2020 - Book - SolvingProblemsInMathematicalA Part 1 PDF
2020 - Book - SolvingProblemsInMathematicalA Part 1 PDF
Tomasz Radożycki
Solving
Problems in
Mathematical
Analysis, Part I
Sets, Functions, Limits, Derivatives,
Integrals, Sequences and Series
Problem Books in Mathematics
Series Editor:
Peter Winkler
Department of Mathematics
Dartmouth College
Hanover, NH 03755
USA
More information about this series at http://www.springer.com/series/714
Tomasz Radożycki
Solving Problems
in Mathematical Analysis,
Part I
Sets, Functions, Limits, Derivatives, Integrals,
Sequences and Series
Tomasz Radożycki
Faculty of Mathematics and Natural
Sciences, College of Sciences
Cardinal Stefan Wyszyński University
Warsaw, Poland
This Springer imprint is published by the registered company Springer Nature Switzerland AG.
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
v
vi Preface
There is also the question of language used in this book. I tried to maximally
simplify it and—in place of abstract terms—use the notions, which are intuitively
clear (and even used in everyday life). Someone may, and would be right, formulate
the objection that they are not precise enough. However, my intention was to
present the issues in such a way that the student, without much effort, could
translate difficult concepts to notions that are more understandable and assimilable.
This observation comes from many years of work at universities. The students’
understanding depends to a large extent on the choice of a simple language,
especially in the first few years of study. To increase the level of abstraction, there
will be time in their further course of study. At the beginning, it is helpful to make
the students aware that many new concepts can be mastered with their present
knowledge and intuition.
With the hope that this set will help to better understand (from a practical point
of view) certain issues of mathematical analysis, I encourage the reader to use other
textbooks that provide exercises for independent work and that certainly cannot be
replaced by this set.
vii
viii Contents
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
Definitions and Notation
Below, some definitions and notation are gathered in order to avoid their repetition
in each applicable chapter.
• Positive integers (i.e., without 0) will be marked with N:
N = {1, 2, 3, . . .} ,
and will be called “naturals.” If we wish to include the 0 in this set, we will
simply write N ∪ {0}, and call it “naturals with zero.”
• The real, rational, and positive integer numbers will be denoted as R+ , Q+ , and
Z+ , respectively. One naturally has Z+ = N. Similarly, symbols R− , Q− , and Z−
will refer to negative numbers.
• If a special notation is not introduced in a particular problem, the symbol X will
mean the whole space.
• In all problems, apart from those contained in Chap. 3 and the last problem
of Sect. 6.1, the Euclidean metric is used as default, based on the Pythagorean
theorem, discussed in detail in Problem 1 of Sect. 3.1. In the case of the set R,
it reduces to “natural metric,” so that the distance of the two numbers x and y is
given by d(x, y) = |x − y|.
• It is assumed that a ball is open. For example, the ball centered at some point x0
and of a radius r is a set of points x satisfying the condition: d(x0 , x) < r. If, in
any problem, a closed ball is needed, it will be written explicitly.
• The function f as a mapping of a set X into Y , formally speaking, apart from the
assignment itself (e.g., the formula for f (x)) requires also the definitions of sets
X and Y . We accept the rule that if in the specific exercise they are not given,
the largest sets for which the formula y = f (x) makes sense is taken. What is
meant always results from the context of the discussed issues. For example, in
the textbook, which is generally concerned with real numbers, we will certainly
not expand the logarithmic function to the complex plane. Similarly, if the set Y
is not given, we assume it to be identical to the image of the function, i.e., f (X).
• The domain of a function will generally be denoted as D or sometimes as X.
xi
xii Definitions and Notation
a(x)
lim = 1.
x→∞ b(x)
Chapter 1
Examining Sets and Relations
In this chapter, we deal with sets and perform various operations on them. We will
also get acquainted with the notion of the “relation.” For the convenience of the
reader, a number of basic properties of sets is first collected, which will be then
used in the remainder of this chapter to help solve problems. The same applies to
the subsequent chapters.
A set is a common notion referring to an object that constitutes a collection of
certain elements. If a given element x belongs to the set A, it is denoted by x ∈ A.
The opposite is x ∈ A. The sets are often defined by their elements, by simply
enumerating them:
A = {a1 , a2 , . . . , an } (1.0.1)
(A ∩ B) = A ∪ B , (1.0.4)
(A ∪ B) = A ∩ B . (1.0.5)
The subset of a given set A is a certain set B composed of the elements of A only
(not necessarily all of them). It is denoted by B ⊂ A. The empty set, i.e., the set
not containing any elements, is a subset of any set: ∅ ⊂ A. Naturally also A ⊂ A.
The two principal rules of the set algebra are
• the distribution of the intersection over the union
A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C), (1.0.6)
A ∪ (B ∩ C) = (A ∪ B) ∩ (A ∪ C). (1.0.7)
Numerical sets can have lower and upper bounds. A lower bound of a set A is a
certain number x satisfying
∀a∈A x ≤ a. (1.0.8)
The largest of the numbers x (if it exists) is called the infimum of A or simply its
greatest lower bound. An upper bound of a set A is a certain number y satisfying
∀a∈A y ≥ a. (1.0.9)
The smallest of the numbers y (if it exists) is called the supremum of A or simply
its least upper bound.
A relation in a certain set A is a subset of the Cartesian product A × A:
R ⊂ A × A.
1.1 Demonstrating Simple Identities 3
Let a and b are two elements of A. They can be R-related, if (a, b) ∈ R. The
notation aRb can also be used. The important case to be considered constitutes the
so-called equivalence relation for which the following three conditions hold:
• Reflexivity: for each a ∈ A there is (a, a) ∈ R.
• Symmetricity: for each a, b ∈ A the implication
holds.
• Transitivity: for every a, b, c ∈ A the implication
holds. Each equivalence relation defines the partition of the set A onto the
equivalence classes. They are subsets of A containing elements equivalent to one
another. Every element of A belongs to exactly one class. The class of a certain
element x will be denoted with [x]R . Any other element of a given class may be
chosen to represent it.
Problem 1
A \ C ⊂ (A \ B) ∪ (B \ C), (1.1.1)
Solution
When solving problems in the calculus of sets, similar to the one considered below,
it is useful to start with a schematic drawing that will allow us to imagine what we
wish to prove. Sometimes such a figure can even save us from the pointless efforts to
prove a false thesis. In Fig. 1.1, three sets in the form of circles: A, B, and C in some
particular configuration have been drawn. On the left the set A \ C and on the right
(A \ B) ∪ (B \ C) are marked in gray. The figure suggests that the inclusion (1.1.1)
really takes place.
Naturally, such a figure is drawn for illustrative purposes only. It would be
reasonable to perform other properly modified sketches for several various config-
4 1 Examining Sets and Relations
urations of sets (e.g., when one or all of them are disjoint). However, even with the
figure drawn as it is allows us to gain some intuition. These types of drawings are
called Venn diagrams.
Let us now recall the rules of set algebra formulated at the beginning of this
chapter:
A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C), (1.1.2)
A ∪ (B ∩ C) = (A ∪ B) ∩ (A ∪ C). (1.1.3)
If we imagine that all the considered sets are subsets of a certain space X, a
notion of a complement set can be introduced. As we remember for a set S, it is the
collection of all elements of X that do not belong to S and is denoted with S . In
accordance with this definition, S = X \ S. The differences of sets occurring in the
problem content may now, with the use of the complement, be written as follows:
A \ B = A ∩ B .
S = T ∪ [· · · ], (1.1.4)
but the set hidden under the symbol [· · · )] is completely irrelevant. In our proof we
will go just this way: using (1.1.2) and (1.1.3), we will try to transform the right-
hand side of (1.1.1) so as to obtain the needed expression (A \ C) ∪ [· · · ]:
(A \ B) ∪ (B \ C) = (A ∩ B ) ∪ (B ∩ C ) = [A ∪ (B ∩ C )] ∩ [B ∪ (B ∩ C )]
= [(A ∪ B) ∩ (A ∪ C )] ∩ [(B ∪ B) ∩ (B ∪ C )],
(1.1.5)
1.1 Demonstrating Simple Identities 5
where the rule (1.1.3) has been used twice. As one knows, the intersection of sets is
associative:
(R ∩ S) ∩ T = R ∩ (S ∩ T ). (1.1.6)
This means that in the expression (1.1.5) one can skip all square brackets because,
no matter how one puts them, one always gets the same result. In addition, it
should be noted that the expression (B ∪ B) is a sum of a set and its complement.
Such a sum is, naturally, the entire space X because each element of the space
belongs either to B or to B . One then may write instead of the right-hand side
of (1.1.5)
(A ∪ B) ∩ (A ∪ C ) ∩ X ∩ (B ∪ C ), (1.1.7)
and simply omit X, since for each set S one has S ∩ X = S. The result is
(A \ B) ∪ (B \ C) = (A ∪ B) ∩ (A ∪ C ) ∩ (B ∪ C ). (1.1.8)
The idea is to get on the right-hand side the expression A \ C, i.e., A ∩ C , but
nothing of that kind has so far appeared. One has A ∪ C , but certainly it is not
the same. However, very often when transforming different expressions, one can
manipulate the equation by inserting the desired expression, if known, all the while,
still maintaining the equality (e.g., by adding and subtracting the same). In our case
we will use the following identity:
A = A ∩ X = A ∩ (C ∪ C) = (A ∩ C ) ∪ (A ∩ C), (1.1.9)
−−−−−−
where the interesting expression has been underlined. The right-hand side will
replace A in the first bracket on the right-hand side of (1.1.8). In that way one gets
The union and intersection are associative, so one is allowed to add braces, which
were absent in (1.1.8). We will now apply (1.1.3) in relation to the union and
intersection of sets marked with arrows in the last formula. As a result, one will
obtain
(A \ B) ∪ (B \ C) = [(A ∩ C ) ∩ (A ∪ C ) ∩ (B ∪ C )] (1.1.11)
∪ [{(A ∩ C) ∪ B} ∩ (A ∪ C ) ∩ (B ∪ C )],
after having omitted the nonessential braces. The expression has the form of the
union of two sets, each of which is placed in square brackets. It will be shown
below that the former simply equals A ∩ C , and the latter is irrelevant (i.e., it is
what in the formula (1.1.4) was marked as [· · · ] ).
6 1 Examining Sets and Relations
[(A ∩ C ) ∩ (A ∪ C )] ∩ (B ∪ C ) = (A ∩ C ) ∩ (B ∪ C ). (1.1.12)
(A ∩ C ) ∩ (B ∪ C ) = A ∩ C = A \ C . (1.1.13)
[A \ B] ∪ [B \ C] = [A \ C] ∪ [· · · ] ⊃ A \ C,
Now just compare the fourth and the last column to see that the implication:
holds. In the language of logic it means the same as (1.1.1) in the language of set
theory. We assume here that the reader is familiar with basic laws concerning logical
statements and knows that the implications 0 ⇒ 0, 0 ⇒ 1, 1 ⇒ 1 are true
and 1 ⇒ 0 is false.
1.1 Demonstrating Simple Identities 7
Problem 2
Let A, B, and C be certain sets and let the symbol denote the “symmetric
difference” of sets:
Solution
The first equation to start with can be called the associativity of the symmetrical
difference. The second one is just the distribution of the intersection over
symmetrical difference.
Identity (a)
Let us begin with the left-hand side of the equation (1.1.16) and transform it,
using (1.1.2) and (1.1.3). The associativity and commutativity for the union and
intersection will be useful too:
R ∪ (S ∪ T ) = (R ∪ S) ∪ T , R ∪ S = S ∪ R, (1.1.19)
R ∩ (S ∩ T ) = (R ∩ S) ∩ T , R ∩ S = S ∩ R, (1.1.20)
(S ∩ R) = S ∪ R , (1.1.21)
(S ∪ R) = S ∩ R . (1.1.22)
As we remember from the previous example, the symbol S means the complement
of the set S to the whole space X: S = X \ S. One also knows that the expression
S \ R can be written as S ∩ R . Hence one has
The idea of our proof will now consist of expanding the right-hand side of the
above equality and the appropriate rearranging and grouping of the resulting terms.
To start with, let us convert the first term in curly brackets by applying de Morgan’s
laws several times
The square brackets could be omitted because of the associativity of the intersection.
By expanding this expression further, this time with the use of (1.1.2), we get
As one can see above, there have appeared two empty sets as a result of the
intersection of a given set with its complement: B ∩ B = C ∩ C = ∅. It is clear that
the intersection of an empty set with any other is again an empty set, and ∅ in any
union may be disregarded. (When “adding zero” nothing is changed.) As a result,
one has
Let us now come back to the second term in curly brackets in the for-
mula (1.1.23). First we are going to expand it using the property of the distribution
of the intersection over union, and then the associativity of the sum:
Now on the right-hand side of (1.1.23) the obtained formulas are inserted:
A(BC) = (A ∩ B ∩ C ) ∪ (A ∩ C ∩ B) ∪ (A ∩ B ∩ C ) ∪ (A ∩ B ∩ C).
−−−−−−−−−− −−−−−−−−−−
(1.1.28)
We then have the expression, which is the sum of four sets. Two of them are
underlined and will be referred to below. This is not yet the final result, but a more
perceptive eye could consider the proof as practically complete. Why? Well, it is
because the right-hand side is fully symmetric with respect to the exchange of the
set names A, B, and C. Furthermore, the operation is symmetric too. Therefore,
on the left-hand side, instead of A(BC) one could equally write (BC)A.
Then, using the symmetry, one would easily justify that this expression must be
equal to (AB)C and thus equality (1.1.16) holds.
1.1 Demonstrating Simple Identities 9
A(BC) (1.1.29)
= (A ∩ B ∩ C ) ∪ (A ∩ B ∩ C ) ∪ (A ∩ C ∩ B) ∪ (A ∩ B ∩ C)
= {[(A ∩ B ) ∪ (A ∩ B)] ∩ C } ∪ {[(A ∩ B) ∪ (A ∩ B )] ∩ C}
= [(AB) ∩ C ] ∪ {[(A ∩ B) ∪ (A ∩ B )] ∩ C}.
−−−−−−−−−−−−−−
It is visible that we are already at the end if it could be justified that the underlined
expression is simply (AB) . It is actually the case because
(AB) (1.1.30)
= [(A ∩ B ) ∪ (A ∩ B)] = (A ∩ B ) ∩ (A ∩ B) = (A ∪ B) ∩ (A ∪ B )
= (A ∩ A) ∪(A ∩ B ) ∪ (B ∩ A) ∪ (B ∩ B ) = (A ∩ B ) ∪ (B ∩ A),
∅ ∅
Identity (b)
They are all clear. Certain explanation requires only adding the set A in the
locations indicated by arrows. Well, (A ∩ B) is a subset of A, so it has empty
intersection with its complement: (A ∩ B) ∩ A = ∅. Similarly, (A ∩ C) ∩ A = ∅.
This means that in the marked places the empty sets have been inserted.
10 1 Examining Sets and Relations
To complete our proof, we simply use the de Morgan’s law (1.1.21), and then
compress the expression (1.1.31):
Identity (c)
In the last case the expression on the left-hand side will be converted, first using the
definition (1.1.15):
AB = (A \ B) ∪ (B \ A) = (A ∩ B ) ∪ (B ∩ A ) , (1.1.33)
and then the distribution of the union over intersection (1.1.3) and the de Morgan’s
law (1.1.21), which entails
AB = (A ∪ B) ∩ (A ∪ A ) ∩ (B ∪ B) ∩(B ∪ A ) = (A ∪ B) ∩ (B ∪ A )
X X
= (A ∪ B) ∩ (B ∩ A) = (A ∪ B) \ (A ∩ B) , (1.1.34)
Problem 3
Solution
In Fig. 1.2 the sets appearing on the left- and right-hand sides of (1.1.35) are depicted
in gray. At least for this exemplary configuration of A, B, C, and D we see that the
1.1 Demonstrating Simple Identities 11
set on the left-hand side is actually a subset of that on the right. It would be useful if
the reader, as a part of the solution, made similar drawings for other configurations
too.
The method of proof chosen in the present case will be to expand both sides
of (1.1.35) and to check whether they are identical. Let us start with the left-hand
side, using sequentially
1. the definition of given in the previous exercise,
2. de Morgan’s second law (1.1.22),
3. the distribution of the intersection over union (1.1.2),
4. the associativity of the intersection (1.1.20) and union (1.1.19).
In order not to interrupt the course of transformations under each equality, the
property used at a given place is marked.
(A ∪ B)(C ∪ D) = [(A ∪ B) ∩ (C ∪ D) ] ∪ [(A ∪ B) ∩ (C ∪ D)]
(1)
= (A ∩ C ∩ D ) ∪ (B ∩ C ∩ D ) ∪ (A ∩ B ∩ C) ∪ (A ∩ B ∩ D)].
(4)
We will leave for now the obtained expression and in a similar way transform the
right-hand side of (1.1.35):
= (A ∩ C ) ∪ (A ∩ C) ∪ (B ∩ D ) ∪ (B ∩ D) (1.1.37)
(4)
= (A ∩ C ) ∪ (B ∩ D ) ∪ (A ∩ C) ∪ (B ∩ D),
where, at the end, the components of the sum are rearranged, thanks to its
commutativity (1.1.19).
12 1 Examining Sets and Relations
Comparing now the above expressions (1.1.36) and (1.1.37), one sees that both
are sums of four sets. What’s more, each of the terms of the union (1.1.36) is a
subset of the corresponding term of (1.1.37):
A ∩ C ∩ D ⊂ A ∩ C ,
B ∩ C ∩ D ⊂ B ∩ D , (1.1.38)
A ∩ B ∩ C ⊂ A ∩ C,
A ∩ B ∩ D ⊂ B ∩ D. (1.1.39)
Problem 1
Solution
A solution of this kind of task usually consists of two steps. First, using the
appropriate illustration and one’s imagination, we specify the preliminary thesis
(i.e., in our case the concrete form of B and C). The second step is a strict
demonstration of the formulated thesis.
To accomplish the first step, several sets At for various values of t are shown in
Fig. 1.3. The higher value of t, the darker color is used to draw At . The figure was
constructed for t = 1, 2, 4, and 50. (We start with t = 1, because of the definitions
of B and C.) The relative location of A1 , A2 , A4 , and A50 gives us some idea of
how their intersection and union may look like. First of all, it can be easily inferred
from the figure, that for 1 < t1 < t2 < t3 < . . . the following inclusion holds:
y tx2
A1 A2 A4 A50
y tx2 1
t 1
t 2
t 4
0 t 50 x
1 1
that is, all the subsequent sets are contained in the first one (A1 ). This means that
one should have:
B= At = A1 . (1.2.4)
t∈[1,∞[
This is our first conclusion, to be strictly demonstrated below. The second one
concerns the intersection. From the inclusion (1.2.3), one can deduce that
A1 ∩ At1 = At1 ,
A1 ∩ At1 ∩ At2 = At1 ∩ At2 = At2 , (1.2.5)
A1 ∩ At1 ∩ At2 ∩ At3 = At2 ∩ At3 = At3 ,
and so on. One can then infer that the set C will have the form of an open segment
]0, 1[ lying on the y axis, i.e.,
C= At = {0}×]0, 1[. (1.2.6)
t∈[1,∞[
Any point lying outside (1.2.6) will not belong to At for certain t, and if so, it cannot
be an element of the intersection (1.2.2).
We are now going to turn to the second part of our job, that is, to the strict
demonstration of both proposed theses. The first one is essentially a matter of the
proof of the inclusion (1.2.3), which can be perceived from the figure. Let us then
choose two arbitrary parameters t1 and t2 satisfying t2 > t1 ≥ 1. It will be verified
that if the point on the plane with the coordinates (x, y) belongs to At2 , it must also
belong to At1 . One has:
Similarly for t2 > t1 , the inequalities −t2 x 2 ≤ −t1 x 2 and −t2 x 2 + 1 ≤ −t1 x 2 + 1
hold. One, therefore, gets:
Both obtained inequalities mean jointly that (x, y) ∈ At1 . Consequently, the
following implication holds:
for any x and y. This is what was exactly needed: At2 ⊂ At1 . The first thesis is then
proven. For if At2 is located inside At1 , the same is true for all sets At with t ≥ 1,
and hence (1.2.4).
We move now to the second claim. The two properties of the set C simply need
to be demonstrated:
1. All elements of the set {0}×]0, 1[ belong to C.
2. Any element not belonging to {0}×]0, 1[ does not belong to C either.
Confirming both of these properties will imply the equality of sets {0}×]0, 1[ and C.
Let us start with the first. If a point (x, y) belongs to the set {0}×]0, 1[, this signifies
that these coordinates are de facto of the form (0, y), where 0 < y < 1. Let us then
insert these coordinates into the inequalities defining the set At . One then gets:
But this is, after all, precisely our assumption. The above two inequalities are,
therefore, true, irrespective of the value of t. As a consequence, a point (0, y), where
0 < y < 1, belongs to all At and hence also to their intersection. And if so, it must
also belong to C, which was to be proved.
Now let us turn to the second property. One has to choose any point lying outside
the set {0}×]0, 1[. Its coordinates x and y are fixed. The following situations are
possible (not necessarily mutually exclusive): x = 0 or y ≥ 1 or y ≤ 0. They will
be examined in turn.
• For x = 0, the inequality y > tx 2 can be rewritten as t < y/x 2 . Can it be satisfied
for any t ≥ 1? Of course not because the right-hand side is fixed (it is a concrete
number) and t on the left-hand side can be given an arbitrarily large value. This
means that there exists such t ≥ 1, that (x, y) ∈ At . As a consequence (x, y)
cannot belong to all sets, and hence, nor to their intersection C.
1.2 Finding Sets on a Plane 15
• For y ≥ 1, one is not able to satisfy inequalities y < −tx 2 +1 because the number
on the right is at most equal to 1 (remember that t is positive), and the inequality
is strict. Such a point inevitably lies beyond At for t ≥ 1, and, therefore, is not
in C.
• For y ≤ 0, our reasoning is carried out in a similar way. This time the inequality
y > tx 2 cannot be satisfied, as the number on the right-hand side is nonnegative.
The conclusion is: no point lying outside the set {0}×]0, 1[ belongs to C, and
any point belonging to it belongs also to C. In this way, the equality (1.2.6) has been
shown.
Problem 2
Solution
As in the previous example, one is dealing with sets on the plane, which can be
drawn relatively easily. For each fixed value of t, the set At is a rectangle (the edges
and interior) with vertices located at the points with coordinates:
According to (1.2.13), one is interested in values of t lying in the interval [0, 1]; in
Fig. 1.4 several such rectangles are depicted, starting from t = 0 (darkest rectangle),
up to t = 1 (brightest rectangle).
As shown in the figure, the sum of all sets (i.e., the set B) should be a hexagon
of vertices
(−1, 0), (1, 0), (2, 1), (2, 4), (−2, 4), (−2, 1). (1.2.15)
On the other hand, the common part (set C) seems to be the rectangle with corners
y
4
x
2 1 0 1 2
Below, it will be proved that it is a correct assumption. Let us start with the set B.
The point on the plane belongs to B, in so far as it lies in one of rectangles At for at
least one t ∈ [0, 1]. This statement may be given the form of the following sentence:
This means that for a given (x, y) ∈ B there must exist a parameter t satisfying the
six inequalities:
t ≥ 0, t ≤ 1, t ≥ −x − 1, (1.2.18)
t ≥ x − 1, t ≤ y, t ≥ y/2 − 1.
At this point, one should realize that the particular value of t is unimportant. It is
only necessary for t to exist! If so, one has (x, y) ∈ B. If it does not exist, the point
(x, y) ∈ B. When is one then able to find any t satisfying (1.2.19)? The answer is
quite clear: one is, if the following inequality holds:
If, however, the largest of the numbers on the left-hand side is to be less than
(or equal to) the smallest number on the right-hand side, this is equivalent to the
statement that any of the numbers on the left-hand side must be less than (or equal
1.2 Finding Sets on a Plane 17
to) any number on the right-hand side. So we come to the following inequalities,
which must be satisfied simultaneously:
0 ≤ 1, x − 1 ≤ 1, x − 1 ≤ 1, y/(2 − 1) ≤ 1,
0 ≤ y, x − 1 ≤ y, x − 1 ≤ y, (2 − 1) ≤ y/y. (1.2.21)
The obtained inequalities no longer contain parameter t and define the set B.
There are many, but some of them may be omitted because they are either met
automatically, or result from others. In this way one gets
B = {(x, y) ∈ R2 | − 2 ≤ x ≤ 2 ∧ 0 ≤ y ≤ 4 ∧ y ≥ −x − 1 ∧ y ≥ x − 1}.
(1.2.22)
After having found the points of intersection for the pairs of lines
x = −2, x = 2, y = 0, y = 4, y = −x − 1, y = x − 1, (1.2.23)
it is easy to convince oneself that the set in request is really the hexagon (1.2.15),
depicted in the figure.
We are going to solve now the second part of our problem and find the set C. If
a logical sentence similar to (1.2.17) was to be written, there would be one change:
now a point (x, y) belongs to C, as long as it lies not in one, but in all At ’s. Thus,
one has
The four inequalities are now obtained, which must be met for each t ∈ [0, 1]:
t ≥ −x − 1, t ≥ x − 1, t ≤ y, t ≥ y/2 − 1. (1.2.25)
The most demanding conditions arise when setting in the inequalities of the type
t ≥ . . ., the value of t equal to 0, and in the inequalities of the type t ≤ . . ., the value
of t equal to 1. For these values of t, the system (1.2.25) has to be met because they
belong to the interval [0, 1]. And then it will also be met automatically for all other
parameters t in this interval. This reasoning leads us to the set C in the form of
As one can see, this set is actually a rectangle with corners (1.2.16).
18 1 Examining Sets and Relations
Problem 1
The least upper and the greatest lower bounds (supposing they exist) of the set:
3|y| − 1
X := x ∈ R | x = ∧ y∈R (1.3.1)
5|y| + 1
will be found. It will also be checked, if they belong to X.
Solution
Consider first the least upper bound. In order that a numerical set might have
supremum or infimum, it certainly must be bounded above or below, respectively.
This raises the obvious question of whether our set is bounded above. To check this,
let us transform the expression for an element x as follows:
The latter inequality is of course “strict,” which will later prove to be important.
From this estimate it follows that the set X must be bounded above, since no matter
what value is put under y, the expression will always be less than 3/5.
The question arises as to whether this number is also a supremum or at most one
of the many (i.e., infinitely many) upper bounds. One knows that a supremum is the
smallest upper bound, if it exists. In the space of real numbers it is known that it
certainly exists, but not necessarily in the space of rational numbers. As an example
√can consider the set {q ∈ Q | q < 2}, the extremal bounds
of such a situation, one 2
of which (equal to ± 2) do not belong to the space Q (which means, thatde facto
neither the supremum nor infimum exists).
Returning to our task, it must be said that the key is now to examine whether the
upper bound of the set X can be any number smaller than 3/5, i.e., 3/5−, for some
small positive . If such a number is not available, then the supremum shall be equal
to 3/5. But if it exists, one must inevitably have:
3|y| − 1 3
∀y∈R x = ≤ − . (1.3.3)
5|y| + 1 5
By performing transformation analogous to (1.3.2), this condition can be given the
form:
8
∀y∈R ≥ . (1.3.4)
5(5|y| + 1)
1.3 Finding Lower and Upper Bounds of Numerical Sets 19
It is clear that it is impossible to meet this requirement. The right-hand side (i.e.,
) is fixed, and the left-hand side can be made arbitrarily small by selecting large
y so that the inequality cannot be true for every y ∈ R. Therefore, there is no
satisfying (1.3.3). The conclusion is: the least upper bound of X does exist and
equals 3/5. It is worth determining yet whether this supremum belongs to X or
remains outside it. The answer to this question is provided by (1.3.2). It has already
been noted above that this inequality is strict, which means that ∀y∈R x = 3/5.
Therefore, this number does not belong to the set X.
The search for the greatest lower bound must begin by determining whether the
set is bounded below. This time the expression for x is transformed as follows:
As one can see, the set is actually bounded below by the number −1. Of course there
are infinitely many lower bounds for the set. To determine the infimum of X, one
has to find the largest of them (naturally, if it exists). This task is very simple due to
the fact that inequality (1.3.5) is not strict. By selecting y = 0, one finds x = −1. So
we have finally two conclusions: the greatest lower bound equals −1 and it belongs
to X. Summarizing, it has been found
3
sup X = ∈ X, inf X = −1 ∈ X. (1.3.6)
5
A certain puzzle for the reader may constitute the question, how we knew from
the beginning that the expressions (1.3.2) and (1.3.5) should have been transformed
in a way to extract 3/5 and −1. Well, this comes from the formula for x and from our
intuition: one can see without any calculations that because of −1 in the numerator
and +1 in the denominator the fraction is less than 3/5 and approaches this value for
very large y. So if the value 3/5 is extracted, it remains only to determine whether
the second term is positive or negative. The identification of the sign is generally
much simpler than finding a specific value. The same applies to the infimum, where
the substitution y = 0 is conspicuous, since |y| takes then the smallest value.
Problem 2
The least upper and the greatest lower bound of the set:
Solution
a 2 + b2 a b
a 2 − 2ab + b2 ≥ 0 ⇐⇒ a 2 + b2 ≥ 2ab ⇐⇒ = + ≥ 2.
ab b a
(1.3.9)
On the other hand, when looking at the definition of the set Y , one sees that the
number y can be given the form:
1 1 a b
y = (a + b) + = + + 2. (1.3.10)
a b b a
Regarding the supremum of the set Y , it is easy to show that it cannot exist
because the set is not bounded above. It happens very often when investigating
expressions dependent on several variables (in our case on a and b) that it is
convenient to fix all variables except one and examine the behavior of the expression
in this single variable only. So let b = 1, which corresponds to the examination of
certain subset of Y . If one is able to prove that this subset is unbounded, surely the
same will refer to Y . One then has
a b 1
y|b=1 = + + 2 =a+2+ . (1.3.13)
b a b=1 a
Problem 1
is an equivalence relation. If so, the equivalence classes will be found and the graph
of the relation will be performed.
Solution
As the reader knows well from lectures of analysis and from the theoretical
introduction at the beginning of this chapter, a relation R in set X is simply a subset
of the Cartesian product X × X:
R ⊂ X × X.
If one chooses two elements of that set, e.g., a and b, they can be R-related, if
(a, b) ∈ R. One uses also the alternative notation aRb. Naturally, one needs to
remember that if a couple (a, b) belongs to such a subset, it does not mean that
the couple (b, a) does. The relation for which this condition would be true is called
symmetric. Therefore, aRb in general is not the same as bRa.
In mathematics, but also in physics, a particularly important role is played by
the so-called equivalence relations. Their definition is going to be recalled below. A
relation R is called the “equivalence relation” if, and only if, it satisfies the following
three conditions:
1. For each x ∈ X one has (x, x) ∈ R. The relation with this property is called
reflexive.
2. For every x, y ∈ X the implication
holds. As discussed above, in this case one is talking about the symmetric
relation.
3. For every x, y, z ∈ X one has
Now we are going to examine whether the relation (1.4.1) satisfies these
conditions.
• Reflexivity. One needs to check whether or not (x, x) ∈ R, i.e., whether
x − 4x
∈ Z. (1.4.4)
3
This condition is obviously fulfilled, because
x − 4x −3x
= = −x, (1.4.5)
3 3
and by supposition x is an integer. The relation is, therefore, reflexive.
• Symmetry. Now we have to make sure that if n := (x − 4y)/3 is an integer, the
same can be said about m := (y − 4x)/3. To this goal let us calculate the sum of
these two numbers:
x − 4y y − 4x x − 4y + y − 4x
n+m = + =
3 3 3
3x + 3y
=− = −(x + y) ∈ Z. (1.4.6)
3 x,y∈Z
x − 4y y − 4z x − 4z 3y
n+m−l = + − =− = −y ∈ Z. (1.4.7)
3 3 3 3
This immediately leads to the conclusion that l ∈ Z, i.e., (x, z) ∈ R. The relation
is then transitive.
In conclusion, one finds that the relation defined in the text of the present problem
is the equivalence relation. As the reader certainly knows from the lecture, the
equivalence classes form the partition on Z. These classes will be found below.
If two integers x and y are in relation with each other, then naturally k := (y −
4x)/3 is an integer. This equation can be rewritten as
y = 4x + 3k. (1.4.8)
If one now substitutes for x a certain fixed value belonging to Z and start to choose
different integers k, one gets elements y that are equivalent to x. In this way, taking
k = 0, ±1, ±2, . . ., the class of the element x = 0 is obtained:
Similarly:
and
We are still left to perform a graph of the relation. It will be a collection of all
points on a plane whose (integer) coordinates are R-related to each other. These
points are shown in Fig. 1.5 as black dots. As we know already from (1.4.8), they
lie on straight lines y = 4x + 3k. These auxiliary lines are marked as dashed on
the graph. (Remember that not the whole lines, but only discrete points, form the
graph of the relation.) Naturally, thanks to the symmetry of the relation, the points
must be simultaneously situated on the lines of the type (1.4.8), where x and y are
3
y x
2
x
3 2 1 1 2 3
interchanged. That is really the case. These additional lines are drawn as dotted. The
gray line corresponds to y = x. Because of the reflexivity all the points of integer
coordinates on this line must be marked as belonging to R (each x is R-related to
y = x). In turn the symmetry property manifests itself on the graph as the invariance
under reflection with respect to this gray line.
Problem 2
is the equivalence relation. If so, the equivalence classes will be found and the graph
of the relation will be performed.
Solution
φ(x) = x 4 − 4x 2 , (1.4.14)
Below, it will be proved that this is in fact an equivalence relation. The appropriate
definition is already known from the previous problem, so one can proceed now
with checking the conditions.
• Reflexivity. Naturally each element x ∈ R is R-related to itself, because the
condition φ(x) = φ(x) is obviously satisfied.
• Symmetry. From φ(x) = φ(y) it follows that φ(y) = φ(x), so the relation is
symmetrical.
• Transitivity. The implication
y x
x
-3 2 2 3
y x
-3
form of the function φ(x) has been used and our conclusions are then valid for any
relation defined by a function (Fig. 1.6).
Suppose that one wants to determine which elements are R-related to the chosen
element x. To do this, one has to consider the equality (1.4.15), where x is fixed,
and y is unknown. It is, therefore, necessary to solve the equation for y:
y 4 − 4y 2 = x 4 − 4x 2 . (1.4.17)
y 2 − x 2 = 0, (1.4.19)
the two straight lines on the plane are: y = x and y = −x. The second factor
y2 + x2 − 4 = 0 (1.4.20)
is simply the equation of the circle with the center at the origin and with a radius
equal to 2. On this basis, the class for the item x can be immediately provided. It
will consist of two to four elements.
26 1 Examining Sets and Relations
• If |x| > 2, there are no solutions of (1.4.20) because the left-hand side is always
positive, so only solutions obtained from (1.4.19) belong to the class:
• For x = 2, the equation (1.4.20) has exactly one solution y = 0, and additionally
one has y = ±2 from (1.4.19), so
which might be denoted as [0]R . The same result is naturally obtained for x =
−2.
• If 0 < |x| < 2, then we have a total of four solutions, as such is the number of
class elements:
[x]R = {x, −x, 4 − x 2 , − 4 − x 2 }, (1.4.23)
√
except for the specific case x = ± 2, where solutions (1.4.19) and (1.4.20)
overlap:
√ √ √
2 = { 2, − 2}. (1.4.24)
R
Then a set of equivalence classes has been found and what remains is to create a
graph. It is very simple because one only has to mark on the plane all pairs (x, y),
for which the coordinates satisfy the equation (1.4.18). This is, however, equivalent
to the logical disjunction of (1.4.19) and (1.4.20). The graph will then consist of
two straight lines and the circle, mentioned above. Again one recognizes here the
elements on which the attention was drawn in the previous exercise: the graph
contains the line y = x (reflexive relation), and is symmetric with respect to this
line (symmetric relation).
The classes found earlier can be very easily read off from the drawing. All one
has to do is to choose some x0 on the horizontal axis and draw from it an auxiliary
(vertical) line: x = x0 . This line, depending on the value of x0 , crosses the graph
in two, three, or four points. The coordinates y of these points are just elements
that form the class [x0 ]R . Naturally among them we will find also x0 itself from the
intersection of the lines x = x0 and y = x.
Problem 3
Solution
2n2
= 2n ∈ N. (1.4.26)
n
Reflexivity then is not violated.
• Symmetry. Verifying the second property does not present any problem, since
a condition in (1.4.25) has a symmetric form, which is not modified when
interchanging n and m. If this condition is fulfilled for couples (n, m), it must also
be so for couples (m, n). It is simply a well-known property of the commutativity
of the logical disjunction in (1.4.25).
• Transitivity. That the considered relation is not transitive, and thus is not an
equivalence relation, one can ascertain taking the concrete exemplary numbers
n and m. There is no doubt that (2, 3) ∈ R, because the sentence
2 · 22 8 2 · 32
= ∈N ∨ =9∈N (1.4.27)
3 3 2
is true (in spite of the fact that the first component of the logical disjunction is
false). Likewise, it is true that
2 · 22 8 2 · 52
= ∈N ∨ = 25 ∈ N, (1.4.28)
5 5 2
which means that (2, 5) ∈ R. If the relation was to exhibit the transitivity
property, there would also have to be (3, 5) ∈ R. However, the sentence:
2 · 32 18 2 · 52 50
= ∈N ∨ = ∈N (1.4.29)
5 5 3 3
is false in a clear way.
Concluding, R is not an equivalence relation.
28 1 Examining Sets and Relations
where
(a) At = [t, t + 1] × [−t, −t + 1].
(b) At = [−2t − 2, t] × [t − 1, 2t].
(c) At = {(x, y) ∈ R2 | (tx − 1)2 + t 2 y 2 < 1}.
Answers
(a) B is a hexagon with vertices (0, 0), (1, −1), (2, −1), (2, 0), (1, 1),
(0, 1); C = {(0, 0)}.
(b) B is a hexagon with vertices (1, 0), (1, 2), (−4, 2), (−4, 0), (−2, −1),
(0, −1); C = [−2, 0] × {0}.
(c) B = R+ × R; C = K((1, 0), 1).
Exercise 3 Check if the following subsets of R are bounded, and if so, find their
suprema and infima:
(a) A = {(x + 1)/(|x| + 2) | x ∈ R}.
(b) B = {2/n − 3/m | n, m ∈ N}.
(c) C = {x ∈ R, | | |x − 1| − |x − 2| | < 2}.
Answers
(a) Bounded below and above, sup X = 1 ∈ X, inf X = −1 ∈ X.
(b) Bounded below and above, sup Y = 2 ∈ Y , inf Y = −3 ∈ Y .
(c) Unbounded.
Exercise 4 Check if the given relations are equivalence relations. If so, find the
equivalence classes and—for the first two cases—draw graphs of the relations.
1.5 Exercises for Independent Work 29
Answers
(a) Equivalence relation. Classes: e.g., [0]R = {0, −2π, 2π }, [π ]R =
{π, −π, −3π, 3π }.
(b) Equivalence relation. Classes: [x]R = {x, 1 − x} for x = 1/2 and
[1/2]R = {1/2}.
(c) Equivalence relation. Each set C ⊂ X \ X0 defines certain equivalence
class.
Chapter 2
Investigating Basic Properties of
Functions
f : X ∈ x −→ y ∈ Y. (2.0.1)
The subset of Y containing all and only values that correspond to at least one
argument is denoted by f (X) and is called the image of the function f .
The level sets are subsets of the domain for which the value of the function is
constant:
If all level sets (i.e., for any h ∈ Y ) are one-element, then the function is called
a bijection. If some of them are empty, the function is an injection. The name
surjection is reserved for the case where there are no empty level sets regardless of
the number of contained elements. From these definitions it results that an injective
surjection is a bijection.
The bijective function has its inverse. The inverse function is defined on the set
f (X) (in the case of a bijection one has f (X) = Y ) with the values in X as
provided f (x) = y.
The image or range of a set A denoted as f (A) is defined in the following way:
Using this notation the level set (2.0.2) can be defined as Dh = f −1 ({h}). In this
last definition and in (2.0.5) the symbol f −1 does not refer to any inverse function.
The so-called Darboux’s theorem (or property) for the continuous function (see
Chap. 8) f : R → R states that the image of an interval is an interval. This property
is generalized onto other spaces and then it asserts that the image of a connected set
(see Chap. 6) is connected.
For X, Y ⊂ R, one can talk about the monotonic functions. In particular by the
term increasing function a real function satisfying
If a function fulfills only the conditions f (x1 ) ≤ f (x2 ) or f (x1 ) ≥ f (x2 ), one has
a nondecreasing or nonincreasing function.
Problem 1
5 2
f (x, y) = (x + y 2 ) + 3xy. (2.1.1)
2
The image of this mapping will be examined and exemplary level sets will be drawn.
Solution
The function f defined on the set D = R2 takes its values in R, but this does not
mean that this is the range of the function. One has to accurately determine f (D).
Let us first try to simplify the expression by putting y = x. This means that instead
2.1 Looking for Ranges (Images) and Level Sets 33
5 2
f (x, x) = (x + x 2 ) + 3x 2 = 8x 2 ≥ 0. (2.1.2)
2
This expression is always nonnegative and takes all values from the interval [0, ∞[.
Then the question arises, whether this function can also take negative values (if
x = y ), or whether one simply has f (D) = [0, ∞[. The answer shall be found, if
one transforms the formula for f (x, y) as follows:
5 2 3
f (x, y) = (x + y 2 ) + 3xy = (x 2 + y 2 ) + 3xy + x 2 + y 2 (2.1.3)
2 2
3 3
= (x 2 + y 2 + 2xy) + x 2 + y 2 = (x + y)2 + x 2 + y 2 ≥ 0.
2 2
The obtained expression is certainly nonnegative, as a sum of three squares. Thus,
it appears that indeed f (D) = [0, ∞[.
In the second part of the solution we are going to find and draw some level sets
of the function f . When doing it, the image will be created again in an alternative
way. As we know, level sets of a function are such subsets of D (let them be marked
with Dh ), for which the function takes fixed values (equal to h):
Such subsets are, e.g., isobars or isotherms on a map or points with the same altitude
above sea level. In a similar way the level sets can be defined for D = Rn (In this
case a level value will constitute n values.) or for other sets, which are nonessential
for the present problem. The definition equivalent to (2.1.4) is Dh := f −1 ({h}),
with the symbol on the right-hand side being simply the inverse image of a (one-
element) set, not requiring that the function be reversible. This is simply a collection
of all arguments, for which the function values lie in the set {h}, which means that
they are simply equal to h (Fig. 2.1).
So one has to examine and draw the equation:
5 2
(x + y 2 ) + 3xy = h. (2.1.5)
2
Below, a method that is worth remembering will be used. Well, we know how to
relatively easily draw graphs for quadratic equations which do not contain “mixed”
terms xy. There are only four well-known curves of that kind: a circle, an ellipse, a
parabola, and a hyperbola, and their standard expressions are well known. Hence, if
in the formula product terms are found as well, one first has to get rid of them. To
do this, the axes of the coordinate system must be tilted. Because the coefficients
accompanying x 2 and y 2 in (2.1.5) are identical, this will simply reduce to the
34 2 Investigating Basic Properties of Functions
y
u
x
h 0
h 1
h 2
h 3
Fig. 2.1 Level sets of the function (2.1.1). Auxiliary axes u and v are drawn in gray
rotation of both axes by the angle ±π/4. Let us then define in place of x and y
two new variables:
1 1
u= (x + y), v= (x − y), (2.1.6)
2 2
with which will be connected the new (secondary) lines of the coordinate system.
It is worth explaining at this point how one would proceed if the coefficients in
x 2 and y 2 were different. In such a case one could introduce u and v as:
u = αx + βy, v = α x + β y, (2.1.7)
and so choose the values of constants α, β, α , and β to “kill” the “mixed” term
(now in the variables u, v). In order to simplify the calculations, one can first put
α = α = 1, and seek only β and β . In general in this situation, the axes of a new
coordinate system, which are defined by the equations v = 0 and u = 0, can be
“bent,” i.e., it may happen that each axis is rotated by a different angle (in relation
to the old axes x and y).
Using definitions (2.1.6), out of which x and y may be obtained, the equation for
the level sets (2.1.5) can be rewritten in the form:
5
((u + v)2 + (u − v)2 ) + 3(u + v)(u − v) (2.1.8)
2
5
= (u2 + 2uv + v 2 + u2 − 2uv + v 2 ) + 3u2 − 3v 2 = 8u2 + 2v 2 = h.
2
2.1 Looking for Ranges (Images) and Level Sets 35
Problem 2
x2 − y2 + 1
f (x, y) = . (2.1.10)
x2 + y2 + 1
Level sets of this function will be examined and drawn and its range will be found.
y
36 2 Investigating Basic Properties of Functions
Solution
In this exercise, unlike the previous one, we begin with determining level sets. As
usual the domain is denoted with D and the level set corresponding to the value h
with the symbol Dh . The equation for the level set has the form:
x2 − y2 + 1
= h. (2.1.11)
x2 + y2 + 1
After having rearranged the terms one obtains a form, which can now be examined:
h+1 2
x2 + y = −1. (2.1.14)
h−1
h 0.2
h 0.2
h 0.7
h 1 x
h 0.7
h 0.2
h 0.2
One could come to this result independently, using appropriate upper and lower
estimates of the expression (x 2 − y 2 + 1)/(x 2 + y 2 + 1)
2
x − y2 + 1 x2 + y2 + 1 x2 − y2 + 1
≤ = 1 ⇐⇒ −1 ≤ ≤ 1, (2.1.16)
x2 + y2 + 1 x2 + y2 + 1 x2 + y2 + 1
which implies that a set of values is bounded (by ±1). Since one knows that for
x = y = 0,
x2 − y2 + 1 1
= = 1, (2.1.17)
x +y +1
2 2 1
x2 − y2 + 1 −x 2 − y 2 − 1 + 2x 2 + 2 2x 2 + 2
= = −1 + . (2.1.18)
x2 + y2 + 1 x2 + y2 + 1 x2 + y2 + 1
domain, i.e., the entire plane R2 is connected—see Chap. 6) is also a connected set
(i.e., a set that cannot be broken down into two separate and simultaneously open
subsets). As the reader surely knows from the lecture, the only connected sets in R
are intervals. Since the extremal bounds are already known, the only possibility is
f (D) =] − 1, 1].
Problem 1
Solution
In the previous section, level sets of various functions were examined. They helped
us to determine the images, but as we will see below, they also allow to determine
whether the function is single-valued (i.e., whether it is an injection). In accordance
with the appropriate definition a single-valued function for different arguments
always assumes different values. The following implication must then hold:
where symbol D, as usual, stands for the domain of f . Referring to the “language”
of levels, it means that each level set may be at most a singleton. The presence of
two or more elements in a given level set would violate the implication (2.2.1), since
the same value would be assigned to all these elements.
For this reason, we are going to begin our discussion of the present problem by
examining the level sets Dh , assuming first that D = R. To this end, one has to
consider the equation:
1 x
cosh x = (e + e−x ) = h. (2.2.2)
2
By multiplying both sides by the factor 2ex , which never equals zero, a quadratic
equation is obtained, in which the role of the unknown variable is played by t = ex .
t 2 − 2ht + 1 = 0. (2.2.4)
When calculating the discriminant of this equation, one obtains = 4(h2 − 1).
The first conclusion can be drawn on this basis immediately: for
−1 < h < 1 one has < 0 and the equation (2.2.4) has no solutions. Any level set
for these h is then empty. If one looks at cosh x as a function of values in codomain
Y = R, and not only in [1, ∞[, one can already say that this function is not “onto”
since the range does not exhaust the entire codomain Y (i.e., the function is not a
surjection).
For other values of h, i.e., for |h| ≥ 1, the equation (2.2.4) has always certain
solutions, but remember that under the letter t, ex is hidden. For real values of x,
this expression must be positive. Therefore, it is still possible that (2.2.4) does have
solutions, but (2.2.3) does not. We need then to look at the solutions thoroughly.
• h = 1. The only solution is here t = 1, which corresponds to ex = 1, i.e., x = 0.
The level set is, therefore, a singleton.
• h = −1. Here one has t = −1 or equivalently ex = −1. This equation has no
real solutions. The level set is empty.
• h > 1. This time there are two solutions:
t1 = h + h2 − 1, t2 = h − h2 − 1. (2.2.5)
are singletons (the second solution, i.e., x2 , does not belong to the domain). Both of
these properties together mean that the function is a bijection and is reversible. The
formula for the inverse function f −1 : [1, ∞[→ [0, ∞[ results from (2.2.6) and has
the form
x = f −1 (y) = ln(y + y 2 − 1). (2.2.8)
Solving this problem, one could be convinced of the importance, when defining a
function f : X → Y , of specifying both the rule of assignment f and sets X and Y .
Problem 2
(−1)n (2n − 1) + 1
f (n) = . (2.2.9)
4
Its injectivity and surjectivity will be examined. If the function turns out to be a
bijection, the formula for its inverse will be found.
Solution
To gain some idea about the function under consideration let us first calculate its
values for a few initial natural numbers:
(−1)1 (2 · 1 − 1) + 1 (−1)2 (2 · 2 − 1) + 1
f (1) = = 0 , f (2) = = 1,
4 4
(−1)3 (2 · 3 − 1) + 1 (−1)4 (2 · 4 − 1) + 1
f (3) = = −1 , f (4) = = 2,
4 4
(−1)5 (2 · 5 − 1) + 1 (−1)6 (2 · 6 − 1) + 1
f (5) = = −2 , f (6) = = 3.
4 4
(2.2.10)
On the basis of these few terms we suspect that the mapping transforms even
integers onto naturals and the odd integers onto nonpositive integers. Symbolically
one could write
On the basis of the above observations, one can clearly see that it is convenient
to divide the whole domain, the entire set of natural numbers, onto two subsets: odd
and even natural numbers, and to examine separately the behavior of the function
for each of them. So let us first assume that n = 2k, where k = 1, 2, 3, . . .. One has
then
(−1)2k (2 · 2k − 1) + 1 2 · 2k − 1 + 1 4k
f (2k) = = = = k. (2.2.12)
4 4 4
For each even natural number, the function reduces then to the division by two.
Hence, the image of the set, denoted with 2N, is simply N. All level sets Dh for
h ∈ N are singletons: Dh = {2h}. For h ≤ 0 the level sets are empty. This means
that the function f when reduced to the above-mentioned subset is single valued.
Now let us turn to the subset of odd natural numbers, i.e., let us put n = 2k − 1,
for k = 1, 2, 3, . . .. Then,
f −1 : Z → N. (2.2.15)
(−1)n (2n − 1) + 1
m= , (2.2.16)
4
where m ∈ Z and n ∈ N. Now it should be inverted in order to determine the relation
n(m). The equation above may be given the form
4m − 1 = 2n − 1 ⇐⇒ n = 2m. (2.2.18)
This result is consistent with the assumption that n is even. Now let us assume that
m ≤ 0. The left-hand side of the equation (2.2.17) is negative, so the same refers to
the right-hand side. This is, however, possible only on the assumption that the factor
(−1)n = −1 < 0, and thus, that n is odd. Then,
which is in fact an odd number. (It should be remembered that one has m =
0, −1, −2, . . ..)
In conclusion, the inverse function f can be defined as follows:
2m for m ∈ N,
f −1 (m) = (2.2.20)
1 − 2m for m ∈ Z− ∪ {0}.
|4m − 1| + 1
f −1 (m) = . (2.2.21)
2
Problem 3
Its injectivity and surjectivity will be examined. If the function turns out to be a
bijection, the formula for its inverse will be found.
Solution
Let us assume for definiteness that m1 ≥ m2 . If this condition was not met, then,
in what follows, the pairs of numbers (m1 , n1 ) and (m2 , n2 ) would only swap their
roles, which is irrelevant for further conclusions.
After having transformed (2.2.24) one can write
m1 = m2 , n1 = n2 , (2.2.27)
Now, as we know, the number h can be written in the form h = 2k (2l − 1), where k
and l are given. In place of (2.2.28), the equation analogous to (2.2.24) is obtained:
f −1 : N → N2 . (2.2.30)
44 2 Investigating Basic Properties of Functions
The exact recipe for this inverse function is based on the procedure that was used
to find level sets. If one wished to find the value of the inverse function for some
p ∈ N, one needs to perform the two following steps:
1. to write the number p in an unequivocal form p = 2k (2l − 1),
2. to read from this form the numbers k and l and write f −1 (p) = (k + 1, l).
At the end, it is worth noticing that the fact that there exists a bijection between
sets N2 and N means that they are equinumerous. Couples (n, m) can, therefore, be
numbered with consecutive naturals. In such a case it may be said that the set is
enumerable.
Problem 1
f (x, y) = x 2 − 4y 2 , (2.3.1)
the image of the set A := {(x, y) ∈ R | x 2 + y 2 ≤ 1} and the inverse image of the
interval [1, 2], i.e., f −1 ([1, 2]) will be found.
Solution
f (x, y) = x 2 − 4y 2 = 5x 2 − 4x 2 − 4y 2 (2.3.2)
= 5x 2 − 4(x 2 + y 2 ) ≥ 5x 2 − 4 ≥ −4,
x 2 +y 2 ≤1
f (x, y) = x 2 − 4y 2 = x 2 + y 2 − 5y 2 ≤ 1 − 5y 2 ≤ 1.
x 2 +y 2 ≤1
It is already known, therefore, that the set f (A) is bounded, but it still must be
checked whether or not the obtained numbers −4 and 1 are the “best” restrictions,
i.e., they are extremal bounds. With a bit of luck, one can easily resolve this dilemma
by considering if for some specific choices of (x, y) ∈ A, the (not strict) inequalities
in (2.3.2) turn into true equalities. If yes, no better estimates can be found.
2.3 Finding Images and Inverse Images of Sets 45
When looking at the formula of the function f (x, y) = x 2 −4y 2 , one can see that
f (x, y) actually does assume the value −4, for example at the point of coordinates
x = 0, y = 1. This point undoubtedly belongs to A, because 02 + 12 = 1 ≤ 1, so
Now we will verify whether the upper bound can be “improved.” It turns out
that again it is not possible. It can be easily seen that by taking x = 1 and y =
0, the second of the inequalities (2.3.2) may also be transformed into the equality
f (1, 0) = 1, and the chosen point belongs to A. So we come to the conclusion
A set A ⊂ R2 is a circle, and, therefore, a connected set. Its image must also be a
connected set in R, i.e., an interval. The only interval with the extremal bounds −4
and 1 (which belong to it) is [−4, 1]. Thus one has:
In the second part of our job, we are going to find the inverse image of the interval
[1, 2]. It is defined as follows:
The symbol f −1 does not apply in this case to the concept of the inverse function,
so this function does not need to be reversible. One has
The inverse image of the interval [1, 2] is, therefore, the closed area contained
between curves of the two hyperbolas:
y2 x2 y2
x2 − = 1, and − = 1, (2.3.9)
1/4 2 1/2
x2 y2
1
2 1 2
x
Problem 2
Solution
The values of the function f lie on the plane R2 . The corresponding coordinates
will be denoted with (ξ, η). Thus one has
1 x
ξ= , η= . (2.3.11)
1 + x2 1 + x2
It can easily be seen that if one calculates the value of the expression ξ 2 + η2 , one
will get
2
2
1 x 1 + x2 1
ξ +η =
2 2
+ = = =ξ . (2.3.12)
1 + x2 1 + x2 (1 + x 2 )2 1 + x2
2.3 Finding Images and Inverse Images of Sets 47
After having moved all terms to the left-hand side, one can see that the equation of
a circle has been obtained:
1 1
ξ 2 + η2 = ξ ⇐⇒ (ξ − )2 + η2 = . (2.3.13)
2 4
Of course at this stage, one cannot say yet that the image of the function f , and
even less so the set f ([1, ∞[), is a circle on the plane R2 , but, at most, that they
are contained in it. In the image there are actually no points lying beyond the
circle (2.3.13), but not necessarily all of the points of the circle belong to f (D).
(D, as usual, denotes the domain of the function.) In order to precisely determine
the range of the function as well as the image of the set [1, ∞[, a new variable
t ∈] − π, π ] will be introduced and the circle (2.3.13) will be parametrized in the
following way:
1 1 1
ξ(t) = + cos t , η(t) = sin t, (2.3.14)
2 2 2
where in the equation for ξ we have added 1/2 to account for the position of the
center. For such a parameterization, the equation (2.3.13) is fulfilled automatically,
as a result of the Pythagorean trigonometric identity. The parameter t acts as an
angle, as is shown in Fig. 2.5. This angle uniquely identifies points on the circle, so
one can use it to define level sets. By Dt we will then understand D(ξ(t),η(t)) :
1 1 x 1
Dt = x ∈ R | = (1 + cos t) ∧ = sin t . (2.3.15)
1 + x2 2 1 + x2 2
t x
1
2 1 1
x 2
h2
2 4
1
2
from which it follows that t/2 ≥ π/4, i.e., t ≥ π/2. The set f ([1, ∞[) is then
a quarter of the circle, corresponding to the parameter (angle) t ∈ [π/2, π [. In
Fig. 2.5, it has been drawn in bold line.
Once having the knowledge of the range and all level sets, it is easy to answer the
question what the set f −1 ({(a, b)}) is. Surely, it is nothing other than D(a,b) . So, if
the point (a, b) does not lie in the range, then D(a,b) = ∅, and if it does, it follows
from the equations:
1 x
a= , b= . (2.3.18)
1 + x2 1 + x2
b
After dividing the second by the first, D(a,b) = .
a
Exercise 1 Find and sketch level sets and determine ranges of the following
functions:
(a) f : R+ 2 → R, where f (x, y) = x/y + y/x.
(b) f : R2 → R, where f (x, y) = (x − y)(x + y).
(c) f : R2 \ {(0, 0)} → R), where f (x, y) = (x + y)/(x 2 + y 2 ).
2.4 Exercises for Independent Work 49
Answers
(a) Level sets: the union of two half-lines (for h > 2), one half-line (for
h = 2), or an empty set (for h < 2). Range: [2, ∞[.
(b) Level sets: hyperbolas x 2 − y 2 = h, except for h = 0, for which it is the
union of two lines y = ±x. Range: R.
(c) Level sets: circles (x − 1/(2h))2 + (y − 1/(2h))2 = 1/(2h2 ) without the
point (0, 0), except for h = 0, for which it is the line y = −x (with one
point excluded). Range: R.
Exercise 2 Find and eventually draw the image of the set A ⊂ X under the function
f : X → Y , for
(a) X = R2 , Y = R, A = {(x, y) ∈ X | x 2 + y 2 = 4},
f (x, y) = (x − y)(x + y).
(b) X = R+ 2 , Y = R2 , A = {(x, y) ∈ X |x + y = 2},
√
f (x, y) := (u, v) = x y, 1 + x y .
(c) X = R2 , Y = R2 , A = A = {(x, y) ∈ X | 2x − y = 0},
f (x, y) := (u, v) = x + y, x 2 − y 2 .
Answers
(a) f (A) = [−4, 4].
(b) f (A) is a piece of hyperbola
√ u2 − v 2 = −1 contained between two
points (0, 1) and (1, 2).
(c) f (A) is a parabola u2 + 3v = 0.
Answers
(a) Injection—yes, surjection—no, bijection—no.
(b) Injection—yes, surjection—yes, bijection—yes.
(c) Injection—no, surjection—yes, bijection—no.
Chapter 3
Defining Distance in Sets
The present chapter is devoted to the notion of a metric. We will learn how to
check whether a given function is a metric and draw the special sets called balls
and segments.
The definition of this object is as follows. A metric or, in other words, distance
in a certain set X is a function d : X × X → R, which satisfies the following
conditions (axioms of a metric) for all x, y, z ∈ X:
1. d(x, y) = 0 ⇐⇒ x = y.
2. d(x, y) = d(y, x) (symmetry).
3. d(x, z) ≤ d(x, y) + d(y, z) (triangle inequality).
If one defines such function in a certain set X, this set is called a metric
space. From the given conditions, it follows that for any x, y ∈ X this function
is nonnegative: d(x, y) ≥ 0. This result can be easily obtained from the last axiom,
if one puts z = x and makes use of the first two properties.
Given a metric in some metric space, one can define a ball and a segment. Their
appearance is quite different dependent on the metric chosen even if the set X
remains the same. This is because the metric space (X, d) is different from (X, d ),
where d and d denote different metrics.
An open ball with the center at certain x0 ∈ X and radius r > 0 is the collection
of all points x ∈ X satisfying
This equality should be compared to the third axiom (i.e., the triangle inequality)
satisfied by any point y if d is a metric. A space endowed in a metric is called the
metric space.
In the following problems we will be acquainted with several specific examples
of metrics.
Problem 1
Solution
The definition of metric and its axioms are given just above. The function (3.1.1)
defines the so-called Euclidean metric, sometimes also called Pythagorean. Below,
it will be proved that in fact it satisfies all three required conditions. Let us consider
them in turn.
1. Let us choose on the plane any point x with coordinates (x1 , x2 ) and calculate its
distance from itself, i.e., dE (x, x). We get
dE (x, x) = (x1 − x1 )2 + (x2 − x2 )2 = 02 + 02 = 0. (3.1.2)
is satisfied. However, if the sum of the squares is zero, then each of them
separately must be also equal to zero. We have, therefore, x1 = y1 and x2 = y2 ,
or simply x = y. Hence, the implication in both directions holds, and the first
axiom is fulfilled.
3.1 Examining Whether a Given Function Is a Metric 53
2. The second axiom is also very easy to check, since one has
dE (x, y) = (x1 − y1 )2 + (x2 − y2 )2 (3.1.4)
= (y1 − x1 )2 + (y2 − x2 )2 = d(y, x).
where a1,2 , b1,2 ∈ R. If one now adds a12 b12 + a22 b22 to both sides, it is possible to
rewrite this inequality in the form:
(a12 + a22 )(b12 + b22 ) ≥ a12 b12 + 2a1 b2 a2 b1 + a22 b22 = (a1 b1 + a2 b2 )2 (3.1.6)
This result represents the special case of the so-called Schwarz inequality. It will
turn useful in a while, but now let us go back to the metric and transform the
expression:
(dE (x, y) + dE (y, z))2 = dE (x, y)2 + dE (y, z)2 + 2dE (x, y)dE (y, z)
= (x1 − y1 )2 + (x2 − y2 )2 + (y1 − z1 )2 + (y2 − z2 )2 (3.1.8)
+ 2 (x1 − y1 )2 + (x2 − y2 )2 (y1 − z1 )2 + (y2 − z2 )2 .
We are now going to make use of (3.1.7) by inserting a1,2 = x1,2 − y1,2 and b1,2 =
y1,2 − z1,2 and getting
The absolute value present in (3.1.7) has been omitted, since for any real number w,
the inequality |w| ≥ w holds. Transforming (3.1.9) further, one finds
54 3 Defining Distance in Sets
(dE (x, y) + dE (y, z))2 ≥ x12 + y12 − 2x1 y1 + x22 + y22 − 2x2 y2 + y12 + z12
− 2y1 z1 + y22 + z22 − 2y2 z2 + 2x1 y1 − 2x1 z1 + 2y1 z1
− 2y12 + 2x2 y2 − 2x2 z2 + 2y2 z2 − 2y22
= x12 + z12 − 2x1 z1 + x22 + z22 − 2x2 z2 (3.1.10)
= (x1 − z1 ) + (x2 − z2 ) = dE (x, z) ,
2 2 2
Problem 2
Solution
Zero value has been obtained, as it had been expected. Now let us assume
One would like to conclude that a = b. Since the largest number among
nonnegative |ak − bk | is equal to zero, this means that they all vanish:
ak − bk = 0 for k = 1, 2, . . . , n, (3.1.14)
and thus one comes to the obvious conclusion that indeed a = b. The first axiom
is, therefore, fulfilled.
2. Now, let us verify the symmetry property:
The last inequality is dictated by the fact that the index k has now become
common for both expressions inside absolute values. If the largest numbers
|ak − bk | and |bk − ck | correspond to different values of k, one has inequality,
and if they correspond to the same values, one has equality. Hence, the use of the
symbol ≥ is justified.
Let us now make use of the known fact
Problem 1
Balls and segments in the so-called “river” metric defined in the set R2 with the
formula:
(x1 − y1 )2
+ (x2 − y2 )2 if x2 · y2 ≥ 0
dm (x, y) = 2 , (3.2.1)
x1 + x2 + y12 + y22
2 if x2 · y2 < 0
will be drawn.
Solution
Let us begin by explaining the peculiar name of this metric. Well, imagine the
abscissa on the plane as a river and a bridge located at the origin. Suppose one
wants to get to a certain point y of coordinates (y1 , y2 ) starting from x of coordinates
(x1 , x2 ). If the two points are on the same side of the river, which means that x2 and
y2 have the same sign, one simply follows the shortest path, using the Euclidean
metric known to us from the first exercise of the previous section (the top formula
of (3.2.1)). However, if x2 and y2 are of different signs, i.e., points x and y are
located on opposite sides of the river, one first needs to go, using the shortest path,
from x to the bridge (that is to the origin), again using the Euclidean metric, and
then in the same way from the bridge to the point y. This is described by the lower
formula of (3.2.1).
When solving the present problem it will no longer be checked if the considered
function is in fact a metric. This was the subject of the previous section. Here it is
already assumed that all three axioms of a metric are met. Instead, we will focus on
balls and segments.
• Ball. A closed ball with the center at x and radius r is defined by the formula:
dm (x, y) ≤ r. (3.2.2)
One can say that it is a collection of all such points y, which one is able to reach,
when going with a constant velocity within the specified time. By choosing the
speedappropriately, one can agree that r is just the number of walking hours. If
r ≤ x12 + x22 , which means that we are able to walk, at most, up to the bridge,
the ball has a simple form:
with the difference that all points lying on the other side of the river are
removed—perhaps they are even close to x (in the common sense), but one are
unable to reach the points at any given time. This situation is shown in Fig. 3.1.
The white point is the starting point, and at the same time, the center of
the ball (x). Its exemplary coordinates are (2, 2). Balls with r = 1 and r = 2
are completely contained in the upper half-plane
√ and are
√ quite “normal.” When
increasing r, but in such a way that r ≤ 22 + 22 = 2 2, we note that ordinary
balls would cross into the lower half-plane, and those parts must be√removed.
However, if r exceeds the distance between x and the bridge (i.e., 2 2 ), there
occurs a part of the ball also on the other border. In the figure it corresponds to
the balls with radii 4 and 6. For example, when r = 6 all points y satisfying the
condition
r 6
r 4
r 2
r 1
x
bridge
1
-1 1
z
58 3 Defining Distance in Sets
i.e.,
√ √
y12 + y22 ≤ (6 − 2 2)2 = 4(11 − 6 2) ∧ y2 < 0. (3.2.7)
It is then the ordinary semi-circle lying on the lower side of the river with the
center in the origin and with a relatively small radius because some part of 6 h
has already been used to get to the bridge.
• Segment. A line segment connecting points x and z (it will simply be denoted
with xz) is, for any metric d, a collection of all points y, for which (not strict)
triangle inequality will be realized as equality
It is interesting to note here the fact, which will turn out useful in a moment,
that if w belongs to the segment xz, then the whole of segments xw and wz are
contained in it. For, one has then
Now, if a certain point y lies in the segment of xw, then the condition
Please note that we have the symbol ≥ rather than ≤. The only way to reconcile
it and the triangle inequality (which must be met, because d is, after all, a metric)
is to accept that the following equality holds:
and that means that in addition to w (which previously was assumed) y must
belong to the segment xz. Similar reasoning can be carried out for the segment
wz. It becomes apparent that it must also be contained in xz as a whole. Let us
note that nowhere in the above considerations any specific form of a metric has
been referred to; hence, these properties apply to not only dm but also any other
metric.
If the points x and z are located on the same side of the river, it is not necessary
to go over the bridge and the segment appears in a traditional way. In order to
explain the reasons for this, let us for a moment look back to Problem 1 in the
previous section, to the point where triangle inequality for the Euclidean metric
was being proved. If we trace the reasoning carried out there, we will come to
3.2 Drawing Balls and Segments 59
the conclusion that equality (3.2.8) will be true only if a1,2 and b1,2 introduced
in the formula (3.1.5) are chosen so that
a1 b2 = a2 b1 . (3.2.13)
This is an equation of the straight line (in variables y1 , y2 ) passing through the
points x and z, which can easily be checked by substituting the coordinates. The
segment, in this case, is actually a piece of the line connecting the points x and z.
Things look differently if these points lie on opposite sides of the river. An
exemplary segment in this case is drawn as a dashed white line. To this segment
belong all points lying on the straight line between x and the “bridge” and on the
straight line between the “bridge” and z. It is easy to justify this conclusion. The
path from x to y is composed of two pieces, and on each of them one moves,
using the ordinary Euclidean metric. But for the Euclidean metric it has already
been checked what the segment looks like and we can now make use of this
knowledge. The origin (“bridge”) undoubtedly belongs to the segment, because
The last equality is simply the lower equation from the definition (3.2.1). Our
segment must then be composed of two ordinary Euclidean segments that extend
between x and (0, 0) and between (0, 0) and z. It has been justified a few
moments earlier for any metric. One can now choose the “bridge” as w and use
the conclusions obtained after formula (3.2.8).
Problem 2
Balls and a segment in the so-called “post office” metric defined on the plane R2 as
x12 + x22 + y12 + y22 if x1 y2 = y1 x2
dr (x, y) = , (3.2.16)
(x1 − y1 )2 + (x2 − y2 )2 if x1 y2 = y1 x2
will be drawn.
60 3 Defining Distance in Sets
Solution
The “post office” metric owes its name to the post system where all packages have
to go through the central post office (in abbreviation CPO), except those that may be
delivered along the way. (Sometimes this exception is not included in the definition.)
By analyzing (3.2.16), one can easily see that the role of CPO is played by the frame
origin. If the two points lie on a straight line passing through it, the distance between
them is a simple Euclidean distance (lower formula). This is because they lie on the
same path to CPO. However, if they are not located on such a line, the package first
needs to go (in the Euclidean way)) to CPO, and from there a different path to the
end point is chosen (upper formula).
Now, it will be examined how balls and segments in this metric occur.
• Ball. As we know, a set of points y forming a closed ball is described by the
inequality
dr (x, y) ≤ r, (3.2.17)
where x plays the role of the center of the ball and r its radius. For the purposes
of illustration below, sample values assumed: x1 = 1 and x2 = 3. Now we will
consider two situations. If the radius r is smaller than the Euclidean distance
from the point x to CPO, i.e., if
√
r< (1 − 0)2 + (3 − 0)2 = 10, (3.2.18)
then the ball boils down to a simple straight line or segment passing through the
origin and point x. Its center is just this point, and the length is equal to 2r. This
is due to the fact that it is not possible to meet the upper formula of (3.2.16)
with the condition (3.2.18). Otherwise the following equation would have to be
satisfied:
√
1 + 3 + y1 + y2 = r ⇐⇒
2 2 2 2 y12 + y22 = r − 10, (3.2.19)
which, naturally, leads to a contradiction, since the value on the right-hand side
is negative. One is, therefore, restricted to the lower formula of (3.2.16) and to
the points y satisfying
after having taken into account that x1 = 1 and x2 = 3. The equation (3.2.20)
describes just such a straight line that we have been talking about. The segment
is limited by the condition
3.2 Drawing Balls and Segments 61
(x1 − y1 )2 + (x2 − y2 )2 = (1 − y1 )2 + (3 − y2 )2 (3.2.21)
= (1 − y1 )2 + 9(1 − y1 )2 = 10(1 − y1 )2 ≤ r
y2 =3y1
r r r
⇒ |y1 − 1| ≤ √ ⇒ 1 − √ ≤ y1 ≤ 1 + √ .
10 10 10
√ When the radius of the ball (or our walk time) is large enough (i.e., greater than
10), one can successfully reach CPO and then one can either begin a walk along
the other radial path or one can move forward along the same one. The ball will
have the form of two sets: a simple (Euclidean) segment, spoken of before, and
an ordinary (Euclidean)
√ ball with the middle at the origin and a radius resulting
from (3.2.19): r − 10 > 0. A few examples of balls with different radii are
shown in Fig. 3.2.
• Segment. One can use the knowledge from the previous exercise. When the two
points x and y lie on a common straight line passing through the origin, the
distance is measured with the Euclidean metric, and hence one gets an ordinary
segment contained on this line. On the other hand, if these three points are not
collinear, one must first draw a straight line from x to the frame origin, and then
another one from there to y. An example of a typical segment is shown in Fig. 3.2
with a dashed white line. All the arguments, which lead to this conclusion, are
identical to those of the previous problem where we had to pass through the
post office
1
-1 1
r 5
r 6 y
r 7
62 3 Defining Distance in Sets
“bridge.” But this time one has to go by the central post office. We already
know that if any point w belongs to the segment xy (e.g., the origin), the whole
segments xw and wy must be contained in it.
ρ(x, y) = | log(x/y)|.
Answers
(a) Metric.
(b) Metric.
(c) Metric, a sample ball K(x0 , r) for x0 = 2 and r = 1 is a segment
]2/e, 2e[.
defines a metric in the set RN . If so, find out how balls in this metric occur.
Answers
Metric, exemplary ball K(x0 , r) for x0 = (1, 1, . . . , 1) and r = 1 is
]0, 2[×]0, 2[× · · · ×]0, 2[.
Chapter 4
Using Mathematical Induction
In this chapter we learn how to use mathematical induction in various proofs. The
method of proving different claims, identities, and inequalities, which is called the
mathematical induction, can be formulated as follows. Assume a certain thesis is
to be demonstrated for all n ∈ N. Then the inductive proof is composed of two
steps:
1. First, one must check that the thesis to be shown is true for n = 1. In general the
verification of this fact is very simple.
2. The second step is to demonstrate the veracity of the following claim: if one
assumes that the thesis is true for certain k ∈ N (the inductive hypothesis), then
it is also true for k + 1 (the inductive thesis). This part of the proof is generally
much more difficult.
The implementation of these two steps means that the inductive proof is accom-
plished and it allows for the conclusion that the thesis is true for every natural n.
It can happen that the property to be proved does not hold for all n, but for
“almost all,” i.e., all n starting from some m ∈ N. In such a case it is easy to adjust
the above-mentioned inductive steps to this situation: at first the veracity of our
claim for n = m is verified, and second one proves the implication for k ≥ m.
The arithmetical mean of certain numbers x1 , x2 , . . . , xn is defined in the
following way:
1
A(x1 , x2 , . . . , xn ) = (x1 + x2 + . . . + xn ), (4.0.1)
n
n
where the symbol , called the binomial coefficient, denotes
k
n n!
= . (4.0.5)
k k! (n − k)!
Problem 1
1
an = · 36n + 10 · 3n (4.1.1)
12
is divisible by 33.
Solution
1
· 36k+1 + 36 · 10 · 3k = 36 · 33 · l1 . (4.1.5)
12
Now, one needs to somehow convert the expression on the left-hand side to generate
ak+1 . It is known, what we want to achieve, so we write
1 1
· 36k+1 + 36 · 10 · 3k = · 36k+1 + 12 · 3 · 10 · 3k (4.1.6)
12 12
1 1
= · 36k+1 + 12 · 10 · 3k+1 = · 36k+1 + (1 + 11) · 10 · 3k+1
12 12
1
= · 36k+1 + 10 · 3k+1 + 110 · 3k+1 = ak+1 + 110 · 3k+1 = 36 · 33 · l1 .
12
The last equality implies that
The term in brackets is obviously an integer (remember that, in accordance with the
inductive hypothesis l1 ∈ N) or even natural, because we know from (4.1.4) that
ak+1 > 0. It is then the number l2 that has been looked for. We have resulted with
the inductive thesis, and, therefore, our proof is complete.
Problem 2
Solution
Unlike the previous example, now we will move from the left-hand side of the
induction claim, i.e., from the expression for pk+1 (x). Where this turns out to
be useful, the inductive assumption, i.e., the formula (4.1.10) will be exploited.
However, in order to have the opportunity for this, we need to transform the
expression in such a way as to perceive within it (or actually to get within it,
by rearranging terms, or even if necessary by adding and subtracting them) the
polynomial pk (x):
+ x k+1 + x 2 − x − x 2 + x + x − 1. (4.1.12)
−−−−−
A glance at the above formula shows that this objective can be achieved. Simply
collecting together all underlined terms, it is seen that their sum is simply x pk (x).
The other ones can also be rearranged leading to the result:
Under the inductive assumption one knows that the factor (x − 1)3 can be extracted
from pk (x). The second term explicitly contains (x − 1)2 , and the subsequent factor
(x − 1) comes from x k+1 − 1, which also vanishes for x = 1. One can make use of
the known formula:
In this way one comes to the final expression, in which the induction thesis is
recognized:
The conclusion is that the polynomial pn (x) defined by the formula (4.1.8) for any
natural n has a triple zero at x = 1.
68 4 Using Mathematical Induction
Problem 1
Solution
From the previous examples we know already the procedure to be applied. First the
formula for n = 1 is to be verified. To this end let us evaluate independently both
sides and compare them:
⎫
1
⎪
⎬
L= l · l! = 1 · 1! = 1
⇒ L = R. (4.2.2)
l=1 ⎪
⎭
R = (1 + 1)! − 1 = 2! − 1 = 1
Now one has to perform the inductive step: the inductive thesis should be proved
with the use of the inductive assumption. The inductive assumption has the form:
k
l · l! = (k + 1)! − 1. (4.2.3)
l=1
k
k+1
l · l! + (k + 1) · (k + 1)! = l · l! = (k + 1)! − 1 + (k + 1) · (k + 1)!. (4.2.5)
l=1 l=1
Since the left-hand side has already the form as in (4.2.4), i.e.,
k+1
l · l!,
l=1
4.2 Proving Equalities and Inequalities 69
it is sufficient to check that the same refers to the right-hand side. This calculation
is very simple
Problem 2
Solution
At the beginning we proceed as in the previous (and all other) example: the validity
of (4.2.7) is verified for n = 1 :
⎫
1
⎪
L= l =1 =1
3 3 ⎪
⎪
⎪
⎬
l=1 ⇒ L = R. (4.2.8)
1 ⎪
⎪
R=( 2 2 ⎪
m) = 1 = 1 ⎪
⎭
m=1
k
k
l3 = ( m)2 , (4.2.9)
l=1 m=1
It is visible that the left-hand side of (4.2.10) can be obtained from (4.2.9) by adding
the missing term in the form of (k +1)3 . Therefore, we add it to both sides of (4.2.9):
k
k+1
k
l + (k + 1) =
3 3
l =(
3
m)2 + (k + 1)3 . (4.2.11)
l=1 l=1 m=1
70 4 Using Mathematical Induction
The next step to be carried out now is to demonstrate that the obtained expression on
the right-hand side is identical to the right-hand side of the induction thesis (4.2.10).
This is so, in fact, as it is shown below.
k
k+1
( m)2 + (k + 1)3 = ( m − (k + 1))2 + (k + 1)3 (4.2.12)
m=1 m=1
k+1
k+1
=( m)2 − 2(k + 1) m + (k + 1)2 + (k + 1)3 ,
m=1 m=1
where the short multiplication formula has been made use of:
k+1
(a − b)2 = a 2 − 2ab + b2 for a = m and b = k + 1.
m=1
The expression k+1 m=1 m is the sum of first k + 1 terms of the arithmetic sequence
an with the first term a1 = 1 and common difference r = 1. This sum may
be calculated using the arithmetic mean of the first term (i.e., 1) and the last one
(i.e., k + 1 ), and then multiplying it by the number of terms (i.e., k + 1):
k+1
1 (k + 1)(k + 2)
m= (1 + k + 1)(k + 1) = . (4.2.13)
2 2
m=1
One should remember, however, that this method applies only for an arithmetic
sequence! The above formula could be, if someone wished to demonstrate it, treated
as a separate—and very simple—exercise to use mathematical induction.
Let us now insert (4.2.13) into the right-hand side of (4.2.12), getting
k
( m)2 + (k + 1)3 (4.2.14)
m=1
k+1
(k + 1)(k + 2)
=( m)2 − 2(k + 1) + (k + 1)2 + (k + 1)3
2
m=1
k+1
k+1
=( m)2 − (k + 1)2 (k + 2) + (k + 1)2 (k + 1 + 1) = ( m)2 .
m=1 m=1
The induction thesis has then been demonstrated and, in consequence, the for-
mula (4.2.7) too.
4.2 Proving Equalities and Inequalities 71
Problem 3
It will be shown that for every n ∈ N and for any positive numbers a and b the
inequality
is satisfied.
Solution
Contrary to the previous examples, this time we are going to prove the inequality
but our procedure will be similar with quite obvious modifications. We begin, as
usual, with verifying (4.2.15) for n = 1:
L = (a + b)1 = a + b
⇒ L < R, (4.2.16)
R = 21 (a 1 + b1 ) = 2(a + b)
since a, b > 0. Now let us write down explicitly the inductive hypothesis:
Contrary to the former Problems 1 and 2, the left-hand sides above are not sums of
terms, but products, and the inductive thesis has one more factor of (a + b). It is
then obvious that starting from the left-hand side of (4.2.17), in order to obtain the
left-hand side of (4.2.18), one needs to multiply the former by (a +b). This factor, in
accordance with the values of a and b, is positive and this operation does not change
the inequality. One obtains:
(a+b)k (a+b) = (a+b)k+1 < 2k (a k +bk ) (a+b) = 2k (a k+1 +bk+1 +a k b+bk a)).
(4.2.19)
On the left-hand side it has already been obtained (a + b)k+1 , but the expression on
the right-hand side is still not like that of the inductive thesis. Remember, however,
that this time we are proving not equality, but inequality. Therefore, all we have to
do is to demonstrate that
At first glance it is not clear how to achieve this. There is, however, a certain standard
method in such situations. It consists of transforming the expression on the left-hand
side (of (4.2.20)) to obtain that of the right-hand side plus possibly some additional
term. Then, to know whether the inequality holds or not, it will be sufficient to
determine the sign of this additional term. It is worth remembering that, generally,
it is easier to establish the sign of an expression than to search for its specific value.
In our case, the modification of the left-hand side in (4.2.20) will consist of adding
the missing 2 in front of a k+1 and bk+1 (on the left there is 2k , and on the right
2k+1 ). Of course it is not allowed to add anything to only one side of an equation
or inequality, so the added expressions must be simultaneously subtracted. In the
formula below, the changes have been marked with an underscore.
=2 k+1
(a k+1 + bk+1 ) + 2k (a k b + bk a − a k+1 − bk+1 )
= 2k+1 (a k+1 + bk+1 ) + 2k (a k − bk )(b − a). (4.2.21)
As one can see, we were able to incorporate our plan, since the expression became
actually the right-hand side of the inductive thesis plus an additional term. Moreover,
it is clear that
(a k − bk )(b − a) ≤ 0. (4.2.22)
This is because for a = b, this expression is always the product of one positive and
one negative number, and when a = b, the two factors are equal to zero. We have
obtained then
and this entails the inductive thesis. The procedure of the inductive proof is complete
and thus the inequality (4.2.15) is proved.
Problem 4
n
1 √
√ ≥ n (4.2.24)
m
m=1
is satisfied.
4.2 Proving Equalities and Inequalities 73
Solution
If L = R, the logical sentence L ≥ R is also true and hence the first step of induction
has been completed. As usual, the inductive hypothesis and the inductive thesis are
now formulated:
k
1 √
I.H. : √ ≥ k, (4.2.26)
m
m=1
k+1
1 √
I.T. : √ ≥ k + 1. (4.2.27)
m
m=1
We have already learned how to proceed when both left-hand sides differ in one term
of the sum: most often one simply adds it to the both sides of inductive
√ hypothesis.
The missing component, in the present case, has the form of a 1/ k + 1. One then
obtains
k
1 1
k+1
1 √ 1
√ +√ = √ ≥ k+ √ . (4.2.28)
m=1
m k + 1 m=1 m k+1
√ 1 √
k+ √ ≥ k + 1. (4.2.29)
k+1
The expression
√ on the left-hand side is certainly greater than that in which the first
term (i.e., k) would be multiplied by a certain factor ξ smaller than 1:
√ 1 √ 1
k+ √ >ξ· k+ √ for ξ < 1. (4.2.30)
k+1 k+1
The choice of this factor will become clear if we ponder what we want to achieve.
One should always have in front of his or her eyes the goal of given transformations
and expressions one is aiming to acquire. Chaotic transformations of formulas,
in general, shall not lead to the objective; similarly a chess player performing
accidental moves has little chance to achieve a checkmate. So in the first place, in
74 4 Using Mathematical Induction
√
order to simplify the expression, we want to have a common denominator k + 1.
Second, it would be helpful to get rid of the square root of k in the numerator. Third,
we would like to ensure that the numerator has the denominator as a factor because
on the right-hand side of the inequality (4.2.29) the denominator is absent. All of
these objectives will be realized if the numerator is brought to the form k + 1 in the
following way:
√
√ 1 √ k 1 k+1 √
k+ √ > k· √ +√ =√ = k + 1. (4.2.31)
k+1 k+1 k+1 k+1
ξ
√ √
The factor ξ < 1 in (4.2.29) turned out to be the fraction k/ k + 1.
In this way, the inequality (4.2.29) has been proved and consequently the
inductive thesis too.
Problem 5
holds.
Solution
Recalling (4.0.5), one can easily check that the inequality (4.2.32) is satisfied for
n = 1, since one has
⎫
= 2⎪
2 2!
L= = ⎪
⎬
1 1! (2 − 1)!
⇒ L = R ⇒ L ≥ R. (4.2.33)
41 ⎪
⎪
R= √ =2 ⎭
2 1
2(k + 1) 4k+1
I.T. : ≥ √ . (4.2.35)
k+1 2 k+1
In order to decide how to proceed further, one has to expand the expressions on
the left-hand sides in accordance with the formula (4.0.5) and determine how they
differ. Then it will be known what operation should be carried out on the inequality
being the inductive assumption.
2k (2k)! 2(k + 1) (2k + 2)!
= , = (4.2.36)
k k! k! k+1 (k + 1)! (k + 1)!
It is easy to notice that the second expression can be obtained from the first one
by multiplying it by the factor
One still has to transform, in some way, the expression on the right-hand side in
order to show that it is not less than
4k+1
√ .
2 k+1
If it is not obvious how to start, one can always try to isolate on the right-hand
side (4.2.37) the needed factor in a “mechanical” way. Then, it is sufficient to
determine whether the remaining factor is greater or less than one. Proceeding in
this way, one obtains
We arrived at the right-hand side of the inductive thesis multiplied by the additional
factor
k 2 + k + 1/4
> 1.
k2 + k
76 4 Using Mathematical Induction
4k 2(2k + 1) 4k+1
√ · > √ . (4.2.39)
2 k k+1 2 k+1
By combining this inequality with (4.2.37), one gets (4.2.35), and in effect, we see
that the claim was demonstrated.
Problem 1
It will be proved that for every n ∈ N and x ≥ −1, the so-called Bernoulli’s
inequality
(1 + x)n ≥ 1 + nx (4.3.1)
is satisfied.
Solution
The indicated Bernoulli’s inequality turns out to be very useful in different estimates
or proofs and it is worth being demonstrated as an exercise for mathematical
induction. The first such application will appear in Problem 3 of this section.
At the outset, it should be noted that for the special case x = −1 the
inequality (4.3.1) takes the form
0n = 0 ≥ 1 − n. (4.3.2)
For all natural n’s, the inequality is obviously satisfied. For this reason, we will
focus only on the inductive proof in the case x > −1.
In the first step, one verifies the veracity of the inequality (4.3.1) for n = 1. Let
us compare both sides:
L = 1 + x, R = 1 + x ⇒ L = R ⇒ L ≥ R. (4.3.3)
The inequality is satisfied, so we can proceed to the second part of our proof. As
one knows, it consists of demonstrating the implication:
for any k ∈ N. It is easy to see that the left-hand side of the inductive hypothesis
and that of the inductive thesis differ only by the factor (1 + x). This observation
determines how to proceed. The important fact is that this factor is positive, since
the case x = −1 has already been excluded. Hence, if both sides of the inductive
assumption are multiplied by (1 + x), the inequality sign is not reversed.
Problem 2
Solution
The binomial coefficient has already been recalled in the equation (4.0.5), so the
formula to be demonstrated is clear. We verify if the equation (4.3.7) is satisfied for
n = 1. As usual, the left-hand and the right-hand sides are compared:
⎫
L = (a + b)1 = a + b ⎪
⎪
1
⎪
⎬
1 1 1
R= a b =
1−l l
a b +
1−0 0 1−1 1
a b ⇒ L = R.
l 0 1 ⎪
⎪
l=0 ⎪
⎭
= 1·a·1+1·1·b =a+b
(4.3.8)
Newton’s formula gives then the correct result for n = 1. Now let us proceed with
the second step and, to this goal, the inductive hypothesis and thesis are stated:
78 4 Using Mathematical Induction
k
k
I.H.: (a + b) = k
a k−l bl , (4.3.9)
l
l=0
k+1
k+1
I.T.: (a + b) k+1
= a k+1−l bl . (4.3.10)
l
l=0
The experience already gained immediately tells us how to use the induction
hypothesis in our proof: both sides of (4.3.9) have to be multiplied by the factor
(a + b) with the temporary assumption that it is different from zero, i.e., a = −b.
As a result of this operation, the left-hand side of the inductive thesis is obtained.
What remains is to write down the resulting right-hand side in such a way to be able
to identify it as the right-hand side of (4.3.10). Thus one has
k
k
(a + b)k+1 = (a + b) · a k−l bl (4.3.11)
l
l=0
k
k
k k
= a k−l+1 bl + a k−l bl+1 .
l l
l=0 l=0
Now the question arises how to further transform this expression. The answer
to this question is contained in another question: what do we want to obtain? A
glance at the inductive thesis (4.3.10) makes us aware that we finally need power
law expressions of the form a k+1−l bl . In the first component of the sum, there is
such a product, but in the second one the exponents disagree. One has to introduce
(only in the second sum) a new summation variable l = l + 1 and then instead of
a k−l bl+1 one will have what is needed, i.e., a k+1−l bl . Naturally, we remember that
this new variable does not run from 0 to k but from 1 to k + 1. One obtains, therefore
k
k+1
k k
(a + b) k+1
= a k−l+1 l
b + a k+1−l bl . (4.3.12)
l l−1
l=0 l=1
Note that in the second sum the prime in l has been omitted since l was only a
summation variable, i.e., “dummy” variable, which can be freely called l as well. In
addition, there was an obvious change in the binomial coefficient:
k k
−→ .
l l−1
From the formula (4.3.12), one can see that both terms are similar in structure.
One should now write them under a common symbol of a sum, but this cannot be
done automatically because of the different summation limits: in the first term from
0 to k, and in the second one from 1 to k +1. This trouble will be managed, however,
4.3 Demonstrating Some Important Formulas 79
by considering the summation from 1 to k only (which is present in both terms), and
separating the term for l = 0 (from the first sum) and that for l = k + 1 (from the
second one). In that way one gets
k
k k k
(a + b)k+1 = a k+1−0 b0 + + a k−l+1 bl
0 l l−1
l=1
k
+ a k+1−(k+1) bk+1 . (4.3.13)
k
We have come to the binomial symbol in the form as in the inductive thesis (4.3.10)
and our expression can be rewritten as
k
k k+1 k
(a + b) k+1
= a b +
k+1 0
a b +
k−l+1 l
a 0 bk+1 . (4.3.15)
0 l k
l=1
Now one would like to absorb the first and the last terms into the sum in the middle,
extending limits of summation: the lower one to zero and the upper to k + 1. It is
easy to see that this is really possible because
k k+1 k k+1
=1= and =1= , (4.3.16)
0 0 k k+1
and constants a and b have exactly such powers, as they should. In this way, the
inductive thesis is finally obtained:
k+1
k+1
(a + b) k+1
= a k−l+1 bl (4.3.17)
l
l=0
The only thing that is left is to verify the veracity of Newton’s formula for the
special case a + b = 0, which was postponed. Let us think whether the steps,
which we have performed, would be “legal” for a + b = 0 as well. Tracing all the
transformations makes us aware that only in one place a dubious step was made:
it was while multiplying equation (4.3.9) by the factor (a + b). But the equation
obtained in this way (4.3.11) may also be observed from another point of view.
Imagine that we can manipulate the left-hand side of the induction thesis, i.e., of the
expression:
(a + b)k+1 = (a + b) · (a + b)k ,
and make use of the inductive hypothesis, replacing (a + b)k with the right-hand
side (4.3.9). This is the alternative way of transformations, which does not require
any multiplication by 0, and which leads to the same equation (4.3.11) and to
all its consequences. Therefore, the inductive thesis, and the entire claim, can be
demonstrated for the case a + b = 0 as well.
Problem 3
It will be shown that for each n ∈ N and n ≥ 2 and for any positive numbers
a1 , a2 , . . . , an , the following inequality between the arithmetic and geometric
means holds:
1 √
(a1 + a2 + . . . + an ) ≥ n a1 · a2 · . . . · an . (4.3.18)
n
It bears the name of Cauchy’s inequality.
Solution
Both means
1
A(a1 , a2 , . . . , an ) := (a1 + a2 + . . . + an ), (4.3.19)
n
√
G(a1 , a2 , . . . , an ) := n a1 · a2 · . . . · an , (4.3.20)
(a1 + a2 )2
⇒ (a1 + a2 )2 ≥ 4a1 a2 ⇒ ≥ a1 a2
4
a1 + a2 √
⇒ ≥ a1 a2 ⇒ A(a1 , a2 ) ≥ G(a1 , a2 ). (4.3.21)
2
A(a1 , a2 , . . . , ak )
in order to obtain
A(a1 , a2 , . . . , ak , ak+1 ).
In such a situation, the easiest way is to start from the left-hand side of the inductive
thesis and to transform it using the inductive assumption. This method of action was
spoken of at the end of the previous exercise. Let us then write
a1 + a2 + . . . + ak + ak+1
A(a1 , a2 , . . . , ak , ak+1 ) = (4.3.24)
k+1
k · (a1 + a2 + . . . + ak )/k + ak+1
= .
k+1
The objective of this form of writing is clear: the terms have been so manipulated to
produce in the numerator the left-hand side of the inductive hypothesis which will
allow us to use the inequality (4.3.22):
A(a1 , a2 , . . . , ak+1 )
k · A(a1 , a2 , . . . , ak ) + ak+1 k · G(a1 , a2 , . . . , ak ) + ak+1
= ≥
k+1 k+1
(k + 1 − 1) · G(a1 , a2 , . . . , ak ) + ak+1
=
k+1
ak+1 − G(a1 , a2 , . . . , ak )
= G(a1 , a2 , . . . , ak ) + (4.3.25)
k+1
ak+1 − G(a1 , a2 , . . . , ak )
= G(a1 , a2 , . . . , ak ) 1 + .
(k + 1)G(a1 , a2 , . . . , ak )
82 4 Using Mathematical Induction
ak+1 − G(a1 , a2 , . . . , ak )
≥0 (4.3.26)
(k + 1)G(a1 , a2 , . . . , ak )
because the geometric mean of any numbers a1 , . . . , ak certainly is not greater than
the largest term, and hence cannot be greater than ak+1 . This is important because
it means that the assumptions required for the Bernoulli’s inequality are satisfied
where x represents the left-hand side of (4.3.26).
Now we have to think how to further transform the expression on the right-hand
side of (4.3.25) in order to obtain the inductive thesis. Our objective is to obtain,
at this point, the geometric mean of the numbers from a1 to ak+1 . Since the factor
outside the square brackets is already the product of the numbers from a1 to ak
(under the root), the missing number ak+1 should be extracted from these brackets.
In fact, this may be possible if a way is found to identify a common factor with
k + 1 in the denominator. If so, the unities inside will be reduced and we will get
the needed factor. Well, the method is already known: it is the use of the Bernoulli’s
inequality (4.3.1) proved earlier. Let us raise to the power k+1 both sides of (4.3.25):
The product of all numbers ai has actually been obtained. To get the inductive thesis,
it remains now only to take the root of degree k + 1 on both sides:
It simply means that its inverse is the arithmetic mean of the inverses of the numbers
a1 , a2 , . . . , an :
1 1 1 −1
H (a1 , a2 , . . . , an ) = A( , , . . . , ) . (4.3.30)
a1 a2 an
1 1 1
H (a1 , a2 , . . . , an ) ≤ G( , , . . . , )−1 = G(a1 , a2 , . . . , an )
a1 a2 an
≤ A(a1 , a2 , . . . , an ) . (4.3.31)
Problem 4
It will be proved that for each n ∈ N and n ≥ 2 and for arbitrary nonnegative
numbers q1 , q2 , . . . , qn satisfying the condition q1 + q2 + . . . + qn = 1 and for any
convex function f : [a, b] → R, the so-called Jensen’s inequality holds:
Solution
We must start by recalling what function is a convex one. The formal definition is
which means that (for x ∈]x1 , x2 [) the graph of the function lies below the secant as
shown in Fig. 4.1.
84 4 Using Mathematical Induction
f x
qf x1 1 q f x2
f q x1 1 q x2
x
a x1 qx1 1 q x2 x2 b
f (x2 ) − f (x1 )
( qx1 + (1 − q)x2 − x1 ) + f (x1 ) (4.3.35)
x2 − x1
f (x2 ) − f (x1 )
= (1 − q)(x2 − x1 ) + f (x1 )
x2 − x1
= (1 − q)(f (x2 ) − f (x1 )) + f (x1 ) = (1 − q)f (x2 ) + qf (x1 ).
The comparison of these two values explains the meaning of the definition (4.3.33).
One can now proceed with the inductive proof of the inequality (4.3.32). For
n = 2, it is satisfied automatically because a convex function is being considered
for which the inequality (4.3.33) by definition is satisfied. Hence, it remains to
implement only the second inductive step. The inductive hypothesis has the form
k
k
f( qi xi ) ≤ qi f (xi ), (4.3.36)
i=1 i=1
q1 + q2 + . . . + qk = 1
4.3 Demonstrating Some Important Formulas 85
k+1
k+1
f( qi xi ) ≤ qi f (xi ), (4.3.37)
i=1 i=1
q1 + q2 + . . . + qk + qk+1 = 1.
Both sets of numbers qi do not have anything in common and are completely
independent.
As in the previous example, we are going to proceed starting from the left-hand
side of the inductive thesis and transform it, using the inductive hypothesis. Let us
first separate the last term from the sum extending from 1 to k + 1:
k+1
k
k
k
qi xi
f( qi xi ) = f ( qi xi + qk+1 xk+1 ) = f ( ql k + qk+1 xk+1 ).
i=1 i=1 l=1 i=1 m=1 qm
(4.3.38)
k
The coefficient inserted in front of i=1 qi xi is simply a unity, since
k k
l=1 ql = m=1 qm . Next let us introduce the auxiliary denotation:
k
k
qi xi
q= ql , qk+1 = 1 − q, x = k , x = xk+1 . (4.3.39)
l=1 i=1 m=1 qm
k+1
f( qi xi ) = f (qx + (1 − q)x ) (4.3.40)
i=1
and, eventually, use of the convexity of our function. It should be noted that x ∈
[a, b], since all xi ∈ [a, b] for i = 1, 2, . . . , k. Denoting
one has
k
k
qi xi k
qi xmin qi
x = k ≥ k = ki=1 xmin = xmin ,
i=1 m=1 qm i=1 m=1 qm m=1 qm
k
k
qi xi k
qi xmax qi
x = k ≤ k = ki=1 xmax = xmax .
i=1 m=1 qm i=1 m=1 qm m=1 qm
(4.3.41)
86 4 Using Mathematical Induction
Thus xmin ≤ x ≤ xmax , and since both xmin and xmax belong to [a, b], this must
also be true for x . On the other hand, x ∈ [a, b] by assumption in the formulation
of the problem. This means that in (4.3.40) the definition of the function convexity
can be used which yields
k+1
f( qi xi ) ≤ qf (x ) + (1 − q)f (x ) (4.3.42)
i=1
k
q1 qk
= ql f ( k x1 + . . . + k xk ) + qk+1 f (xk+1 ).
l=1 m=1 qm m=1 qm
At this point we have arrived at the crucial moment of our proof. Introducing the
designation:
qi
q̃i = k for i = 1, . . . , k,
m=1 qm
k+1
k
f( qi xi ) ≤ ql f (q̃1 x1 + . . . + q̃k xk ) + qk+1 f (xk+1 ), (4.3.43)
i=1 l=1
it can be seen that one can use the inductive assumption. The following inequality
emerges
k+1
k
k
f( qi xi ) ≤ ql q̃i f (xi ) + qk+1 f (xk+1 )
i=1 l=1 i=1
k
k
qi
= ql k f (xi ) + qk+1 f (xk+1 )
l=1 i=1 m=1 qm
k
k+1
= qi f (xi ) + qk+1 f (xk+1 ) = qi f (xi ), (4.3.44)
i=1 i=1
in which the inductive thesis can be recognized. The claim is then demonstrated.
1
1 · 2 · 3 + 2 · 3 · 4 + . . . + n(n + 1)(n + 2) = n(n + 1)(n + 2)(n + 3).
4
Exercise 4 Prove that for every n ∈ N, the following inequality holds:
1 3 2n − 1 1
· · ... · <√ .
2 4 2n 3n + 1
Exercise 5 Prove that for every n ∈ N and n ≥ 2, the following inequality holds:
1 1 1 13
+ + ... + > .
n+1 n+2 2n 24
The sequence can be specified by an explicit formula for an , which can be treated
as a function a(n), whose domain is the set of natural numbers N, or by recursion.
In this latter case, one has to fix the value of the first element a1 and provide the
dependence an+1 = f (an ). The more complicated dependence, for instance an+2 =
f (an+1 , an ), requires fixing two first elements, a1 and a2 , etc.
The principal notion for the infinite sequence is its limit. A sequence is called
convergent to a certain number g if
If such number g exists it is called the limit of the sequence, which is written as
lim an = g. (5.0.6)
n→∞
If it does not, the sequence is divergent. A sequence that is bounded and monotonic
is convergent. There are several tests for the investigation of the convergence of
sequences which are discussed in detail when solving particular problems.
For convergent sequences, the following equations hold:
• lim an ± lim bn = lim (an ± bn ),
n→∞ n→∞ n→∞
• ( lim an ) · ( lim bn ) = lim (an · bn ),
n→∞ n→∞ n→∞
• ( lim an )/( lim bn ) = lim (an /bn ), if lim bn = 0,
n→∞ n→∞ n→∞ n→∞
and inserting into the second equation a constant sequence for bn , i.e., bn = c, one
also gets the relation lim c an = c lim an .
n→∞ n→∞
It can also be proved that a sequence in the special form
an = (1 + bn )cn , (5.0.7)
where
lim bn = 0
n→∞
lim an = eg . (5.0.9)
n→∞
i.e., they are the lower and upper bounds of the set of all cluster points. For
convergent sequences, both extremal limits are equal to each other and to the limit
of the sequence.
Problem 1
Solution
The sequence an is typical for problems where one has to find the limit of the kind
∞ − ∞. The procedure in all these cases is very similar. It comprises the following
steps:
1. First one should have to look at both the diverging terms in an to assess whether
their degrees of divergence are identical. A chance (but not certainty) for the
finite limit exists only when leading terms behave identically for n → ∞, since
this allows for the cancellation of troublesome expressions.
2. If these terms actually behave in the same way, one should, with the use of
appropriate transformations, bring about their real deletion. Then, most often, the
limit will no longer be of the type ∞–∞ because—one can say—the infinities
“subtracted” from each other.
3. At the end, one finds the limit of simplified expression.
92 5 Investigating Convergence of Sequences and Looking for Their Limits
A similar procedure, still involving the separation of leading terms, will be met
again in the calculation of limits in Problem 4 of the current section and Problem 1
in Sect. 5.5, and now we shall pursue it for an defined by (5.1.1).
Let us look at the first term. Under the fourth root the polynomial n4 + 2n3
is
√ found. For √ very large n the leading term is n4 , so one can expect that
4
n + n n4 = n. The behavior of the second term is similar. One can,
4 4 3
therefore, hope for cancellation of both divergent expressions and for the finite limit
of an .
One should be aware that the above reasoning is not strict and does not prejudge
the existence of the limit. Even if leading terms do cancel, this limit still may not
exist. Equally true, it may exist and be finite or even be equal to zero. Examples
of such different behaviors for apparently similar sequences will be given below.
Despite the lack of full strictness in reasoning carried out above, it plays an
important role: it indicates to us the possible solution to the problem. It is, as already
mentioned, the precise cancellation of unwanted expressions.
How then can we accurately reduce the diverging terms? To do this, of course,
we have to get rid of the symbols of roots. This can be done if one recalls the
familiar short multiplication formula (a − b)(a + b) = a 2 − b2 . Let our a and b be,
respectively, the first and the second terms in the formula for an . Then one can write
4
4
an = n4 + 2n3 − n4 + n3
√ √
4 4
4 4 n + 2n3 + n4 + n3
4 4 4
= ( n + 2n − n + n ) √
3 3 √
n + 2n3 + n4 + n3
4 4 4
√ √
n4 + 2n3 − n4 + n3
= √ √ . (5.1.2)
n + 2n3 + n4 + n3
4 4 4
For now, one has failed to reduce the unwanted terms, but still one has succeeded:
instead of roots of the fourth degree in the numerator the square roots have occurred.
This means that if one repeats this procedure once again both roots in the numerator
will disappear and the expected cancellation will take place. One should also note
here that roots in the denominator do not pose any problem, because they add to
each other, and not subtract!
Let us then write
√ √ √ √
n4 + 2n3 − n4 + n3 n4 + 2n3 + n4 + n3
an = √ √ ·√ √
n + 2n3 + n4 + n3 n4 + 2n3 + n4 + n3
4 4 4
n3
= √ √ √ √ .
( n + 2n3 + n4 + n3 )( n4 + 2n3 + n4 + n3 )
4 4 4
5.1 Some Common Tricks Useful for Calculating Limits of Sequences 93
This quotient expression is much easier to examine. One needs only to extract the
highest power of n in the numerator and in the denominator and one can calculate
the required limit:
n3 1 1
lim an = lim · √ √ √ √ = .
n→∞ n→∞ n3 ( 4 1 + 2/n + 4 1 + 1/n)( 1 + 2/n + 1 + 1/n) 4
(5.1.4)
Its value has been established, having regard to the following items:
nm
• lim = 1.
nm
n→∞ √
k
• lim 1 + a/n = k lim (1 + a/n) = 1 = 1, where the first equality follows
k
n→∞ n→∞
from the continuity of the “root” function, thanks to which the limit can be moved
into the argument.
• The following operations can be carried out on the limits of sequences:
bn lim bn
n→∞
lim = ,
n→∞ cn lim cn
n→∞
if all these limits exist and, in the latter case, the limit of cn (and, therefore, almost
all its terms) is not equal to zero.
At the end, it is worth emphasizing that in this type of example, it is nonleading
terms (that is, those that remain after cancellation of diverging expressions) that
are essential for the existence of the limit and its value. Let us, for example, make
apparently unimportant changes in the definition of sequence and write:
4 4 √ 4 √
an = n + 2 n · n − n4 + n · n3 .
3 (5.1.5)
√
It would seem that additional n is of no concern to the limit because the behavior
of the sequence is determined by n4 . Remember, however, that it is precisely this
main term which finally disappears. Proceeding in the same way as in (5.1.2)
and (5.1.3), we get:
√
n · n3
an = (5.1.6)
n3
1
· √ √ √ √ −→ ∞.
( 1 + 2/ n + 1 + 1/ n)( 1 + 2/ n + 1 + 1/ n) n→∞
4 4
94 5 Investigating Convergence of Sequences and Looking for Their Limits
n2 1
an = 3
· −→ 0.
n ( 1 + 2/n2 + 1 + 1/n2 )( 1 + 2/n2 + 1 + 1/n2 ) n→∞
4 4
(5.1.8)
One has then to be very careful while formulating conclusions concerning the
existence and values of limits only on the basis of leading terms.
Problem 2
1 1 1
an = √ +√ + ... + √ (5.1.9)
n +1
2 n +2
2 n +n
2
Solution
As usual, we are going to start solving the problem by carefully looking at the
formula for an . It is easy to realize that when n → ∞ each of the terms separately
tends to zero. However, it would be a serious mistake to conclude that an goes to zero
as well. Why? Well, because with the decrease of individual terms their total number
increases. The conclusion that an −→ 0 might actually be drawn if the quantity of
n→∞
ingredients remained bounded. It would be true, for example, if we considered a
sequence whose n-th term would have the form
1 1 1
√ +√ + ... + √ , (5.1.10)
n2 + 1 n2 + 2 n2 + k
for k ∈ N fixed. On the contrary, for an defined by (5.1.9), the value of the limit is the
result of the interplay between the number of terms and the rate of their convergence
to zero.
One of the standard ways of dealing with limits of that kind is to use the so-called
“squeeze theorem.” It says that if two sequences bn and cn are found with a common
limit g, satisfying the inequality bn ≤ an ≤ cn for almost all n, then lim an = g.
n→∞
5.1 Some Common Tricks Useful for Calculating Limits of Sequences 95
Thus, it remains now to choose sequences bn and cn . How can one guess them?
Well, they must converge to the common limit, which will also be the limit of the
tested sequence an . Therefore, we should have at least a certain idea of what the
value of that limit would be. Finding the answer to this question is our first task.
When looking at an , one sees that for large n each of the terms behaves as
√
1/ n2 = 1/n, and their total number is equal to n. We suspect, therefore, that
lim an = lim n/n = 1 = g. It remains to strictly demonstrate this.
n→∞ n→∞
Now one has to look for bn and cn . We already know that they should converge
to g = 1. In the formula (5.1.9), all terms are in descending order. This suggests a
certain idea: let us replace all n terms in the sum with the largest one, that is the first
one, and a sequence obtained in this way will be called cn :
1 1 1 1
cn = √ +√ + ... + √ =n √ . (5.1.11)
n +1
2 n +1
2 n +1
2 n +1
2
n 1
lim cn = lim · = 1. (5.1.12)
n→∞ n→∞ n 1 + 1/n2
Similarly, one can find sequence bn , this time by n-fold duplication of the smallest
term:
1 1 1 1
bn = √ +√ + ... + √ =n √ . (5.1.13)
n2 +n n +n
2 n +n
2 n +n
2
n 1
lim bn = lim ·√ = 1. (5.1.14)
n→∞ n→∞ n 1 + 1/n
In this way, it is clear that the assumptions of the squeeze theorem are satisfied. The
conclusion is then the lim an = 1.
n→∞
It is worth noting that the key to the solution was the fact that all of the
components of (5.1.9) behave for large n identically (as 1/n ). If certain seemingly
minor changes in the definition of an are made and one writes
1 1 1
an = √ +√ + ... + √ , (5.1.15)
n +1
2 2 n +2
2 2 n + n2
2
√When n → ∞,
then the difficulties in finding the limit will be much more serious.
the first term and the last term tend to zero as 1/n, and as 1/(n 2) respectively.
Therefore, one is not able to predict the limit. A purely “mechanical” application of
previous definitions gives
96 5 Investigating Convergence of Sequences and Looking for Their Limits
1 1 1
bn = n √ −→ √ , and cn = n √ −→ 1. (5.1.16)
n +n
2 2 n→∞ 2 n2 + 12 n→∞
Hence the sequences do not have a common limit, and the squeeze theorem cannot
be applied.
Problem 3
where k ∈ N, and λi > 0, i = 1, 2, . . . , k, will be proved and its limit will be found.
Solution
Finding the above limit is a typical exercise for the application of the squeeze
theorem. Just as in the previous example, we have to consider above all whether it is
possible to formulate any predictions as to the value of the limit. This is necessary
for choosing the two auxiliary sequences, which—as we know—both must converge
to it.
For a given n in the sum under the root, there is a fixed number of components
equal to k. One can assume then that among them there is a largest and a smallest.
Possibly there may even be a few if λi ’s are not all different for various i, but this
has no effect on our reasoning. Let us denote
The exponentials of the type λn quickly grow (for λ > 1) or decrease (for λ < 1 );
therefore, the behavior of the whole expression for large n shall be determined by
the largest number of all λi ’s. Therefore, we suspect that
an n λnmax = λmax −→ λmax . (5.1.19)
n→∞
Assuming that the value of the limit has been guessed correctly, one must now
indicate two sequences bn and cn , still convergent to λmax and satisfying—for almost
all n—the inequalities bn ≤ an ≤ cn .
The choice of bn is clear. Since it is anticipated that the limit of an is determined
by the greatest number of λi ’s only, then the same value of the limit should be
obtained if one disregards all other terms under the root. We are going to propose,
therefore,
5.1 Some Common Tricks Useful for Calculating Limits of Sequences 97
bn = n
0 + . . . + 0 + λnmax + 0 + . . . + 0 = λmax . (5.1.20)
Since all omitted terms were positive, obviously the inequality bn < an holds.
In turn,
√ in order to select cn , we remember that for any fixed positive α one has
lim n α = 1. This suggests that one might put all λi ’s equal to λmax :
n→∞
√
n
cn = n
λnmax + . . . + λnmax + . . . + λnmax = n
kλnmax =
k λmax −→ λmax .
n→∞
(5.1.21)
Of course an < cn and the squeeze theorem gives the expected result:
lim an = λmax .
n→n
It is worth noticing that estimates similar to those of the previous example (i.e.,
the sum of k smallest terms and the sum of k largest ones) would fail. For, leaving
the string cn without changes, and choosing bn as
bn = λmin + . . . + λnmin + . . . + λnmin ,
n n
(5.1.22)
one gets
√
n
bn = n
kλnmin = kλmin −→ λmin . (5.1.23)
n→∞
If not all λi are equal, then bn and cn do not converge to the common limit.
When solving these examples, one should realize that not only the selection of the
appropriate criteria (in this case, the squeeze theorem), but also the way of further
proceeding is dictated by the special form of the sequence under study.
Problem 4
Solution
In the first exercise of this section, we already encountered the situation where
leading terms diverging to infinity canceled, and a finite “remainder” determined the
limit. A similar case is dealt with in the present example. Someone might ask where
the place for cancellation of infinities is if in our formula for an ’s, there appears no
98 5 Investigating Convergence of Sequences and Looking for Their Limits
subtraction of any divergent expressions at all. Well, this cancellation results from
the periodicity of the sine function. It is true that the argument of this function goes
to infinity, but one can always reduce it to the interval [0, 2π [ by subtracting from it
the appropriate multiple of the period equal to 2π . For the sine function, then, it is
not the value of the argument itself that is important, but rather the remainder after
this reduction. This is reflected in our further steps.
If one looks at an , one can immediately
√ see that, for n large enough, the argument
2
of the sine function behaves as π n2 = nπ . That is just the value that should be
removed from the argument. This can be done by writing
an = sin π a 2 + n2 = sin π a 2 + n2 − nπ + nπ . (5.1.25)
Using now the formula for the sine of the sum of two angles,
one gets
an = sin π a 2 + n2 − nπ cos nπ + cos π a 2 + n2 − nπ sin nπ
= (−1)n sin π a 2 + n2 − nπ , (5.1.26)
Since sin 0 = 0 in the equation (5.1.26), one has the product of a bound
expression, and that convergent to zero. The product of this type is convergent to
zero as well, which can be easily justified. Adopting such notation that bn means a
bounded expression (by a number M > 0) and cn converges to zero, one has
Please note that for the final result, the vanishing limit of cn , i.e., in our case of
the expression:
sin π a 2 + n2 − nπ ,
Problem 5
Solution
The latter runs to zero since, due to the continuity of inverse cosine function, one
can write
1 1 π
lim arccos = arccos lim = arccos 0 = . (5.1.32)
n→∞ n n→∞ n 2
The limit (5.1.31) is, therefore, of the type ∞ · 0. To determine its value, one needs
to know how fast the expression in brackets decreases to zero. One of the possible
methods of proceeding in such circumstances is to examine the function arccos x
for x → 0+ , e.g., by using its expansion in the Taylor series. We will, however, use
another method, consisting of getting rid of the function that causes trouble—the
cyclometric function. To achieve this goal we write
an π 1
= − arccos . (5.1.33)
n 2 n
As we know, the right-hand side of this equation is convergent to zero, hence also
an /n −→ 0. Now let us calculate the sine of both sides of (5.1.33). Why do we
n→∞
choose sine and not cosine, if in the formula one has arccos? Well, it is because on
the right-hand side there is an additional π/2, which, after the application of the
reduction formula, will change the function sine into cosine:
an π 1 1 1 an 1
sin = sin − arccos = cos arccos = = · . (5.1.34)
n 2 n n n n an
From this, one can compute 1/an and find the limit:
1 sin(an /n)
= −→ 1 ⇒ an −→ 1, (5.1.35)
an an /n n→∞ n→∞
has been exploited for φ, expressed in radian measure. Since the latter limit exists
and is equal to 1, so, in accordance with Heine’s definition of the limit of a
function (see (7.0.1)), the same result must hold for each sequence of arguments
φn convergent to zero. In particular, one can choose φn = an /n obtaining (5.1.35).
But how does one know the result (5.1.36) is true? The easiest way to justify it is
to use Fig. 5.1, performed for 0 < φ < π/2.
It shows a circle of radius 1 and two rectangular triangles: AOB and COD with
the common angle φ. From AOB one has
|AB| |AB|
sin φ = = = |AB|. (5.1.37)
|OB| 1
5.1 Some Common Tricks Useful for Calculating Limits of Sequences 101
tanf
1
sinf
f
x
O A C
|CD| |CD|
tan φ = = = |CD|. (5.1.38)
|OC| 1
The length of the arc CB is just (in this measure) the value of the angle φ. One,
therefore, has
Combining these will result in the form cos φ < sin φ/φ < 1. Because the functions
cos φ and sin φ/φ are even, this formula is valid also for −π/2 < φ < 0. Taking
now an arbitrary sequence of arguments φn −→ 0, one can apply the squeeze
n→∞
theorem, one is already familiar with:
sin φn
1 ←− cos φn < < 1 −→ 1, (5.1.41)
n→∞ φn n→∞
and the desired conclusion (5.1.36) is reached. This result will turn out very useful
in many different problems and it is worth remembering.
102 5 Investigating Convergence of Sequences and Looking for Their Limits
Problem 1
pk (n)
an = (5.2.1)
ql (n)
Solution
It is worth solving one example referring directly to the definition of the limit. For
this goal, we chose the above limit; a step that appears as a component in many
different problems and one absolutely must know it. Estimations, performed below,
are not quite obvious. Let us start with the reminder of the definition of the limit:
However, it should be emphasized that if one wishes to use this definition in the
proof, one must first be able to guess the value of g.
Let us now write (5.2.1) in the form
The sequence will behave differently depending on which polynomial (that in the
numerator or that in the denominator) has a higher degree, or whether the degrees
are identical. Therefore, we are going to proceed with three independent cases
Case 1: l > k
Due to the fact that l − k is positive, by choosing a sufficiently large value of n, one
will certainly be able to make the factor nk /nl = 1/nl−k as small as one wishes. So
5.2 Using Various Criteria 103
now one has to focus on evaluating the second factor in (5.2.3). With the numerator,
the matter is simple: since n ≥ 1, then
α
0 α1 αk−1
k + k−1 + . . . + + αk ≤ |α0 | + |α1 | + . . . + |αk−1 | + |αk |. (5.2.5)
n n n
For the denominator, we need a reverse estimate—from below. It is not evident at
all and one has to ponder over the expression
β0 β1 βl−1
+ l−1 + . . . + + βl .
nl n n
It is intuitively felt that for suitably large values of n, the first l terms of the sum
will be small (as modules) compared to the last component, i.e., βl . Let us use
this intuition to find an appropriate estimate. One simply has to determine how
large n should be. For instance we might like the sum of the former l terms to
be smaller than, say, a half of the last one, i.e., |βl |/2. Then it could be written that
the denominator is estimated from below as follows:
β0
+ β1 + . . . + βl−1 + βl > |βl | − 1 |βl | = 1 |βl |. (5.2.6)
nl nl−1 n 2 2
Let us remember this restriction. We know from the text of the exercise that βl = 0,
so this expression is plausible. With the above assumption for n, one has:
For large n, this expression becomes arbitrarily small for any positive constant C. It
is anticipated, therefore, that this limit of the sequence is 0. Hence, if we choose, in
accordance with (5.2.2), any small > 0, it ought to be shown that there exists such
N ∈ N, that for n ≥ N one has
When looking at (5.2.8), it is seen that this requirement could really be satisfied, by
taking
n > l−k C/. (5.2.10)
104 5 Investigating Convergence of Sequences and Looking for Their Limits
With this choice, (5.2.9) is actually fulfilled and the proof is complete.
Case 2: l < k
Now, one will need inverse estimates of the numerator and the denominator
in (5.2.3). However, the transformations performed in case 1 can be reused, swap-
ping the roles of the numerator and the denominator. So, we have the counterpart of
the formula (5.2.5):
β0
+ β1 + . . . + βl−1 + βl ≤ |β0 | + |β1 | + . . . + |βl−1 | + |βl | (5.2.12)
nl nl−1 n
|αk |/2
|an | > nk−l · =: D · nk−l . (5.2.15)
|β0 | + |β1 | + . . . + |βk−1 | + |βk |
we obtain |an | > M. Almost all terms of the sequence |an | are greater than
the arbitrary constant! This sequence cannot, therefore, be bounded. It must be
divergent, and the same must refer to the sequence an itself.
5.2 Using Various Criteria 105
Case 3: l = k
Our purpose will be to demonstrate that the limit constitutes the number g = αk /βk .
Therefore, let us consider the difference an − g, i.e.,
One has managed to have the same issue as the first case: the polynomial in the
numerator has a lower degree than that in the denominator! We already know that
by choosing a sufficiently large N (for any previously indicated ), the inequality
an − αk < (5.2.18)
βk
can be satisfied. The indicated number is actually the limit. This result should be
memorized: if the degrees of polynomials in the numerator and the denominator are
equal, the limit of the sequence is the quotient of the coefficients accompanying the
highest powers of n.
Problem 2
will be examined.
Solution
In accordance with the text of the problem, we are going to solve it using Cauchy’s
criterion called
√ also the “nth root test.” As we know, in this case, one needs to
calculate n |an | and then examine the limit of this expression for n → ∞, and if
• lim n |an | < 1, then lim an = 0.
n→∞ n→∞
• lim n |an | > 1, then the sequence is divergent.
n→∞
106 5 Investigating Convergence of Sequences and Looking for Their Limits
• lim n |an | = 1, this criterion does not determine convergence and the sequence
n→∞
ought to be examined with other methods.
One can ask the question, how does one know that this criterion is the most
convenient to study the expression (5.2.19). Well, it is suggested by its structure
in which the most “troublesome” part seems to be the following factor in the
numerator:
2
1 n
1+ .
n
In the absence of Cauchy’s criterion, one should find some nontrivial estimate for
it. It should be noted that the other expressions in (5.2.19) are relatively easy to
estimate from above or below if necessary. Guided by the principle of adapting the
criterion to the most difficult part of an expression, we choose Cauchy’s criterion,
for which the nth root appears and
2
n 1 n 1 n
1+ = 1+ −→ e. (5.2.20)
n n n→∞
For now, we do not care about the other elements of √the formula (5.2.19) because one
can see that we will certainly handle them. One has n n −→ 1 and with expressions
n→∞
similar to that in the denominator we learned how to deal with solving Problem 3 in
Sect. 5.1.
The above considerations constitute only a motivation for the choice of this and
not of the other criterion. Now it is time to apply them to the relevant calculations.
One finds
√
n n (1 + 1/n)n
2 n
n (1 + 1/n)n
n
|an | = = √ . (5.2.21)
n + n2 + 3
2 n n
n + n2n + 3n
n 2
We would like now to make use of the rule that the limit of the product (or
quotient) is equal to the product (quotient) of limits, in so far as they all exist.
Sequences in the numerator have limits respectively equal to 1 and e, which has
been discussed above. It only remains to verify the limit of the sequence in the
denominator. For our transformations to make sense, it must be obviously different
from zero. The leading expression is naturally 3n , so for sufficiently large n one can
estimate
n 2 √ √
n
√n
n + n2n + 3n < 3n + 3n + 3n = 3 · 3n = 3 3 −→ 3 · 1 = 3,
n
n→∞
n √ √
n
n2 + n2n + 3n > 0 + 0 + 3n = 3n = 3 −→ 3.
n
(5.2.22)
n→∞
5.2 Using Various Criteria 107
lim an = 0. (5.2.25)
n→∞
Problem 3
(n/2)n (n + 1)
an = (5.2.26)
(2n + 1)!!(n + 2)
will be examined.
Solution
Let us first recall the d’Alembert criterion. First, one creates the quotient an+1 /an
and looks for its limit as n → ∞. Then, if
an+1
• lim < 1, then lim an = 0.
n→∞ an n→∞
an+1
• lim > 1, the sequence an is divergent.
n→∞ an
an+1
• lim = 1, this criterion does not determine convergence and the sequence
n→∞ an
ought to be examined with other methods.
108 5 Investigating Convergence of Sequences and Looking for Their Limits
(n/2)n
an = . (5.2.27)
(2n + 1)!!
and hence, by virtue of the d’Alembert criterion, it can be deduced that lim an = 0.
n→∞
As one can see, the most important element is the ability to choose a suitable
criterion to a specific problem. Nothing can supersede here the experience coming
from solving multiple problems. It is then easier to recognize in the expression the
important elements and those without any influence on the selection of the criterion.
In this example, the key elements were undoubtedly factors present in (5.2.27), and
certainly not a fraction (n + 1)/(n + 2).
At the end of this solution, let us consider what may happen in case the expression
|an+1 /an | has no limit at all. If |an+1 /an | −→ ∞, then for an arbitrarily large
n→∞
number M and for almost all n we have |an+1 /an | > M. The sequence |an | behaves
“worse” than a geometric sequence of quotient M. |an | (and, therefore, also an ) must
be divergent. However, if |an+1 /an | has no limit, but still does not tend to infinity,
5.2 Using Various Criteria 109
then even if lim supn→∞ |an+1 /an | = q > 1, the sequence an can be convergent
(even for q = ∞), which can be found out while considering the example of bn
with the following consecutive terms:
1 2 2 4 4 8
1, , , , , , , .... (5.2.30)
3 3 9 9 27 27
The odd and even subsequences are independent geometric sequences with quotient
equal to 2/3, and, therefore, both tend to zero, while we have
bn+1 bn+1 1
lim sup = 2, lim inf = . (5.2.31)
n→∞ bn n→∞ bn 3
Problem 4
Solution
The Stolz–Cesàro criterion (or theorem) allows us to find the limits of sequences of
the type
bn
an = ,
cn
and both these limits shall be equal. Interestingly, this criterion is also applicable
when an = bn /cn −→ ∞.
n→∞
There appears a quite natural question: why should it be more comfortable
studying the behavior of the fraction (bn+1 − bn )/(cn+1 − cn ) than the original
expression bn /cn ? Well, there are many examples of sequences that contain sums
of many terms in the numerator or denominator. Creating a difference of the type
bn+1 − bn , we hope that a large number of them will reduce, and then a simpler
expression remains. This is just the case for our problem. Additionally, one can
also hope that degrees of divergences in the numerator and the denominator will
decrease. In the present example,
In this way, our expression has been significantly simplified and reduced (5.2.32) to
the limit
(2n + 3)1/4
lim . (5.2.37)
n→∞ (n + 1)5/4 − n5/4
A potential problem here still can appear in the denominator in which the difference
of two expressions diverging to infinity is present. However, we have already learned
how to deal with this type of situation in the Example 1 of Sect. 5.1. Following the
idea elaborated there, we write
one sees that terms with n5 in the denominator are canceled. Extracting the highest
powers of n from the numerator and from the denominator, one gets
bn+1 − bn
lim = (5.2.39)
n→∞ cn+1 − cn
n4 (2 + 3/n)1/4 (1 + 1/n)5/4 + 1 (1 + 1/n)5/2 + 1 29/4
= lim = ,
n→∞ n4 5 + 10/n + 10/n2 + 5/n3 + 1/n4 5
Problem 5
Solution
When setting about solving similar problems, one should not be subject to the false
impression that it is too complicated or even unsolvable. As will be seen below, one
needs only to know how to separate, what is important from, what is insignificant,
and the example turns out to be very simple. But first one must recall the theorem
that will be used. We know from the lecture of analysis that if a sequence an is of
the special form
an = (1 + bn )cn , (5.2.41)
lim an = eg . (5.2.43)
n→∞
112 5 Investigating Convergence of Sequences and Looking for Their Limits
If one now looks at our expression (5.2.40), it will easily noticed that it has the
desired character. A fraction in brackets, despite its complex form, in an obvious
way goes to 1, due to the fact that the numerator and denominator have identical
leading terms (i.e., n3 ). This unity may be separated in order to reach the structure
such as in (5.2.41). The exponent, in turn, diverges to infinity, which gives hope to
meet (5.2.42). Hence, let us assume that:
n3 + 100 sin n2 + 1 n2
bn = − 1, cn = . (5.2.44)
n3 + n2 + log n 5 n+1
it can be seen that the last term decreases very quickly to zero as compared with the
first ones, so again, one could skip it and leave the denominator in the simplified
form of n3 (1 + 1/n).
Now the question may arise: how do we know that terms such as n2 (or worse)
appearing with n3 are important, while others can be omitted with impunity?
Well, the answer is given by the form of the sequence cn together with the
condition (5.2.42). For large n one has cn n, and a finite limit g in (5.2.42)
would be obtained when bn 1/n. Since both leading expressions in the numerator
and the denominator are of the form n3 , these giving (at the end) finite limit are
n3 · 1/n = n2 . In the possible case of vanishing of such terms—which is not the
case here—one eventually has to take into account further ones.
So, the following conclusion occurs: instead of calculating the limit of the
complicated sequence an , we are going to look for the limit of
n2 /(n+1)
n2 /(n+1)
n2 /(n+1)
n3 n3 + n2 − n2 1
= = 1− ,
n3 + n2 n3 + n2 n+1
(5.2.46)
since they are identical. And the latter may be obtained almost mentally as equal to
e−1 = 1/e, because
−1 n2
g = lim · = −1. (5.2.47)
n→∞ n + 1 n + 1
5.3 Examining Recursive Sequences 113
Let us now return to the full formula for an because the replacement of the
expression (5.2.40) with (5.2.46) was based only on our intuition and not on strict
arguments. One has
n3 + 100 sin n2 + 1 n2
lim bn cn = lim −1 (5.2.48)
n→∞ n→∞ n3 + n2 + log5 n n+1
−n2 − log5 n + 100 sin n2 + 1 n2
= lim ·
n→∞ n3 + n2 + log5 n n+1
n4 −1 − 1/n2 · log5 n + 100 sin n2 /n2 + 1/n2
= lim · = −1,
n→∞ n4 (1 + 1/n + 1/n3 · log5 n)(1 + 1/n)
1
lim an = e−1 = . (5.2.49)
n→∞ e
Problem 1
1 1
an+2 = an+1 + an , (5.3.1)
4 8
for n = 0, 1, 2, . . ., will be calculated, where the two initial members are chosen as
a0 = 5 and a1 = −1/2.
Solution
The idea is the following: let us substitute for the sequence terms an = λn , where
λ is a certain number other than zero. Now it should be verified if there exists such
value of λ, for which the equation is satisfied. For an chosen in that way, the increase
of n simply means multiplying by a constant, and this is, essentially, the sense of
the equation (5.3.1).
After the above substitution and simplifying both sides, one gets the quadratic
equation
1 1
λ2 − λ − = 0, (5.3.2)
4 8
which has two solutions: λ1 = 1/2 and λ2 = −1/4. “Forgetting” for a moment that
one has to comply with initial conditions, there are two possibilities:
n
1 1 n
an = and an = − . (5.3.3)
2 4
where α and β are constants. They can be fixed with the use of initial conditions.
Substituting n = 0, one has α + β = 5, and substituting n = 1, one finds 1/2 · α −
1/4 · β = −1/2. Thus, α = 1 and β = 4 and the complete solution is given by the
expression
n
1 1 n
an = +4 − . (5.3.5)
2 4
Since in both terms the bases are smaller than 1, the limit of an for n → ∞ is equal
to 0.
What complications can occur while solving this type of problem?
• The equation for a constant λ, similar to (5.3.2), may have a double root λ = λ0 .
In this case, one of the solutions has still the form an = λn0 , but apparently
the second solution is missing. In general, in order to meet the initial con-
ditions, we must have at our disposal two constants α and β, as in (5.3.4).
This, however, means that one needs two solutions. Substituting expression
an = nλn0 into the recursive equation, which then would have to be in the form
an+2 = 2λ0 an+1 − λ20 an (only in this case λ0 is a double root), one can easily
see that this equation is in fact satisfied. So, in place of (5.3.4), we have now the
formula
5.3 Examining Recursive Sequences 115
The recursive equation (5.3.1) may be now written in the matrical form
vn+1 = Mvn , where M is created with the coefficients read off from (5.3.1):
1/4 1/8
M= . (5.3.8)
1 0
1 1
φ(λ) = λ2 − λ−
4 8
116 5 Investigating Convergence of Sequences and Looking for Their Limits
and two eigenvalues: λ1 = 1/2 and λ2 = −1/4. Consequently, the equation (5.3.2)
turns out to be simply a characteristic equation for the matrix M, and parameters
λ1,2 —its eigenvalues!
If one denotes the eigenvectors with symbols u1 and u2 , the subsequent steps of
the solution are already clear from the course of algebra: v0 is expanded in the basis
of eigenvectors v0 = c1 u1 + c2 u2 (constants c1,2 are then known) and it can be
written as
vn = M n v0 = M n (c1 u1 + c2 u2 ) = c1 M n u1 + c2 M n u2
= c1 (λ1 )n u1 + c2 (λ2 )n u2 . (5.3.10)
The solution (5.3.5) can now be read off from (5.3.10), comparing the lower
components of the vector vn on both sides.
Problem 2
3an
an+1 = , (5.3.11)
an + 1
Solution
In this example, the recursive formula an+1 = f (an ) is nonlinear. In this case, we
do not have at our disposal any universal method leading to a formula for the general
sequence member, but one can still try to find the limit. First of all, it should be noted
that if this limit exists (let us denote it with g ), one is entitled to execute n → ∞ on
both sides of the equation (5.3.11). Naturally, the limits of an and an+1 are identical,
and the function on the right-hand side is continuous for arguments that interest us.
(If a0 > 0, the recurrence will never bring us out of positive values an , for which
the right-hand side of (5.3.11) is well defined.) So one can execute the limit within
its argument. In that way, one gets the equation
3g
g= , (5.3.12)
g+1
y x
y f x
a3
a2
a1
a0
a0 a1 a2 a3 x
2
we get the answer: the sequence is increasing and bounded above. It is, therefore,
convergent! Below it will be demonstrated in a strict way.
1. We prove the boundedness of the sequence.
This proof is going to be made by induction. It has already been assumed that
0 < a0 < 2, so it remains to demonstrate the following implication for any
index k:
3ak 3ak ak − 2
ak+1 = f (ak ) = =2+ −2=2+ . (5.3.14)
ak + 1 ak + 1 ak + 1
3ak 3
ak+1 = f (ak ) = = · ak . (5.3.15)
ak + 1 ak + 1
If 0 < ak < 2, which we already know, the factor 3/(ak + 1) > 1, and then
ak+1 > ak . The sequence is increasing.
The conclusion is that the sequence is convergent and its limit must be the number 2.
Now we are going to consider what changes in our reasoning occur if a0 > 2.
Let us again refer to the figure: the graph of the function f in this interval lies
above the line y = 2 (i.e., the sequence is bounded below by the number 2) and
below the line y = x (i.e., the sequence is decreasing, because f (an ) < an ). So
again, the sequence has to converge and its limit must be the number 2. A strict
proof is analogous to that carried out above and one can even use formulas (5.3.14)
and (5.3.15). Now, for ak > 2 one has (ak − 2)/(ak + 1) > 0, and, therefore, also
ak+1 > 2. Furthermore 3/(ak + 1) < 1, and in consequence ak+1 < ak .
Of course not all functions f behave in such a way as shown in the figure. An
interesting situation is the case where a sequence is convergent but not monotonic.
(This is the case we will look at in the next problem.) In other cases, it may happen
that the terms “escape” from the fixed point of a function f . Such a sequence is
divergent, unless there is another fixed point somewhere else.
5.4 When a Sequence Oscillates 119
Problem 1
6
an+1 = , (5.4.1)
2an + 1
Solution
As in the previous example, we start this exercise by drawing the appropriate figure.
The function defining the recursion will be plotted—in our example it is the function
f (x) = 6/(2x +1)—as well as the straight line y = x. It is sufficient to limit oneself
to positive values of x. This is because if recurrence starts at a certain a0 > 0, from
the equation (5.4.1), one quickly sees that each subsequent term will be positive too.
The x coordinate of the point of intersection of these two graphs (in our case
equal to 3/2 ), i.e., the fixed point of function f is a candidate for a limit g, which is
known from the previous example. If the limit of an exists, it must be equal to 3/2.
In the previous problem we were proving that the sequence was monotonic
and bounded, and hence one could conclude that it was convergent. In the present
example, the effort put into this kind of evidence would be vain—our sequence is
not monotonic! To realize this, let us refer to Fig. 5.3 where the first few terms are
shown. We begin with a0 marked on the x-axis and move as is indicated by the
arrows. Let us chose a0 < g = 3/2. When one calculates f (a0 ), it will be seen
that it is above the line y = 3/2, and, therefore, a1 will be located on the x-axis
to the right of 3/2. The next step will lead us again to the left of the fixed point. It
is clear that our sequence will oscillate around this point (which is due to the fact
that the function f is decreasing). Whether the number 3/2 will be “attracting” the
subsequent terms, as is the case in our figure, or “repelling” them as in the next
example, depends on a particular function f .
Assuming that our figure has been drawn with enough precision, one can see
the contents of the convergence proof of an . It will be composed of the following
steps:
1. We separate an into two subsequences: one with even indexes and one with odd
ones. (They are respectively called the even-numbered sequence and the odd-
numbered sequence.)
2. For each subsequence, the recursive formula with a certain new function f˜ is
written down.
120 5 Investigating Convergence of Sequences and Looking for Their Limits
a1
y x
a3
3
2
a4 y f x
a2
a0
a0 a2 a4 a3 a1 x
3
2
3. It is proved that each of these subsequences is monotonic and bounded, and hence
convergent.
4. The limits of both subsequences are fixed points of the relevant functions. Both
these limits are next verified to be identical.
Since both subsequences “consume” all the terms of an , their possible common
limit will be the limit of the sequence an .
At this point, one still has to explain the adopted assumption 0 < a0 < 3/2.
Well, if one took a0 > 3/2, then in the first step of the recursion one would just
get 0 < a1 < 3/2. So all one has to do in this case is to discard the first term with
no impact on the limit, and one returns to the sequence considered here. In turn, the
value a0 = 3/2 leads to the trivial constant sequence.
Now let us carry out the delineated program. The formula (5.4.1) may be given
the form
6
an+2 = . (5.4.2)
2an+1 + 1
5.4 When a Sequence Oscillates 121
Now, again using formula (5.4.1), one can eliminate an+1 for an , obtaining
6 2an + 1
an+2 = =6 . (5.4.3)
2 · 6/(2an + 1) + 1 2an + 13
This recurrence is “by 2” and binds separately even terms and odd terms with one
another:
2a2k + 1
a2k+2 = 6 , (5.4.4)
2a2k + 13
2a2k−1 + 1
a2k+1 = 6 . (5.4.5)
2a2k−1 + 13
The aforementioned function f˜ for both subsequences has, therefore, the form
2x + 1
f˜(x) = 6 .
2x + 13
Its only positive fixed point is still x = 3/2. First, the even-numbered subsequence
will be considered. The figure suggests that it is necessary to try to prove that this
subsequence is bounded and increasing.
1. We prove the boundedness.
As in the previous example, the mathematical induction is going to be used.
Given 0 < a0 < 3/2, it now ought to be demonstrated that, for any k,
3 3
0 < a2k < ⇒ 0 < a2k+2 < . (5.4.6)
2 2
The following series of equalities takes place:
2ak + 1 3 2ak + 1 3
a2k+2 = f˜(a2k ) = 6 = +6 −
2ak + 13 2 2ak + 13 2
3 a2k − 3/2
= +9 . (5.4.7)
2 2a2k + 13
It is clear that under the inductive assumption, the second component is negative,
and, therefore, a2k+2 < 3/2. Of course a2k+2 is positive too, which was
mentioned at the beginning. It follows that the even-numbered subsequence is
in fact bounded.
122 5 Investigating Convergence of Sequences and Looking for Their Limits
2a2k + 1 12a2k + 6
a2k+2 = f˜(a2k ) = 6 =
2a2k + 13 2a2k + 13
2 + 13a − 2a 2 − a + 6
12a2k + 6 2a2k 2k 2k 2k
= a2k = a2k
2a2k + 13a2k
2 2a2k + 13a2k
2
For 0 < a2k < 3/2, which has already been shown, the numerator of the fraction
is positive, and, therefore, the entire expression in brackets is larger than 1. Thus,
a2k+2 > a2k , i.e., the even-numbered subsequence is increasing.
The conclusion is that the even-numbered subsequence is convergent and its limit
must be the number 3/2.
For the odd-numbered subsequence we prove that it is bounded below by 3/2
and decreasing. Since the recurrence is described by the identical function f˜, some
expressions derived previously can still be used. Rewriting the formula (5.4.7) and
adjusting indexes, one has
3 a2k−1 − 3/2
a2k+1 = +9 . (5.4.9)
2 2a2k−1 + 13
In this case, the inductive assumption has the form a2k−1 > 3/2, which means that
the second component is positive. One then gets the inductive thesis: a2k+1 > 3/2.
Naturally, in this case the condition a1 > 3/2 required for induction is met, which
is due to assumptions 0 < a0 < 3/2. For one has
6 6 3
a1 = > = . (5.4.10)
2a0 + 1 2 · 3/2 + 1 2
2
2a2k−1 + 13a2k−1 + 2(a2k−1 + 2)(3/2 − a2k−1 )
a2k+1 = a2k−1
2
2a2k−1 + 13a2k−1
! "
(a2k−1 + 2)(3/2 − a2k−1 )
= 1+2 a2k−1 , (5.4.11)
2
2a2k−1 + 13a2k−1
5.4 When a Sequence Oscillates 123
and noting that this time the numerator is negative, one concludes that
a2k+1 < a2k−1 . The odd-numbered subsequence is decreasing and bounded, just
what we wanted to show. The limit must constitute the positive fixed point of the
function f˜, which is the number 3/2.
Since both limits are identical, and both subsequences “consume” all terms of
an , then the sequence cannot have other cluster points and one has
3
lim an = . (5.4.12)
n→∞ 2
Problem 2
Solution
From the previous examples, we already know that the eventual limit of an can only
be a fixed point of the function
from which it can be seen that other solutions do not exist because a trinomial in
brackets is always positive. (Its discriminant = −7 < 0.)
So if a0 = 1, then all an = 0 and the sequence is constant. This case is not
interesting and its solution—obvious. Therefore, we are going to continue with the
assumption a0 = 1. It should be noted that in this case for all subsequent terms, one
still has an = 1. This is because if one accepted that some an+1 = 1, the following
equation would have to be satisfied:
and it follows that also an = 1. By repeating this argumentation now for an and
an−1 , one gets an−1 = 1, and all previous terms, including a0 , would have to be
equal to unity too.
124 5 Investigating Convergence of Sequences and Looking for Their Limits
The situation faced around the fixed point x = 1 is shown in Fig. 5.4. One can
see that the subsequent members “escape” from this point. Below this conclusion is
demonstrated in a strict way, which also means that an has no limit.
One must remember that if the number g is the limit of this sequence, then for any
> 0 almost all its members must satisfy the condition |an − g| < (see (5.2.2)).
“Almost all” means “all starting from some (generally large) N ”. Therefore, let us
choose some small and suppose that certain an for large n lies in the vicinity of 1.
Let us examine what happens to an+1 .
Now notice that for an lying close to 1, and even for any nonnegative an we have
|an2 + an + 1| ≥ 1, which means that |an+1 − 1| ≥ |an − 1|. On the other hand, if
(for some n) an were negative, then it would also have to be
since all terms on the right-hand side would be negative. Such a sequence—with
infinitely many members less than zero—naturally could not be convergent to g = 1.
a4
y x
a2
a0
1
a1
y f x
a3
x
a3 a1 1a0 a2 a4
Fig. 5.4 The subsequent terms of the divergent sequence defined by (5.4.13)
5.5 Demonstrating Divergence of Sequences 125
From the inequality |an+1 − 1| ≥ |an − 1|, it follows, therefore, that the sequence
cannot have the limit equal to 1. This number “repels” the subsequent terms as is
shown in the figure. As discussed previously, other limits of the sequence an cannot
exist. It is then a divergent sequence (for a0 = 1).
Problem 1
Solution
A similar sequence has already been met—it was in Example 4 of Sect. 5.1. We are
dealing with a periodic (cosine) expression, the argument of which goes to infinity
with n. Out of this argument, as we have learned, one ought to extract the leading,
divergent term. For large n, naturally, one has
π n3 π
n. (5.5.2)
2n + n
2 2
This quantity should be isolated. It will be achieved by subtracting it and then
adding it to the argument of the cosine function. Using the well-known formula
cos(α + β) = cos α cos β − sin α sin β, one can write
π n3 π n3 πn πn
an = cos = cos − + (5.5.3)
2n2 + n 2n2 + n 2 2
πn
πn
π n3 πn π n3 πn
= cos − cos − sin − sin .
2n2 + n 2 2 2n2 + n 2 2
For the purpose of simplification, let us write this expression in the form
πn πn
an = cn cos − sn sin , (5.5.4)
2 2
126 5 Investigating Convergence of Sequences and Looking for Their Limits
Now let us transform the new argument of these trigonometric functions as follows:
π n3 πn πn n2 π n2 π n
− = − 1 = − · =− · .
2n2 + n 2 2 n2 + n/2 4 n2 + n/2 4 n + 1/2
(5.5.7)
Its infinite limit constitutes the number −π/4, and this—together with the continuity
of trigonometric functions—entails
π √2 π √
2
lim cn = cos − = , lim sn = sin − =− . (5.5.8)
n→∞ 4 2 n→∞ 4 2
It is important that these limits are different from zero. This suggests that the
sequence (5.5.4) will not have any limit because the coefficients accompanying
cn and sn , i.e., cos (π n/2) and sin (π n/2), oscillate, taking alternate values:
. . . 0, 1, 0, −1, 0, 1, 0, −1, 0, . . .. In order to demonstrate the divergence
of an , all one must do is to indicate two subsequences convergent to different limits.
To start, let us take a subsequence for which n will be constantly even (i.e., n = 2m,
m ∈ N). Then
where the known facts that sin mπ = 0, and cos mπ = (−1)m have been used.
The subsequence a2m still has no limit due to the oscillatory factor (−1)m . Our
choice was not good enough (after all we wanted to indicate different subsequences
that do converge to different limits), but we already know how to improve it: instead
of considering all m’s, one should choose only even, i.e., n divisible by 4 ( n = 4k,
k ∈ N):
The second subsequence, convergent to a different limit, is easy to find if one comes
back to (5.5.9) and select m as an odd number, i.e., n = 2m = 2(2k − 1) = 4k − 2,
for k ∈ N:
√
2
a4k−2 = (−1) 2k−1
c4k−2 = −c4k−2 −→ − . (5.5.12)
k→∞ 2
In√that way, two subsequences convergent to distinct limits have been found:
± 2/2, which ends the proof of the divergence of an . Comparing (5.5.11) and
(5.5.12), one can see why it was so important that the limit of cn was nonzero. For
zero value of the limit the oscillations of coefficients would not be relevant and in
both cases the limit would be simply ±0 = 0.
Since in this exercise we are interested in extreme limits, let us construct another
subsequence, exploiting so far unused members of an : these are the ones with odd
n (i.e., n = 2m − 1, m ∈ N):
1 1
a2m−1 = c2m−1 cos m − π − s2m−1 sin m − π = (−1)m s2m−1 .
2 2
(5.5.13)
As before, there are now two options: to take m either even (equal to 2k) or odd
(equal to 2k − 1). One gets
√
2
a2(2k)−1 = a4k−1 = (−1) s4k−1 = s4k−1 −→ −
2k
, (5.5.14)
k→∞ 2
√
2
a2(2k+1)−1 = a4k+1 = (−1)2k+1 s4k+1 = −s4k+1 −→ . (5.5.15)
k→∞ 2
If the set of cluster points (i.e., the set of limits of all convergent subsequences) had
more than two elements found in the solution, then to find its extreme limits, we
would simply take the least upper and the greatest lower bounds.
128 5 Investigating Convergence of Sequences and Looking for Their Limits
Problem 2
Solution
The experience gained in the previous example tells us immediately that the
sequence (5.5.17) should be divergent. One sees in fact that an is in the form of
the product of bn = n/(2n + 1) whose limit is not zero, but the number 1/2, and of
the oscillating factor sin (2π n/3). If so, one should be able to indicate at least two
subsequences convergent to different limits. What guides us while choosing these
subsequences? Well, as in the previous example, one would like to get rid of the
oscillating factor because only then the subsequent will be convergent.
Looking at the factor sin (2π n/3), we conclude that we should start with n being
a multiple of 3, in order to cancel the denominator. Therefore, let n = 3k for k ∈ N:
In this way the constant sequence has been obtained, all members of which are equal
to zero and consequently
n + a n + 1 − n2 + b n + 2
2
(b) an = √ √ , where a, b, c, d > 0.
n2 + c n + 3 − n2 + d n + 4
Answers
(a) Convergent, limit equal to 0.
(b) Convergent, limit equal to (a − b)/(c − d).
Exercise 2 Using the known criteria, examine the convergence and, eventually, find
the limits of sequences:
130 5 Investigating Convergence of Sequences and Looking for Their Limits
1 1 1
(a) an = √ +√ + ... + √ .
n +1
4 n +2
4 n + n2
4
2
n+1 n 5n n!
(b) an = √ , b n = .
n n
2 (2n)n
(c) an = sin(π n2 + 2n + 2), bn = cos(π n2 + 2n + 2).
1
(d) an = (1 · 1! + 2 · 2! + . . . + n · n!).
(n + 1)!
!√ "n2 /(n+1)
n+1+3 π n
(e) an = √ , bn = cos √ .
n+3 n
Answers
(a) Convergent, limit equal to 1.
(b) an divergent; bn convergent, limit equal to 0.
(c) an convergent, limit equal to 0; bn divergent.
(d) Convergent, limit equal to 1.
√
an convergent, limit equal to e; bn convergent, limit equal to e−π /2 .
2
(e)
Answers
(a) Convergent, limit equal to 0.
(b) Convergent, limit equal to 1.
(c) Convergent, limit equal to 2.
(d) Divergent.
In this chapter we deal with basic topological properties of sets, which are necessary
for the proper formulation of limit and continuity of real functions.
A neighborhood (denoted U (x0 , )) of a given point x0 in some metric space is
an open ball satisfying
An internal point of a given set A is the point contained in A together with its
certain neighborhood, i.e.,
The collection of all points of that kind constitutes the interior of the set A. A set
composed only of the internal points is an open set.
A point belonging to A for which such a neighborhood as in (6.0.3) cannot be
found is a boundary point. An open set cannot contain any of its boundary points,
and a closed set contains all of them. If a set contains some, but not all of its
boundary points, it is neither open nor closed.
The collection of real numbers R is an example of the set that is simultaneously
open and closed. Another such example is the empty set.
A set A in some metric space is called bounded if it is contained in a certain ball,
i.e.,
A set is compact if for each sequence composed of its elements, one is able to
select a subsequence convergent to a limit, which also belongs to this set. In the
special case of the space R (or Rn ), a set is compact if and only if it is closed and
bounded, provided the Euclidean metric is used.
A cluster point of a set A is such a point x that any of its neighborhoods contains
at least one other point of the set A. This is equivalent to saying that one is able to
construct a sequence composed of elements of A convergent to x. Therefore, the
cluster points are also called limit points. The closed set contains all its cluster
points. The closure of a set A is the union of A and the set composed of all cluster
points of A.
A set A is said to be connected if it cannot be written as the union of two
nonempty disjoint sets that are open. If such sets can be found, A is a disconnected
set.
Problem 1
Solution
The solution of this exercise will show us that intuition does not always lead to the
right conclusions. Let us consider a set defined as a sum of certain other sets, which
are closed. It may be, for example, the set A in the text of this problem. Seemingly,
we would expect that the sum of closed sets should be a closed set too. As will be
seen, this statement may not be true, if the sum is infinite.
The set A is the sum of closed intervals Pn = [1/2n, 1/(2n − 1)]. Substituting
the subsequent values of n starting from 1, one has
1 1 1 1 1
P1 = ,1 , P2 = , , P3 = , ,.... (6.1.2)
2 4 3 6 5
If one marked Pn ’s on the x-axis, each consecutive interval would lie to the left of
the previous one. What’s more, these sets are pairwise disjoint. If we take
6.1 Examining Openness and Closeness of Sets 133
1 1 1 1
Pn = , and Pn+1 = , ,
2n 2n − 1 2n + 2 2n + 1
it is seen that these sets do not overlap, because 1/(2n + 1) < 1/(2n). Indeed, it is
obvious that any number of x belonging to Pn , i.e., satisfying the inequalities
1 1
≤x≤ ,
2n 2n − 1
1 1
≤x≤ .
2n + 2 2n + 1
P4P3 P2 P1
x
0 11 1 1 1 1
65 4 3 2
Now we are going to study the set B. Let us denote the successive intervals of
the intersection with symbols Qn = ]−1/n, (n + 1)/n[. One has
1 3 1 4
Q1 = ]−1, 2[ , Q2 = − , , Q3 = − , ,.... (6.1.3)
2 2 3 3
In Fig. 6.2, several sets Qn are marked, the darker colors corresponding to larger
values of n. It can be seen that each of the subsequent intervals is fully contained in
the previous one. After comparing Qn and Qn+1 , we have
1 1 n+1 n+2
− <− , and > . (6.1.4)
n n+1 n n+1
1 n+2
− <x< ,
n+1 n+1
x
1 1 1 0 1 4 3 2
2 3 3 2
in all Qn . One sees, therefore, that another interesting conclusion has been obtained:
the infinite intersection of open sets can constitute a closed set.
Problem 2
is open or closed.
Solution
for example, with respect to the variable y, one could plot or explicitly write out
expressions for both curves y(x) and on this basis examine the properties of A.
However, for given expressions it is not possible and one has to use another method.
Let us denote
D := {(x, y) ∈ R2 | x > y}
which implies that f (D) = R. The interval ]0, 4[ is then fully contained in f (D)
and the set ]0, 4[∩f (D) =]0, 4[ is open. This property will be used below. The set
of inequalities in (6.1.5),
136 6 Dealing with Open, Closed, and Compact Sets
x 2 + 2y + log(x − y) > 0,
x 2 + 2y + log(x − y) < 4, (6.1.8)
simply defines the inverse image of this open set for the function f :
This simple observation practically terminates solving this problem. In fact, one can
use a property known from the lecture of analysis concerning continuous functions:
the inverse image of an open set is open. The conclusion is then: A is an open set.
It is worth noting that in order to obtain this result, we did not have to go too
much into the details of the formula for f . It was important only to determine that
it was continuous and that the set that was examined is an inverse image of an open
set. Therefore, one should not be “scared” of a seemingly complicated expression
in the text of this problem. To decide the question of the openness of the inverse
image, it is enough to really know very little about the function.
Now, the question arises whether the set A can, at the same time, be closed. We
know that such open-closed sets do exist, e.g., the entire space or the empty set.
However, it can be shown that in the matter under consideration here, i.e., the space
R2 with the Euclidean metric, they are the only sets with this property. Obviously,
the set A is neither empty (because Darboux’s property would be violated—in the
set there must exist points for which the function takes its values in the range ]0, 4[,
since its image is R), nor is it the whole space (because in the domain there are
points for which the values of the function lie outside the interval ]0, 4[). One can
conclude, therefore, that the set A cannot be closed.
Problem 3
was introduced. It will be examined whether the set A = [2, 3[ is open or closed.
Solution
In the course of solving this problem, we will see that the property of “openness” or
“closeness” of a set is not absolute, but depends on a metric introduced in the space.
The notion of a metric and axioms to be satisfied were recalled in Chap. 3.
6.1 Examining Openness and Closeness of Sets 137
At the beginning, let us note that if in our space the natural metric was introduced,
i.e., defined by the formula d(x, y) = |x − y|, the set A would be neither open nor
closed. Neither open, since the point 2 belongs to A, but A does not contain any of
its neighborhoods, nor closed because 3 does not belong to A even though it is the
cluster point. As it will be seen below, this situation may change if another metric is
chosen in space.
Before we start to solve this problem, which proves to be indeed very easy, let
us make sure that the function d(x, y) defined in the text of the problem actually
satisfies the metric axioms. As we know, they are
1. d(x, y) = 0 ⇐⇒ x = y .
2. d(x, y) = d(y, x) (symmetry).
3. d(x, z) ≤ d(x, y) + d(y, z) (triangle inequality).
The first condition is clearly met by (6.1.10). For x = y, d(x, y) > 0, and
obviously d(0, 0) = 0. In addition, the function d(x, y) is, in a visible way,
symmetrical while interchanging x ↔ y, so the second condition is also met. One
is only left with the triangle inequality. Let us calculate d(x, y) + d(y, z). First,
assume that two (or even three) of the numbers x, y, z are equal. Then,
In all the above cases, the triangle inequality is satisfied. One still has to consider
the case when all three arguments are different:
1 1 1
d(x, y) + d(y, z) = 1 + +1+ > 2 ≥ 1+ = d(x, z). (6.1.12)
x·y y·z x·z
The last inequality holds due to the fact that for x, z ∈ X one has 1/(x · z) ≤ 1.
Thus, it is seen that the function (6.1.10) is actually a good metric in space X.
Now the openness and closeness of A can be considered.
• Openness. The set is open if every point is contained in it with a certain
neighborhood, e.g., a ball of radius and center at x. If adequately small is
chosen, it is easy to see that the only point remaining within it is x itself because
its distance from other elements is, in accordance with the lower formula (6.1.10),
greater than 1. One can, therefore, conclude that any point of the set A is
contained in it with its suitably small neighborhood. A set A is, therefore, open
(in this metric).
138 6 Dealing with Open, Closed, and Compact Sets
• Closeness. A closed set is a set to which belong all its cluster points. The question
arises, how, in space X, does one construct a sequence of elements yn convergent
to a certain element x? After all, any point yn = x satisfies d(x, yn ) > 1, and the
limit would require d(x, yn ) < for each positive and suitably large values of
n. Having thought about it for a moment, we conclude that the only sequences
that are convergent in our space are constant sequences. Therefore, there is no
sequence created out of elements of A which might be convergent to a number
outside A and, consequently, this set must be closed.
Problem 1
A := {x ∈ R | x = α n , n ∈ N} (6.2.1)
is compact.
Solution
definition, it can easily be concluded that from each infinite sequence one can
select infinite constant subsequence (all members of which are equal to 1 or to
−1 ), i.e., convergent to a limit in A.
• 0 < |α| < 1. In this case, the set has an infinite number of elements and has
the form A = {α, α 2 , α 3 , . . . , α n , . . .}. Using these elements one can create the
sequence an = α n , n = 1, 2, . . ., easily seen as converging to 0. This point is
then the cluster point of the set A, but it does not belong to A. This means that
the set A is not closed and, therefore, it cannot be compact. The same conclusion
would be obtained from the definition referring to sequences. Well, since the
sequence an is convergent to 0, this number must be also the limit of any infinite
subsequence. Because of this, one cannot select any subsequence convergent to
the limit in A.
• |α| > 1. The set A has again an infinite number of components:
A = {α, α 2 , α 3 , . . . , α n , . . .}.
Problem 2
A := {(x, y) ∈ R2 | ax 2 + by 2 = 1} (6.2.2)
will be examined.
Solution
From the first exercise, we already know the meaning of compactness. The only
difference is that now the considered sets will not lie on a straight line but on a plane.
We are going to follow the previous example and consider all possible cases.
140 6 Dealing with Open, Closed, and Compact Sets
an = (n, (1 − an2 )/b) = (n, (1 + |a|n2 )/b).
Answers
A = [−1, 1[—neither open nor closed; B = [−4/3, 1]—closed.
Answers
As —compact; Bs —noncompact.
Answer
For q > 1/2 open, closed, compact; for 0 < q ≤ 1/2 nonopen, closed,
compact; for −1/2 < q ≤ 0 nonopen, closed, noncompact; for q ≤ −1/2
open, closed, noncompact.
Chapter 7
Finding Limits of Functions
This chapter is devoted to the notion of the limit of a function at a given point.
This notion is necessary for the formulation of the continuity and differentiability
of functions dealt with in the following chapters. Here we will learn how to find the
limits using some special tricks or substitutions.
There are two main definitions of the limit. Heine’s definition of the limit of a
function, defined in certain domain D, at a given point x0 , which is a cluster point
for D (see the theoretical summary from the preceding chapter) has the following
form:
#
lim f (x) = g ⇐⇒ ∀(xn ) (xn ∈ D ∧ xn = x0 , n = 1, 2, 3, . . .)
x→x0
$
∧ lim xn = x0 ⇒ lim f (xn ) = g . (7.0.1)
n→∞ n→∞
One-sided limits can be defined as follows. For the left limit denoted with the
“minus” sign at x0 one has
#
lim f (x) = g ⇐⇒ ∀(xn ) (xn ∈ D ∧ xn < x0 , n = 1, 2, 3, . . .)
x→x0−
$
∧ lim xn = x0 ⇒ lim f (xn ) = g . (7.0.3)
n→∞ n→∞
For the right limit denoted with “+,” these definitions respectively become
#
lim f (x) = g ⇐⇒ ∀(xn ) (xn ∈ D ∧ xn > x0 , n = 1, 2, 3, . . .)
x→x0+
$
∧ lim xn = x0 ⇒ lim f (xn ) = g , (7.0.5)
n→∞ n→∞
and
#
lim f (x) = g ⇐⇒ ∀>0 ∃δ>0 ∀x∈D ∧ x>x0 0 < |x − x0 | < δ
x→x0+
$
⇒ |f (x) − g| < . (7.0.6)
Problem 1
Using Heine’s and Cauchy’s definitions of the limit of a function, it will be proved
that
sin x
lim = 1. (7.1.1)
x→0 x
Solution
Heine’s definition of the limit of a function has been recalled above. As one can see,
this definition allows us to replace the calculation of the limit of the function f (x)
at x0 with the well-known limit of the sequence f (xn ) as n → ∞. We see that the
obtained result must be the same for all possible sequences xn −→ x0 satisfying
n→∞
the above assumptions. One may not assume anything more about xn .
Now, let us go back to the present problem. In Example 5 of Sect. 5.1, we justified
the inequalities
sin x
cos x < < 1, (7.1.2)
x
7.1 Some Common Tricks Useful for Calculating Limits of Functions 145
which will be used again. Since it is true for every x in the deleted neighborhood
of 0 (i.e., in the neighborhood from which the point 0 itself has been removed), the
same refers also to almost all terms of any sequence xn convergent to zero.
sin xn
cos xn < < 1. (7.1.3)
xn
Then the squeeze theorem can be applied. The sequence on the right-hand side is
constant (all the terms are equal to 1), so this constant constitutes its limit. For this
criterion to work, the sequence on the left-hand side must be convergent to unity
too. This is really the case because we remember that
i.e.,
xn x 2 xn
n
| sin2 |≤ −→ 0 ⇒ 1 − 2 sin2 −→ 1 . (7.1.7)
2 2 n→∞ 2 n→∞
Finally, one can draw the conclusion that
sin xn
lim = 1, (7.1.8)
n→∞ xn
and as a consequence,
sin x
lim = 1. (7.1.9)
x→0 x
Now we are going to try to get the same result using Cauchy’s definition of the
limit formulated in the theoretical summary (see (7.0.2)). As it can be seen, the given
condition means, roughly speaking, that the difference between |f (x) − g| can be
made arbitrarily small if one keeps x in a respectively small neighborhood of x0 .
Using the equations (7.1.2) and (7.1.4), one can write
x sin x sin x x x2
1 − 2 sin2 < < 1 ⇐⇒ 0 < 1 − < 2 sin2 ≤ , (7.1.10)
2 x x 2 2
146 7 Finding Limits of Functions
from which
2
1 − sin x < x (7.1.11)
x 2
is obtained.
Now let us choose any small > 0. The question is whether one can find such a
small δ > 0 that for |x − 0| < δ the following inequality holds:
1 − sin x < . (7.1.12)
x
Problem 2
The limit
x3 − x2 − x − 2
lim (7.1.13)
x→2 x 3 + x 2 − 4x − 4
will be found.
Solution
Below, we apply Heine’s definition of the limit recalled in the previous problem. To
learn a little more about the function, which is presently being studied (let us denote
it with f ), let us factorize the numerator and the denominator:
x 3 − x 2 − x − 2 = (x − 2)(x 2 + x + 1),
x 3 + x 2 − 4x − 4 = (x − 2)(x + 2)(x + 1). (7.1.14)
It can be seen that, at the limit point (i.e., for x → 2), both numerator and
denominator vanish. Therefore, the function is not defined at this point. Now
consider any sequence xn of real numbers convergent to 2 and suppose that xn = 2
for n = 1, 2, . . .. The corresponding sequence of values f (xn ), provided by Heine’s
definition, is
(xn − 2)(xn2 + xn + 1)
f (xn ) = . (7.1.15)
(xn − 2)(xn + 2)(xn + 1)
7.1 Some Common Tricks Useful for Calculating Limits of Functions 147
Since xn = 2 for any n, one can definitely cancel (nonzero) factors (xn − 2), getting
xn2 + xn + 1
f (xn ) = . (7.1.16)
(xn + 2)(xn + 1)
We are now ready to proceed with n to infinity, i.e., with xn to zero. The limit of
a sum of sequences is equal to the sum of their limits, and the limit of a product
(quotient) of sequences equals the product (quotient) of their limits, provided all
these limits do separately exist, and that resulting in the denominator in addition is
different from zero. Therefore, one gets
xn2 + xn + 1 4+2+1 7
lim f (xn ) = lim = = . (7.1.17)
n→∞ n→∞ (xn + 2)(xn + 1) (2 + 2)(2 + 1) 12
7
From the definition of the limit, it can be deduced that lim f (x) = .
x→2 12
Problem 3
will be found.
Solution
In Sects. 5.1 and 5.5 we encountered similar expressions while calculating limits of
sequences. The essence of the solution was then the separation of the leading term in
the argument of sine or cosine and making use of the formula for the trigonometric
function of the sum of angles.
√ One could do the same in this example, noting first
that for large x one has x 2 + a 2 x, and writing
cos x 2 + a 2 = cos( x 2 + a 2 − x + x) = cos( x 2 + a 2 − x) cos x
− sin( x 2 + a 2 − x) sin x. (7.1.19)
However, one may find it convenient to use the formula for the difference of two
cosines:
α+β β −α
cos α − cos β = 2 sin sin . (7.1.20)
2 2
148 7 Finding Limits of Functions
a2
√ −→ 0,
2(x + x 2 + a 2 ) x→∞
and since the sine is a continuous function, then:
a2
lim sin √ = sin 0 = 0. (7.1.23)
x→∞ x + x 2 + a2
It is seen, therefore, that the function f in (7.1.22) can be written in the form of
the product of the bounded function:
√
x 2 + a2 + x
f1 (x) := −2 sin , ( |f1 (x)| ≤ 2 ) (7.1.24)
2
and the one convergent to zero:
a2
f2 (x) := sin √ . (7.1.25)
2(x + x 2 + a 2 )
Thus, this product, f1 (x) · f2 (x), must have the limit equal to zero. For sequences,
we justified it in detail in Problem 4 in Sect. 5.1 (see formula (5.1.28)), and in the
present case, this justification is identical:
lim |f1 (x) · f2 (x)| ≤ lim 2 · |f2 (x)| = 2 · lim |f2 (x)| = 2 · 0 = 0. (7.1.26)
x→∞ x→∞ x→∞
7.1 Some Common Tricks Useful for Calculating Limits of Functions 149
Problem 4
The limit
tan x − sin x
lim (7.1.28)
x→0 x3
will be found.
Solution
Since
then one is dealing with the limit of the kind 0/0. For the time being, we do not
intend to use the l’Hôpital’s rule, which for such limits is particularly convenient,
but requires the skill of differentiation of functions. To this method, Sect. 10.4 will
be devoted, so now, we are going to try to solve this problem by other means.
First of all, it should be noted that in the numerator the sine function is explicitly
present, but the same function is also hidden in the tangent: tan x = sin x/ cos x.
Therefore, this common factor can be extracted together with x in the first power. In
this way one separates the expression sin x/x, whose limit for x → 0 is well known
to us from the first exercise and equals 1:
In order to further simplify (7.1.30) and rewrite it as the product of known limits,
we are going to use the identity that appeared in the Problem 1:
x x x
cos x = cos2 − sin2 = 1 − 2 sin2 . (7.1.31)
2 2 2
Then
2
1 − cos x 2 sin2 x/2 1 1 sin x/2
= = · · . (7.1.32)
x 2 cos x x 2 cos x 2 cos x x/2
One now collects all factors and make use of the fact that
sin x/2 sin t
lim = lim = 1, (7.1.33)
x→0 x/2 t=x/2 t→0 t
and also that the limit of a product (under conditions known from the previous
examples) is the product of limits:
2
1 sin x 1 sin x/2 1 1 1
f (x) = · · · −→ · 1 · · 12 = . (7.1.34)
2 x cos x x/2 x→0 2 1 2
Problem 5
The limit
√ √ √
| x − y| + |x − y|
lim , (7.1.35)
x→y |x 2 − y 2 |
Solution
√
In the denominator, the situation
is obvious: this factor has the form of |x − y|
and can be easily extracted out of |x 2 − y 2 |:
|x 2 − y 2 | = |x − y| · |x + y| = |x − y| · |x + y|. (7.1.36)
The expression has been led to the form of a product of two factors, each of which
has a well-defined limit for x → y:
√ √ √ √ √ √ √
| x − y| + |x − y| |x − y| |x − y| + x + y
lim = lim √ · √ √ √
x→y |x 2 − y 2 | x→y |x − y| |x + y|( x + y)
√ √ √ √
|x − y| |x − y| + x + y
= lim √ · lim √ √ √ (7.1.38)
x→y |x − y| x→y |x + y|( x + y)
√ √
0+ y+ y 1
= 1· √ √ √ =√ .
y + y( y + y) 2y
Problem 1
tan(kx + x 2 /π )
lim , (7.2.1)
x→π tan(nx + x 2 /π )
Solution
x = y + π. (7.2.2)
tan(kx + x 2 /π ) tan(k(y + π ) + (y + π )2 /π )
lim = lim (7.2.3)
x→π tan(nx + x 2 /π ) y→0 tan(n(y + π ) + (y + π )2 /π )
The tangent function is periodic with period equal to π , so the expressions such as
(k + 1)π or (n + 1)π , which is seen within its arguments, can be skipped. The limit
will be then rewritten as follows:
tan((k + 2)y + y 2 /π ) tan((k + 2)y + y 2 /π )
lim = lim (7.2.4)
y→0 tan((n + 2)y + y /π )
2 y→0 (k + 2)y + y 2 /π
(k + 2)y + y 2 /π (n + 2)y + y 2 /π
· · .
(n + 2)y + y 2 /π tan((n + 2)y + y 2 /π )
As one can see, we brought the issue to the calculation of three straightforward
limits. The second one does not pose any problems:
The first and the last limit in (7.2.4) are of similar character, so it is sufficient to deal
with only one of them. Denoting
(k + 2)y + y 2 /π =: z, (7.2.6)
one gets
Identically
(n + 2)y + y 2 /π
lim = 1, (7.2.8)
y→0 tan((n + 2)y + y 2 /π )
7.2 Using Substitutions 153
and after gathering all factors, one obtains the final result:
Problem 2
e x − e x0
lim (7.2.10)
x→x0 x − x0
will be calculated.
Solution
To solve this problem, we use the value of the limit well known from the lecture of
analysis:
x
1
lim 1+ = e. (7.2.11)
x→∞ x
Notice that the limit x → ∞ corresponds to y → −∞ and that the factor (1 + 1/y)
goes to 1. We conclude that apart from (7.2.11), one also has
x
1
lim 1+ = e. (7.2.13)
x→−∞ x
Now let us transform the expression given in the text of this problem in such
a way that only the difference x − x0 appears in it. This will allow us, after the
154 7 Finding Limits of Functions
introduction of the new variable t = x − x0 , to study the limit at 0, and not for
finite x0 :
e x − e x0 ex−x0 − 1 et − 1
lim = lim ex0 · = ex0 · lim . (7.2.14)
x→x0 x − x0 x→x0 x − x0 t=x−x0 t→0 t
The next step, thanks to which we will be able to use formulas (7.2.11)
or (7.2.13), is the substitution of a new variable for the whole expression in the
numerator: u = et − 1, which means that t = log(1 + u). The limit for t → 0
corresponds to the limit for u → 0. Therefore,
e x − e x0 u
lim = ex0 lim . (7.2.15)
x→x0 x − x0 u→0 log(1 + u)
For the logarithmic function, one has a log b = log ba for b > 0, which enables
us to rewrite the fraction above in the form
u 1 1
= = . (7.2.16)
log(1 + u) 1/u · log(1 + u) log(1 + u)1/u
Since the limit can be written in the denominator, one has to find
lim log(1 + u)1/u = log lim (1 + u)1/u (7.2.17)
u→0 u→0
because of the continuity of the logarithmic function. When introducing the variable
v = 1/u, this expression will be similar to (7.2.11) or (7.2.13).
Please note, however, that a certain delicate point is encountered here. When u →
0, can one write v → ∞ or v → −∞? If one-sided limits u → 0+ or u → 0− were
considered, this would be true. But in general u may approach zero passing through
“mixed,” negative and positive values. How to deal with this situation? Intuitively
one feels that the result in this case should be the same, i.e., equal to e, but it must
be shown in a strict way.
The easiest way to achieve it is to use Heine’s definition of the limit (see (7.0.1)).
Consider a sequence un convergent to 0 when n → ∞. Since the case of one-sided
limits in u is not problematic, we are going to restrict ourselves to such situations
where this sequence has an infinite number of negative terms and an infinite number
of positive ones. For example, it may oscillate around zero. Then it can be broken
apart into two infinite subsequences, u+ − + −
k and uk , such that uk > 0 and uk < 0 for
all k ∈ N. Since
1 v
lim (1 + u)1/u = lim 1 + = e, (7.2.18)
u→0+ v→+∞ v
by virtue of (7.2.11), for each sequence of arguments convergent to zero from the
right side (and, therefore, for u+
k ), the same result must be obtained. Identically
7.2 Using Substitutions 155
1 v
lim (1 + u) 1/u
= lim 1 + = e, (7.2.19)
u→0− v→−∞ v
and simply come back to the formula (7.2.15), also taking into account (7.2.16) as
well as the fact that log e = 1. The final result is
e x − e x0
lim = e x0 . (7.2.22)
x→x0 x − x0
Problem 3
Solution
At first glance, the essence in this exercise seems to be the removal of the
troublesome n-th root by making the substitution
√
t= 1 + αx − 1.
n
(7.2.24)
1
x= [(1 + t)n − 1]. (7.2.25)
α
156 7 Finding Limits of Functions
When t approaches zero, all terms in the denominator, except for the first one,
disappear. There may be many (when n is very large), but their number is always
fixed . Thus, one can apply the rule that a limit of a sum of expressions is the sum of
individual limits. In that way one gets
√
n
1 + αx − 1 α α
lim = = . (7.2.29)
x→0 x n + 0 + ... + 0 n
Problem 4
x 1/n − x 1/m
lim , (7.2.30)
x→1 x 1/k − x 1/ l
where n, m, k, l ∈ N, will be found, with the assumptions that n < m and k < l.
7.3 Exercises for Independent Work 157
Solution
In the previous example, with the use of substitutions, we managed to get rid of
an irrational function from the limit. The question is whether it is also possible in
the present exercise. One has here, however, as many as four roots. Is it possible to
dispose of them all at once? As it turns out, the answer to this question is positive:
one simply has to introduce a variable t using the formula x = t r , where r ∈ N
should be divisible by all four numbers: n, m, k, l. For example, one can choose:
r = n · m · k · l.
With this substitution the limit x → 1 gets converted into t → 1, so one obtains
The first fraction, i.e., t nkl /t nmk obviously converges to 1. The second requires
a moment of reflection. Let us introduce two parameters: p = (m − n)kl and
q = (l − k)m. Both are natural numbers. If so, the familiar formula can be used
t p − 1 = (t − 1)(tp−1 + t p−2
+ . . . + 1), (7.2.32)
p terms
and the analogous one for t q − 1. If one now goes with t to 1, each of p terms in
brackets tends to 1 too, and therefore, all of them together tend to the number p.
Hence one has
tp − 1 t − 1 t p−1 + t p−2 + . . . + 1 p
lim = lim · =1· , (7.2.33)
t→1 t q − 1 t→1 t − 1 t q−1 + t q−2 + . . . + 1 q
x→4 x 2 − 16
x +1
2 1
(c) 1◦ . lim sin x − sin . 2◦ . lim − cot x .
x→∞ x x→0 sin x
log(1 + x) − x
(d) 1◦ . lim x (log(1 + 2x) − log(2x)). 2◦ . lim .
x→∞ x→0 x
eax − ebx
(e) 1◦ . lim , where a, b = 0. 2◦ . lim ( cosh x − sinh x).
x→0 sin x x→∞
◦ cosh x − 1 ◦ 5 3
(f) 1 . lim . 2 . lim − .
x→0 cos x − 1 x→2 x 2 + x − 6 x2 − x − 2
Answers
(a) 1◦ . limx→2+ f (x) = limx→−2+ f (x) = π/2;
limx→2− f (x) = limx→−2− f (x) = −π/2. 2◦ . Limit equals 0.
(b) 1◦ . There is no limit (the function is divergent to ∞).
2◦ . Limit equals −1/96.
(c) 1◦ . Limit equals √0. 2◦ . Limit equals 0.
(d) 1 . Limit equals e. 2◦ . Limit equals 0.
◦
Answers
(a) 1◦ . Limit equals 1. 2◦ . Limit equals log√2.
(b) 1◦ . Limit equals −1. 2◦ . Limit equals 2.
(c) 1◦ . Limit equals 27(log 3 − 1). 2◦ . Limit equals α/2.
(d) 1◦ . Limit equals −1/(2π ). 2◦ . Limit equals 1/e.
Chapter 8
Examining Continuity and Uniform
Continuity of Functions
or in Cauchy’s version
or equivalently
The reader should be aware of the essential difference between (8.0.2) and (8.0.7).
In the latter case, one has to fix the value of δ independently of later chosen x and
x . This condition is more difficult to satisfy than what is required by (8.0.2). It will
be explained in detail in Sect. 8.3.
Problem 1
It will be proved that the function f (x) = tan x is continuous for any argument x in
the interval ] − π/2, π/2[.
Solution
As we know from the lecture of analysis and from the theoretical summary above,
the continuity of a function f defined on a set D at a certain point x0 ∈ D means
8.1 Demonstrating the Continuity of Functions with Heine’s and Cauchy’s. . . 161
that the limit of this function, when x → x0 , equals its value at this point. The
formal definitions have been formulated above.
Now let us take a look at the text of our exercise concentrating first on Heine’s
definition. Proving the continuity of simple functions is generally very easy, while
the difficulty lies in understanding and precisely executing a few formal steps, which
are required of us by the definition (8.0.1). First, we take a certain x0 ∈]−π/2, π/2[.
Now, as n → ∞, the sequence of function values, i.e., tan xn is expected to tend to
tan x0 . It will be convenient to consider the difference of these two quantities and
check that it converges to zero. We will use the formula
At this point, one might use the continuity of the sine or cosine functions and say that
the right-hand side goes to 0/ cos2 x0 , that is to 0, if the continuity of these functions
at x0 has been previously demonstrated. We are going, however, to proceed another
way and bring our proof to the end without relying on the continuity of other
trigonometric functions.
In order to estimate the numerator of (8.1.2), a well-known inequality for x close
to zero can be used:
In turn, to estimate the denominator (this time from below), it should be noted first
that, since xn −→ x0 , almost all terms xn , including all of them starting from a
n→∞
certain N ∈ N, must lie in the neighborhood of x0 . This neighborhood may be
chosen arbitrarily small if N is increased correspondingly. Let us then choose as the
radius of this neighborhood (i.e., an open circle) the number δ satisfying
π |x0 | π
δ= − >0 (since |x0 | < ). (8.1.4)
4 2 2
Cosine is an even function and decreasing on the interval [0, π/2[, so one can write
π |x0 |
cos xn = cos |xn | > cos(|x0 | + δ) = cos |x0 | + −
4 2
|x0 | π
= cos + > 0. (8.1.5)
2 4
162 8 Examining Continuity and Uniform Continuity of Functions
π |x0 | π π
≤ + < .
4 2 4 2
It is obvious since the radius δ was adjusted with the use of the formula (8.1.4) in
such a way that the entire neighborhood fits in the interval ] − π/2, π/2[. Taking
into account (8.1.2), (8.1.3), and (8.1.5), one gets the estimation
|xn − x0 |
| tan xn − tan x0 | < . (8.1.6)
cos (|x0 |/2 + π/4) cos x0
in other words the tangent function is continuous at any point x0 lying in the interval
] − π/2, π/2[.
Below, we are going to prove once again the continuity of this function, now
using Cauchy’s definition (8.0.2). Let us then choose a certain very small > 0.
Again the estimate (8.1.6) found above can be used to obtain
|x − x0 |
| tan x − tan x0 | < . (8.1.8)
cos (|x0 |/2 + π/4) cos x0
Remember that for the correctness of (8.1.6), it was only needed for xn to be very
near x0 (see (8.1.4)). Therefore, in order to be able to take advantage of it now, the
same must be required from x:
π |x0 |
|x − x0 | < − . (8.1.9)
4 2
Now, to the values of and x0 , a suitably small δ > 0 should be assigned. It
is worth noticing here that when, in the rest of this chapter, one is speaking of the
uniform continuity in a certain range, this δ will have to be universal for all x0 ’s in
the interval, and will not be allowed to change from point to point. However, in this
exercise we are dealing with the simple continuity, so the expression dependent on
x0 may be used. There is still a question of a specific form of δ. One can accept, for
example, δ = cos (|x0 |/2 + π/4) cos x0 , but remembering the condition (8.1.9),
one should rather write
8.1 Demonstrating the Continuity of Functions with Heine’s and Cauchy’s. . . 163
|x0 | π π |x0 |
δ = min cos + cos x0 , −
2 4 4 2
|x0 | π
≤ cos + cos x0 . (8.1.10)
2 4
(The smallest number of a certain numerical set is of course not larger than any
number from among them.) Such a choice will ensure the correctness of the
inequality:
δ cos (|x0 |/2 + π/4) cos x0
| tan x − tan x0 | < ≤ = ,
cos (|x0 |/2 + π/4) cos x0 cos (|x0 |/2 + π/4) cos x0
(8.1.11)
and at the same time the fulfillment of the definition (8.0.2).
Problem 2
It will be proved that the function f (x) = log x is continuous for any x ∈]0, ∞[.
Solution
Thanks to the detailed solution of the previous example, it will be relatively easy
for us to cope with the current problem. We start with demonstrating the continuity
of the function using Heine’s definition. Then one has a certain x0 ∈]0, ∞[ and
chooses any sequence of arguments xn convergent to x0 . For our expression to be
meaningful, it is assumed that all terms xn also belong to the range ]0, ∞[. Now one
needs to find the value of log xn − log x0 for n → ∞. Let us then write
xn xn xn − x0
| log xn − log x0 | = log = log 1 + − 1 = log 1 + .
x0 x0 x0
(8.1.12)
To estimate this expression, the inequality (13.2.2) is used:
Does it mean that one can insert modules on both sides of (8.1.13) to obtain
xn − x0 xn − x0
| log(1 + x)| ≤ |x|, or log 1 +
x0 ≤ x ? (8.1.15)
0
Clearly, this is only possible for x > 0, when both sides of (8.1.13) are positive, i.e.,
for xn > x0 . When −1 < x < 0, one has to propose another estimate for log(1 + x),
naturally from below. It can be obtained from (8.1.13) by writing
1 1+x−x
log(1 + x) = − log = − log (8.1.16)
1+x 1+x
x −x x
= − log 1 − ≥− = ,
1+x 1+x 1+x
noting that
x
− > −1.
1+x
Thus, it appears that
x
| log(1 + x)| ≤ , (8.1.17)
1+x
since both sides of the inequality (8.1.16) are now negative. One has, therefore, the
estimation
⎧
⎪ xn − x0
⎪
⎪ x , when xn > x0 ,
⎪
⎪
⎪
⎨ 0
If certain > 0 has already been chosen, it is enough now to define δ = /(1+)·x0 ,
and from the assumption that |x − x0 | < δ one concludes that | log x − log x0 | < :
⎧
⎪
⎪
δ
=
·
x0
=
⎪
⎪ < ,
⎪
⎪ x0 1 + x0 1+
⎪
⎨
| log x − log x0 | < δ δ x0 (8.1.21)
⎪
⎪ < = ·
⎪
⎪ x x0 − δ 1 + x0 − /(1 + ) · x0
⎪
⎪ x0
⎪
⎩ = = .
x0
Problem 1
Solution
The procedure for solving problems where the continuity of functions with one or
more “gluing” points is studied consists, in general, of two steps. First, one verifies
that inside the intervals on which the domain breaks, the function is continuous.
Within each of the intervals the function is defined with a different formula, so each
needs to be considered individually. With some experience, this first step may be
reduced to a minimum if one uses the knowledge that polynomials, trigonometric
functions, power functions, exponential functions, logarithms, etc. are continuous
within their domains of definiteness. Recalling this fact, one can go directly to the
second step, which is the study of “gluing” points themselves.
This first example will, however, be solved from the beginning to the end, and
in the next one, we will assume that the continuity of the different elementary
functions is known. The two examples of such argumentation have already been
given (Problems 1 and 2 in the previous section).
166 8 Examining Continuity and Uniform Continuity of Functions
The domain of the function (8.2.1) falls into three intervals: ] − ∞, −1[, [−1, 1],
and ]1, ∞[. Each of them will be analyzed separately.
• x < −1. In this range, the function is given by the formula f (x) = ax +b, which
describes a simple linear function. Its continuity is not in doubt, but nevertheless
let us take some small > 0 and, using Cauchy’s definition (8.0.2), write
For the last inequality in (8.2.3) to hold, this time δ = 2/π has been chosen.
We come again to the conclusion that the function f is continuous inside the
interval ] − 1, 1[.
• x > 1. Here, f (x) = c(x − 2)2 , i.e., one is dealing with a polynomial anew. Its
continuity, as in the first interval, can easily be proved:
Unless we are talking about uniform continuity, δ may be adjusted both to and
to x0 . Since we expect
we have to solve this equation for δ (one can assume that c = 0, otherwise
|f (x) − f (x0 )| = 0 and the desired result emerges immediately). One of its
solutions is negative, so it must be rejected, and the other, positive, becomes:
(
δ= |x0 − 2|2 + − |x0 − 2|. (8.2.6)
|c|
This choice of δ leads to the expected result |f (x) − f (x0 )| < , and the
continuity of f inside the last interval is demonstrated regardless of c.
Now one can move on to the second step, which is to examine of the continuity
at the “gluing” points: x = −1 and x = 1. The continuity means that the limit of
a function at a given point is equal to its value. These values can be read off the
central line in (8.2.1): f (−1) = sin (−π/2) = −1, and f (1) = sin(π/2) = 1. It
remains to see if one gets the same, calculating the limits of the function at these
points. Of course, one must first check whether these limits do exist.
At this point it is necessary to pay attention to a certain circumstance. While
proving the continuity of functions within each interval separately, nowhere did we
make use of the fact that x0 is restricted to a given interval. Each of the three above
proofs would be equally correct for any x0 ∈ R. There is nothing surprising because
polynomials and sine functions are continuous on the entire R. But since these
functions are continuous, the limits calculated for them at each point do exist and are
equal to the function’s values. The whole process then boils down to comparing the
function’s values obtained from formulas applicable below and above each “gluing”
point. This is equivalent to comparing the left- and right-side limits. For x = −1,
one has
In conclusion, it has to be noted that when both these conditions are satisfied,
the function f is continuous on the entire real axis. However, if one or both are
not met, the function at the appropriate point or points is discontinuous. Apart from
arguments ±1, the function is always continuous, regardless of the parameters.
Problem 2
Solution
As announced in the previous exercise, this time we are not going to examine in
detail the continuity in the intervals beyond the “gluing” points referring to known
facts about continuity of elementary functions. Everywhere, outside x = ±π/2, the
following formula for the function f holds:
sin2 x − 1
f (x) = . (8.2.10)
(4x 2 − π 2 )2
We now use the fact that the limit of a product is the product of limits of independent
factors (if all these limits exist). The limit of each fraction may then be bounded
separately. First let us examine the simpler case:
sin x − 1 −1 − 1 2
lim = = − 2. (8.2.12)
x→−π/2 (x − π/2) 2 (−π/2 − π/2) 2 π
The formula (7.1.9) has been used here. As a result, one has
1 2 1 1
lim f (x) = · − 2 · =− , (8.2.14)
x→−π/2 16 π 2 16π 2
so for the function f to be continuous at −π/2, the following condition must hold:
1
α=− .
16π 2
As to the point x = π/2, the calculations look very similar with the difference
that t = x−π/2. We are going to proceed with our calculation to the end, although it
could be guessed at this stage that the continuity will require satisfying the condition
β = −1/(16π 2 ) because of the parity of the upper expression in (8.2.9) in the
variable x.
and further
sin x + 1 1+1 2
lim = = 2, (8.2.16)
x→π/2 (x + π/2) 2 (π/2 + π/2)2 π
170 8 Examining Continuity and Uniform Continuity of Functions
which leads to
sin x − 1 sin (t + π/2) − 1
lim = lim (8.2.17)
x→π/2 (x − π/2)2 t=x−π/2 t→0 t2
sin t cos(π/2) + cos t sin(π/2) − 1
= lim
t→0 t2
0 · sin t + 1 · cos t − 1 1 − cos t 1
= lim = − lim =− .
t→0 t2 t→0 t2 2
Collecting all the results, one can see that in order that function f be continuous
at the point π/2, the parameter β must be equal to
1 2 1 1
β= · · − =− .
16 π 2 2 16π 2
Problem 3
Solution
One could, at the beginning, ask a question, why this exercise is placed in the section
concerning the “gluing” points, since such points cannot be seen in (8.2.18). Thanks
to a more detailed examination of this formula, we realize that the presence of the
limit with respect to n leads to different formulas for the function f for various
values of x, i.e., the function could be given as a piecewise function, as in the two
previous examples.
Now, let us investigate the various interesting values of a.
• −1 < a ≤ 1. For these values, the problem of eventual zero in the denominator
does not appear because always | sin x| ≤ 1 (and sin x = −1 because of
the domain). One can easily go to the limit in the formula (8.2.18), obtaining
sinn x −→ 0 wherever | sin x| < 1 and sinn x −→ 1, when sin x = 1:
n→∞ n→∞
⎧
⎪ π
⎨ x for x = + kπ , where k ∈ Z,
f (x) = 2 (8.2.19)
⎪ 1 π
⎩ x for x = + 2kπ , where k ∈ Z.
1+a 2
8.2 Examining Functions in Their “Gluing” Points 171
Now the formula (8.2.19) can be analyzed as we did in the previous problems.
Apart from points x = π/2 + kπ , the function is a polynomial of the first degree
and consequently it is continuous everywhere. It is then sufficient to consider
only the “gluing” points: xk = π/2+2kπ and compare the values of the function
with the limits at these points. From the lower formula of (8.2.19), one has
1 1 π
f (xk ) = xk = + 2kπ , (8.2.20)
1+a 1+a 2
1 π π π
1
+ 2kπ = + 2kπ ⇒ + 2kπ 1−
1+a 2 2 2 1+a
π a
= + 2kπ = 0 ⇒ a = 0.
2 1+a
(8.2.22)
1
1 + a sinn x = 0, that is sinn x = − . (8.2.23)
a
The function sinn x for each n ∈ N is continuous and assumes values in the
interval ] − 1, 1] (remember that x = −π/2 + 2kπ does not belong to
D). And, |1/a| < 1, so one can always find an argument x for which the
equation (8.2.23) is satisfied, i.e., the denominator of our function vanishes.
However, this issue ought to be approached in a different way. If we wonder
whether the point x0 may belong to the domain, we must first fix the argument
setting x = x0 , and only then can we examine the existence of the limit with n
going to infinity. One is not allowed to proceed similarly as while searching for
possible solutions of the equation (8.2.23) where the argument x was adjusted
for each value of n.
Let us then consider some specific x0 and consider the limit when n → ∞.
The expression a sinn x0 decreases to zero regardless of how big a is (the case
172 8 Examining Continuity and Uniform Continuity of Functions
sin x0 = 1 may be ruled out because then the equation (8.2.23) cannot be satisfied
and one directly sees that the function is well defined). Since a sinn x0 −→ 0,
n→∞
starting from a certain value n, one has for sure |a sinn x0 | < 1. For such n, the
denominator (8.2.23) cannot be equal to 0 and our expression makes sense. It
may be ill-defined for a certain specific finite value n, but for the existence of this
limit is not important and that is what we understand by the expression (8.2.18).
One can simply say that, for each x0 , out of the terms of the sequence
x
,
1 + a sinn x
just this one that is not defined at this point is thrown away. Always, if this kind
of doubt arises, one has to think how to understand a given expression so that it
makes sense.
After these considerations, one sees that for |a| > 1, the formula (8.2.19)
is applicable, and thus, the conclusions are too. The function is continuous at
the “gluing” points only for a = 0, which conflicts with the relevant condition
|a| > 1.
In conclusion, one can write that for a = 0 the function is continuous at all points
of its domain, and for a = 0 (and naturally a = −1), it is discontinuous at the points
xk = π/2 + 2kπ .
Problem 4
Solution
As a supplement to the text of the exercise, let us note that for all integers, i.e.,
numbers of the form m/1, one already deals with irreducible fractions with k = 1.
In that case, for integer arguments (assume that also for 0), the function takes the
value equal to 1. Let us also add that this function should not be confused with the
famous Riemann zeta function.
8.2 Examining Functions in Their “Gluing” Points 173
The problem concerning the continuity of the Riemann function belongs to this
type of exercise that at first sight seem to be difficult but after some reflection and the
precise application of the continuity definition turn out to be very easy. Especially
useful will be Heine’s definition (8.0.1), which says that a function is continuous
at a given point x0 ∈ D if and only if, for each sequence of arguments xn ∈ D
convergent to x0 , the function values converge to f (x0 ).
When looking at (8.2.24), one can see differences between the values assumed
by the function for irrational arguments (f (x) = 0), and those for rational ones
(f (x) = 0). This allows us to immediately resolve the problem of continuity at
rational points. We have simply to consider if, for a given rational number q, one
can construct a sequence of irrational numbers xn , convergent to q. If it proved to
be possible, the sequence of values would be a constant one: f (xn ) = 0 and of
course then one would have f (xn ) −→ 0, i.e., the sequence could not converge
n→∞
to f (q) = 0.
Finding such a sequence is straightforward. Consider, for example
1√
xn = q + 2. (8.2.25)
n
√
The argument xn is here a sum of a rational (i.e., q) and irrational (i.e., 1/n · 2)
number. Such a sum is obviously irrational and simultaneously convergent to q. This
observation implies that the Riemann function is not continuous for any rational
argument.
Now let us assume that x0 ∈ R \ Q. Since the number x0 is irrational, then
f (x0 ) = 0. Consider any sequence xn −→ x0 . Its terms may be irrational or
n→∞
rational. These cases will be processed separately.
• If a given term xn ∈ R \ Q, then f (xn ) = 0, which implies immediately that
|f (xn ) − f (x0 )| = 0.
• If a given term xn ∈ Q, i.e., xn = mn /kn (where the fraction is already
irreducible), then f (xn ) = 1/kn . So one has |f (xn ) − f (x0 )| = 1/kn .
Thus, if one wants to show that for each sequence xn one has f (xn ) −→ 0, it
n→∞
is enough to justify it for rational numbers. In this way our conclusions will hold
for both the purely rational sequence, and for that of mixed terms (the case of a
sequence of all or almost all irrational terms is obvious). As for any irrational term,
f (xn ) = 0 by definition, so assume xn = qn ∈ Q for n = 1, 2, . . . and qn −→ x0 .
n→∞
Now one needs to consider what form the fraction qn = mn /kn has when
n → ∞, i.e., when qn approaches the irrational number x0 . Naturally, there exists
a sequence of that kind and as an example one can take the subsequent digits of the
(infinite!) decimal expansion of the number x0 . For example, if our x0 were equal to
π , the first terms would be:
174 8 Examining Continuity and Uniform Continuity of Functions
3 31 314 157
q1 = 3 = , q2 = 3.1 = , =
q3 = 3.14 = ,
1 10 100 50
3141 31415 6283
q4 = 3.141 = , q5 = 3.1415 = = ,
1000 10000 2000
314159
q6 = 3.14159 = etc. (8.2.26)
100000
Since the sequence of numbers mn /kn convergent to x0 does exist and its limit can-
not be written as a m/k, the only possibility is that |mn |, kn −→ ∞. The approval
n→∞
of this fact requires only to become aware that inside an epsilon neighborhood of
x0 there can be only a finite number of fractions with denominators kn ≤ M, where
M is arbitrarily large but a fixed natural number. Taking appropriately small , one
can leave them all outside. As a consequence, inside the sphere K(x0 , ) there are
only fractions mn /kn with denominators greater than M. This is also visible in the
above example of the number π , where the value of kn equals , 10, 50, 1000, 2000,
100000, . . .. And this means that
1
lim |f (qn ) − f (x0 )| = lim = 0. (8.2.27)
n→∞ n→∞ kn
Problem 1
Solution
The definition of the uniform continuity has been formulated in the theoretical
summary. Since this notion is more difficult to understand (and to imagine) than the
point continuity, the definition is recalled below with some words of explanation:
1 2
x= , x = , (8.3.2)
nπ (2n + 1)π
x
2 2 1 2
5 3
-1
176 8 Examining Continuity and Uniform Continuity of Functions
since the expression 1/(n(2n + 1)) tends to zero for n → ∞. At the same time one
has
1
1 π
|f (x) − f (x )| = sin − sin = sin nπ − sin nπ +
x x 2
= |0 − (−1)n · 1| = 1 > . (8.3.4)
Problem 2
Solution
The behavior of the function f (x) is entirely different for distinct values of (integer)
exponent n, so one has to consider individual cases separately.
• n = 0. In this case, the function is always equal to 1, so the estimation does not
present any difficulty. One has of course
for any > 0 and for arbitrary x, x ∈]0, ∞[. In accordance with defini-
tion (8.3.1), this type of a function is uniformly continuous.
• n = 1. Now we are dealing with the linear function f (x) = x, for which:
|f (x) − f (x )| = |x − x |. (8.3.6)
Hence, if one chooses δ = , then for all arguments x and x , such that
|x − x | < δ, one has |f (x) − f (x )| < . One can then choose δ, depending
only on , so the function turns out to be uniformly continuous.
8.3 Investigating Whether a Function Is Uniformly Continuous 177
|f (x) − f (x )| = |x n − x n |. (8.3.7)
Now suppose that for a small one has succeeded to find δ > 0, common for all
x, such that for x, x satisfying |x − x | < δ, the inequality |x n − x n | < holds.
In particular, for x one can choose
x + δ/n.
k
1
n δ n δ
= x n−k > x n−1
k n k=1 1 n
k=1
δ
= nx n−1 = δx n−1 , (8.3.8)
n
where the inequality is due to the fact that all terms in the sum over k are positive,
but only one of them has been selected (that for k = 1). A glance at the right-hand
side of the obtained formula immediately explains that the function, in this case,
cannot be uniformly continuous. For any small δ the quantity |f (x) − f (x )| can
always be made bigger than by taking a very large x (since the power of x is
positive). The universal δ does not exist.
It is worth pointing out that uniform continuity could take place if we
considered the function defined not on the interval ]0, ∞[ but ]0, a], where a
is arbitrarily large, but fixed number. In this case, it would not be possible to
move to the right with x as far as we wished. It is important to note, however,
that in order to demonstrate the uniform continuity, one would not be allowed
to put x = x + δ/n, but one would have to demonstrate the veracity of
|f (x)−f (x )| < for all x in the neighborhood of x. To prove only the absence
of uniform continuity, it was enough to choose one example of x for which the
estimate was violated.
• n ≤ −1. We are going to argue that there is no uniform continuity, so following
the above considerations, we take
δ
x = x + .
2|n|
The presence of 2 in the denominator ensures the inequality |x − x | < δ even for
n = −1. In the previous section we had n ≥ 2 and there was no need to include
it. One has
178 8 Examining Continuity and Uniform Continuity of Functions
|f (x) − f (x )|
n 1 1 1 1
n
= x − x = −n − −n = |n| −
x (x ) x (x + δ/(2|n|)) |n|
k
|n| |n| δ
x |n|−k
− x |n|
(x + δ/(2|n|))|n| − x |n| k=0 k
2|n|
= |n| = |n|
x |n| (x + δ/(2|n|)) x |n| (x + δ/(2|n|))
|n|
k
|n| |n|−k δ |n| δ 1
x x |n|−1
k 2|n| 1 2|n|
k=1
= >
x |n| (x + δ/(2|n|))|n| k=1 x |n| (x + δ/(2|n|))|n|
δ
= . (8.3.9)
2x (x + δ/(2|n|))|n|
This last expression, for each fixed δ, can be made arbitrarily large (and hence
greater than ) by selecting appropriate small x. The function is not uniformly
continuous. It could possibly be so on the interval [a, ∞[ for a > 0 because one
would not be able to reach zero with x.
Problem 3
√
It will be proved that the function f (x) = x is uniformly continuous on the
interval ]0, ∞[.
Solution
As we know from the previous examples, one needs to estimate the quantity |f (x)−
f (x )| for |x − x | < δ. In the present problem, this estimate is very simple and does
not pose any difficulties.
√ √
√ √ √ √ | x + x |
|f (x) − f (x )| = | x − x | = | x − x | · √ √
| x + x |
√
|x − x | |x − x |
= √ √ = |x − x | · √ √ . (8.3.10)
x+ x x + x
Now we denote the larger of the numbers x and x with the symbol x+ and observe
that the last fraction is certainly smaller than unity:
8.4 Exercises for Independent Work 179
√
|x − x |
√ √ < 1, (8.3.11)
x + x
δ = 2, (8.3.13)
then from the inequality |x − x | < δ, one concludes that |f (x) − f (x )| < .
It should be stressed that the selected δ in (8.3.13) does not depend on x but only
on the previously fixed . This means that the function on the interval ]0, ∞[ is
uniformly continuous.
Answers
(a) The function is continuous everywhere if a − b = π/2 + 2kπ , where
k ∈ Z. When a + b = π/2 + 2kπ and b = 0, it is continuous at x = nπ
and discontinuous at x = (n + 1/2)π . For a + b = −π/2 + 2kπ , the
function is continuous at x = (n + 1/2)π and discontinuous at x = nπ .
(b) The function is continuous at x = 2 if b = −2a. It is never continuous
at x = −2. For x = ±2, the function is continuous irrespective of the
values of a and b.
(c) The function is continuous always except x = nπ . In addition, if a = 0
and b = 1, it is continuous also at “gluing” points.
Answers
(a) Function is not uniformly continuous.
(b) Function is uniformly continuous.
(c) Function is not uniformly continuous.
Chapter 9
Finding Derivatives of Functions
The main subject of the present chapter constitutes the notion of the differentiability
of functions. We will learn how to verify whether or not there exists a derivative of
a given function and we will find derivatives from the definition. Also, some less
trivial examples are considered.
Given a function f : R ⊂D → R continuous at a certain point x0 ∈ D and on its
neighborhood U (x0 , r) ⊂ D, a difference quotient is the quantity
f (x1 ) − f (x0 )
, (9.0.1)
x1 − x0
f (x0 + h) − f (x0 )
. (9.0.2)
h
If the limit
df f (x0 + h) − f (x0 )
f (x0 ) := := lim (9.0.3)
dx x0 h→0 h
exists, it is called the derivative of the function f at the point x0 . The function
is then called differentiable at x0 . The function differentiable at all points of its
domain D is simply a “differentiable function.” The geometrical meaning of f (x0 )
is the slope of the line tangent to the graph of the function at the point (x0 , f (x0 )).
From the above definition, the following rules for the differentiable functions
f (x) and g(x) can be derived:
• [c · f (x)] = c · f (x), where c is a constant.
• [f (x) ± g(x)] = f (x) ± g (x).
[f = f (x) · g(x) + f (x) · g (x).
• (x) ·g(x)]
f (x) f (x) · g(x) − f (x) · g (x)
• = , if g(x) = 0.
g(x) g(x)2
• [f (g(x))] = fg (g(x)) · g (x), where subscript g denotes the differentiation with
respect to the quantity g and not x. This formula bears the name of the chain
rule.
Given the derivative of a function f (x), the derivative of its inverse equals
−1 1
[f (x)] = . (9.0.4)
fy (y)
y=f −1 (x)
f (x0 + h) − f (x0 )
f− (x0 ) := lim , (9.0.5)
h→0− h
provided the function is continuous on some interval ]x0 − r, x0 ], where r > 0 (at
x0 only the left continuity is required) and the limit exists. Similarly for the right
derivative, one has
f (x0 + h) − f (x0 )
f+ (x0 ) := lim (9.0.6)
h→0+ h
Problem 1
√
The derivative of the function f (x) = 1 + x 2 for x ∈ R will be found.
9.1 Calculating Derivatives of Functions by Definition 183
Solution
f (x + h) − f (x) f (x + h) − f (x)
tan α = = . (9.1.1)
x+h−x h
This quantity is called the difference quotient. Note that if the function is increasing
(as shown) and h > 0, then f (x + h) > f (x) and the quotient (9.1.1) becomes
positive. This means that the secant points upwards. If, on the other hand, the
function is decreasing, one has f (x + h) < f (x), the angle α is negative together
with the value of the difference quotient.
We are not, however, interested in a secant line, but in the tangent to the graph at
the point (x, f (x)), shown in the figure with gray line intersecting the x axis at an
angle β. The derivative at this point, indicated by the symbol f (x), is simply tan β,
i.e., the slope of the tangent line.
A
C
x
x x h
184 9 Finding Derivatives of Functions
The closer the point B is to the point A, the more the secant line approaches the
tangent one. At the limit, when B → A, from the expression (9.1.1) the slope of the
tangent is obtained, i.e., the derivative:
f (x + h) − f (x)
f (x) := tan β = lim . (9.1.2)
h→0 h
Of course, it is possible that this limit does not exist, and then we will say that the
function is not differentiable at a given point. In this exercise, this kind of a problem
will not occur.
Let us then form the difference quotient:
√
f (x + h) − f (x) 1 + (x + h)2 − 1 + x 2
=
h h
√ √
1 + (x + h) − 1 + x
2 2 1 + (x + h)2 + 1 + x 2
= · √
h 1 + (x + h)2 + 1 + x 2
1 + (x + h)2 − 1 − x 2 2xh + h2
= √ = √
h( 1 + (x + h)2 + 1 + x 2 ) h( 1 + (x + h)2 + 1 + x 2 )
2x + h
= √ .
1 + (x + h)2 + 1 + x 2
(9.1.3)
Problem 2
The derivative of the function f (x) = a x for a > 0 and x ∈ R will be found.
Solution
a x+h − a x ah − 1
= ax , (9.1.5)
h h
9.2 Examining the Differentiability of a Function 185
it can be seen that the problem of calculating the derivative is essentially finding a
limit, very similar to that considered in Problem 2 of Sect. 7.2. Thus, one can use
the method applied there. First, we are going to introduce a new variable t = a h − 1,
where h = loga (t + 1), and then we convert the limit with respect to h into the limit
with respect to t:
ah − 1 t 1 1
lim = lim = lim = lim
h→0 h t→0 loga (1 + t) t→0 1/t · loga (1 + t) t→0 loga (1 + t)1/t
1 1
= =
. (9.1.6)
lim loga (1 + t) 1/t
t→0 loga lim (1 + t) 1/t
t→0
The continuity of the logarithmic function has been used here. The limit we obtained
is already known to us from Sect. 7.2:
ah − 1 1
lim = . (9.1.8)
h→0 h loga e
Using now the known formula for converting the base of a logarithm
1
logα β = , for α, β ∈ R+ \ {1}, (9.1.9)
logβ α
the formula for the required derivative at a point x has been obtained:
f (x) = a x = a x loge a = a x log a. (9.1.10)
Problem 1
Solution
f (0) = 0n = 0.
One has then a continuous function on the entire set R and its differentiability can be
tested. One could proceed at this point as in Exercises 2 and 3 in Sect. 8.2 and refer
to the known facts that the function x n (polynomial) is differentiable everywhere,
and that the function e−1/x is the composition of two differentiable functions (except
x = 0) so also differentiable (everywhere beside 0). Then, only the “gluing” point
would remain to be examined. That is how we will proceed in the next example,
but the present one is solved step-by-step. Let us consider the following individual
cases:
• x < 0. Here one has f (x) = x n . We create the difference quotient and check if
it has a limit:
9.2 Examining the Differentiability of a Function 187
f (x + h) − f (x) (x + h)n − x n
=
h h
x n + nx n−1 h + n(n − 1)/2 · x n−2 h2 + . . . + hn − x n
=
h
n(n − 1)
= nx n−1 + x n−2 h + . . . + hn−1
2
−→ nx n−1 + 0 + . . . + 0 = nx n−1 . (9.2.4)
h→0
When making these transformations, first Newton’s binomial formula (4.3.7) was
used, and then also the fact that the limit of a finite sum of expressions is equal to
the sum of their limits, provided they all exist. The result that has been obtained
not only proves the differentiability of the function x n , but also gives the formula
for its derivative:
t = eh/(x(x+h)) − 1 (9.2.7)
and replace the limit h → 0 with that t → 0. To this end, (9.2.7) must be solved
for h:
x 2 log(1 + t)
h= (9.2.8)
1 − x log(1 + t)
and then (9.2.7) and (9.2.8) must be substituted into the formula for the difference
quotient. Taking the limit, one finds
188 9 Finding Derivatives of Functions
f (x + h) − f (x) t
lim = e−1/x lim 2
h→0 h t→0 (x log(1 + t))/(1 − x log(1 + t))
where we made use of the continuity of the logarithmic function and a limit
already known (see (7.2.21)):
e−1/ h − 0 1 −1/ h
lim = lim e = lim t e−t = 0,
h→0+ h h→0+ h t=1/ h t→∞
hn − 0
lim = lim hn−1 = 0, (for n > 1), (9.2.12)
h→0− h h→0−
hn − 0 h
lim = lim = 1, (for n = 1).
h→0− h h→0− h
As one can see, the limit of the difference quotient (i.e., derivative) does exist
at this point for all n > 1. However, when n = 1, its limits on both sides are
different so the function is not differentiable at this point.
In this way, all cases have been investigated. For n > 1, the function (9.2.1) has
proved to be differentiable everywhere, but if n = 1 only for x ∈ R \ {0}.
9.2 Examining the Differentiability of a Function 189
Problem 2
Solution
We already know the procedure from the previous example. Solving the problem,
we start with taking a close look at the formulas in both intervals. For x < 0, the
function f (x) is the sum of hyperbolic functions which are known to be continuous
and differentiable. Therefore, it does not require any detailed examination here. In
turn, for x > 0, one is dealing with a polynomial, so again the function is continuous
and differentiable. It remains then to inspect the “gluing” point.
1. We verify the continuity of the function at x = 0.
The limits appearing here are similar to those already known for trigonometric
functions. It turns out that, for hyperbolic functions, they have the same values,
which is obvious for someone familiar with complex algebra and the resulting
connections between the functions sin and sinh or cos and cosh. However, in
order to demonstrate it without referring to this knowledge, one can write
so we have
a cosh h + b sinh h − a
lim = a · 0 + b · 1 = b. (9.2.19)
h→0− h
Both limits (the left and the right one) of the differential quotient must give
the same result for a differentiable function. This again leads to the condition
a = b.
In conclusion, it is found that if a = b, the function f is differentiable anywhere
besides at the point x = 0, and if a = b, it is differentiable on the entire set R.
9.3 Finding Derivatives of Inverse Functions 191
Problem 1
The derivative of the function f (x) = loga x will be found for x ∈]0, ∞[, a > 0,
and a = 1.
Solution
As we know, the logarithm is defined as the inverse operation with respect to the
exponentiation of the same base. With this fact, it may be relatively easy to find
its derivative. For each x lying in the image of a reversible function g, one has the
obvious identity:
for which the equality between the left-hand and the right-hand sides is most often
fulfilled only for particular values of x simply representing its solution, (9.3.2)
is satisfied for any x. Assuming that the functions f and f −1 are differentiable
wherever they are defined—and such is, in fact, the case of the exponential and
logarithmic functions—one can differentiate both sides of (9.3.2). In the case of
a normal equation, it would not be possible: the equality of two functions at a
particular point does not mean the equality between their derivatives. This is because
their graphs can intersect at a certain nonzero angle and tangents to them are inclined
differently.
Differentiating the second of the equations (9.3.2), we have
a f (x) = x = 1. (9.3.4)
The derivative on the left-hand side will be obtained with the use of the formula for
differentiation of a composite function:
where the symbol h g (g(x)) means differentiation over the argument g, and not
over x:
d
hg (g(x)) := h(y) . (9.3.6)
dy y=g(x)
1 1
a f (x) log a · f (x) = 1 ⇒ f (x) = [loga x] = , =
a f (x) log a
x log a
(9.3.7)
where we have used the known and already demonstrated (see (9.1.10)) formula:
d y
a = a y log a. (9.3.8)
dy
In particular, when a = e and one is dealing with the derivative of the natural
logarithm, we have
1 1 1
[loge x] = [log x] = = = . (9.3.9)
ef (x) log e x log e x
Problem 2
The derivative of the function f (x) = arcsin x will be found for x ∈] − 1, 1[.
Solution
The method discussed in the previous example will be used. The following obvious
identity
1 1
f (x) = [arcsin x] = = . (9.3.12)
cos f (x) cos(arcsin x)
9.3 Finding Derivatives of Inverse Functions 193
We have found the formula for the derivative, but it would not be elegant to leave
it in this form, which certainly can be simplified. Since one has
sin(arcsin x) = x, (9.3.13)
It remains only to decide which sign should stand in front of the square root: +
or −. We manage to establish it very easily if we remember that the domain of the
function arcsin x constitutes the interval [−1, 1] and the image is [−π/2, π/2]. In
this interval the cosine function is nonnegative, so one has to choose “+.” Thereby,
one has
1
f (x) = [arcsin x] = √ , (9.3.15)
1 − x2
for x ∈] − 1, 1[.
Problem 3
The derivative of the function f (x) = arctanh x will be found for x ∈] − 1, 1[.
Solution
and differentiating both sides over x. Let us recall that the derivative of the
hyperbolic tangent is
1
[tanh x] = . (9.3.17)
cosh2 x
This leads to the equation
1
2
· f (x) = x = 1 ⇒ f (x) = cosh2 f (x). (9.3.18)
cosh f (x)
194 9 Finding Derivatives of Functions
to be inserted into (9.3.18). In that way, one obtains the formula that has been looked
for:
1
f (x) = [arctanh x] = . (9.3.22)
1 − x2
It is worth mentioning that this derivative can be found equally well in another
way by using the fact that the function arctanh x can be simply expressed through
the natural logarithm. The equation (9.3.16) can be given in the form
and then one can get the explicit formula for f (x):
Of course, the expression under the logarithm must be positive and it actually is,
since −1 < x < 1. This fact is in turn due to the following estimate for (9.3.23):
ef (x) − e−f (x) ef (x) + e−f (x)
|x| = f (x) < = 1, (9.3.25)
e + e−f (x) ef (x) + e−f (x)
9.4 Solving Several Intricate Problems 195
and is a well-known property of the function tanh. This inequality is indeed strict,
as expression e±f (x) is positive.
Now it remains to calculate the derivative of (9.3.24) according to the rules
applicable to the differentiation of a composite function:
1 1 1 · (1 − x) − (1 + x) · (−1)
f (x) = · ·
2 (1 + x)/(1 − x) (1 − x)2
1 1−x 2 1
= · · = . (9.3.26)
2 1 + x (1 − x)2 1 − x2
As one can see, again the same result has been obtained.
Problem 1
Solution
In this section, we are going to learn how to differentiate more complex functions
with the assumption that we already know the derivatives of various elementary
functions. The essence of the solution consists always of reducing the problem to
the “prime factors” which, treated individually, tend to be very simple.
In the present example, one has such a case: a multiply composed function. This
composition can be seen in the following diagram:
h 1 g 1 f
x −→ −→ sin −→ 5 sin(1/x) . (9.4.1)
x x
It corresponds to a doubly composed function, i.e., to the expression
f (g(h(x))), (9.4.2)
The symbols appearing above have been defined previously, so this expression
should be clear (see (9.3.6)). Now the formula (9.4.3) is applied to our example.
In this way, we obtain
1 1 1
5 sin(1/x)
=5 sin(1/x)
log 5· sin =5 sin(1/x)
log 5·cos · − 2 , (9.4.4)
x x x
where we have used the well-known elementary derivatives:
x 1 1
a = a x log a, (sin x) = cos x, = − 2.
x x
Then, the exercise was simplified for the use of known derivatives of the
exponential function with base 5, of the sine function and of 1/x and for the twofold
application of the rule (9.3.5).
Problem 2
The derivative of the function f (x) = logx cos(π x/2) will be found for x ∈]0, 1[.
Solution
The function, dealt with in this example, cannot be differentiated according to the
rule (9.3.5), since this is not a usual composite function. One cannot directly use
formulas for a derivative of a logarithm with a variable base either, i.e., [loga x]
because now a also depends on the argument x, over which one should differentiate.
If one, notwithstanding, wanted to write an expression for the difference quotient of
logh(x) g(x), it would have to take the form
Problem 3
x)
The derivative of the function f (x) = x sin(x will be found for x > 0.
Solution
The present problem is, in a sense, similar to the previous one. An expression of
the kind x x can be differentiated neither as a x nor as x a because in the difference
quotient one has to perform simultaneous shifts of the base and of the exponent:
(x + h)x+h − x x
. (9.4.11)
h
198 9 Finding Derivatives of Functions
Let us now differentiate the left-hand side, using the formula for a composite
function (9.3.5):
1
[log(x x )] = · [x x ] , (9.4.13)
xx
and then the right-hand side of (9.4.12):
1
[log(x x )] = [x log x] = 1 · log x + x · = log x + 1. (9.4.14)
x
We come to the formula, which will turn out to be useful while further resolving
this exercise (in equation (9.4.17)),
Now let us return to the function given in the text of the exercise and use again the
method of taking a logarithm in order to convert the second power into the product:
x
log(f (x)) = log(x sin(x ) ) = sin(x x ) · log x. (9.4.16)
The product of two functions has been obtained: the composite function, the deriva-
tive of which may be calculated using (9.3.5), and the logarithm whose derivative
is well known. Differentiating (separately) both sides of the equation (9.4.16), one
gets
# $ 1
log(f (x)) = · f (x), (9.4.17)
f (x)
# $
sin(x x ) log x = [sin(x x )] · log x + sin(x x ) · [log x]
1
= cos(x x ) · [x x ] · log x + sin(x x ) ·
x
1
= cos(x x )x x (log x + 1) log x + sin(x x ) .
x
The comparison of both these formulas gives the final result:
x 1
f (x) = x sin(x ) = x sin(x ) cos(x x )x x (log x + 1) log x + sin(x x )
x
.
x
(9.4.18)
9.5 Exercises for Independent Work 199
Answers
(a) f (x) = cos x, g (x) = − sin x.
(b) f (x) = cosh x, g (x) = sinh x .
(c) f (x) = −2/x 3 .
(d) f (x) = 1/x.
(e) f (x) = 1/(x 2 + 1).
Answers
(a) The function is differentiable apart from the√point x = π/2 √ regardless
of parameter values. In addition, for a = − 2/2 and b = − 2/4 it is
differentiable also at x = π/2.
(b) Apart from the point x = 1, the function is always differentiable. In
addition, when a = 1/2 and b = π/4 − 1/2, it is differentiable also at
x = 1.
(c) Apart from the point x = 1, the function is always differentiable. In
addition, when a = 3 and b = 1/e, it is differentiable also at x = 1.
200 9 Finding Derivatives of Functions
Answers
√
(a) f (x) = −1/ 1 − x 2 .
(b) f (x) = 1/(x 2√+ 1).
(c) f (x) = 1/(n x n−1 ).
n
Answers
√
(a) f (x) = 1/ x 2 + 1 .
(b) f (x) = cos x log(tan x) + (tan x)sin x / cos x .
(c) f (x) = (2x 2 + 1)/(x(log x + x 2 )2 + 1) .
Chapter 10
Using Derivatives to Study Certain
Properties of Functions
The derivative of a function embodies the powerful tool for the investigation of
function properties. Some ideas are dealt with in the present chapter, but this subject
will continue in Chap. 12, where higher derivatives come into play.
The following properties of a differentiable function f : R ⊂]a, b[ −→ R can be
easily inferred from its derivative:
• If f (x) > 0 for x ∈]a, b[, the function f is increasing in ]a,b[.
• If f (x) < 0 for x ∈]a, b[, the function f is decreasing in ]a,b[.
• If f (x) = 0 for x ∈]a, b[, the function f is constant in ]a,b[.
Rolle’s theorem says that if a function f : R ⊂D −→ R has the following
properties:
1. is continuous on a closed interval [a, b] ⊂ D,
2. is differentiable on ]a, b[,
3. f (a) = f (b),
then there exists at least one c ∈]a, b[ for which f (c) = 0.
The Lagrange’s theorem also called the mean value theorem is a generaliza-
tion of Rolle’s theorem. It states that under the same assumptions as above, except
the third one, there exists at least one c ∈]a, b[ for which
f (b) − f (a)
f (c) = . (10.0.1)
b−a
From this theorem, Rolle’s theorem can be easily derived upon setting f (a) = f (b).
L’Hospital’s rule, which is a helpful tool when investigating limits of functions
of the special kinds discussed in detail in the specific problems below, has the
following form. Given two real functions f (x) and g(x) defined in a deleted
neighborhood S of a point x0 . If
then
f (x) f (x)
lim = lim . (10.0.2)
x→x0 g(x) x→x0 g (x)
This theorem may be applied also in the case when functions in the numerator and
denominator diverge to infinity instead of going to zero and for the improper limits
x → ±∞.
Problem 1
1−x π
arctan x + arctan = . (10.1.1)
1+x 4
Solution
1−x
f (x) = arctan x + arctan . (10.1.2)
1+x
The identity, which is to be proved, simply states that the function f within the
interval ] − 1, ∞[ is constant. Its graph must, therefore, be a horizontal line. The
derivative of such a function equals zero, because the tangent to the graph is
horizontal too. This observation suggests a certain possible way of solution: let us
10.1 Proving Identities and Inequalities 203
find f (x) and examine whether it is in fact equal to zero. We already know that the
derivative of the function arctan has the form:
1
[arctan x] = , (10.1.3)
1 + x2
so, after having differentiated f (x), one gets only an algebraic expression without
any trace of cyclometric functions. Therefore, it should be actually easy to check if
f (x) = 0. Thus, one has
1−x
f (x) = [arctan x] + arctan
1+x
1 1 1−x
= + ·
1 + x2 1 + ((1 − x)/(1 + x))2 1+x
1 1 −1 · (1 + x) − (1 − x) · 1
= + ·
1+x 2 1 + (1 − x) /(1 + x)
2 2 (1 + x)2
1 1 −2
= + · (10.1.4)
1 + x2 1 + (1 − x)2 /(1 + x)2 (1 + x)2
1 −2
= +
1+x 2 (1 + x) + (1 − x)2
2
1 −2 1 −2
= + = +
1 + x2 1 + 2x + x 2 + 1 − 2x + x 2 1 + x2 2 + 2x 2
1 1
= − = 0,
1 + x2 1 + x2
where, in the first step, the formula for the derivative of composite functions was
used (see (9.3.5)). Actually one has obtained zero, so the differentiated function
equals a constant:
1−x
f (x) = arctan x + arctan =C for x ∈] − 1, ∞[. (10.1.5)
1+x
It remains only to check whether this constant really equals π/4, as it is given in the
formula (10.1.1). To do this, it is sufficient to pick one specific point x0 , such that
−1 < x0 < ∞, for which we know how to easily calculate f (x0 ). The easiest is to
simply choose x0 = 0:
1−0 π π
C = f (0) = arctan 0 + arctan = 0 + arctan 1 = 0 + = . (10.1.6)
1+0 4 4
The proof of the identity (10.1.1) then has been completed. Since the value of the
(constant) function f at zero equals π/4, then it must be so at all other points in the
interval ] − 1, ∞[.
204 10 Using Derivatives to Study Certain Properties of Functions
At this place, there appears an interesting and important question. What would
be the values of the function f (x), if it was determined not only on ] − 1, ∞[, but
wherever the formula (10.1.1) makes sense, i.e., on D = R \ {−1}? Within the
interval ] − ∞, −1[, the calculation of a derivative is the same as above, so again it
turns out to be equal to zero. Therefore, one has:
Is one now allowed to still write that f (x) = π/4? The answer is negative. On
the basis of our former considerations, we may only conclude that the function is
constant on each of the two intervals separately and not on their sum. Our function
has, therefore, the form:
f (x) = C1 for x ∈] − ∞, −1[,
f (x) = π (10.1.8)
f (x) = C2 = for x ∈] − 1, ∞[.
4
Constants C1 and C2 can, but do not have to be the same. It can be verified by
determining explicitly the constant C1 . The other one, C2 , is already known. Which
point −∞ < x1 < −1 is the best to choose in order to find C1 ? Naturally, the one
for which it is easy to calculate f (x1 ). Unfortunately, at first glance, such a point is
not visible. So, instead of fixing C1 based on a specific point, one can equally well
examine the limit of both sides of the equation:
1−x
arctan x + arctan = C1 , (10.1.9)
1+x
when x → −∞. Since (10.1.9) is an identity for x < −1, the equality of limits
must also take place:
1−x
lim arctan x + arctan = lim C1 = C1 . (10.1.10)
x→−∞ 1+x x→−∞
As the limit of the sum is equal to the sum of all limits (since they all will prove to
exist), and since one can also apply the limit to the argument of the function arctan
(it is a continuous function), we obtain
π 1−x π π π 3π
− + arctan lim =− + arctan(−1) = − − = − = C1 .
2 x→−∞ 1 + x 2 2 4 4
(10.1.11)
As a result two identities are found:
1−x 3π
arctan x + arctan =− for x ∈] − ∞, −1[,
1+x 4
1−x π
arctan x + arctan = for x ∈] − 1, ∞[. (10.1.12)
1+x 4
10.1 Proving Identities and Inequalities 205
Finally, in accordance with what had been announced, we show how one can
independently demonstrate the identity (10.1.1), using relations between trigono-
metric or cyclometric functions. Well, in order to simplify the expression one can
use the known formula
a+b
arctan a + arctan b = arctan , (10.1.13)
1 − ab
which holds for a and b satisfying the condition ab < 1. (10.1.13) is a consequence
of the expression for the tangent function of the sum of angles:
tan α + tan β
tan(α + β) = . (10.1.14)
1 − tan α tan β
1−x x + (1 − x)/(1 + x)
f (x) = arctan x + arctan = arctan
1+x 1 − x(1 − x)/(1 + x)
(1 + x)x + 1 − x x + x2 + 1 − x
= arctan = arctan
1 + x − x(1 − x) 1 + x − x + x2
1 + x2 π
= arctan = arctan 1 = . (10.1.15)
1+x 2 4
One needs only to check whether the condition ab < 1 is met, which means here
1−x
x < 1. (10.1.16)
1+x
For x > −1, as in the text of the exercise, one can multiply both sides by (1 +
x), without inverting the inequality symbol, and in this way obtain an equivalent
inequality:
which is always satisfied. It should be stressed here that passing from (10.1.16)
to (10.1.17), we were allowed to write symbols of equivalence, and not only that of
implication, so our reasoning may be done in the reverse direction, starting from an
obvious inequality x 2 + 1 > 0 and leading to (10.1.16).
There remains a question, how to, if necessary, find values of the function f for
x < −1. This time the other inequality would be of use:
1−x
x > 1. (10.1.18)
1+x
206 10 Using Derivatives to Study Certain Properties of Functions
We would have ab > 1 and a < 0 so instead of (10.1.13) one could write:
a+b
arctan a + arctan b = −π + arctan . (10.1.19)
1 − ab
Problem 2
x n − nx + n − 1 ≥ 0 (10.1.20)
Solution
As in the previous problem, denote the left-hand side of the inequality with the
symbol f (x):
f (x) = x n − nx + n − 1 (10.1.21)
Consider now the sign of this derivative. Because n is even, and n − 1 odd, there
exists only one point at which the derivative equals 0. Namely,
x n−1 = 1, or x = 1. (10.1.23)
It should be noted that for odd n (i.e., for even n − 1) there would be two such points
(±1).
When x < 1, the derivative is negative—as seen from (10.1.22)—i.e., the
function f is decreasing, and when x > 1 the derivative is positive, i.e., the function
10.1 Proving Identities and Inequalities 207
is increasing. If the function to the left of 1 decreases and to the right it increases,
it is obvious that at x = 1 (which belongs to the domain), it assumes its minimal
value. One exception of this situation is possible, if the function is discontinuous at
this point. However, this is not the case here. The function (10.1.21) is ultimately a
polynomial, so it is not only continuous, but even differentiable on the entire R.
The conclusion is, therefore, that the function f for x = 1 actually takes its
smallest value:
f (1) = 1n − n · 1 + n − 1 = 1 − n + n − 1 = 0 (10.1.24)
Problem 3
is satisfied.
Solution
We are going to proceed in the same way as in the previous example. We start with
moving the whole expression onto the left-hand side and introduce the function f
as
π x
f (x) = tan x − (10.1.26)
4 π − 2x
defined on the interval [0, π/2[. We will try to show that the minimal value assumed
by this function is equal to 0. Since it is easy to find that
π 0
f (0) = tan 0 − = 0 − 0 = 0, (10.1.27)
4 π −2·0
our task will prove to be particularly easy, if one is able to demonstrate that on
the interval ]0, π/2[ the function is increasing (or at least not decreasing), i.e., its
derivative is positive (or at least nonnegative). In such a case, the function on the left
end of the interval would equal 0, and to the right of it, it could take only positive
(or at least nonnegative) values. Therefore, let us calculate:
208 10 Using Derivatives to Study Certain Properties of Functions
π 1 · (π − 2x) − x · (−2) π π − 2x + 2x
f (x) = − = −
4 cos2 x (π − 2x)2 4 cos2 x (π − 2x)2
π π π
= − = (π − 2x)2 − 4 cos2 x .
2
4 cos x (π − 2x) 2 4(π − 2x) cos x
2 2
(10.1.28)
As a result, one can conclude that f (x) > 0 in the considered interval. This, in
turn, entails the required inequality (10.1.25).
Problem 4
Solution
This exercise is seemingly unrelated to the subject of this section. However, as we
will see in a moment, this issue can easily be solved by a study of inequalities,
similar to Problems 2 and 3. Consider the following function, defined on the interval
]0, ∞]:
10.2 Using Rolle’s and Lagrange’s Theorems 209
ex
f (x) = . (10.1.31)
xe
One naturally has f (e) = ee /ee = 1. The question arises, what the value of this
function for x = π is. Formally f (π ) = eπ /π e , so if it is found to be greater than
1, then eπ > π e , and if smaller, then eπ < π e . At this point, instead of using our
calculator, our goal is to make use of differential calculus.
Since we know that f (e) = 1 and π > e, let us calculate the derivative of the
function f in order to check if, on the right of e, the function is monotonic:
[ex ] x e − ex [x e ] ex x e − ex e x e−1
f (x) = 2e
= (10.1.32)
x x 2e
ex x e 1 − xe ex e ex
= = 1 − = (x − e).
x 2e xe x x e+1
It is clear that for x > e, one has f (x) > 0. This in turn means that the function is
increasing there, so for x = π its value must be greater than one:
eπ ee
f (π ) > f (e) = 1 ⇐⇒ > = 1. (10.1.33)
πe ee
This result indicates that the following inequality is true:
eπ > π e . (10.1.34)
Problem 1
Solution
First, we are going to recall Rolle’s theorem, which was formally formulated at the
beginning of this chapter, because it will be helpful. This theorem states that if a
real function f is continuous on an interval [a, b] and differentiable in ]a, b[, then
inside this interval there exists a point c, for which
f (c) = 0. (10.2.2)
210 10 Using Derivatives to Study Certain Properties of Functions
Now suppose, contrary to the text of the exercise, that the polynomial wn (x) has
three (or more) roots. Let us denote them by x1 , x2 , x3 supposing that
x1 < x2 < x3 .
a = x1 , b = x2 , or a = x2 , b = x3 , (10.2.4)
the assumptions of Rolle’s theorem are met. Thereby, it can be concluded that wn (x)
must vanish at least twice: inside the interval ]x1 , x2 [ and inside the interval ]x2 , x3 [.
Let us calculate the derivative of the polynomial wn (x) and verify how many roots
it can possess.
In brackets we have the square trinomial, whose discriminant has the form:
Examining the roots of (10.2.6), it is easy to see that for n ≥ 3—and in the case
of our example—it is constantly negative. Consequently, the polynomial in square
brackets in (10.2.5) cannot vanish. This means that the only root of the polynomial
wn (x) is x = 0 coming from the factor x n−3 (where n > 3) or there are no roots
at all (if n = 3). Both, the former and the latter situations stay in contradiction with
Rolle’s theorem. Our initial assumption, that
thereby, must be wrong. Because of this the following conclusion emerges: wn (x)
can have at most two roots (if n > 3) or has exactly one (when n = 3): the
polynomial is then of odd degree, so unavoidably one root does exist.
In the first case, i.e., when n > 3, we are able to further establish that if these two
roots exist, they are of different signs. wn (x) vanishes then at x = 0 but from Rolle’s
theorem it follows that this must take place between x1 and x2 , so x1 < 0 < x2 .
10.2 Using Rolle’s and Lagrange’s Theorems 211
Problem 2
It will be proved that for any x > 0, the following inequality is satisfied:
x
< log(x + 1) < x. (10.2.7)
x+1
Solution
In order to solve this problem, one will need Lagrange’s theorem, which says that if
a real function f is continuous on [a, b] and differentiable in ]a, b[, then inside this
interval there exists a point c, for which
f (b) − f (a)
f (c) = . (10.2.8)
b−a
As one can see, this theorem is a generalization of Rolle’s theorem applied in
the previous problem, since in the particular case f (a) = f (b) one gets (10.2.2).
Below, it will be seen how it can be used to demonstrate the inequality (10.2.7).
In these types of exercises, the main difficulty lies in guessing what function
f to choose in order to obtain the desired inequality. However, if one looks at the
expressions in (10.2.7), one sees that the function log(x +1) appears, and in addition
on the left-hand side there is also the fraction 1/(x + 1), which is, after all, its
derivative. So the idea imposed in the first step is to try
1 log(b + 1) − log(a + 1) 1
< < . (10.2.12)
b+1 b−a a+1
212 10 Using Derivatives to Study Certain Properties of Functions
If we now put a = 0 and b = x > 0, then (10.2.12) will take the form
1 log(x + 1) − log(0 + 1) 1
< < , (10.2.13)
x+1 x−0 0+1
i.e.,
1 log(x + 1)
< < 1. (10.2.14)
x+1 x
After multiplying both sides by x (we remember that x > 0), one obtains the
inequalities (10.2.7), which were to be demonstrated.
Problem 3
It will be proved that for any a, b ∈ R such that b > a, the following inequalities
b−a b−a
< arctan b − arctan a < (10.2.15)
1+b 2 1 + a2
are satisfied.
Solution
In this exercise, there should not be any problem identifying the function f .
Immediately, it can be seen that in (10.2.15) there appears arctan x and its derivative
1/(1 + x 2 ). We, therefore, choose f (x) = arctan x. This function is obviously
continuous and differentiable on the entire R, so for any a and b such that a < b,
the assumptions of Lagrange’s theorem (10.2.8) are met. Thus one can write
1 arctan b − arctan a
∃c∈]a,b[ = . (10.2.16)
1+c 2 b−a
1 1 1
< < . (10.2.17)
1+b 2 1+c 2 1 + a2
1 arctan b − arctan a 1
< < . (10.2.18)
1 + b2 b−a 1 + a2
After multiplying both sides by the (positive) number (b − a), one gets the required
inequalities (10.2.15).
As one can see in the last two examples, proving these types of inequalities on the
basis of Lagrange’s theorem is generally very simple, and the entire problem reduces
to the proper guess of the function f . In the case of complicated expressions, finding
function f may become troublesome. It should not be expected that all of these
types of inequalities can be demonstrated with the use of Lagrange’s theorem. For
this reason, it is necessary to learn various methods, including those of Sect. 10.1
and those based on the convexity property of functions.
Problem 1
It will be proved that the curve y = eax sin bx for a, b > 0 is tangent to the curves
y = eax and y = −eax at all common points.
Solution
The geometric interpretation of the derivative—as a slope of the straight line tangent
to the graph of a function—can be applied to the fundamental properties of curves on
the plane. One can find angles of their intersection, and possibly determine whether
they are tangential—which constitutes the aim of this exercise.
Solving this problem will, therefore, be composed of the following three steps.
1. First one has to find all the common points of the curves
2. Then, in each of them, the values of derivatives for both curves must be
determined. If they turn out to be equal, the curves will be tangential.
3. Finally, the above procedure ought to be repeated for the second pair of curves
Of course, one must note that all curves in this problem are described by differ-
entiable expressions. In Fig. 10.1, the graphs are shown for a specific choice of
214 10 Using Derivatives to Study Certain Properties of Functions
4 ex
2 ex sin 4x
x
1 1 2
2
4
ex
Fig. 10.1 The curves from the text of the problem plotted with the exemplary values of parameters
a = 1, b = 4
parameters a and b. For the other (positive) values of a and b the general course
of these curves does not change.
Now consider the first pair of curves. Points in common have coordinates (x, y)
which are solutions of the equations
y = eax sin bx,
. (10.3.1)
y = eax
Since the exponential function is always positive, the only possibility to meet the
above equation is
π
sin bx = 1 ⇐⇒ bx = + 2nπ, where n ∈ Z. (10.3.3)
2
Labeling the solutions with a parameter n, we have
1 π
xn = + 2nπ . (10.3.4)
b 2
The solution of the system (10.3.1) formally requires also a suitable value yn . One
naturally has yn = eaxn , but this will not be needed in further parts of our work.
To check that the two curves are tangential at the designated points, one can now
calculate and compare both derivatives:
10.3 Examining Curves on a Plane: Tangency and Angles of Intersection 215
# ax $
e sin bx = a eax sin bx + b eax cos bx x=x
x=xn n
At xn , one has of course sin bxn = 1 and cos bxn = 0, and that gives
# ax $
e sin bx = a eaxn · 1 + b eaxn · 0 = aeaxn . (10.3.6)
x=xn
Since also
# $
eax = a eax x=x = a eaxn , (10.3.7)
x=xn n
the derivatives of both functions are identical at all common points of their graphs.
Thus, the results indicate that y = eax sin bx and y = eax actually describe
tangential curves.
For the second pair of functions, i.e., for y = eax sin bx and y = −eax , we have
a similar system of equations:
y = eax sin bx,
(10.3.8)
y = −eax ,
This entails
π
sin bx = −1 ⇐⇒ bx = − + 2nπ, where n ∈ Z. (10.3.10)
2
If, as before, the solutions are labeled with the parameter n and, to be distinguished
from the former ones, are given the symbol “tilde” over x, one has
1 π
x̃n = − + 2nπ . (10.3.11)
b 2
For completeness we note that ỹn = −ea x̃n . Now let us find and compare the
derivatives of both functions:
# ax $
e sin bx = a eax sin bx + b eax cos bx x=x̃
x=x̃n n
Both functions have identical derivatives at all common points of their graphs. The
conclusion is then that the curves y = eax sin bx and y = −eax are tangential to
each other at these points.
Problem 2
where C1 , C2 ∈ R and C12 + C22 = 0, are perpendicular to each other at all points of
intersection.
Solution
In Fig. 10.2, the two families of curves for several values of parameters C1 and C2
have been drawn, respectively, in black and gray. The figure shows that they actually
intersect at right angles. Below, it will be demonstrated in a strict, analytical way.
First, the common points for both curves (10.3.15) must be found with the
assumption that the parameters C1 and C2 are fixed and different from zero. This
also means that x = 0 and y = 0. The case when one of the constants C1 , C2
vanishes will be dealt with later. Thus we have
⎧ x
⎪
⎨ 2 = C1 ,
x + y2
−y (10.3.16)
⎪
⎩ 2 = C 2 ,
x + y2
from which it follows, after dividing these equations by each other, that
x C1
− = . (10.3.17)
y C2
Determining x (i.e., writing x = −y · C1 /C2 ) and inserting it into the second of the
equations (10.3.16), one obtains
−y −y
= C2 ⇐⇒ 2 2 = C2 . (10.3.18)
(−y · C1 /C2 ) 2
+ y2 C1 /C2 + 1 y 2
10.3 Examining Curves on a Plane: Tangency and Angles of Intersection 217
y
C2
x2 y2
x
C1
x2 y2
x
Then, we already have the points of intersection, and now let us calculate the
appropriate derivatives. It may be said that each of the equations (10.3.15) defines a
certain function y(x). Such functions are called implicit functions and there exist
appropriate theorems about their existence and differentiability, but these issues
will appear only in the second part of this book. Sometimes such equations can be
unraveled, i.e., the dependence y(x) can explicitly be found. This function can then
be differentiated over x and the slope of the tangent to the graph can be determined.
It could be done in this exercise too. Since the situation where x = 0 and y = 0 is
considered, the equations from (10.3.15) are equivalent to some quadratic equations
of y, which naturally could be solved. From the first equation, we get two solutions
(i.e., two functions), denoted by y1 (x) and y2 (x). From the second one, two other
functions are obtained: ỹ1 (x) and ỹ2 (x). Considering a specific point of intersection
of the graphs, we would quickly realize that only one of the functions yi has a graph
that intersects with only one function ỹj , and, therefore, it is always known which
pair of functions comes into play. After having unraveled the equations (10.3.15),
our exercise becomes very similar to the previous one.
218 10 Using Derivatives to Study Certain Properties of Functions
Not always, however, do equations of the type (10.3.15) allow to explicitly find
the function y(x). It is possible that they are genuinely implicit and then one has to
cope in some other way. In what follows, we are going to proceed as if we had to
deal with such a situation.
If the first of the equations (10.3.15) defines a certain (unknown) function y(x),
then after having inserted it back in the place of y, one must get the identity
x
= C1 . (10.3.20)
x 2 + y(x)2
y(x)2 − x 2
y(x)2 − x 2 − 2xy(x)y (x) = 0 ⇐⇒ y (x) = . (10.3.22)
2xy(x)
A question could arise here, whether, writing down (10.3.21), one may assume that
x 2 + y(x)2 = 0. This requires some comment. According to the text of the exercise,
C12 + C22 = 0 which means that x and y that constitute a solution of (10.3.15) cannot
simultaneously vanish. The function y(x) is considered only on the neighborhood of
this point (i.e., on the neighborhood of x constituting a solution). Since it has been
assumed that it is a differentiable function, it must also be continuous. The same,
therefore, refers to the expression x 2 + y(x)2 , and since it has a nonzero value for x
in question, then it cannot vanish at least on its certain small neighborhood too.
To find the derivative, i.e., the slope of the tangent at the point of intersection
(denoted by tan α), we simply place into (10.3.22) the values (10.3.19):
2 2
−C2 /(C12 + C22 ) − C1 /(C12 + C22 )
tan α = y (x)x=C
=
1 /(C1 +C2 )
2 2
2C1 /(C12 + C22 ) · (−C2 )/(C12 + C22 )
C12 − C22
= . (10.3.23)
2C1 C2
Please note that we have found the derivative at the point of intersection without
knowing explicitly the formula for the function! Now it should be repeated for the
second curve. We have the identity (naturally the symbol y(x) has a new meaning—
this function is defined now by the second of the equations (10.3.15)):
10.3 Examining Curves on a Plane: Tangency and Angles of Intersection 219
s(x)
= C2 . (10.3.24)
x 2 + y(x)2
2xy(x)
y (x)(x 2 − y(x)2 ) − 2xy(x) = 0 ⇐⇒ y (x) = − . (10.3.26)
y(x)2 − x 2
2C1 C2
tan β = y (x)x=C =− . (10.3.27)
1 /(C1 +C2 )
2 2
C12 − C22
1
tan α = − , i.e., tan α · tan β = −1. (10.3.28)
tan β
The inclinations of the curves, and hence the angles α and β too, are restricted
to the interval [−π/2, π/2], or rather ] − π/2, π/2[, since we are dealing with
differentiable functions for which the tangent line may not be vertical. In this case,
one can rewrite (10.3.28) in the form
sin α sin β
= −1 ⇐⇒ sin α sin β + cos α cos β = 0 ⇐⇒ cos(α − β) = 0,
cos α cos β
(10.3.29)
which shows that α − β = ±π/2, and, therefore, both curves intersect at a right
angle.
Now we have to come back to the omitted values of C1 and C2 .
• C1 = ∓C2 = 0. At the points of intersection, one has now y = ±x, which
is due to (10.3.17). The derivative (10.3.22) equals zero. This obviously implies
α = 0. The tangential line at each of these points is horizontal. In turn, the
220 10 Using Derivatives to Study Certain Properties of Functions
derivative (10.3.26) at these points does not exist and the same is true for the
function y(x) defined by second equation (10.3.15). However, we are able to
overcome this obstacle by treating this equation as one defining the function
x(y) instead of y(x)! The derivative x (y) := (d/dy(x(y)) naturally equals
zero, which can be concluded without any calculations because this equation
is analogous to (10.3.20)—they differ by the names of the variable only (x ↔ y)
and a constant on the right-hand side. The tangential line is, therefore, parallel to
the axis of the arguments, which this time is the y-axis and the angle β = ±π/2.
In consequence, both cures are perpendicular.
• C1 = 0, C2 = 0. Now the first curve is described by the equation x = 0 (except
for the point (0, 0) which is removed from this curve). Points of intersection with
the second curve are such for which x = 0 and
−y 1
= C2 , or y=− . (10.3.30)
02 + y 2 C2
Due to (10.3.26), one immediately sees that the derivative of the second curve
is equal to zero, and, therefore, its tangent is horizontal. It must be then
perpendicular to the line x = 0.
• C1 = 0, C2 = 0. This case does not need to be considered separately. In
comparison with the previous case, the two curves simply interchange their roles
and, therefore, intersect at right angles.
• C1 = C2 = 0. This case leads to contradictory equations and may be omitted.
At the end, it is worth to mention that when, in the course of studying analysis,
we are already accustomed with holomorphic functions (these are differentiable
functions of complex variable z), this exercise will become quite trivial. The two
equations given in the text will take the form of
where f (z) = 1/z is just such a holomorphic function (except z = 0). Such func-
tions have the property (which is easy to prove using the so-called Cauchy-Riemann
conditions) that curves defined by equations (10.3.31) are always perpendicular.
Problem 1
The limit
x − arcsin x
lim (10.4.1)
x→0 x3
will be found.
10.4 Calculating Limits Using l’Hospital’s Rule 221
Solution
In this section, we are going to learn how to find limits of functions that have the
form of a quotient f (x)/g(x) when both numerator and denominator tend to zero.
A tool to be used is the so-called l’Hospital’s rule, formulated in the theoretical
summary at the beginning of this chapter. It is assumed that the two real functions
f (x) and g(x) are defined in a certain deleted neighborhood S of a point x0 , and for
any x ∈ S, g(x) = 0. In addition, we require
It is also assumed that both functions are differentiable in this (deleted) neighbor-
hood, and for any x ∈ S, g (x) = 0. Then, if the limit
f (x)
lim (10.4.3)
x→x0 g (x)
f (x)
lim , (10.4.4)
x→x0 g(x)
does exist too, and both are equal. This rule also can be used when functions in
the numerator and denominator go to infinity instead of zero and when one deals
with improper limits x → ±∞ as well. As we will see in further examples of this
section, besides limits 0/0 and ∞/∞, l’Hospital’s rule turns out to be useful for
finding limits of the type ∞ − ∞, 00 and 1∞ .
Let us now look at the limit (10.4.1). In the numerator, we have the function
f (x) = x − arcsin x, and in the denominator g(x) = x 3 . For x → x0 = 0, both of
them go to zero and also both are differentiable in a deleted neighborhood of x0 . In
addition, the denominator does not vanish. In accordance with l’Hospital’s rule, one
can now calculate:
√
f (x) [x − arcsin x] 1 − 1/ 1 − x 2
lim = lim = lim . (10.4.5)
x→0 g (x) x→0 [x 3 ] x→0 3x 2
Performing this step, we have not explicitly found the value of the limit yet, but
still some success has been achieved. It consists of the fact that in the numerator,
the inverse sine function has been discarded, and now a purely algebraic expression
remains. In the denominator, in turn, we have managed to reduce the degree of zero
by 1 (from 3 to 2). The limit (10.4.5), however, still has the character typical for
l’Hospital’s rule, i.e., 0/0. Now there are two possible ways to proceed. One can
treat the new expression as another problem for the application of l’Hospital’s rule
with certain new functions:
222 10 Using Derivatives to Study Certain Properties of Functions
1
f˜(x) = 1 − √ , g̃(x) = 3x 2 , (10.4.6)
1 − x2
and differentiate the numerator and the denominator once again. The order of zero
in the denominator will be again reduced by one. If the rule was applied for the
third time—although this may be unnecessary—it would be found that the limit is
no longer of the kind 0/0 and should be able to calculate it explicitly.
The second way is to apply to the algebraical expression (10.4.5) some of
the “tricks” used in Sect. 7.1 and transform it into a form of a limit which we
have already learned how to calculate. A fairly standard method and known to us,
applicable in the case of the expression (10.4.5), is to get rid of the square root by
using the formula:
√
1 − x2 + 1 1 − x2 − 1 −x 2
1 − x2− 1 = ( 1 − x − 1) · √
2 =√ =√ .
1 − x2 + 1 1 − x2 + 1 1 − x2 + 1
(10.4.7)
Since this section is devoted to l’Hospital’s rule, we are going to exploit the first
method. Functions in the expression (10.4.5) meet the required assumptions, so one
has
√
f˜(x) f˜ (x) 1 − 1/ 1 − x 2 −x/ (1 − x 2 )3
lim = lim = lim = lim
x→0 g̃(x) x→0 g̃ (x) x→0 [3x 2 ] x→0 6x
−1/ (1 − x 2 )3 1
= lim =− . (10.4.8)
x→0 6 6
As one can see, thanks to having reduced x in the numerator and in the denominator,
further differentiation of both functions was not necessary. In general, however,
unless some extraordinary simplification occurs, one needs as many steps as the
order of zeros (in our case, both in the numerator and denominator, one had initially
zeros of the third degree).
At this point, one should pay attention to one issue: after each differential step,
it is necessary to look at the emerging expression and consider if there exists a
way to simplify it. The “mechanical” differentiation of functions without such a
preparation can cause—rather than simplification—significant complication of the
derived expressions. For example, applying the described method to the limit
1 − arcsin x/x
lim , (10.4.9)
x→0 x2
which is, after all, the same limit as (10.4.1), one would not be able to get rid
of arcsin and would not get any algebraical expression. It is clear that, before
differentiation, the expression has to be transformed in such a way that the
most troublesome function, i.e., arcsin, does not form a fraction. Identifying such
bothersome function (insofar as it is present in the expression) and separating it
10.4 Calculating Limits Using l’Hospital’s Rule 223
is, therefore, the first step in solving limits by this method. An example of such
reasoning appears in the next problem.
Going back to the present exercise, using (10.4.8), (10.4.5), and l’Hospital’s rule,
the final result is obtained:
x − arcsin x 1
lim =− . (10.4.10)
x→0 x3 6
Problem 2
The limit
Solution
Before starting to solve this example, one must note that l’Hospital’s rule formulated
in the previous problem is applicable to one-sided limits as well. Naturally, the
appearing derivatives will also be “one-sided,” and instead of a deleted neighbor-
hood of the point x0 , we will have to deal with an interval ]x0 , x0 + r[ or ]x0 − r, x0 [
for some r > 0. It should be noted also that (10.4.11) is neither a limit of the type
0/0 nor ∞/∞, but 0 · ∞, because
In order to be able to exploit the rule, one needs to transform our expression into
the required form. There are two possible ways. Either we rewrite (10.4.11) in the
form:
xm
lim x m logn x = lim (10.4.13)
x→0+ x→0+ 1/ logn x
and deal with the limit 0/0, or the other way round:
logn x
lim x m logn x = lim (10.4.14)
x→0+ x→0+ 1/x m
and we get ∞/∞. The important question arises at this point: which of these two
options should be chosen and what is the rationale for this? Maybe both ways are
224 10 Using Derivatives to Study Certain Properties of Functions
equally good? For, in both cases, the required assumptions (in the one-sided version)
are met, i.e., there is no formal reason to reject either (10.4.13) or (10.4.14).
The answer to this question is provided by the discussion that occurred at the end
of the previous problem: we must choose such an option that after having applied
l’Hospital’s rule, i.e., after differentiation (maybe repeated) of the numerator and
the denominator, we can get rid of the “unwanted,” i.e., the most “troublesome,”
function. In the previous exercise, such was undoubtedly the function arcsin x. Here,
if we are to choose between the logarithmic functions or the power one, certainly we
prefer the former. The calculation of limits of power expressions, generally, poses
no problems.
Let us now take a look at the first option, i.e., the formula (10.4.13). In the
denominator one has to differentiate a certain power of the inverse of logarithm:
1 −n 1
= · . (10.4.15)
logn x logn+1 x x
n logn−1 x
=− lim . (10.4.17)
m x→0+ 1/x m
10.4 Calculating Limits Using l’Hospital’s Rule 225
We have managed to reduce the power of the logarithm, and the exponent in the
denominator has not changed. In general, it is clear that in order to entirely get rid
of the logarithm, this procedure must be repeated n − 1 times. It is always necessary
to check whether the new appearing functions satisfy the assumptions, but it is quite
obvious here, since the nature of the functions does not change.
By following the steps above, one obtains
n
n − 1
1
1
lim x m logn x = − · − · ... · − · lim
x→0+ m m m x→0+ 1/x m
n!
= (−1)n lim x m = 0. (10.4.18)
mn x→0+
We have come to the well-known result that the power behavior dominates over the
logarithmic one regardless of the values n, m > 0. It is worth knowing that in a
similar way an exponential expression dominates over a power expression.
Problem 3
The limit
π
lim x − arctan x (10.4.19)
x→∞ 2
will be bound.
Solution
does not solve the problem, since differentiating the numerator and the denominator,
even repeatedly, one will not get rid of the “troublesome” function arctan. Therefore,
we must rather write
π π/2 − arctan x
x − arctan x = , (10.4.21)
2 1/x
226 10 Using Derivatives to Study Certain Properties of Functions
which leads to the limit 0/0, and then apply l’Hospital’s rule. The assumptions
of this theorem are clearly fulfilled, since both arctan x and 1/x are defined and
differentiable in the neighborhood of infinity, i.e., in an interval ]M, ∞[, where
M ∈ R+ . In addition, the denominator of (10.4.21) is different from zero and this
remains true after differentiation. Thus one has
Problem 4
The limit
1 1
lim − (10.4.23)
x→0 x(x + 1) log(x + 1)
will be found.
Solution
The above limit is still of the kind 0/0 and the assumptions are again satisfied.
Therefore, the successive identical step has to be performed and we come to the
final result:
−1/(x + 1)2 − 2 −1 − 2 3
= lim = =− .
x→0 2 log(x + 1) + (2x + 1)/(x + 1) + 1 2·0+1+1 2
Problem 5
The limit
will be found.
Solution
In this example, there appears yet another type of limit to which l’Hospital’s rule
can be applied. Since
the limit is of the kind 00 . The question arises, how can one transform it to the
form 0/0 or ∞/∞. Here, the logarithmic function is helpful because it allows one
to replace a power with a product. Instead of calculating the limit of a function
f (x), the limit of the expression log f (x) will be found. An important property of
the logarithm, which plays a crucial role in our reasoning, is its continuity, which
allows us to write the following equation:
lim log f (x) = log lim f (x) , (10.4.29)
x→0 x→0
provided f (x) > 0 in an interval surrounding point x = 0 and lim f (x) > 0. If
x→0
one then finds the limit of the left-hand side to be
The solution has just been outlined, and therefore, one can now proceed to its
realization. Since in our case f (x) = (sin2 x)1−cos x , we calculate
log(sin2 x)
lim log(sin2 x)1−cos x = lim . (10.4.34)
x→0 x→0 1/(1 − cos x)
The obtained limit still has a character of 0/0, but it is not reasonable to automati-
cally apply l’Hospital’s rule for the second time. There is no sense to differentiate
the whole expression in the numerator of (10.4.35) if the limit of the function cos x
in zero is well known and equals 1. It is more convenient to calculate separately
(1 − cos x)2
lim , (10.4.36)
x→0 sin2 x
and then make use of the fact that the limit of a product is equal to the product of
limits. The l’Hospital’s rule is then applied only to (10.4.36) with the result
10.4 Calculating Limits Using l’Hospital’s Rule 229
# $
(1 − cos x)2 (1 − cos x)2 2(1 − cos x) sin x
lim = lim = lim
x→0 sin2 x x→0 [sin2 x] x→0 2 sin x cos x
(1 − cos x) 0
= lim = = 0. (10.4.37)
x→0 cos x 1
Finally we find
where “1” refers to the limit of the omitted cosine. Thus, it appears that
Problem 6
The limit
2)
lim (cos x)1/ log(1+x (10.4.40)
x→0
will be calculated.
Solution
Yet another kind of limit to calculate using l’Hospital’s rule may be symbolically
written as 1∞ . To convert it to the form typical for l’Hospital’s rule, the logarithmic
function must again be used as in the previous example. Therefore, let us first write
2) log cos x
lim log(cos x)1/ log(1+x = lim (10.4.41)
x→0 x→0 log(1 + x 2 )
and see that we have directly got the needed form 0/0. The required assumptions
are satisfied, so one has:
sin x
lim =1 (10.4.43)
x→0 x
and use the fact that the limit of the product of functions in (10.4.42) equals the
product of their limits, obtaining
1
= e−1/2 = √ .
2)
lim (cos x)1/ log(1+x (10.4.45)
x→0 e
Answer
10001001 > 10011000 .
Answers
(a) Limit equals (b2 − a 2 )/2.
(b) 1◦ . Limit equals 1/3. 2◦ . Limit equals −1/3.
(c) 1◦ . Limit equals 0. 2◦ . Limit equals −1/8.
(d) 1◦ . Limit equals 1/2. 2◦ . Limit equals 11/6.
(e) 1◦ . Limit equals 1/e. 2◦ . Limit equals 1.
(f) 1◦ . Limit equals 6. 2◦ . Limit equals 1/e2 .
Chapter 11
Dealing with Higher Derivatives and
Taylor’s Formula
The present chapter is concerned with higher derivatives and Taylor’s formula. It is
also shown how to use the latter to easily find some special limits of functions.
Given a function f : R ⊂D → R differentiable in a certain open subset U ⊂ D.
Let us assume that the function g(x) := f (x) is again differentiable on U . Then
the object
d 2f d df
f (x) := [f (x)] , or := (11.0.1)
dx 2 dx dx
is called the second derivative of the function f , and the function is said to be twice
differentiable.
In a recursive way, the nth derivative of the function f can be defined and
denoted as
d nf
f (n) (x) or . (11.0.2)
dx n
h h2 hn (n)
f (a+h) = f (a)+ f (a)+ f (a)+. . .+ f (a)+Rn (a, h), (11.0.3)
1! 2! n!
where the remainder Rn (a, h) can be written, inter alia, in the following forms:
hn+1
• Lagrange’s form: Rn (a, h) = f (n+1) (a + θ h),
(n + 1)!
hn+1
• Cauchy’s form: Rn (a, h) = (1 − θ )n f (n+1) (a + θ h).
n!
The parameter θ does not need to be identical in both cases. The so-called Peano
Rn (a, h)
form simply specifies that lim = 0.
h→0 hn
A series
∞
hn (n)
f (a) (11.0.4)
n!
n=0
is called the Taylor series. The convergence of series is dealt with in Chaps. 13
and 15. If a = 0 the Taylor series slightly simplifies and bears the name of the
Maclaurin series.
Problem 1
The so-called Leibniz formula for the nth derivative of a product of functions:
n
n
(f (x) g(x)) (n)
= f (n−k) (x)g (k) (x) (11.1.1)
k
k=0
Solution
The binomial coefficient was introduced in the equation (4.0.5) and so there is no
need to restate it here. The proof will be made by the method of induction, which
we were accustomed to in Chap. 4. In particular the calculations performed while
demonstrating Newton’s binomial formula in Exercise 2 in Sect. 4.3 will be strongly
applicative.
Let us start the inductive proof by determining whether the equation (11.1.1) is
satisfied for n = 1. As usual, both sides are compared:
11.1 Demonstrating by Induction Formulas for High Order Derivatives 235
where f (0) (x) ≡ f (x) and g (0) (x) ≡ g(x). (11.1.2) shows that L = R. For n = 1,
the Leibniz formula is, therefore, fulfilled. In the second step, we begin by explicitly
writing the inductive hypothesis and thesis:
k
k
I.H. : (f (x) g(x))(k) = f (k−l) (x)g (l) (x), (11.1.3)
l
l=0
k+1
k+1
I.T. : (f (x) g(x))(k+1) = f (k+1−l) (x)g (l) (x).
l
l=0
(11.1.4)
As we know, the derivative of a finite sum of functions equals the sum of their
derivatives, in so far as each of these functions is separately differentiable. The
derivative of the product is calculated according to the general rules as in (11.1.2).
In this way, it is obtained
k
k
(f (x) g(x))(k+1) = (f (k−l) (x)g (l) (x)) (11.1.6)
l
l=0
k
k
k k
= f (k−l+1) (x)g (l) (x) + f (k−l) (x)g (l+1) (x).
l l
l=0 l=0
Next, our calculations carried out when proving Newton’s binomial formula can
be followed step by step. It may be noticed from the inductive thesis that one needs
expressions of the type
that is to say, similar to the first term of the sum (11.1.6). In the second term, the
orders of derivative disagree, but we already know that the solution is to introduce
(only in the second sum) a new summation variable: l = l + 1. It is clear that the
summation in this new variable runs now not from 0 to k, but rather from 1 to k + 1.
One, therefore, arrives at:
k
k
(f (x) g(x))(k+1) = f (k+1−l) (x)g (l) (x) (11.1.7)
l
l=0
k+1
k
+ f (k+1−l) (x)g (l) (x),
l−1
l=1
where the irrelevant prime in the dummy summation variable l has been omitted.
Naturally Newton’s symbol has also changed:
k k
−→ .
l l−1
Both terms in the formula (11.1.7) already have the similar structure, except for
the summation: in the first, it runs from 0 to k, and in the second from 1 to k + 1. In
Sect. 4.3, this difficulty was averted by separating the summation from 1 to k (which
is contained in both terms) from the particular terms for l = 0 (first sum) and for
l = k + 1 (second sum). Now we proceed similarly, obtaining
k
k k
(f (x) g(x)) (k+1)
= f (k+1−0)
(x)g (x) +
(0)
0 l
l=1
k k
+ f (k+1−l)
(x)g (x) +
(l)
f (k+1−k−1) (x)g (k+1) (x).
l−1 k
(11.1.8)
The expression in square brackets was already simplified when proving Newton’s
binomial formula (see (4.3.14)):
k k k+1
+ = . (11.1.9)
l l−1 l
One can now make use of this result. We obtain in this way Newton’s symbol of
exactly such indexes as present in the inductive thesis (11.1.4). Our expression can
now be rewritten in the form
11.1 Demonstrating by Induction Formulas for High Order Derivatives 237
k
(f (x) g(x)) (k+1)
= f (k+1−0) (x)g (0) (x) (11.1.10)
0
k
k+1 k
+ f (k+1−l)
(x)g (x) +
(l)
f (k+1−k−1) (x)g (k+1) (x).
l k
l=1
The first and the last terms may now be included in the sum by extending the
summation to 0 from below, and to k + 1 from above. The known identities are
used here:
k k+1 k k+1
=1= , =1= . (11.1.11)
0 0 k k+1
Collecting all the results, one sees that the inductive thesis has been reached:
k+1
(f (x) g(x)) (k+1)
= f (k+1−0) (x)g (0) (x) (11.1.12)
0
k
k+1
+ f (k+1−l) (x)g (l) (x)
l
l=1
k+1
+ f (k+1−k−1) (x)g (k+1) (x)
k+1
k+1
k+1
= f (k+1−l) (x)g (l) (x),
l
l=0
Problem 2
The nth derivative of the function f (x) = eax sin bx defined on R for a, b ∈ R \ {0}
will be found.
Solution
The solution of this problem consists of two steps. In the first step, we will try to
guess the formula for f (n) (x), and in the second one, it will be strictly demonstrated
by the method of mathematical induction.
238 11 Dealing with Higher Derivatives and Taylor’s Formula
How can one guess a formula for the nth derivative? If one does not know how to
start, it is usually useful to find explicitly the first few derivatives and maybe some
idea will occur. We, therefore, calculate
f (x) = a eax sin bx + b eax cos bx = eax (a sin bx + b cos bx). (11.1.13)
Now f (x) should be found. However, the direct differentiation of the above
expression gives
# $
f (x) = eax (a sin bx + b cos bx) = a eax (a sin bx + b cos bx) (11.1.14)
+ eax (ab cos bx − b2 sin bx) = eax ((a 2 − b2 ) sin bx + ab cos bx).
It can be seen that after subsequent differentiations the coefficients of the trigono-
metric functions become more and more complicated and it would be difficult
to deduce on this basis the general formula for f (n) (x). For this reason, before
differentiating the second time, the expression (11.1.13) should be slightly prepared.
First of all, one must realize that the complications come from the appearance of the
cosine function. If one could get rid of it, then f (x) would only subtly differ from
the initial function f (x) and one would easily find the successive derivatives.
So, how can one remove the cosine from the expression (11.1.13)? Helpful here
is the known formula for the sine of the sum of angles:
In the formula for f (x), this structure is now recognized in brackets, provided one
denotes α = bx. One might be tempted also to write that
however, it is not possible for arbitrary parameters a and b. Sine and cosine
functions, after all, assume values from the interval [−1, 1] only and also meet
the Pythagorean trigonometric identity. This indicates that one could eventually
introduce an auxiliary angle β using equations (11.1.16), provided the parameters
satisfy the condition: a 2 + b2 = 1. In general, however, it is not true. How can one
deal with this situation? Well, (11.1.13) can be rewritten in the form
a b
f (x) = a2 + b2 eax √ sin bx + √ cos bx , (11.1.17)
a 2 + b2 a 2 + b2
a b
cos β = √ and sin β = √ (11.1.19)
a + b2
2 a2 + b2
This expression is similar to f (x) and√there is no problem with its subsequent dif-
ferentiation. A multiplicative constant a 2 + b2 can be moved out of the derivative,
and the shift of the sine function argument does not lead to any complication either.
By repeating the whole procedure again, one gets
2
f (x) = a 2 + b2 eax sin(bx + β) = a 2 + b2 eax sin(bx + β + β).
(11.1.21)
At this time we are in a position to postulate the formula for nth derivative:
n
f (n) (x) = a 2 + b2 eax sin(bx + nβ). (11.1.22)
Now one can proceed to the second part of the solution, i.e., the inductive proof
of the formula (11.1.22). We need not examine its correctness for n = 1, since we
have chosen it to match for initial values of n. Therefore, it is sufficient to show that
from the inductive hypothesis
k
f (k) (x) = a 2 + b2 eax sin(bx + kβ) (11.1.23)
This proof is similar to that of the previous exercise. One starts from the equa-
tion (11.1.23) and differentiate both sides. In this way we obtain
k
f (k+1)
(x) = a2 + b2 e ax
sin(bx + kβ) (11.1.25)
k
= a 2 + b2 a eax sin(bx + kβ)
+ ( a 2 + b2 )k b eax cos(bx + kβ) = ( a 2 + b2 )k+1 eax
a b
× √ sin(bx + kβ) + √ cos(bx + kβ)
a 2 + b2 a 2 + b2
240 11 Dealing with Higher Derivatives and Taylor’s Formula
k+1
= a 2 + b2 eax sin(bx + kβ + β)
k+1
= a 2 + b2 eax sin(bx + (k + 1)β).
As one can see, the right-hand side of the inductive thesis (11.1.24) has been
reached, so the formula (11.1.22) has been proved.
Problem 1
x2 x3
ex ≥ 1 + x + + (11.2.1)
2 6
is satisfied.
Solution
This problem can be solved by the methods of Sect. 10.1, but we will use Taylor’s
formula (11.0.3). Let us briefly revisit this formula. Suppose that one is dealing with
a real function f of the class C n on a certain interval [a, a + h], where h > 0. There
exists then a certain parameter θ ∈]0, 1[, such that
h h2 hn (n)
f (a+h) = f (a)+ f (a)+ f (a)+. . .+ f (a)+Rn (a, h), (11.2.2)
1! 2! n!
where the so-called “remainder” Rn (a, h) may be given different forms. For our
goal, we choose the Lagrange’s form:
hn+1
Rn (a, h) = f (n+1) (a + θ h). (11.2.3)
(n + 1)!
The function f (x) = ex clearly meets the above assumptions on [0, x]. What is
more, the derivatives can be calculated trivially, as [ex ] = ex . We will thus be able
to apply the formula (11.2.2), with the substitution a = 0 and h = x. One has
x 0 x2 0 xn 0
e0+x = e0 + e + e + ... + e + Rn (0, x). (11.2.4)
1! 2! n!
11.2 Expanding Functions 241
x x2 x3 x 4 θx
ex − 1 − − − = R3 (0, x) = e . (11.2.5)
1! 2! 3! 4!
The specific value of the parameter θ on the right-hand side is not known, but
whatever it is, one certainly has x 4 eθx ≥ 0 and we get the desired inequality:
x x2 x3
ex − 1 − − − ≥ 0. (11.2.6)
1! 2! 3!
Problem 2
Taylor’s formula for the function f (x) = (1 + x)α will be written√down, where
α ∈ R, in the neighborhood of x = 0, and on this basis the value 5 1.04 will be
evaluated.
Solution
Taking this into account, it can be seen that Taylor’s formula (or more properly
Maclaurin’s formula) has the form
(α)n+1
Rn (0, x) = (1 + θ x)α−n−1 x n+1 , (11.2.11)
(n + 1)!
1
x = 0.04, and α= . (11.2.12)
5
As an example, let us calculate this value, taking the first five terms of Taylor’s
expansion:
√
5 (1/5)1 (1/5)2 (1/5)3
1.04 = 1 + · 0.04 + · (0.04)2 + · (0.04)3
1! 2! 3!
(1/5)4 1 2
+ · (0.04)4 + R4 (0, 0.04) = 1 + · 0.04 − · (0.04)2
4! 5 25
6 21
+ · (0.04)3 − · (0.04)4 + R4 (0, 0.04) = 1 + 0.008
125 625
− 0.000128 + 0.000003072 − 0.000000086016 + R4 (0, 0.04)
= 1.007874985984 + R4 (0, 0.04). (11.2.13)
To determine the error committed in this formula if one ignored R4 (0, 0.04), the
remainder is going to be evaluated:
(1/5)5
R4 (0, 0.04) = · (0.04)5 · (1 + 0.04 · θ )1/5−5 (11.2.14)
5!
(1/5)5 1
= · (0.04)5 ·
5! (1 + 0.04 · θ )24/5
(1/5)5 399
< · (0.04)5 = · (0.04)5 ≈ 0.0000000026.
5! 15625
11.2 Expanding Functions 243
In the text of the exercise, however, it was not indicated that one should use Taylor’s
formula for n = 5. In practical estimations, one sets n so that the omitted remainder
be sufficiently small. Usually one must start with the determination of the maximal
error () acceptable, and then one chooses the properly large n. Thus we require
(1/5)n+1 1
Rn (0, 0.04) = · (0.04)n+1
(n + 1)! (1 + 0.04 · θ )−1/5+n+1
(1/5)n+1
< (0.04)n+1 < . (11.2.15)
(n + 1)!
This inequality cannot be solved for n, but substituting subsequent values into the
left-hand side, it can be easily recognized which n would satisfy |Rn | < .
Problem 3
Taylor’s formula for the function f (x) = log(1 + x) will be written down in the
neighborhood of x = 0 and the remainder will be evaluated. Then it will be proved
that
1 1 1
1− + − + . . . = log 2. (11.2.16)
2 3 4
Solution
1
f (x) = , (11.2.17)
1+x
−1 1
f (x) = = (−1)1 ,
(1 + x) 2 (1 + x)2
(−1) · (−2) 1·2
f (x) = = (−1)2 ,
(1 + x)3 (1 + x)3
···
(n − 2)!
f (n−1) (x) = (−1)n−2 ,
(1 + x)n−1
(n − 1)!
f (n) (x) = (−1)n−1 .
(1 + x)n
244 11 Dealing with Higher Derivatives and Taylor’s Formula
Substituting a = 0 and h = x into (11.2.2) and using the above expressions, one
obtains
1 1! 2 2! 3 (n − 1)! n
log(1 + x) = log 1 + x− x + x + . . . + (−1)n−1 x
1! 2! 3! n!
x2 x3 xn
+ Rn (0, x) = x − + + . . . + (−1)n + Rn (0, x).
2 3 n
(11.2.18)
x2 x3 x4
log(1 + x) = x − + − + .... (11.2.21)
2 3 4
In order to obtain (11.2.16), one would have to let x = 1 for Lagrange’s
remainder. This case is considered first:
n+1
(−1)n 1
Rn (0, 1) = . (11.2.22)
n+1 1+θ
This means that the formula (11.2.16) has actually been demonstrated, since
1 1 1 (−1)n−1
Rn (0, 1) = log 2 − 1 − + − + . . . + . (11.2.24)
2 3 4 n
Below we consider for what other values of x the formula (11.2.21) remains
correct. There is no doubt that for all x ∈ [0, 1[ the following inequalities are
satisfied:
x
0≤ <1
1 + θx
11.2 Expanding Functions 245
because the numerator is smaller and the denominator larger than 1. In this case,
obviously, the equation (11.2.20) is true, and consequently also (11.2.21).
When x > 1, the opposite inequality can be met:
1+x
> 1.
θx
This happens for very small values of θ , about which we know, after all, nothing
more than 0 < θ < 1. The convergence of Rn to 0 does not even help the
denominator n + 1, since the behavior of the expression of the type nα β n at infinity
is always determined by the exponential factor, i.e., β n (except for the specific case
β = ±1) and the value of β (and not α) dictates, whether this limit is equal to 0 or
to ∞.
And what is the behavior of the remainder when x < 0 ? For arguments x ≤ −1,
the expression log(1 + x) does not make sense, so we have to confine ourselves to
the interval ] − 1, 0[. The remainder (11.2.19) takes the form
n+1
1 |x|
Rn (0, x) = − , (11.2.25)
n+1 1 − θ |x|
but not knowing the value of θ , one is not in a position to find the limit for n → ∞.
It can be either zero or infinity. The remainder in the form of Lagrange does not
give us any unequivocal answer. But in place of (11.2.3), the so-called Cauchy’s
remainder may be used equally well:
hn+1
Rn (a, h) = (1 − θ )n f (n+1) (a + θ h). (11.2.26)
n!
The parameter θ is again in the interval ]0, 1[, but it need not be the same as the
parameter in Lagrange’s remainder. In our case (11.2.26) takes the form:
n+1
n+1
x (−1)n x(1 − θ )
Rn (0, x) = (−1)n (1 − θ )n = ,
1 + θx 1−θ 1 + θx
(11.2.27)
and for x ∈] − 1, 0[ one can write
n+1
1 |x|(1 − θ )
Rn (0, x) = − . (11.2.28)
1−θ 1 − θ |x|
One has now to determine whether the expression |x|(1 − θ )/(1 − θ |x|) is larger or
smaller than one. It can be transformed as follows:
|x|(1 − θ ) |x| − θ |x| + 1 − 1 1 − |x|
0< = =1− < 1. (11.2.29)
1 − θ |x| 1 − θ |x| 1 − θ |x|
246 11 Dealing with Higher Derivatives and Taylor’s Formula
for 0 < a, b < 1. It follows that Cauchy’s remainder goes to zero for x ∈] − 1, 0[.
Let us summarize the obtained results. It has been found that the for-
mula (11.2.21) can be used for x ∈] − 1, 1]. For x > 1, the remainder in Lagrange’s
form can diverge. It is easy to show that the same is true for Cauchy’s remainder
(e.g., when θ is very small). However, one must be aware that our reasoning does
not rule on the validity of the formula (11.2.21) for x > 1, but only means that
the method used does not provide any answer to this question. The fact that the
series (11.2.21) is really divergent in this case can be seen when examining the
so-called necessary condition for the convergence of a series, which is not met (the
subsequent terms do not tend to zero). We will look into this later in Sect. 13.1.
Problem 1
The limit
cos x − cosh x + x 2
lim . (11.3.1)
x→0 x 3 sin3 x
will be found.
Solution
In this problem, we will be acquainted with a very easy and quick way of
finding limits of a function, which may be treated as “competitive” with respect to
l’Hospital’s rule known from Sect. 10.4. As the condition of its application, however,
one must know in advance the Taylor formulas to a specific order for all functions
in the expression. This will be assumed below. Actually, it does not constitute any
restriction, since expansions of elementary functions are well known and available
in almost every textbook on differential calculus.
In the case specified in this exercise, one needs to know these formulas for the
following functions: cos x, cosh x, and sin x. Before writing them out, let us think a
moment to which order these functions should be expanded. First of all, one needs
to determine the degrees of zeros we are dealing with. There is no doubt that in the
denominator it has the sixth order since for x 0, one has sin x x. Therefore,
11.3 Using Taylor’s Formula to Calculate Limits of Functions 247
for this sine function, one simply takes into account only the term linear in x and in
consequence the formula (11.2.2) for n = 1 can be used.
In the numerator, the situation is a little more complicated, but in order to start
we will use our knowledge of the behavior of the denominator. Consistently, we
are going to keep terms up to x 6 (i.e., n = 6). If they all, coming from various
functions, are canceled, then formally the higher powers should be taken, although
it may be predicted that the limit must then be equal to zero (the degree of zero in
the numerator would be higher than that in the denominator).
As regards to the remainder Rn (a, h), it is sufficient for us to take it in the so-
called Peano version. One does not have to use any specific formula, and one needs
only to know that
Rn (a, h)
lim = 0. (11.3.2)
h→0 hn
Now let us write the Taylor formula for each function, taking into account as
many terms as we need and setting a = 0 and h = x:
x2 x4 x6
cos x = 1 − + − + R6 (0, x),
2! 4! 6!
x2 x4 x6
cosh x = 1 + + + + R̃6 (0, x),
2! 4! 6!
sin x = x + R̄1 (0, x). (11.3.3)
In the light of the above remarks, the three indicated remainders satisfy
By inserting into the formula (11.3.1) all needed functions stored in the
form (11.3.3), one gets
cos x − cosh x + x 2 1 − x 2 /2! + x 4 /4! − x 6 /6! + R6 (0, x)
lim 3
= lim 3
x→0 x 3 sin x x→0 x 3 · x + R̄1 (0, x)
− 1 + x 2 /2! + x 4 /4! + x 6 /6! + R̃6 (0, x) + x 2
+ 3 (11.3.5)
x 3 · x + R̄1 (0, x)
−2 · x 6 /6! + R6 (0, x) − R̃6 (0, x)
= lim
x→0 x 3 · x 3 + 3x 2 R̄ (0, x) + 3x R̄ (0, x) 2 + R̄ (0, x) 3
1 1 1
248 11 Dealing with Higher Derivatives and Taylor’s Formula
1/360 − 0 + 0 1
= −1 · =− .
1+3·0+3·0+0 360
As one sees, the required limit has been found. Apparently, it may seem that
this calculation is quite long. Remember, however, that if one wanted to apply here
l’Hospital’s rule, it is likely that six steps would be needed, since such is the order
of zero in the denominator (and in the numerator). On the other hand, with some
practice and knowing Taylor’s formulas (at least the first few terms) for elementary
functions, limits of the type (11.3.1) can often be found—with the use of the above
outlined method—even mentally.
Problem 2
The limit
will be found.
Solution
This time the limit has a type of 1∞ . This kind of limit was encountered in Sect. 10.4
where l’Hospital’s rule was used. For that problem the expressions such as (11.3.6)
were successfully converted into products with the use of the logarithmic function.
We are going to proceed now in the same manner and, exploiting the property of
continuity of this function, instead of calculating the limit of f (x) = (tan x)tan 2x ,
we find the limit of
log tan x
log f (x) = log(tan x)tan 2x = tan 2x log tan x = . (11.3.7)
cot 2x
The cotangent function, seen in the denominator, vanishes for the argument 2 ·
π/4 = π/2, and this is the first-order zero, which can be established as shown
below:
cos(2x) sin (π/2 − 2x) sin [2(x − π/4)]
cot(2x) = = =− . (11.3.8)
sin(2x) cos (π/2 − 2x) cos [2(x − π/4)]
11.3 Using Taylor’s Formula to Calculate Limits of Functions 249
The value of cosine at 0 equals 1, so cot(2x) for x → π/4 behaves like sine
close to zero. Indeed the zero of the denominator (11.3.7) is then of the first order.
Therefore, Taylor’s formula (11.2.2) will be used (apart from the remainder) with
only terms of the first degree of (x − π/4) incorporated. Since one has
R1 (0, y)
tan y = y + R1 (0, y), where lim = 0, (11.3.9)
s→0 y
Then, one has to write Taylor’s formula for the external function, i.e., the logarithmic
function (where in a moment tan x will act as y):
R̄1 (0, y)
lim = 0. (11.3.14)
s→0 y
250 11 Dealing with Higher Derivatives and Taylor’s Formula
Hence, the expansion will be done in two steps. In the first one, we will use
formula (11.3.13), and then take into account the fact that under the symbol y the
expression (11.3.11) is hidden:
π π π
log tan x = log 1 + 2 x − + R̃1 ,x − (11.3.15)
4 4 4
y
π π π π π π
=2 x− + R̃1 ,x − +R̄1 0, 2 x − + R̃1 ,x −
4 4 4 4 4 4
y y
and we will have all components of (11.3.16). Due to the second of the equa-
tions (11.3.17), one can freely add in the denominator the expression 1/2 ·
R̃1 (π/4, x − π/4), and then designate
Answers
(a) f (n) (x) = (−1)n n!/(2a)[1/(x − a)n+1 − 1/(x + a)n+1 ].
(b) For even n: f (n) (x) = (−1)n/2 (x sin x − n cos x);
for odd n: f (n) (x) = (−1)(n−1)/2 (x cos x + n sin x).
252 11 Dealing with Higher Derivatives and Taylor’s Formula
Exercise 2 Write down Taylor’s (Maclaurin’s) formulas for the following func-
tions (use known expansions for cos x, log x, sin x, and 1/(1 − x)):
(a) f (x) = log cos x, up to x 6 inclusive.
1
(b) f (x) = , up to x 5 inclusive.
1 + sin x
Answers
(a) f (x) = −x 2 /2 − x 4 /12 − x 6 /45 + R6 (0, x).
(b) f (x) = 1 + x + x 2 + 5x 3 /6 + 2x 4 /3 + 61x 5 /120 + R5 (0, x) .
Exercise 3 Find limits (a)–(e) of Problem 5 in Sect. 10.5 using Taylor’s formula.
Chapter 12
Looking for Extremes and Examine
Functions
The present chapter is devoted to the applications of the differential calculus to the
comprehensive investigation of the behavior of functions. In particular, we will learn
how to find monotonicity intervals, extremal points, points of inflection, etc. The
monotonicity has already been touched in the theoretical summary at the beginning
of Chap. 10.
Consider some differentiable function f : R ⊂D → R. It is said to have
at a certain point x0 ∈ D a local maximum if there exists some neighborhood
U (x0 , r) ⊂ D such that
• The function f has a local maximum at x0 if and only if there exists r > 0 such
that
◦ f (x0 ) = 0,
◦ f (x) > 0 for x ∈]x0 − r, x0 [,
◦ f (x) < 0 for x ∈]x0 , x0 + r[.
• The function f has a local minimum at x0 if and only if there exists r > 0 such
that
◦ f (x0 ) = 0,
◦ f (x) < 0 for x ∈]x0 − r, x0 [,
◦ f (x) > 0 for x ∈]x0 , x0 + r[.
For a twice differentiable function on U (x0 , r) with the continuous second
derivative, the necessary and sufficient conditions can be formulated as follows.
• The function f has a local maximum at x0 if and only if f (x0 ) = 0 and
f (x0 ) < 0.
• The function f has a local minimum at x0 if and only if f (x0 ) = 0 and f (x0 ) >
0.
In the case of f (x0 ) = 0, this test does not rule on extremes.
A function f defined on a certain interval [a, b] is called convex if and only if
∀x1 ,x2 ∈[a,b] ∀q∈[0,1] f (qx1 + (1 − q)x2 ) ≤ qf (x1 ) + (1 − q)f (x2 ), (12.0.3)
which was explained in Fig. 4.1. On the other hand, a function f is called concave
if and only if
∀x1 ,x2 ∈[a,b] ∀q∈[0,1] f (qx1 + (1 − q)x2 ) ≥ qf (x1 ) + (1 − q)f (x2 ). (12.0.4)
The point of inflection is a point at which the function changes its character from
concave to convex or vice versa. For a twice differentiable function on ]a, b[, the
necessary and sufficient condition to be convex (concave) on this interval is f (x) >
0 (f (x) < 0) for x ∈]a, b[.
Problem 1
Solution
By the cone in the text of the exercise, naturally the right one is meant, i.e., such
that the top is situated above the center of the base. As everybody knows, the volume
should be calculated using the formula:
1
V = π r 2 h, (12.1.1)
3
where r is the radius of the base, and h is the height of the cone. If the right cone
is inscribed in a ball of fixed radius (R), these two quantities (i.e., r and h) are not
independent of each other: when h has a specific value in the interval [0, 2R], then
r is no longer free. This means that the volume V can be regarded as a function of
one variable only (for instance h). Therefore, from the formula (12.1.1), one should
eliminate r in favor of h. This can be easily done if one looks at Fig. 12.1.
The question may arise, why of the two potential variables (r and h) has h been
chosen as the independent variable. Would it be wrong to consider V (r) rather than
V (h)? Of course not! We were guided only by our future convenience. Choosing r
as an independent variable would require to solve (12.1.2) with respect to h and to
insert the obtained result into (12.1.1). Then V (r) would be expressed by a square
root. As a result, one would also obtain a slightly more complicated expression
for the derivative V (r) too. On the contrary, if one chooses h as an independent
variable, V (h) is simply a polynomial (see (12.1.4)) due to the fact that in the
expression (12.1.1) one finds r 2 and not r. All calculations thereafter are very
simple.
According to the chosen way, let us now find r. The Pythagorean theorem gives
Naturally, of the two solutions, this one is chosen that makes geometrical sense, i.e.,
r ≥ 0. Since h ≤ 2R, the expression under square root is nonnegative. This result
can now be plugged into (12.1.1) and leads to
1 1
V (h) = π( 2hR − h2 )2 h = π(2h2 R − h3 ). (12.1.4)
3 3
The function V (h) is defined and continuous on the interval [0, 2R] and
differentiable inside. Moreover, one has V (0) = V (2R) = 0, and for 0 < h < 2R,
the function V is positive. The maximum of V corresponds to such value of h inside
the interval for which V (h) = 0. Therefore, we calculate the derivative and look
for its roots:
1 4
V (h) = π(4hR − 3h2 ) = π h R − h = 0. (12.1.5)
3 3
Problem 2
Given a parabola y = ax 2 , where a > 0. We will find a point on the graph such
that the length of the normal segment intersecting this point, contained inside the
parabola, is the shortest.
12.1 Finding the Smallest and the Largest Values of a Function on a Given Set 257
Solution
Let us start with a figure for a certain exemplary value of the parameter a. The
normal lines to the parabola contained inside the graph are drawn as thick, solid
lines. The symmetry of the graph means that there should be two such segments
intersecting the points located symmetrically with respect to the axis y (Fig. 12.2).
How to find the normal to the graph at a given point? First a tangent line should
be issued, and for this purpose the derivative can be used. Next one should take
advantage of the fact that the slopes m1 and m2 of the two straight lines orthogonal
to each other (apart from some special situation when these lines are parallel to the
axes of the coordinate system) satisfy the condition
m1 · m2 = −1. (12.1.7)
Let us assume that we want to issue a normal line at (s, y(s)), i.e., (s, as 2 ). As the
derivative of the function y in s is equal to
y 2 x2
3
2
2 1 2 1
y x y x
2 2 1 2 2
x
1 1 1 1
2 2
1
2
1 1
m2 = − =− . (12.1.9)
m1 2as
The equation for the normal straight line issued from the point of coordinate x = s
has, therefore, the form
1
y=− (x − s) + as 2 . (12.1.10)
2as
Now one needs to find the (second) intersection point of this line with the parabola.
To do this, the following set of equations has to be solved:
⎧
⎨ y = ax 2 ,
1 (12.1.11)
⎩y = − (x − s) + as 2 .
2as
We obtain, therefore,
1 1 1
ax 2 = − (x − s) + as 2 ⇐⇒ ax 2 + x− − as 2 = 0
2as 2as 2a
1
⇐⇒ a(x − s) x + s + 2 = 0.
2a s
(12.1.12)
The point of intersection that look for has then the following coordinates (denote
them x̃ and ỹ):
2
1 1 1 1
x̃ = −s − , ỹ = a x̃ 2 = a s + = as 2 + + 3 2.
2a 2 s 2a 2 s a 4a s
(12.1.13)
)
2
2 *1/2
1/2 1 1 1
d(s) = (s − x̃) + (as − ỹ)
2 2 2
= 2s + + + 3 2
2a 2 s a 4a s
)
2
2 *1/2
1 1 1
= 4s 1 +
2
+ 2 1+
4a 2 s 2 a 4a 2 s 2
)
2 *1/2
1 1
= 4s + 2
2
1+ 2 2 (12.1.14)
a 4a s
1 1/2 1 2 2 1 3/2
= 4s + 2
2
1+ 2 2 = 2 s + 2 .
a 4a s s 4a
This function is differentiable for any s = 0. Let us calculate below its derivative:
4 2 1 3/2 6 2 1 1/2 2 2 1 1 1/2
d (s) = − 3 s + 2 + s + 2 = 3 s − 2 s + 2
2
.
s 4a s 4a s 2a 4a
(12.1.15)
It vanishes, when
1 1
s2 = ⇐⇒ s = ± √ . (12.1.16)
2a 2 2a
Then, there are two solutions, which had been expected from the beginning by
symmetry. The minimal distance can be found from one of them, for instance the
one in which s > 0. That it is actually a minimum may be seen after rewriting the
derivative in the form
2 1 1 1 1/2
s−√ s+√ s + 2
2
. (12.1.17)
s3 2a 2a 4a
√
It is clear that at s = 1/ 2a, it changes its sign from negative to positive, or
equivalently the function changes from decreasing to increasing.
It is now sufficient to insert the value of s into (12.1.14) to find the length of the
shortest segment dmin :
)
2 *3/2 √
1 2 1 1 3 3
dmin = d √ = √ 2 √ + 2 = . (12.1.18)
2a 2a 4a 2a
1/ 2a
260 12 Looking for Extremes and Examine Functions
Problem 1
Solution
If one is interested in the full behavior of a function and in its complete graph, and
not only in its extremes as in the previous exercises, one has to perform a more
systematic study of the differential calculus. To arrange the whole procedure, it is
helpful to keep a certain outline of subsequent steps, but not all of them must be
carried out in the sequence chosen below.
1. The domain of the function. If the domain D in a given task is not explicitly
stated, one has to assume that it includes all the arguments for which the formula
of the function makes sense. In our example one can see that D = R, since both
polynomial and arctan functions are well defined for all real arguments.
2. Special properties (e.g., periodicity, parity). Before continuing into a detailed
examination of the function, it is worth considering whether or not it has any
special properties, which will make our job easier. For example, if the function
was even, then finding limits at the ends of intervals, roots, or extremes can be
facilitated by exploiting the symmetry of x ↔ −x. In our example the function
is odd, because
3. Points of intersection with the axes. In order to precisely draw up a graph one,
obviously, needs points where the function vanishes and the point at which its
graph intersects the y axis. One of the solutions of the equation
x − 4 arctan x = 0 (12.2.3)
is, naturally, x = 0. However, the question arises whether there are other
solutions. For sure we shall not find them analytically, but it is worth to establish
whether they exist at all. We are going to come back to this question in a moment.
12.2 Examining the Behavior of Functions from A to Z 261
4. Limits at the ends of intervals. The domain D in our case constitutes one interval
] − ∞, ∞[, so only limits at x → ±∞ come into play, and, due to the odd parity
of the function, it is sufficient to consider only one of them:
f (x)
lim = a. (12.2.5)
x→∞ x
If this limit did not exist, it would mean that there is no asymptote. On the other
hand, the finite value of a still does not prejudge the existence of an asymptote,
√
which can be found out considering an exemplary function g(x) = x + x, for
which
√
g(x) x+ x
lim = lim = 1, (12.2.6)
x→∞ x x→∞ x
but no asymptote exists. This is because the second condition must also be met:
f (x) x − 4 arctan x
lim = lim = 1 =: a1 ,
x→∞ x x→∞ x
lim (f (x) − a1 x) = lim (x − 4 arctan x − 1 · x) = lim (−4 arctan x)
x→∞ x→∞ x→∞
π
= −4 = −2π =: b1 . (12.2.8)
2
This function has, therefore, for x → ∞ the slant asymptote: y = x − 2π .
Because, as we already know, the function f (x) is odd, i.e., f (x)/x is even, then
the asymptote for x → −∞ must have the same slope: a2 = a1 = 1, since
f (x) f (x)
lim = lim . (12.2.9)
x→∞ x x→−∞ x
262 12 Looking for Extremes and Examine Functions
a vertical asymptote is indicated (it may happen, however, that this is a one-sided
asymptote).
6. Derivative and its domain. Further properties as extremes, monotonicity, points
of inflection, convexity, concavity are investigated with the use of differential
calculus. For any x ∈ R, the derivative of the function f exists (so D = R) and
is equal to
4
f (x) = 1 − . (12.2.11)
1 + x2
On the grounds of this expression, one easily notes the following facts:
√ √
• f (x) > 0, if and only if x ∈] − ∞, − 3[ ∪ ] 3, ∞[. However, it would be
wrong to draw the conclusion that the function is increasing on the sum of
these intervals. Yes, it is, but for each
√interval
√ separately!
• f (x) < 0, if and only if x ∈] − 3, 3[. In this interval the function is
decreasing. √
• f (x) = 0 for x = ± 3.
We already have sufficient information
√ to determine that the function f√(x) has
a maximum at the point x = − 3 and a minimum at the point x = 3. It is
worthy to calculate the function values at these points:
√ √ √ √ 4π
f ( 3) = 3 − 4 arctan 3 = 3 − < 0,
3
√ √ √ 4π
f (− 3) = −f ( 3) = − 3 + > 0, (12.2.13)
3
where the property of the odd parity has been used.
12.2 Examining the Behavior of Functions from A to Z 263
√ √
x −∞ −x1 − 3 0 3 x1 ∞
√ 4π √ 4π
f (x) −∞ # 0 # − 3+ $ 0 $ 3− # 0 #∞
3 3
f (x) +++ + + 0 − − − 0 + + +++
f (x) −−− − − − − 0 + + + + +++
On the grounds of the obtained results and of the table, it is now easy to plot the
graph, which is shown in Fig. 12.3.
264 12 Looking for Extremes and Examine Functions
4
3
3
y x 2
y x 2
3 x1
x
x1 3
4
3
3
y x 4arctan x
Problem 2
x2 − x + 4
f (x) = √ (12.2.15)
x2 − x + 1
Solution
In the previous problem, a certain scheme was formulated and so we follow the
indicated steps.
1. The domain of function. In the text of this problem, the domain is given explicitly
as D = R, but as an exercise, one can look for which values of x the
expression (12.2.15) makes sense, i.e.,
x 2 − x + 1 > 0. (12.2.16)
1 2 1
x− + > 0, (12.2.17)
2 4
from which it follows that it is always satisfied. In fact, one can assume then that
D = R.
2. Special properties (e.g., periodicity, parity). The function f does not have
any specific properties, although more skillful eye can recognize the identical
structures in the numerator and denominator: x 2 − x. If one introduced a new
variable x̃ = x − 1/2, the function would become even: f (−x̃) = f (x̃). It is
easy to establish after having rewritten the function in the form
(x − 1/2)2 + 15/4
f (x) = . (12.2.18)
(x − 1/2)2 + 1/4
This means that the graph will be symmetric with respect to the straight line
x = 1/2. We will not benefit directly from this property, but one should be aware
of it as it allows to catch potential mistakes.
3. Points of intersection with the axes. From the equation (12.2.18) it may be seen
that the function is always positive, so there are no roots. The point of intersection
with the y axis will be found, by calculating the value f (0) = 4.
4. Limits at the ends of intervals. Only two limits are important, x → ±∞:
x2 − x + 4 x 2 1 − 1/x + 4/x 2
lim f (x) = lim √ = lim · = ∞.
x→∞ x→∞ x 2 − x + 1 x→∞ x 1 − 1/x + 1/x 2
(12.2.19)
Similarly,
x2 − x + 4 x 2 1 − 1/x + 4/x 2
lim f (x) = lim √ = lim · = ∞.
x→−∞ x→−∞ x 2 − x + 1 x→−∞ |x| 1 − 1/x + 1/x 2
(12.2.20)
Both limits are equal, which is a reflection of the symmetry with respect to the
line x = 1/2.
5. Asymptotes. For x → ∞ the function goes to infinity, so we are going to verify
whether there is an oblique asymptote: y = ax + b. Let us calculate first
√
f (x) (x 2 − x + 4)/ x 2 − x + 1
lim = lim (12.2.21)
x→∞ x x→∞ x
x 2 1 − 1/x + 4/x 2
= lim · = 1 =: a1 ,
x→∞ x 2 1 − 1/x + 1/x 2
and next
266 12 Looking for Extremes and Examine Functions
lim (f (x) − a1 x)
x→∞
√
x2 − x + 4 x2 − x + 4 − x x2 − x + 1
= lim ( √ − 1 · x) = lim √
x→∞ x2 − x + 1 x→∞ x2 − x + 1
√ √
x2 − x + 4 − x x2 − x + 1 x2 − x + 4 + x x2 − x + 1
= lim √ · √
x→∞ x2 − x + 1 x2 − x + 4 + x x2 − x + 1
(x 2 − x + 4)2 − x 2 (x 2 − x + 1)
= lim √ √
x→∞ x 2 − x + 1 (x 2 − x + 4 + x x 2 − x + 1)
−x 3 + 8x 2 − 8x + 16
= lim √ √
x→∞ x 2 − x + 1 (x 2 − x + 4 + x x 2 − x + 1)
1
=− =: b1 . (12.2.22)
2
The slant asymptote at x → ∞ has then the form y = x − 1/2. Now one has
to repeat this calculation for x → −∞, and find a2 :
√
f (x) (x 2 − x + 4)/ x 2 − x + 1
lim = lim
x→−∞ x x→−∞ x
x2 1 − 1/x + 4/x 2
= lim · (12.2.23)
x→−∞ x · |x| 1 − 1/x + 1/x 2
x 1 − 1/x + 4/x 2
= lim · = −1 =: a2 ,
x→−∞ |x| 1 − 1/x + 1/2
and b2 :
x2 − x + 4
= lim ( √ + 1 · x)
x→−∞ x2 − x + 1
√
x2 − x + 4 + x x2 − x + 1
= lim √
x→−∞ x2 − x + 1
√ √
x2 − x + 4 + x x2 − x + 1 x2 − x + 4 − x x2 − x + 1
= lim √ · √
x→−∞ x2 − x + 1 x2 − x + 4 − x x2 − x + 1
(x 2 − x + 4)2 − x 2 (x 2 − x + 1)
= lim √ √
x→−∞ x 2 − x + 1 (x 2 − x + 4 − x x 2 − x + 1)
12.2 Examining the Behavior of Functions from A to Z 267
−x 3 + 8x 2 − 8x + 16
= lim √ √
x→−∞ x 2 − x + 1 (x 2 − x + 4 − x x 2 − x + 1)
x3 −1 + 8/x − 8/x 2 + 16/x 3
= lim ·
x→−∞ x 2 · |x| 1 − 1/x + 1/x 2
1 1
· = =: b2 .
1 − 1/x + 4/x 2 + 1 − 1/x + 1/x 2 2
The last denominator has been transformed in the following way (remember that
x < 0):
( (
1 1 1 1
−x x2 − x + 1 = −x|x| 1 − + 2 = x 1 − + 2 .
2
x x x x
The second asymptote is, therefore, the straight line defined by: y = −x + 1/2.
6. The derivative and its domain. We calculate now the derivative of the function:
√ √
(2x − 1) x 2 − x + 1 − (x 2 − x + 4)(2x − 1)/(2 x 2 − x + 1)
f (x) = √
( x 2 − x + 1)2
2(2x − 1)(x 2 − x + 1) − (x 2 − x + 4)(2x − 1)
= √
2( x 2 − x + 1)3
4x 3 − 4x 2 + 4x − 2x 2 + 2x − 2 − (2x 3 − 2x 2 + 8x − x 2 + x − 4)
= √
2( x 2 − x + 1)3
2x 3 − 3x 2 − 3x + 2
= √ . (12.2.25)
2( x 2 − x + 1)3
The denominator is always positive, so the sign of the derivative is the same as
that of the numerator. A schematic graph of the derivative is then similar to that
of the third degree polynomial and it is a kind of wavy line shown in Fig. 12.4.
268 12 Looking for Extremes and Examine Functions
This figure serves only for identifying the sign of the derivative. It can be
easily determined on its basis that
• f (x) > 0, if and only if x ∈] − 1, 1/2[ ∪ ]2, ∞[. One must remember,
however, that the function f does not have to be increasing on the sum of
these intervals, but only in each interval separately.
• f (x) < 0, if and only if x ∈] − ∞, −1[ ∪ ]1/2, 2[. On each of these intervals
(separately) the function is decreasing.
• f (x) = 0 for x = −1, x = 1/2 or x = 2.
Accordingly, the conclusion to be drawn is that the function f (x) at the point
x = −1 has a minimum, at x = 1/2 a maximum and at x = 2 a minimum again.
Note that these points are located symmetrically in relation to the line x = 1/2.
The function values at these points are given below:
√ 1 5√ √
f (−1) = 2 3, f( ) = 3, f (2) = 2 3. (12.2.27)
2 2
8. The second derivative and its domain. After having differentiated (12.2.25) and
simplified the expression one gets
27x(x − 1)
f (x) = √ . (12.2.28)
4( x 2 − x + 1)5
x2 x 4
y
x2 x 1
5
1 3 1
y x 2 y x
2 2
2 3
x
1 1 2
2
2
1
x −∞ −1 0 1 2 ∞
2
√ 5√ √
f (x) ∞$ 2 3 # 4 # 3 $ 4 $ 2 3 #∞
2
f (x) −−− 0 + + + 0 − − − 0 +++
f (x) +++ + + 0 − − − 0 + + +++
Exercise 1 Find the radius R of the base of a right cone of the smallest volume,
circumscribed of a sphere of radius r.
Answer
√
R = r 2.
Exercise 2 Find the quotient of the height h and the radius r of the base of a
cylinder of the smallest surface and the fixed volume.
Answer
H /r = 2.
270 12 Looking for Extremes and Examine Functions
Exercise 3 Thoroughly examine the following functions and draw their graphs:
(a) f (x) = 5x 2/3 − x 5/3 , for x ∈ R.
3 2
(b) f (x) = x − + 2 , for x ∈ R \ {0}.
x x
x3 √ √
(c) f (x) = 2 + 2x, for x ∈ R \ {− 3, 3}.
x −3
Chapter 13
Investigating the Convergence of Series
In this chapter we deal with numerical series. We will learn how to check their
convergence by the definition or by applying various tests.
A numerical series is the concept of adding infinitely many numbers (terms) an ,
which is denoted by
∞
an . (13.0.1)
n=1
The sequence constructed from the terms of the above series in the following way
N
SN := a1 + a2 + . . . + aN = an (13.0.2)
n=1
lim SN (13.0.3)
N →∞
exists. The value of this limit constitutes simultaneously the sum of the series.
Otherwise, it is called a divergent series. To SN , all tests of the convergence dealt
with in Chap. 5 may be applied.
The necessary condition of the convergence of a series (13.0.1) is
lim an = 0.
n→∞
It is not, however, a sufficient condition, which can be noted when considering the
harmonic series:
1 ∞
1 1 1
1+ + + + ... = , (13.0.4)
2 3 4 n
n=1
If all an > 0, one deals with a series of positive terms. If, in turn, a series has
the form
∞
(−1)n an , (13.0.6)
n=1
(−1)n ∞
1 1 1
1− + − + ... = . (13.0.7)
2 3 4 n
n=1
There are several convergence tests which will be discussed in full detail in the
following problems.
Problem 1
will be examined.
13.1 Using Estimates 273
Solution
We know well from the lecture of analysis that this is a divergent series, although
bn := 1/n −→ 0. This series often serves as a comparative series when studying
n→∞
other series.
Let us now have a look at (13.1.1) and consider how the general term behaves for
very large n. Omitting constants in the denominator, which will be then very small
in comparison with n, one can write
1 1 1 1
an := √ √ = = .
3
(2n − 1)(2n + 1)(2n + 3) 3
2n · 2n · 2n 3
(2n) 3 2n
(13.1.3)
Obviously, the factor 1/2 does not affect the convergence of the series. Thus, one
sees that (13.1.1) behaves for n → ∞ as the harmonic series, and, therefore, it
should be divergent. One only needs an accurate estimate. Thanks to the above
considerations, we already know whatwe want to prove and what estimate to look
for. One has to propose such a series n cn of positive terms that is divergent, and
at the same time satisfies the following inequality:
an ≥ cn , (13.1.4)
for almost all n. Then, if the series n cn diverges, the same must hold for n an .
This method is called the comparison test. Because of the form of the right-hand
side of (13.1.3), a good candidate for comparative series will be the harmonic one.
This surely does not mean that one can immediately write
1 1
√ ≥ , (13.1.5)
3
(2n − 1)(2n + 1)(2n + 3) 2n
274 13 Investigating the Convergence of Series
since constants in the denominator have been omitted without paying attention
to their signs. It is necessary to make the estimation more carefully. We know
that a numerical ratio multiplying 1/n is not relevant for convergence. The idea
would be then that whenever a constant in the denominator cannot be omitted to
maintain inequality (13.1.4), the coefficients are modified in such a way that it can
be extracted from the brackets:
1 1
√ > √
3
(2n − 1)(2n + 1)(2n + 3) 3
(2n)(2n + 2)(3n + 3)
1 1 1 1 1 1
= √ · > √ · = √ · .
3
12 3
n(n + 1)2
3
12 3
(n + 1)3
3
12 (n + 1)
(13.1.6)
The series ∞ n=1 1/(n + 1) is simply the ordinary harmonic series with the first
term removed. Rejection of one term, or even any finite number of them, does not
∞
affect
∞ the convergence. Since n=1 1/n is divergent, the same can be said about
n=1 1/(n + 1) and vice versa. As it has been already mentioned, the multiplicative
constant in (13.1.6) is not relevant for convergence, and this means that the required
estimate (13.1.4) has been found. The conclusion then is that the considered series
is divergent.
Problem 2
will be examined.
Solution
The idea of solving this exercise will be to evaluate the partial sums
m
n+1
Sm := (−1)n log (13.1.8)
n
n=1
and to examine its convergence (as a usual sequence). If it is found that Sm has a
limit, the same will refer to the series (13.1.7).
13.1 Using Estimates 275
The factor (−1)n suggests that it is worthy to consider two situations: when m
is odd (i.e., m = 2k − 1, for k = 1, 2, . . .) and when m is even (i.e., m = 2k, for
k = 1, 2, . . .). Thus, one has
2k−1
n+1 2 3 4 2k − 2
S2k−1 = (−1)n log = − log + log − log + . . . − log
n 1 2 3 2k − 3
n=1
2k − 1 2k
+ log − log , (13.1.9)
2k − 2 2k − 1
2k
n+1 2 3 4 2k − 2
S2k = (−1)n log = − log + log − log + . . . − log
n 1 2 3 2k − 3
n=1
2k − 1 2k 2k + 1
+ log − log + log . (13.1.10)
2k − 2 2k − 1 2k
It is clear that formulas for S2k−1 and S2k differ only by one component of the
sum, namely log(2k + 1)/(2k), which is absent in the first expression. Let us note
at this point an important fact that this term is positive because the argument of the
logarithm is greater than one. This implies the inequality S2k−1 < S2k .
We are going to show below that the sequence S2k−1 is increasing in k. It is
also bounded above by any of the terms of S2k (since the latter will be shown in
a moment to be decreasing). As we know from lectures of analysis, the monotonic
and bounded sequence does converge. This fact was used when examining recursive
sequences in Sects. 5.3 and 5.4.
When k is increased by 1, the number of terms of (13.1.9) increases by 2 and
constantly remains odd. So let us separate the first term, and all other group by two:
2 3 4 2k − 1 2k
S2k−1 = − log + log − log + . . . + log − log
1 2 3 2k − 2 2k − 1
2 9 (2k − 1)2 2 9
= − log + log + . . . + log = − log + log
1 8 2k(2k − 2) 1 8
4k 2 − 4k + 1
+ . . . + log . (13.1.11)
4k 2 − 4k
As it is seen, aside from the first term, all others successively added are positive,
since after having used the formula log a − log b = log(a/b) the argument of
logarithm turns out to be greater than one. Hence, one has
2k + 1 2k + 2
S2(k+1)−1 = S2k+1 = S2k−1 + log − log
2k 2k + 1
(2k + 1)2 4k 2 + 4k + 1
= S2k−1 + log = S2k−1 + log
2k(2k + 2) 4k 2 + 4k
276 13 Investigating the Convergence of Series
1
= S2k−1 + log 1 + 2 > S2k−1 . (13.1.12)
4k + 4k
S2k =
2 3 2k − 2 2k − 1 2k 2k + 1
− log + log − . . . − log + log − log + log
1 2
2k − 3 2k − 2
2k − 1 2k
Both these sequences are bounded by numbers S1 and S2 , as seen when using
inequalities:
and thus prove to be convergent to some limits: g1 (for S2k−1 ) and g2 (for S2k ).
Now, the question arises whether these two subsequences (even and odd) of the
sequence of partial sums Sn have the same limit, i.e., whether g1 = g2 . Because they
have exhausted all the terms of the sequence, this would mean that the sequence Sn
is convergent to the same limit, i.e., the series (13.1.7) from the text of the exercise
too. The answer to this question is provided by the relation
2k + 1
S2k = S2k−1 + log , (13.1.15)
2k
noticed at the beginning. Since with k → ∞ both S2k and S2k−1 are convergent,
and log(2k + 1)/(2k)) goes to zero, one can equate the limits of both sides:
g2 = g1 + 0 ⇒ g2 = g1 ⇒ Sn is convergent. (13.1.16)
In this way, the convergence of the series (13.1.7) has been proved.
13.2 Using Various Tests 277
It is important to note at the end that in our demonstration the specific form
of terms was inessential. The essential ingredients were the presence of the factor
(−1)n and the fact that an = log[(n + 1)/n] was monotonically decreasing to zero.
For any other series of the same characteristics, the proof would look exactly the
same. The only exploited method was grouping together the terms by two (ai −
ai+1 ) and establishing the sign of this difference. For each decreasing sequence this
difference is naturally positive and the conclusions (13.1.12) and (13.1.14) remain
valid. Subject to the condition an −→ 0, (13.1.16) is also met.
n→∞
The observation made in this problem constitutes the essence for the so-called
Leibniz’s convergence test for series. We will come back to it in Exercises 6 and 7
of Sect. 13.2, as well as in Exercise 2 of Sect. 13.3.
Problem 1
will be examined.
Solution
In order to solve this problem, the comparison test will be used. For this purpose,
one will need the inequality
to be proved at the outset. From the lecture of analysis, we certainly know that
x n
lim 1+ = ex for x ∈ R. (13.2.3)
n→∞ n
Let us now take a look at an = (1 + x/n)n and use Bernoulli’s inequality which
was demonstrated in Problem 1 in Sect. 4.3:
x n x
an = 1 + ≥ 1 + n = 1 + x. (13.2.4)
n n
278 13 Investigating the Convergence of Series
This inequality, in accordance with the assumptions given in (4.3.1), is true for
x/n ≥ −1, which is satisfied for any natural n where x ≥ −1. This is just the
case that concerns this problem (see (13.2.2)).
Let us now have a look at the inequality (13.2.4). On the left-hand side, there
are terms of the sequence an , which converges to ex , and on the right-hand side, the
expression is independent of n. Since all terms satisfy the inequality an ≥ 1 + x,
one can go to the limit on both sides, obtaining
lim an = ex ≥ 1 + x. (13.2.5)
n→∞
One must remember at this point that if the inequality in (13.2.4) was strict,
it would not necessarily lead to a strict inequality between the appropriate limits
in (13.2.5). This can be easily seen if one, for instance, takes the sequence of terms
1/n > 0 whose limit is equal to zero (and not greater than zero). In general, if there
are two convergent sequences bn and cn and also bn > cn for almost all n, then it
follows only that lim bn ≥ lim cn .
n→∞ n→∞
The inequality (13.2.5) was demonstrated only for x ≥ −1, but it is very easy
to see that it will remain true for any x ∈ R. For each real x, one can always
find sufficiently large N , such that for n > N, there is x/n > −1 and thus the
inequality (13.2.4) is met. In any case, from (13.2.5), it is completely clear that for
x < −1 the left-hand side is positive and the right-hand side negative.
Coming back to our proof, we use now the inequality (13.2.5) in order to
demonstrate (13.2.2). It is sufficient to notice that the natural logarithm is an
increasing function. If so, the following implications are true:
ex ≥ 1 + x ⇒ log ex ≥ log(1 + x) ⇒ x ≥ log(1 + x), (13.2.6)
it has been put 1/n. This gives us an estimate from below. To find an estimate from
above, the inequality (13.2.2) is again applied, rewritten as follows:
n+1 n n+1−1 1
log = − log = − log = − log 1 −
n n+1 n+1 n+1
−1 1
≥− = . (13.2.8)
n+1 n+1
This time for x, it was inserted −1/(n + 1). Now from (13.2.8) it follows that
1 n+1 1 1 1 1
− log ≤ − = < 2. (13.2.9)
n n n n+1 n(n + 1) n
13.2 Using Various Tests 279
These estimates, thanks to the comparison test, are sufficient to establish that the
series in the text of the problem is in fact convergent. For, it is a series of nonnegative
terms (which is due to (13.2.7))2 and bounded from above by appropriate terms of
the convergent series ∞ n=1 1/n .
Problem 2
Solution
This time, the so-called limit comparison test will be used. Suppose that one is
dealing with a series of positive
terms. The idea of this test consists of selecting a
certain comparative series n bn (of which the convergence or divergence may be
easily established) in such a way that
an
lim =ξ for ξ = 0 and ξ = ∞. (13.2.11)
n→∞ bn
Then one can, not very precisely, say that for very large n both series behave
identically. If so, the convergence
of n bn entails also theconvergence of n an .
Similarly, the divergence of n bn implies a divergence of n an .
If ξ = 0 or ξ = ∞, this test may not decide on the convergence of n an . It is
then useful in only two cases:
1. ξ = 0 ∧ bn − convergent ⇒ an − convergent, since the series
n n
n an behavesin the infinity “better” thancertain convergent series,
2. ξ = ∞ ∧ bn − divergent ⇒ an − divergent, since the series
n n
n bn behaves in the infinity “worse” than certain divergent series.
It is clear that in other cases, when our series is “worse” than a convergent series or
“better” than a divergent series, one cannot judge.
Since this criterion ismost effective when ξ turns out to be a finite number, let us
try to find such a series n bn to obtain this result. We need to exploit, at this point,
a little of our imagination in order to get any ideas as to the behavior of
280 13 Investigating the Convergence of Series
α
1
an = 1 − n sin (13.2.12)
n
for very large n. Let us temporarily substitute a real variable x for 1/n and consider
the expression
if the variable x → 0. Now one can use the Taylor expansion for the sine function.
We know that sin x = x − 1/3! · x 3 + . . ., and hence,
x − sin x x − x + 1/3! · x 3 1
= x2. (13.2.14)
x x 6
We expect, therefore, that for very large n
α
1 1
an , (13.2.15)
6 n2
so the comparative series should have the terms: bn = 1/n2α . From the lecture of
analysis, we know that such a series is convergent if and only if the exponent is
larger than 1 (i.e., α > 1/2 ).
At this point, it is essential to make two remarks:
1. The coefficient 1/3! in the formula (13.2.14) is, at this stage, completely
inessential. The important fact is only that it is not equal to zero. Because in
the expression x − sin x the variable x in the first power cancels, one needs to
determine a leading (for small x) nonzero term. From the odd parity of the sine
function only, we know that it cannot be a x 2 , as odd functions have only odd
powers in the expansion in Taylor’s series around zero (i.e., Maclaurin’s series).
The next term is surely x 3 and it is enough to know that the coefficient with it is
nonzero.
2. The above reasoning, which leads to guessing a form of the comparative series,
is not strict but rather intuitive. That we have been in fact able to predict bn
correctly, we will be convinced below, in the second part of the solution, when
calculating (13.2.11).
Now one applies the limit comparison test, i.e., calculate
an [1 − n sin(1/n)]α 1/n − sin(1/n) α
lim = lim = lim
n→∞ bn n→∞ 1/n2α n→∞ 1/n3
α
1/n − sin(1/n)
= lim , (13.2.16)
n→∞ 1/n3
13.2 Using Various Tests 281
where in the last equality the continuity of the power function has been used. In
order to find the limit
1/n − sin(1/n)
lim , (13.2.17)
n→∞ 1/n3
l’Hospital’s rule (dealt with in Sect. 10.4) is going to be applied in the following
way. Instead of (13.2.17), we consider
x − sin x
lim , where x ∈ R. (13.2.18)
x→0 x3
This limit is of the type 0/0 and the assumptions of l’Hospital’s rule are met, so one
has
The latter limit can be found either by subsequent applications of l’Hospital’s rule
or by transforming it in a way already applied several times:
The limit (13.2.18) is, therefore, equal to 1/6. By virtue of the application of Heine’s
definition, the same result must be obtained for any sequence of arguments xn
convergent to zero. In particular, it has also to hold for xn = 1/n:
1/n − sin(1/n) 1
lim 3
= . (13.2.21)
n→∞ 1/n 6
It is important at this point to caution the reader against the direct use of
l’Hospital’s rule to calculate limits of sequences, i.e., differentiation over a discrete
variable n. Remember that the sequence cn is a function defined only for n =
1, 2, . . ., and not on their neighborhoods! One is then not able to construct the
difference quotient (cn+h − cn )/ h and examine its limit for h → 0. The correct
way of reasoning was presented above.
As a result, one concludes from (13.2.16) that ξ = 1/6α . We have thus been
able to accurately identify a comparison series because ξ = 0 and ξ = ∞, and,
therefore, the test unequivocally decides about the convergence. There are now two
possible cases:
• If α > 1/2, the series n bn = n 1/n2α is convergent, so the same is true
for (13.2.10).
282 13 Investigating the Convergence of Series
• If α ≤ 1/2, the series n bn = n 1/n
2α is divergent, so the same is true
for (13.2.10).
Problem 3
Solution
As we know from Problem 3 in Sect. 5.2, if the general term contains a lot of
multiplicative factors, which have a chance to cancel in the quotient an+1 /an , the
test particularly relevant to use is the so-called ratio test (or d’Alembert’s test). In
Sect. 5.2, it was formulated with respect to the limits of sequences. Now the version
useful for testing the convergence of series is going to be formulated. As required
by this criterion, one finds first the limit an+1 /an when n → ∞. Then, if
an+1
• lim < 1, then the series an is convergent,
n→∞ an
n
an+1
• lim > 1, then the series an is divergent,
n→∞ an
n
an+1
• lim = 1, this criterion does not rule on the convergence and one must
n→∞ an
examine it with other methods.
In the formula (13.2.22), the factors that to a large extent will be canceled within
the quotient can be recognized: these are n! and cn . Therefore, let us write
so one has
an+1 c
lim = . (13.2.25)
n→∞ an e
This result indicates that one has to deal with three cases:
an+1
• c > e, so lim > 1, so (13.2.22) is divergent.
n→∞ an
an+1
• 0 < c < e, so lim < 1, so the series is convergent.
n→∞ an
an+1
• c = e, so lim = 1, so the criterion does not rule on the convergence.
n→∞ an
Is it then the end of the story and we can move to the next problem? Certainly
not. The ratio test does not work for c = e, but that does not mean that one cannot,
in this particular case, examine the convergence using some other method, shown
below. The easiest way to proceed is to refer to the fact known from the lecture of
analysis that the sequence bn = (1 + 1/n)n is increasing. Since it is simultaneously
convergent to the number e, then all its terms must be smaller than e:
1 n
∀n∈N bn = 1 + < e. (13.2.26)
n
A precondition for convergence is not met for c = e, and if so, the series (13.2.22)
is divergent in this case.
Problem 4
a + cos n n
(13.2.28)
b + cos n
n=1
Solution
Having solved the problems of Sect. 5.2, we know that if the general term contains
the n-th power of a certain expression, it is reasonable to attempt to apply Cauchy’s
test for the study of convergence. It also has √ its counterpart for series and is often
called the root test. Let us find the limit of n |an | for n → ∞ and if
• lim n |an | < 1, the series an is convergent,
n→∞
n
• lim n |an | > 1, the series an divergent,
n→∞
n
• lim n |an | = 1, the test does not rule on the convergence and another method
n→∞
should be used.
Let us then calculate
n
a + cos n a + cos n
n
|an | = n
= . (13.2.29)
b + cos n b + cos n
One does not know how to easily find the limit of this expression (in so far as it
exists at all). Naturally, one would be satisfied with the knowledge of whether this
limit is greater or smaller than one. The relevant estimates can be made very easily
if we realize that the function
a+x
f (x) = (13.2.30)
b+x
defined for x ≥ −1 (because b > 1) is increasing for a < b, decreasing for a > b,
and constant when a = b. For, we have
a + x1 a + x2
f (x1 ) > f (x2 ) ⇐⇒ > (13.2.31)
b + x1 b + x2
⇐⇒ (a + x1 )(b + x2 ) > (a + x2 )(b + x1 )
⇐⇒ ab + ax2 + bx1 + x1 x2 > ab + ax1 + bx2 + x1 x2
⇐⇒ (a − b)(x2 − x1 ) > 0.
This means that the two expressions in brackets are either simultaneously positive
(that is, for a > b one has x1 < x2 and the function is decreasing) or negative (i.e.,
for a < b one has x1 > x2 and the function is increasing). Since −1 ≤ cos n ≤ 1,
for those cases the estimates of the expression (13.2.29) are as follows:
If one could assume that the limit (13.2.29) does exist, from the above estimate one
would get
a + cos n a+1
lim n
|an | = lim ≥ > 1, (13.2.33)
n→∞ n→∞ b + cos n b+1
or
n
|an | ≥ c > 1 (then the series is divergent).
In our case, such number has been found, thanks to the estimates (13.2.32)
and (13.2.34):
a+1
c= ,
b+1
so for a > b the series in the text of the exercise is divergent, and when a < b—
convergent. In the particular case—not addressed to until now—when a = b, all
terms are simply equal to 1 and the series is clearly divergent.
Problem 5
Solution
an = 1/n1+ ,
where > 0, is convergent even for very small values of . However, if one puts
= 0, this series turns out to be divergent. The series in this exercise is a kind of
intermediate between the two harmonic series mentioned. For suitably large values
of n, the following inequalities hold:
1 1 1
< < . (13.2.37)
n1+ n log n n
It is because that although log n tends to infinity with n, it does it more slowly than
any power of n, e.g., n . The same estimate for large n will be also valid for α > 0:
1 1 1
< < . (13.2.38)
n1+ n logα n n
An
interesting question arises whether the “improving” of the harmonic series
n 1/n by introducing a logarithm in the denominator is sufficient to achieve its
convergence.
The test, which is especially useful when one is dealing with logarithmic function
in the denominator, is called Cauchy’s condensation test. It has the following form.
Let us assume that we are studying convergence of n an whose terms form a
sequence (monotonically) decreasing to zero. As we remember, the convergence to
zero of the terms an is necessary for the convergence of a series, but monotonicity
not. It constitutes a certain supplementary condition. Then,
an is convergent ⇐⇒ 2k a2k is convergent. (13.2.39)
n k
The other version of this test may be obtained by changing 2k to 3k and so on. Let
us now apply this test to our series. Instead of (13.2.36) we are going to study the
convergence of
1 1 1 1
2k a2k = 2k α k = = . (13.2.40)
2k log (2 ) (k log 2) α logα 2 kα
k k k k
A series well known to converge for α > 1 and diverge for α ≤ 1 has been
obtained. Because of the equivalence in (13.2.39) the identical conclusion can be
drawn for (13.2.36). So for (13.2.36) to be convergent, the exponent α must be
greater than one.
13.2 Using Various Tests 287
It is interesting to note at the end that there is another very convenient test for
this type of series: the so-called integral test. It can be given some attention once we
have been accustomed to improper integrals.
Problem 6
Solution
When solving this problem, one can utilize the results obtained in Example 4 from
Sect. 5.1. It was shown there that the limit of an defined as
an := sin π a 2 + n2 , (13.2.42)
It turns out that (13.2.41) is in fact an alternating series. In this case the Leibniz test,
in a natural way, is imposed on us. It applies precisely to the series of the form
(−1)n bn . (13.2.44)
n
lim bn = 0, (13.2.45)
n→∞
is met, and bn is a monotonic sequence (say from certain n ), the series (13.2.44) is
convergent. If so, the present exercise is reduced to demonstrate the monotonicity
of
288 13 Investigating the Convergence of Series
a2
cn := sin π √ . (13.2.46)
a 2 + n2 + n
For very large values of n the argument of the sine function becomes very small and
certainly belongs to the interval [0, π/2]. In this interval, this function is increasing.
In turn, the expression
a2
√ (13.2.47)
a 2 + n2 + n
obviously monotonically decreases with n:
a2 a2
cn+1 = <√ = cn . (13.2.48)
a 2 + (n + 1)2 + n + 1 a 2 + n2 + n
The composition of an increasing function (denoted by f ) with a decreasing one
(g) is again a decreasing function, which can be deduced from the following
implications:
y1 > y2 ⇒ f (y1 ) > f (y2 )
(13.2.49)
x1 > x2 ⇒ g(x1 ) < g(x2 )
⇒ [x1 > x2 ⇒ g(x1 ) < g(x2 ) ⇒ f (g(x1 )) < f (g(x2 ))] .
Problem 7
will be examined.
13.2 Using Various Tests 289
Solution
In this example, we are going to see how important, when applying Leibnitz’s test
used in the previous exercise to a series of the form
(−1)n bn , (13.2.51)
n
its convergence would not raise any√ doubts. It is ultimately an alternating series and
the sequence of the general term 1/ n monotonically tends to zero. It might seem,
therefore, that the seemingly “tiny” modification in the numerator:
√ √
n n + 1/1000 · (−1)n
−→ =: bn , (13.2.53)
n n
√
which consists of adding to a large (for large n) number n a very small expression
1/1000 · (−1)n , should not affect the conclusions. However, as it is easy to see, one
has
∞
√ ∞
n + 1/1000 · (−1)n n 1/1000 · (−1)n
(−1)n = (−1)n √ + (−1)n
n n n
n=1 n=1
∞
1 1 1
= (−1) √ +n
· . (13.2.54)
n 1000 n
n=1
So one can say that terms of our series are sums of those of the convergent series
(spoken of above) and of the harmonic series divergent to infinity. In such a case,
the series (13.2.54) must be divergent and nothing will be cured here by the small
number 1/1000. The sequence of partial sums,
N
1 1 1
SN := (−1)n √ + · (13.2.55)
n 1000 n
n=1
N
1 N
1 1
= (−1)n √ + · = sN + s̃N ,
n 1000 n
n=1 n=1
sN s̃N
290 13 Investigating the Convergence of Series
is, in fact, the sum of a bounded and unbounded series, so it must be unbounded too.
There remains a question, why such an apparently inessential modification
(13.2.53) caused such important implications and resulting with the series ceasing
to converge. At which point have the assumptions of the Leibniz test been violated?
In a visible way, the sequence formed out of members of the series goes to zero,
so there remains the only possibility that it is no longer monotonic. It is not easy
to recognize it at first glance, but in fact it is no longer monotonic. To check this
property, let us calculate the difference bn+1 − bn :
√ √
n + 1 + 1/1000 · (−1)n+1 n + 1/1000 · (−1)n
bn+1 − bn = −
n+1 n
n
1 1 (−1) 1 1
= √ −√ − + . (13.2.56)
n+1 n 1000 n + 1 n
Monotonicity would require that the above expression has a fixed sign, independent
of n (at least starting from a certain value n). This property depends on the
comparison of modules of the first and the second constituents of the sum above.
Therefore, let us create the quotient:
√ √
1/ n + 1 − 1/√n √
n(n + 1)( n + 1 − n)
= 1000 √ √ (13.2.57)
1/1000 · |1/(n + 1) + 1/n| n n + 1(2n + 1)
√ √ √ √
n(n + 1)( n + 1 − n) n+1+ n
= 1000 √ √ ·√ √
n n + 1(2n + 1) n+1+ n
n(n + 1)(n + 1 − n)
= 1000 √ √ √ √
n n + 1(2n + 1)( n + 1 + n)
√ √
n n+1
= 1000 √ √
(2n + 1)( n + 1 + n)
√
n 1 + 1/n
= 1000 √ · √ .
n n (2 + 1/n)( 1 + 1/n + 1)
This expression tends to zero when n → ∞, so for almost all n it is smaller than
1. This fact does not depend on how large of a factor would be written instead
of 1000. It means that at some point the second term in (13.2.56) dominates over
the first one (as to its modulus). Due to the presence of (−1)n , the sequence bn
ceases to be monotonic and starts to oscillate. This example constitutes a warning
against “disrespectful” treatment of the assumptions of a theorem. The other similar
situation will appear in Exercise 2 of the next section.
13.3 Solving Several Interesting Problems 291
Problem 1
Solution
Before we begin the examination of (13.3.1), we have to solve the auxiliary exercise:
to prove the convergence and find the limit of the following sequence
√
an = n n
a−1 , (13.3.2)
where a > 0. Once again the application of the squeeze rule, considered in Sect. 5.1,
is met. At the same time, we shall see that finding the right comparative sequence
may be a difficult matter.
We begin by citing the inequalities that can serve for the appropriate estimations.
The first one, i.e.,
ex ≥ 1 + x (13.3.3)
has already been proved for any x ∈ R (see (13.2.5)), and there is no need to do
it again. The second one can be obtained from (13.3.3) by substituting −x in place
of x:
1
e−x ≥ 1 − x ⇒ ex ≤ . (13.3.4)
for x<1 1−x
1 x
x ≤ ex − 1 ≤ −1= for x < 1. (13.3.5)
1−x 1−x
So far their relationship with the content of our exercise is not visible. It will,
however, become clear if one substitutes
log a
x= . (13.3.6)
n
292 13 Investigating the Convergence of Series
For fixed a and sufficiently large n, the inequality log a/n < 1 is undoubtedly
satisfied, so one can use (13.3.5), writing
√ n log a
log a ≤ n( n a − 1) ≤ −→ log a. (13.3.9)
n − log a n→∞
Someone could ask now how we guessed at the beginning what inequalities that
would be useful and would lead to the required estimation (13.3.9). Unfortunately,
there is no good answer to this question. The choice depends mainly on the
imagination and experience of the student. A large number of various examples
solved enriches this imagination, but still in more complex and nonstandard
problems, one needs to use the laborious method of “trial and error.”
Now let us continue to the central part of the exercise, i.e., the examination of the
convergence of (13.3.1). If one looks at (13.3.10), this task becomes very simple.
This is because
√ √ √ √
n n
lim n n a − b = lim n n a − 1 + 1 − b
n→∞ n→∞
√ √
n
= lim n( n a − 1) − n( b − 1)
n→∞
√ √
n
= lim n( n a − 1) − lim n( b − 1)
n→∞ n→∞
a
= log a − log b = log . (13.3.11)
b
What has been achieved when calculating this limit? Well, once again a little
imagination is useful. The answer becomes clear if the above equation is rewritten
in the form
√ √ √
√n
n
a− nb a
lim n n
a − b = lim = log . (13.3.12)
n→∞ n→∞ 1/n b
This formula has the form of the limit comparison test, known to us from Example 2
of the previous section, in which, as a comparative series, the harmonic series
13.3 Solving Several Interesting Problems 293
n 1/n has been used. As we know, this one is divergent. Apart from the specific—
and non-interesting—case a = b, the right-hand side of (13.3.12) is finite and
different from zero. It is, therefore, this lucky case, where the test clearly rules on
the convergence. A comparative series was divergent, so the series (13.3.1) is as
well.
Problem 2
Solution
(−1)n
an = , (13.3.14)
n + α sin nx
one might get an impression that the series (13.3.13) is ideal for the use of the
Leibniz test (see Exercises 6 and 7 in the previous section). The series is alternating,
and it seems that for large values of n the presence of α sin nx in the denominator
does not matter because this expression is restricted to the range [−|α|, |α|] ⊂
] − 1, 1[. This suggests that the considered series is convergent. However, we have
already learned that one has to be very cautious in formulating similar conclusions
and one must precisely examine whether the assumptions are really met. It is not
enough to have a superficial assessment, such as presented above, since we are not
absolutely certain as to the monotonicity of 1/(n + α sin nx), although it may not
be ruled out.
To strictly investigate the convergence of the series, let us write it in the form
∞
∞
(−1)n 1 1 1
= (−1) n
− + (13.3.15)
n + α sin nx n + α sin nx n n
n=1 n=1
∞
∞
1 1 (−1)n
= (−1) n
− + .
n + α sin nx n n
n=1 n=1
294 13 Investigating the Convergence of Series
The last equality, in which the series has been written as the sum of two others,
does not have to be automatically correct and requires that both series on the right-
hand side are convergent. Only then the series of sums is equal to the sum of series.
The last series on the right-hand side does not present any problem. The expression
1/n goes monotonically to zero and in combination with factor (−1)n allows one
to use the Leibniz test. A series of that kind is called “anharmonic” and naturally is
convergent. In Exercise 3 of Sect. 11.2, one even managed to find the value of this
sum (log 2).
For the first of the two series we write
1 1 |α sin nx| |α|
(−1)n − = ≤ . (13.3.16)
n + α sin nx n n(n + α sin nx) n(n − |α|)
To estimate the quotient from above, the largest value of the numerator and the
smallest one of the denominator were used. Next, one can write
∞
1 1
(−1)n − . (13.3.18)
n + α sin nx n
n=1
We remember that for the convergence it does not matter whether the sum starts
from n = 1, n = 2 or even n = 100. What is important is its behavior for large
values of n. The convergence of (13.3.18) implies also the convergence of
∞
1 1
(−1) n
− . (13.3.19)
n + α sin nx n
n=1
In such a case,
one says that the series isabsolutely convergent, i.e., not only the
expression n an makes sense, but even n |an |.
So, one can see that the series (13.3.13) has proved to be the sum (13.3.15) of
two convergent series and, as such, is convergent too. Our first superficial proposal
has proved to be correct, but the convergence of the alternating series is often very
delicate and one should never relay solely on the intuition.
13.3 Solving Several Interesting Problems 295
Problem 3
will be examined.
Solution
From Example 6 in the previous section and from those in which the convergence
of series was examined, we learned that if the argument of sine or cosine increases
to infinity, it is convenient to separate the leading part and to examine only the
remainder. However, in the cited examples, this leading part was a multiple of π
or of its fraction as π/2, which guaranteed that one was able to calculate the value
of the appropriate trigonometric function. A completely different situation, we have
to deal with in the current example. The sine function for integer arguments does
not assume any “elegant” values. When trying to uncouple the leading part, which,
moreover, is simply n2 , we will come to nothing. One has to use some other way.
There exists one test, which—it seems—could work in this case: it is a necessary
condition for the convergence of the series, which was also required when solving
previous examples. If one was able to show that
an −→
0,
n→∞
The argument of sine increases with each step by a natural number, which, of course,
is not a multiple or π (which, after all, is an irrational number!) or even a fraction
of it. Our imagination suggests that the obtained values of (13.3.21) should be
randomly scattered within the image of the sine function, i.e., between [−1, 1]. And
since they are scattered randomly, they cannot form a sequence convergent to zero.
296 13 Investigating the Convergence of Series
The above reasoning is purely intuitive and does not constitute a strict proof, but
tells us how it should be performed. One should, namely, pass from an to an+1 ,
an+2 , and eventually further, to see if all these values may lie around zero.
We are going to lean on the definition of the limit of a sequence (5.2.2). Let us
choose a very small > 0 and suppose that there exists such N ∈ N, that for all
n > N one has
Since the sine is a continuous function together with its inverse (in neighborhoods
of zeros), this means that its argument must be very close to k0 π , where k0 is a
certain natural number. Thus, one can write
n2 = k0 π + 0 , (13.3.23)
(n + 1)2 = k1 π + 1 , (13.3.26)
(n + 2)2 = k2 π + 2 , (13.3.27)
2n + 1 = (k1 − k0 )π + 1 − 0 . (13.3.28)
2n + 3 = (k2 − k1 )π + 2 − 1 . (13.3.29)
Let us now think if the above condition can be satisfied. If k2 − 2k1 + k0 = 0, then
one would have
2 = 2 − 21 + 3 , (13.3.31)
13.4 Exercises for Independent Work 297
one can see that the left-hand side is a number in the interval ]1, 3[, and the right-
hand side is greater than 3 (because |k2 − 2k1 + k0 | ≥ 1 ). Thus, we have again
a contradiction. This means that the limit of an cannot not be zero; the necessary
condition for the convergence of (13.3.20) is violated and the series is divergent.
Answers
(a) 1◦ . Convergent. 2◦ . Divergent.
(b) 1◦ . Convergent. 2◦ . Divergent.
(c) 1◦ . Convergent. 2◦ . Divergent.
(d) 1◦ . Divergent. 2◦ . Convergent. 3◦ . Convergent.
(e) 1◦ . Convergent. 2◦ . Convergent.
(f) 1◦ . Divergent. 2◦ . Divergent.
298 13 Investigating the Convergence of Series
Answers
(a) 1◦ . Convergent. 2◦ . Divergent.
(b) 1◦ . Convergent. 2◦ . Convergent.
(c) 1◦ . Divergent. 2◦ . Convergent.
Chapter 14
Finding Indefinite Integrals
The principal notion of the present chapter is that of the indefinite integral. We will
become acquainted with the definition and we will learn to use several methods to
calculate the integrals.
The indefinite integral of a given function f : P ∈ x −→ f (x) ∈ R , where
P ⊂ R is an interval, is a certain differentiable function F (x) satisfying
Due to its definition, it is also called the antiderivative or primitive function. The
procedure of calculating this function is called (indefinite) integration. Symboli-
cally it is written as
+
F (x) = f (x) dx, (14.0.2)
There are no universal methods to find indefinite integrals. Some of the formulas
for the fundamental functions are collected below, and for more complicated
functions, one can try to use properties 3 and 4. Still more sophisticated methods
dedicated to special types of integrand functions are discussed in detail in the
following exercises.
, ,
f (x) f (x) dx f (x) f (x) dx
0 C cosh x sinh x + C
1 x+C tanh x log(cosh x) + C
2 x +C log | sinh x| + C
1 2
x coth x
x2 1
3 x3 + C sin x − cos x + C
x α (α = −1) 1
α+1 x α+1 + C cos x sin x + C
1
x log |x| + C tan x − log | cos x| + C
ex ex + C cot x log | sin x| + C
log a a + C arctan x + C
1 1
ax x
1+x 2
sinh x cosh x + C √1 arcsin x + C
1−x 2
Problem 1
will be found.
Solution
One of the most important tools used to find integrals is the method of “integration
by parts.” It proceeds according to the formula:
+ +
f (x)g(x) dx = f (x)g(x) − f (x)g (x) dx (14.1.2)
14.1 Integrating by Parts and by Substitution 301
upon the assumption that, all derivatives and integrals exist. The idea is that rather
than directly
, calculating f (x)g(x) dx, which for some reason is difficult, we try
to find f (x)g (x) dx, which may (but does not have to) turn out to be easier. In
some cases, it happens that the procedure defined by (14.1.2) must be used several
times repeatedly to reach the relatively simple integral. In this problem we are
dealing with such a case.
The first difficulty that appears when one faces a problem of calculating an
integral is the choice of the integration method. Sometimes it is possible to use
a number of alternative methods, and sometimes none of the standard ones leads
directly to the goal. Of course, it is important to remember that there are integrals
that cannot be explicitly calculated by any method, defining the so-called special
functions. However, if, for some reason, one has decided to use the integration by
parts, one needs to be able to recognize in the integrand the product of a certain
function (i.e., g) and of the derivative of another function (f ). In some cases this is
quite obvious, and in others not.
Analyzing our example, one sees in the first place that the integral is complex
enough that any straight way of calculating it is not visible. Second, the integrand
is a product (and not, for example, a composition of functions), which suggests
that one should try integration by parts. If so, one must decide how to divide the
integrand between functions f (x) and g(x). This can be done in many different
ways, but there are two factors one should keep in mind. The first is that if we decide
to mark a certain expression as f (x), we need to be able to relatively easily find the
primitive function f itself because it is required by the right-hand side of (14.1.2).
The second important factor is that after having applied (14.1.2) the function under
the integral should simplify, and not become more complicated!
When looking at (14.1.1), one can see that these two conditions will be satisfied
if the integrand expression is written in the form
3
log x 1
= · log3 x
x x 3 . (14.1.3)
f (x) g(x)
The antiderivative for 1/x 3 can be found mentally: it is simply −1/(2x 2 ). In turn,
the differentiation of the function g, after having used the fact that [log x] = 1/x,
leads to the simplification of the integrand: in place of one of the logarithms, there
appears a factor 1/x which is already present under the integral, so at most only its
power changes. If one managed to discard the logarithmic function completely—
and this will be our aim in the next steps—only a power expression would remain
to be integrated, and this we know how to do without difficulty.
Since we already have a plan of all steps, we can write
+
3 + +
log x 1 −1
= I dx = log3 x dx = log3 x dx (14.1.4)
x x3 2x 2
302 14 Finding Indefinite Integrals
+
1 1 1 3 1
= − log3 x + log x dx = − 2 log3 x
int. b. p. 2x 2 2 x2 2x
+ +
1 1 3 1 3 1
+ 2
· log 2
x dx = − 2
log 3
x + log2 x dx.
2 x x 2x 2 x3
As one can see, some success has been achieved: the degree of the logarithm has
been reduced by 1 without making the integrand more complicated in any way.
Performing two analogous steps, one can get rid of the logarithm completely. First
calculate
+ +
3 1 3 −1 2
log 2
x dx = log x dx (14.1.5)
2 x3 2 2x 2
+
3 3 1 2
= − 2 log x + 2
log x dx
i. by p. 4x 4 x2
+
3 3 1 2
= − 2 log2 x + · log x dx
4x 4 x2 x
+
3 3 1
= − 2 log x + 2
log x dx,
4x 2 x3
and then
+ +
3 1 3 −1
3
log x dx = log x dx
2 x 2 2x 2
+ +
3 3 1 # $ 3 3 1 1
= − log x + log x dx = − log x + · dx
i. by p. 4x 2 4 x2 4x 2 4 x2 x
+
3 3 1 3 3
= − 2 log x + 3
dx = − 2 log x − 2 + C, (14.1.6)
4x 4 x 4x 8x
1 3 3 3
I =− log3 x − 2 log2 x − 2 log x − 2 + C. (14.1.7)
2x 2 4x 4x 8x
At the end, it is always helpful to make sure that no mistakes have been made by
differentiating (14.1.7) and checking that it really gives the integrand in (14.1.1).
And so it does, in fact, in our case.
14.1 Integrating by Parts and by Substitution 303
Problem 2
will be found.
Solution
From the previous example, we already know what to be guided by when separating
the integrand between functions f (x) and g(x) if one wants to integrate by parts.
The same method will be used in this example. The reader might oppose here as
the expression under the integral (14.1.8) does not have a form of a product. Which
factors do we want to separate then? However, if one reflects, then one comes to the
conclusion that one can always write:
In this way, our expression not only the form of the left-hand side of (14.1.2) is
given (where f (x) = x and g(x) = arctan x) but moving the derivative one easily
gets rid of the troubling arctan function, since
1
[arctan x] = . (14.1.10)
1 + x2
Coming back for a moment to the trick contained in the equation (14.1.9), it is
worth noting that in some other case, it may turn out to be beneficial to include in
the integral not only a unity, but even the whole missing expression f (x), i.e., to
replace the whole integral:
+ +
1
g(x) dx with f (x) · · g(x) dx, (14.1.11)
f (x)
as long as one aptly chooses a subsidiary function f (x).
We can now proceed with our calculations which, after having used (14.1.9), turn
out to be very simple:
+ + +
I = arctan x dx = [x] arctan x dx = x arctan x − x [arctan x] dx
i. by p.
+
1
= x arctan x − x dx. (14.1.12)
1 + x2
304 14 Finding Indefinite Integrals
One can be sure that it is more convenient to integrate rational functions than
inverse trigonometric functions. Moreover, in Sect. 14.3, it will be seen that there
exists a systematic way of integrating any rational function, provided one knows
how to decompose the denominator into elementary factors. However, in the case
of (14.1.12), the problem is particularly simple. For, if the integrand is rewritten as
1 1 2x
x = · , (14.1.13)
1+x 2 2 1 + x2
2x = [1 + x 2 ] . (14.1.14)
Such expressions can be integrated mentally, since the formula for the differentiation
of the composite function immediately gives : h (x)/ h(x) = [log(h(x))] for any
positive and differentiable function h(x) (in our case its role is played by 1 + x 2 ).
This means that the integral we are looking for is equal to
+ +
1 1 2x
I = x arctan x − x dx = x arctan x − dx
1+x 2 2 1 + x2
1
= x arctan x − log(1 + x 2 ) + C. (14.1.15)
2
Problem 3
Indefinite integrals
+ +
Iss = sin ax sin bx dx and Isc = sin ax cos bx dx, (14.1.16)
Solution
It often happens when integrating by parts that after having applied this method
one obtains on the right-hand side of the equation not the final result, but again
the original expression, i.e., the same that one has on the left-hand side. Then,
the success of our calculations depends on whether it has appeared there with a
coefficient different from 1. Otherwise the unknown quantity cancels on both sides
of the equation and it cannot be found. If, however, the coefficients are distinct, one
gets the equation with one unknown quantity and there is a chance to determine it.
14.1 Integrating by Parts and by Substitution 305
We are going to deal with examples of that kind in this problem. In addition, the
obtained results will be helpful in Problem 3 of the next section.
Let us first apply the method of integration by parts to the integral denoted as Iss :
+ +
−1
Iss = sin ax sin bx dx = [cos ax] sin bx dx
a
+
1 1
= − cos ax sin bx + cos ax [sin bx] dx
i. by p. a a
+
1 b
= − cos ax sin bx + cos ax cos bx dx. (14.1.17)
a a
The single application of the method has not led to our goal, so one needs to do it
again:
+
1 b 1
Iss = − cos ax sin bx + [sin ax] cos bx dx
a a a
+
1 b b
= − cos ax sin bx + 2 sin ax cos bx − 2 sin ax [cos bx] dx
i. by p. a a a
+
1 b b2
= − cos ax sin bx + 2 sin ax cos bx + 2 sin ax sin bx dx
a a a
1 b b2
= − cos ax sin bx + 2 sin ax cos bx + 2 Iss . (14.1.18)
a a a
If a 2 = b2 the unknown Iss does not disappear from the equation and can be
determined:
1
Iss = [b sin ax cos bx − a cos ax sin bx] . (14.1.19)
a 2 − b2
Because the integral is indefinite, one can add any constant to the result, as usual. In
the case when a = ±b, calculating Iss does not pose any problems if one uses the
identity
1
sin2 α = (1 − cos 2α). (14.1.20)
2
Then one obtains (again up to a constant)
+ +
1
Iss = ± sin2 ax dx = ±
(1 − cos 2ax) dx
2
1 1 x 1
=± x− sin 2ax = ± ∓ sin 2ax. (14.1.21)
2 2a 2 4a
306 14 Finding Indefinite Integrals
It is easy to verify that the same result would be obtained from the formula (14.1.19)
by executing the limit b → ±a.
Now let us look at the second of the integrals (14.1.16). We will proceed
identically to the case of Iss :
+ +
−1
Isc = sin ax cos bx dx = [cos ax] cos bx dx
a
+
1 1
= − cos ax cos bx + cos ax [cos bx] dx
i. by p. a a
+
1 b
= − cos ax cos bx − cos ax sin bx dx. (14.1.22)
a a
for b = ±a and
+
x 1
cos ax cos ax dx = + sin 2ax. (14.1.28)
2 4a
At the end, it is worth saying that the integrals (14.1.16) could also be found
using another simpler way and provided one knows the formulas given below. For
the rest, it is quite a common case that various alternative methods are possible to
use for an integration. We have namely the relations
α+β β −α
cos α − cos β = 2 sin sin ,
2 2
α+β β −α
sin α + sin β = 2 sin cos . (14.1.29)
2 2
One can now choose α and β to obtain (α + β)/2 = ax and (β − α)2 = bx and
instead right-hand sides of (14.1.29) to integrate the left-hand ones, from which a
product of trigonometric functions disappeared.
Problem 4
will be found.
Solution
This time we are going to use the method of integration by substitution, although
in the second part of the solution, it will be found that the integration by parts may
also be useful. We begin by quoting the formula to be applied. Suppose that there
exists an indefinite integral of a real function f defined on an interval P and a
differentiable real function g with values in P . Then,
+ +
f (x) dx = f (g(t))g (t) dt. (14.1.31)
When looking at the integrand in (14.1.30), we have to perceive some hint which
would allow us to propose the function g(t). First of all, one should note that the
range of variation of x is restricted to the interval [−1, 1]. Therefore, it can be
advantageous to choose x = g(t) = sin t, where t ∈] − π/2, π/2[. The expression
under the root simplifies then because one can use the Pythagorean trigonometric
identity:
1 − x 2 = 1 − sin2 t = cos2 t = | cos t|. (14.1.32)
In the interval ] − π/2, π/2[, the cosine function is positive, so in the upcoming
calculations, the symbol of the absolute value may be omitted. To use (14.1.31),
one still needs to know the derivative of the function g(t), but this one is obvious:
g (t) = [sin t] = cos t. We can now write
+ + +
I= 1 − x 2 dx = t · cos
cos t dt = cos2 t dt. (14.1.33)
f (g(t)) g (t)
1
cos 2α = cos2 α − sin2 α = 2 cos2 α − 1 ⇒ cos2 α = (cos 2α + 1),
2
(14.1.34)
thanks to which one gets
+
1 1 1 1 t
I= (cos 2t+1) dt = sin 2t + t +C = sin 2t+ +C. (14.1.35)
2 2 2 4 2
1 t 1
I= · 2 sin t cos t + + C = x 1 − x 2 + arcsin x + C. (14.1.36)
4 2 2
Now, let us think whether this result could have been obtained by integration by
parts. Under the integral there is not any product of functions, so we are going to
use the trick known from the previous exercise:
+ + +
I = 1 − x 2 dx = 1· 1 − x 2 dx = [x] 1 − x 2 dx
+ +
−2x
= x 1 − x 2 − x[ 1 − x 2 ] dx = x 1 − x 2 − x √ dx
i. by p. 2 1 − x2
+
x2
= x 1−x + √ 2 . (14.1.37)
1 − x 2 dx
14.1 Integrating by Parts and by Substitution 309
What has the application of the formula (14.1.2) shown us? Apparently not much,
but with a little imagination, we can notice that in the integral on the right-hand side
one can easily obtain again the quantity I , which allows us to determine its value.
This is because one can write:
+ + 2 +
x2 x −1+1 1
√ dx = √ dx = − 1 − x2 + √ dx
1 − x2 1 − x2 1 − x2
= −I + arcsin x + C, (14.1.38)
Problem 5
Solution
In this problem, our first goal is to get rid of the exponential function from the
denominator. Therefore, the method of integration that is the first to impose on us is
the integration by substitution and transition from the integral over x to that over z,
where z = ex . However, if one rewrites the expression under the root in the form:
it can be seen that a more efficient first step would be to make a straight substitution:
z = ex + 1. (14.1.44)
Now we are going to use the formula (14.1.31), in which one only has to change
the names of the variables that appear, i.e., to write it in the form:
+ +
f (z) dz = f (z(x))z (x) dx. (14.1.45)
I1 I2
The first of the above integrations is now easy to calculate because in the
numerator the derivative of the polynomial (up to the factor of 2) has appeared
under the root: 2z = [z2 + 1] . This means that
z 1 2z
√ = ·√ = z2 + 1 , (14.1.47)
z2 + 1 2 z2 + 1
and then
I1 = z 2 + 1 + C1 . (14.1.48)
The second of the integrations in (14.1.46) will not be so easy. The consecutive
substitution√will have to be used. When one looks at the denominator in I2 , the
expression
√ z2 + 1 is found. As we know from the previous exercise, in the case
of −z + 1, it would be convenient to apply the trigonometric substitution because
2
of the possibility of using the Pythagorean identity. This time, unfortunately, one has
the opposite sign at z2 , so it cannot be directly applied here. We have, however, at
our disposal, the analogous identity referring to the hyperbolic functions in which
this sign is reversed (see (9.3.20)):
This is exactly what we need if we make the substitution z = sinh u. The image of
the hyperbolic sine is the interval ] − ∞, ∞[, so there are no restrictions on z. For
each value of z the corresponding u does exist.
14.1 Integrating by Parts and by Substitution 311
There was no problem with the sign when taking the square root, since cosh u is
always positive. Going back to the variable z, one gets
I2 = arsinh z + C2 . (14.1.51)
The function u = arsinh z means the inverse to z = sinh u and can be explicitly
found by solving the following equation for u (in an analogous way it was done in
Exercise 1 of Sect. 2.2):
1 u
z = sinh u = (e − e−u ) ⇐⇒ e2u − 2zeu − 1 = 0. (14.1.52)
2
There remains only to return to the variable x, writing z = ex + 1, and the final
result is obtained:
I = (ex + 1)2 + 1 − log(ex + 1 + (ex + 1)2 + 1) + C (14.1.56)
= e2x + 2ex + 2 − log(ex + 1 + e2x + 2ex + 2) + C.
312 14 Finding Indefinite Integrals
Problem 6
Solution
it can be seen that—again due to (14.1.31)—our integral I may be given the form
+ +
2 u (t) 2 1
I= dt = du. (14.1.62)
n u (t) − 1
2 n u2 −1
14.2 Using the Method of Recursive Formulas 313
This last integration is already very easy to execute if one makes the partial fractions
expansion:
+
2 1 1 1 1
I = − du = (log |u − 1| − log |u + 1|) + C
n 2 u−1 u+1 n
1 u − 1
= log + C. (14.1.63)
n u + 1
At the end, one needs only to return to the original variable x, writing
√ √
u= t +1= x n + 1. (14.1.64)
Finally we obtain
√
xn + 1 − 1
1
I = log √ + C. (14.1.65)
n xn + 1 + 1
For
√ an even value√of n, the absolute value symbol can be omitted, as the expression
( x n + 1 − 1)/( x n + 1 + 1) is nonnegative regardless of the value of x (of course
one has x = 0). If n is odd, this cannot be done because the numerator and the entire
expression is negative for −1 < x < 0.
Problem 1
Solution
One of the methods for calculating integrals with which one should be acquainted
is the so-called recursive method. In what circumstances can it be applied? We are
going to answer this question using the present example. Imagine that one is faced
with the necessity of calculating the integral (14.2.1), where n is undetermined or
equal to a large natural number (by “large” we mean here even n = 4, 5, 6, etc.).
314 14 Finding Indefinite Integrals
The straight calculation of such an integral is, in this case, quite cumbersome or at
least tedious. On the other hand, for n = 1 our job is very simple because (14.2.1)
becomes then the known integral leading to the inverse tangent function (the
integration constant is omitted):
+
1 1 x
I1 = dx = arctan . (14.2.2)
x2 +a 2 a a
This result is obtained very easily when making in I1 the substitution x = at:
+ +
1 1 1 1
I1 = adt = dt = arctan t, (14.2.3)
(at)2 + a 2 a t2 + 1 a
1 1
. (14.2.4)
(x 2 + a 2 )n (x 2 + a 2 )n+1
The simplest way is to rewrite the integral (14.2.1) in the following manner:
+ +
1 x 2 + a2
In = dx = dx (14.2.5)
(x 2 + a 2 )n (x 2 + a 2 )n+1
+ + +
a2 x2 x2
= dx + = a 2
In+1 + dx.
(x 2 + a 2 )n+1 (x + a )
2 2 n+1 (x + a 2 )n+1
2
We have found In+1 but still do not know the value of the integral
+
x2
dx, (14.2.6)
(x 2 + a 2 )n+1
which does not have a form of Ik . So, if one wants to get a closed recursive equation,
one has to find it explicitly or bring it back to the form (14.2.1). As it will be seen
below, the latter can be done if one performs the integration by parts:
14.2 Using the Method of Recursive Formulas 315
+
x2
dx (14.2.7)
(x 2 + a 2 )n+1
+ +
1 −2nx 1 1
=− x dx = − x dx
2n (x 2 + a 2 )n+1 2n (x 2 + a 2 )n
+
1 x 1 1
= − · 2 + [x] dx
i. by p. 2n (x + a 2 )n 2n (x 2 + a 2 )n
+
1 x 1 1 1 x 1
=− · 2 + dx = − · 2 + In .
2n (x + a ) 2 n 2n (x + a )
2 2 n 2n (x + a ) 2 n 2n
1 x 1
In = a 2 In+1 − · + In , (14.2.8)
2n (x 2 + a 2 )n 2n
2n − 1 1 x
In+1 = In + · . (14.2.9)
2na 2 2na 2 (x 2 + a 2 )n
The recursive formula has been obtained! Using it with simple algebraic
operations, one can easily find more and more complicated integrals:
1 1 x 1 1 x 1 x
I2 = I1 + 2 · 2 = 2 · arctan + 2 · 2
2a 2 2a x + a 2 2a a a 2a x + a 2
1 x 1 x
= 3 · arctan + 2 · 2 , (14.2.10)
2a a 2a x + a 2
3 1 x 3 1 x 1 x
I3 = 2 I2 + 2 · 2 = arctan + ·
4a 4a (x + a 2 )2 4a 2 2a 3 a 2a 2 x 2 + a 2
1 x 3 x 3 x 1 x
+ · 2 = 5 arctan + 4 · 2 + 2· 2 ,
4a (x + a )
2 2 2 8a a 8a x + a 2 4a (x + a 2 )2
and so on.
Finally, some attention should be paid to a certain fact. The formula (14.2.9)
describes the recursion “by one,” which means that in order to find I2 , I3 , . . ., it
was sufficient to calculate explicitly the easiest integral, i.e., I1 . If the obtained
formula defined the recursion “by two” (i.e., In+2 were expressed by In and possibly
also by In+1 ), then, in addition to I1 , one would also have to find I2 (by direct
integration). With the same situation, we dealt in Problem 1 of Sect. 5.3. Higher
recursive formulas requiring calculation of several “initial” integrals are possible
too.
316 14 Finding Indefinite Integrals
Problem 2
Solution
In this exercise, obviously the recurrence refers to the parameter n and not α. We
are going to endeavor to associate In with In+1 and an imposing way is to make use
of the equality
1 1 1
= −n · · . (14.2.12)
logn x x logn+1 x
This observation suggests how to solve our problem: the factor x α should be written
in the form of a derivative and then, when integrating the expression by parts, this
derivative can be moved onto the logarithm, as in (14.2.12). Finally one can try to
save the result as In+1 . First we are going to consider the case α = −1, for which
one has
1 α+1
xα = x . (14.2.13)
α+1
Consequently,
+
1 1 1 x α+1
In = x α+1 dx = ·
α+1 logn x i. by p. α + 1 logn x
+
n 1 1 1 x α+1
+ x α+1 · dx = · (14.2.14)
α+1 x logn+1 x α + 1 logn x
+
n 1 1 x α+1 n
+ xα = · + In+1 ,
α+1 log n+1
x α + 1 logn x α+1
α+1 1 x α+1
In+1 = In − · . (14.2.15)
n n logn x
14.2 Using the Method of Recursive Formulas 317
Now the case α = −1 has to be considered. The integral has then the form
+
1 1
· dx, (14.2.16)
x logn x
and there is no need for a recursive formula because it can be calculated directly. In
the integrand expression, the derivative of a composite function is recognized (1/x
ultimately is the derivative of the natural logarithm):
⎧
+ + ⎨− 1 1
1 1 1 · for n > 1,
· dx = [log x] dx = n − 1 logn−1 x
x logn x logn x ⎩
log(log x) for n = 1.
(14.2.17)
This case does not have to be dealt with.
Coming back to the recursive formula (14.2.15), one sees that in order to find In
by algebraic methods, one must at the outset perform one integration and calculate
I1 . All higher In will then be easily obtained by using only multiplications and
additions. There emerges here a difficulty which can be encountered in practical
calculations. It appears that, apart from certain specific values α (such as α = −1 ),
the integral
+
xα
I1 = dx (14.2.18)
log x
The obtained integral (up to an additive constant) defines a special function (the
so-called exponential integral Ei), which is well known and described in various
textbooks. This issue is beyond the scope of this book, but for us it is enough to
know that
+ u
e
I1 = du = Ei(u) + C = Ei((α + 1)t) + C = Ei((α + 1) log x) + C.
u
(14.2.20)
Together with (14.2.15), this gives a complete solution to our problem.
318 14 Finding Indefinite Integrals
The reader may not be fully satisfied that the obtained result (14.2.20) and that
the formulas for In have not been expressed by elementary functions. However,
one has to get accustomed to the situation—not at all uncommon—that results of
integration happen to be special functions. The choice of a method does not affect
it. Each method applied leads to the same result (barring an additive constant or
different forms of the same expression).
Problem 3
Solution
To obtain a recursive formula, one first writes the expression for In+2 and makes
use of the well-known formula for the sine of the sum of angles:
+ +
sin(n + 2)x sin nx cos 2x + cos nx sin 2x
In+2 = dx = dx. (14.2.22)
sin x sin x
Under the integral there are two fractions, the second of which should not cause any
trouble. If we write that sin 2x = 2 sin x cos x, then the denominator will reduce
with the sine in the numerator and for integrating there will remain only the product
of trigonometric functions considered in Exercise 3 of the previous section. To
integrate the first fraction we are going to apply the formula (14.1.20). Thus we
have
+ +
(1 − 2 sin2 x) sin nx
In+2 = dx +2 cos x cos nx dx . (14.2.23)
sin x
a
In+2 b
In+2
1
b
In+2 = [n cos x sin nx − sin x cos nx] for n > 1, (14.2.24)
n2 −1
14.2 Using the Method of Recursive Formulas 319
a
while In+2 splits into two integrals:
+ +
sin nx
a
In+2 = dx − 2 sin x sin nx dx. (14.2.25)
sin x
The former is nothing other than In , and the latter has already been determined and
is given by the formula (14.1.19). Thus one can write
2
a
In+2 = In + [n sin x cos nx − cos x sin nx] for n > 1. (14.2.26)
n2 −1
By combining these results, one gets the following recursive relation, starting from
n = 2:
2
In+2 = In + [n sin x cos nx − cos x sin nx]
n2 − 1
2
+ 2 [n cos x sin nx − sin x cos nx] (14.2.27)
n −1
2
= In + 2 (n − 1)[cos x sin nx + sin x cos nx]
n −1
2 sin(n + 1)x
= In + .
n+1
and try to get the recursion “by 1”? Looking, however, at (14.2.29), one can see
that the second integration really does not present any problem, but the first one
would be neither easy to calculate, nor easily expressed by In , since it would not be
possible to get rid of the cosine function from the numerator.
This difficulty does not appear when calculating In+2 , as instead of cos x there
appears cos 2x which can be directly expressed by the sine function, reducing with
the denominator. The chosen method leads, therefore, certainly easier and faster to
the goal.
Problem 1
will be found.
Solution
x x ax + b c
= 2 = 2 + . (14.3.3)
x3 + x2 + x + 1 (x + 1)(x + 1) x +1 x+1
The most general case of such an expansion will be dealt with in the following
exercise. In the current example, the integrand function is relatively simple and there
are only two such fractions because the denominator is a product of two irreducible
factors. One only needs to take care that the polynomials in the numerators be of
lower degree than those in the respective denominators and to establish constants
a, b, and c. One can easily find the constants by comparing the left- and right-hand
sides of (14.3.3):
14.3 Integrating Rational Functions 321
a + c = 0, a + b = 1, b + c = 0. (14.3.5)
Problem 2
will be calculated.
322 14 Finding Indefinite Integrals
Solution
From the previous example, we already know that any integral of a rational
function can be calculated, provided the partial fractions expansion (plus possibly a
polynomial) of the integrand is available. Therefore, the first step should be to write
down the denominator in (14.3.11) in the form of a product of irreducible factors. As
we know from algebra, each real polynomial can be given the form of multiplicative
factors of at most second degree. Finding them, however, is in general a complicated
issue, and such a task can turn out to be unfeasible. In this situation, the method of
integrating presented in this section fails. Therefore, we proceed further with the
assumption that these factors are known or can be easily found. Then there is a
guarantee that one will succeed in finding the integral, although it can sometimes be
quite tedious.
In the example given in the text of this exercise, one has
The second step is to write out all possible partial fractions which may appear
in the expansion of the integrand. They are determined by factors appearing in the
denominator. In our case the following fractions may arise:
a b cx + d ex + f
, , , , (14.3.13)
x−1 (x − 1)2 x2 + 4 (x 2 + 4)2
cx + d c(x − 1) + d + c c d +c
= = + . (14.3.14)
(x − 1)2 (x − 1)2 x − 1 (x − 1)2
Including expressions such as (cx + d)/(x − 1)2 leads, therefore, only to redefining
constants in (14.3.13). This reasoning is not, however, applicable to (cx + d)/(x 2 +
4), where the denominator is indecomposable into linear factors and, therefore,
cannot be reduced. In the same way, one can argue that there is no need to take
into account in the numerators expressions with higher powers of x other than those
given.
If one already has a complete list (14.3.13), the next step is to determine all
constants. For this purpose, we write the equation
14.3 Integrating Rational Functions 323
3x 5 − 4x 4 + x 3 + x 2 − 24x − 2 a b cx + d ex + f
= + + 2 + 2
(x − 1) (x + 4)
2 2 2 x − 1 (x − 1) 2 x +4 (x + 4)2
(14.3.15)
and reduce the right-hand side to a common denominator. Then, on both sides,
fractions with identical denominators are obtained, and, therefore, one can equate
their numerators only:
3x 5 − 4x 4 + x 3 + x 2 − 24x − 2 =
= (a + c)x 5 + (−a + b − 2c + d)x 4 + (8a + 5c − 2d + e)x 3
+(−8a + 8b − 8c + 5d − 2e + f )x 2 + (16a + 4c − 8d + e − 2f )x
−16a + 16b + 4d + f. (14.3.16)
The solution is tedious but does not present any major difficulties. One gets a = 0,
b = −1, c = 3, d = 3, e = −8, and f = 2. After having inserted these constants
into the integrand (14.3.15), one can see that we are left with three simple integrals
partially known to us from the previous exercise:
+ + +
1 x+1 4x − 1
I =− dx + 3 dx − 2 dx. (14.3.18)
(x − 1) 2 x +4
2 (x 2 + 4)2
Each of them will be found separately (constants of integration are temporarily
omitted):
+
1 1
dx = − ,
(x − 1)2 x−1
+ + +
x+1 1 2x 1
dx = dx + dx
x +4
2 2 x +4
2 x +4
2
1 1 x
=log(x 2 + 4) + arctan ,
2 2 2
+ + +
4x − 1 2x 1
dx = 2 dx − dx (14.3.19)
(x + 4)
2 2 (x + 4)
2 2 (x + 4)2
2
2 1 x 1 x
=− − · − arctan ,
x 2 + 4 8 x 2 + 4 16 2
324 14 Finding Indefinite Integrals
where (14.2.10) has been used. Collecting all terms together, one gets the final
result:
13 x 3 4 1 x 1
I= arctan + log(x 2 +4)+ 2 + · + +C. (14.3.20)
8 2 2 x + 4 4 x2 + 4 x − 1
For irreducible factors one has q − p2 /4 > 0, so the formula similar to (14.3.19)
can be used.
Problem 1
will be calculated.
14.4 Integrating Rational Functions of Trigonometric Functions 325
Solution
It is well known that the integral of any rational function of two variables u = sin x
and v = cos x can be reduced to an ordinary rational function of t if one applies
a universal substitution t = tan(x/2). Rational functions, in turn, can be integrated
with methods of the previous section. However, the above-mentioned substitution
generally leads to expressions that require a lot of work due to the relatively high
degree of polynomials in the numerator and in the denominator, obtained as a result
of substitution (although this is not, of course, an absolute rule). As a consequence,
the function is expanded in a variety of partial fractions that all need to be integrated;
one has to find plenty of constants in this expansion, etc. For this reason, before
attempting to “mechanically” use the universal substitution, one should always
consider whether any of the specific substitutions can be used.
In the case of our present exercise, it can be seen that if in the integrand function
we substitute
which means that the function is odd in the variable cos x (but not in the variable
x!). In this situation, a convenient substitution has the form
t = sin x. (14.4.2)
It leads to a rational function of the variable t in a relatively simple form. For one
has
and other than even powers of cosine cannot occur (because of the above-mentioned
oddness of the integrand function). We have, therefore,
+ +
sin2 x cos x t2
I= dx = dt. (14.4.5)
sin x + cos2 x t + 1 − t2
t2 t2 − t − 1 + t + 1 t +1
= = −1 − 2 . (14.4.6)
t +1−t 2 t + 1 − t2 t −t −1
326 14 Finding Indefinite Integrals
This expression can be expanded into fractions, as this was done in the previous
examples, but one can make the task easier by complementing the numerator to the
derivative of the denominator, which is [t 2 − t − 1] = 2t − 1:
+ +
t +1 1 2t − 1 + 3
I = −1 − dt = −1 − · dt
t2 − t − 1 2 t2 − t − 1
+
1 2t − 1 3 1
= −1 − · 2 − · dt. (14.4.7)
2 t − t − 1 2 t2 − t − 1
and then,
+ +
1 2t − 1 1
I2 := − dt = − log |t 2 − t − 1| dt
2 t2 − t − 1 2
1
= − log |t 2 − t − 1| + C2 , (14.4.9)
2
+ +
3 1 3 1
I3 := − dt = − dt
2 t −t −1
2 2 (t − 1/2)2 − 5/4
+
3 1
= − √ √ dt (14.4.10)
2 t − 1/2 − 5/2 t − 1/2 + 5/2
+
3 1 1
= − √ √ − √ dt
2 5 t − 1/2 − 5/2 t − 1/2 + 5/2
) √ √ *
3 1 5 1 5
= − √ log t − − − log t − + + C3
2 5 2 2 2 2
√
t − 1/2 − √5/2 √ 2t − 1 − √5
3 5 3 5
= − log √ + C3 = − log √ + C3 .
10 t − 1/2 + 5/2 10 2t − 1 + 5
Problem 2
will be calculated.
Solution
A glance at the integrand function allows us to conclude that it does not have any
special property of parity. Neither when substituting:
nor
and when applying both of them together. We are then doomed to use the universal
substitution: t = tan(x/2). First, it must be established how the components of the
integrand are expressed by the variable t. One has
x x x
sin x = sin 2 · = 2 sin cos (14.4.13)
2 2 2
and
x x x
cos x = cos 2 · = cos2 − sin2 , (14.4.14)
2 2 2
which implies that both trigonometric functions of the half-angles are needed, given
the tangent of full angle (equal to t). One can find them in the standard way by
writing
x sin(x/2)
t = tan , or t= , (14.4.15)
2 cos(x/2)
squaring both sides of the latter equation and using the Pythagorean trigonometric
identity:
x t2 x x t2 1
sin2 = , cos2 = 1 − sin2 = 1 − = . (14.4.17)
2 1 + t2 2 2 1+t 2 1 + t2
This is sufficient. To find cos x, the obtained expression should be directly inserted
into (14.4.14):
1 t2 1 − t2
cos x = − = . (14.4.18)
1 + t2 1 + t2 1 + t2
To find sin x, which is expressed in accordance with (14.4.13) with the half-
angle functions in first powers, one needs to resolve the question of proper signs
when taking square roots of (14.4.17). This can be done very easily if one realizes
that what we need are not, in fact, separate expressions for sin(x/2) and cos(x/2),
but only their product. And as we know, a product has the same sign as a quotient,
which here is tan(x/2), i.e., simply t. This means that when taking roots one has to
choose signs such that the product in (14.4.13) has the same sign as t (i.e., one takes
two pluses or two minuses). In this way we get
2t
sin x = . (14.4.19)
1 + t2
The last element still needed is dx. Let us differentiate over x both sides of the
equation t = tan(x/2):
dt 1 1 1 + t2
= · 2
= . (14.4.20)
dx 2 cos (x/2) 2
2
dx −→ dt. (14.4.21)
1 + t2
+ + 2 +
2 (t + 1)2 t + 2t + 1 2t
· dt = − dt = − dt = − 1+ dt
1 + t2 1 + t2 1 + t2 1 + t2
+
=− 1 + [log(1 + t 2 )] dt = −t − log(1 + t 2 ) + C. (14.4.22)
Problem 1
will be found.
Solution
The concept of the so-called Euler substitutions is similar to those used in the
previous examples: they remove an irrationality from the integrand and convert
it into a rational function which can always be integrated (with the assumption
that one knows how to write the polynomial in the denominator as a product of
elementary factors). In the present case, it is convenient to use the so-called first
Euler substitution. The other Euler substitutions will be dealt with in the next
problem.
Euler substitutions can be used when the integrand expression is a rational
function of two variables. One of them is simply x, and the other has the form
√
ax 2 + bx + c, where a, b, c are certain constants, with a = 0. The first Euler
substitution has the following form:
330 14 Finding Indefinite Integrals
√
ax 2 + bx + c = a(t − x). (14.5.2)
It is clear that it may be applied if the coefficient a > 0. It defines a new variable
t, over which the integral will now be performed. Using this equation, it is easy
to get rid of the square root from the integrand—instead of it the right-hand side
of (14.5.2) appears—but still the variable x has to be removed. Therefore, the
question arises whether or not, when determining x(t) from the equation above,
any new irrationality will be introduced, this time in the variable t. Such a danger
in fact does exist, since after squaring both sides of (14.5.2), which has to be done,
there appears the variable x in the second power. In order to determine x, it can
be necessary to take the root of a certain expression—in other words, to solve a
quadratic equation—dependent on t, i.e., the irrationality in t can appear.
There is, however, one case where this problem does not occur. This happens
when the square terms (in the variable x) on both sides have coefficients chosen so,
that they cancel and x 2 disappears entirely from the equation. It may be easily found
that this takes place with the first Euler substitution (in the second and third ones
too) because both sides of the equation contain the identical square terms ax 2 .
The subsequent procedure will be followed with the use of the current example.
The first Euler substitution has now the form
√
4x 2 + x + 1 = 4(t − x) = 2(t − x). (14.5.3)
4x 2 + x + 1 = 4t 2 − 8tx + 4x 2 , (14.5.4)
4t 2 − 1
x= . (14.5.5)
8t + 1
4t 2 − 1 4t 2 + t + 1
4x 2 + x + 1 −→ 2 t − =2 . (14.5.6)
8t + 1 8t + 1
In accordance with the formula (14.1.31), one still has to replace dx by x (t) dt.
Differentiating (14.5.5) over t,
14.5 Using Euler’s Substitutions 331
4t 2 + t + 1
dx −→ 8 dt. (14.5.8)
(8t + 1)2
Gathering all of it, one can write the integral in the new variable in the following
form:
+
1 4t 2 + t + 1
I = ·8· dt
(4t − 1)/(8t + 1) · 2 · (4t + t + 1)/(8t + 1) (8t+ 1)2
2 2
x √ x (t)
4x 2 +x+1
+
1
=4 dt. (14.5.9)
4t 2 −1
We are then left only with (14.5.3) and, in the place of t, substituting
1 2
t =x+ 4x + x + 1.
2
In that way, we come to the final result
2x + √4x 2 + x + 1 − 1
I = log √ + C. (14.5.11)
2x + 4x 2 + x + 1 + 1
332 14 Finding Indefinite Integrals
Problem 2
will be calculated.
Solution
As one can see in the formula (14.5.12), under the integral, we have a rational
function of x and the square root of a trinomial. In such a situation, known to us
from the previous exercise, one can use Euler substitutions. It is impossible to apply
here the formula (14.5.2), since the coefficient of x 2 is negative. However, there are
two other substitutions (the so-called substitution II and III) at our disposal:
√ √
II : ax 2 + bx + c = xt + c, when c > 0, (14.5.13)
√
III : ax 2 + bx + c = t (x − x1 ), when (14.5.14)
ax 2 + bx + c = a(x − x1 )(x − x2 ).
In the latter case, it is assumed that both roots of the trinomial are real and that
x1 < x2 .
These three substitutions exhaust all possibilities. For, if it happens that neither I
nor II may be used (when a < 0 and c < 0 simultaneously), one can √ always apply
III. This is due to the fact that if an integrand containing the term ax 2 + bx + c
is to have any meaning, there must exist such a range of variation of x for which
ax 2 + bx + c > 0. For a < 0, the parabola is, however, facing down, so positive
values may appear only when the trinomial has two distinct roots, and then one can
use (14.5.14).
From (14.5.12) one can see that the parameter c is positive, so the second
substitution is chosen. We ask the reader to examine whether and how (14.5.14)
may be used. With our choice one has
−x 2 + 2x + 1 = xt + 1. (14.5.15)
t −1
x(t) = −2 , (14.5.17)
t2 + 1
the relation (14.5.15) will be met and the integrand function proves to be a rational
function of t. For, we have
t −1 −t 2 + 2t + 1
−x 2 + 2x + 1 −→ −2t +1= , (14.5.18)
t +1
2 t2 + 1
and
t 2 − 2t − 1
dx −→ x (t) dt = 2 dt. (14.5.19)
(t 2 + 1)2
After having applied the above substitution, we get the integral in the form
+
2 −1
t −1 −t + 2t + 1 t 2 − 2t − 1
I = −2 · 2 +2 · 2 · dt
t +1 t2 + 1 (t 2 + 1)2
+ 2 + 2
t −t +2 t +1−t +1
= −4 dt = −4 dt
(t + 1)
2 2 (t 2 + 1)2
+
1 t 1
= −4 − + 2 dt. (14.5.20)
t 2 + 1 (t 2 + 1)2 (t + 1)2
and the last one has already been dealt with in Problem 1 in Sect. 14.2:
+
1 1 1 t
dt = arctan t + · 2 + C3 . (14.5.22)
(t 2 + 1)2 2 2 t +1
334 14 Finding Indefinite Integrals
1 t
I = −4 arctan t − 2 − 2 arctan t − 2 t 2 +C
+1
t2 t +1
t +1
= −6 arctan t − 2 2 + C, (14.5.23)
t +1
where the symbol C denotes a new constant: C = −4C1 + 4C2 − 4C3 . The variable
t still has to be eliminated in favor of x with the use of the relation
√
−x 2 + 2x + 1 − 1
t= . (14.5.24)
x
In this way, one obtains
√ √
−x 2 + 2x + 1 − 1 −x 2 + 2x + 1 + x − 1
I = −6 arctan +x √ + C,
x −x 2 + 2x + 1 − x − 1
(14.5.25)
which one can try to further simplify, but we leave it as it is.
In the last two exercises, we have been accustomed with Euler substitutions, but it
should be remembered that often integrals can be calculated with the use of various
alternative methods. That is exactly the case in the example under consideration.
One can, for instance, rewrite the integrand function as follows:
x+2 1 −2x + 2 3
√ =− ·√ +√ , (14.5.26)
−x + 2x + 1
2 2 −x + 2x + 1
2 −x + 2x + 1
2
and the second term (apart from the coefficient), after shifting and rescaling the
integration variable, can be written as
1
√ . (14.5.28)
1 − t2
Problem 1
will be found.
Solution
(
q
(x − p) = y, (14.6.4)
a
which gives
( 2
q a 2 q
ax 2 + bx + c −→ a y + q = |q| y + . (14.6.5)
a |a| |q|
(both of these parameters may not be simultaneously negative), and manipulate the
expression to one of the cases (14.6.2).
Now this procedure is to be implemented for the integral (14.6.1). At first, the
expression under the square root is written as
− x 2 − 4x − 3 = −x 2 − 4x − 4 + 1 = 1 − (x + 2)2 , (14.6.7)
and then one defines y = x + 2. The integral assumes now the form
+
1 − y2
I= dy. (14.6.8)
y−1
Now we are ready to use the trigonometric substitution: y = sin t. The integral I
has its meaning only in a domain in which −1 ≤ y < 1, so let us assume that
−π/2 ≤ t < π/2. In this interval, cos t ≥ 0, and hence
1 − y 2 −→ 1 − sin2 t = cos2 t = | cos t| = cos t. (14.6.9)
In addition,
It should be stressed that a rational function of two variables has been obtained:
sin t and cos t. In Sect. 14.4, we learned how to handle similar expressions, but it is
not reasonable to “mechanically” apply methods presented there (or any other). For,
the integrand can easily be simplified, and then integrated if one writes
14.6 Making Use of Hyperbolic and Trigonometric Substitutions 337
+ +
1 − sin2 t (1 − sin t)(1 + sin t)
I = dt = dt
sin t − 1 sin t − 1
+
= − (1 + sin t) dt = −t + cos t + C. (14.6.12)
At the end, one needs only to come back to the variable x by substituting
Problem 2
Solution
This time, the expression under the root, by appropriate substitution, can be
manipulated into y 2 + 1. For, we have
x 2 + 2x + 2 = (x + 1)2 + 1 = y 2 + 1, (14.6.17)
y=x+1
y = sinh t. (14.6.19)
338 14 Finding Indefinite Integrals
The hyperbolic cosine is always positive, so there appears no sign problem when
passing from the first to the second line. If one now uses the identity
then,
sinh 2t = 2 sinh t cosh t = 2(x + 1) x 2 + 2x + 2, (14.6.24)
and consequently,
1 3
I = (x + 1) x 2 + 2x + 2 − 2 x 2 + 2x + 2 + arsinh(x + 1) + C
2 2
1
= (x + 1) x 2 + 2x + 2 − 2 x 2 + 2x + 2
2
3
+ log(x + 1 + x 2 + 2x + 2) + C, (14.6.25)
2
where (14.1.54) has been used.
14.7 Exercises for Independent Work 339
Exercise 3 Expanding rational functions onto partial fractions, find the integrals:
+
2x − 1
(a) I = 2 + 1)2 (x + 2)
dx.
+ (x
2x + x − 1
2
(b) I = dx.
+ (x + 2x + 5)(x 2 + 3x + 2)
2
x2 + x + 4
(c) I = dx.
+ x +x +x+1
4 3
−x + 4
(d) I = dx.
x − 2x 3 + 2x − 1
4
In Chaps. 5 and 13, we were dealing with numerical sequences and series. This
chapter is concerned with sequences and series whose terms are real functions and
not ordinary numbers. We will be interested in their convergence and investigate the
properties of limiting functions.
Given the functions fn : D → R for n = 1, 2, . . .. An infinite sequence of
functions is the sequence in the form
If for any fixed x ∈ D this sequence is convergent to some number denoted by f (x),
it is said to be pointwise convergent to the function f (x), i.e.,
also written as
fn ⇒ f. (15.0.4)
• Assume that (almost) all fn ’s are differentiable functions on P , and for some x0 ∈
P , the finite limit lim fn (x0 ) exists. Assume also that the sequence composed
n→∞
of the derivatives f n converges uniformly on P . Then, fn uniformly converges
on P to a certain function f and
for any x ∈ P .
• Assume that (almost) all fn ’s are continuous functions on [a, b] and fn ⇒ f on
[a, b]. Then the function f (x) is integrable (for this notion see the first chapter
of the second part in this book series) on [a, b] and
+b +b
f (x) dx = lim fn (x) dx.
n→∞
a a
N
FN (x) = fn (x). (15.0.6)
n=1
Now the series (15.0.5) is said to be pointwise convergent if the series FN (x) is
pointwise convergent, and the series (15.0.5) is uniformly convergent if the series
FN (x) is uniformly convergent.
The properties of uniformly convergent series are similar to those formulated for
sequences.
• Fn ⇒ f on P ⇒ Fn → f on P . The converse is not true.
• If (almost) all fn ’s are continuous on P and Fn ⇒ f on P , then f is continuous
on P .
• Assume that (almost) all fn ’s are differentiable functions on P , FN ⇒ f and
∞
f n is uniformly convergent on P . Then, f (x) is a differentiable function on
n=1
P and
∞
f (x) = f n (x)
n=1
for any x ∈ P .
15.1 Finding Limits of Sequences of Functions 343
+b ∞ +
b
A special type of a functional series is the power series (e.g., the Taylor or
Maclarin series) defined as
∞
an (x − x0 )n . (15.0.7)
n
Various tests for the uniform convergence applicable for different types of series
are discussed in detail within the following problems.
Problem 1
Solution
In the current problem and the subsequent problem, we are going to deal with the
pointwise convergence of sequences, the subsequent terms of which are functions
defined on some subset of R and with their values in R. For each fixed argument x,
one has then to do with an ordinary number sequence fn (x), which can be tested
344 15 Investigating the Convergence of Sequences and Series of Functions
with all methods considered in Chap. 5. Therefore, a question may emerge, what
the purpose of discussing again the convergence of sequences is. The answer is
as follows. First, we will be interested in the properties of the limit, which this
time is not a number, but a whole function f (x). Second, this section should be
considered as an introduction to the next one, in which there will appear the concept
of a uniform convergence. After examining both of them, the conclusions regarding
the limit of such sequences (understood again as a function) that can be drawn in
both cases will become clear.
The sequence in the text of the problem will then be treated as a usual sequence
of numbers with two parameters: x and a. When x = 0, this is very simple, since
one has fn (0) = 1/na and the limit is
0 for a > 0,
lim fn (0) = (15.1.2)
n→∞ 1 for a = 0,
The sequence behaves differently for different values of a, so all important cases
need to be considered separately.
• a > −1/2. In this case the exponent of the variable n in denominator is positive,
which means
• a < −1/2. Here the exponent is negative and the limit does not exist for any
x > 0.
• a = −1/2. Now the terms of the sequence simplify to
1
fn (x) = √ √ , (15.1.5)
x + 1/n + x
1
lim fn (x) = √ . (15.1.6)
n→∞ 2 x
15.1 Finding Limits of Sequences of Functions 345
Problem 2
Solution
For each fixed value of x the limit of (15.1.7) can be found by the methods explored
in Exercise 3 in Sect. 5.1. In order to make the required estimates easy, several cases
will be considered, depending on the variable x.
• x = 0. For this value of x all terms equal zero (fn (0) = 0, regardless of n), so
the limit is f (0) = 0.
• 0 < x ≤ 1/4. In this interval, one has
which allows the use of the squeeze test in the following way:
√ √
n √ √
n
x= x = 0 + x n < fn (x) = 4n x 2n + x n ≤ n x n + x n = 2 · x.
n n n
√ (15.1.9)
Since n 2 −→ 1, the sequences on both sides converge to x. For considered
n→∞
values of x, the squeeze test gives
In this way, the entire form of the limiting function f (x) has been obtained:
x for x ∈ [0, 1/4],
f (x) = (15.1.14)
4x 2 for x ∈]1/4, ∞[.
In contrast to the situation faced in the previous exercise, this function is now
continuous, which can easily be verified with the methods of Sect. 8.2. There
appears, however, another puzzling property that will be explored below. All of
the functions fn (x), as compositions of polynomials and a root function, are
differentiable for every x > 0 and in particular at the point x = 1/4. Let us check
now if the resulting limiting function f (x) is also differentiable there. This is the
“gluing point,” so one can proceed as in Sect. 9.2 where we learned how to examine
these types of functions. First, we write the difference quotient
f (1/4 + h) − f (1/4)
Qh := . (15.1.15)
h
When h < 0, 1/4 + h < 1/4, the upper formula of (15.1.14) is used. We have then
1/4 + h − 1/4
Qh = = 1 −→ 1. (15.1.16)
h h→0
15.2 Examining Uniform Convergence of Functional Sequences 347
The left derivative at this point is, therefore, equal to 1. In order to find the right
derivative and determine if it is again equal to 1—which is required for it to be a
differentiable function—one takes h > 0 in (15.1.15) and use the lower formula
in (15.1.14):
4 (1/4 + h)2 − 1/4 4 1/16 + 1/2 · h + h2 − 1/4
Qh = =
h h
1/4 + 2h + 4h2 − 1/4
= = 2 + 4h −→ 2. (15.1.17)
h h→0
The right derivative does exist and equals 2. Consequently the function f (x) is
not differentiable at this point, as derivatives at both sides differ. We see again that
the requirement of only pointwise convergence of the functional sequence is too
weak to expect the property of continuity or differentiability of all fn (x)’s to be
automatically transferred to the limiting function. These observations lead us to the
concept of the uniform convergence dealt with below.
Problem 1
Solution
Let us first recall the definition of the uniform convergence of the sequence of
functions fn : D → R, formulated already at the beginning of this chapter. It
takes place if
In comparison with the definition of the normal (i.e., pointwise) convergence, there
appears now one important change: it is the order of quantifiers in (15.2.2). The
symbol ∃N ∈N is now put before ∀x∈D , which means that the chosen value of N is to
be common for all x ∈ D.
348 15 Investigating the Convergence of Sequences and Series of Functions
The difference between the pointwise and uniform convergence will be traced
below. At the beginning, let us consider the former, which means that we can simply
examine the convergence of fn (x) for each fixed value of x separately. It is very easy
to find that
x 2
∀x∈]0,1[ f (x) := lim fn (x) = lim n · √ · (1 − x 2 )n = 0, (15.2.3)
n→∞ n→∞ 1−x 2
since we are dealing with the product of a power factor (i.e., n) and an exponential
2
with a base less than one (i.e., (1 − x 2 )n ). As we know, the value of the limit is
dictated by the latter. The sequence of functions is then pointwise convergent, and
the limiting function is identically equal to zero (i.e., for all x ∈]0, 1[). In addition,
for each n, one has fn (0) = 0, so f (0) = 0.
Now we are going to use the definition (15.2.2) to check whether the uniform
convergence takes place too. If certain > 0 is fixed, then for each x and n > N, one
must have |fn (x) − f (x)| = |fn (x)| < , if sufficiently large N had been chosen.
Since this condition must be met by all x, this is enough to focus our attention on the
least favorable situation, when |fn (x)−f (x)| = |fn (x)| reaches absolute maximum.
If, at these points, the condition referred to is satisfied, then it will be satisfied in all
others too. Functions fn (x) are positive in the interval ]0, 1[ and vanish on its ends.
In addition, they are continuous and differentiable, so maxima can be found, by
calculating fn (x). To make it easier, it is worth writing these functions as
x
(1 − x 2 )n = nx(1 − x 2 )n −1/2 .
2 2
fn (x) = n √ (15.2.4)
1 − x2
Then,
1
fn (x) = n (1 − x ) 2 n2 −1/2
− x · 2x n − 2 2 n2 −3/2
(1 − x )
2
= n(1 − x 2 )n −3/2 1 − x 2 − (2n2 − 1)x 2
2
(15.2.5)
= n(1 − x 2 )n −3/2 1 − 2n2 x 2 ,
2
1 f1 x
2e
f2 x
f3 x
f6 x
f15 x
x
x15 x6 x3 x2 x1 1
It is worth noting that the above expression has the following limit as n → ∞:
2
1 1 n 1 1
lim fn (xn ) = lim 1− 2 = √ · e−1/2 = √ = 0,
n→∞ n→∞ 2 − 1/n2 2n 2 2e
(15.2.7)
where we have used (5.2.43). This limit is nonzero, and, therefore, for small > 0,
one cannot satisfy the inequality |fn (xn )| < , for almost all n.
Let us summarize the results. A uniform convergence would require, for each ,
the existence of a certain universal of N (i.e., common for all x), for (15.2.2) to be
met. However, it has been shown that for every n > N, one can select the argument
x = xn , for which this condition is violated. The functional series (15.2.1) is thus
convergent to f (x) = 0 only in the pointwise sense and not uniformly.
In Fig. 15.1, several first functions of fn (x) are drawn. The points xn for which
functions reach their maxima are marked as well. It can be seen that at these points
the height of the “hump” is not decreasing to zero, and thus the convergence cannot
be uniform. If someone fixed a very small > 0 and an arbitrarily large N, one still
would be able to find xn with index n > N, such that
Problem 2
arctan(nx)
fn (x) = (15.2.9)
x
for x ∈]0, 1[ and for x ∈ [1, ∞[ will be examined.
350 15 Investigating the Convergence of Sequences and Series of Functions
Solution
For all positive values of x it is easy to find the limit of (15.2.9), since one has
π
lim arctan(nx) = , (15.2.10)
n→∞ 2
which implies that the limiting function f (x) = π/(2x). It remains then only to
decide whether this convergence is uniform or simply pointwise. To this end let us
estimate the difference |fn (x) − f (x)|:
arctan nx
π 1 π
|fn (x) − f (x)| = − = arctan nx − . (15.2.11)
x 2x x 2
We are going to use here the well-known identity (true for positive values of y):
1 π
arctan y + arctan = , (15.2.12)
y 2
1 1
|fn (x) − f (x)| = arctan . (15.2.13)
x nx
Omitting the absolute value symbol, the fact that the function arctan of positive
arguments is also positive was used.
Now let us look more closely at the obtained result (15.2.13). If x > 0 is fixed,
then choosing a very large n, one is able to make the right-hand side as small as we
wish. This is due to the fact that
arctan y −→ 0.
y→0
However, this sequence of our actions, i.e., fixing x at the beginning and only then
selecting the appropriate value of N , proves only the pointwise convergence, already
established in the first part of the solution. We are now interested rather in the
uniform convergence.
Let us look again at (15.2.13). Now x is not fixed. Taking arbitrarily large n and
then approaching with x to zero, one can make the right-hand side as large as we
wish since 1/x −→ ∞, and the inverse tangent function for large arguments tends
x→0+
to π/2. Or, alternatively, one can put x = xn = 1/n ∈]0, 1[. Then,
15.2 Examining Uniform Convergence of Functional Sequences 351
f x
f5 x
5
4 f4 x
f x
3 3
f x
2 2
f x
1 1
x
1 1
2
1 1 nπ
|fn (xn ) − f (xn )| = arctan = n arctan 1 = −→ ∞. (15.2.15)
xn nxn 4 n→∞
The convergence on the interval ]0, 1[ (or any other interval ]0, a[, where a > 0
(including infinity)) cannot be uniform because no universal (common for all x) N ,
required by the definition (15.2.2), exists. The situation one is dealing with is shown
in Fig. 15.2.
The limiting function f (x) goes to +∞, when x → 0+ . In turn, each of the
functions fn (x) has at zero a specific and finite limit:
arctan(nx)
lim fn (x) = lim = n. (15.2.16)
x→0+ x→0+ x
This is due to the fact that if one denotes nx = tan y, then the limit x → 0+
corresponds to y → 0+ , and one can then write
arctan(nx) arctan(tan y) y
lim = lim = n lim
x→0+ x y→0 + 1/n · tan y +
y→0 tan y
y
= n lim cos y = n · 1 · 1 = n. (15.2.17)
y→0+ sin y
The difference (15.2.13) can actually be made large (although it would suffice to be
larger than a certain positive number) when approaching with x sufficiently close to
zero. In this interval, the sequence of functions does not converge uniformly.
In the interval [1, ∞[, matters differ. The estimate (15.2.13) can be further
transformed:
1 1 1 1 1
|fn (x) − f (x)| = arctan ≤ arctan = arctan . (15.2.18)
x nx 1 n·1 n
352 15 Investigating the Convergence of Sequences and Series of Functions
y
p
2 f5 x f x
f4 x
f3 x
1 f2 x
f1 x
1
2
x
1 3 6
0
Fig. 15.3 The convergence of the sequence (15.2.9) in the interval [1, ∞[
We have used here the fact that both the factor 1/x and arctan(1/(nx)) are
decreasing in the variable x, so the largest value of their product corresponds to
x = 1. Note that if one had wished to use an analogous estimate in ]0, 1[, then in
place of x one would have had to put 0, obtaining a completely useless estimate:
To satisfy the definition (15.2.2), one can now choose N > cot (independent
of x !) and we get (15.2.14). In the interval [1, ∞[, the convergence is, therefore,
uniform. Moreover, the same can be told about each interval of the type [a, ∞[,
where a > 0. The situation one is dealing with is shown in Fig. 15.3.
In the least favorable point, i.e., for x = 1, the limit equals π/2. It is also clear
that one has
π
lim fn (1) = lim arctan n = , (15.2.20)
n→∞ n→∞ 2
thanks to which the convergence is uniform.
Problem 1
will be examined, for x ∈]0, a[ and for x ∈ [b, ∞[, where a, b > 0.
15.3 Examining Uniform Convergence of Functional Series 353
Solution
n
Sn (x) := fk (x). (15.3.3)
k=1
We assume that the pointwise convergence does take place (which eventually can
easily be checked, e.g., by the limit comparison test), and, therefore, there exists the
sum
∞
S(x) := fn (x) for every x ∈ X ⊂ R. (15.3.4)
n=1
A proof would be, therefore, to make sure that one could meet
n
|Sn (x) − S(x)| = fk (x) − S(x) < for n > N (15.3.5)
k=1
for almost all n ∈ N and for any x ∈ X, then the functional series (15.3.2) uniformly
converges on the set X.
354 15 Investigating the Convergence of Sequences and Series of Functions
is convergent, then the identical conclusion may be pulled out of the series (15.3.8).
It is, therefore, the majorant one was looking for and guarantees the uniform
convergence of the series (15.3.1) on the considered interval.
If x ∈]0, a[, things differ. Here, the series is convergent only in weaker
(pointwise) sense, so Weierstrass criterion is useless. One is obliged to directly use
the definition (15.2.2), applying it to partial sums. Since all terms of the series are
positive, any sum of them is larger than a single term. Thanks to this, one has the
following estimate:
n
∞
|Sn (x) − S(x)| = fk (x) − S(x) = fk (x) > fn+1 (x)
k=1 k=n+1
1
= . (15.3.10)
3
1 + (n + 1)4 x
Imagine now that a small has been chosen and an arbitrarily large, but fixed n.
Taking xn = 1/(n + 1)4 , one gets
1 1
|Sn (xn ) − S(xn )| > = √ > . (15.3.11)
· (n + 1)−4
3
3
1 + (n + 1)4 2
As one can see, there is no universal value of N , since always reducing x
accordingly, one can violate the inequality (15.3.5). On the interval ]0, a[, one then
deals only with the pointwise convergence.
15.3 Examining Uniform Convergence of Functional Series 355
Problem 2
Solution
In this exercise, similar methods as those applied in the previous one will be
used. First, we are going to determine whether one has to do with the pointwise
convergence. For this purpose, it is sufficient to consider if (15.3.12) does converge
for each x ∈]0, ∞[. Since x = 0, for large values of n the first factor in (15.3.12)
behaves as x/n. The identical behavior has also the arctan function: arctan(x/n) ∼
x/n.
We expect, therefore, that the terms of the series decrease with n as x 2 /n2 , i.e.,
fast enough to guarantee the convergence of (15.3.12). The above reasoning is not
strict, but it gives a hint about which comparative series to choose for the limit
comparison test. The best choice is naturally
∞
1
, (15.3.13)
n2
n=1
convergent due to the fact that the exponent of n in the denominator is greater than
1. The parameter ξ defined in (13.2.11) equals
nx/(x 2 + n2 ) · arctan(x/n) nx x
ξ = lim = lim 2 n2 arctan
n→∞ 1/n2 n→∞ x + n2 n
n2 x arctan(x/n) x
= lim · x =1· · x · 1 = x2
n→∞ n2 x 2 /n2 + 1 x/n 0+1
(15.3.14)
i.e., neither zero nor infinity. In this case the criterion used clearly tells that the series
of (15.3.12) is pointwise convergent.
Let us consider now the uniform convergence. First we assume that x ∈]0, 1[.
We will try to use the Weierstrass M-test and find a majorant, as was shown in the
condition (15.3.6). To this end, let us note that the functions
nx x
fn (x) := arctan (15.3.15)
x2 +n 2 n
356 15 Investigating the Convergence of Sequences and Series of Functions
are strictly increasing (in the considered interval), which can be established by
calculating the derivative:
nx x nx + (n2 − x 2 )arctan(x/n)
f n (x) = arctan =n . (15.3.16)
x 2 + n2 n (x 2 + n2 )2
For 0 < x < 1, both numerator and denominator are positive, i.e., one has f (x) >
0. Consequently in this interval, for all n, the following inequality holds:
n 1 n 1 1 1 1
fn (x) < fn (1) = arctan < arctan ≤ · = 2 . (15.3.17)
1+n2 n 0+n2 n n n n
arctan y ≤ y,
easy to be obtained from (5.1.39). Thus a majorant has been found and on the
interval x ∈]0, 1[ the convergence is uniform.
The situation is different in the interval [1, ∞[. Here the derivative (15.3.16)
changes its sign, and moreover, the interval is unbounded, so it would not be easy
to find an estimate similar to (15.3.17). For the rest, as will be found out soon,
such an estimate simply does not exist. Therefore, we are going to use at this
point the definition (15.3.5) and the fact that all functions fn (x) are positive for
positive arguments. We were proceeding similarly in the previous exercise in order
to get (15.3.10)
n
∞
|Sn (x) − S(x)| = fk (x) − S(x) = fk (x) > fn+1 (x)
k=1 k=n+1
(n + 1)x x
= arctan . (15.3.18)
x2 + (n + 1)2 n+1
Let us rewrite this estimate, substituting certain specific values for the argument
xn = n + 1 ∈]1, ∞[:
(n + 1)2 n+1
|Sn (n + 1) − S(n + 1)| > arctan
(n + 1)2 + (n + 1)2 n+1
1 1 π π
= arctan 1 = · = . (15.3.19)
2 2 4 8
It is obvious that the right-hand side—and, therefore, the left-hand side too—cannot
be made arbitrarily small by choosing large N and considering only n > N, since
it is a constant, independent of n. The convergence in this interval has then only the
pointwise character.
15.4 Calculating Sums of Series 357
It is worth noticing at the end that if, instead of [1, ∞[, an interval [1, a]
was considered, where a is any number greater than 1 (even very large), the
uniform convergence could be proved. By selecting N > a, the positivity of the
derivative (15.3.16) in this interval for all n > N would be guaranteed and one
could write an inequality similar to (15.3.17)
an a an a 1 a a2
fn (x) < fn (a) = arctan < arctan ≤ a · · = .
a 2 + n2 n 0 + n2 n n n n2
(15.3.20)
It can be seen that a majorant on this interval does really exist and has the form
∞
a2
, (15.3.21)
n2
n=1
Problem 1
Solution
As we know from the lecture of analysis, power series within their domain of
convergence are uniformly convergent and may be differentiated or integrated term
by term. This property will be used to find the value of the sum (15.4.1) in the
following way. At the beginning, let us define a new function f (x) with the formula
∞
x 2n+1
f (x) := . (15.4.2)
(2n + 1)(n + 1)
n=0
358 15 Investigating the Convergence of Sequences and Series of Functions
The right-hand side is just a power series and its domain of convergence is the
interval ] − R, R[, where R is given by the well-known expression
) *−1
1
R= lim 2n+1
= 1. (15.4.3)
n→∞ (2n + 1)(n + 1)
The degree of the root (i.e., 2n + 1) is dictated by the power of x in (15.4.2), since
the coefficient accompanying x 2n+1 should be in fact called a2n+1 and not an . The
above value (i.e., R = 1) is easy to obtain, as
(
1 1 1 1
· = 2n+1 √ · 2n+1
√
2n+1
.
2n + 1 n + 1 2n + 1 n+1
√ √
The sequence of terms 1/ 2n+1 2n + 1 is simply a subsequence of 1/ n n, which has
the limit
√ equal to 1 (just as all of its subsequences). In turn from the squeeze rule
1/ 2n+1 n + 1, also goes to 1, since
1 1 1
1 ←− √ < √ < √
n→∞ 2n + 1
2n+1 2n+1
n+1 n + 1/2
2n+1
√
2n+1
1 2 1
= 2n+1
√ = 2n+1
√ −→ = 1,
1/2(2n + 1) 2n + 1 n→∞ 1
is convergent too, then the value of the latter can be found by calculating lim S(x).
x→R −
In other words, the function S(x) is, in fact, defined on the set ] − R, R], and at the
point x = R, it is (left) continuous.
15.4 Calculating Sums of Series 359
Someone could ask a question, why should it be easier to find a sum (15.4.2) than
to find the initial one, given in the content of the exercise. It apparently seems that
including certain powers of x under the infinite sum the expression was made more
complex than before. Of course, this is true, but for the price of this complexity, a
very effective tool has been gained which could not be applied to the series (15.4.1):
it is now possible to differentiate and integrate (15.4.2) over x. At this point, the
question of why, when defining (15.4.2), we put x 2n+1 instead of x n finds its answer.
The exponent was simply selected in such a way that, after having differentiated over
x, one of the factors in the denominator canceled:
)∞ * ∞
x 2n+1 x 2n+1
f (x) = =
(2n + 1)(n + 1) (2n + 1)(n + 1)
n=0 n=0
∞
∞
(2n + 1)x 2n x 2n
= = . (15.4.4)
(2n + 1)(n + 1) n+1
n=0 n=0
The property that a series (within its range of convergence) may be differentiated
term by term has been used here. With this operation, the expression under the sum
has been significantly simplified. One has a power series with the same domain
of convergence (which can easily be checked) and thus again the property of
differentiability may be used. We now want to get rid of the second factor in the
denominator. However, the automatic differentiation of (15.4.4) will not lead to our
purpose, since in the result one would obtain under the sum the fraction 2n/(n + 1),
which cannot be simplified. Therefore, one must first change the power of x from 2n
to 2n + 2 = 2(n + 1), by multiplying f (x) by x 2 :
)∞ * ∞ 2n+2 ∞
x 2n+2 x (2n + 2)x 2n+1
2
[x f (x)] = = =
n+1 n+1 n+1
n=0 n=0 n=0
∞
∞
=2 x 2n+1 = 2x x 2n . (15.4.5)
n=0 n=0
As a result of these operations, one has a geometrical series with the common ratio
x 2 < 1. We are now able to immediately find the sum of it, finally obtaining the
equation
2x
[x 2 f (x)] = . (15.4.6)
1 − x2
Our goal was, however, to find f (x). The next step is then the integration of both
sides of the above equation, which leads to the result (remember that |x| < 1):
+ +
2x
x 2 f (x) = dx = − log(1 − x 2 ) dx = − log(1 − x 2 ) + C1 .
1 − x2
(15.4.7)
360 15 Investigating the Convergence of Sequences and Series of Functions
Let us now put x = 0 on both sides. This point belongs to the domain of convergence
both of the function f and of its derivative f . From the formula (15.4.7)
in connection with (15.4.4), one can see that C1 = 0. This fact will be taken into
account in the subsequent steps.
In this way, we have found f (x), which will now be used to find the function
f (x) itself. The formula
1
f (x) = − log(1 − x 2 ) (15.4.8)
x2
may not, however, be used for x = 0. On the other hand, one knows from (15.4.4)
that f (0) does exist and it is equal to unity. So one should formally write
⎧
⎨1
−
log(1 − x 2 ) for x ∈] − 1, 0[ ∪ ]0, 1[,
f (x) = x2 (15.4.9)
⎩1 for x = 0.
This function is both continuous and differentiable (including 0), as can be easily
seen using the methods of Sects. 8.2 and 9.2. It had to be so because it is given by the
power series. Formal separating of the point x = 0 has no effect on the subsequent
integrations. One can now find the function f with the use of the formula
+
1
f (x) = − log(1 − x 2 ) dx. (15.4.10)
x2
We have then
1 1+x
f (x) = log(1 − x 2 ) + log + C2 . (15.4.12)
x 1−x
15.4 Calculating Sums of Series 361
lim (1 − x) log(1 − x) = 0,
x→1−
resulting from (10.4.18) and based on both a new variable y = 1 − x and the idea
that the limit x → 1− corresponds to y → 0+ .
Problem 2
will be examined, depending on x, and in the domain of its convergence, its sum
will be found.
362 15 Investigating the Convergence of Sequences and Series of Functions
Solution
We start with delimiting the domain of convergence. As one knows, for this goal,
one has to calculate the quantity
−1
R= lim n
|an | , (15.4.16)
n→∞
n √ 3 √ 3
lim n3 = lim n n = lim n n = 13 = 1, (15.4.17)
n→∞ n→∞ n→∞
which implies that R = 1, and the series S(x) is convergent for x ∈] − 1, 1[.
In order to find the value of (15.4.15), we first define the function:
∞
x
f (x) := xn = , (15.4.18)
1−x
n=1
where the formula for the sum of the geometric series has been used. We are now
going to try to carry out such operations on this expression as to obtain S(x). The
power series (15.4.18) is convergent for x ∈]−1, 1[ too and for all arguments within
this interval it can be differentiated or integrated term by term.
As one can see in (15.4.15), under the sum the factor n3 is needed. It can be
obtained by threefold differentiation of the function f . After having taken each
derivative, one has to multiply both sides by x in order to bring back the power of x
to the value n. Each subsequent series obtained as a result of differentiation is again
a power series of the same domain of convergence, so all of them can similarly be
differentiated term by term. By doing so, one finds
) ∞
* ∞ ∞ ∞
# n $
f (x) = x n
= x = nx n−1 ⇒ xf (x) = nx n .
n=1 n=1 n=1 n=1
(15.4.19)
Subsequent differentiations allow us to obtain still higher powers of n:
)∞ * ∞ ∞
# $ # $
xf (x) = nx n
= nx n = n2 x n−1
n=1 n=1 n=1
∞
# $
⇒ x xf (x) = n2 x n , (15.4.20)
n=1
15.5 Exercises for Independent Work 363
)∞ *
# $ ∞
∞
x xf (x) = n2 x n = n2 x n = n3 x n−1
n=1 n=1 n=1
# $ ∞
⇒ x x xf (x) = n3 x n = S(x). (15.4.21)
n=1
In this way, the differential formula for the sum S(x) is obtained:
# $ x
S(x) = x x xf (x) , where f (x) = , (15.4.22)
1−x
Finally,
# $ x(1 + 4x + x 2 )
S(x) = x x xf (x) = . (15.4.23)
(1 − x)4
At the end, let us stress again: the operations such as integration or differentiation
of a series term by term could be easily done in the last two exercises, since we were
dealing with power series. Such series, together with their derivatives, inside their
domains of convergence are uniformly convergent. For other series, before applying
these types of operations one has to make sure that this property in fact takes place.
nx
(d) fn (x) = on ]0, 1[ and on ]1, ∞[.
1 + n2 x 4
1
(e) fn (x) = on R.
1 + x 2 /n2
1
(f) fn (x) = n sin on ]1, ∞[.
nx
Answers
(a) Nonuniformly convergent to the function f (x) = 1, 1/2, 0 for x < 1,
x = 1, and x > 1 correspondingly.
(b) Nonuniformly convergent on ]0, 1[ and uniformly convergent on ]1, ∞[
to the function f (x) = π/(2x).
(c) Nonuniformly convergent to the function f (x) = 0.
(d) Nonuniformly convergent on ]0, 1[ and uniformly convergent on ]1, ∞[
to the function f (x) = 0.
(e) Nonuniformly convergent to the function f (x) = 1.
(f) Uniformly convergent to the function f (x) = 1/x.
Answers
(a) Uniformly convergent.
(b) Uniformly convergent.
(c) Nonuniformly convergent on [0, ∞[, uniformly convergent on [a, ∞[.
(d) Nonuniformly convergent.
15.5 Exercises for Independent Work 365
Answers
√
(a) 1◦ . 3π/9 + log 2/3. 2◦ . 3e2 .
(b) 1◦ . 1 + (1/x − 1) log(1 − x). 2◦ . π/2 − 1 − log 2.
Index
A Codomain, 31, 39
Absolutely convergent series, 272, 294 Compact set, 131–141
Alternating series, 272, 287, 289, 294 Comparison test, 273, 277, 279, 280, 292,
Anharmonic series, 272, 294 353–355
Arithmetical mean, 63 Complement of set, 2, 4, 5, 8
Asymptote, 261, 262, 265–267 Concave function, 263, 268
Concavity, 262, 263, 268
Conditionally convergent series, 272
B Connected set, 32, 38, 44, 45, 132
Ball, xi, 51, 56–62, 131, 133, 137, 255, 343 Continuity, 96, 100, 126, 131, 135, 143, 154,
Bernoulli’s inequality, 76, 82, 277 159–180, 182, 185, 186, 189, 227, 248,
Bijection, 31, 38–44, 49 281, 345, 347, 361
Binomial coefficient, 64, 77, 78, 234 Continuous function, 32, 44, 136, 148, 159,
Boundary point, 131 169, 186, 204, 296, 299, 342, 343
Bounded sequence, 89, 90, 117–119, 275, 276 Convergent sequence, 90, 91, 140, 278, 341
Bounded set, 18, 19, 37, 140 Convergent series, 272, 273, 279, 289, 293,
342, 353, 354
Convex function, 83–85
C Convexity, 86, 213, 262, 263, 268
Cartesian product, 2, 15, 22
Cauchy’s condensation test, 107, 203, 286
Cauchy’s definition of the continuity, 159, 162, D
166, 175 d’Alembert’s test (criterion) for sequences,
Cauchy’s definition of the limit, 143–145 282
Cauchy’s form of a remainder, 245, 246 d’Alembert’s test (criterion) for series, 282
Cauchy’s test (criterion) for sequences, 105 Darboux’s theorem, 32, 38, 44, 136
Cauchy’s test (criterion) for series, 285 Decreasing function, 32, 288
Chain rule, 182 Decreasing sequence, 89, 277, 288
Characteristic equation, 116 Deleted neighborhood, 131, 145, 201, 221, 223
Class of a function, 24 de Morgan’s law, 2, 7, 8, 10, 11
Closed ball, xi, 51, 56, 60 Derivative, 181–231, 233–252, 254–257, 259,
Closed set, 131–136, 138, 140 262, 263, 267, 268, 301, 303, 304, 308,
Closure, 132 310, 316, 317, 326, 334, 342, 347, 356,
Cluster point, 90, 91, 99, 123, 125, 127–129, 357, 360, 362, 363
132, 133, 137–140, 143 Difference of sets, 4, 7
Difference quotient, 181, 183, 184, 186, 188, Integrand, 299–304, 307, 308, 317, 320, 322,
189, 196, 197, 281, 346 323, 325, 327, 329, 330, 332–336, 340
Differentiable function, 181, 183, 186, 190, Integration by parts, 301, 305–307, 314
195, 210, 218, 219, 253, 254, 299, 304, Integration by substitution, 307, 310
342, 347 Interior, 15, 131, 133
Disconnected set, 132 Internal points, 131, 133, 134
Divergent sequence, 124, 125 Intersection of sets, 5
Divergent series, 271–273, 279 Interval, 15, 32, 98, 132, 160, 182, 201, 233,
Domain, xi, 31, 36–41, 47, 90, 136, 143, 159, 253, 288, 299, 349
165, 166, 170–172, 181, 193, 207, 253, Inverse function, 31, 32, 38–45, 191–195, 200
260–264, 267, 268, 336, 357–363 Inverse image of a function, 32, 33, 44–46, 136
Irrational number, 173, 295
E
Equivalence classes, 3, 21–29 J
Equivalence relation, 3, 21–29 Jensen’s inequality, 83
Euler’s substitution, 329–335
Even function, 161
Extremal limits of a sequence, 90, 91 L
Lagrange’s form of a remainder, 242, 244–246
Lagrange’s theorem (= mean value theorem),
F 201
Function (= mapping), xi, 31 Left continuity, 182
Left derivative, 182, 347
Left limit, 143
G Leibnitz’s test, 289
Geometrical mean, 63, 181 Leibniz formula, 234, 235, 237
Global extreme, 253 Level set, 31–44, 46–49
Global maximum, 253 l’Hospital’s rule, 201, 220–230, 246, 248, 281
Global minimum, 253 Limit comparison test, 279, 280, 292, 353, 354
Limit of a function, 143, 144, 152, 158, 228
Limit of a sequence, 296
H Limit point, 132, 146
Harmonic mean, 64, 273 Local extreme, 253
Harmonic series, 271–274, 286, 289, 292 Local maximum, 253, 254
Heine’s definition of the continuity, 163, 173 Local minimum, 253, 254
Heine’s definition of the limit, 100, 143, 144, Lower bound, 2, 18–20, 37, 127
146, 154
M
I Maclaurin series, 234, 280
Image (range) of a function, x, 31, 47 Maclaurin’s formula, 242, 252
Image (range) of a set, 41, 44 Mathematical induction, 63–87, 121, 177, 237,
Increasing function, 32, 278, 288 251
Increasing sequence, 283 Metric (distance), 51–62, 131, 132, 136–138,
Indefinite integral (antiderivative, primitive 205, 238
function), 299–347 Metric space, 51, 52, 54, 131
Inductive hypothesis, 63, 65, 66, 71, 73, 74, Monotonic function, 32
76–77, 80, 81, 84, 85, 235, 239 Monotonic sequence, 117, 118, 286, 287
Inductive thesis, 63, 65, 66, 68, 69, 71–82, 85,
87, 122, 235–237, 239, 240
Infimum, 2, 18, 19 N
Infinite sequence, 89, 90, 139, 140, 341 Natural number, 40, 41, 66, 90, 157
Injection, 31, 38–44, 49 Necessary condition of the convergence of a
Integer number, xi series, 271
Index 369