ak+1 = 1 − 1 − ak . Prove that (ak )k is conver-
gent and find its limit.
1 · 1! + 2 · 2! + · · · k · k!
lim .
k→∞ (k + 1)!
1 · 1! + 2 · 2! + · · · k · k! (k + 1)! − 1
lim = lim = 1.
k→∞ (k + 1)! k→∞ (k + 1)!
a1 + a2 + · · · ak
lim = a.
k→∞ k
28 REAL ANALYSIS
Exercise 1.83 Let (ak )k be a positive sequence which converges to a. Prove that
√
lim k
a1 · a2 · · · ak = a.
k→∞
1 1
a”k = 1 + ... + + ,
2 k
is divergent.
Solution. Obviously,
inf aj = 0 and sup aj = ∞.
j≥k j≥k
Therefore,
lim ak = 0 and lim ak = ∞.
k→∞ k→∞
Exercise 1.86 Find the limit superior and the limit inferior for the following se-
quences:
1) ak = (−1)k ; 2) ak = k (−1)k ;
( )
1
3) ak = k + k (−1)k ; 4) ak = (−1)k a + , a ≥ 0.
k
Chapter 2
Multidimensional Spaces
We start this chapter with a brief review of some useful theoretical results about
metric spaces.
Remark 2.2 Any distance takes only positive values, i.e. d(x, y) ≥ 0, for all x, y ∈
X. Hence, d : X × X → R+ .
29
30 REAL ANALYSIS
is a distance on X, called the discrete metric. In this case, the space (X, d) is called
a discrete metric space or a space of isolated points.
∑
n
d1 (x, y) = |xi − yi | (the Manhattan distance);
i=1
∫b
d(f, g) = |f (x) − g(x)| dx
a
Solution. We notice that due to the fact that d is a distance, i.e. d(x, y) ≥ 0, ∀x, y ∈
M , d1 is well defined. Let us verify that it satisfies all the three axioms in Definition
2.1.
(i) We have to prove that d1 (x, y) = 0 ⇔ x = y. Obviously, d1 (x, y) = 0 ⇔
ln(1 + d(x, y)) = 0 ⇔ 1 + d(x, y) = 1 ⇔ d(x, y) = 0 ⇔ x = y.
(ii) We have to prove that d1 (x, y) = d1 (y, x). But d(x, y) = d(y, x), and hence,
ln(1 + d(x, y)) = ln(1 + d(y, x)), i.e. d1 (x, y) = d1 (y, x).
(iii) We shall prove that for any x, y, z ∈ X, we have d1 (x, z) ≤ d1 (x, y)+d1 (y, z).
Indeed, using the monotonicity of the ln and the fact that d satisfies the triangle
inequality, we have
is a distance.
|x − y|
d(x, y) =
1 + |x − y|
is a distance.
32 REAL ANALYSIS
Exercise 2.13 Let d(x, y) = |x2 − y 2 |, for any x, y ∈ R. Show that d is not a
distance on R.
Exercise 2.14 Let d be the Euclidean metric on R2 . Compute d(x, y) for x = (1, 2)
and y = (3, 1) and draw the open ball Br (a) for a = (0, 0) and r = 2.
Exercise 2.15 Let d1 be the Manhattan (taxicab) metric on R2 . Compute d1 (x, y),
for x = (3, 2) and y = (4, 1), and sketch the open ball Br (a) for this metric when
a = (0, 0) and r = 2.
Exercise 2.17 Let d be the usual metric on R2 and 0 be the origin. We define ρ
on R2 as follows:
{
d(x, y), if x and y are collinear with 0,
ρ(x, y) =
d(x, 0) + d(0, y), otherwise.
Exercise 2.18 Let ρ be the Greek airline metric on R2 and let x = (1, 2) and
y = (3, 4). Compute ρ(x, y) and sketch Br (a) for this metric when a = (0, 0) and
r = 2.
is called the open ball with center at the point x and having the radius r.
Definition 2.21 Let (X, d) be a metric space. Also, let x ∈ X and r ≥ 0. The set
B(x, r) = {y ∈ X | d(x, y) ≤ r}
is called the closed ball with center at the point x and having the radius r.
Definition 2.23 Let (X, d) be a metric space, a ∈ X and r > 0. The set
S(a, r) = {x ∈ X | d(a, x) = r}
is called the sphere centered at the point a and having the radius r.
Definition 2.28 Let (X, d) be a metric space and A, V ⊆ X. The set V is called a
neighbourhood of A if there exists an open set D such that A ⊆ D ⊆ V . If A = {x},
V is called a neighbourhood of the point x. We shall denote by V(A) the collection
of all the neighbourhoods of A and by V(x) the collection of all the neighbourhoods
of x.
We shall denote by Int A or by Å the set containing all the interior points of A.
Let us notice that the interior of a set depends upon the distance we are working
with. Usually, in what follows, if not otherwise stated, our examples will be taken
in the Euclidean space Rn , endowed with the standard distance.
A = {x ∈ X | x is an adherent point of A}
A ′ = {x ∈ X | x is a limit point of A}
Example 2.35 Let X = R2 , with its usual metric, and let A = ((0, 1) × (0, 1)) ∪
{(2, 7)} ⊆ R2 . Then A ′ = [0, 1] × [0, 1] and Ā = ([0, 1] × [0, 1]) ∪ {(2, 7)}.
∂A = A ∩ CA
Ext A = Int (X \ A)
Example 2.37 In the Euclidean space R, endowed with the standard distance, if
A = (0, 1) ⊂ R, then ∂A = {0, 1} and Ext A = (−∞, 0) ∪ (1, ∞).
Definition 2.38 A point which is not a limit point for A is called an isolated point
of A. A set A which contains only isolated points is said to be discrete. We shall
denote by ′ A the set of all the isolated points of A.
Definition 2.40 Let (X, d) be a metric space and A ⊆ X. The number diam A =
sup{d(x, y) | x, y ∈ A} is called the diameter of the set A.
Definition 2.41 Let (X, d) be a metric space and A ⊆ X. The set A is called
bounded if diam A < ∞.
Definition 2.43 Let (X, d) be a metric space and A, B ⊆ X. The number d(A, B) =
inf{d(a, b) | a ∈ A, b ∈ B} is called the distance between the sets A and B.
Definition 2.44 A metric space (X, d) is called compact if for any family of open
∪
sets (Dα )α∈A , Dα ⊆ X, such that X = Dα , there exists a finite set J ⊆ A such
∪ α∈A
that X = Dα .
α∈J
Definition 2.45 A set K ⊆ X in a metric space (X, d) is called compact if for any
∪
family of open sets (Dα )α∈A , Dα ⊆ X, such that K ⊆ Dα , there exists a finite
∪ α∈A
set J ⊆ A such that K ⊆ Dα .
α∈J
Example 2.46 Let us consider the metric space (R, d), where d(x, y) = |x − y|.
Then, if a, b ∈ R, with a < b, the closed interval [a, b] is a compact set. Also, notice
that R is not compact and every finite set in (R, d) is compact.
Remark 2.50 Let us consider the metric space (R, d), where d(x, y) = |x − y|. A
nonempty set A ⊆ R is connected if and only if A is an interval. As a consequence,
R is connected, the open interval (a, b), with a, b ∈ R, a < b, is also connected, but
Q is not a connected set.
diam A = sup{d(x, y) | x, y ∈ A}
is called the diameter of the set A. Notice that diam A ∈ [0, ∞] and diam (A) =
diam A.
Definition 2.52 Let (X, d) be a metric space and A ⊆ X. The set A is called
bounded if diam A < ∞.
In a metric space, any ball (open or closed) with a finite radius is a bounded
set. As a matter of fact, we could define a bounded set as being a set which is
contained in a ball having a finite radius. For instance, the balls B(0, 3) and B(0, 3)
are bounded sets in R3 .
d(x, A) = inf{d(x, y) | y ∈ A}
d(A, B) = inf{d(a, b) | a ∈ A, b ∈ B}
Exercise 2.57 For A = [0, 2) ∪ {3} ∪ {7}, show that Int A = (0, 2), A = [0, 2] ∪
{3} ∪ {7}, A ′ = [0, 2], ∂A = {0} ∪ {2} ∪ {3} ∪ {7}, ′ A = {3} ∪ {7}.
Exercise 2.58 Show that in R2 endowed with the discrete metric, the open ball
B(x, r) contains only one point, its center x, if 0 < r < 1, and contains all the
points of R2 , if r ≥ 1.
Exercise 2.59 Prove that in R2 endowed with the Chebyshev distance the open ball
B(x, r) consists of all the points in the interior of a square centered at the point x
and with sides parallel to the coordinate axes and of length equal to 2r.
Exercise 2.60 Prove that in R2 endowed with the Euclidean distance the open ball
B(x, r) consists of all the points of the disk centered at the point x and having the
radius equal to r.
Solution. Following the definition of the Euclidean distance, we get B1 (−1) = {−1},
B1 (0) = [0, 1), and B2 (0) = {−1} ∪ [0, 2).
Therefore,
∪
A= Dx ,
x∈A
Exercise 2.63 Let (X, d) be a metric space and x ∈ X. The set V(x) possesses the
following properties:
(i) if V1 ∈ V(x) and V1 ⊆ V2 , then V2 ∈ V(x);
(ii) if V1 , V2 ∈ V(x), then V1 ∪ V2 ∈ V(x), V1 ∩ V2 ∈ V(x).
In fact, the union of any collection of neighbourhoods is also a neighbourhood. Also,
the intersection of any finite number of neighbourhoods is a neighbourhood, too.
Exercise 2.65 Let X = R2 equipped with its usual metric and A = Q2 . Show that
Int A = ∅.
Hint. There exists no open ball in R2 that contains only points with both coordinates
rational.
Exercise 2.66 Identify the interior, the closure and the set of all the accumulation
points of each of the following sets: (i) the integers Z in R; (ii) the rational
numbers Q in R; (iii) the complex numbers with rational real and imaginary parts
in C; (iv) the set A = [0, 1) ∪ {3} in R.
Remark 2.73 Any metric space has the following property: for any x, y ∈ X with
x ̸= y, there exist U ∈ V(x) and V ∈ V(y) such that U ∩ V = ∅. A space with such
a property is called separated or Hausdorff.
Remark 2.74 A convergent sequence in a metric space has only one limit.
Remark 2.78 The converse implication in Proposition 2.77 is, in general, not true.
As a counterexample, let (R, d), with d(x, y) = | arctan y−arctan x|. It is not difficult
to see that the sequence (xk ), with xk = k, is fundamental, but not convergent.
Definition 2.79 A metric space (X, d) is called a complete space if in this space
any Cauchy sequence is convergent.
Example 2.81 (i) The metric space (R, d), with d(x, y) = |x − y|, is complete.
Also, the Euclidean space Rn and the unitary space Cn (endowed with the usual
Euclidean distances) are complete.
(ii) The space lp , with 1 ≤ p < ∞, and the space l∞ are complete.
Example 2.82 (i) In X = (0, 1), the sequence xk = 1/k, with k ≥ 2, is Cauchy,
but it is not convergent (in X).
(ii) The set of all rational numbers Q with the usual metric given by d(x, y) =
|x − y|, for x, y ∈ Q, is not complete.
Definition 2.85 Two metric spaces X and Y are said to be isometric if there exists
a bijective isometry from X to Y .
MULTIDIMENSIONAL SPACES 41
Remark 2.87 Every metric space has a completion. Such a completion is unique
up to isometry.
Exercise 2.88 Let (xk )k be a convergent sequence in a metric space (X, d). If there
exists A ⊆ X and k0 ∈ N such that xk ∈ A, for any k ≥ k0 , then lim xk ∈ A.
k→∞
Exercise 2.89 Let (xk )k be a sequence in a metric space (X, d) and let x ∈ X.
Prove that x = lim xk if and only if for all ε > 0, there exists kε ∈ N such that
k→∞
d(xk , x) < ε, for all k ≥ kε (ε-criterion).
Exercise 2.90 Let C([0, 1]) be the space of all the continuous real-valued functions
∫ 1
on [0, 1], endowed with the metric d(f, g) = |f (t)−g(t)| dt. Show that (C([0, 1]), d)
0
is not a complete metric space.
Exercise 2.91 Let M be a nonempty subset of a metric space (X, d). Show that
x ∈ M if and only if there exists a sequence (xk )k in M convergent to x. Also, prove
that M is closed if and only if the limit x of any convergent sequence (xk )k ⊆ M
belongs to M .
Exercise 2.95 Let (X, d) be a metric space and M a dense subset of X with the
property that every Cauchy sequence in M is convergent in X. Show that X is
complete.
Let us notice that we can replace axiom 3) in the above definition by the following
one: ⟨x, x⟩ > 0, ∀x ∈ X \ {0X }. Also, notice that the nonnegativity condition
∀x ∈ X, ⟨x, x⟩ ≥ 0 makes sense, because ⟨x, x⟩ ∈ R, for any x ∈ X.
An alternative definition, encountered especially in physics, would require to
consider the inner product to be linear in the second argument rather than in the
first one. In this case, the first argument becomes conjugate linear.
As typical examples, we shall consider the real Euclidean space Rn and the complex
Euclidean space Cn .
Remark 2.101 In (2.1), we have an equality if and only if the vectors x and y are
linear dependent, i.e. there exists λ ∈ K such that x = λ y.
Remark 2.102 If we take X = Rn with the usual Euclidean scalar product, then,
from (2.1), for any x = (x1 , . . . , xn ), y = (y1 , . . . , yn ) ∈ Rn , we obtain the following
well-known inequality:
( n )1/2 ( n )1/2
∑
n ∑ ∑
xi y i ≤ x2i yi2 ,
i=1 i=1 i=1
a a a
44 REAL ANALYSIS
Then, in this case, for any two sequences x = (xi )i and y = (yi )i in l2 , we get
∞
∑ (∑
∞ )1/2 ( ∑
∞ )1/2
xi y i ≤ |xi |2 |yi |2 .
i=1 i=1 i=1
Definition 2.104 Let x and y be nonzero vectors in Rn . The angle between x and
y is the unique number θ ∈ [0, π], given by
⟨x, y⟩
cos θ = √ √ .
⟨x, x⟩ ⟨y, y⟩
⟨x, y⟩
−1 ≤ √ √ ≤ 1.
⟨x, x⟩ ⟨y, y⟩
π
Let us notice that for θ = , we get ⟨x, y⟩ = 0, which justifies Definition 2.103.
2
Exercise 2.105 Prove that the set of all ordered n-tuples of real numbers
Rn = {(x1 , x2 , . . . , xn ) | xi ∈ R, i = 1, . . . , n},
α · x = (αx1 , . . . , αxn ), ∀α ∈ R, x ∈ Rn ,
MULTIDIMENSIONAL SPACES 45
∑
n
x= xi ei .
i=1
Exercise 2.106 Show that Rn is a real linear space and Cn is a complex linear
space.
Exercise 2.110 Let X = C([a, b]) = {f : [a, b] → R | f continuous on [a, b]}. Prove
that the map ⟨·, ·⟩ : X × X → R defined by
∫b
⟨f, g⟩ = f (x) g(x) dx (2.3)
a
Exercise 2.113 Let x, y be two given vectors in Rn . Prove that if ⟨z, x⟩ = ⟨z, y⟩,
for any z ∈ Rn , then x = y.
Remark 2.115 Any norm takes only positive values, i.e. ∥x∥ ≥ 0, for all x ∈ X.
Thus, ∥ · ∥ : X → R+ . Also, as a consequence of the second axiom, we have ∥x∥ =
0 ⇔ x = 0X .
Example 2.116 For p ∈ [1, ∞), the space of p-summable sequences lp is defined
as being the collection of all the sequences x = (xi )∞
i=1 for which
∞
∑
|xi |p < ∞.
i=1
MULTIDIMENSIONAL SPACES 47
Definition 2.119 Two norms ∥ · ∥ and ∥ · ∥ ′ are called equivalent if there exist
α, β > 0 such that
∥x∥ ≤ α ∥x∥ ′ ≤ β ∥x∥, ∀x ∈ X.
We write ∥ · ∥ ∼ ∥ · ∥ ′ .
Remark 2.120 Let us notice that two norms on a linear space will generate the
same topology if and only if they are equivalent.
Remark 2.121 Let X be a normed space with dim K X = finite. One can prove
that in such a finite dimensional normed space, any two norms are equivalent.
Definition 2.122 Let X be a normed space. We shall say that X has Heine-Borel’s
property if every bounded sequence in X possesses a convergent subsequence, whose
limit belongs to X.
Definition 2.124 A normed space (X, ∥·∥) which is complete is said to be a Banach
space.
Example 2.125 For any fixed real number p ≥ 1, the Banach space Lp ([a, b]) is
the completion of the normed space of all the continuous real-valued functions on
[a, b], with the norm given by
(∫ )1/p
b
∥f ∥p = |f (x)| dx
p
.
a
48 REAL ANALYSIS
Definition 2.126 A pre-Hilbertian space (X, ⟨·, ·⟩) which is complete is called a
Hilbert space.
Example 2.127 The Euclidean spaces Rn and Cn are Banach spaces. In fact, any
finite dimensional normed space X is a Banach space.
So, if (X, ∥·∥) is a normed space and Y ⊂ X is a linear subspace, then (Y, ∥·∥) is a
normed space.
Exercise 2.132 Prove that the following map is a norm on C([a, b]):
√
∫ b
∥f ∥ = f 2 (x) dx.
a
Solution. Indeed, from a given inner product, we can naturally define a norm by:
√
∥x∥ = ⟨x, x⟩. (2.4)
So, any pre-Hilbertian space is a normed space. Let us remark that the map ∥ · ∥ :
X → R+ is well-defined, since ⟨x, x⟩ ≥ 0, for any x ∈ X. We shall prove that (2.4)
verifies all the axioms in Definition 2.115. Indeed, we have:
√
1) ∥x∥ = 0 ⇔ ⟨x, x⟩ = 0 ⇔ ⟨x, x⟩ = 0 ⇔ x = 0X ;
√ √ √
2) ∥αx∥ = ⟨αx, αx⟩ = αα⟨x, x⟩ =| α | ⟨x, x⟩ = |α| ∥x∥, ∀α ∈ K, ∀x ∈ X;
3) ∥x + y∥2 = ⟨x + y, x + y⟩ = ∥x∥2 + ∥y∥2 + ⟨x, y⟩ + ⟨x, y⟩ = ∥x∥2 + ∥y∥2 +
2Re⟨x, y⟩ ≤ ∥x∥2 + ∥y∥2 + 2|⟨x, y⟩|. Hence, ∥x + y∥2 ≤ ∥x∥2 + ∥y∥2 + 2|⟨x, y⟩|.
Using (3.4), we get ∥x + y∥2 ≤ ∥x∥2 + ∥y∥2 + 2∥x∥ ∥y∥ = (∥x∥ + ∥y∥)2 , which gives
∥x + y∥ ≤ ∥x∥ + ∥y∥.
Exercise 2.134 For m, p ∈ N, we introduce the space of matrices Mm,p of all the
m × p matrices with complex entries. Then, prove that Mm,p is an inner product
space with the trace inner product defined as
∑
m ∑
p
⟨A, B⟩ = aij bij .
i=1 j=1
The norm generated by this inner product on Mm,p is called the Hilbert-Schmidt or
the Frobenius norm of matrices:
(∑
m ∑
p )1/2
∥A∥ = |aij |2 .
i=1 j=1
Remark 2.135 The converse implication in Exercise 2.134 is false. Most real and
complex normed vector spaces do not have inner products. For instance, in R2 , the
norm
∥x∥∞ = max{|x1 |, |x2 |}
50 REAL ANALYSIS
If the norm satisfies this identity, the associated inner product is given by the so-
called polarization identity. In the real case, this identity, which generates the scalar
product, is:
1( )
⟨x, y⟩ = ∥x + y∥2 − ∥x − y∥2 .
4
In the complex case, the polarization identity is given by
1( )
⟨x, y⟩ = ∥x + y∥2 − ∥x − y∥2 + i∥x + iy∥2 − i∥x − iy∥2 .
4
Additionally, the inner product generated by a given norm is unique, as a consequence
of the polarization identity.
Exercise 2.136 Let X be a real pre-Hilbertian space and let x, y ∈ X. Prove that
and
⟨x, y⟩ = ∥x∥ · ∥y∥ ⇔ ∃α ∈ R such that y = αx.
∥ · ∥1 ∼ ∥ · ∥2 ∼ ∥ · ∥∞ .
Exercise 2.141 Prove that if X and Y are normed spaces, then the product space
X × Y is also a normed space. For instance, the map
is a norm on X × Y . Notice that one can also consider other norms, such as
or √
∥(x, y)∥X×Y = ∥x∥2X + ∥y∥2Y .
Exercise 2.142 Let X be a compact topological space and let us consider the space
Exercise 2.143 Prove that the space of all bounded sequences (real or complex)
x = (xk )k≥1 with the norm
∥x∥ = sup ∥xk ∥
k
is a Banach space.
Exercise 2.144 Let X be the space of all the continuous real functions on [0, 1]
and let d : X × X → R+ defined by
∫ 1
d (f, g) = |f (x) − g(x)| dx.
0
Exercise 2.145 Show that the space c of convergent sequences in C, endowed with
the supremum norm, is a Banach space.
Exercise 2.146 Prove that the space l∞ of bounded sequences (with the supremum
norm) is a Banach space.
Exercise 2.147 Show that the space of all the continuous real functions on [0, 1],
with the norm ∫ 1
∥f ∥L1 = |f (x)| dx
0
given by the Riemann integral of the absolute value, is not a complete normed space.
Exercise 2.148 (Minkowski Inequality in lp ) Let p ∈ [1, ∞). Prove that for any
two sequences x, y ∈ lp , we have
(∑
∞ )1/p (∑
∞ )1/p (∑
∞ )1/p
|xi + yi |p ≤ |xi |p + |yi |p .
i=1 i=1 i=1
Exercise 2.149 (i) Show that the set c of convergent sequences and the set c0 of
sequences converging to zero are closed subspaces of l∞ .
(ii) Let p ∈ [1, ∞). Prove that the set of p-summable sequences lp is a closed
subspace of l∞ and a dense subspace of c0 .
Exercise 2.150 Let us consider a Banach space X and let Y be a linear subspace
of X. Prove that Y is Banach if and only if Y is closed in X.
Exercise 2.151 Show that the space l∞ of all bounded sequences of real numbers,
with the supremum norm, is a Banach space.
Exercise 2.152 Prove that the space of polynomials P (x) on the interval [0, 1] is
not a Banach space with respect to the supremum norm.
Exercise 2.153 Let x = (1, 1, 0) and y = (2, 0, 0). Compute ⟨x, y⟩, ∥x∥, ∥y∥ and
the angle between x and y.
d(x, y) = ∥x − y∥,
it is easy to verify that this map is indeed a distance on X. The metric d is said to
be induced by the norm ∥ · ∥. Hence, (X, d) is a metric space.
Exercise 2.157 Show that in R3 , the set of all the vectors whose Euclidean norm
is a given constant forms the surface of a sphere, the set of vectors whose 1-norm is
a given constant forms the surface of a cross polytope and the set of vectors whose
∞-norm is a given constant forms the surface of a hypercube.
∥ · ∥1 ∼ ∥ · ∥2 ∼ ∥ · ∥∞ .
Exercise 2.159 Prove that if X and Y are normed spaces, then the product space
X × Y is also a normed space.
or √
∥(x, y)∥X×Y = ∥x∥2X + ∥y∥2Y .
∑
k
Sk = xi , (3.1)
i=0
then the pair ((xk )k≥0 , (Sk )k≥0 ) is called a series (the series associated to the sequence
(xk )n≥0 ).
Sk is called the kth partial sum of the series (3.2), while xk is said to be its general
term.
S = lim Sk ,
k→∞
then the series (3.2) is called convergent. In this case, S is said to be its sum.
55
56 REAL ANALYSIS
∑
If the series xk converges to S, we shall write:
k≥0
∞
∑
S= xk .
k=0
Definition 3.5 A series which is convergent, but not absolutely convergent, is called
semi-convergent or conditionally convergent.
∑ ∑
Proposition 3.6 If the series xk is absolutely convergent, then the series xk
k≥0 k≥0
is convergent. The converse implication is, in general, not true.
∑ (−1)k
Example 3.7 The series is only semi-convergent.
k≥1
k
∑
Proposition 3.8 If the series xk is convergent, then the sequence (xk )k is con-
k≥0
vergent and
lim xk = 0.
k→∞
1
xk = → 0.
k
∑
Proposition 3.11 (Cauchy’s Criterion) The series xk is convergent in X if
k≥0
and only if for any ε > 0, there exists kε ∈ N such that
|xk | ≤ yk , ∀ k ≥ k0 .
∑ ∑
If the series yk is convergent, then the series xk is absolutely convergent.
k≥0 k≥0
0 ≤ xk ≤ y k , ∀ k ∈ N .
∑ ∑
(i) If yk is convergent, then xk is convergent.
k≥0 k≥0
∑ ∑
(ii) If xk is divergent, then yk is also divergent.
k≥0 k≥0
xk
(iii) If xk , yk ∈ R such that xk , yk > 0 and there exists lim = L ∈ (0, ∞),
k→∞ yk
∑ ∑
then the series xk and yk have the same nature.
k≥0 k≥0
Then, we have:
∑
(i) if L < 1, the series xk is absolutely convergent;
k
∑
(ii) if L > 1, the series xk is divergent.
k
then:
58 REAL ANALYSIS
∑
(i) if L < 1, the series xk is absolutely convergent;
k
∑
(ii) if L > 1, the series xk is divergent;
k
∑
(iii) if L = 1, we cannot decide upon the nature of the series xk .
k
lim yk = 0,
k→∞
∑
then the series xk yk is convergent.
k0
SERIES 59
If there exists
L = lim Rk ,
k→∞
then:
∑
(i) if L > 1, the series xk is convergent;
k≥0
∑
(ii) if L < 1, the series xk is divergent;
k≥0
∑
(iii) if L = 1, we cannot decide upon the nature of the series xk .
k≥0
∑ ∑
Definition 3.22 Let xk and yk be series of real (or complex) numbers. Let
k≥0 k≥0
∑
us consider the series zk , with
k≥0
∑
zk = xi y j .
i+j=k
∑ ∑
This series is called the Cauchy product series of the series xk and yk .
k≥0 k≥0
60 REAL ANALYSIS
∑ ∑
Theorem 3.23 (Mertens) Let xk and yk be convergent series of real (com-
k≥0 k≥0
plex) numbers. If at least one of our two series is absolutely convergent, then the
∑
Cauchy product series zk , with
k≥0
∑
zk = xi yj
i+j=k
is convergent and ( )
∞
∑ ∞
∑ ∞
∑
zk = ( xk ) yk .
k=0 k=0 k=0
∑ ∑
Theorem 3.24 (Cauchy) Let xk and yk be absolutely convergent series of
k≥0 k≥0
∑
real (or complex) numbers. Then, the Cauchy product series zk is absolutely
k≥0
convergent.
Remark 3.25 In general, a product series of two convergent series is not conver-
gent. For instance, if we take
(−1)k
xk = y k = √ ,
k
∑
then the Cauchy product series zk is divergent.
k≥0
Therefore, since
lim Sk = 1,
k→∞
our series is convergent and its sum is equal to 1.
k+1
Sk = ln .
2k
Thus, since
lim Sk = − ln 2,
k→∞
our series is convergent and its sum is equal to − ln 2.
1 1
xk = arctan − arctan .
k k+1
By induction,
1
Sk = arctan 1 − arctan ,
k+1
π π
which converges to . Hence, our series is convergent and its sum is equal to .
4 4
∑ 1
a) ;
k≥1
4k 2−1
∑ 1
b) , p ∈ N∗ ;
k≥1
k (k + p)
√
∑ k − k2 − 1
c) √ .
k≥1
k (k + 1)
∑1
Exercise 3.30 Prove that the series is divergent.
k≥1
k
Solution. We have
1 1 1 1 1
|S2k − Sk | = + + ··· + ≥k = .
k+1 k+2 2k 2k 2
Therefore, condition (3.3) is not satisfied and, so, our series is divergent.
∑ ∑
Exercise 3.31 If xk and yk are convergent series in X and if α, β ∈ K, show
k≥0 k≥0
that the series
∑
(αxk + βyk )
k≥0
is convergent and
∞
∑ ∞
∑ ∞
∑
(αxk + βyk ) = α xk + β yk .
k=0 k=0 k=0
SERIES 63
zk = xk + i yk .
Show that
∑ ∑ ∑
zk is convergent ⇔ xk and yk are convergent.
k≥0 k≥0 k≥0
Exercise 3.34 (The geometric series) Let q ∈ R such that |q| < 1. Prove that the
∑ k
series q is convergent and
k≥0
∞
∑ 1
qk = .
k=0
1−q
∑
If |q| ≥ 1, show that the series q k is divergent.
k≥0
∑
Exercise 3.35 If z ∈ C and |z| < 1, show that the series z k is convergent and
k≥0
∞
∑ 1
zk = .
k=0
1−z
1 − q k+1
Sk = 1 + q + · · · + q k = .
1−q
Therefore,
1
lim Sk = .
k→∞ 1−q
If |q| ≥ 1, then
|xk | = |q k | ≥ 1.
Hence, xk 9 0 and the series is divergent.
64 REAL ANALYSIS
∑ k!
Exercise 3.36 Prove that the series is convergent.
k≥1
(2k)!
∑ 1
Exercise 3.37 Show that the series √ is divergent.
k≥1
k(k + 1)
∑ 1 ∑ 1
Exercise 3.38 Show that the series and have the same nature.
k≥1
k+1 k≥1
5k + 2
∑ 1
Solution. By comparing with the series , which is convergent, it turns out
k≥1
k3
that the given series is convergent.
∑ 1
Exercise 3.40 Prove that the series (−1)k is convergent for p > 0 and diver-
k≥0
kp
gent for p ≤ 0.
SERIES 65
k≥1
kα
is convergent for α > 1 and divergent for α ≤ 1. This series is known as the
generalized harmonic series or Riemann’s series.
∑ 1
Solution. Using the condensation test, the series is convergent if and only if
k≥1
kα
∑ 1
the series 2k is convergent. But
k≥0
(2k )α
( )k
1 1
2k = .
(2k )α 2α−1
1
If α > 1, then < 1. Hence, our series is convergent.
2α−1
( )k
1
If α = 1, then = 1 9 0. Therefore, our series is divergent.
20
1 1
If α < 1, then k α < k and α
> . So, the given series is divergent.
k k
Exercise 3.42 Test the following series for convergence:
∑ 1
, a > 0.
k≥1
k (1 + a + · · · + ak)
Solution. We have to distinguish between three cases: a = 1, a > 1 and 0 < a < 1.
1
For a = 1, the general term of our series becomes and, obviously, the series
k (k + 1)
is convergent.
For a > 1, we have
1 1
< k , for any k ≥ 1.
k (1 + a + · · · + a )
k a
∑ 1 1
Since the geometric series is convergent, having the ratio < 1, using the
k≥1
ak a
comparison test we conclude that our series is convergent.
66 REAL ANALYSIS
In the case 0 < a < 1, we shall compare our series with the classical harmonic series,
which is divergent. Indeed, we have:
1
k (1 + a + · · · + ak ) 1
lim = lim = 1 − a ∈ (0, ∞).
k→∞ 1 k→∞ 1 − ak+1
k 1−a
xk+1 (k + 1)! k k kk 1
lim = lim = lim = < 1.
k→∞ xk k→∞ k!(k + 1)k+1 k→∞ (k + 1)k e
∑ ak
Exercise 3.45 Prove that the series , defined for a > 0, is convergent.
k≥0
k!
xk+1 ak+1 k! a
lim = lim k
= lim = 0 < 1.
k→∞ xk k→∞ (k + 1)! a k→∞ k + 1
∑ 1
Exercise 3.46 Show that the series is convergent.
k≥2
(ln k)k
∑
Exercise 3.47 Prove that the series (ln k)k is divergent.
k≥2
Exercise 3.49 Using the ratio and the root test, discuss the nature of the following
series:
∑ ak ∑ ak ∑ 1
a) , a > 0; b) , a, b > 0; c) ;
k! kb (ln k)k
k≥1 k≥1 k≥2
∑ k! ∑ k 2 k2
d) , e) (2 + sin k) (1 − ) .
kk k
k≥1 k≥1
∑ (−1)k ∑ 1
a) ; b) √ ;
k≥1
(3k + 1)2 k≥1
k+ k+1
∑ k!
c) , a > 0;
k≥1
a(a + 1) · · · (a + k)
∑ sin k + 1 ∑
e−k , β ∈ R.
β
d) ; e)
k≥2
k(ln k)2 k≥2
Exercise 3.54 Discuss, in terms of p and q, the nature of the following series:
∑ 1
p (ln k) q
.
k≥2
k
∑ √ ∑ 1 (k)k
c) ( (k + 1)(k + a) − k) k , a > 0; d) , a > 0.
k≥1 k≥1
k! a
Chapter 4
f : M → Y.
A number l ∈ Y is called the limit of the function f at the point a if for any V ∈ V(l),
there exists U ∈ V(a) such that, for any x ∈ U ∩ M \ {a}, it follows that f (x) ∈ V .
Remark 4.2 If a function f possesses a limit l at a given point a, then this limit
is unique.
′
Theorem 4.3 Let (X, d) and (Y, ρ) be metric spaces, M ⊆ X, a ∈ M . Also, let
f : M → Y . The following statements are equivalent:
(i) l = lim f (x);
x→a
(ii) ∀ ε > 0 ∃ δε > 0 such that ∀x ∈ M, x ̸= a, with d(x, a) < δε , it follows that
ρ(f (x), l) < ε;
(iii) ∀ (xn )n ⊆ M, xn ̸= a, such that xn → a, it follows that f (xn ) → l.
′
Theorem 4.4 (Cauchy’s Criterion) Let f : E ⊆ Rn → Rm and let a ∈ E . Then,
there exists l = lim f (x) if and only if for any ε > 0, there exists Uε ∈ V(a) such
x→a
′ ′′ ′ ′′
that for any x , x ∈ Uε ∩ E \ {a}, it follows that d(f (x ), f (x )) < ε.
69
70 REAL ANALYSIS
lu = lim f (x).
x→au
Notice that, for the existence of the above limit, we need that a+h u ∈ E, for h → 0.
Remark 4.6 If the global limit lim f (x) exists, then all the directional limits lu
x→a
exist, for all u ̸= 0.
The converse implication is, in general, false. As a concrete example, let us consider
the function f : R2 \ {(0, 0)} → R, defined by
x2 y
f (x, y) = .
x4 + y 2
It is not difficult to see that lim f (x, y) doesn’t exist, but all the directional
(x,y)→(0,0)
limits limu for all u ̸= 0 exist.
Remark 4.7 Usually, directions are taken to be normalized, although the above
definition works for arbitrary vectors u ̸= 0. So, in such a case we shall work with
unit vectors (also called by some authors direction vectors), i.e. with vectors with
∥u∥ = 1.
Remark 4.8 We recall now the concept of iterated limits for a two-variable func-
tion. Let f : A ⊆ R2 → R and let (a, b) be an accumulation point of A. The
limits ( )
l12 = lim lim f (x, y)
x→a y→b
and ( )
l21 = lim lim f (x, y)
y→b x→a
′
Remark 4.9 Let f : A ⊆ R2 → R, (a, b) ∈ A . If there exists
lim f (x, y) = l
(x,y)→(a,b)
then l = l12 . So, if the global limit and one of the iterated limits exist at a given
point, then the values of these limits must be equal.
Let us notice that if the repeated limits exists, but are not equal, then the global
limit doesn’t exist.
Example 4.11 If
2x2
f (x, y) = ,
x2 + y 2
then ( )
2x2
lim lim f (x, y) = lim =2
x→0 y→0 x→0 x2
and ( )
lim lim f (x, y) = lim 0 = 0.
y→0 x→0 y→0
Thus, the global limit lim(x,y)→(0,0) f (x, y) again does not exist, despite the fact that
both of the iterated limits exist.
72 REAL ANALYSIS
a) lim (f ± g)(x) = lim f (x) ± lim g(x); b) lim (c f )(x) = c lim f (x);
x→a x→a x→a x→a x→a
lim f (x) = l.
x→a
′
Exercise 4.15 Let f : E ⊆ Rn → C, f = u + iv and let a ∈ E . Then,
lim (Ref )(x) = Re l,
x→a
lim f (x) = l ⇔
x→a
lim (Imf )(x) = Im l.
x→a
′
Exercise 4.16 Let f1 , f2 : E ⊆ Rn → R and let a ∈ E . If we assume that
and if
f1 (x) ≤ φ(x) ≤ f2 (x), ∀x ∈ U ∩ E \ {a}, U ∈ V(a),
then
lim φ(x) = l.
x→a
x2 − y 2
Exercise 4.17 Let f (x, y) = , ∀(x, y) ∈ R2 \ {(0, 0)}. Prove that the limit
x2 + y 2
of this function, for (x, y) → (0, 0), doesn’t exist.
LIMITS AND CONTINUITY 73
( )
1 α
Solution. If we consider the sequence (xk , yk ) = , , with α > 0, which is an
k k
admissible one, i.e. (xk , yk ) → (0, 0), for k → ∞, we get
1 − α2
lim f (xk , yk ) = .
k→∞ 1 + α2
So, the limit of f (xk , yk ) for k → ∞ depends on α and, hence, the limit of f (x, y)
for (x, y) → (0, 0) doesn’t exist.
x2 y 2
Exercise 4.18 Let f (x, y) = , ∀(x, y) ∈ R2 \ {(0, 0)}. Prove that
x2 + y 2
lim f (x, y) = 0.
(x,y)→(0,0)
x2 ≤ x2 + y 2 , y 2 ≤ x2 + y 2 , ∀(x, y) ∈ R2 .
Therefore, we obtain
|f (x, y)| ≤ x2 + y 2 .
Obviously, since
lim (x2 + y 2 ) = 0,
(x,y)→(0,0)
we get
lim f (x, y) = 0.
(x,y)→(0,0)
x2 y 1
l= lim 2 2
cos 2 .
(x,y)→(0,0) x + y x + y2
Solution. Since
x2 y
lim =0
(x,y)→(0,0) x2 + y 2
and the cosine function is bounded, the limit l exists and it is equals to 0.
Exercise 4.20 Compute the directional limit at the point a = (1, 2, 3), in the di-
rection u = (1, 1, 1), of the function f : R3 → R, defined by f (x, y, z) = x2 + y + yz.
74 REAL ANALYSIS
x−y
f (x, y) = xy .
x2 + y 2
Test this function for the existence of directional limits at the point (0, 0).
x−y | x2 y | | xy 2 |
0 ≤ xy ≤ + ≤| y | + | x | .
x2 + y 2 x2 + y 2 x2 + y 2
1
Exercise 4.22 Prove that lim sin doesn’t exist.
x→0 x
xy
Exercise 4.23 Show that the limit lim doesn’t exit.
(x,y)→(0,0) x2 + y 2
xy
lim √ = 0.
(x,y)→(0,0) x2+ y2
x−y
Exercise 4.25 Show that the limit lim doesn’t exist, but the iterated
(x,y)→(0,0) x + y
limits exist and
lim (lim f (x, y)) ̸= lim ( lim f (x, y)).
x→0 y→0 y→0 x→0
1 1
Exercise 4.26 Prove that the limit lim (x + y) sin sin exists, but we don’t
(x,y)→(0,0) x y
have iterated limits.
LIMITS AND CONTINUITY 75
Exercise 4.27 Compute the directional limits in the direction u = (1, 2), at the
point (0, 0), for the function
xy
f (x, y) = .
x2 + y2
Exercise 4.28 Compute the directional limits, at the point (0, 0), in the direction
(1, 2) for the function
x2 y
f (x, y) = 2 .
x + y2
Exercise 4.29 Show that
1
lim (x2 + y 2 ) sin = 0.
(x,y)→(0,0) x2 + y2
x2 − y 2 + z 2
f (x, y, z) = .
x2 + y 2 + z 2
Prove that the limit lim f (x, y, z) doesn’t exist.
(x,y,z)→(0,0,0)
x2 y 2
f (x, y) = .
x2 y 2 + (x − y)4
Prove that the global limit lim f (x, y) doesn’t exist, but the iterated limits
(x,y)→(0,0)
exist.
Exercise 4.33 Compute the directional limits at the point (π/4, 1), in the direction
→
− √ √
u = ( 2/2, 2/2), for the function
x
f (x, y) = (x2 + x cos y, ln sin ).
y
76 REAL ANALYSIS
Exercise 4.34 Compute the directional limits at the point (1, 1), in the direction
→
− √ √
u = ( 2/2, 2/2), for the function
1 − cos(x3 + y 3 )
f (x, y) = .
x2 + y 2
Test this function for the existence of directional limits at the point (0, 0).
Definition 4.37 Let (X, d) and (Y, ρ) be metric spaces and f : X → Y . The
function f is said to be continuous on X if it is continuous at each point a ∈ X.
Theorem 4.38 Let (X, d) and (Y, ρ) be metric spaces, a ∈ X. Also, let f : X → Y .
The following statements are equivalent:
(i) f is continuous at the point a;
(ii) for any ε > 0, there exists δε > 0 such that for any x ∈ X with d(x, a) < δε , it
follows that ρ(f (x), f (a)) < ε;
(iii) ∀(xn )n ⊆ X such that xn → a, it follows that f (xn ) → f (a).
Notice that, for the existence of the above limit, we need that a+h u ∈ E, for h → 0.
Remark 4.43 Usually, as in the case of directional limits, we shall work with unit
vectors, i.e. we shall consider that ∥u∥ = 1.
Definition 4.44 Let (X, d) and (Y, ρ) be two metric spaces. A function f : (X, d) →
(Y, ρ) is called uniformly continuous on X if for any ε > 0, there exists δε > 0 such
that for any x ′ , x ′′ ∈ X with d(x ′ , x ′′ ) < δε , it follows that ρ(f (x ′ ), f (x ′′ )) < ε.
| f (x) |≤ a | x | + b,
for any x ∈ R.
Definition 4.47 Let K = R, C and X and Y be two vector spaces over K. A map
T : X → Y is called linear if
T (0) = 0.
78 REAL ANALYSIS
Exercise 4.52 If (X, ∥ · ∥) is a normed space, show that the map ∥ · ∥ is continuous.
2x
f (x) = +x
1+x
is uniformly continuous on R.
2x1 2x2
f (x1 ) − f (x2 ) = + x1 − − x2
1 + x1 1 + x2
2
≤| x1 − x2 | 1 + < 3 | x1 − x2 | .
(x1 + 1)(x2 + 1)
ε
Therefore, for any ε > 0, there exists δε = > 0 such that for any x1 , x2 ∈ (0, ∞)
3
with | x1 − x2 |< δε , we obtain
1
f (x) = ,
x2
is not uniformly continuous on (0, 3).
Solution. To prove that f is not uniformly continuous on the interval (0, 3), it suffices
to prove that there exists ε0 > 0 and there exist two sequences (xn ) and (yn ), with
d(xn , yn ) < δ, but with ρ(f (xn ), f (yn )) ≥ ε0 , for any n ≥ n0 . Indeed, taking ε0 = 1
and
1 1
xn = √ , y n = √ ,
n n+1
it follows that | xn − yn |→ 0, while | f (xn ) − f (yn ) |= 1 = ε0 > 0. So, the function
f is not uniformly continuous (0, 3).
Exercise 4.55 (Heine-Cantor) Let (X, d) and (Y, ρ) be metric spaces. If (X, d) is
compact and f : (X, d) → (Y, ρ) is continuous on X, then f is uniformly continuous
on X.
Since X is compact, there exists a subsequence (xnk ) of the sequence (xn ) and there
exists an element x ∈ X such that
In a similar manner, there exist a subsequence (ynk ) of the sequence (yn ) and an
element y ∈ X such that
lim ynk = y. (4.4)
k→∞
80 REAL ANALYSIS
Since
d(ynk , x) ≤ d(ynk , xnk ) + d(xnk , x),
ε0
ρ(f (xnk ), f (ynk )) < , ∀k ≥ n0 ,
2
Remark 4.56 Let us notice that a contraction mapping between two metric spaces
is uniformly continuous.
Solution. Let M > 0 such that | f ′ (x) |≤ M , for all x ∈ I. Using Lagrange’s
theorem, it follows that for any x, y ∈ I there exists θ between x and y such that
So,
| f (x) − f (y) |≤ M | x − y |, for all x, y ∈ I.
ε
Let ε > 0 and δε = . Therefore, for any x, y ∈ I with | x − y |< δε , we obtain
M
ε
| f (x) − f (y) |≤ M = M,
M
f (x) = 7x + cos x,
is uniformly continuous on R.
It follows that
∪ ∪
K ⊆ f −1 (f (K)) ⊆ f −1 ( Fα ) = f −1 (Fα ).
α∈A α∈A
Since f is continuous and (Fα )α∈A is a family of open sets, it follows that f −1 (Fα ) ∈
D. Using the fact that K is compact, there exists J finite, J ⊆ A such that
∪
K⊆ f −1 (Fα )
α∈J
and, so,
∪
f (K) ⊆ Fα .
α∈J
Thus, f (K) is compact.
82 REAL ANALYSIS
So, A is not connected, which contradicts our hypotheses. Thus, our assumption is
false and, hence, f (A) is connected.
Exercise 4.64 Show that every product of a family of connected spaces is also
connected.
Solution. Let x0 ∈ X. Since T is continuous at the point a, it follows that for any
ε > 0 there exists δε > 0 such that for any x ∈ X with ∥x − a∥ < δε , we have
Thus,
T (x) − T (a) = T (t) − T (x0 ),
∥T (x)∥ ≤ M ∥x∥, ∀x ∈ X.
Solution. Let us prove the implication (i) ⇒ (ii). This is equivalent, in fact, to
proving that non (ii) ⇒ non (i). Thus, let us suppose that for any M ≥ 0, there
exists xM ∈ X such that
∥T (xM )∥ > M ∥xM ∥.
84 REAL ANALYSIS
So, ( )
1
∥T xn ∥ > 1, ∀n ∈ N∗ ,
n∥xn ∥
which means that
1
T( xn ) 9 0.
n∥xn ∥
On the other hand,
1
lim xn = 0.
n→∞ n∥xn ∥
Hence, since ( )
1
T xn 9 T (0),
n∥xn ∥
it follows that T is not continuous at the point 0.
(ii) ⇒ (i). Let a ∈ X. Since T is linear, we get
∥T (x − a)∥ ≤ M ∥x − a∥,
∑
n ∑
n
∥T (x)∥ = ∥T ( xi ei )∥ = ∥ xi T (ei )∥ ≤
i=1 i=1
∑
n ∑
n ∑
n
∥xi T (ei )∥ = |xi |∥T (ei )∥ ≤ ∥x∥ ∥T (ei )∥.
i=1 i=1 i=1
∑
n
Denoting by M = ∥T (ei )∥, we get
i=1
Test this function for continuity and directional continuity at the point (0, 0).
LIMITS AND CONTINUITY 87
Test this function for continuity and directional continuity at the point (0, 0).
Test this function for continuity and directional continuity at the point (0, 0).
is uniformly continuous on R.
f (x) = sin x,
Multivariable Differential
Calculus
89
90 REAL ANALYSIS
So, f is derivable at the point a if there exist f ′ (a) ∈ L(E, F ) and ωf,a : U → F
such that
and
lim ωf,a (x) = 0F . (5.3)
x→a
Notice that due to the uniqueness of the derivative f ′ (a) ∈ L(E, F ), we can define
a map
f ′ : U → L(E, F ).
Also, notice that the domain of the linear mapping f ′ (a) is the entire space E,
despite the fact that the function f is defined only locally about the point a.
f (x) − f (a) − f (x − a) 0
= = 0, ∀x ∈ E \ {a}.
∥x − a∥ ∥x − a∥
f ′′ : U → L(E, L(E, F ))
Putting
C 0 (U, F ) = {f : U → F | f is continuous on U },
we can define
∩
C ∞ (U, F ) = C n (U, F ).
n∈N
A function f ∈ C ∞ (U, F ) is said to be infinitely (indefinitely) derivable on U .
Definition 5.13 Let X be a vector space and let A ⊆ X. The set A is said to be
convex if for any x, y ∈ A the segment [x, y] ⊆ A, i.e. for any x, y ∈ A and for any
t ∈ [0, 1], we have
(1 − t)x + ty ∈ A.
∥f (b)−f (a)−f ′ (c)◦(b−a)∥ ≤ ∥b−a∥ sup ∥f ′ (x)−f ′ (c)∥, ∀a, b ∈ U, c ∈ [a, b].
x∈[a,b]
E = Rn , F = Rm .
Exercise 5.19 Let ∅ ̸= U ⊆ Rn be an open set and a ∈ U . Show that if the functions
f, g : U → R are derivable at the point a, then f ± g, λf, λ ∈ R, f g, f /g, g ̸= 0,
are derivable at the point a. Moreover, if f is injective and continuous on U and
f ′ (a) ̸= 0, prove that the inverse function f −1 is derivable at the point b = f (a) and
′ 1
(f −1 ) (f (a)) = ′ (a)
.
f
Proposition 5.20 (Chain Rule) Let U, V ⊆ R be open nonempty sets and let f :
U → R and g : V → R such that f (U ) ⊆ V . If f is derivable at the point a ∈ U and
g is derivable at the point b = f (a) ∈ V , then g ◦ f is derivable at the point a and
2x
2 arctan x + arcsin = π sgn x, ∀ | x |≥ 1.
1 + x2
94 REAL ANALYSIS
Let us consider now briefly another important case, i.e. the case in which E = Rn
and F = Rm . So, let ∅ ̸= U ⊆ Rn be an open set, f : U → Rm and a ∈ U . If
f = (f1 , . . . , fm ),
pi : Rn → R, pi (x) = xi , ∀x = (x1 , . . . , xn ) ∈ Rn
Remark 5.25 pi are linear and continuous. Thus, they are are derivable.
If m = n, the determinant
Jf (a) = det Mf (a)
∂(f1 , ..., fn )
Jf (a) = (a).
∂(x1 , . . . , xn )
Exercise 5.28 Compute the Jacobi matrix at the point (1, 1) for the function f :
R2 → R2 defined by
f (x, y) = (x2 + y 2 , 2xy).
pi = dxi
∑
n
∂f
f ′ (a) = (a) dxi , (5.4)
i=1
∂xi
where
∂f f (a1 , . . . , ai−1 , xi , ai+1 , . . . , an ) − f (a1 , . . . , ai−1 , ai , ai+1 , . . . , an )
(a) = lim .
∂xi xi →ai xi − ai
96 REAL ANALYSIS
In fact, {p1 , . . . , pn } is the canonical basis for L(Rn , R) and, so, (5.4) is just the
unique representation of the linear map f ′ (a) in this basis. The partial derivatives
∂f
(a) are exactly the coefficients of f ′ (a) in the canonical basis.
∂xi
Since f ′ (a) ∈ L(Rn , R) ≃ Rn , the derivative f ′ (a) can be also regarded as being the
vector
∂f ∂f
f ′ (a) ≃ ( (a), . . . , (a)).
∂x1 ∂xn
Remark 5.30 Let ∅ ̸= U ⊆ Rn , U open, f : U → R derivable at the point a ∈ U .
Then, there exists a unique vector y ∈ Rn such that
This vector is denoted by grada f and it is called the gradient of the function f at
the point a.
Solution. If f would be differentiable at the point (0, 0), then its partial derivatives
at the point (0, 0) would exist. In this case,
∂f f (x, 0) − f (0, 0) |x|
(0, 0) = lim = lim
∂x x→0 x−0 x→0 x
and
∂f f (0, y) − f (0, 0) |y|
(0, 0) = lim = lim .
∂y y→0 y−0 y→0 y
But these limits don’t exist and, hence, f is not differentiable at the point (0, 0).
Solution. If f would be differentiable at the point (0, 0), then f would be continuous
at the point (0, 0), i.e. lim f (x, y) = f (0, 0). But lim f (x, y) doesn’t
(x,y)→(0,0) (x,y)→(0,0)
even exist and, hence, f is not differentiable at the point (0, 0). Still, the partial
∂f ∂f
derivatives (0, 0) and (0, 0) exist and are equal to zero.
∂x ∂x
Exercise 5.33 Test the following function for differentiability:
x2 y
, (x, y) ̸= (0, 0),
f (x, y) = x2 + y 2
0, (x, y) = (0, 0).
∂f ∂f 2ab3 a2 (a2 − b2 )
f ′ (a, b) = ( (a, b), (a, b)) = ( 2 , ),
∂x ∂y (a + b2 )2 (a2 + b2 )2
∂f f (x, 0) − f (0, 0) 0
(0, 0) = lim = lim = 0
∂x x→0 x−0 x→0 x
∂f
and, in a similar manner, (0, 0) = 0. Therefore, if the function f would be
∂y
differentiable at the point (0, 0), then the derivative of f at this point should be the
zero map and
i.e.
x2 y
x2 + y 2
lim √ = 0.
(x, y) → (0, 0) x2 + y 2
But this limit doesn’t exist and, hence, f is not differentiable at the origin.
98 REAL ANALYSIS
Show that f is not continuous at the point (0, 0), but both partial derivatives
∂f ∂f
(0, 0) and (0, 0) exist.
∂x ∂x
Solution It is easy to see that if we let (x, y) to approach the origin along the line
x = y, then
lim f (x, y) = 1 ̸= f (0, 0)
(x,y)→(0,0)
and this implies the fact that f is not continuous at the origin. However, both
partial derivatives exist at the origin. Indeed,
∂f f (x, 0) − f (0, 0) 0
(0, 0) = lim = lim = 0
∂x x→0 x−0 x→0 x
∂f
and, in a similar manner, (0, 0) = 0.
∂y
Let f : U ⊆ Rn → R. Then,
In (5.5),
∂f ∂f
(a1 , . . . , xi , . . . , an ) − (a1 , . . . , ai , . . . , an )
∂2f ∂xj ∂xj
(a) = lim
∂xi ∂xj xi →ai xi − ai
and pi ⊗ pj is the tensor product of the linear maps pi and pj , defined by
In fact, {pi ⊗ pj }1≤i,j≤n is the canonical basis of L2 (Rn , Rn ; R). So, (5.5) is the
unique representation of the bilinear map f ′′ (a) in this basis. The mixed partial
∂2f
derivatives (a) are the coefficients of f ′′ (a) in the canonical basis.
∂xi ∂xj
Just as the Jacobi matrix was defined for representing (in the canonical bases)
the first derivative f ′ (a) of a function f : U ⊆ Rn → Rm which is derivable at the
point a ∈ U , in the case of the second derivative, for a function f taking values in
R, we have the following analogue representation of the second derivative:
100 REAL ANALYSIS
f (x, y, z) = xy 2 z 3 .
Then,
0 2yz 3 3y 2 z 2
Hf (x, y, x) = 2yz 3 2xz 3 6xyz 2 .
3y 2 z 2 6xyz 2 6xy 2 z
∂2f ∂2f
(a) = (a), ∀i, j = 1, n.
∂xi ∂xj ∂xj ∂xi
Remark 5.43 The existence of the mixed partial derivatives of the second order for
a given function doesn’t imply their symmetry.
∂2f
Indeed, let us compute (0, 0). We get:
∂x∂y
∂f ∂f
(x, 0) − (0, 0)
∂2f ∂y ∂y
(0, 0) = lim .
∂x∂y x→0 x−0
But
∂f f (x, y) − f (x, 0)
(x, 0) = lim =x
∂y y→0 y−0
and
∂f f (0, y) − f (0, 0)
(0, 0) = lim = 0.
∂y y→0 y−0
So,
∂2f x−0
(0, 0) = lim = 1.
∂x∂y x→0 x − 0
∂2f
On the other hand, computing (0, 0), we obtain:
∂y∂x
∂f ∂f
∂2f (0, y) − (0, 0)
(0, 0) = lim ∂x ∂x .
∂y∂x y→0 y−0
But
∂f f (x, y) − f (0, y)
(0, y) = lim = −y
∂x x→0 x−0
and
∂f f (x, 0) − f (0, 0)
(0, 0) = lim = 0.
∂x x→0 x−0
So,
∂2f −y − 0
(0, 0) = lim = −1.
∂y∂x y→0 y − 0
Hence,
∂2f ∂2f
(0, 0) ̸= (0, 0).
∂x∂y ∂y∂x
pi = dxi .
102 REAL ANALYSIS
Also, we shall usually omit to write explicitly the symbol ” ⊗ ”. So, we have:
∑
n
∂2f
f ′′ (a) = (a) dxi dxj .
i,j=1
∂xi ∂xj
∂mf ∂mf
(a) = (a),
∂x1 · · · ∂xm ∂xσ(1) · · · ∂xσ(m)
∑
n
∂mf
f (m)
(a) = (a) pi1 ⊗ · · · ⊗ pim .
i1 ,...,im =1
∂xi1 · · · ∂xim
∂2f ∂2f
+ = 0.
∂x2 ∂y 2
∂2f
= ex cos y
∂x2
and
∂2f
= −ex cos y.
∂y 2
Therefore, their sum is zero.
Exercise 5.46 Compute the second derivative at the point (1, 1) of the function
f : R2 → R, defined by:
2 2
f (x, y) = ex +y .
MULTIVARIABLE DIFFERENTIAL CALCULUS 103
Solution. Obviously, f ∈ C 2 (R2 ) and it is not difficult to compute its second order
partial derivatives at the point (1, 1). As a result, we get
f ′′ (1, 1) = 6 e2 dx2 + 4 e2 dx dy + 4 e2 dy dx + 6 e2 dy 2 .
Exercise 5.47 Compute the first and the second differential for the function f :
R2 → R, f (x, y) = x3 + y 3 + ln (2 + x2 + y 2 ) at the point (1, 1).
Exercise 5.48 Compute the second derivative at the point (2, 2) of the function
f : R2 → R2 , defined by
f (x, y) = (x4 + y 4 , x2 + y 2 ).
prove that
∂2f ∂2f
(0, 0) ̸= (0, 0).
∂x∂y ∂y∂x
Exercise 5.50 Compute the first and the second differential at the point (1, 1) for
the function f : R2 → R, defined by
f (a + tu) − f (a)
δf (a, u) = lim ∈ Rm ,
t→0 t
then δf (a, u) is called the derivative of the function f , at the point a, in the direction
u.
104 REAL ANALYSIS
δf (a, u) = f ′ (a)(u), ∀u ̸= 0.
is continuous at the point (0, 0), but it is not Gâteaux differentiable at this point.
On the other hand, if we consider the function f : R2 → R, defined by
x2
, y ̸= 0,
f (x, y) = y
0, y = 0,
MULTIVARIABLE DIFFERENTIAL CALCULUS 105
it is not difficult to see that f is Gâteaux differentiable at the point (0, 0), but it is
not continuous at this point.
It is worth noting that the Gâteaux derivative is not, in general, linear, unlike
the Fréchet derivative. Moreover, let us mention that a similar notion can be defined
for functions between more general spaces, such as locally convex topological vector
spaces or Banach spaces.
If we take
t = xj − aj ,
then
∂f f (a + tej ) − f (a) ∂f
(a) = lim = (a),
∂ej t→0 t ∂xj
where
ej = (0, . . . , 1, . . . , 0).
106 REAL ANALYSIS
Hence, partial derivatives can be seen as particular directional derivatives along the
standard basis vector directions.
So, the slopes in all directions are known from slopes in n directions (given by the
partial derivatives).
f (x, y) = 4 − 2x2 − y 2
and
1
u = − √ (1, 1),
2
then
∂f √
(1, 1) = 3 2.
∂u
Exercise 5.59 Let U be a nonempty open set in Rn and let f : U → Rm . Also, let
u ∈ Rn , u ̸= 0. Show that if f is derivable at the point a in the direction u, then f
is derivable at a in the direction −u and
∂f ∂f
(a) = − (a).
∂(−u) ∂u
∂f ∂f
(a) = α (a).
∂(αu) ∂u
Exercise 5.60 Let U be a nonempty open set in Rn and let f : U → Rm . Also, let
u ∈ Rn , u ̸= 0. Prove that if f is derivable at the point a in the direction u, then f
is continuous at a in the direction u.
MULTIVARIABLE DIFFERENTIAL CALCULUS 107
Prove that f is not differentiable at the point (0, 0), despite the fact that it possesses
directional derivatives at this point with respect to any direction u ̸= 0.
If m = n, the determinant
Jf (a) = det Mf (a)
was called the Jacobian of the function f at the point a and it was also denoted by
∂(f1 , . . . , fn )
Jf (a) = (a).
∂(x1 , . . . , xn )
F (x, y) = g(x2 + y 2 ).
∂F ∂F
Compute and .
∂x ∂y
Solution. If we denote by
u(x, y) = x + y
and by
v(x, y) = x3 + y 3 ,
then
∂F ∂g ∂u ∂g ∂v
=y+ + ,
∂x ∂u ∂x ∂v ∂x
i.e.
∂F ∂g ∂g
=y+ + 3x2 .
∂x ∂u ∂v
In a similar manner, we get
∂F ∂g ∂g
=x+ + 3y 2 .
∂y ∂u ∂v
∂F ∂F
where g : R3 → R is an arbitrary given function of class C 1 . Compute , and
∂x ∂y
∂F
.
∂z
Solution. If we denote
u(x, y, z) = x + y + z,
v(x, y, z) = x2 + y 2 + z 2
and
w(x, y, z) = xyz,
then
∂F ∂g ∂u ∂g ∂v ∂g ∂w
= yz + + + ,
∂x ∂u ∂x ∂v ∂x ∂w ∂x
110 REAL ANALYSIS
i.e.
∂F ∂g ∂g ∂g
= yz + + 2x + yz .
∂x ∂u ∂v ∂w
In a similar manner, we get
∂F ∂g ∂g ∂g
= xz + + 2y + xz .
∂y ∂u ∂v ∂w
and
∂F ∂g ∂g ∂g
= xy + + 2z + xy .
∂z ∂u ∂v ∂w
Exercise 5.71 Let
F (x, y) = xy + g(x2 + y 2 , xy),
∂2F ∂2F
where g : R2 → R is an arbitrary given function of class C 2 . Compute ,
∂x2 ∂y 2
∂2F
and .
∂x∂y
Solution. If we denote by
u(x, y) = x2 + y 2
and by
v(x, y) = xy,
then
∂2F 2
2∂ g ∂2g 2
2∂ g ∂g
2
= 4x 2
+ 4xy + y 2
+2 ,
∂x ∂u ∂u∂v ∂v ∂u
∂2F ∂2g ∂2g ∂g ∂g
= 1 + 4xy 2 + (2x2 + 2y 2 ) + xy 2 +
∂x∂y ∂u ∂u∂v ∂v ∂u
and
∂2F 2
2∂ g ∂2g 2
2∂ g ∂g
2
= 4y 2
+ 4xy + x 2
+2 .
∂y ∂u ∂u∂v ∂v ∂u
Exercise 5.72 Compute the first and the second differential for the function
∂2F ∂2F
∆F = 2
+ .
∂x ∂y 2
Exercise 5.76 Prove that the function
z(x, y) = f (x + φ(y)),
∂z ∂ 2 z ∂z ∂ 2 z
= .
∂x ∂x∂y ∂y ∂x2
Exercise 5.77 Prove that the function
where φ, ψ are given functions of class C 2 and a > 0, satisfies the equation:
∂2g 2
2 ∂ g
= a ,
∂t2 ∂x2
called the equation of the vibrating string.
∑
n
∂
∇= ei
i=1
∂xi
∂
∇= ei
∂xi
∂f
(a) = ⟨∇f (a), u⟩ = ∥∇f (a)∥ cos θ,
∂u
where θ is the angle between the vectors ∇f (a) and u. Therefore, the rate of change
of the function f in the direction u depends on u and varies from −∥∇f (a)∥ (when
θ = π, i.e. u points in a direction opposite to ∇f (a)) to ∥∇f (a)∥ (when θ = 0, i.e.
u points in the same direction as ∇f (a)). The vector ∇f (a) points in the direction
of the greatest rate of increase of the function f at the point a and the greatest rate
of change is exactly the magnitude of this vector, i.e. ∥∇f (a)∥.
MULTIVARIABLE DIFFERENTIAL CALCULUS 113
f (x, y, z) = 2x + y 2 − cos z.
f (x, y, z) = x y z.
∂2f ∂2f
∆f = 2 + ··· + 2 .
∂x1 ∂xn
114 REAL ANALYSIS
So, the Laplace operator is a second order differential operator defined as the sum
of all the unmixed second partial derivatives:
∑
n
∂2
∆= .
i=1
∂x2i
f (x, y) = ln(x2 + y 2 ).
∆f = 0.
∂2f 2(y 2 − x2 )
= .
∂x2 (x2 + y 2 )2
In a similar manner,
∂2f 2(x2 − y 2 )
= .
∂y 2 (x2 + y 2 )2
Hence,
∂2f ∂2f
∆f = 2
+ 2 = 0.
∂x ∂y
This proves that our function f is harmonic.
∂v1 ∂vn
div −
→
v = + ··· + .
∂x1 ∂xn
→
−
Exercise 5.89 Let −
→
r = (x, y, z) ∈ R3 , −
→ ̸ 0 and
r =
r = ∥−
→
r ∥,
i.e. √
r= x2 + y 2 + z 2 .
Prove that the divergence of the field
→ q→
− − r
E = 3
r
−
→
is zero. Hence, E is a solenoidal field.
→
−
Solution. Let E = (Ex , Ey , Ez ), where
qx qy qz
Ex = 3
, E y = 3 , Ez = 3 .
r r r
Since, by the chain rule,
∂r3
= 3xr,
∂x
it is easy to see that
∂Ex r3 − 3x2 r
=q .
∂x r6
In a similar manner,
∂Ey r3 − 3y 2 r
=q
∂y r6
and
∂Ez r3 − 3z 2 r
=q .
∂z r6
Therefore,
→
−
div E = 0.
curl →
−
v =∇×−
→
v,
i.e.
−
→
e1 −
→
e2 −
→
e3
∂ ∂ ∂
curl →
−
v =
∂x1 ∂x2 ∂x3
v1 v2 v3
and
curl (f (r)−
→
r ) = 0,
∗
for any f ∈ C 1 (U ), U ⊆ R+ .
−
→ −
→ → −
Exercise 5.96 Let V = V0 + −
ω ×→r be the velocity field in a rigid solid. Show
that
−
→
curl V = 2−
→
ω.
and
−
→
div V = 0.
−
→
v = −∇Φ.
ggrad f − f grad g
c) grad (f /g) = , g(x) ̸= 0, ∀x ∈ U .
g2
Exercise 5.99 Show that
div (α −
→
u +β−
→
v ) = α div −
→
u + β div −
→
v,
for all α, β ∈ R, ∀−
→
u,−
→
v ∈ C 1.
div (φ−
→
v ) = φ div −
→
v + (grad φ) · −
→
v,
for all φ, −
→
v ∈ C 1.
div (−
→
u ×−
→
v)=−
→
v · rot −
→
u −−
→
u · rot −
→
v,
for all −
→
u,−
→
v ∈ C 1 . As a consequence, if →
−
u and −
→
v are irrotatational vector fields,
→
− −
→
then u × v is solenoidal.
118 REAL ANALYSIS
div (grad φ) = △ φ,
for all φ ∈ C 2 .
for all φ ∈ C 2 .
rot (α −
→
u +β−
→
v ) = α rot −
→
u + β rot →
−
v, ∀α, β ∈ R,
for all →
−
u,−
→
v ∈ C 1.
rot (φ−
→
v ) = φ rot −
→
v −−
→
v × grad φ,
for all φ, −
→
v ∈ C 1.
rot (rot →
−
u ) = grad (div −
→
u ) − △−
→
u,
for any −
→
u ∈ C 2.
for any −
→
v ∈ C 2.
MULTIVARIABLE DIFFERENTIAL CALCULUS 119
∑
p
f (k) (a) f (p+1) (ξp )
f (x) = (x − a)k + (x − a)p+1 . (5.7)
k=0
k! (p + 1)!
If we denote by
∑
p
f (k) (a)
Tp (x, a) = (x − a)k
k=0
k!
f (p+1) (ξp )
Rp (x) = (x − a)p+1
(p + 1)!
∫ x
1
Rp (x) = (x − t)p f (p+1) (t) dt
p! x0
Remark 5.110 If in Theorem 5.107 we expand about the origin, i.e. we take a = 0,
we get the so-called Maclaurin’s formula.
120 REAL ANALYSIS
Let us write down Maclaurin’s series for some important elementary functions.
f (x) = x3 ex .
Prove that ∞
∑ xk+3
f (x) = .
k=0
(k)!
f (x) = ln (1 + x).
Prove that ∞
∑ xk+1
f (x) = (−1)k .
k=0
k+1
MULTIVARIABLE DIFFERENTIAL CALCULUS 121
1
f (x) = ,
1−x
show that ∞
∑
f (x) = xk .
k=0
Exercise 5.119 Using the binomial expansion, deduce the expression for the clas-
sical kinetic energy from the formula of total relativistic energy.
Solution. The total relativistic energy of a particle of mass m and velocity v is given
by
( )−1/2
v2
E = mc 2
1− 2 .
c
122 REAL ANALYSIS
1 3 v2
E = mc2 + mv 2 + mv 2 2 + · · · .
2 8 c
So, ( )
1 3 v2
2
E = mc + mv 2 1 + + ··· .
2 4 c2
The first term, mc2 is the rest mass energy. Then,
( )
1 3 v2
Ekinetic = mv 2 1 + + ··· .
2 4 c2
If the particle velocity is much smaller than the velocity of light, i.e. if v ≪ c, the
expression in the parentheses reduces to unity and we see that we obtain the classical
result for the kinetic part of the total relativistic energy.
1 ∑ n
∂mf
+ (a)(xi1 − ai1 ) . . . (xim − aim )
m! i ,...,i =1 ∂xi1 . . . ∂xim
1 m
1 ∑
n
∂ m+1 f
+ (ξ)(xi1 − ai1 ) . . . (xim+1 − aim+1 )
(m + 1)! i ∂xi1 . . . ∂xim+1
1 ,...,im+1 =1
or
∑
n
∂f
f (x1 , . . . , xn ) = f (a1 , . . . , an ) + (a1 , . . . , an )(xj − aj )
j=1
∂xj
MULTIVARIABLE DIFFERENTIAL CALCULUS 123
1 ∑ n
∂2f
+ (a1 , . . . , an )(xj − aj )(xk − ak )
2! j,k=1 ∂xj ∂xk
1 ∑ ∂mf
+··· + (a1 , . . . , an )(x1 − a1 )j1 . . . (xn − an )jn
m! j
1 +···+jn =m
∂xj11 . . . ∂xjnn
1 ∑ ∂ m+1 f
+ (ξ)(x1 − a1 )k1 . . . (xn − an )kn .
(m + 1)! k
1 +···+kn =m+1
∂xk11 . . . ∂xknn
∑
m
1
f (x) = f (p) (a) (x − a)p + εf,a (x) ∥x − a∥m , (5.11)
p=0
p!
with
lim εf,a (x) = 0.
x→a
124 REAL ANALYSIS
Exercise 5.124 Write Taylor’s formula of the third order for the function f : U →
R, f (x, y) = ln(1 + x + y), about the point (0, 0). Here,
Solution. Obviously, f is four times differentiable on the convex set U and, comput-
ing the needed derivatives, we obtain
1 1 1
ln(1 + x + y) = x + y − (x + y)2 + (x + y)3 − (x + y)4 (1 + ξ + η)−4 ,
2 3 4
where (ξ, η) = (tx, ty), with t ∈ (0, 1).
Exercise 5.125 Write Taylor’s formula of the third order for the function f : R2 →
R, f (x, y) = ex+y , about the point (1, −1).
Solution. The function f is four times differentiable on the convex set R2 . Computing
the needed derivatives, we obtain
1 1 1 eξ+η
ex+y = 1 + (x + y) + (x + y)2 + (x + y)3 + (x + y)4 ,
1! 2! 3! 4!
where (ξ, η) = (1 + tx, −1 + ty), with t ∈ (0, 1).
Exercise 5.126 Write Taylor’s formula of the order 2 about the point (x0 , y0 ) =
(0, 0) for the function f : R2 → R, defined by f (x, y) = ex sin y.
( )
π
Exercise 5.127 Write Taylor’s formula of the order 2 at the point (x0 , y0 ) = 0,
2
for the function f : R2 → R, defined by f (x, y) = e2x cos y.
( )
π
Exercise 5.128 Write Taylor’s formula of the order 2 at the point (x0 , y0 ) = 0,
2
for the function f : R2 → R, defined by f (x, y) = cos x cos y.
Exercise 5.129 Write Taylor’s formula of the order 2 at the point (x0 , y0 ) = (1, 1)
for the function f : (0, ∞) × R → R, defined by
f (x, y) = xy .
MULTIVARIABLE DIFFERENTIAL CALCULUS 125
we have:
∂(f1 , f2 )
| = ρ.
∂(ρ, θ) (ρ,θ)
It is easy to see that f ∈ C 1 (U ). Also, for any ρ ∈ (0, ∞), f ′ (ρ, θ) ∈ Isom (U, V ).
So, f is a C 1 -diffeomorphism on U .
∂h
(a, b)
φ ′ (a) = − ∂x . (5.13)
∂h
(a, b)
∂y
2) fi (a, b) = 0, i = 1, m;
∂(f1 , . . . , fm )
3) (a, b) ̸= 0.
∂(y1 , . . . , ym )
Then, there exist ∅ ̸= U × V ⊆ W , U × V open, with (a, b) ∈ U × V , and gi : U →
R, i = 1, m, gi ∈ C 1 such that the system:
f1 (x1 , . . . , xn , y1 , . . . , ym ) = 0,
........................................,
fm (x1 , . . . , xn , y1 , . . . , ym ) = 0
Moreover,
bj = gj (a1 , . . . , an ), j = 1, m
and
∂(f1 , . . . , fm )
(x, g(x))
∂yj ∂(y1 , . . . , yj−1 , xi , yj+1 , . . . , ym )
(x) = − ,
∂xi ∂(f1 , . . . , fm )
(x, g(x))
∂(y1 , . . . , ym )
for any x = (x1 , ..., xn ) ∈ U, i = 1, n, j = 1, m, g = (g1 , . . . , gm ).
The above corollary gives conditions for the solvability of a system of functional
equations and offers formulas for calculating the partial derivatives of the implicitly
specified functions.
∂F ∂F ∂F
(x, y, z) · (x, y, z) · (x, y, z) ̸= 0.
∂x ∂y ∂z
∂F
Since F is of class C 1 on D and (x, y, z) ̸= 0, it follows that locally, around
∂x
the point (x, y, z), we can express x in terms of y and z, i.e. there exists X such
that
x = X(y, z)
and
F (X(y, z), y, z) = 0.
Therefore,
∂
[F (X(y, z), y, z)] = 0,
∂y
MULTIVARIABLE DIFFERENTIAL CALCULUS 129
i.e.
∂F ∂X ∂F
+ = 0,
∂x ∂y ∂y
which implies that
∂F
∂X ∂y
=− .
∂y ∂F
∂x
In a similar manner, we get
∂F ∂F
∂Y ∂Z
= − ∂z , = − ∂x .
∂z ∂F ∂x ∂F
∂y ∂z
So, if
F (x, y, z) = 0,
then, locally,
∂X ∂Y ∂Z
· · = −1. (5.14)
∂y ∂z ∂x
Formula (5.14) is known as the triple product rule.
Exercise 5.142 Compute the derivative of the function y = f (x) implicitly defined
by the equation
x2 + 4y 2 + 2xy = 12
∂h
(2, 1) 1
φ ′ (2) = − ∂x =− .
∂h 2
(2, 1)
∂y
130 REAL ANALYSIS
Prove that, in a neighbourhood of the point (1, 1, −1), we can uniquely determine y
dy dz d2 y d2 z
and z as functions of x. Also, compute (1), (1), 2
(1), (1).
dx dx dx dx2
Solution. If we take f = (f1 , f2 ) : R × R2 → R2 , defined by
f1 (x, y, z) = x + y + z − 1,
f (x, y, z) = x3 − y 3 + z 3 + 1,
2
and
g1 (1) = 1, g2 (1) = −1.
So, on U , we have
x + g1 (x) + g2 (x) = 1,
x3 − g 3 (x) + g 3 (x) = −1.
1 2
So, by differentiation, we get
dg1 dg2
(1) = 0, (1) = −1.
dx dx
In a similar manner, we obtain
d2 g1 d2 g2
(1) = 0, (1) = 0
dx2 dx2
MULTIVARIABLE DIFFERENTIAL CALCULUS 131
Exercise 5.144 Compute the derivative of the function y = f (x) implicitly defined
by the equation x2 − y − cos y = 0 in a neighbourhood of the point (1, 0).
Exercise 5.145 Compute the first and second order partial derivatives for the func-
tion z=f(x,y) defined implicitly by the equation x3 + y 3 + z 3 − 3xyz = 0 in a neigh-
bourhood of the point (−1, 1, 0).
∂2f
Exercise 5.146 Compute (1, 0) for the function z = f (x, y) defined by the
∂x∂y
equation x2 + y 2 − z 2 + xz − yz − 1 = 0 in a neighbourhood of the point (1, 0, 1).
Exercise 5.147 Compute the first and second order partial derivatives for the func-
tion z=f(x,y) defined implicitly by the equation x3 + y 3 + z 3 − 3xyz = 0 in a neigh-
bourhood of the point (−1, 1, 0).
Exercise 5.148 Compute the differential d f (0, 0) for the function z=f(x,y) defined
implicitly by the equation x + y + z − sin xyz = 0 in a neighbourhood of the point
(0, 0, 0).
x3 − xz + y 3 = 0.
Prove that, in a neighbourhood of the point (1, −1, 0), we can uniquely determine x
dx dy
and y as functions of z. Compute (0), (0).
dz dz
We shall use the terms global or absolute maximum and, respectively, global or
absolute minimum, to refer to the overall maximum and minimum values of the
function on the range under consideration. Also, the term local is synonymous with
relative. Moreover, extremum is a term that will be used to denote both a maximum
and a minimum: a local extremum will be a local or relative maximum or minimum,
and a global extremum will be a global or absolute maximum or minimum.
For instance, the point (0, 0) is a strict local minimum for the function f (x, y) =
x2 + y 2 , but not for g(x, y) = (x + y)2 .
It is not difficult to see that, exactly like in the one-dimensional case, we have
the following result:
Remark 5.153 The points a at which f ′ (a) = 0 are called stationary points.
several real variables, these notions coincide, and, so, the condition that a point is
a critical one for f is equivalent to the fact all the partial derivatives of f must be
zero at that point.
Also, let us notice that exactly like in the one-dimensional case, one can prove
that if we consider a function f : Rn → R which is continuous on an open set U and
if S is a nonempty closed and bounded subset of U , then f has a maximum value
and a minimum value on S. Notice that this result needs no reference to derivatives.
Moreover, let us notice that global extrema of a given function f on a closed
and bounded set S can occur only at boundaries, non-differentiable points and sta-
tionary points. But, sometimes, dealing with the boundary of a given set and with
non-differentiable points requires a significant amount of work. Therefore, in this
paragraph, we shall work only with smooth functions defined on open sets and, in
such a case, as already mentioned, local extrema can be found among the stationary
(critical) points of the given function. Still, we have to keep in mind that this is
only a necessary condition, since not all the stationary points are local extrema. So,
a local extremum must occur at a stationary point, but the converse may not be
true. A stationary point which is not a local extremum is called a saddle point.
e
h(x) = h(x, x), ∀x ∈ E.
∑
n
h(x) = aij xi xj ,
i,j=1
det(aij ) ̸= 0.
∑
n
∂2f
(a)xi xj ≥ 0, ∀x ∈ Rn .
i,j=1
∂xi ∂xj
∑
n
∂2f
(a)xi xj ≤ 0, ∀x ∈ Rn .
i,j=1
∂xi ∂xj
∂2f ∂2f
(a) (a)
∂x2 ∂x∂y
D2 (a) = ̸= 0.
∂2f ∂2f
(a) (a)
∂y∂x ∂y 2
If D2 (a) < 0, then a is not a local extremum for f . If D2 (a) > 0, then a is a strict
∂2f
local minimum if D1 (a) = (a) > 0 and a strict local maximum if D1 (a) < 0.
∂x2
Let us notice that if the point a is degenerate, then the test is inconclusive.
∂2f ∂2f
2 (a) · · · (a)
∂x1 ∂x1 ∂xk
Dk (a) = ······················ .
∂2f ∂2f
(a) · · · (a)
∂xk ∂x1 ∂x2k
Then:
a) if ∆1 (a) > 0, ∆2 (a) > 0, . . . , ∆n (a) > 0, then Hf (a) is positive definite and
a is a strict local minimum;
136 REAL ANALYSIS
b) if ∆1 (a) < 0, ∆2 (a) > 0, . . . , (−1)n ∆n (a) > 0, then Hf (a) is negative definite
and a is a strict local maximum.
Notice that if Hf (a) is positive or negative semi-definite, we cannot decide upon the
nature of the point a using this method. The test is inconclusive and this is an open
case. Also, notice that if the Hessian Hf (a) is indefinite, then the point a is a saddle
point for f . If the point a is degenerate, no conclusion can be drawn.
Solution. Obviously, f ∈ C ∞ (R2 ). The critical points for the function f are the
′
solutions of the equation f (x, y) = 0, i.e. solutions for the following system:
∂f
(x, y) = 0,
∂x
∂f
(x, y) = 0.
∂y
So, computing the partial derivatives of f , we have to solve the system
2 2
x +y =5
(5.15)
xy = 2
which, obviously, has four solutions: (1, 2), (2, 1), (−1, −2), (−2, −1). So, we have
four critical points and for deciding which of them are local extrema for the function
f we shall use Theorem 5.162. So, we have to compute the partial derivatives of the
second order for f . We obtain:
Now, we shall take look at each critical point and we shall decide, using Theorem
5.162, if it is a local extremum or not.
For the point (1, 2), we obtain D2 (1, 2) < 0 and, so, this point is not a local
extremum. Also, the point (−1, −2) is not a local extremum, because D2 (−1, −2) <
0.
For the point (2, 1), we get D2 (2, 1) > 0 and D1 (2, 1) > 0. Therefore, this point is
a local minimum and fmin = f (2, 1) = −28. In a similar manner, we obtain that the
point (−2, −1) is a local maximum, because D2 (−2, −1) > 0 and D1 (−2, −1) < 0.
Moreover, fmax = f (−2, −1) = 28.
f (x, y, z) = x2 + y 2 + z 2 + 4x + 2y − 8z.
Solution. It is not difficult to see that the only critical point of the function f is
(−2, −1, 4). For this point, D3 (−2, −1, 4) > 0, D2 (−2, −1, 4) > 0, D1 (−2, −1, 4) >
0. Hence, the point (−2, −1, 4) is a local minimum. Moreover, fmin = −21.
f (x, y) = x2 + 2xy − 4x − y 2 ,
has no extrema.
Solution. The point (1, 1) is the only stationary point of f . The Hessian of f at this
point is
2 2
Hf (1, 1) = .
2 −2
Therefore, (1, 1) is a saddle point. The function f has no extrema.
Exercise 5.167 Determine the nature of the stationary points of the function f :
R2 → R, defined as being
2 +y 2
f (x, y) = (x2 + y 2 )2 ex .
138 REAL ANALYSIS
In other words, the extrema of f on S can be found among the stationary points of
the auxiliary function f − Λ ◦ h.
The map Λ being in L(Rm , R), it follows that there exists λj ∈ R such that
∑
m
Λ= λj q j ,
j=1
∂f ∑m
∂hj
(c) − λj (c) = 0, 1 ≤ i ≤ n + m. (5.17)
∂xi j=1
∂xi
∂f ∑m
∂hj
(c) − λj (c) = 0, 1 ≤ i ≤ n + m, hj = 0, 1 ≤ j ≤ m (5.18)
∂xi j=1
∂xi
Let us notice that Lagrange’s method only provides candidates for extrema. We
must check if indeed the function f possesses an extremum and then we have to
carefully evaluate f at every critical point in order to decide upon the nature of all
theses critical points and to determine correctly the extrema.
Also, it is worth mentioning that Lagrange multipliers can fail to identify all the
conditional extrema if ∇h = 0. So, additionally to Lagrange’s method, we have to
check carefully the points with ∇h(x, y) = 0.
Solution. Let
The stationary points of the Lagrangian F can be found by solving the following
system:
1 − 2λx = 0
−2 − 2λy = 0
(5.19)
2 − 2λz = 0
2 2
x +y +z =92
( )
1 1 1
We get = , i.e. λ = ± . Hence, the stationary points of F are
λ2 1, −2, 2;
( 4 ) 2 2
1
and −1, 2, −2; − .
2
MULTIVARIABLE DIFFERENTIAL CALCULUS 141
1
For λ = , the associated quadratic form is negative definite, i.e. the point
2
(1, −2, 2) is a local maximum for F and a local conditional maximum for the function
f , the maximum value being fmax = 9.
1
For λ = − , the associated quadratic form is positive definite, i.e. the point
2
(−1, 2, −2) is a local minimum for F and, of course, a local conditional minimum
for f , the minimum value being fmin = −9.
0 < pi ≤ 1
and
p1 + · · · + pn = 1. (5.20)
∑
n
H(p1 , . . . , pn ) = − pj log2 pj . (5.21)
j=1
Obviously,
H(p1 , . . . , pn ) ≥ 0, ∀p1 , . . . , pn .
Let us notice that by this method it is not always possible to find all the conditional
extrema of a given function (see, for instance, [19] and [34]), i.e. the Lagrange
equations ∇f (x, y) = λ∇h(x, y), h(x, y) = 0 do not give us all the conditional
extrema.
Exercise 5.177 Find the local extrema of the function f : [0, ∞)3 → R, defined by
f (x, y, z) = xyz, subject to the constraint x + y + z = 1.
Exercise 5.179 Find the local extrema of the function f : (0, ∞)3 → R, defined by
f (x, y, z) = xy + yz + zx, subject to the constraint xyz = 1.
Exercise 5.180 Find the local extrema of the function f : (0, ∞)3 → R, given by
f (x, y, z) = x2 + y 2 + z 2 , under the constraint x + y + z = 1.
MULTIVARIABLE DIFFERENTIAL CALCULUS 143
x2 y 2
+ 2 = 1, a>b>0
a2 b
and has the largest perimeter.
x2 y 2 z 2
+ 2 + 2 = 1,
a2 b c
for a > b > c > 0.
144
Chapter 6
The sequence (fk )k≥1 is said to be pointwise convergent at the point a ∈ A if the
number sequence (fk (a))k≥1 is convergent.
The sequence (fk )k≥1 is said to be pointwise convergent on the set A if (fk )k≥1 is
convergent at any point a ∈ A. In this case, we can define a function f : A → R, by
putting
f (x) = lim fk (x), x ∈ A.
k→∞
The function f is called the limit of (fk )k≥1 on A. We shall use the notation fk → f
to denote the fact that f is the limit of the sequence (fk )k .
We shall denote by Ac the set of all the points where (fk )k is pointwise convergent.
145
146 REAL ANALYSIS
sin kx
fk (x) = , k ≥ 1.
k2
for any k ≥ kε .
u
We shall write fk −→ f to denote the fact that the sequence (fk )k≥1 is uniformly
convergent to f .
and
lim ak = 0,
k→∞
i.e. ak 9 0. Therefore, the sequence (fk )k is not uniformly convergent to f on [0, 1].
Example 6.10 The pointwise convergence does not preserve, in general, the proper-
ties of the functions fk . For instance, the pointwise limit of a sequence of continuous
functions may not be continuous. As an example, let us consider the sequence of
continuous functions
fk (x) = xk , x ∈ [0, 1],
which is pointwise convergent to the function
{
0, if x ∈ [0, 1),
f (x) =
1, if x = 1.
∑
k
Sk (x) = fi (x). (6.1)
i=1
The pair ((fk )k≥1 , (Sk )k≥1 ) is said to be a series of functions (the series associated
to the sequence (fk )k≥1 ).
∑ ∑
We use the notation fk . Sk is called the kth partial sum of the series fk and
k≥1 k≥1
fk is said to be its general term.
∑
Definition 6.18 The series fk (x) is said to be convergent at x ∈ A if the se-
k≥1
quence (Sk (x))k≥1 is convergent. The limit S(x) of (Sk (x))k≥1 is called the sum of
∑
the convergent series fk (x) at the point x.
k≥1
We shall use the following symbol to denote the sum of a convergent series:
∞
∑
S= fk .
k=0
∑
then there exists f : I → R such that fk is uniformly convergent to f on I, f is
k≥1
derivable and
f ′ = g.
sin kx
fk (x) = , for p > 1 and x ∈ R.
kp
∑
The series fk is absolutely and uniformly convergent on R, because
k≥1
sin kx 1
fk (x) = ≤ p
kp k
∑ 1
and the series is convergent for p > 1.
k≥1
kp
∑ cos kx
Exercise 6.30 Show that the series is uniformly convergent on R.
k≥1
x2 + k 2
Solution. Since
cos kx 1 1
≤ 2 ≤ 2
x2 + k 2 x + k2 k
∑ 1
and the number series is convergent, it follows immediately that our series is
k2 k≥1
uniformly convergent on R.
∑ 2x
b) fk , where fk : R → R, fk (x) = arctan ;
k≥1
x2 + k4
∑ (−1)k
c) fk , where fk : [0, ∞) → R, fk (x) = .
k≥1
x+k
152 REAL ANALYSIS
fk (x) = ak (x − x0 )k , k ≥ 0,
1
R= √ ,
lim k
|ak |
k→∞
√ √
with the convention that R = 0 if lim k | ak | = ∞ and R = ∞ if lim k |ak | = 0.
k→∞ k→∞
Then:
1) if R > 0, the power series (6.2) is absolutely convergent for any x such that
|x − x0 | < R and divergent for any x with |x − x0 | > R;
2) if R > 0, the power series (6.2) is uniformly convergent over any compact
set [x0 − r, x0 + r] with 0 < r < R;
Remark 6.34 The non-negative quantity R is called the radius of convergence for
the series (6.2).
The set
Dc = {x ∈ R | |x − x0 | < R}
is called the interval or the set of convergence for the series (6.2). At the end points
of the interval of convergence, i.e. at the points x − x0 = R and x − x0 = −R, the
power series (6.2) may be both convergent or divergent.
SEQUENCES AND SERIES OF FUNCTIONS 153
ak
Remark 6.35 If there exists lim , then
k→∞ ak+1
ak
R = lim .
k→∞ ak+1
ak
Moreover, if there exists lim , then
k→∞ ak+1
ak
R = lim .
k→∞ ak+1
∑
Example 6.36 The radius of convergence of the series k! xk is equal to 0. There-
k≥0
fore, this series is convergent only for x = 0, i.e. Dc = {0}.
∑ xk
Example 6.37 The radius of convergence for the series is equal to ∞. Thus,
k≥0
k!
the set of convergence is Dc = R.
Exercise 6.38 Find the radius and the domain of convergence for the series
∑ (x − 5)k
(−1)k−1 .
k≥1
k 3k
Solution. It is easy to see that the radius of convergence for this series is equal to
3. Therefore, using Cauchy-Hadamard theorem, we see that our series is convergent
for x ∈ (2, 8) and divergent for x ∈ (−∞, 2) ∪ (8, ∞). Let us analyze what happens
∑ (−1)2k−1
at the end points x = 2 and x = 8. For x = 2, our series becomes ,
k≥1
k
∑ (−1)k−1
which is a divergent series. For x = 8, our series becomes and, using
k k≥1
Leibniz criterion, it is a convergent series. So, the interval of convergence of the
given power series is
Dc = (2, 8].
Remark 6.39 Every convergent power series defines on its interval of convergence
Dc a function
∞
∑
f (x) = ak (x − x0 )k .
k=0
154 REAL ANALYSIS
When two functions f and g are decomposed into power series about the same point
x0 , the power series of their sum or difference can be obtained by termwise addition
or subtraction. More precisely, if:
∞
∑
f (x) = ak (x − x0 )k
k=0
and ∞
∑
g(x) = bk (x − x0 )k ,
k=0
then ∞
∑
(f ± g)(x) = (ak ± bk )(x − x0 )k .
k=0
Also, we can define the product of the power series f (x) and g(x) as being:
∞ ∑
∞ ∞
( k )
∑ ∑ ∑
(f g)(x) = ai bj (x − x0 )
i+j
= ai bk−i (x − x0 )k .
i=0 j=0 k=0 i=0
The sequence
∑
k
ck = ai bk−i
i=0
Remark 6.40 Usually, we shall work with power series about the point x0 = 0, i.e.
with series of the form
∑
ak xk ,
k≥0
is said to be invertible if there exists a power series g(x) such that f (x) g(x) = 1,
for any x ∈ Dc .
SEQUENCES AND SERIES OF FUNCTIONS 155
∞
∑
Remark 6.42 The power series ak xk is invertible if and only if a0 ̸= 0.
k=0
Example 6.43 The power series 1 − x is invertible and its inverse is the geometric
∑
series xk .
k≥0
Remark 6.44 If a power series having the radius R is uniformly convergent over
[x0 − r, x0 + r], 0 < r < R, then its sum defines a continuous function at any point
x with |x − x0 | < R.
Remark 6.45 Let R be the radius of convergence for the series (6.2). Then, the
∑
series k ak (x − x0 )k−1 (obtained by differentiating (6.2) with respect to x term
k≥1
by term) has the same radius of convergence.
Remark 6.46 Let R > 0 be the radius of convergence for the series (6.2) and let f ,
defined on (x0 − R, x0 + R), be the sum of this series. Then, f is indefinite derivable
(term by term) on this interval.
Example 6.47 For any value of x between −1 and +1, we may write
1
= 1 − x2 + x4 − x6 + · · · + (−1)k x2k + · · · .
1 + x2
Integrating this series term by term between the limits 0 and x, where |x| < 1, we
find
x x3 x2k+1
arctan x = − + · · · + (−1)k + ···.
1 3 2k + 1
Since the new series converges also for x = 1, it follows that
π 1 1 1
= 1 − + − · · · + (−1)k + ···.
4 3 5 2k + 1
Exercise 6.48 Compute the radius and the domain of convergence for the following
series:
∑ xk ∑ xk ∑ ∑ xk
k
a) ; b) ; c) k! x ; d) ;
k≥0
k! k≥1
kk k≥0 k≥1
k2
∑ xk ∑ (x − 5)k ∑ (x − 1)2k
e) ; f) (−1)k−1 ; g) .
k≥1
k 2k k≥1
k 3k k≥1
k 9k
156 REAL ANALYSIS
This definition characterizes the class of the functions which can be locally approx-
imated by convergent power series.
Let us notice that every power series with a positive radius of convergence is
analytic on the interior of its region of convergence. Also, notice that sums and
products of analytic functions are analytic.
For an analytic function, the coefficients ak can be computed as
f (k) (x0 )
ak = .
k!
So, every analytic function can be locally represented by its Taylor series. The global
form of an analytic function is completely determined by its local behavior, i.e. if
f and g are two analytic functions defined on the same open nonempty interval I,
and if there exists an element c ∈ I such that f (k) (c) = g (k) (c), for all k ≥ 0, then
f (x) = g(x) for all x ∈ I.
SEQUENCES AND SERIES OF FUNCTIONS 157
|f (k) (x)| ≤ M k ,
lim Rk (x) = 0.
k→∞
Elementary functions
Using power series, we can rigorously define the most common used elementary
functions of real arguments.
P (x) = a0 + a1 x + · · · + ak xk ,
158 REAL ANALYSIS
with ai ∈ R. The radius of convergence of this power series is ∞ and the polynomial
function is indefinitely derivable on R.
P (x)
f (x) = ,
Q(x)
where P and Q are polynomial functions in x and Q is not the zero polynomial. The
domain of f is the set of all points x for which the denominator Q(x) is not zero.
∑ xk
exp(x) = .
k≥0
k!
∑ x2k+1
sin x = (−1)k .
k≥0
(2k + 1)!
∑ x2k
cos x = (−1)k .
k≥0
(2k)!
∑ x2k+1
sinh x = .
k≥0
(2k + 1)!
∑ x2k
cosh x = .
k≥0
(2k)!
∑ xk+1
8) ln (1 + x) = (−1)k , |x| < 1.
k≥0
k+1
1 1 1
ln 2 = 1 − + − + ···
2 3 4
∑ α(α − 1) · · · (α − k + 1)
9) (1 + x)α = 1 + xk , α ∈ R, x ∈ (−1, 1).
k≥1
k!
The above series is called the binomial series. If α ∈ N, we get the well-known
Newton’s formula.
Exercise 6.56 Prove that the above elementary functions gave the following prop-
erties:
1) ex+y = ex · ey , ∀x, y ∈ R;
′
2) (ex ) = ex , ∀x ∈ R;
eix − e−ix
4) sin x = , ∀x ∈ R;
2i
eix + e−ix
5) cos x = , ∀x ∈ R;
2
ex − e−x
6) sinh x = , ∀x ∈ R;
2
ex + e−x
6) cosh x = , ∀x ∈ R;
2
160 REAL ANALYSIS
7) sin2 x + cos2 x = 1, ∀x ∈ R;
8) cosh2 x − sinh2 x = 1, ∀x ∈ R;
′ ′
9) (sin x) = cos x, (cos x) = − sin x, ∀x ∈ R;
′ ′
10) (sinh x) = cosh x, (cosh x) = sinh x, ∀x ∈ R;
[9] K. Ciesielski Set Theory for the Working Mathematician, Cambridge Uni-
versity Press, 1997.
161
162 REAL ANALYSIS
[12] R. Courant, Differential and Integral Calculus, Wiley, New York, 1992.
[14] L. Elsgolts, Differential Equations and the Calculus of Variations, Mir Pub-
lishers, Moscow, 1980.
[18] P. Halmos, Naive set theory, Princeton, NJ: D. Van Nostrand Company,
1960; reprinted by Springer-Verlag, New York, 1974.
[28] Gh. Sireţchi, Series. Set Theory. Elementary Functions, Tipografia Univer-
sităţii Bucureşti, 1984 (in Romanian).
[29] Gh. Sireţchi, General Topology, Tipografia Universităţii Bucureşti, 1983 (in
Romanian).