Documente Academic
Documente Profesional
Documente Cultură
An Excursion
through Elementary
Mathematics,
Volume I
Real Numbers and Functions
Problem Books in Mathematics
Series Editor:
Peter Winkler
Department of Mathematics
Dartmouth College
Hanover, NH 03755
USA
An Excursion through
Elementary Mathematics,
Volume I
Real Numbers and Functions
123
Antonio Caminha Muniz Neto
Mathematics
Universidade Federal do Ceará
Fortaleza, Ceará, Brazil
This is the first of a series of three volumes (the other ones being [4] and [5])
devoted to the mathematics of mathematical olympiads. Generally speaking, they
are somewhat expanded versions of a collection of six volumes, first published
in Portuguese by the Brazilian Mathematical Society in 2012 and currently in its
second edition.
The material collected here and in the other two volumes is based on course
notes that evolved over the years since 1991, when I first began coaching students
of Fortaleza to the Brazilian Mathematical Olympiad and to the International
Mathematical Olympiad. Some 10 years ago, preliminary versions of the Portuguese
texts also served as textbooks for several editions of summer courses delivered at
UFC to math teachers of the Cape Verde Republic.
All volumes were carefully planned to be a balanced mixture of a smooth and
self-contained introduction to the fascinating world of mathematical competitions,
as well as to serve as textbooks for students and instructors involved with math clubs
for gifted high school students.
Upon writing the books, I have stuck myself to an invaluable advice of the
eminent Hungarian-American mathematician George Pólya, who used to say that
one cannot learn mathematics without getting one’s hands dirty. That’s why, in
several points throughout the text, I left to the reader the task of checking minor
aspects of more general developments. These appear either as small omitted details
in proofs or as subsidiary extensions of the theory. In this last case, I sometimes
refer the reader to specific problems along the book, which are marked with an *
and whose solutions are considered to be an essential part of the text. In general, in
each section, I collect a list of problems, carefully chosen in the direction of applying
the material and ideas presented in the text. Dozens of them are taken from former
editions of mathematical competitions and range from the almost immediate to real
challenging ones. Regardless of their level of difficulty, we provide generous hints,
or even complete solutions, to virtually all of them.
This first volume concentrates on real numbers, elementary algebra, and real
functions. The book starts with a non-axiomatic discussion of the most elementary
properties of real numbers, followed by a detailed study of basic algebraic identities,
vii
viii Preface
have had the courage to embrace the task of translating more than 1500 pages from
Portuguese into English. I acknowledge all the staff of Springer involved with this
project in his name.
Finally, and mostly, I would like to express my deepest gratitude to my parents
Antonio and Rosemary, my wife Monica, and our kids Gabriel and Isabela. From
early childhood, my parents have always called my attention to the importance of
a solid education, having done their best for me and my brothers to attend the
best possible schools. My wife and kids filled our home with the harmony and
softness I needed to get to endure on several months of solitary nights of work
while translating this book.
xi
xii Contents
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 511
A Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 513
Index . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 643
Chapter 1
The Set of Real Numbers
This first chapter recalls some definitions and results which are essential to all
further developments. We assume from the reader a modest acquaintance with the
most basic concepts of set theory; we also assume that he or she is familiar with the
sets of naturals,
N D f1; 2; 3; 4; : : :g;
integers,
and rationals,
na o
QD I a; b 2 Z; b ¤ 0 ;
b
as well as with the elementary arithmetic operations within these sets.
In what concerns the integers1, given a; b 2 Z, with a ¤ 0, we say that a divides
b if there exists an integer c such that b D ac. Equivalently, to say that a divides b is
the same as to say that the rational number ba is an integer; for example, 13 divides
52, since 52
13
D 4.
If a divides b, we also say that a is a divisor of b, or that b is divisible by a; in
such a case, we denote a j b. If a doesn’t divide b (or, which is the same, if ba … Z),
we denote a − b. An integer n is even if 2 j n; otherwise, n is said to be odd. Hence,
0, ˙2, ˙4, ˙6, . . . are the even integers, while ˙1, ˙3, ˙5, . . . are the odd ones.
1
We refer the reader to Chap. 6 of [5] for a systematic discussion of what follows.
Given natural numbers a and b, it is well known that there exist unique integers
q and r satisfying the following conditions:
The above relation is known as the division algorithm, and the integers q and r
are respectively called the quotient and the remainder of the division of b by a.
Conditions (1.1) are usually condensed in the diagram
b a
r q
In particular, for given naturals a and b and in the above notations, a j b is the
same as r D 0 and q D ba .
Two nonzero integers a and b always have a greatest common divisor, which
will be denoted gcd.a; b/; Moreover, a and b are said to be relatively prime if
gcd.a; b/ D 1; in particular, if b D ka C 1, then a and b are relatively prime.
If r D mn , with m and n integers, is a representation of the nonzero rational
r, then, by cancelling out factors common to m and n (i.e., cancelling gcd.m; n/
out of m and n), we get an irreducible representation for r. For example, since
gcd.12; 18/ D 6, the rational number 12 2
18 has 3 as an irreducible representation,
which was obtained by cancelling out a factor 6 from both 12 and 18.
An integer p > 1 is prime if 1 and p are its only positive divisors; in another way,
an integer p > 1 is prime if, for a 2 N, we have that
p
2 N ) a D 1 or p:
a
It is a well known fact (cf. Chap. 6 of [5]) that the set of prime numbers is infinite.
Below, we list all prime numbers less than 100:
2; 3; 5; 7; 11; 13; 17; 19; 23; 29; 31; 37; 41; 43; 47; 53;
An integer greater than 1 and which is not prime is said to be composite. It is also a
well known fact (cf. Chap. 6 of [5]) that every natural number n > 1 can be written
(or decomposed) as a product of a finite number of powers2 of prime numbers
(its prime factors); also, such a representation of n is unique, up to the order of the
powers. For instance, 9000 D 23 32 53 is the decomposition of 9000 as a product of
powers of primes. The existence and uniqueness of such a decomposition of n > 1
is known as the Fundamental Theorem of Arithmetic.
2
To recall the definition and the main properties of powers of numbers, we refer the reader to
Section 1.2.
1.1 Arithmetic in R 3
1.1 Arithmetic in R
We use to represent rational numbers in decimal notation. For the rational number
1 1
8 , for example, we write 8 D 0:125 as a shorthand for the equality
1 1 2 5
D C 2 C 3;
8 10 10 10
1
D 0:08333 : : : :
12
What does this equality mean? Arguing as in the case of 18 , we are tempted to say
that this equality is a shorthand for
1 8 3 3 3
D 2 C 3 C 4 C 5 C : (1.2)
12 10 10 10 10
This is actually so, provided we correctly interpret the sum with an infinite number
of summands at the right hand side. Rigorously, (1.2) means that, if we fix the a
priori maximum error 101n D 0: „00 ƒ‚
: : : 01
…, then we have
n
1 8 3 3 3 1
0< 2
C 3 C 4 CC k
12 10 10 10 10 10n
for every natural number k n; yet in another way, (1.2) means that all of the
numbers 1082 C 1033 C 1034 C C 103k , with k n, approximate 12 1
by defect with
error less than or equal to 0: „
00 ƒ‚
: : : 01
… . In fact, it will follow from Proposition 3.12
n
that
1 8 3 3 3 1
C 3 C 4 CC k D ;
12 102 10 10 10 3 10k
so that the error in the defect approximation 1082 C 1033 C 1034 C C 103k of 12
1
equals
1 1
310k
, which is always less than or equal to the maximum error 10n , whenever k n.
1
It is precisely in this sense that we should think of the equality 12 D 0:08333 : : :.
4 1 The Set of Real Numbers
In the light of the above discussion, one question suggests itself naturally: for an
arbitrary sequence3 .a1 ; a2 ; a3 ; : : :/ of decimal digits, can we think of
0:a1 a2 a3 : : :
1
(In particular, for the rational number 12 the list is .0; 8; 3; 3; 3; : : :/, which is clearly
periodic.) Therefore, if we are able to exhibit a list of digits that is not periodic from
any point on, we will conclude that the general answer to the question posed above
is no! We show an example of such a list now.
Example 1.1 The sequence of digits .0; 1; 0; 1; 1; 0; 1; 1; 1; 0; : : :/, with infinitely
many 0’s and such that the quantity of digits 1’s after each digit 0 equals the previous
quantity of digits 1 plus one, is not periodic from any point on.
Proof By the sake of contradiction, suppose that the list in the statement is periodic
from some point on, with, say, a block of p digits that repeats itself. The way we
defined the sequence assures that, from some point on, each occurrence of a block of
digits 1 would bring more than p digits 1. Therefore, the block of digits that repeats
itself should be composed only by digits 1, so that the list should be, from some
point on, equal to .1; 1; 1; 1; 1; 1; 1; : : :/. However, if this was so, then we could not
have an infinite number of digits 0, which is a contradiction. t
u
In short, the previous discussion points to the following deficiency of the set
of rationals: every rational number admits a decimal representation, but not every
decimal representation represents a rational number. At this point, in order to fulfill
this deficiency, we postulate4 the existence of a set R, containing Q and having the
following properties:
3
As we shall see in Section 6.1, a sequence of real numbers is just a function f W N ! R; however,
for our purposes here, we can think of a sequence just as an ordered list of numbers, i.e., a list of
numbers in which we specify which is the first number, which is the second, third, and so on. We
will sistematically discuss some important elementary sequences of real numbers in Chap. 3.
4
An axiom or postulate in a certain theory is a property imposed as true. One of the fundamental
characteristics of Mathematics as a branch of human knowledge is the use of the axiomatic
method, i.e., the acceptance of the fact that not every mathematical property can be logically
deduced from previously established properties, being necessary the adoption of an adequate set
of axioms.
1.1 Arithmetic in R 5
5
There are more construtive (and rigorous) ways of introducing the real numbers, as the reader
can find in [6], for example. However, in these notes we chose to follow an approach that was as
close as possible to the previous experience of the medium reader, sacrificing rigor in the name of
understanding.
6 1 The Set of Real Numbers
give us
b D b C 0 D b C .a C b0 /
D .b C a/ C b0 D .a C b/ C b0
D 0 C b0 D b0 :
Hence, the real number a has a unique additive inverse, which from now on will
be denoted by a, as is usually done within Q. It follows from a C .a/ D 0
that a is the additive inverse of a; therefore, according to the notation just
established for additive inverses, we have .a/ D a.
(ii) Uniqueness of the multiplicative inverse: for a given a 2 R n f0g, if b; b0 2 R
are such that a b D 1 and a b0 D 1, then b D b0 . The proof of this fact
is completely analogous to that of the previous item (cf. Problem 3, page 7).
From now on, we will denote the multiplicative inverse of a 2 R n f0g by a1 ,
as is also usually done within Q.
(iii) For a 2 R, we have a 0 D 0: in order to check this, let a 0 D e. The
distributivity of multiplication with respect to addition gives
e D a 0 D a .0 C 0/ D a 0 C a 0 D e C e:
Hence,
e D e C 0 D e C .e C .e//
D .e C e/ C .e/
D e C .e/ D 0;
so that a 0 D e D 0.
In view of the above properties, we adopt the convention (as is usual within Q) to
omit the sign of multiplication, thus writing simply ab instead of ab. Now, observe
that the associativity and commutativity of addition and multiplication in R allow us
to add or multiply an arbitrary finite number of real numbers without worrying with
which summands or factors should be initially operated; the final result will always
be the same6 .
We also define the operations of subtraction () and division () in R as is
usually done in Q: for a; b 2 R, we set
6
Rigorously speaking, the validity of such a statement should be proved as a theorem, what can
be done with the aid of the principle of mathematical induction (cf. Chap. 5). However, in order
to have a less terse reading, we chose not to give a formal proof of this fact, just relying on the
previous experience of the reader.
1.2 The Order Relation in R 7
with b ¤ 0 in the last case; yet in the case of division, and whenever there is no
danger of confusion, we write a=b or ab as synonymous for a b.
a c a˙c
D D :
b d b˙d
(f) Suppose c > 0 (the case c < 0 can be treated analogously). It follows from (6’)
and (7’) that
a > b ) a b > 0 ) .a b/c > 0 ) ac bc > 0 ) ac > bc:
(h) Suppose a > 0. If we had 1a < 0, it would follow from (b) that 1 D a 1a < 0,
which is an absurd (for the order relation on R extends that on Q).
(i) Since a and b have equal signs, it follows from (b) and (c) that ab > 0. Thus,
1
item (h) gives ab > 0. Since b a < 0, we get from either (b) or (c) that
1 1 ba 1
D D .b a/ < 0:
a b ab ab
1
Finally, (6’) assures that this is equivalent to a
< 1b .
t
u
Items (b) and (c) of the former proposition are known as the sign rules for
multiplication of real numbers.
Next, let r 2 R be given. The square of r, denoted r2 , is the real number r2 D rr;
the cube of r, denoted r3 , is the real number r3 D r r r. More generally, for a given
n 2 N, we define the nth power of r, denoted rn , as being r, if n D 1, or the real
number obtained by multiplying r by itself n times, if n > 1:
rn D „
r r ƒ‚
…r :
n
Once more we call the attention of the reader to the fact that the associativity of the
multiplication of reals, together with the principle of mathematical induction (cf.
Chap. 5) allows us to prove that the result of the right hand side of the equality above
does not depend on the order in which we multiply the n copies of r. Therefore, rn
is a well defined real number. Problem 3, page 10, lists some useful properties of
the powers of real numbers.
We collect below an important consequence of the properties of the order relation
on R, listed in the last proposition.
Corollary 1.3 Let r be a positive real number and m and n be natural numbers,
with m > n. Then:
(a) 0 < r < 1 ) rm < rn .
(b) r > 1 ) rm > rn .
Proof
(a) Since r is positive, multiplying both sides of the inequality r < 1 by r, we get
r2 < r. Multipliying both sides of this last inequality by r once more, it follows
that r3 < r2 and, hence, that r3 < r2 < r. Proceeding this way, we arrive at the
desired result, i.e.,
(b) The proof of this item is essentially equal to that of (a), with the only difference
that, in this case, we have r > 1.
t
u
We next illustrate, in an example, how the above corollary can be useful in
comparing certain real numbers.
Example 1.4 In order to compare 2100 C 3100 and 4100 , for instance, it suffices to
see that
a2 C b2 0; (1.4)
a2 C b2 C c2 0;
7
In order to prove properties (a) to (d) in a rigorous way, we have to rely on the principle of
mathematical induction (cf. Chap. 5). Thus, our intention here is simply to make the reader give
some arguments on their plausibility.
1.2 The Order Relation in R 11
(b) rmCn D rm rn .
(c) .rm/n D rmn .
n n
(d) rs D rsn , if s ¤ 0.
4. * For r 2 R n f0g and n 2 N, we extend the notion of powers with natural
exponents to powers with integer exponents by defining rn D r1n . For example,
r1 D 1r , r2 D r12 etc. If we also set r0 D 1, prove that, for all m; n 2 N, one
m
has rrn D rmn .
5. Let a and b be nonzero integers, with b > 1. If the only prime divisors of b are
2 or 5, prove that the decimal representation of ab is finite.
6. * If x ¤ 0 is a real number and n 2 N, prove that xn is positive if n is even, and
has the same sign of x if n is odd.
7. Prove that 12 13 C 14 15 C 99
1 1
C 100 > 15 .
8. Given positive real numbers a and b such that a < b, compare (i.e., decide
which is the greatest of) the numbers aC1 aC2
bC1 and bC2 .
9. (TT) We are given ten real numbers such that the sum of any four of them is
positive. Show that the sum of all ten numbers is also positive.
10. Decide which of the numbers 3111 or 1714 is the greatest one.
11. * Let n 2 N and a; b be positive reals. Prove that:
(a) a < b if and only if a2 < b2 .
(b) a < b if and only if an < bn .
(c) an C bn < .a C b/n .
12. (EKMC - adapted) Let a, b and c be the lenghts of the sides of a right triangle,
c being the hypotenuse. Which one is the greatest: a3 C b3 or c3 ? Justify your
answer.
13. Show that, for every n 2 N, we have 12n C 22n C 32n 2 7n .
14. * Find all natural numbers a, b and c such that a b c and 1a C 1b C 1c is an
integer.
15. (IMO) Explain how to write 100 as a sum of naturals whose product is as large
as possible.
16. (Russia)8 The leftmost digit of the decimal representations of the natural
numbers 2n and 5n is the same. Prove that such digit is equal to 3.
17. (Russia) Let a, b, c and d be positive real numbers. Show that, among the
inequalities
8
For a converse to this problem, see Example 10.60.
12 1 The Set of Real Numbers
For the time being, we postulate that to every sequence .a1 ; a2 ; a3 ; : : :/ of digits,
there corresponds a unique x 2 R, in the following sense: for a fixed maximum
error 101n , with n 2 N, we have
a a2 ak 1
1
0x C C C n;
10 102 10k 10
for every natural k n. In particular, it follows from this inequality that
a1 a2 ak 1
x C 2 C C k C n;
10 10 10 10
for every k n. Taking k D n and recalling that aj 9 for every j, we get
a1 a2 an1 an 1
x C 2 CC C nC n
10 10 10n1 10 10
9 9 9 9 1
C 2 C C n1 C n C n
10 10 10 10 10
9 9 9 1
D C 2 C C n1 C n1
10 10 10 10
9 9 1
D C 2 C C n2
10 10 10
D D 1;
Thus, 0 x 1.
As a shorthand to the above postulate, we say that the set R of real numbers is
complete. In this respect, we refer the reader to Section 7.1, as well as to Problem 9,
page 242, where the completeness of the real number system will be more rigorously
discussed.
Conversely, we also postulate that to every positive real number x there cor-
responds a nonnegative integer m and a sequence .a1 ; a2 ; a3 ; : : :/ of digits, such
that x D m C 0:a1 a2 a3 : : :, in the sense of the previous paragraph. If m > 0 and
m D bn : : : b1 b0 , with the bi ’s being their digits, we write
x D bn : : : b1 b0 :a1 a2 a3 : : :
Up to the present, the only irrational number we have met was the number
0;0101101110 : : : (cf. Example 1.1). In a sense, such an example is pretty dis-
satisfying, for it is difficult to manipulate this number (i.e., it is difficult to make
calculations with it). We will remedy this situation in what follows.
From an arithmetic point of view, one great advantage of the set of reals, in
comparison to the set of rationals, is the possibility of making root extractions of
positive real numbers. More precisely, given a real number x > 0 and a natural
number n, it is possible to prove (and we will do so, in two different ways, in
sections 7.1 and 8.3) that there exists a unique positive real number y such that
yn Dpx. From now on, we shall call this real number y the nth root of x, and write
y D n x; the natural n is the index of the root. In short,
p
y D n x , x D yn :
Proposition 1.7 p Given natural numbers n and k, with k > 1, either n is a k-th
perfect power or k n is an irrational number.
p p p
According to the above proposition, numbers like 2, 3 3, 5 10 etc are all
examples of irrationals (for 2 is not a perfect square, 3 is not a perfect cube and
10 is not a perfect 5th power).
Now we can, at least formally (i.e., without worrying with approximations),
operate with several irrational numbers. Let us see an example where we (indirectly)
apply the last proposition to explain why a certain real number is irrational.
p p p p
Example 1.8 The number 2 C 3 is irrational. In fact, if we let r D 2 C 3,
there are two possibilities: r 2 Q or r 62 Q. Suppose, for the sake of contradiction,
that r 2 Q. Then, since the set of rationals is closed with respect to multiplication,
we would have r2 2 Q. On the other hand, the distributivity of multiplication with
respect to addition gives
p p p p
r2 D . 2 C 3/. 2 C 3/
p p p p p p
D 2. 2 C 3/ C 3. 2 C 3/
p p
D .2 C 6/ C . 6 C 3/
p
D 5C2 6
p 2 p
so that 6 D r 25 . Therefore, 6 would be the quotient of the rational numbers
p
r2 5 and 2 and, as such, 6 would be itself a rational number. This contradicts the
result of Proposition 1.7, so that r … Q.
In order to finish our discussion on rational and irrational numbers, note that the
set of irrationals is not closed with respect to the ordinary arithmetic operations.
For instance, for a given irrational number r, although r is also irrational,
p we have
r C .r/ D 0, which is a rational. On the other hand, if we set r D 2, we get
r r D r2 D 2, which is also a rational. Finally, if r ¤ 0, then the quotient of r by
itself is equal to 1, again a rational number.
Then, extend the above properties to all nonzero reals x and y, provided m and n
are odd naturals.
3. * Let a and b be rational numbers, and let r be an irrational one. If a C br D 0,
prove that a D b D 0.
4. Let a, b, c and d be rational numbers, and let r be an irrational one. If a C br D
c C dr, prove that a D c and b D d.
5. Let r be a positive real and letpk be an integer greater than 1. If r is irrational,
prove that the numbers 1r and k r are also irrational.
p p
6. (Canada) Let a; b and c be rational numbers, such that a C b 2 C c 3 D 0.
Prove that a; b and c are all equal to zero.
7. * Assuming the validity of the Fundamental Theorem of Arithmetic (according
p
to the last paragraph of the introduction to this chapter), prove that 2 is an
irrational number.
p
8. Let p be a prime number, and k > 1 be natural. Prove that k p is irrational.
A quite useful way of thinking geometrically on the set of rational numbers is the
following: we take a line r, and mark on it a point O; then, we choose one of the half-
lines that O determines on r, call it positive (the other being called negative) and set
a line segment ` as unit of measure. Then, we associate to each rational number a
point of r, in the following way: first, we associate 0 to the point O; then (according
to Figure 1.1), given a rational number ab , with a; b 2 N, we mark, starting from O
and on the positive half-line, a line segment OA of length a` (i.e., OA is obtained by
juxtaposition of a segments equal to `). If b D 1, we associate a1 D a to the point A.
If b > 1, we divide OA into b equal line segments, by plotting b 1 points on OA;
from these b 1 points, we let B denote the one closest to O, and associate ab to the
point B. It is not difficult to show (cf. Problem 1) that this geometric construction is
consistent, in the sense that, by changing ab for another equivalent fraction, we get
the same point B on r. Moreover, an analogous construction can obviously be made
for the negative rationals, marked on the negative half-line.
By proceeding as described in the previous paragraph, it happens that there are
lots of points on r which are not associated to any rational number. In order to give a
simple example, let A be the point associated to the number 1, and construct a square
a
b
a
...
O B A r
C B
O A E r
OABC, as shown in Figure 1.2. By using a compass, mark on the positive half-line a
point E such that OE D p OB. Since OA D 1,pit follows from Pythagoras Theorem
9 10
9
Pythagoras of Samos was one of the greatest mathematicians of classical antiquity. The theorem
that bears his name was already known to babylonians, at least two thousand years before he was
born; nevertheless, Pythagoras
p was the first one to prove it. It is also attributed to him the first proof
of the irrationality of 2.
10
We recall that Pythagoras’ Theorem, one of the most celebrated (and important) results of Plane
Euclidean Geometry, says that, in every right triangle, the square of the length of the hypotenuse
equals the sum of the squares of the lengths of the legs. For two different proofs of it, see chapters 4
and 5 of [4].
11
For an axiomatic construction of plane Euclidean Geometry, we refer the reader to [16].
12
We call the reader’s attention to the less common notations Œa; bŒ instead of Œa; b/, a; b instead
of .a; b, a; bŒ instead of .a; b/, Œa; C1Œ instead of Œa; C1/ and 1; aŒ instead of .1; a/.
1.4 The Geometric Representation 17
0 a b R
The rest of this volume, up to Chap. 5, approaches and develops several tools
necessary for an adequate presentation of the material in volumes 2 and 3. We
start by studying, in this chapter, some important algebraic identities, equations and
systems of equations.
Through the rest of these notes, we refer to a varying real number as a real
variable.1 In general, real variables will be denoted by lower case Latin letters,
for example a, b, c, x, y, z etc (an important exception to this usage is mentioned in
the next paragraph).
An algebraic expression, or simply an expression, is a real number formed
from a finite number of real variables, possibly with the aid of one or more
algebraic operations, i.e., additions, subtractions, multiplications, divisions, power
computations and root extractions (whenever the results of these operations make
sense in R). In particular, every real variable can be seen as an algebraic expression.
For another example,
p p
xC y x2 z
C 3 5 x2 yz3 x4
yz
is an algebraic expression which makes sense for all reals x; y; z, such that y > 0
and z ¤ 0 (recall that Problem 1, page 14, assures that we can extract roots of odd
index of any real number). We shall denote algebraic expressions by upper case latin
letters, as E, F etc.
1
In [5], we shall have the opportunity to consider complex variables.
for all values of the real variables x and y. Therefore, by setting F D x2 C 2xy C y2 ,
we obtain the algebraic identity E D F, i.e.,
to which we refer, from now on, as the formula for the square of a sum of two real
numbers.
The following proposition collects some important algebraic identities, which the
reader must keep for future use.
Proposition 2.1 For all x; y; z 2 R, we have:
(a) x2 y2 D .x y/.x C y/.
(b) .x ˙ y/2 D x2 ˙ 2xy C y2 .
(c) x3 ˙ y3 D .x ˙ y/.x2 xy C y2 /.
2.1 Algebraic Identities 21
In order to get the result of item (e), we apply that of item (b), with x C y in the
place of x and z in the place of y:
t
u
The reader has certainly noticed that, in the previous proposition, one either has:
(i) an identity of the form E D F, where E is a product of (at least two) polynomials
and F is the sum of monomials we get from expanding the products in E (this is the
case of the identities of items (b), (d) and (e)); or else (ii) an identity of the form
E D F, where E is a polynomial and F is a product of (at least two) polynomials (as
in items (a) and (c) of the previous proposition). In case (ii), we shall sometimes say
that F is a factorisation of E, or that it is obtained by factoring out expression E.
The coming examples will give us an idea on how to apply the identities collected
in the previous proposition to solve several interesting problems.
Example 2.2 Let x; y; z be real numbers, not all zero, such that xCyCz D 0. Explain
why xy C xz C yz ¤ 0 and, then, compute all possible values of the expression
x2 C y2 C z2
:
xy C yz C zx
22 2 Algebraic Identities, Equations and Systems
x2 C y2 C z2
D 2:
xy C yz C zx
t
u
Our next example shows how to use the algebraic identities we know so far to
prove inequalities.2
Example 2.3 (Poland) For given positive real numbers a and b, prove that 4.a3 C
b3 / .a C b/3 .
Proof By expanding the right hand side with the aid of item (d) of Proposition 2.1,
it is immediate to see that the inequality we want to prove is equivalent to a3 C b3
a2 b C ab2 . It now suffices to see that
a3 C b3 a2 b ab2 D a3 a2 b C b3 ab2 D a2 .a b/ b2 .a b/
D .a2 b2 /.a b/ D .a C b/.a b/.a b/
D .a C b/.a b/2 0;
p r2 a b
ab D ;
2
which would be a rational number too, for, in the right hand side of the above
equality, both the numerator and the denominator are rational numbers. t
u
Example 2.5 (Canada) For each natural number n, prove that
2
We will undertake a thorough discussion of inequalities in Chap. 5 and Sects. 9.7 and 10.8.
2.1 Algebraic Identities 23
p D m2 1 > m2 2m C 1 D .m 1/2 :
Observe that, at the right hand side of the above expression, both the sum S D y C z
and the product P D yz of y and z do appear. An expression of the form x2 Sx C P,
where S and P represent the sum and the product of two numbers or expressions, is
called a second degree trinomial in x. Hence, writing (2.2) backwards, we can also
see it as giving a factorisation for the second degree trinomial x2 Sx C P, where
S D y C z and P D yz:
The above factorisation is sometimes called Viète’s formula, in honor of the French
mathematician François Viète.3
The following example shows us how to apply Viète’s formula.
Example 2.6 (Soviet Union) Let a; b and c be pairwise distinct real numbers. Show
that the number
a2 .c b/ C b2 .a c/ C c2 .b a/
3
François Viète, French mathematician of the XVI century. By his pioneerism in the usage of letters
to represent variables, Viète is sometimes called the father of modern Algebra.
24 2 Algebraic Identities, Equations and Systems
S D a2 .c b/ C b2 a b2 c C c2 b c2 a
D a2 .c b/ C .b2 a c2 a/ C .c2 b b2 c/
D a2 .c b/ C a.b C c/.b c/ C bc.c b/
D .c b/Œa2 a.b C c/ C bc
D .c b/.a b/.a c/;
where we used (2.3) in the last equality. Now, it follows from a ¤ b, b ¤ c and
c ¤ a that a b; c b; a c ¤ 0, so that S ¤ 0. t
u
A useful variant of Viète’s formula is the factorisation for the expression x2 C
Sx C P, where, as before, S D y C z and P D yz:
Proof Applying item (d) of Proposition 2.1 twice, first with x C y in place of x and
z in place of y, we successively get
where, in the last equality, we have used the variant (2.4) of Viète’s formula. t
u
.x3 C y3 C z3 /2 .x3 y3 z3 /2
:
yCz
7. Let x and y be natural numbers such that x2 C 361 D y2 . Find all possible
values of x.
8. Real numbers a and b are such that a C b D m and ab D n. Compute the value
of a4 C b4 in terms of m and n. 2
9. If a2 C b2 D 1, find all possible values of 13.ab/
a6 Cb6
.
10. (EKMC) Let p a, b, c and d be real numbers such that a2 Cb2 D 1 and c2 Cd 2 D1.
If acCbd D 23 , compute the value of adbc, provided it is a positive number.
11. (Brazil) Find all natural numbers x and y such that x C y C xy D 120.
12. * Given positive distinct real numbers x and y, prove that the following
rationalisations4 are valid:
p p
p 1p
x y
(a) D xy .
x˙ y
p
3 2 p
p
1 x 3 xyC 3 y2
(b) p p
3 x˙ 3 y D x˙y
.
p3 x˙ p
3
1 p D x˙y y .
(c) p
3 2 p
x 3 xyC 3 y2
4
In an informal way, one can think of a rationalisation as a way of clearing roots from
denominators.
26 2 Algebraic Identities, Equations and Systems
14. Rationalise p1 p .
2C 2C 3
15. Rationalise p 1p . More precisely, obtain integers a, b, c, d, e, f and g such
2C 3 3
that
1 1 p p p
3
p p
3
p p
3
D Œ.a 2 C b/ C .c 2 C d/ 3 C .e 2 C f / 9:
2C 3 g
16. Let x, y and z be nonzero real numbers, such that x C y C z D 0. Explain why
the sum of any two of them is also nonzero, and compute all possible values
of each of the following expressions:
x2 y2 z2
(a) .yCz/2
C .xCz/2
C .xCy/2
.
x3 y3 z3
(b) .yCz/3
C .xCz/3
C .xCy/3
.
17. Let a and b be distinct integers. Find, in terms of a and b, the quotient of the
division of a64 b64 by .a C b/.a2 C b2 /.a4 C b4 /.a8 C b8 /.a16 C b16 /.
18. * Given n > 1 integer and a; b 2 R, prove that the following factorizations are
valid:
(a) an bn D .a b/.an1 C an2 b C an3 b2 C C bn1 /.
(b) an C bn D .a C b/.an1 an2 b C an3 b2 C bn1 /, provided n is odd.
19. Write x4 C 4y4 as a product of two non constant polynomials in x and y, both
having integer coefficients.
20. (Canada) Let a; b; c 2 Z. Prove that 6 divides a C b C c if and only if 6 divides
a 3 C b 3 C c3 .
21. (Canada) If a, b and c are real numbers for which a C b C c D 0, show that
a3 C b3 C c3 D 3abc.
22. Prove the double radical formula, also known as Bhaskara’s formula5 : for all
positive real numbers a and b, such that a2 b, one has
s p s p
q
p aC a2 b a a2 b
a˙ bD ˙ :
2 2
5
In honor of the Indian mathematician of the XII century Bhaskara II, also known as
Bhaskaracharya (Bhaskara, the professor). The idea behind Bhaskara’s formula is that, if a and
b are
qnaturals for which a2 b is a perfect square, then his formula provides a simpler expression
p
for a ˙ b.
2.2 The Modulus of a Real Number 27
23. Show that there do not exist nonzero real numbers x, y and z such that x C y C
z ¤ 0 and
1 1 1 1
D C C :
xCyCz x y z
24. (Soviet Union) Let a, b and c be pairwise distinct rationals. Prove that
1 1 1
C C
.b c/2 .c a/2 .a b/2
As
p an example, p sincep5 < 0, we have j 5j D .5/ D 5; analogously,
j 3j D . 3/ D 3 etc. More generally, an immediate consequence of the
definition is that jxj 0 for all x 2 R, with equality if and only if x D 0. Moreover,
one always has
x jxj D j xj;
28 2 Algebraic Identities, Equations and Systems
jx aj D b;
where a and b are given real numbers. Since jx aj 0, such an equation does
not admit roots when b < 0. On the other hand, when b 0, it follows from the
definition of modulus that one must have either x a D b or x a D b, from
where we get the roots
x D a C b; a b:
The coming example shows how to solve a more elaborate equation in a single
variable, involving the concept of modulus of a real number.
Example 2.9 Solve equation jx C 1j C jx 2j C jx 5j D 7.
Solution First of all, note that
x C 1; if x 1
jx C 1j D ;
x 1; if x < 1
x 2; if x 2
jx 2j D
x C 2; if x < 2
and
x 5; if x 5
jx 5j D :
x C 5; if x < 5
x 0 R
for
[
n [
n
fmaxfai ; bi g; minfai ; bi gg D fai ; bi g D f1; 2; 3; : : : ; 2ng:
iD1 iD1
30 2 Algebraic Identities, Equations and Systems
and
Therefore,
and
t
u
We continue our study of the concept of modulus with the following important
result, which is known in mathematical literature as the triangle inequality.6
Proposition 2.11 For all real numbers a and b, we have
Moreover, if a; b ¤ 0, then equality holds if and only if a and b have the same sign.
6
At the end of Sect. 5.2, we will give an explanation of why (2.9), as well as the coming
inequality (4.6), are called triangle inequality.
2.2 The Modulus of a Real Number 31
which is clearly true. From the computations above it also follows that ja C bj D
jaj C jbj if and only if ab D jabj; in turn, this happens if and only if ab 0. Finally,
if a; b ¤ 0, then we have the equality if and only if ab > 0. t
u
Corollary 2.12 For all real numbers a and b, we have
Moreover, if a; b ¤ 0, then the equality holds if and only if a and b have the same
sign.
Proof Applying the triangle inequality to a b in place of a, we get
and, hence, jaj jbj ja bj. Repeating the above argument with the roles of a
and b interchanged, it follows that jbj jaj ja bj.
Now, since ja bj jaj jbj; jbj jaj, we get
which is the analogous of (2.9) for three real numbers, instead of two. Therefore,
we shall also refer to this last inequality as the triangle inequality.
If a; b; c ¤ 0 and we have equality in (2.11), then we should also have equality
in all inequalities in (2.10). In particular, we have ja C bj jaj C jbj, and it follows
from Proposition 2.11 that a and b have equal signs. Since we can also reach (2.11)
by writing
Hence, we have just shown that equality holds in (2.11) if and only if a, b and c
have equal signs.
As we shall see in Sect. 4.1 (cf. Problem 7, page 96), inequalities (2.9) and (2.10)
can be easily generalized for n real numbers. For the time being, we end this section
with the following
Example 2.13 Prove that, for every x 2 R, we have
jx 1j C jx 2j C jx 3j C C jx 10j 25:
Hence, grouping the summands at the left hand side in pairs, we get,
jx 1j C jx 10j j10 1j D 9I
jx 2j C jx 9j j9 2j D 7I
jx 3j C jx 8j j8 3j D 5I
jx 4j C jx 7j j7 4j D 3I
jx 5j C jx 6j j6 5j D 1:
jx aj C jx bj D c:
jx 1j C jx 2j C C jx 2nj n2
ax2 C bx C c D 0; (2.13)
where a, b and c are given real numbers, with a ¤ 0. For reasons that will soon
be clear, the left hand side of (2.13) is also known as the second degree trinomial
associated to Eq. (2.13).
In order to solve (2.13), we let (one reads delta) denote the real number
D b2 4ac;
and call it the discriminant of the equation (or of the associated trinomial). As we
shall see in a moment, the sign of discriminates whether or not the equation has
real roots. To this end, we need the following auxiliary result.
Lemma 2.14 Given a; b; c 2 R, with a ¤ 0, one has
" #
2 b 2
ax C bx C c D a x C 2 : (2.14)
2a 4a
This algebraic identity is called the canonical form of the second degree trinomial
ax2 C bx C c.
7
Here, we are using the concept of a sequence of real numbers. For further details in this respect,
we refer the reader to Chap. 3.
2.3 A First Look at Polynomial Equations 35
t
u
Remark 2.15 The idea of adding and subtracting a certain summand out of a given
algebraic expression in order to complete a square, as was done right after the
second equality in the proof of the previous lemma, is very important and should
be learned as a kind of algebraic trick that will be useful in a number of places,
here as well as in [4] and [5]. Later in this chapter, we shall see other interesting
applications of such a technique.
Proposition 2.16 Let a, b and c be given real numbers, with a ¤ 0.
(a) The equation ax2 C bx C c D 0 has p
real rootspif and only if 0. Moreover,
if this is so, then its roots are b2a and bC2a .
(b) If 0, then the sum S and the product P of the roots of ax2 C bx C c D 0 are
given by S D ba and P D ac .
Proof
(a) It follows from (2.14) that
2 b 2
ax C bx C c D 0 , x C D 2: (2.15)
2a 4a
b 2
Since x C 2a
0 for all x 2 R, if the equation has real roots, then one must
p
have 0. In this case, it transpires from (2.15) that x C 2a
b
D˙ 2a , and item
(a) follows.
(b) It suffices to compute
p p
b b C b
C D
2a 2a a
36 2 Algebraic Identities, Equations and Systems
and
b p b C p .b/2 c
D 2
D :
2a 2a 4a a
t
u
Remarks 2.17
p
i. When 0, formulae b˙2a for the roots of ax2 C bx C c D 0 are known as
Bhaskara’s formulae.
ii. The formulas of item (b) are also known as Viète’s formulae.
iii. In the notations of item (a), if D 0 we say that ax2 C bx C c D 0 has two
equal roots.
The coming example shows how one can reduce a seemingly complicated
equation to a simpler one by means of a suitable substitution of variable.
156
Example 2.18 (Brazil) Find all real numbers x such that x2 C x C 1 D x2 Cx
.
Solution By performing the substitution y D x2 Cx, we get the equation yC1 D 156
y
or, which is the same, y2 C y 156 D 0. For this last equation, since D 12
4.156/ D 625 D 252 , it follows that y D 1˙252
D 13 or 12. Thus, we have
reduced the original equation to the second degree equations x2 C x D 13 and
x2 C x D 12. For the first one, we have D 51 < 0, so that there are no real
roots. For the second, D 49 and, hence, x D 1˙7
2
D 4 or 3. t
u
Our next example shows that it is sometimes more useful to algebraically
manipulate a second degree equation than to solve it explicitly.
p p
Example 2.19 Find the numerical value of .3C 2/5 C.3 2/5 without expanding
the powers involved.
p p
Proof Letting u D 3 C 2 and v D 3 2, we have u C v D 6 and uv D 7, so
that u and v are the roots of the equation x2 6x C 7 D 0. Therefore, making x D u
and x D v in this equation gives us u2 6u C 7 D 0 and v 2 6v C 7 D 0, or,
equivalently, u2 D 6u 7 and v 2 D 6v 7. Multiplying the first equality by uk and
the second one by v k , where k 0 is an integer, we get
u2 C v 2 D 6.u C v/ 7 2 D 6 6 14 D 22I
u3 C v 3 D 6.u2 C v 2 / 7.u C v/ D 6 22 7 6 D 90I
u4 C v 4 D 6.u3 C v 3 / 7.u2 C v 2 / D 6 90 7 22 D 386I
u5 C v 5 D 6.u4 C v 4 / 7.u3 C v 3 / D 6 386 7 90 D 1686:
t
u
For the next example, recall that if the sum and the product of two real numbers
are positive, then both numbers are also positive.
Example 2.20 Let p and q be given real numbers. If the equation x2 C px C q D 0
has real, positive and distinct real roots, show that the same is true for the equation
qx2 C .p 2q/x C .1 p/ D 0.
Proof Note initially that q ¤ 0, for otherwise the first equation would reduce to
x2 C px D 0, which has 0 as one of its roots, thus contradicting our hypotheses.
Now, let and 0 be, respectively, the discriminants of x2 C px C q D 0 and
qx2 C .p 2q/x C .1 p/. Let us first show that 0 > 0, which will guarantee that
the second equation has distinct real roots. Since x2 C px C q D 0 has distinct real
roots, we have D p2 4q > 0. Therefore,
0 D .p 2q/2 4q.1 p/
D p2 4q C 4q2
D C 4q2 > 0:
2q p p 1p 1 p
S0 D D2C > 0 and P0 D D C > 0;
q q q q q
as we wished to show. t
u
We finish our discussion of second degree equations with the following important
remark: if a ¤ 0 and ax2 CbxCc D 0 has real roots ˛ and ˇ (not necessarily ˛ ¤ ˇ),
then we have the factorisation
In fact, it follows from item (b) of Proposition 2.16 that, for every real x,
It is instructive to compare (2.16) to (2.2). The right hand side of (2.16) is called the
factorised form of the second degree trinomial ax2 C bx C c.
In turning to more general polynomial equations, it is natural to try to look at
those of degrees n D 3 and n D 4. In these cases, formulas have been built, in
terms of the coefficients of the equations, to compute their real roots, if any.8 As
professor I. Stewart teaches us in Chap. 4 of his very interesting book [25], such
formulas derive from the works of the Italian mathematicians Scipione del Ferro,
Girolamo Cardano, Niccolò Fontana (conhecido como Tartaglia) and Lodovico
Ferrari. However, they are too complicated to be useful, and for this reason we
shall not discuss them here. In order to help convincing the reader, let us just
mention that Cardano’s formula for the roots of the third degree polynomial equation
ax3 C bx2 C cx C d D 0 is the following9:
q p q p
3 3
q C q C .r p / C q q2 C .r p2 /3 C p;
2 2 3
where p D 3a b
, q D p3 C bc3ad
6a2
and r D 3a
c
(however, see Problems 16, 21 and 22).
For polynomial Eq. (2.12) of degree n 5, the Norwegian mathematician Niels
H. Abel10 and the French mathematician Évariste Galois,11 both of the XIX century,
proved that there exists no similar formula, built on the coefficients of the given
polynomial equation, that gives its real (or even complex!) roots. Well understood,
it doesn’t matter how smart someone is; they proved that it is impossible to find such
a formula, simply because it doesn’t exist! For a beautiful and elementary account
of the ideas involved, we refer the reader to [14].
Some particular kinds of polynomial equations of degrees 4 and 6, however,
are sufficiently simple to deserve some attention, specially because appropriate
8
As it happens, these formulas also give the complex roots of the corresponding equations.
However, we shall postpone any considerations involving complex numbers to [5].
9
Cf. http://www.math.vanderbilt.edu/~schectex/courses/cubic.
10
In his 27 years of life, Abel made several deep contributions to Mathematics, among which the
most famous one is perhaps the impossibility of solving general polynomial equations of degree 5.
11
In spite of his premature death, when he was only 21 years old, Galois is considered to be one
of the greatest mathematicians the world has ever seen. His work on the connection between the
solvability of polynomial equations of degrees n 5 and Group Theory constitute the foundations
of what is known today as Galois’ Theory, a branch of modern Algebra with applications to several
distinct areas of Mathematics.
2.3 A First Look at Polynomial Equations 39
in a way quite similar to that used to study biquadratic equations. We refer the reader
to Problem 16 for the corresponding details.
Let us now examine reciprocal polynomial equations of degree 4, i.e., polyno-
mial equations of degree 4 having the form
where a, b and c are given real numbers, with a ¤ 0. As Problem 28 shows, the
name reciprocal applies to a larger class of polynomial equations, and comes from
the fact that x 2 R n f0g is a real root of it if and only if 1x also is.
Initially, note that 0 is not a root of the equation above, for a ¤ 0. Therefore, a
real number x is a root of it if and only if it is a root of the equation
40 2 Algebraic Identities, Equations and Systems
b a
ax2 C bx C c C C 2 D 0; (2.19)
x x
which is obtained from the original equation by dividing both sides of the equality
by x2 . Rewrite the left hand side of the last equation above as
2 1 1
a x C 2 Cb xC C c D 0:
x x
Hence, x2 C x12 D y2 2, so that solving (2.19) amounts to solving the second degree
equation
a.y2 2/ C by C c D 0: (2.20)
However, the above discussion hides a subtlety: it is sure that every real root
x D ˛ of (2.19) generates the real root ˇ D ˛ C ˛1 of a.y2 2/ C by C c D 0.
Nevertheless, the converse statement is not true: not every real root y D ˇ of this
last equation does generate real roots x D ˛ of the initial reciprocal equation. In
fact, once we get a real root y D ˇ of a.y2 2/ C by C c D 0, in order to obtain the
possible real roots of the reciprocal equation corresponding to ˇ, we have to solve
in R the equation
1
xC D ˇ;
x
or, which is the same, x2 ˇx C 1 D 0. Since the discriminant of this last equation
is
D ˇ 2 4;
2.y2 2/ C 5y C 6 D 0:
Since this second degree equation has real roots y D 2 and y D 12 , it follows that
the real roots of the original equation are the real roots of the equations x C 1x D 2
and x C 1x D 12 . The first of these equations is equivalent to x2 C 2x C 1 D 0 and,
hence, has two real roots, both equal to 1. The second is equivalent to 2x2 CxC2 D
0, which has discriminant D 15 < 0; therefore, it has no real roots. t
u
1. Let b and c be given real numbers, such that the equation x2 C bjxj C c D 0 has
real roots. Prove that the sum of these roots is always equal to 0.
2. Solve, for x 2 R, the following equations:
p
(a) px C x C 2p D 10. p
(b) x C 10 2x C p 3 D 1 3x.
(c) x2 C 18x C 30 D 2 x2 C 18x C 45.
p
3. (IMO) In each of the cases (a) A D 2, (b) A D 1 and (c) A D 2, find all x 2 R
for which we have
q q
p p
x C 2x 1 C x 2x 1 D A:
4. A math teacher composed three different quizzes. In the first one, he put
a second degree equation. In the second, he put almost the same equation,
changing just the coefficient of the monomial of second degree. Finally, in the
third quiz, once more he put almost the same equation of the first, this time
changing just the constant coefficient. It is known that the roots of the equation
of the second quiz are 2 and 3, and that those of the third one are 2 and 7.
Decide whether the second degree equation of the first quiz has real roots and,
if this is so, compute them.
5. Let a and b be two distinct, nonzero real numbers. If a and b are the roots of the
equation x2 C ax C b D 0, find all possible values of a b.
42 2 Algebraic Identities, Equations and Systems
6. Let ˛ and ˇ be the roots of x2 13x C 9 D 0, and a and b be real numbers such
that the equation x2 C ax C b D 0 has roots ˛ 2 and ˇ 2 . Compute the value of
a C b.
7. Equation x2 Cx1 D 0 has roots u and v. Find a second degree equation whose
roots are u3 and v 3 .
8. The roots of the equation x2 Sx C P D 0 are the real numbers ˛ and ˇ. Find
a second degree trinomial whose coefficients are expressions built on S and P
and whose roots are the real numbers ˛S C P and ˇS C P.
9. Use the theory of second degree
p equations p developed in this section to compute
the value of the sum .7 C 4 3/5 C .7 4 3/5 .
10. If ˛ is a root of x2 x 1 D 0, find all possible values of ˛ 5 5˛.
11. For which integer values of m does the equation x2 C mx C 5 D 0 have integer
roots?
12. Show that, for every a; b; c 2 R, with a ¤ 0, the equation
1 1 1
C D 2
xb xc a
has exactly two distinct
q real roots.
q
13. Solve equation x D x 1x C 1 1x in the set of real numbers.
14. Show that, for every real number a ¤ 0, the equations
with a; b; c; d 2 Z.
(b) If x and y are real numbers such that x3 3x2 C5x D 1 and y3 3y2 C5y D 5,
compute all possible values of x C y.
16. * Let n 2 N and a; b; c 2 R, with a ¤ 0. Elaborate, for the equation ax2n C
bxn C c D 0, a discussion analogous to that made on the text for biquadratic
equations, and which led us to Proposition 2.21.
17. * Consider the polynomial equation of third degree x3 C ax2 C bx C c D 0,
where a, b and c are given real numbers, with c ¤ 0. If ˛ is a (nonzero) real
root of it, prove that there exist real numbers b0 and c0 such that we have the
factorisation
x3 C ax2 C bx C c D .x ˛/.x2 C b0 x C c0 /:
2.3 A First Look at Polynomial Equations 43
Then, conclude that the original polynomial equation has at most three (not
necessarily distinct) real roots. Moreover, if this is the case, and ˛, ˇ and are
its three real roots, show that
This result generalizes the factorised form (2.16) of a second degree trinomial
12
and is a particular case of the division algorithm.
p . p
3 p 3 p
18. (a) Show that the real number ˛ D 2 C 5 C 2 5 is a root of the
equation x3 C 3x 4 D 0.
(b) Conclude that ˛ is a rational number.
19. * Establish the following version of Girard’s relations13 between roots and
coefficients of a polynomial equation of third degree: if the real numbers a0 , a1 ,
a2 and a3 (a3 ¤ 0) are such that the equation
a 3 x3 C a 2 x2 C a 1 x C a 0 D 0
x1 x2 C x1 x3 C x2 x3 D a1
a3 : (2.21)
:̂ x x x D a0
1 2 3 a3
20. Assume that the equation x3 3x C 1 D 0 has three real roots, say ˛, ˇ and .
Compute the values of ˛ 2 C ˇ 2 C 2 , ˛ 3 C ˇ 3 C 3 and ˛ 4 C ˇ 4 C 4 .
21. Still concerning the third degree polynomial equation x3 C ax2 C bx C c D 0,
with a; b; c 2 R, prove that there exists a real number d such that the variable
substitution y D x d transforms the given equation into one of the form
y3 C py C q D 0, for certain real numbers p and q.
22. Concerning the equation x3 11x C 16 D 0, do the following items:
(a) Substitute x D uCv and get an equivalent equation in the two real variables
u and v.
(b) Impose that uv D 11 3
(i.e., make uv equals 13 times the coefficient of x in
the original equation) and conclude that the equation in u and v of item (a)
3
is equivalent to u6 C 16u3 C 11 3
D 0.
(c) Find u and v, and conclude that one of the roots of the given equation is
s p s p
3 1191 3 1191
8 C C 8 :
9 9
12
For more details concerning this point, as well as for the generalization of the result of
Problem 19, we refer the reader to [5].
13
After Albert Girard, French mathematician of the XVII century.
44 2 Algebraic Identities, Equations and Systems
The items above describe, by means of a specific example, the ideas behind
Cardano’s formula for the roots of a polynomial equation of third degree.
2
23. * Let x be a nonzero real number, such that x C 1x D 3. Compute all possible
values of x3 C x13 .
24. If x is a nonzero real number such that x C 1x D 4, calculate x4 C x14 .
2 2
25. If x2 x 1 D 0, compute all possible values of x 1x C x3 x13 .
26. (Singapore) If x2 4x C 1 D 0, compute all possible values of
1
6
x6 C x C 1x C 2
x6
3 :
x3 C x13 x C 1x
an xn C an1 xn1 C C a1 x C a0 D 0;
14
We shall see in [5] that the same holds for the complex roots of this equation.
2.4 Linear Systems and Elimination 45
is the problem of finding all sequences .a1 ; : : : ; an / of real numbers, such that the
substitutions x1 D a1 , . . . , xn D an solve all of the equations Ej D Fj . As above,
each such sequence .a1 ; : : : ; an / is said to be a solution of the system (2.22). To
solve a system of equations as (2.22) means to find all of its solutions.
In this section and the next one, we shall learn how to solve some simple (though
useful) systems of equations. We are also going to see that, under certain conditions,
an equation E D F (in several real variables) is equivalent to a system of equations
like (2.22), and this will be a source of a number of interesting examples.
The simplest—and, for our purposes, also the most useful—systems of equations
are the linear systems with two (resp. three) equations in two (resp. three) real
unknowns, i.e., systems of equations of one of the forms
8
< a 1 x C b 1 y C c1 z D d 1
a 1 x C b 1 y D c1
or a x C b 2 y C c2 z D d 2 ; (2.23)
a 2 x C b 2 y D c2 : 2
a 3 x C b 3 y C c3 z D d 3
respectively. Here, the real numbers ai ; bi ; ci ; di are given and not all zero, being
called the coefficients of the linear system.
15
Although the reader is probably acquainted with the concept of sequence, we refer to Sect. 6.1
for a rigorous definition.
46 2 Algebraic Identities, Equations and Systems
The most efficient method for solving linear systems like those above is the
elimination algorithm,16 also known as gaussian elimination,17 which is based
in the following result.
Lemma 2.24 Let E and F be given algebraic expressions in the real variables x
and y (or x, y and z). For reals a, b and c, the systems of equations
EDa E C cF D a C cb
and
FDb FDb
16
An algorithm is a finite sequence of precise procedures that, once followed, give an expected
result, also known as the output of the algorithm. Although algorithms will play a very modest
role in this volume, [5] brings a number of very interesting ones.
17
After Joanne Carl Friedrich Gauss, German mathematician of the XVIII and XIX centuries.
Gauss is generally considered to be the greatest mathematician of all times. In the several different
areas of Mathematics and Physics in which he worked, like Number Theory, Analysis, Differential
Geometry and Electromagnetism, there are always very important and deep results or methods that
bear his name. We refer the reader to [26] for an interesting biography of Gauss.
2.4 Linear Systems and Elimination 47
cx C dy D f by C ax D e dy C cx D f
; or ;
ax C by D e dy C cx D f by C ax D e
and Lemma 2.24 immediately assures that the solutions of this new system coincide
with those of the original one. Hence, in order to solve that system, it suffices to
solve this last one. On the other hand, since
c c bc
F E D .cx C dy/ .ax C by/ D d y; (2.24)
a a a
where d 0 D d bc 0
a and f D f a e. Now, we shall consider three different cases:
c
appropriate multiple of the first equation from the second, in order to eliminate the
variable x from it (and that’s where the name elimination comes from).
Example 2.25 Use Gaussian elimination to find all real values of a for which the
system equations
2x C ay D 3
ax C 2y D 32
2
If 2 a2 ¤ 0, which is the same as a ¤ ˙2, then the second equation above
gives us y D 3.1a/
4a2
, so that the first equation furnishes a single value for x, namely,
1 3.4 a/
xD .3 ay/ D :
2 2.4 a2 /
As was done for linear systems in two variables, we can suppose that a1 ¤ 0 (the
other cases being totally analogous). Changing equations E2 D d2 and E3 D d3
respectively by
a2 a2 a3 a3
E2 E1 D d2 d1 and E3 E1 D d3 d1
a1 a1 a1 a1
(in order to eliminate the variable x from the second and third equations), we get the
equivalent system
2.4 Linear Systems and Elimination 49
8
ˆ
< E1 D d1
E2 a2
E
a1 1
D d2 a2
d
a1 1 :
:̂ E a3
D d3 a3
3 a1 E1 a1 d1
Since
a2 a2
E2 E1 D .a2 x C b2 y C c2 z/ .a1 x C b1 y C c1 z/
a1 a1
a2 b1 a 2 c1
D b2 y C c2 z
a1 a1
D b02 y C c02 z;
and, analogously, E3 a3
E
a1 1
D b03 y C c03 z, it suffices to solve the system
8
< a 1 x C b 1 y C c1 z D d 1
b02 y C c02 z D d20 ;
:
b03 y C c03 z D d30
which is also equivalent to the original system. The discussion, then, goes on as
before, and we concentrate our attention in the third equation, c003 z D d300 . Also as
before, we have to distinguish three distinct cases:
• If c003 ¤ 0, then the third equation above gives a single value for z, say z D ;
since b02 ¤ 0, the substitution of this value into the second equation furnishes
y D b10 .d20 c02 z/ D b10 .d20 c02 /. Finally, since a1 ¤ 0, the substitution of the
2 2
values thus obtained for y and z into the first equation give us a single value for
x. Then, we conclude that the system is determined.
• If c003 D 0 and d300 ¤ 0, the system is impossible, for the third equation reduces to
0z D d300 .
• If c003 D d300 D 0, then the third equation reduces to the equality 0z D 0, and the
system as a whole reduces to
a 1 x C b 1 y C c1 z D d 1
b02 y C c02 z D d20
Since a1 and b02 are both nonzero, for each real value z D the system
corresponding to
a 1 x C b 1 y D d 1 c1
b02 y D d20 c02
2nd. Eliminate the variable y (in case b02 ¤ 0 in the last system above) from the
third equation, by adding to it an adequate multiple of the second equation,
thus obtaining a system of the form
8
< a 1 x C b 1 y C c1 z D d 1
b02 y C c02 z D d20 :
:
c003 z D d300
3rd. Analyse equation c003 z D d300 and, after this, the other two equations in
succession, in order to decide whether the original system is determined,
undetermined or impossible.
As before, it is much more useful to keep the general steps above in mind than
to try to memorize any of the expressions obtained through the calculations in the
previous discussions. Let us see one more example to illustrate this point.
Example 2.26 Find all real values of a for which the system of equations
8
< x C y az D 1
2x C ay C z D 1
:
ax C y z D 2
is impossible.
Solution Multiplying the first equation respectively by 2 and by a, and subtracting
(also respectively) the results from the second and third equations, we get the
equivalent system
8
< x C y az D 1
.a 2/y C .1 C 2a/z D 3 :
:
.1 a/y .1 a2 /z D 2 C a
(Observe that we have slightly changed the second step, so that the roles of the
second and third equations of the last system are interchanged—it is the second
equation that, now, has just one variable. Obviously, this is perfectly right, and shows
that the elimination algorithm is quite a flexible one.)
52 2 Algebraic Identities, Equations and Systems
p
Now, if a2 C a 1 D 0, i.e., if a D 1˙2 5 , then the system is impossible,
for a2 3a C 7 ¤ 0 for these values of ap and, hence, the second equation
reduces to an impossible equality. If a ¤ 1˙2 5 , then the second equation gives
us a single possible value for z; in turn, upon substitution of this value of z into the
third equation, we find a single possible value for y; finally, putting these values for
y and z into the first equation, we find a single possible value for x, so that the system
is determined. p
We conclude a D 1 and a D 1˙2 5 are the values of a for which the original
system is impossible. t
u
We finish our discussion of linear systems by observing that Gaussian elimination
algorithm can easily be put to work to the analysis of the general linear system in m
equations and n unknowns
8
ˆ
ˆ a11 x1 C a12 x2 C C a1n xn D b1
ˆ
< a21 x1 C a22 x2 C C a2n xn D b2
:: :: I (2.25)
ˆ
ˆ : :
:̂
am1 x1 C am2 x2 C C amn xn D bm
here, the aij and bi are given real numbers, such that at least one of the aij is nonzero.
Apart from an important particular case of general linear system, which will
make its appearance in Sect. 18:1 of [5] (and will be analysed there, by other
methods), we shall not make a systematic use of such systems along these notes;
hence, we shall not develop the general analysis of the application of the elimination
algorithm to them. If it is the case we have to solve such a linear system (as in the
example below), some clever reasoning, perhaps with the aid of Lemma 2.24, will
suffice.
Example 2.27 (OCM) Find all real solutions x1 , x2 , . . . , x100 of the linear system
8
ˆ
ˆ x 1 C x2 C x3 D0
ˆ
ˆ
ˆ
ˆ x 2 C x3 C x4 D0
ˆ
ˆ
< :: ::
: : :
ˆ
ˆ x98 C x99 C x100 D0
ˆ
ˆ
ˆ
ˆ
ˆ x99 C x100 C x1 D0
:̂
x100 C x1 C x2 D0
Solution Observe that each xj appears in exactly three of the given equations.
Therefore, adding all of them and dividing both sides by 3, we get
x1 C x2 C x3 C C x100 D 0:
2.4 Linear Systems and Elimination 53
For x2 D 0, write
is impossible.
3. Solve the system of equations
82 3 1
<xCy
ˆ z D8
3 5 2
x y C z D 1 :
:̂ 7 6 C 3
D5
x y z
4. For 1 i; j 3, let aij be given real numbers, such that a2i1 C a2i2 C a2i3 D 1 for
1 i 3 and ai1 aj1 C ai2 aj2 C ai3 aj3 D 0, for 1 i; j 3 with i ¤ j. Given real
numbers b1 , b2 , b3 , solve the linear system
54 2 Algebraic Identities, Equations and Systems
8
< a11 x1 C a12 x2 C a13 x3 D b1
a x C a22 x2 C a23 x3 D b2 ;
: 21 1
a31 x1 C a32 x2 C a33 x3 D b3
2.5 Miscellaneous
Let us now turn our attention to the next simplest type of system of equations,
namely, second degree systems. Ultimately, such a system is just a rephrasing of
a second degree equation in terms of a system of two equations in two unknowns.
Nevertheless, the reader will be amazed on how they provide a number of interesting
applications.
Proposition 2.28 Let S and P be given real numbers. The system of equations
xCyDS
(2.26)
xy D P
has real solutions if and only if S2 4P. Moreover, in this case, the solutions are
given by x D ˛, y D ˇ or vice-versa, where ˛ and ˇ are the roots of the second
degree equation u2 Su C P D 0.
Proof First of all, if ˛ and ˇ are the roots of u2 Su C P D 0, then we know from
Viète’s formulas (see item (b) of Proposition 2.16) that ˛ C ˇ D S and ˛ˇ D P.
Therefore, the pairs x D ˛, y D ˇ and x D ˇ, y D ˛ do satisfy the system of
Eq. (2.26).
Conversely, let x D x0 , y D y0 be any solution of that system. Then, the first
equation gives y0 D S x0 , and the second equation then gives P D x0 y0 D x0 .S
x0 /, or, which is the same, x20 Sx0 CP D 0. Therefore, x0 is a root of u2 SuCP D 0,
from where we get x0 D ˛ or x0 D ˇ. Since ˛ Cˇ D S, we have two possibilities:
• If x0 D ˛, then y0 D S x0 D S ˛ D ˇ.
• If x0 D ˇ, then y0 D S x0 D S ˇ D ˛.
t
u
2.5 Miscellaneous 55
The following examples show how we can sometimes reduce the search for
solutions of more complex systems of equations to that of simpler systems, of one
of the types we met so far.
Example 2.29 Find all real solutions of the system of equations
.x2 C 1/.y2 C 1/ D 10
:
.x C y/.xy 1/ D 3
Solution First of all, rewriting the left hand side of the first equation as
.x2 C 1/.y2 C 1/ D x2 y2 C x2 C y2 C 1
D .xy/2 C Œ.x C y/2 2xy C 1;
Now, writing
b2 C a2 2b D a2 C .b2 2b C 1/ 1
D a2 C .b 1/2 1
By squaring the second equation, we transform this last system into one of the
form (2.26), whose unknowns are a2 and c2 , and such that S D 10 and P D 9. Since
the roots of the second degree equation u2 10u C 9 D 0 are 1 and 9, it follows that
a2 D 1 or 9 and, hence, a D ˙1 or ˙3. Then, we have the possibilities
from where
Finally, each of these pairs .a; b/ give us another system of the form (2.26), with
unknowns x and y. Solving the four systems thus obtained, we arrive at the solutions
of the original system:
56 2 Algebraic Identities, Equations and Systems
.x; y/ D .1; 2/; .2; 1/; .1; 2/; .2; 1/; .0; 3/ or .3; 0/:
t
u
Some seemingly complicated equations can be easily solved, provided we find a
way to transform them into systems of equations. Since there is no general procedure
that tells us when or how this can be done, each equation should be analysed
separately. In this respect, the following example shows that a frequently useful
algebraic trick is the introduction of new variables.
Example 2.30 (Israel) Find all real solutions of the equation
p
4
p
4
13 C x C 4 x D 3:
p p
Solution Letting a D 4 13 C x and b D 4 4 x, we get a C b D 3 and a4 C b4 D
.13 C x/ C .4 x/ D 17. Hence, we have reduced the problem of solving the giving
equation to that of solving the system of equations
aCbD3
:
a4 C b4 D 17
17 D a4 C b4 D .a2 C b2 /2 2a2 b2
D Œ.a C b/2 2ab2 2.ab/2
D .9 2ab/2 2.ab/2
D 81 36ab C 2.ab/2 ;
so that
.ab/2 18.ab/ C 32 D 0:
The following lemma states a relatively easy sufficient condition to transform the
search for the roots of an equation in one unknown into that of a system of equations.
Lemma 2.31 If E1 ; E2 ; : : : ; En are expressions involving one or more real vari-
ables, then
8
ˆ
ˆ E1 D 0
ˆ
< E2 D 0
E12 C E22 C C En2 D 0 , : : (2.27)
ˆ ::
ˆ
:̂
En D 0
Proof This is an easy generalization of Corollary 1.5 and of Problem 2, page 10.
t
u
Example 2.32 Find all real roots of the equation
x4 y2 C y2 C 2 D 2y C 2x2 y:
1. (IMO) Find all real numbers x, y and z, such that the sum of any of them with
the product of the other two is always equal to 2.
2. Let a be a nonzero real constant. Find, in terms of a, all real numbers x and y
that solve the system of equations
(
1
Cx D aC1
xCy
:
x
xCy Da
58 2 Algebraic Identities, Equations and Systems
whose unknowns are x1 ; x2 ; x3 , where a; b and c are given real numbers, with
a ¤ 0. If D .b 1/2 4ac, do the following items:
(a) If < 0, then there is no real solution.
(b) If D 0, then there is exactly one real solution.
5. (TT) Find all real solutions of the system of equations
8 3
< x D 2y 1
y3 D 2z 1 :
: 3
z D 2x 1
3 3 3 p p p
xCyCzC C C D2 xC2C yC2C zC2 :
x1 y1 z1
p p
9. 3
Find all real roots of the equation p x C 5 C 3 11 x D 6.
p
10. Find all real roots of the equation 5 5 x D x.
x2
11. (Canada) Find all real roots of the equation x2 C .xC1/ 2 D 3.
Prove that a D b D c D d.
14. * Let E1 ; E2 ; : : : ; En and F1 ; F2 ; : : : ; Fn be given expressions in one or more real
variables, such that E1 F1 , E2 F2 , : : :, En Fn . Prove that the equation
E1 C C En D F1 C C Fn
3.1 Progressions
We say that a sequence .ak /k1 is defined by a positional formula if the values
ak 2 R are given by means of an expression that depends on k.
1
As we will see in Chap. 6, an infinite (resp. finite) sequence of real numbers is, rigorously
speaking, a function f W N ! R (resp. f W f1; 2; : : : ; ng ! R, for some n 2 N). Thus, the above
notation comes from the shortcut of letting ak D f .k/, for every k 2 N (resp., every 1 k n).
However, for the time being, the above definition will be harmless and sufficient for all of our
purposes.
Example 3.1 The sequence .ak /k1 of the perfect squares is the infinite sequence
.12 ; 22 ; 32 ; : : :/. Therefore, we have a1 D 12 , a2 D 22 , a3 D 32 and, more generally,
ak D k2 , for every integer k 1.
An alternative to defining a sequence by means of a positional formula is to give
a recursive definition of it. This means that we should specify a (finite) number
of initial terms of the sequence, as well as a recipe to calculate the other terms by
recurring to those that precede it. Let us see an example.
Example 3.2 Consider the sequence .ak /k1 , defined recursively by a1 D 2, a2 D 5
and
akC1 D ak C r; (3.2)
for every k 1.
In the previous definition, we refer to the real number r as the common
difference of the AP. We also observe that, for an AP .ak /k1 to be completely
determined, one has to give both its first term a1 and its common difference r. For
example, the sequence .ak /k1 such that a1 D 2 and akC1 D ak C 3 for k 1, is the
AP with initial term 2 and common difference 3, and it is completely determined by
the recurrence relation it satisfies and by the value of its first term. However, if we
only knew that akC1 D ak C 3 for k 1, we would not have just one AP, for we
would not know how to start it.
Example 3.4 Let .ak /k1 be an AP of common difference r. Prove that the
sequences .bk /k1 and .ck /k1 , defined by bk D a2k and ck D a2k1 for k 1,
are also AP’s, with common differences equal to 2r.
Proof We look at the sequence .bk /k1 , the case of the sequence .ck /k1 being
totally analogous. By the definition of an AP, it suffices to show that bkC1 bk D 2r,
for k 1. Since .ak /k1 is an AP with common difference r, it follows from the
definition of bk that
as we wished to prove. t
u
Another useful recursive characterization of AP’s is the one given in the
following result.
Proposition 3.5 A sequence .ak /k1 of real numbers is an AP if and only if
Proof
(a) Diagram
Cr Cr Cr Cr Cr
a1 / a2 / a3 / / ak1 / ak
makes it clear that, in order to reach ak from a1 , it is necessary k1 steps, where
each step consists of adding r to a term. Therefore, in order to obtain ak from
a1 , we have to add .k 1/r to a1 , so that ak D a1 C .k 1/r.
(b) From diagram
Cr Cr Cr r r r
a1 / a2 / a3 / o ak2 o an1 o an ;
we conclude that
D n.a1 C an /;
n.n C 1/
1C2C3CCn D :
2
3.1 Progressions 65
1 1 C 2ak 1
bkC1 D D D C 2 D bk C 2;
akC1 ak ak
so that .bk /k1 is the AP of initial term b1 D a11 D 1 and common difference
2. Therefore, this AP coincides with that of the odd positive integers, so that the
previous example gives bk D 2k 1, for every k 1. Hence,
1 1
ak D D :
bk 2k 1
t
u
Another useful class of sequences is the one of geometric progressions, according
to the next definition.
Definition 3.9 A sequence .ak /k1 of real numbers is a geometric progression (we
shall abbreviate GP) if there exists a real number q such that
akC1 D q ak ; (3.4)
for every k 1.
The real number q that appears in the definition of a GP is its common ratio.
Observe that, if q D 0, then ak D 0 for every k > 1. On the other hand, if q D 1,
then ak D a1 for every k 1. As it happens with AP’s, a GP .ak /k1 is completely
determined provided we know its first term a1 and its common ratio q.
Example 3.10 For a fixed real number q ¤ 0, the sequence .ak /k1 such that ak D
qk for every k 1 (i.e., the sequence formed by the powers of q, with natural
exponents) is a GP of common ratio q. If q < 0, then Problem 6, page 11, shows
66 3 Elementary Sequences
that ak is positive if and only if k is even; if 0 < q < 1, then Corollary 1.3 gives
a1 > a2 > a3 > > 0; if q > 1, then, again by that result, we have 0 < a1 < a2 <
a3 < .
Another useful recursive characterization of (almost all) GP’s is the one given by
the following result.
Proposition 3.11 A sequence .ak /k1 of nonzero real numbers is a GP if and only if
for every k 1.
Proof By definition, .ak /k1 is a GP (of common ratio q) if and only if
a2 a3 a4
D D D D q:
a1 a2 a3
In other words, .ak /k1 is a GP if and only if for every integer k 1, we have
akC2 akC1
akC1 D ak , and this is an equivalent way of writing (3.5). t
u
As we have done with AP’s, our next result brings formulas for the general term
and for the sum of the first k terms of a GP.
Proposition 3.12 If .ak /k1 is a GP of common ratio q, then:
(a) ak D a1 qk1 , for every k 1.
anC1 a1
(b) If q ¤ 1, then a1 C a2 C C an D q1 , for every n 1.
Proof
(a) The proof we present parallels that for the general term of an AP (see the proof
of item (a) of Proposition 3.6): diagram
q q q q q
a1 / a2 / a3 / / ak1 / ak
Therefore,
.q 1/Sn D qSn Sn
D .a2 C a3 C C an C anC1 / .a1 C a2 C C an /
D .a2 C a3 C C an / C anC1 a1 .a2 C C an /
D anC1 a1 ;
ckC1 ba b1 qakC1 1
D kC1 D D qakC1 ak D qr :
ck bak b1 qak 1 t
u
The purpose of the next example is to call the reader’s attention to the fact that
the idea used in the proof of item (b) of Proposition 3.12 is as important as the actual
result it collects. Hence, the reader should keep this idea in his/her mind.
Example 3.14 Compute the value of the sum
where, from left to right, the k-th summand equals the product of the k-th term of
the AP 2; 7; 12; : : : ; 502 with the k-th term of the GP 1; 3; 32 ; : : : ; 3100 .
Solution Arguing exactly as in the proof of item (b) of Proposition 3.12, let S
denote the desired sum, and compute 3S (we use the factor 3, for it is the common
ratio of the GP 1; 3; 32 ; : : : ; 3100 ):
Hence,
2S D 3S S
D 2 3 C 7 32 C 12 33 C 17 34 C C 497 3100 C 502 3101
2 1 7 3 12 32 17 33 497 399 502 3100
D .502 3101 2/ 5.3 C 32 C 33 C 34 C C 3100 /
5
D .502 3101 2/ .3101 3/;
2
where we grouped multiples of equal powers of 3 in the third equality, and used the
formula of item (b) of Proposition 3.12 in the fourth one. Finally, grouping together
the summands of the last expression above, we get S D 14 .999 3101 C 11/. t
u
1. Compute the first four terms of the sequence .an /n1 , defined by a1 D 1 and
akC1 D a2k C ak C 1, for k 1.
2. Write a positional formula for each of the following sequences:
(a) .1;
1 2; 3; 4; 5; 6; 7; 8; : : :/.
2 3 4 5 6 7 8
(b) ; ; ; ; ; ; ; ; : : : .
2 13 4 15 6 17 8 1 9
(c) 1; 2 ; 3; 4 ; 5; 6 ; 7; 8 ; : : : .
3. Let .an /n1 be a sequence of positive real numbers satisfying the recurrence
relation akC1 D 3aak C1
k
, for k 1. Find a recurrence relation satisfied by the
sequence .bn /n1 , defined for n 1 by bn D a1n .
4. Write a recurrence relation satisfied for each of the following sequences:
(a) .1; 1; 1; 3; 5; 9; 17; 31; 57; 105; 183; : : :/.
2 22
(b) .1; 2; 22 ; 22 ; 22 ; : : :/.
5. Below, we show the first four lines of an infinite array of natural numbers, such
that, for i > 1, the i-th line starts with the number i and has two more entries
than the .i 1/-th line. Compute the sum of the entries in the n-th line.
1
2 3 4
3 4 5 6 7
4 5 6 7 8 9 10
3.1 Progressions 69
ap C aq D au C av , p C q D u C v
.0; 1; 1; 2; 2; 3; 3; 4; 4; : : : ; r; r; r C 1; r C 1; : : :/:
1 3 5 99
C 2 C 3 C C 50 ;
2 2 2 2
in which the k-th summand from left to right equals the quotient between the
k-th term of the AP 1; 3; 5; : : : ; 99 and the k-th term of the GP 2; 22 ; 23 ; : : : ; 250 .
2
According to the introduction to Chap. 1, we recall that a natural n > 1 is composite if we can
write n D ab, for some naturals a; b > 1.
70 3 Elementary Sequences
18. The sequence .ak /k1 is an arithmetic-geometric progression if, for every
integer k 1, we have ak D bk ck , where the sequences .bk /k1 and .ck /k1
are respectively an AP and a GP. Let r be the common difference of the AP and
q be the common ratio of the GP. Compute, as a function of n, b1 , c1 , r and q,
the value of the sum of the first n terms of .ak /k1 .
19. (Macedonia) In a nonconstant AP of real numbers, the quotient between the
first term and the common difference is an irrational number. Prove that no
three distinct terms of this AP form a GP.
20. Compute, as a function of n, the n-th term of the sequence .ak /k1 , given by
a1 D 2 and akC1 D 2ak 1, for every integer k 1.
21. The sequence .an /n1 satisfies a1 D 1 and akC1 D 3ak 1, for every k 1. Do
the following items:
(a) If bn D an 12 , prove that bkC1 D 3bk for every k 1.
(b) Write down the first five terms of the sequence .bn /n1 and, then obtain a
general positional formula for it.
(c) Find a positional formula for an .
22. Prove that there does not exist a GP having the numbers 2, 3 and 5 as three of
its terms.
At the end of last section, we saw that a sequence .an /n1 is a GP if and only if it
satisfies a recurrence relation of the form (3.4), i.e., if and only if
for every integer k 1, where q is a real constant. For reasons that will be clear in
a few moments, we say that (3.6) is a first order linear recurrence relation with
constant coefficients. There, we also saw that a sequence .an /n1 is an AP if and
only if it satisfies a recurrence relation like (3.3).
In this section, our task is to study the more general class of sequences .an /n1
that satisfy recurrence relations like
where r and s are given real constants, not both zero. If s D 0, (3.7) essentially
reduces to (3.6); if s ¤ 0, it is said to be a second order linear recurrence relation
with constant coefficients.
Problem 5 explains what we mean by the property of linearity of sequences that
satisfy recurrence relations like (3.6) and, more generally, (3.7). For such a sequence
.an /n1 , Theorem 3.16 will explain how to compute an as a function of n. Before we
present it, it is instructive to show the idea behind its proof in a concrete example,
and we start by doing this.
3.2 Linear Recurrences of Orders 2 and 3 71
Example 3.15 Let .an /n1 be the sequence such that a1 D 1, a2 D 7 and akC2 D
8akC1 15ak , for every integer k 1. Find a positional formula for an .
Solution For every integer k 1, we have
Thus, the sequence .bk /k1 , defined by bk D akC1 3ak for every k 1, is a GP of
common ratio 5 and initial term b1 D a2 3a1 D 4. Hence, bk D b1 5k1 D 45k1 .
Analogously, for every integer k 1, we have
so that the sequence .ck /k1 , given by ck D akC1 5ak for every k 1, is a GP
of common ratio 3. Since its initial term is c1 D a2 5a1 D 2, it follows that
ck D c1 3k1 D 2 3k1 .
Therefore, for every integer k 1, we have the system of linear equations
akC1 3ak D 4 5k1
;
akC1 5ak D 2 3k1
ak D 2 5k1 3k1 ; 8 k 1: t
u
The reasoning presented in the solution to the previous example can be easily
generalized to deal with a general second order linear recurrence relation with
constant coefficients. We do this now, referring the reader to Sect. 11.4 for a different
approach.
Theorem 3.16 Let .an /n1 be a sequence of real numbers such that
for every integer k 1, where r and s are given real constants, not both zero. If the
second degree equation x2 Crx Cs D 0 has real roots3 ˛ and ˇ, then there exist real
constants A and B, completely determined by the values of a1 and a2 , such that:
(a) If ˛ ¤ ˇ, then an D A˛ n1 C Bˇ n1 , for every integer n 1.
(b) If ˛ D ˇ, then an D .A C B.n 1//˛ n1 , for every integer n 1.
Proof Recall from Proposition 2.16 that ˛ C ˇ D r and ˛ˇ D s. Thus, (3.7) can
be rewritten as
3
The case of complex roots is dealt with, in a much more general setting, in Chap. 21 of [5].
72 3 Elementary Sequences
or
Now, let us first consider the case ˛ ¤ ˇ. In (3.8), subtracting the first relation
from the second gives
letting
a2 ˇa1 a2 ˛a1
AD and B D
˛ˇ ˛ˇ
for every integer k 1 (see Problem 4); therefore, for fixed A; B 2 R, the sequence
zk D Auk C Bvk is also such that
(see Problem 5). Thus, the idea is to search for real numbers A and B such that, for
every integer k 1, we have ak D zk . Since both sequences .ak /k1 and .zk /k1
satisfy identical linear second order recurrence relations, Problem 13, page 96, will
assure that it suffices to find real numbers A e B for which a1 D z1 and a2 D z2 , i.e.,
such that
a1 D A
:
a2 D .A C B/˛
x2 C rx C s D 0;
In both of these cases, the systems we have to solve are obtained by taking k
respectively equal to 1 and 2 in the formulas of items (a) or (b) of the theorem.
Example 3.17 Let us execute the above procedure to get a positional formula for the
n-th term of the celebrated Fibonacci sequence,4 i.e., the sequence .Fn /n1 given
by F1 D 1, F2 D 1 and
4
After Leonardo di Pisa, also known as Fibonacci, Italian mathematician of the XII and XIII
centuries. As can be seen in Chap. 1 of [5], the Fibonacci sequence plays a relevant role in
Combinatorics.
74 3 Elementary Sequences
Fn D A˛ n1 C Bˇ n1 ;
Multiplying the first equation by ˛ and subtracting the second equation p from the
result, we obtain .˛ ˇ/B pD ˛ 1. Now, since ˛ C ˇ D 1 and ˛ ˇ D 5, we can
write this last equation as 5B D ˇ, so that B D pˇ5 . Analogously, multiplying
the first equation by ˇ and subtracting the result from the second equation, we obtain
A D p˛ , so that
5
˛ ˇ 1
Fn D p ˛ n1 p ˇ n1 D p .˛ n ˇ n /:
5 5 5 t
u
Our next example shows a situation where Theorem 3.16 can be applied, albeit
not straightforwardly.
Example 3.18 Let .an /n1 be a sequence of real numbers satisfying the recurrence
relation akC1 D rak C s for every integer k 1, where r and s are given real
constants, with r ¤ 0; 1 (we need not consider the case r D 1, for in this case
.an /n1 is an AP). For an integer k 1, we have
This is a second order linear recurrence relation with constant coefficients, whose
characteristic equation is
x2 .r C 1/x C r D 0:
3.2 Linear Recurrences of Orders 2 and 3 75
Since its roots are r and 1, it follows from item (a) of Theorem 3.16 that an D
Arn1 C B. If we now recall that a2 D ra1 C s, we see that A and B can be found by
solving the linear system
A C B D a1
:
Ar C B D ra1 C s
an D a1 rn1 C s:
Let us now turn our attention to third order linear recurrence relations with
constant coefficients. As the reader has probably guessed by now, in this case we
have a sequence .an /n1 satisfying a recurrence relation of the form
for every integer k 1, where r, s and t are given real constants, not all zero.
As in the case of second order linear recurrence relations with constant coef-
ficients, we define the characteristic equation of (3.11) as the third degree
polynomial equation
x3 C rx2 C sx C t D 0:
According to Problem 21, page 43, such an equation admits, at most, three real
roots. In what follows, we shall assume that it indeed has three such roots, let us say
˛, ˇ and . Then, we have the following result.
Theorem 3.19 Let .an /n1 be a sequence of real numbers such that
for every integer k 1, where r, s and t are given real constants, not all zero. If the
characteristic equation x3 C rx2 C sx C t D 0 of (3.11) has real roots ˛, ˇ and ,
then there exist real constants A, B and C, completely determined by the values of
a1 , a2 and a3 , such that:
(a) If ˛ ¤ ˇ ¤ ¤ ˛, then an D A˛ n1 C Bˇ n1 C C n1 , for every n 1.
(b) If ˛ D ˇ ¤ , then an D .A C B.n 1//˛ n1 C C n1 , for every n 1.
(c) If ˛ D ˇ D , then an D .A C B.n 1/ C C.n 1/2 /˛ n1 , for every n 1.
Proof We shall only sketch the proof, leaving the details to the reader (see
Problem 6).
Let .bn /n1 be the sequence given by bn D ˛ n1 , for every integer n 1. Since
˛ C r˛ 2 C s˛ C t D 0, we also have ˛ kC2 C r˛ kC1 C s˛ k C t˛ k1 D 0 or, which
3
is the same,
for every integer k 1. Hence, .bn /n1 satisfies the same recurrence relation as
.an /n1 does. Analogously, the sequences .cn /n1 and .dn /n1 , given for an integer
n 1 by cn D ˇ n1 and dn D n1 , also satisfy the same recurrence relation as
.an /n1 .
We now turn our attention to the cases (a)–(c):
(a) For all A; B; C 2 R, the sequence .un /n1 such that un D A˛ n1 C Bˇ n1 C
C n1 for every integer n 1 satisfies the same recurrence relation as .an /n1
does. On the other hand, since ˛ ¤ ˇ ¤ ¤ ˛, we can choose the constants
A, B and C in such a way that u1 D a1 , u2 D a2 and u3 D a3 . Therefore,
by invoking once more the result of Problem 13, page 96, we conclude that
un D an , for every integer n 1.
(b) The result of Problem 21, page 43, gives us
x3 C rx2 C sx C t D .x ˛/2 .x /:
With the aid of this identity, it is easy to show that bn D .n 1/˛ n1 also
satisfies the same recurrence relation as the sequence .an /n1 . Then, as in item
(a), the sequence .un /n1 such that un D .A C B.n 1//˛ n1 C C n1 for n 1
satisfies that same recurrence relation, for all A; B; C 2 R. Moreover, we can
again choose the real constants A, B and C such that u1 D a1 , u2 D a2 and
u3 D a3 . Thus, once more by the result of Problem 13, page 96, we conclude
that un D an for every integer n 1.
(c) This time, the result of Problem 21, page 43, gives
x3 C rx2 C sx C t D .x ˛/3 :
for every integer n 1, where A, B and C are real constants chosen so that a1 D 1,
a2 D 4 and a3 D 14. These initial conditions, in turn, give us the linear system of
equations
8
<A D 1
ACBCC D2 ;
:
A C 2B C 4C D 7=2
1. Let .an /n1 be the sequence given by a1 D 1, a2 D 4 and akC2 D 5akC1 6ak ,
for every integer k 1. Compute an as a function of n.
2. Let .an /n1 be the sequence given by a1 D 3, a2 D 5 and akC2 D 3akC1 2ak ,
for every integer k 1. Prove that an D 2n C 1, for every n 2 N.
3. Let .an /n1 be the sequence given by a1 D 3 and akC1 D 2ak 1, for every
integer k 1. Compute an as a function of n.
4. * If the second degree equation x2 C rx C s D 0 has two real roots equal to ˛,
prove that the sequences uk D ˛ k1 and vk D .k 1/˛ k1 satisfy the recurrence
relations
ukC2 C rukC1 C suk D 0
: (3.12)
vkC2 C rvkC1 C svk D 0
5. * The sequences .un /n1 and .vn /n1 satisfy the recurrence relations (3.12), for
every integer k 1. For any fixed real constants A and B, prove that the sequence
.zn /n1 , given by zk D Auk C Bvk , satisfies an analogous recurrence relation.
6. * Fulfill the details in the proof of Theorem 3.19.
7. * The Lucas sequence5 is the sequence .Ln /n1 given by L1 D 1, L2 D 3 and
LkC2 D LkC1 C Lk , for every integer k 1. Show that, for every integer n 1,
we have:
p p
1C 5 1 5
(a) Ln D ˛ n C ˇ n , where ˛ D 2 and ˇ D 2 .
(b) L2n D L2n C 2.1/n1 .
5
After François Édouard Anatole Lucas, French mathematician of the XIX century.
78 3 Elementary Sequences
(c) L2n 5Fn2 D 4.1/n , where .Fn /n1 is the Fibonacci sequence (cf. Exam-
ple 3.17).
8. (IMO shortlist—adapted) With respect to the sequence .an /n1 , such that a1 D 1
and
1 p
anC1 D .1 C 4an C 1 C 24an /
16
for every integer n 1, do the following items:
p
(a) If bn D 1 C 24an, show that 2bnC1 D bn C 3 for every n 1.
(b) Show that bn D 3 C 24n for every n 1.
(c) Conclude that, for every n 1, we have
1 1 1
an D 1 C n1 1C n :
3 2 2
X
n
a1 ; if n D 1
aj D :
a1 C a2 C C an ; if n > 1
jD1
X
n X
n
aj D c D nc:
jD1 jD1
The main reason for the success of the † notation owes to the fact that it turns it
quite easy to manipulate sums with large numbers of summands, specially when
each such summand is itself a sum. For example, given sequences .ak /k1 and
.bk /k1 of real numbers, the associativity and commutativity of addition of reals
give
with the aid of the † notation, the equality above can be written in the much more
compact form
X
n X
n X
n
aj ˙ bj D .aj ˙ bj /: (3.13)
jD1 jD1 jD1
On the other hand, given c 2 R, the distributivity of the multiplication of reals with
respect to addition gives
an equality that can be written, again with the aid of the † notation, as
X
n X
n
c aj D caj : (3.14)
jD1 jD1
The following example shows, in a simple case, how one could apply the two
identities above in order to simplify the task of calculating the value of a sum.
P
Example 3.22 Compute the value of nkD1 .2k C 1/ in terms of n 2 N.
Solution Successively applying (3.13), (3.14) and the first part of Example 3.7, we
get
X
n X
n X
n X
n
.2k C 1/ D 2k C 1D2 kCn
kD1 kD1 kD1 kD1
n.n C 1/
D2 C n D n2 C 2n:
2
t
u
P
The notation is particularly useful to perform cancellations in sums. More
precisely, given a sequence .ak /k1 , and performing the possible cancellations in
the sum
X
n1
.ajC1 aj / D an a1 : (3.15)
jD1
80 3 Elementary Sequences
X
n
.ajC1 aj / D anC1 a1 : (3.16)
jD1
Any one of (3.15) or (3.16) is known as a telescoping sum. The idea behind
the name is the following: as a telescope shortens the immense distance between
a celestial body and our eyes, the above formulas shorten the way between certain
given sums and their results. P
Telescoping sums are the greatest advantage of having the notation at our
disposal. Let us justify this claim by examining two interesting examples.
Example 3.23 Deduce the formula for the general term of an AP with the aid of
telescoping sums.
Solution Let the sequence .ak /k1 be an AP of common difference r. Then, ajC1
aj D r for each integer j 1, so that (3.16) gives
X
n1 X
n1
an a1 D .ajC1 aj / D r D .n 1/r
jD1 jD1
and stipulate the initial term of the second order AP, say a1 D 2; then, we
successively compute the values of a2 ; a3 ; : : : from the relations a2 a1 D 3,
a3 a2 D 7, a4 a3 D 11 etc. Proceeding this way we obtain, from the AP
given above, the second order AP
The discussion of the previous paragraph makes it clear that a second order AP
is completely determined only if we know its first three terms. In fact, it is only by
knowing its first three terms that we will know the first two terms of the nonconstant
AP formed by the differences of the consecutive terms of the second order AP.
We are now in position to use telescoping sums to deduce a formula for the
general term of a general second order AP.
3.3 The † and … Notations 81
.n 1/.n 2/r
an D a1 C .a2 a1 /.n 1/ C ; (3.17)
2
where r D a3 2a2 C a1 is the common difference of the nonconstant AP formed
by the differences of the consecutive terms of .ak /k1 .
Solution Let bk D akC1 ak for every integer k 1. The formulas for telescoping
sums and for the sum of the terms of a finite AP give us
X
n1 X
n1
.n 1/.b1 C bn1 /
an a1 D .ajC1 aj / D bj D :
jD1 jD1
2
On the other hand, by applying the formula for the general term of an AP, we get
and, thus,
.n 1/.b1 C bn1 /
an D a1 C
2
.n 1/.2.a2 a1 / C .n 2/r/
D a1 C
2
.n 1/.n 2/r
D a1 C .n 1/.a2 a1 / C :
2
Finally, it suffices to observe that
Solution Since the terms of the sequence .ak /k1 are all positive (why?), we can
define the sequence .bk /k1 by setting bk D a1k . The given recurrence relation, then,
gives,
1 1 C kak 1
bkC1 D D D C k D bk C k:
akC1 ak ak
Thus, by applying the formula for telescoping sums, together with the formula for
the sum of the terms of a finite AP, we get
X
n1 X
n1
n.n 1/
bn b1 D .bkC1 bk / D kD :
kD1 kD1
2
Therefore,
n.n 1/ n.n 1/ n2 n C 2
bn D b1 C D1C D ;
2 2 2
1 2
and, hence, an D bn
D n2 nC2
. t
u
For the next example define, for each positive integer n, the factorial of n,
denoted nŠ, as the product of all natural numbers from 1 to n (by convention, 1Š D 1),
i.e., 2Š D 2, 3Š D 6, 4Š D 24, 5Š D 120 etc. Observe that, in general, we have
.k C 1/Š D .k C 1/ kŠ.
P
Example 3.26 (Canada) Compute the value of the sum njD1 j.jŠ/ in terms of n.
Solution In order to use the formula for telescoping sums, we have first to succeed
in writing j.jŠ/ as a difference ajC1 aj , of consecutive terms of a single sequence
.ak /k1 . We do this by observing that
X
n X
n X
n
j.jŠ/ D Œ.j C 1/Š jŠ D .ajC1 aj /
jD1 jD1 jD1
Y
n
a1 ; if n D 1
aj D :
a1 a2 : : : an ; if n > 1
jD1
Q
With the aid of the notation, we can write the factorial of n 2 N (see the
paragraph that precedes Example 3.26) by writing
Y
n
nŠ D j: (3.18)
jD1
P Q
As was the case with the notation, the usefulness of the notation comes
from the fact that it formally commutes with multiplications and divisions. In fact,
given a real number c and sequences .ak /k1 and .bk /k1 , we have
.n C 1/Š .n C 1/ nŠ
D 2n D 2n
nŠ nŠ
D 2n .n C 1/:
t
u
84 3 Elementary Sequences
P Q
Analogously to the case of the notation, the notation is particularly useful
when performing cancellations in certain products. This is the content of the formula
for telescoping products, collected in the following result.
Proposition 3.29 If .ak /k1 is a sequence of nonzero real numbers, then
Y
n
ajC1 anC1
D : (3.19)
jD1
aj a1
Proof As with telescoping sums, it suffices to observe that the intermediate factors
of the product at the left hand side all cancel. In symbols,
Y
n
ajC1 a2 a3 a4 an anC1 anC1
D D :
aj a1 a2 a3 an1 an a1
jD1 t
u
Example 3.30 We take another look at the previous example, this time with the
formula for telescoping products at our disposal. To this end, we first of all note that
Yn Yn
2 kC1
2C D 2n :
kD1
k kD1
k
Now, defining the sequence .ak /k1 by ak D k, it follows from (3.19) that
Yn Yn
2 akC1 anC1
2C D2 n
D 2n D 2n .n C 1/:
kD1
k kD1
a k a 1
1. Prove that a sequence .ak /k1 is a second order AP if and only if akC2 2akC1 C
ak ¤ 0 and akC3 3akC2 C 3akC1 ak D 0, for every integer k 1.
2. Let .ak /k1 be the sequence defined by a1 D 1 and anC1 D an C 3n 1, for
every positive integer n. Compute the n-th term of this sequence in terms of n.
3. The sequence .an /n1 is given by a1 D 1 and anC1 D an C 8n, for n 1.
Compute an in terms of n. P 1
4. * Compute the value of the sum nkD2 .k1/k as a function of n 2 N.
5. The sequence .ak /k1 is an AP. Prove that, for every positive integer n, we have
X
n1
1 n1
D :
kD1
ak akC1 a1 an
3.3 The † and … Notations 85
1 1 1 1 1
2
C 2 C 2 CC 2 < 2 :
1 2 3 n n
P 1
7. Compute the value of the sum n1kD1 .4k1/.4kC3/ as a function of n 2 N.
8. (Romania) Let k and n be positive integers. Prove that:
(a) .2kC1/3 .2k1/3 can always be written as a sum of three perfect squares.
(b) .2n C 1/3 2 can always be written as a sum of 3n 1 perfect squares.
P
9. Compute the value of the sum nkD1 .kC1/Š
k
as a function of the integer n > 1.
10. (Brazil—adapted) Do the following items:
(a) For k 2 N, write .k C 1/2 C k2 C k2 .kqC 1/2 as a perfect square.
P
(b) Compute the value of the sum 99 kD1
1
k2
1
C .kC1/ 2 C 1.
11. * (Brazil) Let .Fk /k1 be the Fibonacci sequence, i.e., the sequence given by
F1 D 1, F2 D 1 and FkC2 D FkC1 C Fk , for every integer k 1. Compute the
P F
value of the sum nkD1 Fk FkC1
kC2
in terms of n 2 N.
12. The sequence .ak /k1 is an AP. Prove that, for every positive integer n, we have
X
n1
1 n1
p p Dp p :
kD1
ak C akC1 a1 C an
999;999
X 1
p p
3 2
p :
kD1
3
.k C 1/ C k 1 C 3 .k 1/2
2
X
n
1
p p :
kD1
.k C 1/ k C k k C 1
16. (Germany) Given a natural number n > 1, compute, as a function of n, the value
Qn 1
of the product jD2 1 j2 .
17. For 0 k 101, let xk D k
101
. Compute the value of the sum
101
X x3k
:
kD0
1 3xk C 3x2k
86 3 Elementary Sequences
Y
n1 3
k C1
:
kD2
k3 1
For the next problem, we recall that the arithmetic mean A of a finite
collection a1 , a2 , . . . , ak of real numbers is the real number
a1 C a2 C C ak
AD :
k
19. (Macedonia) Find all values of n 2 N for which we can write the set A D
f1; 2; 3; : : : ; 4ng as the union of n pairwise disjoint 4-element subsets, such that,
in each of them, one element is the arithmetic mean of the other three.
20. (China—adapted) Use the result of Problem 13, page 26, to compute the value
of the greatest integer that is less than or equal to
1 1 1 1
S D p C p C p CC p :
1 2 3 10;000
With the algebraic background of the previous chapters at our disposal, we devote
the first section of this chapter to the development of a very important topic in
elementary Mathematics, namely, the principle of mathematical induction. It will
considerably improve our ability of elaborating proofs of mathematical statements
that depend on natural numbers. In the other two sections, we shall apply this
principle to deduce Newton’s formula for binomial expansion, as well as some
important related results.
Generally speaking, there are several ways of proving something. For instance,
we can make a direct proof, or a proof by contradiction (for more on Logic and
proof techniques, see the first chapters of [21]). In this sense, the principle of
mathematical induction will provide us yet another way of proving certain types
of mathematical statements.
To understand how this principle works, let’s consider a subset A N such that
1 2 A. Also, suppose we know that, whenever a certain natural number k is in A,
then, so k C 1 also is in A. Therefore, 1 2 A assures that 2 D 1 C 1 2 A; accordingly,
2 2 A allows us to conclude that 3 D 2 C 1 2 A. Continuing this way, we conclude
that A contains all natural numbers, or, which is the same, that A D N. This intuitive
discussion can be formalized as the following axiom, known as the principle of
mathematical induction.
Axiom 4.1 Let A N be a set satisfying the following conditions:
(a) 1 2 A.
(b) k 2 A ) k C 1 2 A.
Then A D N.
At this point, a natural question is: how can we apply the principle of mathemat-
ical induction to prove something in Mathematics? In order to answer it, suppose
there is given a property P.n/ of the natural number n, which we wish to prove to
be true for every n 2 N. Then, we let
A D fk 2 NI P.k/ is trueg
Thus, in order to prove that P.n/ is true for every n 2 N, it suffices to prove that
A D N; yet in another way, by invoking the principle of mathematical induction, it
suffices to prove that:
• 1 2 A;
• k 2 A ) k C 1 2 A.
In turn, the definition of the set A assures that, showing the validity of the two items
above is the same as showing that
• P.1/ is true;
• P.k/ true ) P.k C 1/ true.
The previous discussion can be summarized as the following recipe for a proof
by induction.
Proposition 4.2 If P.n/ is a property of the natural number n, then P.n/ is true for
every n 2 N if and only if the following two conditions are satisfied:
(a) P.1/ is true;
(b) P.k/ true ) P.k C 1/ true.
To understand how a proof by induction works in practice, let’s start by
discussing two examples, the first of which was already considered in the previous
chapter.
Example 4.3 Prove that, for every n 2 N, the sum of the first n odd natural numbers
equals n2 .
Proof Since the kth odd natural number is 2k1, the property P.n/ is, in this case,
the same as
X
n
P.n/ W .2j 1/ D n2 :
jD1
As was said before, in order to prove by induction that P.n/ is true for every n 2 N,
we have to verify that:
4.1 The Principle of Mathematical Induction 89
i. P.1/ is true.
ii. P.k/ true ) P.k C 1/ true.
Checking i. is immediate: the first odd natural number is 1, which is the same
as 12 . In order to verify ii., we have to assume that P.k/ is true, i.e., that
X
k
.2j 1/ D k2 ; (4.1)
jD1
and find a way to deduce from this that P.k C 1/ is also true, i.e., that
X
kC1
.2j 1/ D .k C 1/2 : (4.2)
jD1
Since we are assuming that P.k/ is true, we can substitute (4.1) into the left hand
side of (4.2) to get
X
kC1 X
k
.2j 1/ D .2j 1/ C .2k C 1/ D k2 C .2k C 1/ D .k C 1/2 :
jD1 jD1
1
n.n C 1/.2n C 1/:
6
Proof Since the kth perfect square is the number k2 , the property P.n/ is, in this
case,
X
n
1
P.n/ W j2 D n.n C 1/.2n C 1/:
jD1
6
X
k
1
j2 D k.k C 1/.2k C 1/; (4.3)
jD1
6
90 4 Induction and the Binomial Formula
and find a way to deduce that P.k C 1/ is also true, i.e., that
X
kC1
1
j2 D .k C 1/Œ.k C 1/ C 1Œ2.k C 1/ C 1: (4.4)
jD1
6
X
kC1 X
k
j2 D j2 C .k C 1/2
jD1 jD1
1
D k.k C 1/.2k C 1/ C .k C 1/2
6
1
D .k C 1/Œk.2k C 1/ C 6.k C 1/
6
1
D .k C 1/.k C 2/.2k C 3/:
6
A D fk 2 NI P.k/ is trueg
A D fa; a C 1; a C 2; : : :g
m
P.n/ is true for every natural number n a:
This way, we obtain the following more general recipe for proofs by induction.
4.1 The Principle of Mathematical Induction 91
Proposition 4.6 Let a 2 N. If P.n/ is a property of the natural number n, then P.n/
is true for every natural number n a if and only if the following two conditions
are satisfied:
(a) P.a/ is true;
(b) P.k/ true ) P.k C 1/ true.
The next example illustrates the fact that, sometimes, this more general form of
proof by induction is really necessary.
Example 4.7 Prove that nŠ > 2n , for every natural number n 4.
Proof First of all, observe that we really need to start from n D 4, for the inequality
nŠ > 2n is not true when n D 1, 2 or 3.
The property P.n/ we wish to prove is
P.n/ W nŠ > 2n :
kŠ > 2k :
From this, we wish to deduce the trueness of P.k C 1/, i.e., that .k C 1/Š > 2kC1 .
To this end, we first use the trueness of P.k/ to get
.k C 1/Š D .k C 1/ kŠ > .k C 1/ 2k I
.k C 1/ 2k 2kC1
for every integer k 1. Hence, by combining the two inequalities got above, we
arrive at .k C 1/Š > 2kC1 , thus concluding that P.k C 1/ is indeed true.
Therefore, it follows by induction that the property P.n/ is true for every integer
n 4. t
u
Before we present another example, it is appropriate to make some comments
on terminology: in a proof by induction, the step P.k/ ) P.k C 1/ is generally
called the induction step. In order to execute it, we assume that the property P.k/
is indeed true (and this assumption constitutes our induction hypothesis) and,
then, use this trueness, possibly together with other arguments, to deduce that the
property P.k C 1/ is also true. Thus, a proof by mathematical induction, as stated in
Proposition 4.6, is usually done by executing the following steps:
92 4 Induction and the Binomial Formula
1 1 1
C C D 1:
2 3 6
Now suppose, by induction hypothesis, that, for a certain natural k 3, there
exist natural numbers x1 < x2 < < xk such that
1 1 1
C CC D 1:
x1 x2 xk
1 1
Multiplying both sides of the equality above by 2 and adding 2 to both sides of
the resulting equality, we obtain
1 1 1 1
C C CC D 1:
2 2x1 2x2 2xk
Since 2 < 2x1 < 2x2 < < 2xk , we have got k C 1 pairwise distinct
natural numbers whose sum of inverses is also equal to 1, and this completes the
induction step. t
u
When applying the principle of mathematical induction to prove something,
one must be very careful in the execution of the induction step, to avoid absurd
conclusions. This point is better illustrated with the following classical example and
the subsequent discussion on the reasoning we shall present.
Example 4.9 In a farm, if one horse is white, show that all horses are white.
Proof Let’s “prove” the statement above by induction on the number n of horses.
To the initial step, note that if a farm has just one horse and at least one of its
horses is white, then surely all horses in the farm are white.
By induction hypothesis, assume that in a farm with k horses, if at least one of
them is white, then all are white. Then, consider a farm with k C 1 horses, also with
4.1 The Principle of Mathematical Induction 93
at least a white one. Take one horse from the farm, one that is not that we knew a
priori to be white. Since we are left with k horses in the farm, at least one of which
is white, it follows from the induction hypothesis that all of the k horses are white.
Now, bring back the horse who was taken from the farm, and take from it one that
was left at the first time. Again, we are left with k horses, at least one of which is
white, so that, applying the induction hypothesis one more time, we conclude that
all of these k horses are also white. However, since the horse we took from the
farm this second time was already white, we conclude that all of the k C 1 horses
are white. t
u
It is quite evident that there is some absurd in the “proof” presented above, for the
statement of the example does not reflect the reality. The point is that we did not
succeed in executing the induction step, for our reasoning does not work in a farm
with two horses, as you can readily verify.
There is yet another important form of the principle of mathematical induction,
generally referred to as the strong principle of mathematical induction, or simply
as strong induction. We describe it next.
Axiom 4.10 Let A N be a set satisfying the following conditions:
(a) 1 2 A.
(b) f1; : : : ; kg A ) k C 1 2 A.
Then, A D N.
At this point, the reader should probably find it clear how to use of the strong
principle of induction to compose proofs. Nevertheless, in order to further illustrate
it, we take a look at a couple of examples.
Example
p p - adapted) Show that, for every n 2 N, the number .7 C
4.11 (OCS
4 3/n C .7 4 3/n is a positive even integer.
p p
Proof Let u D 7C4 3 and v D 74 3. Then, uCv D 14 and uv D 1, from where
it follows that u and v are the roots of the second degree equation x2 14x C 1 D 0.
It follows from this that u2 D 14u 1 and v 2 D 14v 1, and hence, for every
integer k 2,
uk D 14uk1 uk2 and v k D 14v k1 v k2 :
Thus, letting sj D uj C v j and adding both relations above, we get, for every integer
k 2, that
sk D 14sk1 sk2 :
Now, observe that s0 D 2 and s1 D u C v D 14, both integers. Suppose, by
induction hypothesis, that sk 2 Z for every integer 1 k < n. Then, the recurrence
relation above gives
sn D 14sn1 sn2 ;
from where we conclude that sn , being the sum of two integers, is also integer.
94 4 Induction and the Binomial Formula
2k n < 2kC1 ;
Using once more the inequality n 2k < 2k , it follows that 2a0 C 2a1 C C 2al < 2k
and, thus, al < k. Hence,
where in the last passage we used the formula for the sum of the terms of a finite
GP. It follows from the above that 2aj < 2bl C1 and, hence, aj < bl C 1, i.e., aj bl .
Changing the roles of aj and bl in the above computations, we analogously get bl
aj and, thus, aj D bl . Letting aj D bl D k, say, we obtain
n 2k D 2a0 C 2a1 C C 2aj1 D 2b0 C 2b1 C C 2bl1 : (4.5)
Since n 2k < n, if we now invoke the uniqueness part of the induction hypothesis
(i.e., if we assume that the binary representation of every natural number less than
n is unique), then (4.5) gives j 1 D l 1 and a0 D b0 , a1 D b1 , . . . , aj1 D bl1 ,
as we wished to prove. t
u
Before you go on and try the proposed problems, there are two remarks that
ought to be made. Firstly, if we want to prove the validity of a certain property P.n/
of the natural number n, then using induction will not always be the best possible
choice. For example, try Problem 1 and, after you are done, compare your proof by
induction to the one given in Example 3.7. Secondly, apart from Problems 8 to 12,
we do not present any applications of induction (and there are lots of them) to either
Combinatorics or Number Theory. Instead, we refer the interested reader to [5].
1. Use the principle of induction to prove that the sum of the first n natural
numbers equals n.nC1/
2
.
2. Prove that, for every natural number n, one has
n.n C 1/ 2
13 C 23 C C n3 D :
2
3. Prove that, for every natural number n, one has
1 1 1 1 1 1
1 C C D C CC :
2 3 2n 1 n nC1 2n 1
1 1
4. (Canada) For n 2 N, let h.n/ D 1 C 2 C 3 C C 1n . Prove that
1
12 C 32 C 52 C C .2n 1/2 D .2n 1/2n.2n C 1/:
6
96 4 Induction and the Binomial Formula
X
n1
1 n1
D :
jD1
aj ajC1 a1 an
17. (Macedonia) Let x be a nonzero real number, such that x C x1 2 Z. Prove that
xn C xn 2 Z, for every integer n.
18. (Brazil) Let .xn /n1 and .yn /n1 be sequences of real numbers such that, for
every integer k 1, we have xkC1 D x3k 3xk e ykC1 D y3k 3yk . If x21 D y1 C 2,
show that x2n D yn C 2 for every integer n 1.
19. (Putnam) Let .xn /n0 be a sequence of nonzero real numbers satisfying, for
every integer n 1, the recurrence relation x2n xn1 xnC1 D 1. Prove that there
exists a real number ˛ such that xnC1 D ˛xn xn1 , for every n 2 N.
4.1 The Principle of Mathematical Induction 97
The next four problems concern the Fibonacci sequence. We refer the reader
to Example 3.17, for a review of its definition.
20. Let .Fn /n1 be the Fibonacci sequence. Prove that, for every n 2 N, the
following identities are true:
(a) F1 C F2 C C Fn D FnC2 1.
(b) F12 C F22 C C Fn2 D Fn FnC1 .
(c) F1 C F3 C C F2n1 D F2n .
(d) F2 C F4 C C F2n D F2nC1 1.
2
(e) FnC1 Fn FnC2 D .1/n .
21. Let .Fn /n1 be the Fibonacci sequence. Prove that, for every m; n 2 N, with
m > 1, we have FmCn D Fm FnC1 C Fm1 Fn .
22. Let .Fn /n1 be the Fibonacci sequence and .Ln /n1 be the Lucas sequence (cf.
Problem 7, page 77). Given m; n 2 N, prove that:
(a) 2FmCn D Fm Ln C Fn Lm .
(b) 2LmCn D 5Fm Fn C Lm Ln .
23. Let .Fn /n1 be the Fibonacci sequence.
(a) Prove thatpFn D n2 if and only if n D 1 or 12.1
(b) If ˛ D 1C2 5 , do the following items:
i. Prove that ˛ n D Fn ˛ C Fn1 , for every integer n 1.
ii. Find all n 2 N such that ˛ n n2 ˛ is an integer.
24. (TT) Let .an /n1 be a sequence of pairwise distinct natural numbers. Prove that
there are infinitely many naturals k such that ak > k.
25. (France) Let .an /n1 be a sequence of positive real numbers such that a1 D 1
and
p p 1
a < xn a C n1 :
2
27. (France) Prove that, for every integer n > 5, there exist n positive integers such
that the sum of the inverses of their squares is equal to 1.
1
More generally, it can be shown that Fn is a perfect square if and only if n D 1 or 12. For a proof
of this fact, see the problems of Section 12.3 of [5].
98 4 Induction and the Binomial Formula
28. Given m; n 2 N, with m > 1, prove that there exist unique nonnegative integers
k, a0 , a1 , : : :, ak such that 0 ai < m for 0 i k, ak ¤ 0 and
n D ak mk C ak1 mk1 C C a2 m2 C a1 m C a0 :
In this case, we write n D .ak ak1 : : : a1 a0 /m and say that the right hand side
above is the representation of n in base m, or the madic representation of n.
29. (Sweden) Prove that, for every natural number n, there exists a unique sequence
.aj /j1 of integers, such that 0 aj j for every j 1 and
n D a1 1Š C a2 2Š C a3 3Š C :
30. * Prove Zeckendorf’s theorem2 : every natural number can be uniquely written
as a sum of Fibonacci numbers of nonconsecutive and greater than 1 indices.
One could guess that it would be much more reasonable to define 0Š as being
equal to 0. Nevertheless, the reasons for letting 0Š be equal to 1 will soon become
evident.
2
After Édouard Zeckendorf, belgian mathematician of the XX century.
4.2 Binomial Numbers 99
for every nonnegative integers n for which the binomial numbers above are defined.
On the other hand, for every integers n and k such that 0 k n, we have
! !
n n
D : (4.7)
k nk
Proof It suffices to apply the definition of binomial number to the right hand side
of the equality above, together with some algebraic manipulations:
! !
n1 n1 .n 1/Š .n 1/Š
C D C
k k1 kŠ.n 1 k/Š .k 1/Š.n k/Š
.n 1/Š 1 1
D C
.k 1/Š.n 1 k/Š k nk
.n 1/Š n
D
.k 1/Š.n 1 k/Š k.n k/
!
nŠ n
D D :
kŠ.n k/Š k
t
u
3
Another proof is the object of Problem 8.
4
After Michael Stifel, german mathematician of the XVI century.
100 4 Induction and the Binomial Formula
0, read
• The entries of column from
top to bottom, are respectively equal to the
binomial numbers 00 ; 10 ; 20 ; 30 ; : : :. As we have already seen, all such numbers
are equal to 1.
• Line 0 is formed solely by the binomial number 00 D 1. Line 1 is composed by
the binomial numbers 10 and 11 , both of which are also equal to 1.
• In general, the entries ofline
n,
read from
left to right, are respectively equal to
the binomial numbers n0 ; n1 ; n2 ; : : : ; nn .
According to the above recipe, we show below the first few lines of Pascal’s
triangle:
Pascal’s Triangle
0
0
1 1
0 1
2 2 2
0 1 2
3 3 3 3
0 1 2 3
4 4 4 4 4
0 1 2 3 4
5 5 5 5 5 5
0 1 2 3 4 5
6 6 6 6 6 6 6
0 1 2 3 4 5 6
With respect to Pascal’s triangle, Stifel’s relation says that, whenever we add, in
line n 1, the entries at columns k 1 and k, we get the entry situated at line n and
column k. This is more difficult to say than to understand and verify,
and allows us
to recursively get the numerical values of the binomial numbers nk . The table below
shows the numerical values of the binomial numbers nk for 0 n 6, obtained
with the aid of Stifel’s relation.
5
After Blaise Pascal, french mathematician of the XVII century. Besides the triangle that bears his
name, there is an important Pascal’s theorem in the theory of conics, which has greatly motivated
the developments of Projective and Algebraic Geometry.
4.2 Binomial Numbers 101
1
1 1
1 2 1
1 3 3 1
1 4 6 4 1
1 5 10 10 5 1
1 6 15 20 15 6 1
··· ··· ··· ··· ··· ··· ··· ···
More gerally, since n0 D 1 and nn D 1 whenever these numbers are defined,
it is not difficult for the reader to convince himself or herself that nk 2 N for all
integers n and k such that 0 k n. We give a formal proof of this fact in the
corollary below.
Corollary 4.16 For all integers n and k such that 0 k n, we have nk 2 N.
Proof Let’s make induction on n 0, the case n D 0 being obvious, for the only
such binomial number is 00 D 1.
Now suppose, by induction hypothesis, that n1 is a natural number for every
j
0 j n 1, and consider a binomial number of the form nk . There are two cases
to consider:
(i) If k D 0 or k D n, then we have already observed that n0 D nn D 1.
(ii) If 1 k n 1, then it follows from the induction hypothesis that both n1
n1 k
and k1 are natural numbers. Hence, Stifel’s relation gives
! ! !
n n1 n1
D C 2 N:
k k k1
t
u
Later (cf. Theorem 4.20), we shall see that one of the main utilities of Pascal’s
triangle is that it provides an easy and quick way to write down the expansion of
the binomial .x C y/n for small values of n. For the time being, let’s establish some
useful identities relating the entries of a row, column or diagonal of it.
The formula of the following proposition is known as the columns’ theorem of
Pascal’s triangle.
Proposition 4.17 In the column n of Pascal’s triangle, the sum of the entries in
rows n, n C 1, . . . , n C k 1 is equal to the entry located at column n C 1 and row
n C k. In symbols,
! !
X
k1
nCj nCk
D : (4.9)
jD0
n nC1
102 4 Induction and the Binomial Formula
t
u
Given an integer n 0, the diagonal n of Pascal’s triangle is formed by the
binomial numbers
! ! ! !
n nC1 nC2 nC3
; ; ; ;::::
0 1 2 3
With respect to them, the following corollary is the diagonals’ theorem of Pascal’s
triangle.
4.2 Binomial Numbers 103
Corollary 4.19 In the diagonal n of Pascal’s triangle, the sum of the entries in rows
0; 1; : : : ; k1 is equal to the entry located at row nCk and column k1. In symbols,
! !
X
k1
nCj nCk
D : (4.10)
jD0
j k1
nCj nCj
Proof Since j
D n
, it follows from the columns’ theorem that
! k1 ! ! !
X
k1
nCj X nCj nCk nCk
D D D :
jD0
j jD0
n nC1 k1
t
u
if n is even, and
! ! ! ! !
n n n n n
< < < D > >
0 1 n1
2
nC1
2
n
if n is odd.
5n
4. Prove that, for every integer n 2, one has 2 4 < 2n
n .
5. Given natural numbers k and m, with m > k, prove that
n
X
m
1
nnC1
k1
D 1 mC1 :
nDk k k k
7. Let .Fk /k1 be the Fibonacci sequence (cf. Example 3.17). Show that, for every
n 2 N, one has
bX !
2 c
n1
nj1
Fn D ;
jD0
j
8. * Givenintegers
n and k such that 0 k n, prove that the set f1; 2; : : : ; ng has
exactly nk subsets of k elements each.
We shall now obtain Newton’s binomial formula,6 i.e., we shall explicitly write
.x C y/n as a sum of monomials of the form xk yl .
Theorem 4.20 Given n 2 N, we have
!
Xn
n nj j
.x C y/ D
n
x y: (4.11)
jD0
j
6
Sir Isaac Newton, english mathematician and physicist of the XVII century, is considered to be
one of the greatest scientists ever. Actually, it is difficult to properly address Newton’s contribution
to the development of science. Known as the father of modern Physics, Newton also created,
together with G. W. Leibniz, the Differential and Integral Calculus. His masterpiece, Philosophiae
Naturalis Principia Mathematica, is one of the most influential books ever written and contains
the cornerstones of both Calculus and Physics.
4.3 The Binomial Formula 105
! ! !
k C 1 kC1 X k C 1 kC1l l
k
k C 1 kC1
x C x y C y
0 lD1
l kC1
It is customary to write
!
n nj j
Tj D x y
j
and say that such a monomial Tj is the general term of the expansion of .x C y/n .
Hence,
.x ˙ y/n D T0 ˙ T1 .˙1/n Tn :
Making
! !
X k kj j p j X k p
cD a b r and d D akj bj rj1 ; (4.12)
0jk
j 0jk
j
2jj 2−j
p p
it is immediate that c; d 2 Q (since rj 2 Q for an even j, and rj1 2 Q for
an odd j).
4.3 The Binomial Formula 107
p
(b) It sufficespto note that the expansion of .a b r/k is essentially equal to that
of .a C b r/k , the difference lying on the signs of half of the summands. More
precisely, in the notations of (4.12), we get
! !
p X k kj j p j p X k kj j p j1
.a b r/k D a b r r a b r
0jk
j 0jk
j
2jj 2−j
p
D c d r:
t
u
Example 4.24 Use Newton’s binomial formula to prove Lagrange’s identity:7
!2 !
Xn
n 2n
D : (4.13)
jD0
j n
Proof Writing .1Cx/2n D .1Cx/n .1Cx/n and applying Newton’s binomial formula
to both sides, we get
2n
! ! !0 n ! 1 ! !
X 2n k Xn
n i @X n j A X n n iCj
n
x D x x D x :
kD0
k iD0
i jD0
j i;jD0
i j
Comparing the coefficients of xn in the first and last expressions above and
using (4.7), we get
! ! ! ! ! !2
2n X n n Xn
n n Xn
n
D D D :
n iCjDn
i j iD0
i ni iD0
i
t
u
Item (a) of the following corollary is known as the lines’ theorem of the Pascal’s
triangle.
Corollary 4.25 Given n 2 N, we have:
Pn n
(a) D 2n .
PjD0 jn P
(b) 0jn
j
D 0jn nj D 2n1 :
2jj 2−j
Proof For item (a), it suffices to set x D y D 1 in the formula for .x C y/n .
7
Joseph Louis Lagrange, french physicist and mathematician of the XVIII century. Lagrange was
one of the greatest scientists of his time, with notable contributions to Physics and Mathematics.
In particular, he was a pioneer in the fields of Calculus of Variations and Celestial Mechanics.
108 4 Induction and the Binomial Formula
P n P n
Now, letting A D 0jn
j
and B D 0jn
j
, it follows from (a) and the
2jj 2−j
above relation that
A C B D 2n
:
ABD0
for kD 3 we
in terms of n, k andr.For instance, can
compute, interms n2 N, the
of
values of the sums n0 C n3 C n6 C , n1 C n4 C n7 C and n2 C n5 C n8 C .
The corresponding deduction uses the multisection formula, which will be obtained
in Chapter 15 of [5], with the aid of complex numbers.
We finish this small chapter with an example that illustrates the use of the lines’
theorem of Pascal’s triangle to the computation of sums.
Example 4.27 Given n 2 N, compute the value of
! ! ! !
n n n n
C2 C3 CCn :
1 2 3 n
t
u
value of n.
1
3. Show that .1; 1/n 200 .n2 C 19n C 200/.
4. (OCM) Without directly evaluating all summands, compute the value of the
sum
1 1 1 1 1 1
C C C C C :
10Š 3Š8Š 5Š6Š 7Š4Š 9Š2Š 11Š
65
5. Find the maximal term in the expansion of 1 C 13 .
6. (Baltic Way). Let a, b, c and d be real numbers for which a2 C b2 C .a C b/2 D
c2 C d2 C .c C d/2 . Prove that a4 C b4 C .a C b/4 D c4 C d4 C .c C d/4 .
22n1
7. Given n 2 N, prove that 2n n > n .
8. In the expansion of x.1Cx/n , divide the coefficient of each term by the exponent
of x in that term. Prove that the sum of all these numbers equals 2 n 1 .
nC1
Pn n j
9. Given a 2 Z n f0g and n 2 N, compute the value of the sums jD0 j ja and
Pn n 2
jD0 j j in terms of a and n.
10. The sequence .ak /k1 is an AP. Prove that, for every integer n > 1, the following
identities are valid:
Pn n
(a) .1/jC1 aj D 0.
n n
PjD0 j
jC1 2
(b) jD0 j .1/ ajC1 D 0.
q
11. * If 0 < q < 1 and n 2 N, prove that qn < qCn.1q/ .
p p
12. Given a; b; n 2 N, with n > 1, prove that the number a C n b is the root of a
polynomial equation with integer coefficients and degree 2n.
13. (Croatia) Let x, y and z be nonzero real numbers, such that x C y C z D 0. Prove
that the value of
x5 C y5 C z5
xyz.xy C yz C zx/
where the sum in the right hand side extends over all possible choices of
nonnegative integers j; k; l such that j C k C l D n.
15. Use the result of the previous problem to expand the trinomials .x C y C z/3 and
.x C y C z/4 .
10
16. Compute, in the expansion of 1 C x C 6x , the term that doesn’t depend on x.
17. Prove the following identities involving trinomial numbers:
P n
jCkClDn j;k;l D 3 .
n
(a)
P
jCkClDn .1/ j;k;l D 1.
l n
(b)
18. * Prove that, for every n 2 N, we have
!
Xn
n
F2nC1j D F2nC1 1;
jD1
j
995
!
X .1/k 1991 k 1
D :
kD0
1991 k k 1991
Chapter 5
Elementary Inequalities
x2 C y2
jxyj; (5.1)
2
aCb p
ab; (5.2)
2
p p
with equality if and only if a D b, i.e., if and only if a D b.
The simplicity of the previous reasoning hides its importance. Actually, (5.2)
is a particular case of a much more general inequality, known as the arithmetic-
geometric mean inequality (see Theorem 5.7). For the time being, let’s see how to
deduce other interesting inequalities from (5.2).
Example 5.1 Given positive real numbers x and y, we have:
(a) x C 1x 2, with equality if and only if x D 1.
(b) 1x C 1y xCy
4
, with equality if and only if x D y.
Proof
(a) Applying (5.2) with a D x and b D 1x , we get
r
1 1
x C 2 x D 2;
x x
x2 C y2 C z2 xy C xz C yz; (5.3)
x 2 C y2 x2 C z2 y2 C z2
xy; xz; yz:
2 2 2
It now suffices to add these three inequalities to reach (5.3).
If x D y D z, then (5.3) clearly becomes an equality. Conversely, if x ¤ y (the
2 2
cases x ¤ z and y ¤ z can be treated analogously), then x Cy 2 > xy. Therefore,
an obvious generalization of item (e) of Proposition 1.2 assures that, adding the
inequalities
5.1 The AM-GM Inequality 113
x2 C y2 x2 C z2 y2 C z2
> xy; xz; yz;
2 2 2
aCbCcCd p
4
abcd; (5.4)
4
with equality if and only if a D b D c D d.
p p
Proof We already know that aCb 2
ab and cCd
2
cd. Therefore,
p p q
aCbCcCd aCb
2
C cCd
2 ab C cd p p p
4
D ab cd D abcd;
4 2 2
where the last inequality above follows from yet another application of (5.2).
Equality happens if and only if we have equality inpall of the inequalities
p above.
For the first one of them, we must have aCb D ab and cCd
D cd, and,
2 p 2 p
hence, a D b and c D d. For the second one,pwe mustp have ab D cd; however,
since a D b and c D d, this is the same as a2 D c2 , i.e., a D c. Therefore, we
have equality if and only if a D b D c D d. t
u
The previous example raises the natural question of whether a similar inequality
is true for three positive real numbers. As we have already anticipated, this is indeed
the case, albeit its deduction will not be as immediate as the one just discussed
(nevertheless, see Problem 11).
Example 5.4 Given positive real numbers a, b and c, we have
aCbCc p
3
abc; (5.5)
3
with equality if and only if a D b D c.
ProofpWe apply the inequality of Example 5.4 to the four positive reals a, b, c and
d D 3 abc to get
p q
a C b C c C 3 abc 4 p
3
p
4
p3
abc abc D d3 d D d D abc:
4
p p p
Then, a C b C c C 3 abc 4 3 abc or, which is the same, aCbCc
3
3 abc.
114 5 Elementary Inequalities
y
z
x
From the above computations it is clear that equality holds if and only if it also
happens in the inequality
p q
aCbCcC 3
abc 4 p
3
abc abc:
4
Therefore,
p it follows from Example 5.3 that equality holds if and only if a D b D
c D 3 abc, i.e., if and only if a D b D c. u
t
At this point, the reader is probably formulating in his/her mind a natural
generalization of inequalities (5.2), (5.4) and (5.5), which we anticipated to be true.
Before we formally state and prove it, let’s see an interesting application of (5.5).
Example 5.5 A box (i.e., a rectangular parallelepiped1) is such that the sum of its
edge lengths equals 48 cm. If the box encloses the largest possible volume, find its
dimensions.
Solution Figure 5.1 shows a rectangular parallelepiped with dimensions x, y and
z (measured in centimeters), such that the sum of the lengths of its twelve edges
equals 48 cm. Since each dimension comprises four equal and parallel edges, this is
the same as saying that x C y C z D 12.
It is a well known fact that the volume V of such a parallelepiped is given by
the formula V D xyz. Hence, algebraically our problem reduces to the one of
maximizing the product xyz, under the restrictions that x; y; z > 0 and xCyCz D 12.
To this end, we apply inequality (5.5) to get
3 3
xCyCz 12
V D xyz D D 64;
3 3
thus concluding that V is at most 64 cm3 . Example 5.4 also shows that the volume
equals 64 cm3 (i.e., that equality holds in the above inequality) if and only if x D
y D z or, which is the same, if and only if x D y D z D 4 cm. Therefore, the box of
largest possible volume is a cube with edge length equal to 4 cm. t
u
1
For the elementary facts on the notion of volume of solids, we refer the reader to [4].
5.1 The AM-GM Inequality 115
Before we state the promised generalization of (5.2), (5.4) and (5.5), we need
some terminology.
Definition 5.6 Given n > 1 positive real numbers a1 ; a2 ; : : : ; an , we define their:
(a) Arithmetic mean as the number 1n .a1 C a2 C C an /.
p
(b) Geometric mean2 as the number n a1 a2 : : : an .
In the context of the previous definition, what we did in (5.2) and in Examples 5.3
and 5.4 was to show that the arithmetic means of two, three or four positive reals
are always greater than or equal to the respective geometric means, with equality in
each case if and only if the given numbers are all equal. We establish the general
case in the coming result, whose proof can be omitted in a first reading. As the title
of this section suggests, inequality (5.6) below is known as the arithmetic mean-
geometric mean inequality.
Theorem 5.7 Given n > 1 positive reals a1 ; a2 ; : : : ; an , their arithmetic mean is
always greater than or equal to their geometric mean. In symbols:
a1 C a2 C C an p
n a1 a2 : : : ; an ; (5.6)
n
with equality if and only if a1 D a2 D D an .
Proof Firstly, let’s prove by induction that the desired inequality is true whenever n
is a power of 2, with equality if and only if a1 D a2 D D an . To this end, we have
to verify the initial case n D 2 (which was already done along the discussion that
established (5.2)), formulate the induction hypothesis (for n D 2j , say) and execute
the induction step (i.e., deduce the case n D 2jC1 from the case n D 2j ). However,
since 2jC1 D 2 2j , it suffices to suppose the inequality to be true for any k positive
reals (with equality if and only if these k numbers are all equal) and, from this, to
deduce that it is also true for any 2k positive reals (with equality if and only if these
2k positive reals are also all equal). Hence, to perform the induction step, consider
2k positive reals a1 ; a2 ; : : : ; a2k . The induction hypothesis, together with the case
n D 2, give
0 1
2k
1 X 1 @1 X 1X
k k
aj D aj C akCj A
2k jD1 2 k jD1 k jD1
1 p p
k
a1 : : : ak C k akC1 : : : a2k
2
q
p p
k a1 : : : ak k akC1 : : : a2k
p
D 2k a1 : : : ak akC1 : : : a2k :
2
The reader maybe find it useful if we observe that the adjective geometric attached to this number
comes from the case n D 2. In this case, inequality (5.2) has a simple geometric interpretation, for
which we refer to the problems of Sect. 4:2 of [4].
116 5 Elementary Inequalities
To have equality, we must have it in all passages above. Then, it must be that
a1 C C ak p akC1 C C a2k p
D k a1 : : : ak ; D k akC1 : : : a2k
k k
and
p p
k a :::a C k a q
1 k kC1 : : : a2k p p
D k a1 : : : ak k akC1 : : : a2k :
2
For the first two equalities, it follows from the induction hypothesis that a1 D
p
D ak and akC1 D D a2k . For the third one, we must have k a1 : : : ak D
p
kC1 : : : a2k ; this last condition, together with the two former ones, implies that
k a
true). We get
q
a1 C C an C a C C a 2k
a1 : : : an a2k n
2k
p
2k
p
2k
D an a2k n D a2k D a:
a1 C a2 C C an p
a D n a1 a2 : : : an :
n
For the equality to hold, we already know that we must have a1 D a2 D D
an D a D D a; in particular, all of the numbers a1 ; a2 ; : : : ; an must be equal.
Finally, it’s immediate to see that, if all of the numbers a1 , a2 , . . . , an are equal, then
equality holds. t
u
The following corollary generalizes item (b) of Example 5.1. For an alternative
proof, see Problem 21.
Corollary 5.8 Given n > 1 positive reals a1 ; a2 ; : : : ; an , we have
1 1 1
.a1 C a2 C C an / C CC n2 ; (5.7)
a1 a2 an
5.1 The AM-GM Inequality 117
In order to have equality in (5.7), we must have equality in (5.6), and we know that
this implies a1 D a2 D D an . Conversely, it is immediate to verify that, if all of
the n given numbers are equal, then we have equality in (5.7). t
u
Remark 5.9 The inequality (5.7) is sometimes referred to as the arithmetic mean-
harmonic mean inequality. This is due to the fact that it can be written as
1
a1 C a2 C C an 1=a1 C 1=a2 C C 1=an
;
n n
and that the right hand side above (i.e., the inverse of the arithmetic mean of the
inverses of a1 , a2 , . . . , an ) is known as the harmonic mean of a1 , a2 , . . . , an .
We now illustrate the use of the inequality on the arithmetic and geometric means
in three examples.
Example 5.10 (Israel-Hungary) Let k and n be positive integers, with n > 1. Prove
that
r !
1 1 1 n k C 1
C CC >n 1 :
kn kn C 1 kn C n 1 k
X
n1 n1
X Xn1
1 1 kn C j C 1
Cn D C1 D
jD0
kn C j jD0
kn C j jD0
kn C j
v r
un1
uY kn C j C 1 n k C 1
> nt
n
Dn ;
jD0
kn C j k
where we’ve applied the inequality on the arithmetic and geometric means once.
Note that, since the numbers knCjC1
knCj
are pairwise distinct, equality never holds. u
t
118 5 Elementary Inequalities
Letting S denote the left hand side of the last expression above, it follows
from (5.6) and (5.7) that
1 1 1
S D .a C b C c/ C C 3
a b c
2 1 1 1 1 1 1 1
D .a C b C c/ C C C .a C b C c/ C C 3
3 a b c 3 a b c
2 3 1
.a C b C c/ p C 93
3 3
abc 3
2.a C b C c/
D p3
:
abc
t
u
Example 5.12 Gabriel has a sheet of cardboard of 2 m by 3 m. In order to assemble
an open box, he cuts four equal squares from the corners of the sheet, folds it along
the cuts and glues the lateral faces of the box along their common edges. If the box
is to have the largest possible volume, what should be the length of the sides of the
squares he cut? Justify your answer.
Solution If x is the common length of the sides of the cut squares, then one must
clearly have 0 < x < 1. Since the box has height x and its bottom is a rectangle of
side lengths 2 2x and 3 2x, the volume Gabriel wants to maximize depends on
x and equals .2 2x/.3 2x/x.
One possibility for him is to try to apply the inequality between the arithmetic
and geometric means to get rid of x and, then, see what the condition for equality
says about the size of x. However, he cannot do this directly, for, although
.2 2x/ C .3 2x/ C x 3
.2 2x/.3 2x/x ;
3
the expression at the right hand side still depends on x.
5.1 The AM-GM Inequality 119
Nevertheless, the following trick does the job: he starts by choosing positive reals
a, b and c such that a.22x/Cb.32x/Ccx doesn’t depend on x and such that there
exists at least one value of x 2 .0; 1/ for which a.2 2x/ D b.3 2x/ D cx. This
amounts to finding a positive solution .a; b; c/ for the linear equation 2aC2bc D 0,
such that this solution, in turn, gives equal solutions for the first degree equations
2.a b/x D 2a 3b and .2a C c/x D 2a, which should belong to the interval .0; 1/.
Hence, we should have
2a 3b 2a
2a C 2b D c and D 2 .0; 1/:
2.a b/ 2a C c
Equality holds if and only if a.22x/ D b.32x/ D cx. However, we already know
2a
that these two equations have x D 2aCc as common solution, so that the maximal
p
2a 5 7
possible volume is attained only for x D x0 D 2aCc
D 6
, and equals
p
1 2a C 3b 3 10 C 7 7
D x0 .2 2x0 /.3 2x0 / D :
abc 3 27
t
u
Later, when we have the methods of Calculus at our disposal, the previous
example will fall into the general framework of the analysis of the first and second
variations of the function V.x/ D .2 2x/.3 2x/x; as such, it will have a
straightforward solution. Nevertheless, the solution we presented above, of choosing
adequate weights prior to applying the inequality between the arithmetic and
geometric means, albeit quite tricky, is an instructive one for several other situations.
120 5 Elementary Inequalities
1 1 1
C C D0
xa xb xc
has exactly two distinct real roots.
4. (Brazil) Let a, b and c be positive real numbers. Prove that
p
.a C b/.a C c/ 2 abc.a C b C c/:
5. (USA) Prove that, for every positive real numbers a, b and c, one has
1 1 1 1
C 3 C 3 :
a3 3
C b C abc 3
b C c C abc 3
c C a C abc abc
For the next two problems, the reader shall need some Euclidean Geometry,
which we review next.3 More precisely (cf. Fig. 5.2), if a D BC, b D AC and
c D AB are the lengths of the sides of a triangle ABC, then there exist x; y; z > 0
such that a D y C z, b D x C z and c D x C y. In fact, it suffices to take x, y
and z as being equal to the lengths of the line segments determined on the sides
of ABC by the points of tangency of its incircle. In the context of inequalities
involving the lengths of the sides of a triangle, the substitution of them by y C z,
x C z and x C y is frequently referred to as Ravi’s transformation.
6. (IMO) If a, b and c are the lengths of the sides of a triangle, prove that
x I z
A x z C
3
For a thorough discussion of the facts that follow, see Sect. 3:4 of [4], for instance.
5.1 The AM-GM Inequality 121
a b c
C C 3:
bCca cCab aCbc
1 x2nC1
.2n C 1/xn :
1x
10. (England) Prove that 3a4 C b4 4a3 b, for all nonzero real numbers a and b,
with equality if and only if jaj D jbj and ab > 0.
11. * Prove directly (i.e., without appealing to (5.5)) that a3 C b3 C c3 3abc.
12. (Soviet Union) Let a, b and c be positive reals. Prove that
x2 y2 z2 y z x
2
C 2 C 2 C C :
y z x x y z
9.a3 C b3 C c3 / .a C b C c/3 :
a4 .1 C b4 / C b4 .1 C c4 / C c4 .1 C a4 / 6a2 b2 c2 ;
17. Let n > 1 be an odd integer and a1 , a2 , . . . , an be negative reals. Show that
a1 C a2 C C an p
n a1 a2 : : : an ;
n
with equality if and only if all of the ai ’s are equal.
122 5 Elementary Inequalities
20. (Poland) If x1 , x2 , . . . , xn are positive reals whose sum equals S, prove that
S S S n2
C CC ;
S x1 S x2 S xn n1
ak11 a k2 a kn
C 2 C C n a1 a2 : : : an ;
k1 k2 kn
12 22 n2 n n1 n2 1
C CC C C CC :
a1 a2 an a1 a2 a1 a3 a2 an an1
with equality if and only if ja1 j D jb1 j, ja2 j D jb2 j and ja3 j D jb3 j. Adding the left
and right hand sides of the above inequalities, we get
.a21 C a22 C a23 / C .b21 C b22 C b23 / D .a21 C b21 / C .a22 C b22 / C .a23 C b23 /
2.ja1b1 j C ja2 b2 j C ja3 b3 j/
2ja1 b1 C a2 b2 C a3 b3 j;
where in the last step we applied the triangle inequality for three real num-
bers, (2.11).
Hence, it follows from a21 C a22 C a23 D 1 and b21 C b22 C b23 D 1 that
ja1 b1 C a2 b2 C a3 b3 j 1:
Equality holds if and only if it holds in the three inequalities (5.8), as well as in
the triangle inequality. Therefore, equality holds if and only if ja1 j D jb1 j, ja2 j D
jb2 j, ja3 j D jb3 j and either a1 b1 , a2 b2 , a3 b3 0 or a1 b1 , a2 b2 , a3 b3 0. However,
it is immediate to check that such conditions are equivalent to a1 D b1 , a2 D b2 and
a3 D b3 .
Now, consider arbitrary real numbers a1 ; a2 ; a3 and b1 ; b2 ; b3 , such that at least
one of a1 ; a2 ; a3 and at least one of b1 ; b2 ; b3 are nonzero. For a positive real number
c, let xi D aci for 1 i 3. Since
jx1 y1 C x2 y2 C x3 y3 j 1;
ja1 b1 C a2 b2 C a3 b3 j
1;
cd
q q
with c D a21 C a22 C a23 and d D b21 C b22 C b23 . Moreover, equality holds if
and only if ai D dc bi for 1 i 3. Finally, note that the last inequality above is
equivalent to
q q
ja1 b1 C a2 b2 C a3 b3 j cd D a21 C a22 C a23 b21 C b22 C b23 :
Up to this point, what we have done was to establish, for n D 3, the famous
Cauchy’s inequality.4 We now turn to the general case.
Theorem 5.13 (Cauchy) Let n > 1 be an integer and a1 , a2 , . . . , an , b1 , b2 , . . . , bn
given real numbers. Then,
ˇ ˇ v v
ˇ n ˇ u n u n
ˇX ˇ uX 2 uX
ˇ a b ˇ t a t b2j ; (5.9)
ˇ j jˇ j
ˇ jD1 ˇ jD1 jD1
with equality if and only if the ai ’s and bi ’s are respectively proportional, i.e., if and
only if there exists a nonzero real number such that
Proof If all of the ai ’s or all of the bi ’s are equal to zero, there is nothing to do.
Otherwise, in order to establish (5.9), it suffices to follow the steps of the particular
case n D 3 discussed above. The only difference is that, whenever convenient, we
have to use (4.6) instead of (2.11) (see Problem 2, page 127). t
u
Later (see Example 6.18), we shall give another proof of Cauchy’s inequality
as an application of the theory of maxima and minima of quadratic functions. For
a geometric interpretation of Cauchy’s inequality for n D 2, see the problems of
Sect. 6:3 of [4].
The coming two examples illustrate how one can apply Cauchy’s inequality.
4
Augustin Louis Cauchy, one of the greatest mathematicians of the XIX century, and maybe of
History. Cauchy was one of the precursors of Mathematical Analysis, an extremely important area
of higher Mathematics. He also has his name attached to several important results in Differential
Equations and Mathematical Physics.
5.2 Cauchy’s Inequality 125
Example 5.14 Let a, b and c be given real numbers. Show that the system of
equations
3.x2 C y2 C z2 / C a2 C b2 C c2 D 6
ax C by C cz D 2
which is an absurd. t
u
Example 5.15 (Romania) Let x1 ; x2 ; : : :, xnC1 be positive reals satisfying x1 C x2 C
C xn D xnC1 . Prove that
p p
x1 .xnC1 x1 / C C xn .xnC1 xn /
p
xnC1 .xnC1 x1 / C C xnC1 .xnC1 xn /:
t
u
For future use, we collect the following corollary of Cauchy’s inequality.
Corollary 5.16 Given real numbers a1 ; : : : ; an and b1 ; : : : ; bn , we have
v v v
uX uX u n
u n u n 2 uX
t .aj C bj /2 t aj C t b2j ; (5.10)
jD1 jD1 jD1
Proof For the sake of clarity, let’s prove the corollary for n D 3, the general case
being totally analogous. Since both sides of (5.10) are nonnegative real numbers,
it suffices to show that the square of the left hand side is less than or equal to the
square of the right hand side. In symbols,
q q 2
2 2 2
.a1 C b1 / C .a2 C b2 / C .a3 C b3 / a21 C a22 C a23 C b21 C b22 C b23 :
Expanding .ai C bi /2 , it follows that the square of the left hand side equals
Cancelling the summand .a21 C a22 C a23 / C .b21 C b22 C b23 / from both sides, we
conclude that (5.10) is equivalent to
q q
2.a1 b1 C a2 b2 C a3 b3 / 2 a21 C a22 C a23 b21 C b22 C b23 ;
OC OA C OB;
where XY stands for the length of the line segment XY. This is the same as the
(geometric) triangle inequality for the (possibly degenerated) triangle OAC, and for
this reason (5.10) is also known as the triangle inequality.
p if npD 1 and we make a1 D a and b1 D b, then (5.10) reduces to
p Note also that,
.a C b/2 a2 C b2 or, which is the same, jaCbj jajCjbj. This is the reason
why (5.10), as well as its generalization (4.6), are known as triangle inequality.
5.2 Cauchy’s Inequality 127
1. Given real numbers x and y such that 3x C 4y D 12, compute the minimum
possible value of x2 C y2 .
2. Prove the general case (5.9) of Cauchy’s inequality.
3. Given positive reals a1 , a2 , . . . , an , we define its quadratic mean as the number
s
a21 C a22 C C a2n
:
n
Prove the inequality between the quadratic and arithmetic means:
s
a21 C a22 C C a2n a1 C a2 C C an
; (5.11)
n n
x2 y2 z2 y z x
C C C C ;
y2 z2 x2 x y z
X
n
a2k 1X
n
ak :
kD1
ak C bk 2 kD1
12. (APMO) Let a, b and c be the lengths of the sides of a triangle. Show that
p p p p p p
aCbcC bCcaC cCab aC bC c
14. (IMO) Let a, b and c be positive reals such that abc D 1. Prove that
1 1 1 3
C 3 C 3 :
a3 .b C c/ b .a C c/ c .a C b/ 2
This section is devoted to the study of other important elementary inequalities. The
first of them is generally attributed to the Bernoulli brothers,5 being known as Ber-
noulli’s inequality. In spite of its simplicity, we will see that it is quite useful in
applications.
Proposition 5.17 (Bernoulli) If n is a natural number and x > 1 is a real number,
then
.1 C x/n 1 C nx;
5
Jacob and Johann Bernoulli, Swiss mathematicians of the XVIII century.
5.3 More on Inequalities 129
with equality if and only if .1 C x/k D 1 C kx and kx2 D 0, i.e., if and only if x D 0.
t
u
Example 5.18 Given a natural number n and positive reals a and b, show that
a n b n
1C C 1C 2nC1 ;
b a
Since 12 C 2b
a
> 1 and 12 C 2a
b
> 1, if we apply Bernoulli’s inequality to both
summands at the left hand side above and add the results, we get
1 a n 1 b n a b
1 C C 1 C 2Cn C 1 :
2 2b 2 2a 2b 2a
Now, it suffices to apply the inequality between the arithmetic and geometric means
to obtain
r
a b a b
C 12 1 D 0;
2b 2a 2b 2a
a1 a2 an and b1 b2 bn ;
6
After Pafnuty Chebyshev, Russian mathematician of the XIX century.
130 5 Elementary Inequalities
then
! !
1X 1X 1X
n n n
ai bi ai bi ;
n iD1 n iD1 n iD1
and for this it suffices to observe that the expression at the left hand side equals
X
n
.ai aj /.bi bj /; (5.12)
i;jD1
Moreover, if k > 1, then equality holds if and only if all of the ai ’s are equal.
Proof Let’s prove the inequality by induction on k 1, noting that (5.13) is trivially
true for k D 1 and all positive reals a1 , . . . , an . By induction hypothesis, let l > 1 be
a natural number such that (5.13) is true for k D l 1 and all positive reals a1 , a2 ,
. . . , an .
Since both sides of the inequality we wish to prove are invariant under permu-
tations of the indices 1, 2, . . . , n, we can suppose, without any loss of generality,
5.3 More on Inequalities 131
Finally, let l > 1 and suppose we have equality in (5.13) when k D l. Then,
the argument of the previous paragraph assures that we must have equality in (5.14)
and (5.15). However, by the condition for equality in Chebyshev’s inequality, the
only way to have equality in (5.14) is having a1 D D an . t
u
The two coming examples illustrate how one can apply Chebyshev’s inequality.
Example 5.21 (Poland) Let a1 ; a2 ; : : : ; an be positive reals with sum equal to s.
Prove that
a1 a2 an n
C CC :
s a1 s a2 s an n1
n ! n !
X
n
ai X 1 1 X
n X 1
D ai ai
iD1
s ai iD1
s ai n iD1 iD1
s ai
! (5.16)
s X 1
n
D :
n iD1 s ai
Pn
However, since iD1 .s ai / D .n 1/s, this last inequality yields
X
n
1 n2
: (5.17)
iD1
s ai .n 1/s
X
n
x5i
;
iD1
x1 C Cb
xi C C xn
where the hat over xi at the denominator of the i-th summand indicates that it
contains all of x1 ; x2 ; : : : ; xn , except xi .
Proof Let s denote the sum of the xi ’s and suppose, without loss of generality, that
1 1 1
x1 x2 xn . Then, x21 x22 x2n and sx 1
sx 2
sx n
.
The expression to be minimized can be written as
X
n
1
SD x5 :
iD1
s xi i
s 1 1
D D 2 :
n n.n 1/s n .n 1/
5.3 More on Inequalities 133
1
In order to guarantee that n2 .n1/ is the least possible value, we ought to show
that
Pn it is attained. To this end, it is enough to see that, in view of the constraint
2
iD1 xi D 1, the condition for equality in Chebyshev’ inequality gives x1 D x2 D
D xn D p1n if equality holds. However, computing with these values for x1 ,
1
. . . , xn , we see that all of the inequalities above become equalities, so that n2 .n1/
is,
indeed, the least possible value. t
u
Next, we present an inequality known as the rearrangement inequality. We
recall that a sequence .x1 ; x2 ; : : : ; xn / is a permutation or a rearrangement of
.a1 ; a2 ; : : : ; an / if these two sequences differ only by the order of their terms.7
Proposition 5.23 Let a1 < a2 < < an be given positive real numbers. If
.x1 ; x2 ; : : : ; xn / is any permutation of .a1 ; a2 ; : : : ; an /, then
X
n1 X
n1 X
n1
ai ani a i xi a2i ;
iD1 iD1 iD1
with equality in the left (resp. right) inequality above if and only if xi D ani (resp.
xi D ai ), for 1 i n.
Proof Let us show how to maximize the sum a1 x1 Ca2 x2 C Can xn . (The argument
to minimize it is completely analogous.)
Since the number of permutations8 .x1 ; x2 ; : : : ; xn / of .a1 ; a2 ; : : : ; an / is finite,
there is at least one of them which maximizes the sum a1 x1 C a2 x2 C C an xn .
If .b1 ; b2 ; : : : ; bn / is such a permutation, we want to show that bi D ai for
1 i n. To this end, it suffices to show that b1 < b2 < < bn . By contradiction,
suppose that there exist indices 1 i < j n for which bi > bj . Define a
permutation .b01 ; b02 ; : : : ; b0n / of the ai ’s by setting
8
< bk ; se k ¤ i; j
b0k D bi ; se k D j :
:
bj ; se k D i
Then,
X
n X
n
ai b0i ai bi D .ai b0i C aj b0j / .ai bi C aj bj /
iD1 iD1
D .ai bj C aj bi / .ai bi C aj bj /
D .ai aj /.bj bi / > 0:
7
In the more precise language of functions (cf. Chap. 6), we say that .x1 ; x2 ; : : : ; xn / is a
permutation of .a1 ; a2 ; : : : ; an / if there is a bijection ' W f1; : : : ; ng ! f1; : : : ; ng, so that
xi D a'.i/ , for 1 i n.
8
Actually, it is easy to show that there are exactly nŠ such permutations. For a proof, see [5], or
provide one yourself, by making an induction argument.
134 5 Elementary Inequalities
X
n1 X
n1 X
n1
ai ani a i xi a2i ;
iD1 iD1 iD1
with equality in the left (resp. right) inequality above if (and no more if, and only
if ) xi D ani (resp. xi D ai ), for 1 i n. The coming example explores this more
general form of the rearrangement inequality.
Example 5.24 Given positive reals a, b and c, show that:
(a) a3 C b3 C c3 a2 b C b2 c C c2 a.
1 1 1
abc a2 C b2 C c2 .
(b) aCbCc
Proof
(a) Suppose, without loss of generality, that a b c. (Other orderings of a,
b and c would give rise to analogous arguments.) A direct application of the
rearrangement inequality gives
a3 C b3 C c3 D a2 a C b2 b C c2 c a2 b C b2 c C c2 a:
(b) The symmetry of both sides with respect to a, b and c allows us to suppose again
that a b c. The inequality to be proved is equivalent to
In order to get this, let’s first observe that the condition 0 < a b c implies
ab ac bc. Therefore, upon applying the general form of the rearrangement
inequality, we obtain
a2 bc C ab2 c C abc2 D ab ac C ab bc C ac bc
.ab/2 C .bc/2 C .ca/2 :
t
u
5.3 More on Inequalities 135
when .x1 ; x2 ; : : : ; xn / varies over all sequences of nonnegative reals such that x1 C
x2 C C xn D 1.
Solution Let
X
En .x1 ; : : : ; xn / D xi xj .xi C xj /
1i<jn
and observe that the symmetry of the right hand side allows us to suppose that
x1 xn 0.
Let’s first of all show that
To this end, set yj D xj for 1 j < n 1, yn1 D xn1 C xn and yn D 0, and note
that y1 C y2 C C yn D 1. For the sake of notation, let En denote the difference
Then, we get
X X
n2
En D yi yj .yi C yj / C yi yn1 .yi C yn1 /
1i<jn2 iD1
X
n1 X
C yi yn .yi C yn / xi xj .xi C xj /
iD1 1i<jn2
X X
xi xn1 .xi C xn1 / xi xn .xi C xn /:
1i<jn2 1i<jn1
136 5 Elementary Inequalities
X
n2
En D xi .xn1 C xn /.xi C xn1 C xn /
iD1
X
n2 X
n1
xi xn1 .xi C xn1 / xi xn .xi C xn /
iD1 iD1
X
n2
D Œxi .xn1 C xn /.xi C xn1 C xn / xi xn1 .xi C xn1 /
iD1
E2 .x1 ; x2 / D x1 x2 .x1 C x2 /;
Finally, we have
1 1 1
En .x1 ; : : : ; xn / E2 ; D :
2 2 4
t
u
5.3 More on Inequalities 137
The last inequality we wish to consider is due to the XIX century Norwegian
mathematician N. H. Abel. For this reason, it is known as Abel’s inequality.
Theorem 5.26 (Abel) Let n > 1 be a natural number and a1 , a2 , . . . , an , b1 , b2 ,
. . . , bn be given real numbers, with a1 a2 an 0. If M and m respectively
denote the maximum and minimum elements of the set of sums fb1 ; b1 Cb2 ; : : : ; b1 C
b2 C C bn g, then
ma1 a1 b1 C a2 b2 C C an bn Ma1 :
Proof Let’s prove the right hand side inequality, the proof of the left hand side one
being totally analogous.
Make s0 D 0 and si D b1 C C bi for 1 i n. Then,
X
n X
n X
n X
n1
ai bi D ai .si si1 / D ai si aiC1 si
iD1 iD1 iD1 iD0
X
n1
D .ai aiC1 /si C an sn
iD1
X
n1
M.ai aiC1 / C Man D Ma1 :
iD1 t
u
For future reference, and in the notations of the proof of Abel’s theorem, the
identity
X
n X
n1
ai bi D .ai aiC1 /si C an sn (5.19)
iD1 iD1
is usually referred to as Abel’s identity. As we shall see here and later (see
Problem 18, page 243), it is almost as useful as the inequality itself.
We finish this section by presenting a beautiful application of Abel’s inequality.
In the course of the proof, we will use the fact that every set with k elements
has exactly 2k distinct subsets. For a proof of this last fact, we refer the reader to
Problem 12, page 96, or to Chap. 1 of [5].
Example 5.27 (Romania) Do the following items:
(a) Let n > 1 be an integer and x1 ; : : : ; xn , y1 ; : : : ; yn positive reals such that x1 y1 <
x2 y2 < < xn yn and, for 1 k n, x1 C C xk y1 C C yk . Prove that
1 1 1 1 1 1
C CC C CC :
x1 x2 xn y1 y2 yn
138 5 Elementary Inequalities
Proof
(a) First of all, observe that
Xn
1 X1
n Xn
xi y i
D : (5.20)
y
iD1 i
x
iD1 i iD1
xi yi
.x1 y1 / C C .xk yk / 0
Xn
1 X1
n
an minfb1 C C bi I 1 i ng 0:
y
iD1 i
x
iD1 i
(b) Suppose, without loss of generality, that a1 < a2 < < an , and let Bk D
fa1 ; : : : ; ak g for 1 k n. The hypothesis on the set A assures that all of the
2k 1 nonempty subsets of Bk have distinct sums of elements. However, since
each of these sums is a natural number and a1 C C ak is the greatest of them,
we conclude that
a1 C C ak 2k 1:
a1 C a2 C C ak 20 C 21 C C 2k1
1 1 1 1 1 1
C CC 0 C 1 C C n1 < 2:
a1 a2 an 2 2 2
t
u
For another proof of item (b) of the previous example, see Chap. 3 of [5].
a m C a n mm C n n :
1 1 1 a 8 C b 8 C c8
C C :
a b c a 3 b 3 c3
4. (OIM) Find all real positive solutions of the system of equations
x1 C x2 C C x1994 D 1994
:
x41 C x42 C C x41994 D x31 C x32 C C x31994
and give necessary and sufficient conditions for the equality. To which particular
case there corresponds Chebyshev’s inequality?
140 5 Elementary Inequalities
a3 b3 c3 d3 1
C C C :
bCcCd aCcCd aCbCd aCbCc 3
9. (IMO shortlist) Let x, y and z be positive reals such that xyz D 1. Prove that
x3 y3 z3 3
C C :
.1 C y/.1 C z/ .1 C x/.1 C y/ .1 C x/.1 C y/ 4
10. (India) Let n > 1 be an integer and x1 , x2 , . . . , xn be given positive reals whose
sum is equal to 1. Prove that
r
X
n
xi n 1 Xp n
p p xi :
iD1
1 xi n1 n 1 iD1
11. (Slovenia) Given 2n positive reals a1 , a2 , . . . , a2n , how should we arrange them
in pairs, such that the sum of the n products of the numbers of each pair is
maximal?
12. (IMO) Let .ak /k1 be a sequence of pairwise distinct positive integers. Prove
that, for every n 2 N, we have
X
n
ak X
n
1
:
kD1
k2 kD1
k
4
x21 x2 C x22 x3 C x23 x4 C C x2n x1 :
27
5.3 More on Inequalities 141
b1 C b2 C C bn a1 C a2 C C an :
Chapter 6
The Concept of Function
With the algebraic background of the previous chapters at our disposal, we now
begin the study of real functions of a single variable. After presenting the basic
concepts of domain, codomain and image, some relevant examples of functions
are given. Then, we introduce the notions of monotonicity and extremal values,
discussing various examples that, in spite of their elementary character, will reveal
themselves to be very useful. The chapter continues with the study of the operations
of composition and inversion of functions, and culminates with the orchestration of
the whole of it to a first study of implicitly defined functions. We end the chapter by
discussing, in its last two sections, graphs of elementary functions.
f
1 a
b
2 c
3 d
1 a
b
2 c
3 d
1 a
b
2 c
3 d
Note, however, that the diagrams in Fig. 6.2 and 6.3 do not correspond to
functions. The situation of Fig. 6.2 is forbidden because there is no arrow departing
from 1 2 X. The situation of Fig. 6.3 is forbidden because more than one arrows
departs from 1 2 X.
The next three definitions isolate some quite useful types of functions.
Definition 6.1 Let X and Y be given nonempty sets. For a fixed element c 2 Y, the
constant function c from X to Y is the function f W X ! Y such that f .x/ D c for
every x 2 X.
6.1 Definitions and Examples 145
Thus, in the extreme case of the constant function equal to c, defined as above,
every x 2 X is associated to a single y 2 Y, namely, y D c. Nevertheless, the
conditions required by the “definition” of function are fully satisfied, i.e., every x 2
X is associated to a unique y 2 Y.
Definition 6.2 Let X be a given nonempty set. The identity function of X is the
function IdX W X ! X, such that IdX .x/ D x, for every x 2 X.
As in the previous definition, it is immediate to see that the conditions required
by the “definition” of function are satisfied, so that IdX is indeed a function from X
to itself.
For the next definition, given an arbitrary n 2 N, we let In be the set whose
elements are the first n natural numbers, i.e., I1 D f1g, I2 D f1; 2g, I3 D f1; 2; 3g
and, more generally,
In D fk 2 NI 1 k ng:
The cartesian product of X and Y (in this order), denoted X Y, is the set
Definition 6.4 Given nonempty sets X and Y, a relation from X to Y (or between
X and Y, in this order) is a subset R of the cartesian product X Y, i.e., R is a set of
ordered pairs .x; y/, with x 2 X and y 2 Y. If R is a relation from X to X, we simply
say that R is a relation on X.
1
For a more thorough account of set theory, we refer the reader to the classic [15].
146 6 The Concept of Function
R D f.x; y/ 2 X YI x yg
is a relation from X to Y, given by R D f.2; 2/; .3; 2/; .3; 3/g. Indeed, these are the
only ordered pairs .x; y/, with x 2 f1; 2; 3g, y 2 f2; 3; 4; 5g and such that x y.
The example above illustrates an obvious procedure we can use to define a
specific relation R between given nonempty sets X and Y (in this order): it suffices
to declare, somehow, a subset of the cartesian product X Y; those ordered pairs of
X Y that satisfy the given prescription will be precisely the elements of R.
If R is a relation from X to Y, then R X Y, by definition. Conversely, once
we choose an ordered pair .x; y/ 2 X Y, then either .x; y/ 2 R or .x; y/ … R. In
the former case, we say that x and y are related by R, and write x R y; in the latter
case, we say that x and y are not related by R, and write x R y. Thus, in symbols,
x R y , .x; y/ 2 R: (6.1)
8 x 2 X; 9 a unique y 2 YI x R y:
f W X ! Y
:
x 7! f .x/
This way, the function of the previous paragraph, that associates to each x 2 R its
square x2 , could have been denoted in the following manner:
f W R ! R
:
x 7! x2
We surely have defined a function, for the expressions that define f .x/ have sense
in R and, although we must apply different formulas, according to the rational x
satisfies x 0 or x > 0,peach rational x p
has a single, well defined image f .x/. Thus,
for example, f .1/ D .1/2 C 1 D 2 (since 1 0), but f .2/ D 2 C 1 D 3
(since 2 > 0).
Observe that we could have defined f .x/ by writing
p
x2 C 1; if x 0
f .x/ D :
x C 1; if x 0
If that were the case, then, albeit conditions x 0 and x 0 cover all rationals, they
wouldn’t be mutually exclusive: x D 0 would satisfy both. However, the formulas
p be applied in one case or the other would give the same result for x D 0
that should
(since 02 C 1 D 0 C 1), thus avoiding any possibility of inconsistence.
We sometimes say that a function f like that of the previous example is defined
by parts. This expression alludes to the fact that there is a formula to compute
f .x/ when x 2 .1; 0, and another one to compute it when x 2 .0; C1/ (or
x 2 Œ0; C1/, as one may wish).
When dealing with a function f W X ! Y, it’s frequently useful to refer to the sets
X and Y as the domain and codomain of the function, respectively; in this context,
we will write X D Dom .f /. For instance, for the function f of Example 6.7, the
domain and codomain are, respectively, Q and R.
We will often work with functions f W X ! R, with X R. In cases like
these, we shall say that f is a real function (alluding to the fact that it assumes real
148 6 The Concept of Function
values – i.e., that its codomain is R) of a real variable (now alluding to the fact
that a generic element x of the domain X of f – the variable of the function – is a
real number).
In cases like those of the previous paragraph, if f .x/ is defined by a formula on
x, then, unless explicitly stated otherwise, we will stick to the usage of taking X as
the largest possible domain. In other words, in such cases we will take X to be the
largest possible subset of R in which the mathematical operations that define the
expression f .x/ have sense. We shall, then, say that X is the maximal domain of
definition, or simply the maximal domain of f .
Example 6.8 Find the maximal domain of definition X R of the function f W X !
1
R, given by f .x/ D px.x1/ .
1
Solution For px.x1/ to be a real number, it’s necessary and sufficient that x.x 1/
be positive. Thus,
n 1 o
X D x 2 RI p 2R
x.x 1/
D fx 2 RI x.x 1/ > 0g:
A product of two real numbers is positive only if both factors have the same sign.
Therefore, we must have x > 0 and x 1 > 0, or else x < 0 and x 1 < 0. Hence,
we must have x > 1 or x < 0, so that
for every x 2 X.
Some remarks are in order. First of all, notice that the addition signs in the first
equality above have distinct meanings: in the left hand side, the C sign is used in
the very definition of the function f C g, while in the right hand side f .x/ C g.x/
stands for the usual addition of the real numbers f .x/ and g.x/. Analogous remarks
are valid for the multiplication signs used in the definitions of the functions f g and
c f . Secondly, as is usual with real numbers, we shall generally omit multiplication
signs, writing fg and cf instead of f g and c f . This usage will not be source of any
kind of confusion.
6.1 Definitions and Examples 149
.f C c/.x/ D f .x/ C c;
for every x 2 X.
Example 6.9 Let f and g be functions from R to R given by f .x/ D x
x2 C1
and
g.x/ D x C 3, for every x 2 R. Then,
x
.f C g/.x/ D f .x/ C g.x/ D C .x C 3/
x2 C1
x C .x2 C 1/.x C 3/ x3 C 3x2 C 3
D 2
D ;
x C1 x2 C 1
x x2 C 3x
.fg/.x/ D f .x/g.x/ D .x C 3/ D
x2 C 1 x2 C 1
and
p
p p p x x 3
. 3f /.x/ D 3f .x/ D 3 D 2 :
x2 C 1 x C1
We leave to the reader the task of verifying that the operations of addition and
multiplication of functions, defined as above, satisfy properties analogous to those
satisfied by the corresponding operations with real numbers. More precisely, for
functions f ; g; h W X ! R and real numbers a, b, we have:
• Commutativity: f C g D g C f ; fg D gf .
• Associativity: f C .g C h/ D .f C g/ C h; f .gh/ D .fg/h; a.bf / D .ab/f .
• Distributivity: f .g C h/ D fg C fh; .a C b/f D af C bf .
Finally, note that the associativity property of the operations of addition and
multiplication of two functions allows us to define, in entirely analogous ways, the
sum and the product of an arbitrary finite number of functions from X to R. For
example, given functions f ; g; h W X ! R, we define f C g C h W X ! R as either
.f C g/ C h or f C .g C h/, since these are equal functions; analogously, fgh W X ! R
can be defined as either .fg/h or f .gh/.
For an additional extension of the discussion of operations on functions, we refer
the reader to Problem 8.
Turning our attention to the codomain of a general function f W X ! Y, it’s
important for the reader to realize that Y doesn’t generally coincide with the set of
the images of the elements of X (which, anyhow, is a subset of Y). Let’s illustrate this
150 6 The Concept of Function
point using again the function f of Example 6.7. We have already observed that its
codomain is the set R of real numbers. On the other hand, it’s certain that its subset
ff .x/I x 2 Qg doesn’t contain real numbers less than 1. Indeed, since x2 C 1 1 for
any real x, we have
p p
f .x/ D x2 C 1 1 D 1;
whenever x < 0 is rational; on the other hand, for a rational x > 0, we have f .x/ D
x C 1 > 1. In any case,
2
For further details on the gcd of two nonzero integers, read again the introduction to Chapter 1 or,
alternatively, read Section 6:2 of [5].
6.1 Definitions and Examples 151
n 1 1 1 o
Im.f / D 1; ; ; ; : : : :
2 3 4 t
u
Unfortunately, there exists no algorithm that allows us to explicitly describe the
image of an arbitrarily given function. Nevertheless, in subsequent chapters we will
solve this problem for several important classes of functions.
We finish this section discussing the important concept of equality of functions.
p
With respect
p to the function f of Example 6.7, it makes no sense to consider f . 2/,
since 2 … Q and the domain of f is Q. What we could do would be to consider,
instead of f , the function g W R ! R, given by
p 2
x C 1; if x 0
g.x/ D :
x C 1; if x > 0
Although the formulas that define f .x/ and g.x/ are the same, for f they can be
applied only to x 2 Q, while for g they can be applied to every real x. Therefore, it
makes no sense to think about f and g as being equal functions, just denoted in two
different ways.
In the positive direction, we have the following
Definition 6.11 Functions f W X ! Y and g W W ! Z are equal if X D W, Y D Z
and f .x/ D g.x/, for every x 2 X.
If f W X ! Y and g W W ! Z are equal functions, we write f D g. We also stress
that, according to the above definition, the equality of f and g doesn’t merely reduce
to f .x/ D g.x/; it also means the equality X D W of the domains of f and g, as well
as the equality Y D Z of the codomains of these two functions. If functions f and
g as above are not equal, we will write f ¤ g and say that f and g are different or
distinct functions.
b c W R ! R
; (6.2)
x 7! bxc
f g W R ! R
; (6.3)
x 7! fxg
We start this section concentrating ourselves in the problem of finding the image of
a given function. To this end, recall that the image of a function f W X ! Y is the set
Im.f / D ff .x/ 2 YI x 2 Xg
D fy 2 YI y D f .x/ for some x 2 Xg:
This second way of declaring Im.f / is particularly useful in case f is a real function
of a real variable, i.e., if f W X ! R, with X R. Indeed, for such an f , if the values
f .x/ are given by a formula on x 2 X, we can look at the problem of finding the
image of f as that of finding the y 2 R for which the equation f .x/ D y has at least
one solution x 2 X. Let’s see some examples.
Example 6.12 An affine function is a function f W R ! R such that f .x/ D ax C b,
for every real x, where a and b are given real numbers, with a ¤ 0. A linear function
is an affine function f as above, such that b D 0.
According to the previous paragraph, the image of an affine function f as above
can be found by searching for the y 2 R for which the equation ax C b D y has
at least one solution x 2 R. Since a ¤ 0, this equation always admits the solution
x D yba . Therefore, we conclude that every y 2 R belongs to the image of f and,
hence, Im .f / D R.
Example 6.13 The function of inverse proportionality is the function f W R n
f0g ! R n f0g given by f .x/ D 1x , for every x 2 R n f0g.
In order to find its image, it suffices to find all y 2 R for which there exists a
real number x ¤ 0 (i.e., x belonging to the domain of f ), such that f .x/ D y, i.e.,
such that 1x D y. If y D 0, it is clear that this equation doesn’t admit solutions; on
the other hand, if y ¤ 0, this same equation admits the solution x D 1y ¤ 0. Hence,
Im .f / D R n f0g.
It is time we define one of the most important classes of elementary real functions
of a real variable.
Definition 6.14 A quadratic, or second degree function is a function f W R ! R
such that f .x/ D ax2 C bx C c for every real x, where a, b and c are given real
numbers, with a ¤ 0. The discriminant of f is the discriminant of the associated
second degree trinomial ax2 C bx C c, so that D b2 4ac.
The problem of finding the image of a quadratic function is sufficiently important
to be collected in the following
Proposition 6.15 With respect to the quadratic function f .x/ D ax2 C bx C c, we
have that:
(a) If a > 0, then Im .f / D 4a ; C1 .
(b) If a < 0, then Im .f / D 1; 4a .
154 6 The Concept of Function
b
f .x/ D ,xD :
4a 2a
Proof Following the general procedure described at the beginning of this section,
it suffices to find all y 2 R for which the equation ax2 C bx C c D y, i.e., the
second degree equation ax2 C bx C .c y/ D 0, has at least one real solution.
As we already know from Section 2.3, a necessary and sufficient condition for the
existence of such (a) root(s) is that this last equation has nonnegative discriminant,
i.e., that b2 4a.c y/ 0. Since we agreed to let b2 4ac D , the y’s we’re
looking for are precisely the solutions of the first degree inequality
C 4ay 0:
Now, we separately consider the cases a > 0 and a < 0. If a > 0, then
4ay C 0 , y ;
4a
and it follows that
Im .f / D y 2 RI y D ; C1 I
4a 4a
if a < 0, then
4ay C 0 , y
4a
and, hence,
Im .f / D y 2 RI y D 1; :
4a 4a
To what was left to do, notice that, for y 2 Im .f /, the solutions of the equation
ax2 C bx C c D y (, ax2 C bx C .c y/ D 0) are
p p
b ˙ b2 4a.c y/ b ˙ C 4ay
xD D : (6.4)
2a 2a
To what comes next, we make the convention of saying that the quadratic
function f .x/ D ax2 C bx C c has constant sign if f .x/ 0 for every x 2 R, or
f .x/ 0 for every x 2 R.
Corollary 6.16 The quadratic function f .x/ D ax2 CbxCc has constant sign if and
only if 0. In this case, we have af .x/ 0 for every x 2 R. In other words:
(a) If 0 and a > 0, then f .x/ 0 for every x 2 R.
(b) If 0 and a < 0, then f .x/ 0 for every x 2 R.
Proof Let’s look at the case a > 0, the case a < 0 being totally analogous. If 0,
it follows from the previous proposition that
f .x/ 0; 8 x 2 R:
4a
Conversely, suppose that a > 0 and that f has constant sign. Again from the
previous proposition, the image of f contains positive numbers, so that we must
have f .x/ 0, for every x 2 R. In particular,
b
Df 0:
4a 2a
Therefore, 0. t
u
Remark 6.17 A simple modification of the argument presented in the proof of the
previous corollary allows us to conclude that
i. If < 0 and a > 0, then f .x/ > 0 for all x 2 R.
ii. If < 0 and a < 0, then f .x/ < 0 for all x 2 R.
From now on, we shall use these results whenever needed, without further com-
ments.
The previous corollary can also be used to give a much simpler proof of Cauchy’s
inequality (cf. Theorem 5.13).
Example 6.18 Let n > 1 be an integer and a1 , a2 , . . . , an , b1 , b2 , . . . , bn be real
numbers, such that at least one of the ai ’s and at least one of the bi ’s is nonzero.
Consider the quadratic function
where A D a21 C a22 C C a2n > 0 (for, at least one of the ai ’s is nonzero), B D
a1 b1 C a2 b2 C C an bn and C D b21 C b22 C C b2n .
Since f .x/ is a sum of squares, we have f .x/ 0 for every x 2 R. On the other
hand, since A > 0, Corollary 6.16 gives D 4.B2 AC/ 0. Hence, B2 AC,
156 6 The Concept of Function
p p
or, which is the same, jBj A C. Substituting the values of A, B and C, we get
Cauchy’s inequality.
According to the above reasoning, equality in Cauchy’s inequality is equivalent,
for the function f , to D 0. In turn, this is equivalent to the existence of a single
˛ 2 R such that f .˛/ D 0. However, since f .˛/ is a sum of squares, the only way
we can have f .˛/ D 0 is if each one of these squares vanishes, i.e., if
a1 ˛ b1 D a2 ˛ b2 D D an ˛ bn D 0:
Therefore, f is increasing.
3
In the notations of this definition, it is worth observing that, for some authors, a function f
satisfying the condition of item (a) (resp. of item (b), (c) or (d)) is said to be strictly increasing
(resp. strictly decreasing, increasing or decreasing).
6.2 Monotonicity, Extrema and Image 157
2
Example 6.21 The function f W Œ0; C1/ ! R, given by f .x/ D xC2 x
, is increasing
in the whole interval Œ0; C1/. To check this, take real numbers 0 a < b. Then,
b2 a2
f .b/ f .a/ D
bC2 aC2
1
D Œb2 .a C 2/ a2 .b C 2/;
.a C 2/.b C 2/
b2 .a C 2/ a2 .b C 2/ D b2 a a2 b C 2.b2 a2 /
D ab.b a/ C 2.b a/.b C a/
D .b a/Œab C 2.b C a/:
Now, since 0 a < b, both factors in the last product above are positive, so that
b2 .a C 2/ a2 .b C 2/ > 0.
Example 6.22 Function f W R ! R, given by f .x/ D x3 C 2x, for every x 2 R, is
increasing. Indeed, for any real numbers a < b, we have
f .b/ f .a/ D b3 a3 C 2b 2a
D .b a/.b2 C ba C a2 / C 2.b a/
D .b a/.b2 C ab C a2 C 2/:
where, in the next to last inequality, we used the fact that j˛j C ˛ 0, for every
˛ 2 R.
The coming proposition solves, for quadratic functions, the problem of finding
the monotonicity intervals.
Proposition 6.23 Let a, b, c 2 R, with a ¤ 0, and f .x/ D ax2 C bx C c for every
x 2 R.
b
(a) If a > 0, then f is decreasing in 1; 2a
b
and increasing in 2a ; C1 .
b
(b) If a < 0, then f is increasing in 1; 2a
b
and decreasing in 2a ; C1 .
158 6 The Concept of Function
Proof Let’s do the proof of item (a), the proof of item (b) being totally analogous.
For x2 > x1 2a
b
, we have
since x2 > x1 2a
b
gives x2 x1 > 0, as well as x2 C x1 C b
a > 0. t
u
The next definition is, in a certain sense, complementary to Definition 6.19.
Definition 6.24 Let I R be an interval and f W I ! R be a given function. We
say that y0 2 R is the minimum value of f in I if the two following conditions are
satisfied:
(a) Im .f / Œy0 ; C1/.
(b) y0 2 Im .f /.
In this case, the real numbers x0 2 I such that f .x0 / D y0 are called the minimum
points of the function f .
Similarly, we define what one means by the maximum value and the maximum
points of a function f W I ! R (I R being an interval). The maximum and
minimum points of a given function (provided they exist) are collectively called its
extreme points; accordingly, the values the function takes at those points are its
extreme values.
In Section 9.6, we shall see how to search extreme points for differentiable
functions, i.e., functions possessing derivatives. For the time being, we shall content
ourselves in analysing some elementary examples, the first of which being an
immediate consequence of Proposition 6.15.
Proposition 6.25 With respect to the quadratic function f .x/ D ax2 C bx C c, if
a > 0 (resp. a < 0), then 2a
b
is the only minimum (resp. maximum) point of f .
Moreover, the minimum (maximum) value of f is 4a .
The proposition above has several interesting applications, two of which are
collected below, for the sake of illustrating its use. For the necessary geometric
background, we refer the reader to [4].
Example 6.26 There is given a semicircle of 1cm of radius. A rectangle is so
situated that one of its sides lies on the diameter of the semicircle, whereas its other
two vertices lie on the semicircle itself. Compute the largest possible value for the
area of the rectangle.
Solution Let Fig. 6.4 account for the described situation, so that AB is the diameter
of the semicircle, O is its center and PQRS is the given rectangle, with PQ AB.
Setting OQ D x and QR D y, the area of PQRS equals 2xy. On the other hand,
applying Pitagoras’ theorem to triangle OQR, we get x2 C y2 D 1 and, hence,
6.2 Monotonicity, Extrema and Image 159
1 y
A P O x Q B
p p p
2xy D 2x 1 x2 D 2 x2 .1 x2 / D 2 x2 x4 :
Now, making the substitution z D x2 , it follows from the last expression above for
the area that it suffices to maximize the quadratic function f .z/ D z z2 , subjected
to the condition 0 < z < 1 (for, x < OR D 1). By Proposition 6.25, such a
function (without any further restrictions) admits z D 12 as its only maximum point.
Moreover, since 12 2 .0; 1/, it follows that the desired maximum value is f 12 D 14 .
q
Therefore, the maximum value for the area is 2 14 D 1. t
u
Example 6.27 Given a triangle ABC in the plane, show that its baricenter is the
2 2 2
only point P in the plane of ABC for which the sum AP C BP C CP attains its
minimum possible value.
Proof Fix a cartesian system of coordinates in the plane, with respect to which
A.x1 ; y1 /, B.x2 ; y2 / and C.x3 ; y3 /. If P.x; y/, then the formula for the distance
between two points in the plane furnishes
2 2 2
AP C BP C CP D f .x/ C g.y/;
where
3
X
f .x/ D .x xi /2 D 3x2 2.x1 C x2 C x3 /x C .x21 C x22 C x23 /
iD1
C1 2
Example 6.28 Let f W Œ0; C1/ ! R be the function given by f .x/ D xxC1 , for
every x 2 R. What is the minimum value of f ? Does f attain a maximum value?
Solution First of all, note that
x2 C 1 x2 1 C 2
f .x/ D D
xC1 xC1
2 2
Dx1C D .x C 1/ C 2:
xC1 xC1
2
Therefore, applying (5.2) (with a D x C 1 and b D xC1
), we get
r
2 2 p
.x C 1/ C 2 .x C 1/ D 2 2;
xC1 xC1
2
with equality if and only if x C 1 D xC1 , i.e., if and only if x2 C 2x 1 D 0. Since
x p 0, we conclude that equality takes place in the above inequality if and only if
x D 2 1. Thus, for x 0 we have
2 p
f .x/ D .x C 1/ C 2 2 2 2;
xC1
p
p 2 2 2 is the minimum possible value for f , and it is attained only at
so that
x D 2 1.
2
To what is left, just notice that, for n 2 N, we have f .n/ D n 1 C nC1 n 1.
Therefore, f does not attain a maximum value. t
u
Example 6.29 Find the maximum value p
and the maximum point(s) of the function
x
f W Œ0; C1/ ! R, given by f .x/ D x2 C16 .
Solution It follows from the inequality between the arithmetic and geometric
means that
16 16 16
x2 C 16 D x2 C C C
3 3 3
r
4 16 16 16 32 p
4 x2 D p x;
3 3 3 4
27
so that
p p
4
x 27
f .x/ D 2 :
x C 16 32
6.2 Monotonicity, Extrema and Image 161
16 p4
Equality holds if and only if x2 D 3 , which is the same as x D 3
(since x 0).
p
4
27 p4 .
Thus, 32 is the maximum value of f and it is attained only at x D 3
t
u
1. Find the image of the function f W R ! R, given by f .x/ D x2 1C1 , for every
x 2 R.
2. * Find the image of the function f W R ! R, given by f .x/ D x C 1x , for every
x 2 R .
3. Let I R be an interval, let a 2 I and f W I ! R be a given function. If f is
increasing (resp. decreasing) in .1; a \ I and decreasing (resp. increasing)
in Œa; C1/ \ I, prove that a is the only maximum (resp. minimum) point of f
in I.
4. * Let X R be a nonempty set, f W X ! R be a given function and c 2 R.
Relate the images of the functions f and f C c. More precisely, if Y D Im .f /,
prove that Im .f C c/ D Y C c, where Y C c denotes the set
Y C c D fy C cI y 2 Yg:
cY D fcyI y 2 Yg:
is the canonical form of the quadratic function f .x/ D ax2 C bx C c. Use this
canonical form to give another proof of Proposition 6.15.
7. Let f .x/ D ax2 C bx C c be a quadratic function for which > 0, and x1 < x2
be the roots of f .x/ D 0. Prove the following items:
(a) If a > 0, then f .x/ < 0 , x 2 .x1 ; x2 /.
(b) If a < 0, then f .x/ < 0 , x … Œx1 ; x2 .
8. Let f .x/ D ax2 C bx C c be a quadratic function. If there exists a real number
x0 for which af .x0 / < 0, prove that > 0 and x0 2 .x1 ; x2 /, where x1 < x2 are
the roots of the equation f .x/ D 0.
162 6 The Concept of Function
9. Among all rectangles of a given perimeter, prove that the one of largest possible
area is the square.
10. The cross section of a tunnel has the shape of a semicircle of radius 5m. The
tunnel has two lanes of traffic, which go in opposite directions and are separated
one from the other by a narrow median. The trucks of a transportation firm are
to cross the tunnel to take goods from one city to another. If the trucks are 18m
long, what should be their widths and heigths, so that they can carry the largest
possible load volume per travel?
11. Compute the maximum value of the function f W R ! R given by f .x/ D x5x1 2 C1 ,
for every x 2 R.
12. Let ˛1 < ˛2 < < ˛n be given reals, and f W R ! R be given by
f .x/ D jx ˛1 j C jx ˛2 j C C jx ˛n j;
for every x 2 R. Prove that f attains a minimum value and compute it in terms
of ˛1 , ˛2 , . . . , ˛n .
13. In each of the following items, use the inequality between the arithmetic and
geometric means to compute the maximum value of the given function:
(a) f W R ! R given by f .x/ D x
2x2 C3
, for every x 2 R.
2x2 C5xC2
(b) f W R ! R given by f .x/ D x2 C1
, for every x 2
R.
3
(c) f W Œ0; 1 ! R given by f .x/ D x.1 x /, for every x 2 Œ0; 1.
14. In each of the following items, use the inequality between the arithmetic and
geometric means to compute the minimum value of the given function:
(a) f W .0; C1/ ! R given by f .x/ D .xC10/.xC2/
xC1 , for every x 2 .0; C1/.
(b) f W .0; C1/ ! R given by f .x/ D x2 C ax , for every x 2 .0; C1/, where a
is a positive real constant.
2
(c) f W .0; C1/ ! R given by f .x/ D x3xCa , for every x 2 .0; C1/.
(d) f W .0; C1/ ! R given by f .x/ D 6x C 24 x2
, for every x 2 .0; C1/.
15. (Romania) Let x; y 2 R be such that x2 xy C y2 2. Show that x4 C y4 8
and explain when the equality holds.
16. (TT) Find all reals x, y, z and t such that
8
ˆ
ˆ y D x3 C 2x
<
z D y3 C 2y
:
ˆ t D z3 C 2z
:̂
x D t3 C 2t
For the next problem, the reader might find it helpful to recall the discussion
on the gcd of two nonzero integers, at the beginning of Chapter 1. alternatively,
look at Chapter 6 of [5].
6.3 Composition of Functions 163
17. (OCM) Let f W N ! N be a function such that f .mn/ D f .m/ C f .n/ whenever
m and n are relatively prime. A natural number m is said to be a strangulation
point of f if n < m ) f .n/ < f .m/ and n > m ) f .n/ > f .m/. If f has
infinitely many strangulation points, show that it is an increasing function.
Roughly speaking, the above definition means that, in order to find the image of
x 2 X by g ı f , it suffices to find the image of f .x/ 2 Y by g. On the other hand,
it is easy to verify that g ı f , as defined above, is indeed a function. Also, observe
that to form the composite of f and g, it is necessary that the domain of g equals the
codomain of f . Let’s see some examples.
Example 6.31 Let X and Y be nonempty sets and f W X ! Y an arbitrary function.
If IdX W X ! X and IdY W Y ! Y are the identity functions on X and Y, respectively,
then
Let’s check the equality f ı IdX D f , the other one being totally analogous. To this
end, it suffices to note that f ı IdX is a function from X to Y such that, for evey x 2 X,
1 1 1
.g ı f /.x/ D g.f .x// D D 2 2 D 4
.f .x//2 C 1 .x / C 1 x C1
164 6 The Concept of Function
and
2
1 1
.f ı g/.x/ D f .g.x// D .g.x//2 D 2
D :
x C1 x4 C 2x2 C 1
x2 C 1 xC2
f .x/ D and .f ı g/.x/ D ;
3x2 3
xC2 g.x/2 C 1
D .f ı g/.x/ D f .g.x// D ;
3 3g.x/2
g.x/2 C1
so that 3g.x/2
D xC2
3
or, which is the same,
h ı .g ı f / D .h ı g/ ı f :
t
u
6.3 Composition of Functions 165
l ı .h ı .g ı f // D .l ı h/ ı .g ı f / D ..l ı h/ ı g/ ı f D ;
so that the order in which we insert parentheses does not alter the composition.
It is not difficult to see that the situation described in the last paragraph remains
true for the composition (whenever possible) of any finite number of functions. In
particular, given a nonempty set X, a function f W X ! X and a natural number
n, this last fact allows us to unambiguously define the nth composite function
f .n/ W X ! X, by setting
f .n/ D f ı f ı ı f :
„ ƒ‚ …
n
we have f .2/ D IdX , the identity function of X D R n f1; 1g. This gives us
In general, suppose (by induction hypothesis) that we have already proved that
f .2k1/ D f and f .2k/ D IdX , for some integer k 1. Then,
and
Therefore, we conclude that f .n/ D f whenever n is odd, and f .n/ D IdX whenever n
is even. u
t
Given a function f W X ! Y, we’ve already seen some examples that illustrate
the fact that the image of f is not necessarily equal to its codomain Y. On the other
166 6 The Concept of Function
hand, we can also have two distinct elements of X with the same image via f . For
an example, consider the quadratic function f .x/ D x2 , with x 2 R; for every x 2 R,
we have f .x/ D x2 D .x/2 D f .x/. We attach special names to functions whose
images coincide with their codomains, or which associate distinct images to distinct
elements of their domains. This is set in the coming
Definition 6.36 A function f W X ! Y is said to be:
(a) Injective, or one-to-one, or an injection, if, for every y 2 Y, there exists at
most one x 2 X such that f .x/ D y.
(b) Surjective, or onto, or a surjection, if, for every y 2 Y, there exists at least one
x 2 X such that y D f .x/. In other words, this is the same as saying that the
image of f is all of Y.
(c) Bijective, or one-to-one onto, or a bijection, if it is simultaneously injective
and surjective.
An efficient way of verifying whether a function f W X ! Y is injective or not is
to verify whether the implication
is true for all x1 ; x2 2 X. Accordingly, in order to prove that f is surjective, one must
be capable of, for each y 2 Y, guarantee the existence of at least one solution x 2 X
for the equation f .x/ D y. Let’s see some examples.
Example 6.37 If X R is a nonempty set and f W X ! X is a function such that
f .f .x// D x for every x 2 X, then f is a bijection.
Proof Let x1 and x2 be elements of X for which f .x1 / D f .x2 /. According to (6.6),
in order to show that f is injective it suffices to prove that x1 D x2 . To this end,
observe that f .x1 / D f .x2 / implies f .f .x1 // D f .f .x2 // and, then (by using the given
hypothesis), x1 D x2 .
The surjectivity of f also follows from the hypothesis: for a fixed y 2 X, taking
x D f .y/ 2 X we get f .x/ D f .f .y// D y, so that y 2 Im.f /. t
u
As a particular case of the previous example, the function of inverse proportion-
ality (cf. Example 6.13) is a bijection from R n f0g into itself.
Example 6.38 Let f W Œ0; 1 ! Œ0; 1 be a surjective function, such that jf .x1 /
f .x2 /j jx1 x2 j for all x1 ; x2 2 Œ0; 1. Prove that there are only two possibilities:
either f .x/ D x for every x 2 Œ0; 1, or f .x/ D 1 x for every x 2 Œ0; 1.
Proof Let a; b 2 Œ0; 1 be chosen in such a way that f .a/ D 0 and f .b/ D 1 (that it
is possible to make such a choice follows from the surjectivity of f ). Then, the given
hypothesis allows us to write
1 D j1 0j D jf .b/ f .a/j jb aj 1;
6.3 Composition of Functions 167
1 D jf .1/ f .0/j
jf .1/ f .c/j C jf .c/ f .0/j
j1 cj C jc 0j
D .1 c/ C c D 1:
Hence, we must have jf .c/ f .0/j D jc 0j and, since c; f .c/ 0, we get f .c/ D c.
Finally, this is true for every c 2 Œ0; 1, so that f .x/ D x for every x 2 Œ0; 1. t
u
The coming proposition teaches us the way injective, surjective and bijective
functions behave with respect to composition.
Proposition 6.39 Let f W X ! Y and g W Y ! Z be given functions. Then:
(a) g ı f injective ) f injective, but the converse is not necessarily true.
(b) g ı f surjective ) g surjective, but the converse is not necessarily true.
(c) g and f injective ) g ı f injective.
(d) g and f surjective ) g ı f surjective.
(e) g and f bijective ) g ı f bijective.
Proof
(a) For x1 ; x2 2 X, we have
where, in the last passage, we used the fact that g ı f is injective. We now have
to give an example in which f is injective but g ı f is not. To this end, take
X D Y D Z D R, f .x/ D x and g.x/ D x2 .
(b) Choosing z 2 Z arbitrarily, the surjectivity of g ı f guarantees the existence of
at least one x 2 X such that z D .g ı f /.x/. Then, z D g.f .x//, so that g is also
surjective. For the second part, take X D Y D Z D R, g.x/ D x and f .x/ D x2 ;
then, g is surjective, but g ı f is not.
(c) Let x1 ; x2 2 X. By using the injectivity of g and, then, that of f , we obtain
t
u
Let’s revisit Example 6.37 in light of the previous proposition.
Example 6.40 Let X be a nonempty set. If f W X ! X is a function such that
f ı f D IdX , then f is a bijection.
Proof Indeed, since the identity function IdX W X ! X is a bijection, it follows from
items (a) and (b) of the previous proposition that f is injective and surjective, hence,
bijective. t
u
The material of this section also allows us to study the important concept of
countably infinite sets, according to the following
Definition 6.41 An infinite set A is said to be countable if there exists an injective
function f W A ! N. In this case, we shall also say that A is a countably infinite set.
If A is countably infinite and B A is infinite, then B is also countably infinite.
Indeed, by composing an injective function f W A ! N with the inclusion W B ! A
(that sends each x 2 B to itself), we get the injective function f ı W B ! N. In
particular, every infinite subset of N is countable.
The set Z of integers is also countably infinite, for one can easily check that the
function f W Z ! N given by
2x; if x < 0
f .x/ D
2x C 1; if x 0
for every natural k > 1. In particular, b1 < b2 < b3 < and h is injective.
If h was not surjective, there would exist b 2 B such that b ¤ b1 ; b2 ; b3 ; : : :. If
bk < b for every k 2 N, then B f1; 2; : : : ; bg, a contradiction to the fact that B
is infinite. Hence, there would exist a natural m > 1 such that bm1 < b < bm .
However, since
h.m/ D min B n fh.1/; : : : ; h.m 1/g
D min B n fb1 ; : : : ; bm1 g ;
is countable, the same is true of the family P of infinite and nonconstant AP’s of
naturals. Hence, we can write P D fs1 ; s2 ; s3 ; : : :g.
Now, let Ak be the set whose elements are the terms of sk , and let A be defined in
the following way: take x1 2 A1 , x2 2 A2 , . . . such that x2 x1 2016, x3 x2
2017, x4 x3 2018, . . . (that this is possible follows from the fact that each Ak is
an infinite set). Then, let B D N n A.
It is clear that A does not contain the terms of an infinite AP of common
difference r 2 N, for the differences xj xj1 are eventually all greater than r.
On the other hand, B does not contain the terms of an infinite and nonconstant AP
either, for, if it did, we should have B
Ak for some k 2 N; however, xk 2 Ak n B. u t
f ı g D g ı f , .a 1/d D .c 1/b:
Gf D f.x; y/ 2 X YI y D f .x/g:
4
Graphs of real functions of a real variable will be one of the main objects of study along these
notes, starting from Section 6.6.
172 6 The Concept of Function
Among all functions f W X ! Y, the case of a bijection is, in a certain sense, the
best possible. Indeed, in this case the elements of X and Y are in a one-to-one onto
correspondence, so that to each element of X there corresponds a single element of
Y via f , and vice-versa. When this happens, we can form a function g W Y ! X by
asking that
f .x/ D y , g.y/ D x:
5
A family is a set whose elements are sets themselves.
6.4 Inversion of Functions 173
At this point, a natural question is this: why cannot we use the same declaration
above to define g when f is not bijective? From an intuitive point of view, if f is not
surjective, then there exists an element y of Y which is not the image of any element
of X via f ; therefore, there is no natural way of using f and y to define g.y/. On the
other hand, if f is not injective, then there exist distinct elements x1 ; x2 2 X with the
same image y 2 Y via f ; if we were to try to define g by using f , there would also
be no natural way of deciding which of x1 and x2 should be equal to g.y/.
Back to the case in which f is bijective, it is not difficult to see that g, defined
as above, is indeed a function, moreover such that .g ı f /.x/ D x, for every x 2 X,
and .f ı g/.y/ D y, for every y 2 Y. Yet in another way, we have g ı f D IdX
and f ı g D IdY . Conversely, if f W X ! Y and g W Y ! X are functions such
that g ı f D IdX and f ı g D IdY , then Proposition 6.39 guarantees that f must
be a bijection, and Problem 1 assures that g is the only function satisfying these
compositions.
We summarize the above discussion in the following
Definition 6.45 Let f W X ! Y be a given bijection. The inverse function of f is
the function g W Y ! X such that, for every y 2 Y,
g.y/ D x , y D f .x/:
From now on, whenever there is no danger of confusion, we shall denote the
inverse (function) of a bijection f W X ! Y by f 1 W Y ! X. Notice that the exponent
1 in the notation of inverse function has no arithmetic meaning; it simply calls the
reader’s attention to the fact that f 1 does the opposite of f , i.e., applies Y into X
instead of X into Y, and reverses the arrows of the correspondences made by f .
Now, it naturally emerges the question of how to effectively compute the inverse
of a given bijection. Although such a computation is generally more complicated
than that of composite functions, in the case of real functions of a real variable f W
X ! Y we can reason in the following simple way: since f 1 .y/ D x , f .x/ D y,
in order to find f 1 .y/ D x it suffices to solve, for x 2 X, the equation f .x/ D y.
If we find a single solution x 2 X for every y 2 Y, then f will be a bijection and
x (computed in terms of y) will give the sought for expression for f 1 .y/. Let’s
implement such a reasoning in some relevant examples.
Example 6.46 Let a and b be given reals, with a ¤ 0, and let f W R ! R be the
affine function given by f .x/ D ax C b, for every x 2 R. Show that f is a bijection
and find the expression of f 1 .
Proof Initially, note that
yb
f .x/ D y , ax C b D y , x D :
a
174 6 The Concept of Function
yb
f 1 .y/ D :
a t
u
Example 6.47 If X is a nonempty set, then, analogously to the former example, the
inverse function of the identity function IdX is this function itself, so that we can
write . IdX /1 D IdX . Nevertheless, the inverse of a bijective function can be the
function itself, even if it is not the identity function of some nonempty set; an
example is furnished by the inverse proportionality function f W R n f0g ! R n f0g
(which we already know to be bijective), for which f .x/ D 1x , for every x 2 R n f0g.
Indeed, since
1 1
f .x/ D y , Dy,xD ;
x y
we conclude that
1
f 1 .y/ D x D
y
and, hence, f 1 D f .
Example 6.48 Let a, b and c be given real numbers, with a > 0. Propositions 6.15
and 6.23 guarantee that the quadratic function f .x/ D ax2 CbxCc, seen as a function
b
f W ; C1 ! ; C1 ;
2a 4a
.g ı f /1 D f 1 ı g1 :
Proof We already know, from item (e) of Proposition 6.39, that g ı f is bijective. On
the other hand, since .g ı f /1 and f 1 ı g1 are both functions from Z to X, in order
to check that .g ı f /1 D f 1 ı g1 it suffices, by the uniqueness of the inverse (cf.
Problem 1), to show that
These verifications are quite straightforward and will be left for the reader as
exercises. u
t
6
Here and in the previous example, we are relying on previous knowledge from the reader, namely,
the existence of square roots of positive real numbers. Therefore, the present discussion will be
rigorously justified only after Theorem 7.9, where we prove the existence of such roots. In this
respect, the reader might also want to take a look at Example 8.36.
176 6 The Concept of Function
f .x C 1/ f .x 1/ D 4x:
6.5 Defining Functions Implicitly 177
Now, we can try to reverse the steps above, asking which functions f W R ! R are
such that
f .x C 1/ f .x 1/ D 4x; 8 x 2 R: (6.7)
Clearly, f .x/ D x2 is not the only one, for, fc .x/ D x2 Cc also satisfies (6.7), whatever
the real constant c.
Now, let f W R ! R be any real function satisfying (6.7), so that f is not explicitly
given by its values. Since we merely know that f must satisfy that relation, we say
that f is implicitly defined.
On the other hand, since we know that (6.7) must be satisfied, it allows us to find
other relations f must also satisfy. For instance, if g W R ! R is given by g.x/ D x2 ,
then
Thus, any function f W R ! R satisfying (6.7) will also satisfy (6.8). Nevertheless,
relation (6.8) may not be very useful to the task of finding all functions f W R ! R
that satisfy (6.7).
It is our purpose in this section to approach the problem of finding all functions
implicitly defined by a certain (finite) set of relations. Since there is no general
theory to accomplish this, we shall content ourselves in analysing a number of
interesting examples. In turn, these examples will provide us with several useful
techniques to be employed in such problems.
Let’s start by looking at two simple examples.
Example 6.51 (Canada) Find all increasing functions f W N ! N, such that f .2/ D
2 and f .mn/ D f .m/f .n/, for every m; n 2 N.
Solution From 1 f .1/ < f .2/ D 2 we get f .1/ D 1; also, f .4/ D f .2/f .2/ D 4
and f .8/ D f .4/f .2/ D 8. Suppose, by induction hypothesis, that f .2k / D 2k for a
certain natural number k. Then,
However, f .2n C 1/, f .2n C 2/, : : :, f .2nC1 1/ are natural numbers too, so that the
only possibility is
Finally, since such a reasoning is valid for whatever natural n, we conclude that
f .m/ D m, for every m 2 N. t
u
Example 6.52 (OIM) If D D R f1; 0; 1g, find all functions f W D ! R such
that, for every x 2 D, we have
1x
f .x/2 f D 64x:
1Cx
1x
Solution First of all, note that, since x ¤ 0, we have f .x/2 f 1Cx ¤ 0 for every
1x
x 2 D. In particular, f .x/ ¤ 0 for every x 2 D. Now, let g.x/ D 1Cx for x 2 D. The
definition of D easily gives that g.D/ D, so that we can compose f and g. This
way, for every x 2 D, we have
2 1 g.x/
f .g.x// f D 64g.x/: (6.9)
1 C g.x/
Substituting the defining formula for g in the above relation, we arrive at the equality
1x
!
1x 1 1Cx 1x
f f 1x
D 64
1Cx 1C 1Cx
1Cx
for every x 2 D. Squaring both sides of the relation given in the statement of the
problem and dividing the result by the last relation above, we find
1x
f .x/3 D 64x2 ;
1Cx
q
1x
so that f .x/ D 4 3 x2 1Cx .
Up to this point, we have only shown that, if there exists a function f W D ! R
satisfying the stated relation, then it must be given by the last expression above.
Hence, it is necessary that we verify that such an f does satisfy that relation, for
every x 2 D. Since such a verification is straightforward, we leave it to the reader.
t
u
6.5 Defining Functions Implicitly 179
Yet in respect to the previous example, a little more practice would have allowed
us to get rid of formally defining g to, then, compose it with f in order to get (6.9).
Instead, we could have just said
1x
Substituting x by in the given relation; we obtain : : : ;
1Cx
Solution Since this relation is valid for every x; y 2 R, it must be valid if we make
x D aCb2 and y D 2 , with a; b 2 R. Substituting these values for x and y, we get
ab
the relation
f .a/ f .b/
D a2 b2 ;
a b
a contradiction to x1 < x2 .
Example 6.55 (Argentina) Let f W R ! R be a decreasing function such that
f .x C f .x// D x C f .x/, for every real number x. Prove that f .f .x// D x, for every
real x.
Proof The hypotheses on f guarantee that x C f .x/ is a fixed point of f , for every
x 2 R. On the other hand, according to the previous discussion, the decreasing
character of f assures the existence of at most one fixed point for it. Therefore, there
must exist a 2 R such that x C f .x/ D a, for every x 2 R, so that f .x/ D a x for
every x 2 R. Hence,
f .f .x// D f .a x/ D a .a x/ D x;
for every x 2 R. t
u
The next example develops a body of ideas which will reveal themselves to be
useful in a number of other situations involving implicitly defined functions. In
particular, the first part of the presented reasoning solves Problem 12, page 152.
Example 6.56 Find all functions f W R ! R such that f .1/ D 1 and, for all
x; y 2 R:
(a) f .x C y/ D f .x/ C f .y/.
(b) f .xy/ D f .x/f .y/.
Solution Let f be a function satisfying the stated conditions. Making x D y D 0 in
(a), we get
n
1 1 1 n
f Df n D nf Dn D :
m m m m m
In order to see what happens with negative rationals, let y D x in item (a), to get
In particular, if x < 0 is rational, it follows from (6.11), together with the fact that
x is a positive rational, that f .x/ D f .x/ D .x/ D x. Therefore, f .x/ D x,
for every x 2 Q.
Since f .x/ D x for every x 2 Q, we suspect that the identity function of R is the
only one satisfying the stated conditions. To confirm that, we now turn our attention
to item (b). First of all, let’s show that, if f .x/ D 0 for some x 2 R, then x D 0.
Indeed, if x ¤ 0, then, making y D 1x in item (b), we would have
1 1
0 D f .x/f Df x D f .1/ D 1;
x x
Finally, suppose that there exists a 2 R such that f .a/ < a. We invoke the
result of Problem 4, page 206, according to which there exists a rational number
between any two real numbers. This allows us to choose a rational number r such
that f .a/ < r < a, and the increasing character of f gives
which is a contradiction. Analogously, we cannot have f .a/ > a, so that the only
possibility is f .a/ D a. However, since a 2 R was arbitrarily chosen, we conclude
that f .x/ D x for every x 2 R. t
u
Our last example shows that, for implicitly defined functions f W N ! N,
elementary divisibility arguments are sometimes useful.
Example 6.57 (Lituania) Find all functions f W N ! N such that, for all natural
numbers m and n, we have
f .f .m/ C f .n// D m C n:
Solution Let’s first prove that f is injective. To this end, let m and n be natural
numbers such that f .m/ D f .n/ D k. Then, f .2k/ D f .f .n/ C f .n// D 2n and,
analogously, f .2k/ D 2m, so that m D n.
Now, let k > 1 be a natural number. From .k 1/ C 2 D k C 1, it follows that
By the injectivity of f , we then get f .k 1/ C f .2/ D f .k/ C f .1/ or, which is the
same, f .k/ f .k 1/ D f .2/ f .1/, for every k > 1. Writing this relation for
k D 2; 3; : : : ; n and adding the corresponding equalities, we arrive at
for every natural n > 1. Letting n D 2f .1/ in the above relation, it comes that
2 D f .f .1/ C f .1// D f .2f .1// D .2f .1/ 1/.f .2/ f .1// C f .1/
and, hence,
2 f .1/
f .2/ f .1/ D :
2f .1/ 1
for all x; y 2 Q.
4. (Austria) Find all functions f W Z n f0g ! Q such that, for all x; y 2 Z n f0g for
which x C y is a multiple of 3, we have
xCy f .x/ C f .y/
f D :
3 2
1 1 1
f .xy/ C f .xz/ f .x/f .yz/ ;
2 2 4
for all x; y; z 2 R.
6. (Spain) Find all increasing functions f W N ! N such that f .n C f .n// D 2f .n/,
for every n 2 N.
7. (OIM shortlist) Find all functions f W R ! Z which satisfy the following set of
conditions:
(a) f .x C a/ D f .x/ C a, for every x 2 R and every a 2 Z.
(b) f .f .x// D 0, for x 2 Œ0; 1/.
8. (Austrian-Polish) Prove that there doesn’t exist a function f W Z ! Z such that,
for all x; y 2 Z, we have
f .x C f .y// D f .x/ y:
f .f .k// C f .k/ D 2k C 3;
for every k 2 Z.
10. (Romania) Let k > 1 be an odd integer and A D fx1 ; x2 ; : : : ; xk g be a set of k
real numbers. Find all injective functions f W A ! A such that
11. (BMO) Let a be a given real number and f W R ! R be a function such that
f .0/ D 12 and
f .x C y/ C f .x y/ D 2f .x/;
for all x; y 2 R.
20. (Poland) Find all functions f W QC ! QC satisfying, for every positive rational
number x, the following set of conditions:
(a) f .x C 1/ D f .x/ C 1.
(b) f .x3 / D f .x/3 .
6.6 Graphs of Functions 185
Gf X Y R R;
and this last set can be identified with the euclidean plane, furnished with a fixed
cartesian coordinate system7 .
Our purpose in these last two sections of the chapter is to examine some
elementary examples and study some simple properties of graphs of functions
f W X ! R, when X R is a finite union of intervals. We postpone the discussion
of the main properties of graphs of continuous and differentiable functions to
chapters 8 and 9.
In all that follows, we fix a cartesian coordinate system is in the plane.
The first point worth observing is that not every subset of the cartesian plane
can be the graph of a function. Indeed, suppose f W X ! R is a real function of a
real variable, X being a finite union of intervals. If .x0 ; y0 / 2 Gf , then, by the very
definition of graph, x0 2 X and y0 D f .x0 /. On the other hand, for a fixed x0 2 X, if
A1 .x0 ; y1 / and A2 .x0 ; y2 / are points on the graph of f , then, again from the definition,
we have
y1 D f .x0 / D y2 ;
7
We refer the reader to Chapter 6 of [4] for an adequate presentation of cartesian coordinate
systems.
186 6 The Concept of Function
(3, 0)
(−3, 0) O x
Fig. 6.5 A subset of the cartesian plane which is not the graph of a function
y
Gf
y = y0
a α β γ b x
an intersection point of the line and the graph. By pertaining to the graph of f , the
point .x0 ; y0 / must be such that x0 2 Œa; b/ and f .x0 / D y0 , so that x0 D ˛, ˇ or .
Conversely, let be given a point .x0 ; y0 / of the cartesian plane, with x0 2 Œa; b/. It is
clear that .x0 ; y0 / belongs to the horizontal line y D y0 ; moreover, if x0 is a solution
of the equation f .x/ D y0 , i.e., if f .x0 / D y0 , then x0 D ˛, ˇ or and we also
have .x0 ; y0 / 2 Gf . Therefore, the horizontal line y D y0 intersects the graph of f
exactly when y0 belongs to the image of f . The reasoning for an arbitrary function
f W X ! R, with X R, is entirely analogous and allow us to conclude that
The image of f is precisely the set of y0 2 R for which the horizontal line
y D y0 intersects the graph of f :
6.6 Graphs of Functions 187
Gf
x0
a b x
y = y0
i.e., it is the horizontal line y D c, which crosses the vertical axis at the point .0; c/
(cf. Fig. 6.8, where we consider the case c > 0).
Example 6.59 Recall (cf. Definition 6.2) that the identity function IdR W R ! R is
given by IdR .x/ D x, for every x 2 R. Therefore, its graph is the set
It is an easy exercise in euclidean geometry to verify that the points of the cartesian
plane of the form .x; x/ or .x; x/ are precisely those situated on the bisectors of the
angles formed by the coordinate axes, those of the form .x; x/ belonging to the first
188 6 The Concept of Function
(0, c) Gf
O x
G IdR
(0, a) (a, a)
(a, 0) x
or third quadrants. Hence, the graph of the identity function IdR is the line shown
in Fig. 6.9, and, from now on, will be called the bisector of odd quadrants. Observe
that the set of points of the form .x; x/, i.e., the bisector of even quadrants, is the
graph of the function f W R ! R given by f .x/ D x.
Example 6.60 The modular function is the function f W R ! R given by f .x/ D
jxj. It follows immediately from the definition of modulus of a real number that
Gf D f.x; jxj/I x 2 Rg
D f.x; jxj/I x 2 RC g [ f.x; jxj/I x 2 R g
D f.x; x/I x 2 RC g [ f.x; x/I x 2 R g:
Since the points .x; x/ and .x; x/ are symmetric with respect to the horizontal axis,
the graph of the modular function is obtained by reflecting, along the horizontal
axis, the portion of the graph of the function IdR situated in the third quadrant (cf.
Fig. 6.10).
6.6 Graphs of Functions 189
G|x|
(x, −x)
(x, x)
B(0, b)
A(−b/a, 0)
O x
Example 6.61 If f .x/ D ax C b is an affine function, then its graph is the subset of
the cartesian plane given by
According to basic analytic geometry (cf. Chapter 6 of [4], for instance), the graph
line of equation y ax b D 0, with slope a and passing through the
of f is the
points A ba ; 0 and B D .0; b/. Fig. 6.11 depicts the graph of f .x/ D ax C b in the
case a; b > 0.
For what comes next, recall (cf. Chapter 6 of [4], for instance) that, given in the
plane a point F and a line d, with F … d, the parabola of focus F and directrix d
(cf. Fig. 6.12) is the locus of the points P in the plane for which
PF D dist.P; d/;
190 6 The Concept of Function
V
d
where dist.P; d/ stands for the distance from P to d. The axis of the parabola is the
line that passes through F and is perpendicular to d, while its vertex is the point V
where it intersects its axis.
We now show that the graph of every quadratic function is a parabola. More
precisely, we have the following result.
Theorem 6.62 Given a; b; c 2 R, with a ¤ 0, the graph of the quadratic function
f .x/ D ax2 CbxCc is the parabola whose axis is the line x D 2a
b
and whose vertex
b
is V 2a ; 4a . Moreover, it “opens upwards” if a > 0, and “opens downwards”
if a < 0.
Proof Let’s look for x0 ; y0 ; k 2 R such that y0 ¤ k and, letting F.x0 ; y0 / and d be
the line y D k, we have
P 2 Gf ” PF D dist.P; d/:
P 2 Gf ” y D ax2 C bx C cI
also, the formula for the distance between two points of the cartesian plane gives
p
PF D dist.P; d/ ” .x x0 /2 C .y y0 /2 D jy kj:
y D ax2 C bx C c , .x x0 /2 C .y y0 /2 D .y k/2
1 x0 x2 C y20 k2
,yD x2 xC 0 :
2.y0 k/ y0 k 2.y0 k/
6.6 Graphs of Functions 191
1 x0 x20 1
D a; D b; C .y0 C k/ D c:
2.y0 k/ y0 k 2.y0 k/ 2
1
y0 k D ; y0 C k D ;
2a 2a
t
u
We finish this section with the coming proposition, which establishes an impor-
tant relation between the graphs of a bijection and its inverse. Then, we use it to
sketch the graphs of two important functions.
Proposition 6.63 Let I; J R be finite unions of interval. If f W I ! J is a bijection,
then the graphs of f and f 1 are symmetric with respect to the bisector of the odd
quadrants of the cartesian plane.
Proof Fix a 2 I and b 2 J. From the definition of inverse function, we have
.a; b/ 2 Gf , b D f .a/
, a D f 1 .b/
, .b; a/ 2 Gf 1 :
However, since the points .a; b/ and .b; a/ are symmetric with respect to the line
y D x, there is nothing left to do. t
u
192 6 The Concept of Function
y g
O
x
p
Fig. 6.13 Sketching the graph of x 7! x
Solution We saw at Example 6.49 that f is the inverse of g W Œ0; C1/ ! Œ0; C1/,
given by g.x/ D x2 for every x 2 Œ0; C1/. Since we already know the graph of g,
it follows from the previous proposition tha the graph of f is the symmetric of the
graph of g with respect to the line y D x (cf. Fig. 6.13). t
u
Example 6.65 Recall that the inverse proportionality function is f W R n f0g !
R n f0g, such that f .x/ D 1x , for every x 2 R n f0g. With the previous discussion
at our disposal, we can accurately sketch its graph. Indeed, f is clearly decreasing
in .0; C1/; also, we already know from the Problem 10, page 171, that f is odd,
so that (this time by the Problem 5, page 193) its graph is symmetric with respect
to the origin of the cartesian plane; on the other hand, since f is its own inverse,
the previous proposition assures that its graph is also symmetric with respect to the
bisector of the odd quadrants; finally, we will see later that its graph is a continuous
curve (i.e., one with no interruptions) which “opens upwards” in .0; C1/.
Since it is intuitively clear (and will be formalized later) that f .x/ D 1x becomes
more and more close to zero as long as x increases without bound, we arrive at
the sketch shown in Fig. 6.14. There, besides the above remarks, we plotted the
auxiliary points n; 1n , for 1 n 4 integer.
6.6 Graphs of Functions 193
y
y=x
O
x
1
Fig. 6.14 Sketching the graph of f .x/ D x
13. Prove that the graph of the inverse proportionality function is obtained by a
trigonometric (i.e., counterclockwise) rotation, of 4 radians, of the hyperbola8
of equation x2 y2 D 2.
8
For the necessary background on the equation of hyperbolas, we refer the reader to Chapter 6 of
[4], for instance.
6.7 Trigonometric Functions 195
The sine function is the function sin W R ! R, that associates to each x 2 R the
sine of an arc of x radians9 :
sin W R ! R
:
x 7! sin x
cos W R ! R
;
x 7! cos x
9
For the necessary background on Trigonometry for this section, we refer the reader to Chapter 7
of [4].
196 6 The Concept of Function
−π − π2 0 π π x
2
−1
0 6 p
4 p
3 2
0 1=2 2=2 3=2 1
With all this at hand, we immediately get Fig. 6.15, first on the interval Œ0; ,
then on the interval Œ; and finally on R.
We now recall one more piece of basic Trigonometry, which assures that
cos x D sin x C
2
for every x 2 R. Hence, item (a) of Problem 10, page 194, guarantees that, once the
graph of the sine function is drawn, we can get a corresponding sketch for the graph
of the cosine function by translating the graph of the sine function of 2 in the
direction of the horizontal axis. In Fig. 6.15, the graph of the cosine function in the
interval Œ; is given by the dashed curve contained in the strip of the cartesian
plane bounded by the (also dashed) lines y D 1 and y D 1.
Let us now look at a relevant application of the Ptolemy’s formulae for the sine
and cosine of the sum and difference of two arcs, which tells us how to proceed to
study a function given as a linear combination of the sine and cosine functions.
Example 6.68 Given positive integers a and b, let f W R ! R be the function such
that
we notice that
2 2
a b
p C p D 1:
a2 C b2 a C b2
2
6.7 Trigonometric Functions 197
Therefore, point P p 2a 2 ; p 2b 2 belongs to the portion of the trigonometric
a Cb a Cb
circle situated in the first quadrant, so that there exists a real number ˛ 2 .0; 2 / for
which
a b
cos ˛ D p and sin ˛ D p
a2 C b2 a2 C b2
(cf. Fig. 6.16). Now, it follows from the formula for the cosine of a difference that
tan W D ! R
:
x 7! tan x
198 6 The Concept of Function
For x 2 D, we have
sin.x C / sin x
tan.x C / D D D tan x;
cos.x C / cos x
and it’s immediate to verify that there doesn’t exist a real number 0 < p < such
that tan.x C p/ D tan x for every x 2 D. Therefore, the tangent function is periodic
of period . Furthermore, since D is a subset of R which is symmetric with respect
to 0 and
sin.x/ sin x
tan.x/ D D D tan x
cos.x/ cos x
− π2 − π4 0 π
4
π
2
x
−1
(b) Show that the restriction of the cosine function to the interval Œ0; defines
a decreasing bijection (which we also denote by cos, whenever there is no
danger of confusion) cos W Œ0; ! Œ1; 1; its inverse is the arc-cosine,
denoted arccos W Œ1; 1 ! Œ0; , which is also decreasing. Then, compute
arccos 12 , arcsin 1 and arcsin.1/.
(c) Show that the restriction of the tangent function to the interval 2 ; 2
defines an increasing bijection (which we also denote by tan, whenever
there is no danger of confusion) tan W 2 ; 2 ! R; its inverse is the arc-
tangent function, denoted arctan W R ! 2 ; 2 , which is also increasing.
p
Then, compute arctan 1, arctan 3 e arctan p13 .
For item (a) of the next problem, recall that the cotangent of x ¤ k (k 2 Z)
1
is the real number cot x, defined by cot x D cos
sin x . In particular, cot x D tan x .
x
2. In each of the items below, sketch the graph of the given function:
(a) f W R n fkI k 2 Zg ! R, given by f .x/ D cot x.
(b) f W Œ1; 1 ! 2 ; 2 , given by f .x/ D arcsin x.
(c) f W Œ1; 1 ! Œ0;, given by f .x/ D arccos x.
(d) f W R ! 2 ; 2 , given by f .x/ D arctan x.
3. * Let a and b given real numbers, not both zero, and f W R ! R be the function
given by
4. * Let a and b be given real numbers, at least one of which is nonzero, and
f W R ! R be the function given by
where is a nonzero real number. Show that f is periodic and compute its
period.
5. Let f W R ! R be the function given by f .x/ D 2 sin x C cos.2x/.
(a) Compute the maximum and minimum values of f , as well as the real
numbers x for which f attains these values.
(b) Show that f is periodic, of period 2.
(c) Sketch the graph of f .
6. (Canada) Compute the number of real solutions of the equation sin x D 10
x
.
7. Find the maximum
p value attained by the function f W Œ1; 1 ! R, given by
f .x/ D 3x C 4 5 x2 .
8. (New Zealand) Let ˛ be a given irrational number. Prove that the function f W
R ! R, defined by
This chapter proceeds with the study of real numbers by presenting the notion of
convergence for (infinite) sequences and series of real numbers. Among other appli-
cations, we shall introduce one of the two most important numbers of Mathematics,1
the number e. We shall also present a famous result of Kronecker on dense subsets of
the real line, which will find several interesting applications, here and in the coming
chapters.
X .1; M:
In this case, we also say that M is an upper bound for X. Similarly, a nonempty set
X R is bounded from below if there exists a real number m such that
X Œm; C1/:
In this case, we say that m is a lower bound for X. Finally, a nonempty set X R
is bounded if X is simultaneously bounded from above and from below.
1
As the reader probably suspects, the other one is the number , which is defined as the numerical
value of the area of a circle of radius 1—see [4], for instance.
a x a .resp. x a; x a/; 8 x 2 X:
On the other hand, a nonempty set X R which is not bounded (resp. not bounded
from above, not bounded from below) is said to be unbounded (resp. unbounded
from above, unbounded from below). In this case, given any positive real number
a, one can always find an element x 2 X such that
Examples 7.1
(a) The set X D f1; 12 ; 13 ; 14 ; : : :g is bounded. Indeed, 0 is a lower bound and 1 is an
upper bound for X.
(b) Bounded intervals (cf. Definition 1.10 and subsequent discussion) are bounded
sets, in the sense of the above discussion.
Example 7.2 If a nonempty set X R is bounded from above, then the subset Y of
R given by Y D fxI x 2 Xg is bounded from below, and conversely. Indeed, a real
number a is an upper bound for X if and only if a is a lower bound for Y.
In spite of its apparent obviousness, we shall list our next example as an axiom
of the natural number system, known as its archimedian property, after the Greek
mathematician Archimedes of Syracuse.2 Some important consequences of it are
collected in Problems 2, 3 and 4, page 206.
Axiom 7.3 The set N of natural numbers is unbounded from above.
Continuing with the development of the theory, fix a bounded from above
nonempty set X R. If M 2 R is an upper bound for X, then X .1; M.
Nevertheless, it may happen that there exists M 0 < M which still is an upper
bound for X, i.e., is such that X .1; M 0 . Indeed, the condition X .1; M
doesn’t guarantee that, for M 0 < M, we have X \ .M 0 ; M ¤ ;; if it happens that
X \ .M 0 ; M D ;, then we will have X .1; M 0 , and M 0 will be another upper
bound for X, which is less than M.
On the other hand, given x 2 X, it is obvious that no M 0 < x is an upper bound
for X, since x 2 X n .1; M 0 , i.e., X 6 .1; M 0 . To put in another way, the set
of upper bounds of a nonempty, bounded from above set X R is bounded from
below.
2
Archimedes, who lived in the III century b.C., was the greatest scientist of his time. Among many
important contributions to Mathematics and Physics, his seminal ideas on how to compute the
area under a parabolic segment anticipated, in some 2000 years, the development of the Integral
Calculus by Newton and Leibniz.
7.1 Supremum and Infimum 203
1
M < x M:
n
Proof Since M is the least upper bound of X and M 1n < M, the real number M 1n
is no longer an upper bound of X. Hence, there exists x 2 X such that x > M 1n .
However, since X .1; M, we must also have x M. t
u
204 7 More on Real Numbers
Problem 10 states an analogous result for the greatest lower bound of nonempty,
bounded below sets.
Proposition 7.7 Let X; Y R be nonempty sets. If x y for all x 2 X and y 2 Y,
then X is bounded above, Y is bounded below and
sup.X/ inf.Y/:
1
yn xn < < m M;
n
which is a contradiction!
Finally, since the assumption that sup.X/ < inf.Y/ leads us to a contradiction,
the only possibility is that sup.X/ D inf.Y/. t
u
We finish this section by presenting a nontrivial (and important) application of
the concept of supremum of a bounded above nonempty set. To set the stage, we
recall that, in the previous chapters, we several times relied upon the existence of
square roots of positive real numbers. The next result establishes existence in a more
general setting.
Theorem 7.9 Given a positive real number x and a natural number n > 1, there
exists a unique positive real number y such that yn D x.
Proof Let’s consider just the case x > 1, referring the reader to Problem 1 for the
case 0 < x < 1 and observing that the case x D 1 is trivial. If
X
n
2n n.2y/n
z <y C
n n
yn D yn C I
jD1
k k
n n
therefore, we have zn < x provided yn C n.2y/ k
< x, i.e., k > n.2y/
xyn
.
1
For the proof of (ii), suppose y > x. Letting z D y k , with k 2 N, then, as
n
in the previous paragraph anterior, we have yn > zn . We claim that zn > x for all
sufficiently large k. To this end, arguing as above we have
n !
1 Xn
n 1 nj
zn D y D yn C .1/j y
k jD1
j kj
!
X n
n 1 nj n.2y/n
>y n
y > y n
I
jD1
j kj k
n.2y/n n.2y/n
hence, zn > x whenever yn k
> x, i.e., k > yn x
. t
u
For another, conceptually simpler, proof of the above theorem, we refer the reader
to Example 8.36.
206 7 More on Real Numbers
p p
8. If X D fj a bjI a; b 2 N and a ¤ bg, compute inf.X/.
9. Let
C1 D .0; 1/
C2 D .0; 1/ n .1=3; 2=3/ D .0; 1=3 [ Œ2=3; 1/
C3 D ..0; 1=3 n .1=9; 2=9// [ .Œ2=3; 1/ n .7=9; 8=9/
D .0; 1=9 [ Œ2=9; 1=3 [ Œ2=3; 7=9 [ Œ8=9; 1/:
More generally, for each n 2 N, obtain CnC1 from Cn , by S erasing the open
middle third of each of the intervals that form Cn . If C D n1 Cn , show that
inf.C/ D 0 and sup.C/ D 1.
10. * Let Y R be a nonempty, bounded below set, with m D inf Y. If n 2 N,
prove that there exists y 2 Y such that
1
my<mC :
n
11. Let X; Y R be nonempty sets, such that X is bounded above, Y is bounded
below and sup.X/ D inf.Y/ D ˛. If ˛ … X [ Y, prove that there exist elements
xn 2 X and yn 2 Y for which yn xn < 1n , for every n 2 N.
12. * Let X R be a nonempty, bounded above set. Given c 2 R, let cX D fcxI x 2
Xg. Prove that:
(a) If c > 0, then sup.cX/ D c sup.X/.
(b) If c < 0, then cX is bounded below and inf.cX/ D c sup.X/.
Then, if X is bounded below, establish properties analogous to the ones listed
above, relating inf X to the sup or the inf of cX, according to whether c < 0 or
c > 0.
13. * Let X; Y R be nonempty sets and X C Y D fx C yI x 2 X and y 2 Yg.
(a) If X and Y are bounded above, prove that X C Y is bounded above and
sup.X C Y/ D sup.X/ C sup.Y/.
(b) If X and Y are bounded below, prove that X C Y is bounded below and
inf.X C Y/ D inf.X/ C inf.Y/.
14. * Let X; Y Œ0; C1/ be nonempty, bounded above sets. If XY D fxyI x 2
X and y 2 Yg, prove that XY is bounded above and such that sup.XY/ D
sup.X/ sup.Y/ and inf.XY/ D inf.X/ inf.Y/.
15. (Hungary) Let .Rn /n1 be a sequence of pairwise distinct rectangles in the
cartesian plane, each of which having all vertices with integer coordinates and
two sides along the axes. Prove that one can find two of them such that one
contains the other.
16. (IMO) Let f ; g W R ! R be functions satisfying, for all x; y 2 R, the relation
f .x C y/ C f .x y/ D 2f .x/g.y/:
If f is not identically zero and jf .x/j 1 for every x 2 R, prove that jg.x/j 1,
for every x 2 R.
208 7 More on Real Numbers
Given a sequence .an /n1 of real numbers, we are interested in recognizing whether
or not their terms are approaching a certain real number l, as n increases. For
instance, if an D 1n , it is reasonable to say that the numbers an become closer and
closer to 0 as n increases, since the result of the division of 1 by n is increasingly
smaller as n increases. This naive point of view is formalized as follows.
Definition 7.10 A sequence .an /n1 of real numbers converges to a real number l
if, given an error > 0 for the value of l, there exists an index n0 2 N such that
jan lj < for every n > n0 .
Alternatively, if the sequence .an /n1 converges to l, we say that it is convergent
and that l is a limit of the sequence, which we denote by writing
n
an ! l or lim an D l:
n!C1
For the reader to get used to the important concept of convergent sequence, we
collect below some elementary examples of convergent and divergent sequences.
Examples 7.11
n
(a) If an D 1n , then an ! 0: indeed, for a given > 0, we have jan 0j <
provided n > 1 ; thus, once we have chosen n0 2 N such that n0 > 1 , we will
have jan 0j < whenever n > n0 .
(b) If an D .1/n , then .an /n1 is divergent: indeed, since the terms of the sequence
are alternately equal to 1 and 1, it is impossible for them to (collectively)
become closer to a single real number l (formalize this intuition).
n
(c) If an D 1 C .1/
n
n
, then an ! 1: this is so because jan 1j D 1n , so that
jan 1j < for n > 1 .
(d) If .an /n1 is a constant sequence, with an D c for every n 1, then an ! c.
n
Example 7.12 If an D qn , with 0 < jqj < 1, then an ! 0.
1 1
Proof Since jqj > 1, we can write jqj D 1 C ˛, with ˛ > 0. Therefore, taking the
first two terms in the binomial expansion formula, we get
1
D .1 C ˛/n 1 C n˛
jqjn
7.2 Limits of Sequences 209
and, hence,
1
jan 0j D jqjn :
1 C n˛
1
Thus, if we wish that jan 0j < , it suffices to impose < or, equivalently,
1Cn˛
n > ˛1 1 1 . t
u
p p
Example 7.13 The sequence .an /n1 , given for n 1 by an D n C 1 n,
converges to 0.
1
Proof Note that an D p p .
nC1C n
Thus, given > 0 and taking n0 2 N such that
1
n0 > 42
, we have
p p p p p 1
n > n0 ) n C 1 C n > n0 C 1 C n0 > 2 n0 > :
Therefore,
1
n > n0 ) jan 0j D p p < :
nC1C n
t
u
The notion of convergent sequence doesn’t make it clear whether the correspon-
dent limit is unique. Yet in another way, in principle it could happen that a certain
sequence converges to more than one limit. The coming result shows that this is not
so.
Proposition 7.14 If the sequence .an /n1 converges, then its limit is unique.
Proof Let l1 and l2 be distinct real numbers, and suppose that the given sequence
simultaneously converges to l1 and l2 . Toking D 12 jl1 l2 j > 0, the definition of
convergence guarantees the existence of n1 ; n2 2 N such that
which is an absurd. t
u
Thanks to the previous result, from now on we speak of the limit of a convergent
sequence. In this respect, the next proposition collects two basic, albeit very
important, properties of limits of convergent sequences. In order to state it properly,
we define a subsequence of a sequence .an /n1 as the restriction of the given
sequence to an infinite subset N1 D fn1 < n2 < n3 < g of N; in this case,
we denote it by .ank /k2N . Since the function j 7! nj from N1 to N is a bijection,
every subsequence can actually be seen as a sequence.
210 7 More on Real Numbers
an < l C D l C .a l/ D a;
which is an absurd.
n
(b) Let > 0 be given. Since an ! l, there exists a natural number n0 such that
jan lj < for n > n0 . Since n1 < n2 < n3 < , there exists an index ni such
that nj > n0 for j i; hence, for all such j, we have janj lj < , which is the
k
same as saying that ank ! l. t
u
In words, item (b) of the previous proposition says that, if the terms of a certain
sequence come closer and closer to l as their indices increase, then the same is true
for the terms of every subsequence of the given sequence. Item (b) of the previous
proposition also has the following immediate corollary, which gives us a sufficient
(and quite useful) condition for the divergence of a sequence. sequência.
Corollary 7.16 If two subsequences of a given sequence converge to distinct limits,
then the original sequence is divergent.
Up to now, except for some very simple examples we haven’t seen how one could
find out the limit of a convergent sequence. In order to remedy this situation, we
need to understand how to perform simple arithmetic operations with convergent
sequences. We turn to this next, starting with an auxiliary result which is important
in its own.
We say that a sequence .an /n1 is bounded (resp. bounded from above,
bounded from below) if the set fa1 ; a2 ; : : :g is bounded (resp. bounded from above,
bounded from below), in the sense of the previous section.
Lemma 7.17 Every convergent sequence is bounded.
Proof If .an /n1 is a convergent sequence with limit l, then there exists n0 2 N such
that
Finally, letting L D maxf1 C jaj; ja1 j; ja2 j; : : : ; jan0 1 jg, we get jan j < L for every
n 2 N, so that the sequence is bounded. t
u
Proposition 7.18 Let .an /n1 and .bn /n1 be convergent sequences, and c be any
real number.
n n
(a) If an ! a, then can ! ca.
n n n n
(b) If an ! a and bn ! b, then an ˙ bn ! a ˙ b and an bn ! ab.
n n
(c) If an ! 0 and .bn /n1 is bounded, then an bn ! 0.
n n n
(d) If an ! a and bn ! b, with b; bn ¤ 0 for every n 1, then abnn ! ab .
Proof
(a) If c D 0, then can D ca D 0, and there is nothing to do. Suppose, then, that
n
c ¤ 0, and let > 0 be given. Since an ! a, there exists n0 2 N such that
n > n0 ) jan aj < jcj . Hence,
n > n0 ) jcan caj D jcjjan aj < jcj D :
jcj
n
(b) For the first part, let’s prove that an Cbn ! aCb (to prove that an bn ! ab
n n
is completely analogous). Given > 0, the convergences an ! a and bn ! b
assure the existence of n1 ; n2 2 N such that
n > n1 ) jan aj < and n > n2 ) jbn bj < :
2 2
Therefore, taking n > maxfn1 ; n2 g, we get
j.an C bn / .a C b/j jan aj C jbn bj < C D :
2 2
For the second part, let L > 0 be such that jbn j < L for every n 2 N. Given
> 0, take n0 2 N for which
n > n0 ) jan aj < and jbn bj < :
2L 2jaj C 1
Then
(c) Let L > 0 be such that jbn j < L for every n 1. Given > 0, let’s take n0 2 N
such that
n > n0 ) jan j < :
L
212 7 More on Real Numbers
Then,
n > n0 ) jan bn 0j D jan jjbn j < L D :
L
1 n
(d) By the second part of item (b), it suffices to show that bn
! 1b . To this end,
start by observing that
ˇ ˇ
ˇ1 ˇ
ˇ 1 ˇ D 1 jbn bj 1 jbn bj ;
ˇb bˇ jbj jbn j jbj jbj jbn bj
n
where we used the triangle inequality in the last passage above. Now, given
> 0, choose n0 2 N such that
jbj
n > n0 ) jbn bj < ; :
2 2
Then, for n > n0 , we have
ˇ ˇ
ˇ1 ˇ
ˇ 1 ˇ 1 jbn bj D 2jbn bj < :
ˇb b ˇ jbj jbj jbj=2
n
t
u
p
Example 7.19 Let a be a positive real number. If .an /n1 is given by an D a for
n
n
every n 1, then an ! 1.
Proof If a > 1, then an > 1. Write an D 1 C bn , so that bn > 0. Since
!
n
a D ann D .1 C bn /n 1 C bn D 1 C nbn ;
1
so that
2
0 < b2n < :
n1
n
Hence, once more from the squeezing principle, the above inequality gives bn ! 0
n
and, thus, an D 1 C bn ! 1. u
t
For what comes next, recall that a sequence .an /n1 of real numbers is just a
function f W N ! R, for which we write an D f .n/. Hence, it is natural to say that
.an /n1 is monotonic increasing (resp. decreasing, nondecreasing, nonincreasing)
provided an < anC1 (resp. an > anC1 , an anC1 , an anC1 ), for every n 1.
The most important result on limits of sequences is the theorem below, which is
known in the mathematical literature as Bolzano-Weierstrass theorem.3
Theorem 7.21 (Bolzano-Weierstrass ) Every monotonic bounded sequence is
convergent.
Proof Suppose that .an /n1 is a nondecreasing bounded sequence (the other cases
can be dealt with similarly), i.e., that
a1 a2 a3 < M;
for some M > 0. Then, M is an upper bound for the set A D fa1 ; a2 ; a3 ; : : :g, so that
n
A has a sup, say sup A D l. We claim that an ! l. Indeed, let > 0 be given; since
l is no longer an upper bound for A, some element of it is greater than l ,
say, an0 > l . Therefore, since an0 an0 C1 an0 C2 , we conclude that
an > l for every n n0 . Thus, for n n0 , we have
l < an l < l C ;
as we wished to show. t
u
The previous theorem, together with the definition of convergence, assures that if
a bounded sequence is monotonic from a certain term on, then it will be convergent.
We explore this comment by revisiting the last two examples.
p
Example 7.22 Let a be a positive real number. If .an /n1 is given by an D n a for
n
every n 1, then an ! 1.
Proof Assume a > 1 (the case 0 < a < 1 can be dealt with as in the Example 7.19).
Then, a1 > a2 > a3 > > 1, and the Bolzano-Weierstrass theorem guarantees
the existence of l D limn!C1 an 1. Item (a) of Proposition 7.15 gives l
1, and item (b) guarantees that every subsequence of .an /n1 also converges to l.
3
After Bernard Bolzano and Karl Weierstrass, German mathematicians of the XIX century.
214 7 More on Real Numbers
k p
k a
Therefore, ak.kC1/ ! l. Now, since ak.kC1/ D p
kC1 a
, it follows from item (d) of
Proposition 7.18 that
p
k
a k l
ak.kC1/ D p ! D 1:
kC1
a l
Therefore, l D 1. t
u
p
Example 7.23 The sequence .an /n1 , given by an D n for every n 1, converges
n
to 1.
p p 3
p
Proof The initial terms
p ofpthe sequence
p pare 2,p 3, 4 4, . . . , and it is easy to
directly show that 2 < 3 3 and 3 3 > 4 4 > 5 5. Since 2n n2 for n 4 (by
induction, for instance), we get a2 an > 1 for n 4, so that the sequence is
bounded; hence, if we show that it is indeed decreasing from the third term on, its
convergence will follow from Bolzano-Weierstrass theorem, with limit l 1.
For what is left to do, for an integer n > 2 we have
p p 1 n
n> n C 1 , nnC1 > .n C 1/n , n > 1 C :
n nC1
Let’s prove the last inequality above. For n D 3, n it’s immediate to check it
numerically; for n > 3, it suffices to show that 1 C 1n < 3. To this end, notice that
! ! !
1 n n 1 n 1 n 1
1C D1C C CC
n 1 n 2 n2 n nn
and
!
n 1 nŠ 1 n.n 1/ : : : .n k C 1/ 1 1
D D < k1 :
k nk kŠ.n k/Šnk kŠ nk kŠ 2
Therefore,
! ! !
1 n n 1 n 1 n 1
1C D1C C CC
n 1 n 2 n2 n nn
1 1 1
<1C1C C 2 C C n1
2 2 2
1
D3 < 3:
2n1
In order to finish, we need to show that l D 1. To this end, first observe
p that
pp the
p
2k
p
2k 2k
subsequence a2k D 2k also converges to l. On the other hand, 2k D 2 k
k,
7.2 Limits of Sequences 215
p k pp k p
2k
with 2 ! 1. Now, it follows from Problem 2 that k k ! l. Therefore, by
applying item (b) of Proposition 7.18, we get
q
p
2k
p
2k
p p
l D lim 2k D lim 2 lim k D l;
k
t
u
Sometimes, we have to show that a given sequence has at least a convergent
subsequence (even if, as a whole, the sequence does not converge). In this sense,
Theorem 7.25 below, also due to Weierstrass, provides a sufficient condition for the
existence of such a subsequence. Before we state and prove it, we need to discuss an
important auxiliary result, known as the lemma of nested intervals. In what follows
if I D Œa; b, with a; b 2 R, we shall let jIj D b a.
Lemma 7.24 For n 2 N, let In D Œan ; bn . If I1
I2T
I3
: : : and
limn!C1 jIn j D 0, then there exists a unique l 2 R such that n1 In D flg.
Proof First of all, note that the intersection of the intervals In , if not empty, has a
single element;Tindeed, if there existed reals a < b in such an intersection, we would
have Œa; b n1 In ; in particular, Œa; b In and, hence, jIn j b a for every
n 2 N, thus contradicting the fact that limn!C1 jIn j D 0.
Secondly, the inclusions I1
I2
I3
: : : give a1 a2 a3
b3 b2 b1 , and the Bolzano-Weierstrass
T theorem assures the existence
of l D limn!C1 an . We claim that l 2 n1 In .
Since l D supfan I n 2 Ng, it follows that an l for every n 2 N. On the
other hand, for a fixed m 2 N, we have an bm for every n 2 N, and item (a)
of Proposition 7.15 gives l D limn!C1 an bm . However, since m was chosen
arbitrarily, we have l bm for every m 2 N. It thus follows that l 2 Œam ; bm D Im
for every m 2 N, as we wished to show. t
u
We are now in position to prove Weierstrass theorem.
Theorem 7.25 (Weierstrass ) Every bounded sequence admits a convergent sub-
sequence.
Proof Let .an /n1 be a given bounded sequence, and I0 D Œa; b be a closed and
bounded
aCb interval containing
all of its terms. One (possibly both) of the intervals
a; 2 and aCb 2 ; b , call it I 1 , also contains infinitely many terms of the sequence.
Do the same with I1 , obtaining a closed and bounded interval I2 I1 such that jI2 j D
1
2 jI1 j and I2 contains infinitely many terms of the sequence .an /n1 . Proceeding
inductively, we construct a nested sequence I1
I2
I3
of closed and
bounded intervals, such that jIkC1 j D 12 jIk j and Ik contains infinitely many terms of
the sequence .an /n1 , for every T k 1. Therefore, by the lemma of nested intervals,
there exists c 2 R such that k1 Ik D fcg.
Now, choose n1 2 N such that an1 2 I1 ; then, after having chosen nj 2 N such
that anj 2 Ij , choose njC1 2 N such that njC1 > nj and anjC1 2 IjC1 (this is possible
216 7 More on Real Numbers
by the way the Ij ’s were defined). This way, we inductively construct a subsequence
.ank /k1 of .an /n1 , such that ank 2 Ik for every k 1. Since jIk j D 21k jI0 j and
1 1 k
ank ; c 2 Ik , we conclude that jank cj jI j,
2k 0
for every k 1; since jI j
2k 0
! 0,
k
we conclude that ank ! c. t
u
The concept of convergent sequence gives a precise meaning to the geometric
intuition that the terms of the given sequence come closer and closer to a certain
real number (the limit of the sequence), as long as their indices increase. However,
it is also reasonable to expect that, if the terms of a given sequence come close
together, then the sequence should also converge. This is indeed the case and, in
order to establish it, we start with the following
Definition 7.26 A sequence .an /n1 is said to be a Cauchy sequence if, for every
> 0, there exists n0 2 N such that
On the other hand, since the sequence is Cauchy, there exists N1 2 N such that
m; n > N1 ) jam an j < :
2
Letting M D maxfN1 ; N2 g and fixing nk > M, we have, for n > M and with the aid
of the triangle inequality,
jan lj jan ank j C jank lj < C D ;
2 2
as we wished to show. t
u
The coming example collects an interesting application of the above result.
Example 7.28 Let .an /n1 be a sequence of real numbers such that
for every n 2 N, where 0 < c < 1 is a real constant. Show that this sequence is
convergent.
Proof By Theorem 7.27, it suffices to show that .an /n1 is a Cauchy sequence. To
this end, iterating the inequality in the statement we get, for every k 2 N,
Let n and p be given natural numbers. The above inequality, together with the
triangle inequality, gives
X
nCp1
X
nCp1
janCp an j jakC1 ak j ck1 ja2 a1 j
kDn kDn
nCp
c n1
c
D ja2 a1 j
1c
1
< ja2 a1 jcn1 :
1c
We now note that, by Example 7.12, the last expression above tends to 0 when
1
n ! C1. Therefore, given > 0, there exists n0 2 N such that 1c ja2 a1 jcn1 <
for every n > n0 . Hence, for n > n0 and p 2 N, we get janCp an j < , so that
.an /n1 is, indeed, a Cauchy sequence. t
u
218 7 More on Real Numbers
1. * Let .an /n1 and .bn /n1 be convergent sequences of real numbers, with
limn!C1 an D a and limn!C1 bn D b. Generalize item (a) of Proposi-
tion 7.15, showing that if an bn for every n 1, then a b.
2. * Let .an /n1 be a sequence of positive real numbers converging to a > 0. Show
p n p
that an ! a.
n
3. * Given a 2 R such that jaj > 1, show that anŠ ! 0 when n ! C1.
4. Generalize the result of Example 7.12, showing that, given k 2 N and a 2 R,
k n
with jaj > 1, we have ann ! 0.
5. Let .an /n1 and .bn /n1 be sequences of real numbers and, for each n 2 N, let
tn 2 Œ0; 1 be given. Denote by .cn /n1 the sequence defined by
cn D .1 tn /an C tn bn ;
n n
for every n 2 N. If an ; bn ! l, show that cn ! l.
6. * Prove the squeezing theorem: let .an /n1 , .bn /n1 and .cn /n1 be sequences
n
of real numbers such that an bn cn , for every n 2 N. If an ; cn ! l, for
n
some l 2 R, show that bn ! l.
7. Compute the following limits:
p
n n
(a) limn!C1 n2 C1
.
p
(b) limn!C1 . n2 C an C b n/, with a; b 2 R.
p
(c) limn!C1 n 1 C qn , where 0 < q < 1 is a real number.
p
(d) limn!C1 n an C bn , with a and b positive reals such that a > b.
8. * This problem extends the concept of limit of sequences to consider infinite
limits. We say that a sequence .an /n1 of real numbers converges to C1 (resp.
1) if, given M > 0, there exists n0 2 N such that n > n0 ) an > M (resp.
an < M). In this case, we denote limn!C1 an D C1 (resp. limn!C1 an D
n n
1), or simply an ! C1 (resp. an ! 1). With respect to this concept,
and given sequences .an /n1 and .bn /n1 of real numbers, do the following
items:
n n
(a) If an ! ˙1 and .bn /n1 is bounded, then an C bn ! ˙1.
n
(b) If an ! ˙1 and bn c > 0 (resp. bn c < 0) for every n 1, then
n n
an bn ! ˙1 (resp. an bn ! 1).
n
(c) If bn ! C1 and there exists c > 0 such that an cbn (resp. an cbn )
n
for every n 1, then an ! ˙1.
9. Let q be a real number and .an /n1 be the sequence defined by an D qn . If
n
q > 1, show that an ! C1. If q < 1, show that .an /n1 does not converge
to either C1 or 1.
7.2 Limits of Sequences 219
n
10. * Given positive reals a and q, with q < 1, show that an q2 ! 0.
n
11. (IMO shortlist) Let .an /n1 be a sequence of real numbers such that, for every
m; n 2 N, we have
2mn
jam an j :
m2C n2
for every k 1. Prove that the sequence converges and compute its limit.
14. Let n > 1 be a given integer and t0 ; t1 ; : : : ; tn be given real numbers, such that
t0 C t1 C C tn D 0. Prove that the sequence .ak /k1 , defined by
p p p
ak D t0 k C t1 k C 1 C C tn k C n
converges to 0. p
15. (Romania) The sequence .xn /n1 is such that xnC1 C 2 xn 2, for every
n 1. Find all possible values of x1986 .
16. (Leningrad) Let .an /n1 be a sequence of real numbers such that
1
jam C an amCn j ;
mCn
Prove that the sequence .an /n2 is decreasing and convergent, and compute its
limit.
18. (Romania) Let k be a fixed natural number and .an /n1 be the sequence defined
by
r q
p
an D kC k CC k;
21. (OIMU) Let c and ˛ be positive real constants4 and Q be a square in the plane.
Prove that there doesn’t exist a surjection f W Œ0; 1 ! Q for which
23. (IMO shortlist) Let .an /n1 be a sequence of positive reals. Show that
p
1 C an > an1 2
n
4
Powers of a positive basis with real exponents will be defined in Sect. 10.7. For the time being,
you may assume that ˛ is a positive rational, if you will.
7.3 Kronecker’s Lemma 221
In this section, we apply some of the ideas exposed so far in this chapter to study
the important concept of dense set, as well as to present an interesting geometric
application of it. We start by recalling the following definition.
Definition 7.29 Given an interval I, we say that a subset X of I is dense (in I) if,
for every a 2 I and > 0, it happens that
X \ .a ; a C / ¤ ;:
G˛ D fm˛I m 2 Zg G: (7.2)
5
After Leopold Kronecker, German mathematician of the XIX century.
222 7 More on Real Numbers
mx a < a C .m C 1/x;
so that
x D .m C 1/x mx .a C / .a / D 2;
0 < ˇ < 2˛ ˛ D ˛:
GC fn˛I n 2 Ng
and, since the opposite inclusion was already established in (7.2), we actually
have GC D fn˛I n 2 Ng. Finally, since G D GC [ f0g [ G , where G D
fxI x 2 GC g, it is immediate to see that
G D fm˛I m 2 Zg D G˛ :
t
u
In the coming corollary, we stick to the notation set forth in (7.1).
7.3 Kronecker’s Lemma 223
˛ D nˇ and 1 C ˛ D mˇ:
Now, since ˛ is irrational, the first equality above would give n ¤ 0 and
ˇ D ˛n … Q. On the other hand, we would also have
.m n/ˇ D .1 C ˛/ ˛ D 1;
1
so that ˇ D mn 2 Q.
We have then reached a contradiction, which came from the supposition that G
is not dense in R. Therefore, G is indeed dense in R, as we wished to show. t
u
Our next corollary refines the conclusion of the previous one.
Corollary 7.33 If ˛ is an irrational number, then the following sets are dense
in R:
(a) A D fm C n˛I m; n 2 Z and m < 0 < ng.
(b) B D fm C n˛I m; n 2 Z and n < 0 < mg.
Proof Without any loss of generality, we can suppose that ˛ > 0. Let’s prove item
(a), the proof of item (b) being totally analogous. Given a 2 R and > 0, we want
to establish the existence of x 2 A such that a < x < a C . Suppose that
a 0 (the remaining cases are entirely analogous), and let ı D minf˛; 2g > 0.
We first claim that there exist m; n 2 Z such that m < 0 < n and m C n˛ 2
A \ .0; ı/. By contradiction, suppose that
m C n˛ 2 A \ .0; ı/ ) n 0:
Choose (by the former corollary) x0 D m0 C n0 ˛ 2 A \ .0; ı/, with the greatest
possible n0 0. Since Kronecker’s lemma guarantees that A \ .0; x0/ is infinite, we
can take x1 D m1 C n1 ˛ 2 A \ .0; x0 /, with n1 < n0 . Then,
which is a contradiction, for, n0 n1 > 0. Hence, we can choose mCn˛ 2 A\.0; ı/,
with n > 0. This being said, if it were m 0, we would have m C n˛ ˛ ı,
which is another contradiction. Therefore, m < 0 and, thus, A \ .0; ı/ ¤ ;.
224 7 More on Real Numbers
Now, take x 2 A \ .0; ı/ and consider all numbers of the form kx, with k 2 ZC .
Letting k0 be the greatest nonnegative integer for which k0 x a , we claim that
.k0 C 1/x 2 .a ; a C /. Indeed, if it were .k0 C 1/x a C , we would have
and, hence,
for all P; Q 2 …. Prove that f is an isometry of …, i.e., prove that, for all P; Q 2 …,
one has d.P; Q/ D d.f .P/; f .Q//, where d.X; Y/ D XY stands for the euclidean
distance between the points X and Y.
Proof For P 2 …, we let f .P/ be systematically denoted by P0 , so that d.P; Q/ D
PQ andpd.f .P/; f .Q// D P0 Q0 . Firstly, let’s show that f must preserve segments of
length 3.
Claim 1
p p
PQ D 3 ) P0 Q0 D 3:
p
Indeed, given points P and Q in the plane for which PQ D 3, let’s construct
points R and S such that both QRS and PRS are equilateral triangles whose side
lengths are equal to 1 (cf. Fig. 7.1).
p
Fig. 7.1 pPQ D 3) Q
P0 Q0 D 3
U
T
S R
P
7.3 Kronecker’s Lemma 225
Fig. 7.2 PQ D 2 ) S T
P0 Q0 D 2
P R Q
It suffices to establish the case n D 2, the general case being totally analogous.
Let P and Q be such that PQ D 2, and let R be the midpoint of PQ, such that
PR D RQ D 1 (cf. Fig. 7.2). Let’s consider points S and T such that PRS, RST
and QRT are equilateral triangles of side lengths equal to 1, all situated on one of
!
the half-planes determined by line PQ. Using claim 1 twice, it’s immediate that
P0 Q0 D 2.
Analogously, we can prove that
p p
PQ D n 3 ) P0 Q0 D n 3:
R
R −mk
√ −mk f
nk 3 √ Q
nk 3
l
P l Q P
Fig. 7.3 P0 Q0 PQ
Fig. 7.4 PQ D l ) X
P0 Q0 D l
P Q
!
Therefore,ptriangle inequality assures the existence of a point R 2 … n PQ such that
PR D nk 3 and RQ D mk ; since we already have PQ D l, there is nothing left
to do.
It follows from what we did above (cf. Fig. 7.3) that P0 Q0 C Rp 0 Q0 P0 R0 or,
(a) Show that at least two of the numbers f˛g, f2˛g, . . . , f.p C 1/˛g belong to
a single interval of the form pk ; kC1
p
, for some integer k satisfying 0 k <
p.
(b) Use the result of (a) to show that there exist m0 ; n0 2 Z such that 0 <
m0 C n0 ˛ < 1p .
(c) Use the result of (b) to show that there exists r 2 Z such that r.m0 C n0 ˛/ 2
.a ; a C /.
3. * Given ˛; l 2 R, with ˛ irrational, show that there exist sequences .mk /k1 and
.nk /k1 of integers satisfying the following conditions:
(a) mk < 0 < nk , for every k 1;
(b) limk!C1 .mk C nk ˛/ D l.
For the next problem, we assume from the reader some familiarity with the
basics of plane analytic geometry and vector algebra in the plane. We refer to
Chaps. 6 and 8 of [4] for the necessary background.
4. A subset X of an euclidean plane … is said to be dense in provided X intersects
every disk of …. Now, let O be a fixed point in …. A nonempty subset X of … is
said to be an additive subgroup of … with respect to O provided the following
! ! !
condition is satisfied: for every A; B 2 X, if OA OB D OC, then C 2 X.
(a) If A1 ; : : : ; An 2 … and
nX
n
! o
XA1 ;:::;An D mk OAk I mk 2 Z; 8 1 k n ;
kD1
form a partition of the natural numbers. To this end, do the following items:
(a) Show that if ˛ > 1 is irrational
n o and n 2 N, then n is a term of the sequence
.bk˛c/k1 if and only if ˛n > 1 ˛1 .
n o n o
(b) Given n 2 N, show that ˛n C ˇn D 1.
n o n o
(c) Conclude that either ˛n > 1 ˛1 or ˇn > 1 ˇ1 , but not both.
(d) Finish the proof.
In the notations above, one says that .ak /k1 and .bk /k1 , given by ak D bk˛c
and bk D bkˇc, are the Beatty sequences6 corresponding to ˛ and ˇ.
C1
X
an ;
nD1
6
After Samuel Beatty, Canadian mathematician of the XX century.
7.4 Series of Real Numbers 229
P
or simply n1 an , we mean the sequence .sn /n1 , where sn D a1 Ca2P C Can for
n 1. The real number sn is called the n-th partial sum of the series n1 an , and
we say that such a series converges to s 2 R if the sequence .sn /n1 of its partial
sums converges to s. In this case, we say that s is the sum of the series and write
X
an D s: (7.3)
n1
P
In other words, whenever we write k1 ak D s, we will be saying that the terms
of the sequence of finite sums sn D a1 C a2 C C an come closer and closer to
the real number s, as long as n ! C1. It is in this sense that equality (7.3) must be
thought of, as a limit.
We shall sometimes have a sequence P .an /n0 of reals, in which case the
corresponding series will be denoted by n0 an . We leave to the reader the
(immediate) task of adapting the former and coming discussions to such a situation.
Our main interest in this section is to find out efficient criteria to decide whether
a given series converges or not. If it doesn’t converge, we shall say that it is a
divergent series. Let’s see two simple examples of divergent series.
P P
Example 7.35 The series k1 k and k1 .1/k diverge.
Proof The first series diverges, for its n-th partial sum is sn D 1 C 2 C C n D
n.nC1/
2 , so that .sn /n1 is a divergent sequence. In the second case, the n-th partial
sum sn of the given series is such that sn D 0 if n is even and sn D 1 if n is odd, so
that .sn /n1 is also a divergent sequence. t
u
P
Given a series n1 an , we refer to a generic term an as the general term of the
series. The following proposition gives a necessary condition on the general term of
a series, if it is to converge.
P n
Proposition 7.36 If the series k1 ak converges, then ak ! 0.
Proof Given > 0, we want to Pprove that there exists n0 2 N such that n > n0 )
jan j < . To this end, let l D k1 ak . By the definition of convergence for series,
there exists n0 2 N such that
n n0 ) j.a1 C a2 C C an / lj < :
2
Therefore, it follows from the triangle inequality that, for n > n0 , we have
The converse to the above proposition is not true, namely, there are divergent
P n
series k1 ak for which ak ! 0. The classical example is that of the harmonic
P 1
series, i.e., the series k1 k , whose divergence is established in the coming
example and will find further use in these notes.
Example 7.37 Given n 2 N, let m be the only natural number such that 2m n <
2mC1 . Then,
X
n
1 m
C 1: (7.4)
kD1
k 2
Hence,
2m
1 X1 1 X1
n
1C C 1C C
2 jD3
j 2 jD3
j
m
1 X 1 1
D1C C C C
2 kD2 2k1 C 1 2k
1 X1
m
m
>1C C D1C :
2 kD2 2 2
t
u
P In what comes next, we shall show that, for every rational r > 1, the series
1
k1 k r converges. To this end, we need to examine the convergence of a geometric
series, i.e., of a series of the form
X
qk1 ;
k1
for a certain nonzero real number q. In this sense, we have the following important
result.
P
Proposition 7.38 Given q 2 R n f0g, the geometric series k1 qk1 converges if
and only if 0 < jqj < 1. Moreover, if this is so, then
X 1
qk1 D :
k1
1q
7.4 Series of Real Numbers 231
Proof If jqj 1, the geometric series diverges, since its general term qk1 doesn’t
converge to 0. Suppose, then, that 0 < jqj < 1, and let sn D 1 C q C C qn1 be
the n-th partial sum of the series. By the formula for the sum of the terms of a finite
GP, we have
1 qn 1 qn
sn D D :
1q 1q 1q
1
Hence, in order to show that the series converges to 1q
, it suffices to show that
n
q ! 0. But this was done in Example 7.12.
n
t
u
We can now discuss the promised example.
P 1
Example 7.39 If r > 1 is rational, then the series k1 kr converges.7
Proof By the Bolzano-Weierstrass P theorem, it suffices to show that the sequence
.sn /n1 of the partial sums sn D nkD1 k1r is bounded. To this end, given n 2 N, take
m 2 N such that 2m > n. Then,
1 1 1 1
sn 1 C C r CC CC m
2r 3 .2m1 /r .2 1/r
1 1 1
< 1C2 C 4 r C C 2m1 .m1/r
2r 4 2
1 1 1
< 1 C r1 C r1 C C .m1/.r1/
2 4 2
X 1
< :
k0
2.r1/k
1
However, since r > 1, we have 0 < 2r1
< 1, and it follows from the previous
proposition that
X 1 2r1
D :
k0
2.r1/k 2r1 1
2r1
Therefore, we conclude that 0 < sn < for every n 2 N, so that the sequence
2r1 1
.sn /n1 is, indeed, bounded. t
u
P
Remark 7.40 For the sake of curiosity, we inform the reader that k1 k1r D .r/,
where W .1; C1/ ! R stands for the famous Riemann’s zeta function.8
7
Had powers kr , with r > 0 real, been defined (this will be done inP Sect. 10.7), the reasoning
presented in the proof would work equally well to show that the series k1 k1r converges.
8
After Bernhard Riemann, German mathematician of the XIX century. For more on Riemann, see
the footnote at page 354.
232 7 More on Real Numbers
2
An elementary computation of .2/ D 6 will be hinted to at Problem 12, page 470.
We also refer to Chap. 9 of [5]. Note, however, that for an odd natural number
m > 1 the computation of the exact numerical value of .m/ is an open problem in
Mathematics.
We finish our initial list of examples of series with an additional application of
Bolzano-Weierstrass theorem to the convergence of series. In the coming example,
we introduce one of the most important constants of Mathematics, the number e,
which will play a major role in Sect. 10.7.
P
Example 7.41 The series k0 kŠ1 converges to an irrational number e, such that
2 < e < 3. In symbols,
X 1
eD : (7.5)
k0
kŠ
Proof Let .sn /n0 be the sequence of the partial sums of the given series, i.e.,
1 1 1
sn D 1 C C CC :
1Š 2Š nŠ
For this sequence, we clearly have 1 D s0 < s1 < s2 < ; on the other hand, since
kŠ > 2k1 for every integer k > 2, we have, for an integer n 4,
Xn
1 1 1 X 1
n
8 X 1 35
sn D <1C1C C C < C D ;
kD0
kŠ 2 6 kD4
2 k1 3 k4
2 k1 12
where we used the formula for the sum of a geometric series in the last equality
above. Hence, the sequence .sn /n0 is increasing and bounded, thus, convergent.
Now, item (b) of Proposition 7.15, together with s2 D 2 and sn < 35 12
for every
integer n 4, gives 2 < e < 3 (for an approximation of e with five correct decimal
places, see Problem 8).
We shall now show that e is irrational.9 To this end, observe initially that, for
natural numbers 1 < n < m, we have
Xm
1 X 1
sm sn D <
kDnC1
kŠ knC1 kŠ
1 1 1
D 1C C C
.n C 1/Š nC2 .n C 2/.n C 3/
1 X 1 1 nC2
< D ; (7.6)
.n C 1/Š k0 .n C 2/ k .n C 1/Š n C 1
9
For the reader’s knowledge, we observe that the number e is, indeed, transcendent, i.e., it is not
the root of a polynomial with rational coefficients. A proof of this fact is beyond the scope of these
notes, and can be found in [11].
7.4 Series of Real Numbers 233
where, once more, we used the formula for the sum of a geometric series in the last
equality above.
Therefore, yet for natural numbers 1 < n < m, it follows from the above
computations that
Xm
1 1 nC2
sm D sn C < sn C ;
kDnC1
kŠ .n C 1/Š n C 1
1 nC2
e D lim sm sn C :
m!C1 .n C 1/Š n C 1
Thus,
1 nC2
sn < e sn C :
.n C 1/Š n C 1
1
Multiplying this last inequality by .n 1/Š and noticing that sn D sn1 C nŠ , we
conclude that
1 1 nC2
.n 1/Šsn1 C < .n 1/Še .n 1/Šsn1 C C ;
n n n.n C 1/2
1 nC2 1 1 2
C D C C
n n.n C 1/2 n .n C 1/2 n.n C 1/2
1 1 2 21
C 2C 2
D < 1;
3 4 34 48
for every integer n > 2, we finally arrive at the estimate
n
Theorem 7.42 e D limn!C1 1 C 1n .
n
Proof Let an D 1 C 1n . Arguing as in the proof of Example 7.23, we get
!
Xn
n 1
an D 2 C
kD2
k nk
Xn
1 n.n 1/.n 2/ : : : .n k C 1/
D 2C
kD2
kŠ nk
Xn
1 1 2 k1
D 2C 1 1 ::: 1
kD2
kŠ n n n
Xn
1 Xn
1 X 1
<2C D < D e:
kD2
kŠ kD0
kŠ k0
kŠ
Xn
1 1 2 k1
an D 2 C 1 1 ::: 1
kD2
kŠ n n n
X
nC1
1 1 2 k1
<2C 1 1 ::: 1 D anC1 :
kD2
kŠ nC1 nC1 nC1
Therefore, .an /n1 is monotone increasing and bounded above, and hence there
exists l D limn!C1 an . In particular, since an < e for every n 2 N, item (b) of
Proposition 7.15 gives l e.
Also from the computations above, given natural numbers n > m 2 we can
write
Xn
1 1 2 k1
an D 2 C 1 1 ::: 1
kD2
kŠ n n n
Xm
1 1 2 k1
>2C 1 1 ::: 1
kD2
kŠ n n n
Xm
1 1 2 m1
>2C 1 1 ::: 1
kD2
kŠ n n n
!
1 2 m1 Xm
1
> 1 1 ::: 1 2C :
n n n kD2
kŠ
7.4 Series of Real Numbers 235
Xm
1 Xm
1
D2C D :
kD2
kŠ kD0
kŠ
Pm 1
However, since m 2 N was arbitrarily chosen, we conclude that l kD0 kŠ
for every m 2 N. Then, letting m ! C1 and invoking again item (b) of
Proposition 7.15, we finally obtain
Xm
1 X 1
l lim D D e:
m!C1
kD0
kŠ k0
kŠ
t
u
Back to the general development of the theory, the next result is the analogue, for
series, of Proposition 7.18, and teaches us how to operate with convergent series.
P P
Proposition 7.43 If k1 ak and k1 bk are convergent series and c is a real
number, then:
P P P
(a) Pk1 cak converges and k1P cak D c k1 ak . P P
(b) k1 .ak C bk / converges and k1 .ak C bk / D k1 ak C k1 bk .
Proof
P
(a) If sn denotes the n-th
Ppartial sum of the series k1 ak , then the n-th partial
sum of the series P k1 cak equals csn . Hence, according to item (a) of
Proposition 7.18, k1 cak converges, and
X X
cak D lim csn D c lim sn D c ak :
n!C1 n!C1
k1 k1
P P
(b) If sn and tn are the n-th partial sums of the series P k1 ak and k1 bk ,
respectively, then the n-th partial sum of the series k1 .ak Cbk / equals sn Ctn .
Therefore, item (b) of Proposition 7.18 assures the convergence of this last
series, with
X X X
.ak C bk / D lim .sn C tn / D lim sn C lim tn D ak C bk :
n!C1 n!C1 n!C1
k1 k1 k1
t
u
236 7 More on Real Numbers
A quick analysis of the arguments presented in Examples 7.39 and 7.41 gives the
following more general result, known as the comparison test for the convergence
of series.
Proposition 7.44 Let .ak /k1 and .bk /k1 be sequences
P of positive real numbers,
such that
P k a b k for every k 1. If the series b
k1 k converges, then so does the
series k1 ak . Moreover,
X X
ak bk :
k1 k1
Pn Pn
Proof Letting sn D kD1 ak and tn D kD1 bk , it follows from 0 < ak bk
that 0 < sn tn , for every n 2 N. Since the sequence .tn /n1 converges, it is
bounded. Hence, the sequence .sn /n1 is monotonic and bounded, thus convergent,
by Bolzano-Weierstrass theorem. To what is left to prove, it suffices to make n !
C1 in the inequality sn tn and apply the result of Problem 1, page 218. t
u
Example P a sequence .ak /k1 of positive real numbers such that both
P 7.45 Is there
series k1 ak and k1 k21ak converge?
Solution Suppose there is such a series. Then, item (b) of Proposition 7.43, together
with the inequality between the arithmetic and geometric means, would give us
X X 1 X X s X2
1 1
ak C 2
D ak C 2 2 ak 2 D :
k1 k1
k ak k1
k ak k1
k ak k1
k
Therefore, by the comparison test for series, the harmonic series would be conver-
gent, which is an absurd. t
u
The coming example uses the theory of series to give a proof of the uncountabil-
ity of R.
Example 7.46 Problem 23, 172, shows that the family F of infinite subsets of N is
uncountable. On the other hand, if A D fm1 < m2 < m3 < g is such a P set, then
the comparison test, together with the convergence of the geometric series j1 21j ,
P
guarantees the convergence of the series k1 2m1k .
Let B D fn1 < n2 < n3 < g be another infinite subset of N. If we show that
X 1 X 1
¤ ;
k1
2 mk k1
2nk
P
then the correspondence A 7! k1 2m1k defines an injection from F into R, and this
guarantees that R is uncountable. (Otherwise, by composing such a function with a
bijection from R to N, we would get an injection from F to N, thus contradicting
the uncountability of F .)
7.4 Series of Real Numbers 237
What is left to do is quite similar to the proof of Example 4.12. Indeed, suppose
we had
X 1 X 1
D :
k1
2 mk k1
2nk
Then,
1 X 1 X 1 X 1 1
< D D n 1 ;
2 m 1
k1
2 m k
k1
2 nk
jn
2 j 21
1
Since .tn /n1 converges, it is a Cauchy sequence; hence, given > 0, there exists
n0 2 N such that m > n > n0 ) jtm tn j < . With these and n0 , it follows from
the above inequality that
and .sn /n1 is also a Cauchy sequence. Therefore, Theorem 7.27 guarantees the
convergence of the sequence .sn /n1 , as we wished to show. u
t
The converse to the previous proposition is not valid, namely, there are conver-
gent series which are not absolutely convergent. The classical example is a direct
application of the coming result, which is known in the mathematical literature as
the Leibniz criterion10 for the convergence of alternate series.
Proposition 7.49 (Leibniz) If .an /n1 is a P
nonincreasing sequence of positive reals
such that an ! 0, then the alternate series k1 .1/k1 ak converges.
Proof For each n 2 N, let sn D a1 Ca2 C Can . Condition a1 a2 a3 > 0
easily gives
s1 s3 s5 s6 s4 s2 : (7.7)
which clearly guarantees, in conjunction with (7.7), that .sn /n1 is a Cauchy
sequence. Therefore, .sn /n1 is convergent, as desired. t
u
P
Example 7.50 The alternate series k1 .1/k converges, by a simple application
k1
of the former proposition (we shall compute its value in Problem 8, page 484).
Nevertheless, the series formed by the absolute values of its terms (the harmonic
series) diverges.
We now discuss quite a useful criterion for the convergence of series of nonzero
real numbers. It is based on the asymptotic behavior11 of the quotient of neighboring
terms of the series, and is known as the ratio test.
ˇProposition
ˇ 7.51 Let .an /n1 be a sequence of nonzero real numbers, such that
ˇ anC1 ˇ P
ˇ an ˇ ! l. If l < 1, then the series k1 ak is absolutely convergent; if l > 1, then
P
the series k1 ak is divergent.
P
Proof Let’s prove that the series k1 ak is absolutely convergent if l < 1 (the
proof of its divergence in case l > 1 is completely analogous).
10
After Gottfried Wilhelm Leibniz, German mathematician and philosopher of the XVII century.
Together with Sir Isaac Newton, Leibniz is considered to be one of the creators of the Differential
and Integral Calculus. Up to this day, we still use some of the notations invented by Leibniz more
than 300 years ago.
11
In this context, this expression refers to the behavior of some expression that depends on n 2 N,
when n ! C1.
7.4 Series of Real Numbers 239
Letting l < 1, we can take a real number q such that l < q < 1. The convergence
janC1 j n
! l assures the existence of n0 2 N such that
jan j
janC1 j
n n0 ) q:
jan j
Hence, for n n0 , we have
Y
n1
jakC1 j
jan j D jan0 j jan0 jqnn0 :
kDn0
jak j
P
Thus, P for n n0 , the terms of the series k1 jak j are majorized by those of the
series k1 jan0 jqnn0 , which converges,P by Propositions 7.43 and 7.38. Therefore,
it follows from
P the comparison test that k1 jak j converges, which is the same as
saying that k1 ak is absolutely convergent. t
u
akC1 k2 k
D 2
! 1;
ak .k C 1/
P 1
while the series k1 k2 converges. On the positive side of things, we present the
following
Example 7.52
P Given a natural number m and a real number q > 1, explain whether
m
the series k1 .1/k1 kqk converges or diverges.
m
Solution Letting an D .1/n1 nqn for n 1, we have
ˇ ˇ
ˇ anC1 ˇ nC1 m 1 n 1
ˇ D .n C 1/
m
ˇ qn
ˇ a ˇ m D ! < 1:
n qnC1 n n q q
Therefore, by the ration test, the given series is absolutely convergent, hence,
convergent. t
u
We close this section by discussing the product of two absolutely convergent
series.
P P
Theorem 7.53 Let i1 ai and j1 bj be absolutely convergent series. If
X X
k1
ck D ai bj D ai bki (7.8)
iCjDk iD1
240 7 More on Real Numbers
P
for k 1, then k1 ck is absolutely convergent and such that
X X X
ck D ai bj :
k1 i1 j1
Proof It suffices to show that, given > 0, there exists n0 2 N such that, for n > n0 ,
we have
ˇX2n X X ˇ ˇ 2n1 X X ˇ
ˇ ˇ ˇX ˇ
ˇ ck ai bj ˇ < and ˇ ck ai bj ˇ < :
kD1 i1 j1 kD1 i1 j1
Let’s guarantee the existence of n0 2 N for which the first inequality above is true
(the analysis of the validity of the second inequality is entirely analogous).
Given n 2 N, it follows from triangle inequality that
ˇX2n X X ˇ ˇ X 2n X X
n X
n ˇ
ˇ ˇ ˇ ˇ
ˇ ck ai bj ˇ ˇ ai bj ai bj ˇ
kD1 i1 j1 kD1 iCjDk iD1 jD1
ˇ X
n X
n X X
n ˇ
ˇ ˇ
Cˇ ai bj ai bj ˇ
iD1 jD1 i1 jD1
ˇ X X
n X X ˇ
ˇ ˇ
Cˇ ai bj ai bj ˇ:
i1 jD1 i1 j1
Let A, B and C denote the first, second and third summands of the right hand side
above, respectively, so that
ˇX ˇ ˇX n ˇ ˇX ˇ ˇX ˇ
ˇ ˇ ˇ ˇ ˇ ˇ ˇ ˇ
BDˇ ai ˇ ˇ bj ˇ and C D ˇ ai ˇ ˇ bj ˇ:
i>n jD1 i1 j>n
P
n
The sequence bj
jD1 , being convergent, is bounded; therefore, there
ˇ P n1 ˇ
ˇ ˇ
exists M > 0 such that ˇ njD1 bj ˇ < M, for every n 1. On the other hand, since
P P
the series i1 ai and j1 bj converge, we have
X X X
n
n X X
ai D ai ai ! ai ai D 0
i>n i1 iD1 i1 i1
P n
and, analogously, j>nPbj ! 0. In order to estimate C, we can suppose, without
loss of generality, that i1 ai ¤ 0. Then, we can choose n1 ; n2 2 N such that
ˇX ˇ ˇX ˇ
ˇ ˇ ˇ ˇ
n > n1 ) ˇ ai ˇ < and n > n2 ) ˇ bj ˇ < ˇP ˇI
3M 3 ˇ ˇ
i>n j>n i1 ai
7.4 Series of Real Numbers 241
ˇX2n X X
n X
n ˇ ˇ X ˇ
ˇ ˇ ˇ ˇ
A Dˇ ai bj ai bj ˇ D ˇ ai bj ˇ
kD1 iCjDk iD1 jD1 maxfi;jg>n
iCj2n
X 2n X
X 2n X
X
jai bj j jai jjbj j C jai jjbj j
maxfi;jg>n iDnC1 jn jDnC1 in
iCj2n
X
2n X X
2n X
jai j jbj j C jbj j jai j :
iDnC1 j1 jDnC1 i1
P P
To estimate A, we can suppose that i1 jai j ¤ 0 and j1 jbj j ¤ 0. Now,
P P Pn
since the series i1 jai j and j1 jbj j converge, the sequences jai j n1 and
P iD1
n
jD1 jbj j are Cauchy; therefore, there exist n3 ; n4 2 N such that
n1
2n
X 2n
X
n > n3 ) jai j < P and n > n4 ) jbj j < P :
iDnC1
6 j1 jbj j jDnC1
6 i1 jai j
ˇX 2n X X ˇ
ˇ ˇ
ˇ ck ai bj ˇ A C B C C < 3 D :
kD1 i1 j1
3
t
u
242 7 More on Real Numbers
2. Sequence
P .an /n1 is a nonconstant AP of nonzero real numbers. Prove that
1
k1 ak akC1 converges and compute its sum.
P
3. Given a real number a > 1, prove that the series k1 2k1 ak
converges and
compute its sum. P 1
4. Decide whether the series k1 p p converges or diverges.
P kC k2 1
5. Prove that the series k>1000 p 3 1 2 converges.
k 1000k
6. Let .an /n1 be an infinite, nonconstant AP of positive terms. Prove that:
P 1
(a) diverges.
Pk1 a1k
(b) k1 a k converges.
2
7. (NMC) Let A be a finite set of naturals, all of the form 2a 3b 5c , for some
nonnegative integers a, b and c. Prove that
X1
< 4:
x2A
x
8. * The purpose of this problem is to show that e Š 2:71828, with five correct
decimal places. To this end, do the following items:
(a) For every integer n > 10, show that
1 1 1 1 1 1 1
C CC < 1C C 2 C C n10 :
10Š 11Š nŠ 10Š 11 11 11
item (a), together with the fact that 10Š > 2 106 , to show that 0 <
(b) Use P
e 10 1 6
kD1 kŠ < 10 .
(c) Conclude from (b) that 2:71828 approximates e with five correct decimal
places.
9. * Given a sequence .a1 ; a2 ; a3 ; : : :/ of digits, prove that there exists a single
x 2 R such that, for a given n 2 N, we have
a a2 ak 1
1
0x C C C n;
10 102 10k 10
7.4 Series of Real Numbers 243
for every natural number k n. In such a case (and as the reader is certainly
used to), we write x D 0:a1 a2 a3 : : : and say that 0:a1 a2 a3 : : : is the decimal
representation of x.
10. Show that every real number x 2 .0; 1/ admits a unique decimal expansion of
the form x D 0:a1 a2 a3 : : :, with an ¤ 0 for infinitely many values of n. Then,
use this fact to construct a surjective function f W Œ0; 1
P! Œ0; 1 Œ0; 1.
11. Let .an /n1 be a sequence of real numbers such that k1 a2k converges. Prove
P
that, for every rational ˛ > 12 , k1 ka˛k also converges. P
12. Let .an /n1 be a sequence
P ofp positive real numbers, such that the series k1 ak
converges. Prove that k1 ak akC1 also converges.
13. Let .Fn /n1 be the Fibonacci sequence, i.e., the sequence defined by F1 D 1,
F2 D 1 and FkC2 D FkC1 C Fk , for every integer k 1. Show that the series
P 1
k1 Fk converges. p
14. Let .an /n1 be a sequence of positive reals such that n an ! l.
P
(a) If l < 1, show that the series Pk1 ak converges.
(b) If l > 1, show that the series k1 ak diverges. P
(c) If l D 1, give examples showing that the series k1 ak may converge or
diverge.
convergence criterion given by the case l < 1 is known
TheP P as the root test.
15. Let k1 ak be an absolutely convergent series, with k1 ak D 0. Show that
X a1 a2 an1
2
C 2
CC 2
n1
.n 1/ .n 2/ 1
The convergence criterion for series stated in the next problem is due to N.
Abel, and is known in the mathematical literature as Abel’s convergence test
or Abel’s convergence criterion.
18. Let .an /n1 and .bn /n1 be two sequences of real numbers satisfying the
following conditions:
(a) The sequence .sn /n1 , defined by sn D a1 C C an for every n 2 N, is
bounded.
244 7 More on Real Numbers
n
(b) b1 b2 b3 > 0 and bn ! 0.
P
Prove that the series k1 ak bk converges.
19. Show that Abel’s criterion implies Leibniz criterion.
20. Do the following items:
(a) Given a; h 2 R, with h ¤ 2l for every l 2 Z, show that
.k1/h
X
k sin a C 2
sin .kC1/h
2
sin.a C jh/ D :
jD0
sin h2
P sin k
(b) Use Abel’s criterion to show that k1 k converges.
21. In a cartesian coordinate system centered at O, let .An /n1 be the sequence of
points such that A1 D .1; 0/ and:
(i) Triangle OAn AnC1 is rectangle at An and such that An AnC1 D 1.
(ii) Triangles OAnC1 AnC2 and OAn AnC1 have disjoint interiors, for every n 1.
!
Prove that, when n ! C1, half-line OAn revolves infinitely many times
around O.
22. Let Tn be a right triangle whose side lengths are 4n2 , 4n4 1 and 4n4 C 1, with
n 2 N. Let ˛n be the measure, in radians, of its internal angle opposite to the
side of length 4n2 . Show that
X
˛k D :
k1
2
Chapter 8
Continuous Functions
Let’s initially consider the function f W R ! R given by f .x/ D fxg (the fractional
part function), whose graph is sketched in Fig. 8.1. After a quick look at it, we
would certainly feel comfortable in saying that such a graph is discontinuous, for,
it presents several (actually, infinitely many) jumps. Also, note that this apparently
doesn’t happen with the graph of the function g W R ! R given by g.x/ D x2 , which
should surely be called continuous.
It does emerge the question of how to find a reasonable criterion to identify
the existence or absence of jumps in graphs, thus discerning between the two
possibilities above. In order to develop some intuition on how to do it, let’s restrict
the domain of the function x 7! fxg to the interval Œ 34 ; 32 . If we denote this new
function still by f , then we easily conclude that
1 3
Im.f / D 0; [ ;1 I
2 4
... ...
−4 −3 −2 −1 O 1 2 3 x
On the other hand, it is easy to verify that this doesn’t happen with the function g
of the first paragraph. More precisely, let’s fix an interval Œa; b R and let g.Œa; b/
denote the image of the restriction of g to Œa; b, i.e.,
In particular, for every y in the closed interval bounded by g.a/ and g.b/, there exists
p
x 2 Œa; b (x D ˙ y, according to the case at hand), such that g.x/ D y.
We are now in position to generalize the previous discussion with the following
definition. To this end, in all that follows, unless we explicitly state otherwise, I
denotes an interval and X a union of intervals of the real line.
Definition 8.1 A function f W X ! R has the intermediate value property (cf.
Fig. 8.2) if, for every interval Œa; b X and every y0 situated in the closed interval
bounded by f .a/ and f .b/, there exists x0 2 Œa; b such that y D f .x/.
The previous discussion assures that the function f .x/ D fxg, x 2 R, does not
have the intermediate value property, while the function g.x/ D x2 , x 2 R, has
such a property. Therefore, it compels us to say that, if a function f W X ! R has
the intermediate value property, then its graph should be continuous, i.e., with no
jumps. Nevertheless, the reader can easily verify that the function f W Œ0; 2 ! R,
given by
x; if 0 x 1
f .x/ D ;
x 12 ; if 1 < x 2
8.1 The Concept of Continuity 247
y
Gf
f (a)
y = y0
a x0 b x
f (b)
y
Gf
P0
a x0 b x
2δ
does satisfy the intermediate value property, albeit its graph presents a jump at
x D 1. Thus, the intermediate value property is not the correct way to formulate
the concept of continuous function.
In order to adequately define the concept of continuous function, let’s analyse the
whole situation from another point of view. Let f W .a; b/ ! R be a given function,
x0 2 .a; b/ be a fixed real number and P0 .x0 ; f .x0 // 2 Gf . Also, let x 2 .a; b/ n fx0 g
and P.x; f .x// 2 Gf . For the graph of f to be called continuous at x0 , our geometric
intuition says that, the closer x is of x0 , the closer P should be of P0 .
More precisely, this closedness means (cf. Fig. 8.3) that, for a fixed, arbitrarily
chosen error r > 0 for the position of the point P0 (i.e., given an arbitrary disk
D.P0 I r/, centered at P0 and with radius r), we should have P 2 D whenever the
abscissa x of P approximates x0 within a sufficiently small error, say less than a
certain ı > 0 (one reads delta). In symbols, given an arbitrary r > 0, there must
exist ı > 0 such that
y
Gf
P0
2
a x0 b x
2δ
Since every disk in the cartesian plane contains arbitrarily small rectangles with
sides parallel to the axis, it is easy to show (cf. Problem 4) that the validity of the
condition (8.1) is equivalent to the following alternative geometric description (cf.
Fig. 8.4): given an arbitrary error > 0 (one reads epsilon) for the value of f at x0 ,
i.e., given an arbitrary horizontal strip
R .f .x0 / ; f .x0 / C /
of the cartesian plane, symmetric with respect to the point P0 .x0 ; f .x0 // (the gray
strip in Fig. 8.4)), there must exist another error ı > 0 such that, for every x 2 X
satisfying jx x0 j < ı, the point P.x; f .x// belongs to the gray strip. This being said,
we can finally state the formal definition of continuity of a function (at a point).
Definition 8.2 A function f W X ! R is continuous at the point x0 2 X if the
following condition is satisfied: given > 0, there exists ı > 0 such that
as we wished to show.
As a last remark, the coming examples will make it clear that, along the execution
of items (i) and (ii), the function E will be generally implied.
Example 8.4 The modular function (cf. Example 6.60) is continuous.
Proof Given x0 2 R, if jx x0 j < ı, then triangle inequality gives
B
P
x
A A
O Q
P
B
p p
Note that pıx0 , ı < x0 . Therefore, taking 0 < ı x0 , we conclude
that jx x0 j < ı ) jf .x/ f .x0 /j < , as wished. t
u
In what comes next, we establish the continuity of the sine and cosine functions.
To this end, we shall need the following auxiliary result.
Lemma 8.7 For every x 2 R, we have j sin xj jxj.
Proof Firstly, let’s show that sin x x whenever 0 x 2 . This inequality is
obvious for x D 0; for 0 < x 2 , mark (cf. Fig. 8.5) point P in the first quadrant of
_
the unit circle such that `.AP/ D x.
!
Letting P0 be the symmetric of P with respect to AA0 and Q the intersection of
_ _
P0 P and A0 A, we have P0 P D 2 QP and `. P0 P/ D 2`. AP/. Since the length of every
arc of a circle is greater than that of the corresponding chord, we get
_ _
2 sin x D PP0 < `.PP0 / D 2`.AP/ D 2x:
Now, since sin.x/ D sin x, it’s immediate that j sin xj jxj for jxj 2
.
Finally, for jxj > 2 , we have
j sin xj 1 < < jxj:
2
t
u
Example 8.8 Sine and cosine functions are continuous.
Proof Fix x0 2 R. If jx x0 j < ı, it follows from the sum-to-product identities,
together with the previous lemma, that
8.1 The Concept of Continuity 251
ˇ x x ˇ ˇˇ x C x ˇˇ
ˇ 0 ˇˇ 0 ˇ
j cos x cos x0 j D 2 ˇsin ˇ ˇsin ˇ
2 2
ˇ x x ˇ ˇx x ˇ
ˇ 0 ˇ ˇ 0ˇ
2 ˇsin ˇ 2ˇ ˇ
2 2
D jx x0 j < ı:
is a finite union of open intervals. This way, we get the (well defined) rational
function
f
W R n Rg ! R:
g
The previous example, together with the discussion at the paragraph that precedes
it, assures the continuity of rational functions, wherever they are defined.
252 8 Continuous Functions
On the other hand, the continuity of f assures the existence of ı 0 > 0 such that
Therefore, it follows from the above relations (with y D f .x/ in (8.3)) that
x 2 X; jx x0 j < ı 0 ) jf .x/ f .x0 /j < ı ) jg.f .x// g.f .x0 //j < ;
is continuous. Indeed, since f is the composition of the square root function (which
x2 j sin.x2 /jC1
is continuous) with the function g W R ! R given by g.x/ D x4 Cx p
2 C2C j sin xj ,
the chain rule assures that it suffices to establish the continuity of g. Now, g is the
8.1 The Concept of Continuity 253
p
quotient of the functions x 7! x2 j sin.x2 /j C 1 and x 7! x4 C x2 C 2 C j sin xj,
which are continuous by the previous examples, together with the chain rule. Finally,
observe that both the numerator and denominator of g.x/ are positive.
Let us finish this section with a result on the set of points of discontinuity of
monotonic functions.
Proposition 8.15 If I R be an interval and f W I ! R is an nondecreasing
function. then the set of points of discontinuity of f is finite or countably infinite.
Proof Let
For each x0 2 D, let mx0 D supff .x/I x 2 I and x < x0 g and Mx0 D infff .x/I x 2
I and x > x0 g. Problem 13 guarantees that mx0 < M x0 . Now, Problem 4, page 206
guarantees that we can choose rx0 2 Q \ mx0 ; Mx0 . On the other hand, if x0 < y0
are both in D, the fact that f is nondecreasing gives
x C y
0 0
Mx0 f my 0 ;
2
so that the intervals mx0 ; Mx0 and my0 ; My0 are disjoint.
Hence, the correspondence x0 7! rx0 defines an injective function from D into Q,
and it follows from Problem 21, page 172, that D is finite or countably infinite. u t
In the notations of the statement of the previous proposition, we call the reader’s
attention to the fact that Problem 14 will guarantee that every countably infinite
subset of I is the set of points of discontinuity of a nonincreasing function.
1. Prove that:
(a) The function f W R n f0g ! R, given by f .x/ D 1x , has the intermediate
value property and is continuous, but the function f W R ! R given by
1
; if x ¤ 0
f .x/ D x
0; if x D 0
doesn’t have such property.
(b) The function f W R ! R, given by
x2 C 1; if x 0
f .x/ D
x; if x < 0
tan W D ! R
:
x 7! tan x
jx x0 j
jf .x/ f .x0 /j D p p p p
. x/
n n1 C . x/
n n2 n
x0 C C . n x0 /n1
1
< p jx x0 j:
. n x0 /n1
1
After Gustav Lejeune Dirichlet, German mathematician of the XIX century. Dirichlet made sev-
eral important contributions to Mathematics, notably to Analysis and Number Theory. Even though
the most famous of these is perhaps Dirichlet’s theorem for primes in arithmetic progressions
(about which we shall have more to say on [5]), Dirichlet made very important contributions to
the theory of Fourier Series, which we will tangentially touch upon in Chap. 11.
2
It is possible to prove that, given an interval I, there doesn’t exist a function f W I ! R continuous
at (every point of) I \ Q and discontinuous at I \ .R n Q/. A proof can be found in [1].
256 8 Continuous Functions
Proof Let’s first assume that f is continuous. Then, given a 2 I and > 0, there
exists ı > 0 such that
Now, let .an /n1 be a sequence of elements of I converging to a. Then, there exists
n0 2 N such that
1
jan aj < and jf .an / f .a/j :
n
Then, the above conditions assure that the sequence .an /n1 thus constructed
converges to a, whereas the sequence .f .an //n1 doesn’t converge to f .a/. t
u
Let us illustrate the importance of the previous result with two examples, the
first of which revisits Example 7.19 and shows that the idea used in the proof of
Example 7.23 is quite natural.
8.2 Sequential Continuity 257
p n
Example 8.17 If a > 0 and an D n
a, then an ! 1.
Proof We make the proof in the case a 1, the proof in the case 0 < a < 1 being
completely analogous. If a 1, then an anC1 1, for every n 2 N. Hence, the
sequence .an /n1 is monotonic and bounded, thus convergent.
Setting l D limn!C1 an , it follows from item (a) of Proposition 7.15 that
l 1. On the other hand, the continuity of the square root function (according
to Example 8.6) and the previous result guarantee that
n p n p n p
an ! l ) an ! l ) a2n ! l;
p
where in the last implication we used the fact that an D a2n .
Since every subsequence of a convergent sequence converges to the same limit
n
of the whole sequence, we havepa2n ! l. Therefore, the uniqueness of limits of
convergent sequences gives l D l, so that l D 1. t
u
Example 8.18 (Sweden) Find all continuous functions f W R ! R such that f .x/ C
f .x2 / D 0, for every x 2 R.
Solution If f is any function satisfying the stated conditions, we shall show that f
is identically zero.
Making x D 0 and x D 1, we get f .0/ D 0 and f .1/ D 0. Now, for x > 0 and
applying the stated condition several times, we get
p p p
f .x/ D f . x/ D f . 4 x/ D f . 8 x/ D :
p
In particular, jf . 2 x/j D jf .x/j for every n 2 N.
n
p n
Letting an D 2 x, we already know that an ! 1. On the other hand, since jf j
n
1
n ax < n C < ay < n C 1;
2
260 8 Continuous Functions
so that
ˇ ˇ ˇ ˇ ˇ ˇ
ˇd.ax/ d.ay/ˇ D ˇ.n C 1 ay/ .ax n/ˇ D ˇ2n C 1 ay axˇ
ˇ 1 1 ˇ
ˇ ˇ
D ˇ n C ay C n C ax ˇ
2 2
ˇ 1 ˇ ˇ 1 ˇ
ˇ ˇ ˇ ˇ
ˇn C ayˇ C ˇn C axˇ
2 2
D jay axj D ajx yj:
For the last part, given > 0 take ı D minf a1 ; a g > 0. Then,
1 ˇ ˇ
jx yj < ı ) jx yj < ) ˇd.ax/ d.ay/ˇ ajx yj < a D :
a a
t
u
The coming result proves that, for a function defined on a closed and bounded
interval, the concepts of continuity and uniform continuity coincide.
Theorem 8.24 Every continuous function f W Œa; b ! R is uniformly continuous.
Proof By contradiction, suppose that f is continuous but (8.4) is not valid. Then,
there exists > 0 such that, for every ı > 0, we can find x; y 2 Œa; b satisfying
1
jxn yn j < and jf .xn / f .yn /j : (8.5)
n
Since .xn /n1 is a sequence in Œa; b, Weierstrass Theorem 7.25 guarantees the
existence of a subsequence .xnk /k1 of .xn /n1 , converging to some x0 2 Œa; b.
Then, it follows from triangle inequality that
1 k
jynk x0 j jynk xnk j C jxnk x0 j < C jxnk x0 j ! 0:
nk
Hence, .ynk /k1 also converges to x0 , and Theorem 8.16, together with the continuity
of x 7! jxj, gives
Now, choose real numbers a D x0 < x1 < < xk D b satisfying xi xi1 < ı
for 1 i k, and let
Finally, since x 2 Œa; b as in the last paragraph was arbitrarily chosen, we get
jf .x/j M C 1 for every x 2 Œa; b, so that f is bounded. u
t
The last property of continuous functions we shall study in this section is also
due to Weierstrass, and states that every continuous function f W Œa; b ! R assumes
extreme values in the interval Œa; b.
Theorem 8.26 (Weierstrass) If f W Œa; b ! R is continuous, then there exist
xm ; xM 2 Œa; b such that
f .xm / D minff .x/I x 2 Œa; bg and f .xM / D maxff .x/I x 2 Œa; bg:
Proof Let’s prove the existence of xm , that of xM being totally analogous. Since f is
bounded, it makes sense to let
The definition of greatest lower bound assures (with the aid of Problem 10,
page 207) the existence of a sequence .xn /n1 in Œa; b such that f .xn / ! m.
Since every bounded sequence possesses a convergent subsequence, we can take
a subsequence .xnk /k1 of .xn /n1 , converging to a real number x0 2 Œa; b. Then,
Theorem 8.16 gives
On the other hand, since the sequence .f .xnk //k1 is a subsequence of .f .xn //n1
(which, in turn, converges to m), we conclude from Proposition 7.15, that .f .xnk //k1
also converges to m. Therefore, the uniqueness of the limit of sequences gives
f .x0 / D m. t
u
for all x; y 2 R, so that both sine and cosine functions are lipschitzian, with
Lipschitz constant equal to 1. Prove that every lipschitzian function f W I ! R
is uniformly continuous.
4. For an integer n > 1, show that the function f W R ! R, given by f .x/ D xn , is
not uniformly continuous.
5. Let I R be an interval and f W I ! R be a continuous function. Suppose
that there exist > 0 and sequences .an /n1 and .bn /n1 in I, such that jan
n
bn j ! 0, but jf .an / f .bn /j for every n 1. Show that f is not uniformly
continuous.
6. Show that the function f W .0; C1/ ! R, given by f .x/ D sin 1x , is not
uniformly continuous.
7. The purpose of this problem is to present another proof of Theorem 8.26. To
this end, recall that Corollary 8.25 assures the boundedness of a continuous
function f W Œa; b ! R, so that there exist
3
It can be shown that the hypothesis of continuity is essential here. More precisely, there exists
infinitely many functions f W R ! R such that f .x C y/ D f .x/ C f .y/ for all x; y 2 R but f is not
linear. A proof of this fact is beyond the scope of these notes and can be found in [13].
4
After Rudolf Lipschitz, German mathematician of the XIX century.
8.2 Sequential Continuity 263
d.PI Gf / D inff A0 PI A0 2 Gf g:
If jy0 f .x0 /j < minfjx0 aj; jx0 bjg, prove that there exists A 2 Gf such that
d.PI Gf / D AP.
10. * Let f W R ! R be the polynomial function given by
5
After Stefan Banach, Polish mathematician of the XX century.
264 8 Continuous Functions
f .x/ D f .x C 1/ D f .x C ˛/;
for every x 2 R.
f .x/ an1 a1 a0
n
D an C C C n1 C n
x x x x
ˇa a0 ˇˇ
ˇ n1 a1
an ˇ C C n1 C n ˇ
x x x
ˇa ˇ ˇ a ˇ ˇa ˇ
ˇ n1 ˇ ˇ 1 ˇ ˇ 0ˇ
an ˇ ˇ ˇ n1 ˇ ˇ n ˇ
x x x
jan1 j ja1 j ja0 j
D an n1 n :
jxj jxj jxj
1 X
n1
f .x/
an jaj jI
x n jxj jD0
1 Pn1
in turn, this last expression is positive for jxj >jD0 jaj j.
an
The argument at the previous paragraph shows that f x.x/
n > 0 at x D ˙A, whenever
n o
1 Pn1
A > max 1; an jD0 jaj j . However, since n is odd, it follows that f .A/ < 0 <
f .A/, and Bolzano’s theorem assures the existence of c 2 ŒA; A such that f .c/ D 0.
t
u
Our next result is a refinement of Bolzano’s theorem, which shows that every
continuous function defined on an interval satisfies the intermediate value property.
For this reason, this result is known as the intermediate value theorem (we
abbreviate IVT).
Theorem 8.29 (IVT) Let f ; g W Œa; b ! R be continuous functions. If f .a/ < g.a/
and f .b/ > g.b/ (or vice-versa), then there exists c 2 .a; b/ such that f .c/ D g.c/.
In particular, if a real number d belongs to the interval with endpoints f .a/ and f .b/,
then there exists c 2 Œa; b such that f .c/ D d.
266 8 Continuous Functions
Proof For the first claim, note that h D f g is continuous and such that h.a/h.b/ D
.f .a/g.a//.f .b/g.b// < 0. Hence, Bolzano’s theorem gives c 2 .a; b/ for which
h.c/ D 0, i.e., such that f .c/ D g.c/.
The particular case of the second claim is obtained by taking g to be the constant
function, equal to d. t
u
As a first application of the IVT, let us show that the image of a continuous
function defined on an interval is also an interval.
Corollary 8.30 If I R is an interval and f W I ! R is a continuous function, then
the image of f is also an interval. If I D Œa; b, then there exist real numbers c d
such that Im.f / D Œc; d.
Proof Let’s first consider the case I D Œa; b. By Weierstrass Theorem 8.26, there
exist xm ; xM 2 Œa; b such that f attains its minimum and maximum values at xm and
xM , respectively. Letting f .xm / D c and f .xM / D d, we have Im.f / Œc; d. On the
other hand, for a fixed y 2 Œc; d, the IVT gives a real x, belonging to the interval of
endpoints xm and xM and such that f .x/ D y. In particular, Im.f /
Œc;S d.
For the general case, note that every interval I can be written as I D n1 Œan ; bn ,
with Œa1 ; b1 Œa2 ; b2 : : : (for instance, if I D .a; b/, we could take an D a C 1n
and bn D b 1n , provided n > 2.ba/ 1
—the remaining cases can be dealt with in
analogous ways). Now, letting Œcn ; dn be the image of the interval S Œan ; bn by f ,
it’s easy to show that Œc1 ; d1 Œc2 ; d2 : : : and, hence, that S n1 Œcn ; dn is an
interval J. On the other hand, it’s also easy to see that, since I D n1 Œan ; bn , we
have
[
Im.f / D Œcn ; dn D J:
n1
t
u
In what follows, we discuss some other interesting applications of the IVT.
Example 8.31 Let f W Œ0; 1 ! Œ0; 1 be a continuous function. Prove that there
exists a real number 0 c 1 such that f .c/ D c (i.e., prove that f has at least one
fixed point).
Proof If f .0/ D 0 or f .1/ D 1, there’s nothing left to do; otherwise, suppose f .0/ >
0 and f .1/ < 1. Letting g W Œ0; 1 ! R be the function given by g.x/ D x, we then
have f .a/ > g.a/ and f .b/ < g.b/. Hence, the IVT gives 0 < c < 1 satisfying
f .c/ D g.c/ or, which is the same, f .c/ D c. t
u
Example 8.32 (Bulgaria) Let m; n 1 be given integers. For x > 0, compute the
number of real solutions of the equation
1 1 1 1 p p p
C 2 C 3 C C m D x C x C 3 x C C n x:
x x x x
Solution We claim that the given equation has just one real solution. To this end,
note first that, for x > 0 and a natural k, the function x 7! x1k is decreasing, whereas
8.3 The Intermediate Value Theorem 267
p
the function x 7! k x is increasing. Now, since a finite sum of increasing functions
on the same domain is increasing and a finite sum of decreasing functions on the
same domain is decreasing (prove these facts!), we conclude that the functions f ; g W
.0; C1/ ! R given by
1 1 1 1 p p p
f .x/ D C 2 C 3 C C m and g.x/ D x C x C 3 x C C n x
x x x x
are decreasing and increasing, respectively. On the other hand, since the solutions
of the given equation correspond to the positive values of x for which f .x/ D g.x/,
Problem 1, page 183, guarantees that such equation has at most one real solution.
In order to show that a real solution indeed exists, let’s use the IVT, initially
observing that both f and g are continuous functions. We then consider two cases
separately:
• for x > 1, we have
1 1 1 1 p p p
< m1 < < 2 < and x > x > > n1 x > n x;
xm x x x
p p
so that f .x/ < mx and g.x/ > n n x. In particular, f .x/ < g.x/ if mx < n n x, i.e., if
n n n o
x > mn nC1 . Hence, f .x/ < g.x/ for x > max 1; mn nC1 .
• for 0 < x < 1, we have
1 1 1 1 p p p
m
> m1 > > 2 > and x < x < < n1 x < n x;
x x x x
p p
so that f .x/ > x and g.x/ < n x. In particular, f .x/ > g.x/ if mx > n n x, i.e., if
m n
n n n o
x < mn nC1 . Hence, f .x/ > g.x/ for 0 < x < min 1; mn nC1 .
Finally, the previous discussion assures that we can take real numbers 0 < a <
1 < b satisfying f .a/ > g.a/ and f .b/ < g.b/, and the IVT provides c 2 .a; b/ such
that f .c/ D g.c/. t
u
Example 8.33 (Romania) There exists a continuous function f W R ! R such that
f .x/ 2 Q , f .x C 1/ … Q‹
We now assert that there exists x0 2 R such that f .x0 / 2 Q. Indeed, take any
real number a; if f .a/ is not rational, it follows from the hypotheses that f .a C 1/ is
rational, and it suffices to take x0 D a or x0 D a C 1.
Finally, for x0 as in the former paragraph, the hypotheses on f guarantee that
Hence, f .x0 C 2/ f .x0 / 2 Q, which contradicts (8.9) and finishes the proof. t
u
For the coming example, recall that a function f W R ! R is even if f .x/ D f .x/,
for every real x.
Example 8.34 Let f W R ! R be a continuous function such that f .f .x// D x2 C 1,
for every x 2 R. Prove that f is even.
Proof First of all, note that for every x 2 R we have
so that ˛ D ˙ˇ and f has at most two zeros. Now, there are three possibilities:
• f .x/ ¤ 0, for every real x: by the IVT, f has a constant sign in R. However, since
f .x/ D ˙f .x/ for every real x, it follows that f .x/ D f .x/ for every real x, and
f is even.
• f has a single zero, say at x D ˛: since f .˛/ D ˙f .˛/ D 0, we must have
˛ D ˛, so that ˛ D 0. On the other hand,
f .f .0// D 02 C 1 D 1 ) f .0/ ¤ 0
which is a contradiction.
• f has exactly two zeros, at x D ˛ and x D ˛, for some ˛ > 0: again by the IVT,
f has a constant sign in .˛; ˛/. Let g.x/ D f .x/ x. If f > 0 in .˛; ˛/, then
so that there exists 0 < c < ˛ for which g.c/ D 0. If f < 0 in .˛; ˛/, then
so that there exists ˛ < c < 0 for which g.c/ D 0. In any case, f admits a fixed
point c. However,
f .c/ D c ) c D f .f .c// D c2 C 1 ) c2 c C 1 D 0;
which is impossible. t
u
We close this section by using the IVT to prove the continuity of the inverse of a
continuous function defined on an interval.
Theorem 8.35 Let I R be an interval and f W I ! R be a continuous
function. Then, f is injective if and only if f is increasing or decreasing. Moreover,
in this case:
(a) The image J of f is an interval of the same type (i.e., open, half-open or closed)
as I.
(b) f 1 W J ! I is continuous.
Proof If f is not injective, then f is neither increasing nor decreasing. Conversely,
if f is neither increasing nor decreasing, then there exist a < b < c in I such that
f .a/ f .b/ f .c/ or f .a/ f .b/ f .c/. Suppose that f .a/ f .b/ f .c/ (the
other case is entirely analogous) and choose d 2 R such that
(b) Let’s look at the case of an increasing f , with I D .a; b/ and Im.f / D .c; d/
(once more, the analysis of the other ones is quite similar). We start by observing
(according to Problem 8, page 176), that f 1 is also increasing. Now, for a fixed
y0 2 .c; d/, let x0 D f 1 .y0 /. Given > 0, we want ı > 0 such that
To this end, let ı0 D minfy0 c; d y0 g and take 0 < ı < ı0 (so that the
condition jy y0 j < ı suffices to guarantee that y 2 .c; d/). Letting x D f 1 .y/,
we have y D f .x/ and can rewrite (8.10) in the following way: we want 0 < ı <
ı0 such that
Notice that we can pick > 0 so small that x0 ˙ 2 .a; b/ (otherwise, change the
originally given > 0 by a smaller one, so that this additional condition is satisfied).
Recalling that f is increasing, take
Then,
and, analogously,
as we wished to show. t
u
As a first application of the previous result, we shall now give an alternative proof
for Theorem 7.9 and Problem 8, page 254.
Example 8.36 Given n 2 N, the n-th power function is f W R ! R given by
f .x/ D xn . If n is even, write also f to denote its restriction to Œ0 C 1/. Since
f is continuous, increasing and f .k/ D kn > k for every k 2 N, the IVT gives
Im.f / D Œ0; C1/. Hence, for every y 0 there exists a single x 0 such that
p
xn D y, and we denote by x D n y. The previous theorem now guarantees the
continuity of the function
f 1 W R ! R
p
y 7! n y
arcsin x D y , sin y D x:
arccos x D y , cos y D x:
Finally, the arc-tangent function arctan W R ! 2 ; 2 is the inverse of the
restriction tan W 2 ; 2 ! R of the tangent
function,
so that it is also continuous
and increasing. Also, for x 2 R and y 2 2 ; 2 , we have
arctan x D y , tan y D x:
satisfies f .0/ < 32 < f .2/ but there doesn’t exist any c in the domain of f such
that f .c/ D 32 . Why does this example do not contradict the IVT?
2. Compute the number of real solutions of each of the following equations:
(a) jxj C 1 D x4 .
272 8 Continuous Functions
(b) cos x D x2 .
(c) sin x D 4x .
3. Let f W R ! R be given by f .x/ D x3 sin x, for every x 2 R. Show that its graph
intersects every non vertical real line an infinite number of times.
4. Let n > 1 be an integer. Write down an explicit expression for a continuous
function f W Œ0; 1 ! Œ0; 1 with exactly n fixed points.
5. * Let I be an interval and X be a nonempty subset of I having the following
properties:
(i) For every x0 2 X, there exists ı > 0 such that I \ .x0 ı; x0 C ı/ X.
(ii) If .an /n1 is a sequence of points of X and l 2 I is such that an ! l, then
l 2 X.
Show that X D I.
6. (Bulgaria) Let n > 1 be an integer and a1 , a2 , . . . , an be given positive real
numbers. Prove that the equation
p p p
1 C a1 x C 1 C a2 x C C 1 C an x D nx
15. (TT) Prove that, for each natural number n, the graph of any continuous
increasing function f W Œ0; 1 ! Œ0; 1 can be covered by n closed rectangles,
each of which having area n12 and sides parallel to the coordinate axis.
16. (Leningrad) Let f W R ! R be a continuous function such that, for every real x,
we have f .x C f .x// D f .x/. Prove that f is constant.
17. (Belarus) Find all functions f ; g; h W R ! R such that, for every x; y 2 R, we
have
f .x C y3 / C g.x3 C y/ D h.xy/:
Chapter 9
Limits and Derivatives
!
Fig. 9.1 Secants AB1 and y B2
! B1
AB2 to the graph of f
f A
a x0 x2 x1 b x
on the other hand, if we can define the line tangent to the graph of f at A, and it is
not vertical, then its equation must be of the form
for some m 2 R.
Comparing (9.1) and (9.2), and taking into account the discussion at the
previous paragraph, we are led to the conclusion that the quotients f .xx11/f x0
.x0 /
should
approximate m better and better, provided x1 approximates x0 better and better. Thus,
we conclude that the line tangent to the graph of f at A is that of Eq. (9.2), where m
is the limit value of the quotients f .xx11/f
x0
.x0 /
, when x1 comes closer and closer to x0 .
This limit value of the quotients f .xx11/f
x0
.x0 /
when x1 approaches x0 is the derivative
of the function f at x0 , which we denote f 0 .x0 / or, in classical notation, dx df
.x0 /.
Writing x in place of x1 , we summarize the previous discussion by “defining”
df f .x/ f .x0 /
.x0 / D lim ;
dx x!x0 x x0
f .x/f .x0 /
where the notation lim means that the closer x is to x0 , the closer xx0
is
x!x0
to df
.x /.
dx 0
Hence, for small values of jx x0 j, we expect to have
df f .x/ f .x0 /
.x0 / Š :
dx x x0
df f ˇˇx
.x0 / Š ˇ :
dx x x0
9.2 Limits of Functions 277
Now, suppose we are in a situation where x represents time and f represents some
quantity that evolves with time. Writing t in place of x and t0 in place of x0 , it follows
from the above that
df f ˇˇt
.t0 / D lim ˇ ;
dt t!t0 t t0
so that dfdt .t0 / can naturally be called the instantaneous rate of change of f with
respect to t, at instant t0 .
As a particular situation of that of the previous paragraph, if f .t/ measures
the displacement of a point object along a real line at instant t, then dfdt .t0 / is the
instantaneous velocity of the point object at instant t0 , and is the starting point of the
discussion of Kinematics at most Physics books.
The circle of ideas we developed up to this point, and much more, bears the
name of Differential Calculus and is considered to be the creation of two of the
most brilliant minds mankind ever produced, Newton and Leibniz.1
The rest of this chapter is devoted to a thorough development of the concepts,
results and applications of the Differential Calculus.
Let be given a continuous function f W .a; b/ ! R and x0 2 .a; b/. Recalling our
intuitive discussion so far, we conclude that if there exists a reasonable notion of
tangent line to the graph of f at the point A.x0 ; f .x0 //, and if such a line is not
vertical, then its equation must have the form (9.2), where
f .x/ f .x0 /
m D lim : (9.3)
x!x0 x x0
Here, the expression at the right hand side above intuitively represents the limit value
of the quotients f .x/f
xx0
.x0 /
as x goes to (i.e., approaches) x0 , provided such a “limit”
exists in a proper sense (to be precised in a while).
Since our discussion rests upon rather naive grounds, we start to fix this now by
setting some preliminaries and giving a rigorous definition of the notion of limit of
a function at a point.
Definition 9.1 Given a 2 R, a neighborhood of a is an interval of the form .a
r; a C r/, where r is a positive real number. In this case, we say that r is the radius
of the neighborhood .a r; a C r/ and that each x 2 .a r; a C r/ approximates a
with error less than r.
1
Gottfried Wilhem Leibniz, German mathematician and philosopher of the XVII century. Together
with Sir Isaac Newton, Leibniz is considered to be one of the creators of the Differential and
Integral Calculus. Some of the notations used in Calculus up to this day go back to Leibniz, having
survived the ruthless test of time.
278 9 Limits and Derivatives
we can summarize the previous discussion in the following definition (see Fig. 9.2).
Definition 9.2 Let I R be an interval, let x0 2 I and f W I n fx0 g ! R be a given
function. We say that f has limit L when x goes to x0 , and write
if, to each given real number > 0, there corresponds a real number ı > 0 such that
In words, (9.5) happens if, to each arbitrarily given error > 0 for L, there
corresponds an error ı > 0 for x0 such that, if x 2 I n fx0 g approximates x0 in I with
error less that ı, then f .x/ approximates L with error less than .
Geometrically, we want that, to each x 2 I sufficiently close to (but different
from) x0 , the point of the graph of f with abscissa x belongs to the gray strip
of Fig. 9.2.
L 2
a x0 b x
2δ
Gf
as wished.
(b) limx!3 x2 D 9: again, let > 0 be given. Starting with x 2 R such that 0 <
jx 3j < ı, we have
where we used the triangle inequality in the next to last passage above. Hence,
if it is possible to choose ı > 0 so that ı.ı C 6/ , we will have
key to find out which ı > 0 works for a given > 0 is to reason in the following
way: starting with an x 2 I such that 0 < jx x0 j < ı, we estimate from above the
error jf .x/Lj in terms of ı, thus getting an inequality of the form jf .x/Lj < E.ı/.
Here, E represents a certain function of ı; in item (a) we got E.ı/ D 2ı, whereas
in item (b) we found E.ı/ D ı.ı C 6/. Then, we impose that such an error E.ı/
does not surpass , thus finding the appropriate values of ı (usually, this second step
reduces to solving, for ı > 0, the inequality E.ı/ ). Finally, if ı > 0 satisfies
E.ı/ , then we clearly have that
as wished.
Notice that the only difference between the discussion above and that of Sect. 8.1
lies in the fact that, here, we have to start by assuming jx x0 j > 0, i.e., x ¤ x0 . On
the other hand, and as the coming proposition explains, such a difference was not
relevant to the previous examples because both functions x 7! 2x C 7 and x 7! x2
are continuous.
Proposition 9.4 Let I R be an interval and f W I ! R be a given function. For
x0 2 I, we have f continuous at x0 if and only if
Proof First of all, let f be continuous at x0 . Definition 8.2 guarantees that, given
> 0, there exists ı > 0 such that
However, since the condition jf .x/ f .x0 /j < is trivially satisfied for x D x0 , we
can certainly write the implication above as
Hence, f is continuous at x0 . t
u
x3 3x2 C2
Example 9.5 Given f W R n f˙1g ! R such that f .x/ D x2 1 for every x ¤ ˙1,
we want to compute limx!1 f .x/. To this end, we initially observe that 1 is a root
of both the numerator and the denominator of the defining expression for f .x/, with
9.2 Limits of Functions 281
.x 1/.x2 2x 2/ x2 2x 2
f .x/ D D :
.x 1/.x C 1/ xC1
x2 2x2
Now, since the function g W R n f1g ! R given by g.x/ D xC1 is continuous,
the previous proposition gives
3
lim f .x/ D lim g.x/ D g.1/ D :
x!1 x!1 2
Proof Suppose L > 0 (the other case is completely analogous). By the definition of
limit, given D L2 > 0 there exists ı > 0 such that
L
x 2 I; 0 < jx x0 j < ı ) jf .x/ Lj < :
2
Thus, in order that j.f C g/.x/ .L C M/j < for x 2 I close to (but different
from) x0 , it suffices for us to have jf .x/ Lj < 2 and jg.x/ Mj < 2 .
Since 2 > 0 and limx!x0 f .x/ D L, limx!x0 g.x/ D M, the definition of limit
guarantees the existence of ı1 ; ı2 > 0 such that
x 2 I; 0 < jx x0 j < ı1 ) jf .x/ Lj <
2
and
x 2 I; 0 < jx x0 j < ı2 ) jg.x/ Mj < :
2
Therefore, letting ı D minfı1 ; ı2 g > 0 and taking x 2 I such that 0 < jx x0 j <
ı, we get
(b) Here again, we estimate j.fg/.x/ LMj by excess in terms of jf .x/ Lj and
jg.x/ Mj. To this end, it follows from triangle inequality that
Therefore, in order that j.fg/.x/LMj < for x 2 I close to (but different from)
x0 , it suffices for us to have jf .x/ Ljjg.x/ Mj, jLjjg.x/ Mj and jMjjf .x/ Lj
all less than 3 . In turn, these inequalities hold provided we have
9.2 Limits of Functions 283
r
jf .x/ Lj; jg.x/ Mj < ;
3
jg.x/ Mj < and jf .x/ Lj < :
3.jLj C 1/ 3.jMj C 1/
and
r
jg.x/ Mj < min ; :
3 3.jLj C 1/
np o np o
Now, letting 1 D min ;
3 3.jMjC1/ > 0 and 2 D min 3 3.jLjC1/ >
;
0, the definition of limit guarantees the existence of errors ı1 ; ı2 > 0 such that
jf .x/ Lj < 1 for every x 2 I such that 0 < jx x0 j < ı1 , and jg.x/ Mj < 2
for every x 2 I such that 0 < jx x0 j < ı2 . Hence, setting ı D minfı1 ; ı2 g > 0,
we conclude that x 2 I and 0 < jx x0 j < ı imply both jf .x/ Lj < 1 and
jg.x/ Mj < 2 , as needed.
(c) The sign-preserving lemma gives ı0 > 0 such that jg.x/j > jMj 2
for all x 2 I
satisfying 0 < jx x0 j < ı0 . Sticking to the notations set forth in the paragraph
that immediately precedes this proposition, we look at gf as the function gf W
J n fx0 g ! R, where J D I \ .x0 ı0 ; x0 C ı0 /.
Since f
g D f 1g , by item (b) it suffices to show that limx!x0 1
g.x/ D 1
M. To this
jMj
end, and taking into account that jg.x/j > for x 2 J n fx0 g, we obtain (for such
2
an x)
ˇ ˇ
ˇ 1 1 ˇˇ jg.x/ Mj 2
ˇ
ˇ g.x/ M ˇ D jg.x/jjMj M 2 jg.x/ Mj:
ˇ ˇ
ˇ 1 ˇ 2
Therefore, in order that ˇ g.x/ M1 ˇ < , we need to have jg.x/ Mj < M2 . To fulfill
the last condition above, we choose (by invoking once more the definition of limit)
ı1 > 0 for which
M2
x 2 J; 0 < jx x0 j < ı1 ) jg.x/ Mj < :
2
Finally, letting ı D minfı0 ; ı1 g > 0 and taking x 2 J such that 0 < jx x0 j < ı,
M2
we have jg.x/j < jMj2 and jg.x/ Mj < 2 , as needed. t
u
284 9 Limits and Derivatives
An easy induction allows us to extend the formulas of items (a) and (b) of the
previous proposition to a finite number of functions. More specifically, if I is an
interval, x0 2 I and f1 ; : : : ; fn W I n fx0 g ! R are such that limx!x0 fj .x/ D Lj for
1 j n, then
and
lim f .x/k D Lk :
x!x0
From now on, we assume the validity of these remarks without further comments.
Our next result is usually referred to as the squeezing theorem. As we shall see
right after its proof, it is quite useful for the actual computation of limits.
Proposition 9.8 Let I be an interval, x0 2 I and f ; g; h W I n fx0 g ! R be such
that g.x/ belongs to the interval of ends f .x/ and h.x/ for every x 2 I n fx0 g. If
limx!x0 f .x/ D limx!x0 h.x/ D L, then limx!x0 g.x/ exists and is also equal to L.
Proof Given > 0, we want to find ı > 0 for which the conditions x 2 I and
0 < jx x0 j < ı imply jg.x/ Lj < . To this end, if f .x/ g.x/ h.x/, then
f .x/ L g.x/ L h.x/ L and is easy to conclude that
If h.x/ g.x/ f .x/, one reaches the inequality above in essentially the same way.
Now, the definition of limit gives ı1 ; ı2 > 0 such that
and
Therefore, letting ı D minfı1 ; ı2 g > 0 and taking x 2 I such that 0 < jx x0 j < ı,
we get jf .x/ Lj < and jh.x/ Lj < , so that
t
u
Concerning the computation of limits, the coming corollary is quite a useful tool.
For its statement, recall that a function f W I ! R is bounded if there exists M > 0
such that
9.2 Limits of Functions 285
jf .x/j M; 8 x 2 I:
Proof Since we wish to compute a limit, we can restrict ourselves to the interval
_ _
jxj < 2 . Suppose first that x > 0. Letting `.AB/ D x be the length of the arc AB
(cf. Fig. 9.3), we have
_
sin x D BD < AB < `.AB/ D x;
so that sinx x < 1. On the other hand, it is well known (cf. Chap. 5 of [5], for instance)
that the area of the circular sector AOB equals 2 x
D 2x , whereas the area of the
O D A
286 9 Limits and Derivatives
_
x D `.AB/ < AC D tan x;
sin x
cos x < <1 (9.9)
x
for 0 < x < 2 . However, since both x 7! cos x and x 7! sinx x are even functions,
these inequalities still hold true for 2 < x < 0. Then, it follows from (9.9),
together with the squeezing theorem and the continuity of the cosine function, that
limx!0 sinx x exists and equals 1. t
u
Remark 9.12 The attentive reader may object the former proof under the following
argument: the very definitions of the sine and cosine functions, as given in
elementary school, rely upon the fact that a circle of radius 1 has length 2; in
turn, as we shall see in Sect. 10.10, a thorough computation of length of a circle
involves the use of the derivative of the sine function, so that we are in presence of
a circular reasoning. Although this is indeed the case, we observe that we will fix
this situation in Sect. 11.5, when we shall construct the sine and cosine functions
without invoking any concepts or results of Euclidean Geometry. In particular, at
that future time we will show, by purely analytical means, that limx!0 sinx x D 1 and
sin0 x D cos x. For the time being, we suggest that the reader takes the proof of the
previous lemma simply as a heuristic argument for justifying the validity of both
these results.
Let be given an open interval I, a point x0 2 I and a function f W I n fx0 g ! R.
In a way entirely analogous to what we have done so far, we can define for f the
concept of one-sided or lateral limits at x0 . More precisely:
(i) If I \ .1; x0 / ¤ ;, we say that L is the left-handed limit of f as x goes to x0 ,
and write limx!x0 f .x/ D L, if, for each given > 0, there exists ı > 0 such
that
L 2
a x0 b x
δ
Gf
limits, for x approaches x0 from the left (resp. from the right), i.e., through values
smaller (resp. larger) than x0 itself. In yet another way, when x ! x0 (resp. x !
x0 C), we have x D x0 .something positive/ (resp. x D x0 C .something positive/).
Figure 9.4 gives a geometric interpretation for the notion of right-handed limit.
Note that, for 0 < jx x0 j < ı, the point .x; f .x// only belongs to the gray strip if
x0 < x < x0 C ı. We left to the reader the task of drawing a figure that provides the
analogous geometric interpretation for left-handed limits.
Yet with respect to one-sided limits, it is immediate to state and prove for them
results analogous to those of Proposition 9.7 (cf. Problem 4). We also stress that,
for a function f W .a; x0 / ! R, the notions of limite and left-handed limit of
f at x0 coincide (for, there is simply no way of approaching x0 from the right,
while belonging to the domain of f ). Thus, in such a case we shall generally write
limx!x0 f .x/ D L instead of limx!x0 f .x/ D L. Obviously, for a given function
f W .x0 ; b/ ! R, analogous remarks are true for limits and right-handed limits at x0 .
In this respect, see also Problem 5.
We close this section by defining the notions of infinite limits and limits at
infinity, and sketching some elementary (though useful) results on them.
Definition 9.13 Let be given an interval I, a point x0 2 I and a function f W I n
fx0 g ! R. We write
lim f .x/ D C1
x!x0
if, for any given M > 0, there exists ı > 0 such that
We leave to the reader the task of stating definitions analogous to the one above
for the concepts of limx!x0 f .x/ D 1 and limx!x0 ˙ f .x/ D ˙1.
288 9 Limits and Derivatives
The next result establishes two simple, yet useful, arithmetic rules for infinite
limits. Since its proof resembles that of Proposition 9.7, we also leave it as an
exercise for the reader.
Proposition 9.14 Let be given an interval I, a point x0 2 I and functions f ; g W
I n fx0 g ! R.
(a) If lim f .x/ D L > 0 and lim g.x/ D ˙1, then lim f .x/g.x/ D ˙1.
x!x0 x!x0 x!x0
(b) If lim f .x/ D L < 0 and lim g.x/ D ˙1, then lim f .x/g.x/ D 1.
x!x0 x!x0 x!x0
The prototype of infinite limits are those collected in the coming example, whose
proof is also an easy exercise for the reader.
Example 9.15
1
(a) If f W .x0 ; C1/ ! R is given by f .x/ D xx 0
, then limx!x0 f .x/ D C1.
1
(b) If f W .1; x0 / ! R is given by f .x/ D xx0 , then limx!x0 f .x/ D 1.
Let’s now turn our attention to limits at infinity.
Definition 9.16 Given f W .a; C1/ ! R and L 2 R, we write limx!C1 f .x/ D L
if, for a given > 0, there exists A > a such that
write
an1 a1 a0
f .x/ D xn an C C C n1 C n
x x x
and, then, apply Proposition 9.14 (with ˙1 in place of x0 ). t
u
9.2 Limits of Functions 289
One of the most important applications of infinite limits and limits at infinity is in
identifying horizontal and vertical asymptotes of (the graph of) a function, according
to the coming
Definition 9.18 If I is an interval, x0 2 I and f W I n fx0 g ! R is a function such
that limx!x0 f .x/ D ˙1, then we say that line x D x0 is a vertical asymptote of
(the graph of) f . Analogously, if f W .a; C1/ ! R (resp. f W .1; b/ ! R) is such
that limx!C1 f .x/ D L (resp. limx!1 f .x/ D L), then we say that line y D L is a
horizontal asymptote of (the graph of) f .
Examples 9.19
(a) Figure 6.17 sketches the graph of the tangent function in the interval . 2 ; 2 /.
In view of the limits limx! 2 sin x D 1 and limx! 2 cos x D 0, it follows from
Proposition 9.14 that limx! 2 tan x D C1. Analogously, limx! 2 tan x D
1, so that lines x D ˙ 2 are vertical asymptotes of this graph.
2x
(b) Figure 9.5 sketches the graph of f W R n f1g ! R, given by f .x/ D xC1 .
This graph can be obtained from that of the inverse proportionality function (cf.
Fig. 6.14) with the aid of Problem 10, page 194. Indeed, since
2x 2x C 2 2 2
D D2 ; (9.12)
xC1 xC1 xC1
(0, 2)
(−1, 0)
O x
2x
Fig. 9.5 Graph of x 7! xC1
290 9 Limits and Derivatives
we can sketch the graph of f in the following way: firstly, we sketch the graph
of x 7! 1x ; then, we translate it one unit to the left, thus getting the graph of
1
x 7! xC1 ; thirdly, we stretch this second graph in the vertical direction by factor
2
2, obtaining the graph of x 7! xC1 ; we then reflect the result along the horizontal
2
axis, arriving at the graph of x 7! xC1 ; we end by translating this last graph
2x
two units above, thus finally obtaining the graph of x 7! xC1 .
1
Since limx!1˙ 2x D 2 and limx!1˙ xC1 D ˙1, it follows from
2x
Proposition 9.14 that limx!1˙ xC1 D 1, so that line x D 1 is indeed a
vertical asymptote of the graph. On the other hand, it easily follows from (9.12)
2x
that limx!˙1 xC1 D 2, so that line y D 2 is a horizontal asymptote of the graph.
Finally, note that these results are in perfect accordance with the geometric intuition
given by the sketch of the graph.
f .x/
a D lim and b D lim .f .x/ ax/:
x!C1 x x!C1
With the concept of limit at our disposal, we return to the discussion that lead
to (9.3).
Definition 9.20 Let I R be an open interval and f W I ! R be a given function.
For a fixed x0 2 I, we say that f is differentiable at x0 if the limit
f .x/ f .x0 /
lim
x!x0 x x0
does exist. In this case, this limit is called the derivative of f at x0 , and is
denoted f 0 .x0 /.
In the settings of the previous definition, we observe that the existence of f 0 .x0 /
is equivalent to the existence of the limit
f .x0 C h/ f .x0 /
lim :
h!0 h
Indeed, on the one hand, letting x0 C h D x we have h D x x0 , and obviously
h ! 0 , x ! x0 . On the other hand, when we write f .x0 C h/ we are tacitly
assuming that h is so small that x0 C h 2 I; however, since we are computing a limit
and I is open, such a supposition doesn’t impose any restriction upon the definition
of derivative.
In short, f being differentiable at x0 2 I, we have
From now on, whenever convenient we refer to any one of the fractions above as the
Newton’s quotient of f at x0 .
The coming examples compute derivatives of some simple functions.
Example 9.21
(a) If f W R ! R is a constant function, then f is differentiable and f 0 .x0 / D 0 for
every x0 2 R.
(b) If n 2 N and f W R ! R is such that f .x/ D xn for every x 2 R, then f is
differentiable, with f 0 .x0 / D nxn1
0 for every x0 2 R.
(c) If n 2 Z is negative and f W R n f0g ! R is given by f .x/ D xn , then f is
differentiable and f 0 .x0 / D nx0n1 for every x0 2 R n f0g.
Proof (a) Let’s apply the first equality in (9.13): letting f .x/ D c for every x 2 R,
we get
(b) Let’s apply the second equality in (9.13): by Newton’s binomial formula, we
get
! !
1 X n nk k
n
f .x0 C h/ f .x0 / 1
D ..x0 C h/ x0 / D
n n
x h x0n
h h h kD0 k 0
! !
Xn
n nk k1 Xn
n nk k1
D x h D nx0 C
n1
x h :
kD1
k 0 kD2
k 0
Now, applying the result of item (b) (with m in place of n), we get
1
f 0 .x0 / D mxm1
0 D mx0m1 D nx0n1 :
x2m
0
t
u
Example 9.22 The sine and cosine functions are differentiable, with sin0 x0 D cos x0
and cos0 x0 D sin x0 for every x0 2 R.
Proof Let’s do the proof for the sine function, the proof for the cosine function
being completely analogous.
294 9 Limits and Derivatives
Since the cosine function is continuous, Proposition 9.7, (9.6) and the fundamental
trigonometric limit (cf. Lemma 9.11) give
sin h2 h
sin0 x0 D lim cos x0 C
h!0 h
2
2
sin h2 h
D lim lim cos x0 C
h!0 h
2
h!0 2
sin h
D lim cos x0 D cos x0 :
h!0 h t
u
Remarks 9.23 Before we proceed with the development of the theory, let’s pause to
make two useful remarks.
1. Since derivatives are limits, given a function f W Œa; b/ ! R we can consider the
right-handed derivative of f at a, i.e., the one-sided limit
f .x/ f .a/
fC0 .a/ D lim ;
x!aC xa
f .x/ f .b/
f0 .b/ D lim
x!b xb
(also as before, provided such a limit exists). Moreover, whenever we say that
a function f W Œa; b ! R is differentiable, we shall implicitly assume that
its derivatives at x D a and x D b are one-sided ones, and shall write simply
f 0 .a/ and f 0 .b/ (instead of fC0 .a/ and f0 .b/) to denote them. From now on, such
conventions will be in force, without further comments.
2. As in Sect. 9.1, we shall sometimes denote the derivative of a function f at a point
df
x0 by using Leibniz’s classical notation dx .x0 /, so that
df f .x/ f .x0 /
.x0 / D lim :
dx x!x 0 x x0
The definition of derivative, together with the heuristic discussion of Sect. 9.1,
allows us to present another important
9.3 Basic Properties of Derivatives 295
O Q x
Example 9.25 Let H be the portion of the curve xy D 1 contained in the first
quadrant of a cartesian plane,2 and P and Q be points along the axes, such that
!
line PQ is tangent to H at the point A (cf. Fig. 9.6). Prove that:
(a) AP D AQ.
(b) If O is the origin of the cartesian plane, then the area of triangle POQ doesn’t
depend on the position of A along H.
2
Such a portion is a branch of an equilateral hyperbola (cf. Chap. 6 of [4], for instance), but we
shall not use this fact.
296 9 Limits and Derivatives
(a) It’s a well know fact (again, see Chap. 6 of [4], for instance) that the midpoint of
PQ has abscissa and ordinate respectively equal to 12 x20 D x10 and 12 2x0 D x0 .
Hence, such a point coincides with A.
(b) Since POQ is a right triangle, we have
1 1 2
A.POQ/ D OP OQ D 2x0 D 2:
2 2 x0
t
u
Continuing with the analysis of the concept of derivative, let’s show that
differentiability is stronger than continuity.
Proposition 9.26 If a function f W I ! R is differentiable at x0 2 I, then f is
continuous at x0 .
Proof Note first that, for x 2 I n fx0 g, we have
f .x/ f .x0 /
f .x/ D f .x0 / C .x x0 /:
x x0
f .x/f .x0 /
On the other hand, since we are assuming the existence of limx!x0 xx0 D
f 0 .x0 /, it follows from Proposition 9.7 that
f .x/ f .x0 /
lim f .x/ D lim f .x0 / C .x x0 /
x!x0 x!x0 x x0
f .x/ f .x0 /
D f .x0 / C lim lim .x x0 /
x!x0 x x0 x!x0
3
Later, in Chap. 11, we shall see an example of a continuous function f W R ! R that is not
differentiable at any point.
9.3 Basic Properties of Derivatives 297
1
Therefore, f 0 .x0 / ¤ 0 and .f 1 /0 .y0 / D f 0 .x0 /
.
298 9 Limits and Derivatives
f 1 .y0 C l/ f 1 .y0 / 1
D ;
l .f .x0 C h/ f .x0 //=h
item (c) of Proposition 9.7, together with the fact that l ! 0 , h ! 0, furnishes
f 1 .y0 C l/ f 1 .y0 / 1
lim D lim
l!0 l l!0 .f .x0 C h/ f .x0 //=h
1 1
D lim D 0 :
h!0 .f .x0 C h/ f .x0 //=h f .x0 /
Therefore, f 1 is differentiable at y0 . t
u
Example 9.29 Let n > 1 be p an integer and f W Œ0; C1/ ! Œ0; C1/ be the n-th
root function, i.e., f .x/ D n x D x1=n . Then, f is not differentiable at x D 0 but is
differentiable at every x > 0, with f 0 .x/ D 1n x1=n1 .
Proof Recall that f D g1 , where g W Œ0; C1/ ! Œ0; C1/ is such that g.y/ D yn .
Hence, it follows from the previous result that f is differentiable at x D g.y/ if
and only if g0 .y/ ¤ 0. However, since g0 .y/ D nyn1 , we have f differentiable at
x D g.y/ if and only if y ¤ 0por, which is the same, x ¤ 0. In this last case, recalling
that x D g.y/ D yn , y D n x, (9.15) gives
1 1 1 1
f 0 .x/ D D D p D x1=n1 :
g0 .y/ ny n1 n
n. x/ n1 n
t
u
For an important generalization of the previous example, see Problem 4,
page 310.
We close this section by presenting the concept of higher order derivatives. To
this end, let be given an interval I and a function f W I ! R.
We say that f is twice differentiable at x0 2 I if there exists a neighborhood
.x0 r; x0 C r/ of x0 such that f is differentiable in I \ .x0 r; x0 C r/ and f 0 W
I \ .x0 r; x0 C r/ ! R is differentiable at x0 2 I. In this case, we also say that
.f 0 /0 .x0 / is the second derivative of f at x0 , and write f 00 .x0 / D .f 0 /0 .x0 /. If f is
twice differentiable at each x0 2 I, we shall simply say that f is twice differentiable
in I, and let f 00 W I ! R denote its second derivative function.
More generally, let k 2 N and suppose we have already defined what is meant
by the k-th derivative f .k/ .x0 / of f at x0 2 I. If f is k times differentiable in
I (i.e., if f .k/ .x0 / exists for every x0 2 I), we let f .k/ W I ! R denote the k-th
9.3 Basic Properties of Derivatives 299
and
for every k 2 N.
4
Since f .k/ is also used to denote the composition of f W I ! I with itself, k times, we will rely on
the context to clear any danger of confusion.
300 9 Limits and Derivatives
The next problem establishes a simple particular case of the chain rule (cf.
Theorem 9.19).
f W .a; b/ ! R be differentiable
2. * Let at x0 2 .a; b/. For c > 0 (resp. c < 0), let
g W ac ; bc ! R (resp. g W bc ; ac ! R) be the function given by g.x/ D f .cx/.
Show that g is differentiable at xc0 , with g0 xc0 D cf 0 .x0 /.
3. * Let I be an interval and k 2 N. If f ; g W I ! R are k times differentiable in
x0 2 I, prove that f C g and fg are k times differentiable at x0 .
4. * In Example 8.37, we defined arcsin W Œ1; 1 ! Œ 2 ; 2 and arccos W
Œ1; 1 ! Œ0; as the inverses of sin W Œ 2 ; 2 ! Œ1; 1 and cos W Œ0; !
Œ1; 1, respectively. Use Theorem 9.28 to prove that the restrictions of arcsin
and arccos to the open interval .1; 1/ are differentiable, with
1 1
arcsin0 x D p and arccos0 x D p ;
1 x2 1 x2
f .bn / f .an /
lim D L:
n!C1 bn an
9.3 Basic Properties of Derivatives 301
Then, let r and r0 be the straight lines that pass through the point .1; 1/ and
2
are tangent to the parabola y D x4 ; use the system above to find the points at
0
which r e r touch the parabola.
For the next problem, the reader may find it helpful to read the proof of
Theorem 6.62 again, as well as to review the elementary facts on conics (cf.
Chap. 6 of [4], for instance).
12. Let P be the parabola with focus F and directrix d. If P is a point on d and
! !
A; B 2 P are such that AP and BP are tangent to P, show that F 2 AB.
For the next problem, the reader may find it convenient to have at his/her
disposal an extension of the definition of limx!x0 f .x/ to the case of a function
f W X ! R, where X R is a nonempty set and x0 2 R is the limit of a
sequence of distinct elements of X (for instance—and this will p essentially be
the case of our interest –, we can have X D Q and x0 D 2). Fortunately,
such an extension is formally identical to the one we have already met: we say
5
We shall have more to say on Newton’s method in Problem 7, page 455.
302 9 Limits and Derivatives
that limx!x0 f .x/ exists and is equal to L if, for every > 0, there exists ı > 0
(depending on ) such that
p
If k 2 N is not a perfect square, show that f is differentiable at x D k.
This section is devoted to the presentation of the usual differentiation rules, i.e.,
formulas that show how to compute derivatives of certain functions constructed in
terms of two given differentiable functions.
For item (c) of the coming proposition, we recall that if g W I ! R is
differentiable at x0 2 I, then g is continuous at x0 ; therefore, if g.x0 / ¤ 0, then
there exists an interval J I, containing x0 and such that g ¤ 0 in J. Thus, we think
of gf as defined in such a J.
Proposition 9.32 If f ; g W I ! R are differentiable at x0 2 I, then:
(a) f ˙ g is differentiable at x0 , with .f ˙ g/0 .x0 / D f 0 .x0 / ˙ g0 .x0 /.
(b) fg is differentiable at x0 , with .fg/0 .x0 / D f 0 .x0 /g.x0 / C f .x0 /g0 .x0 /.
(c) If g.x0 / ¤ 0, then gf is differentiable at x0 , with
0
f f 0 .x0 /g.x0 / f .x0 /g0 .x0 /
.x0 / D :
g g.x0 /2
Proof
(a) Let us prove that f C g is differentiable at x0 , with .f C g/0 .x0 / D f 0 .x0 / C g0 .x0 /
(the claims relative to f g can be proved in quite analogous ways). Since
.f C g/.x/ .f C g/.x0 /
.f C g/0 .x0 / D lim
x!x0 x x0
f .x/ f .x0 / g.x/ g.x0 /
D lim C
x!x0 x x0 x x0
f .x/ f .x0 / g.x/ g.x0 /
D lim C lim
x!x0 x x0 x!x0 x x0
D f 0 .x0 / C g0 .x0 /:
g.x/ g.x0 /
C f .x0 / lim
x!x0 x x0
D f 0 .x0 /g.x0 / C f .x0 /g0 .x0 /:
From the equality above and invoking the differentiability (and continuity) of g
at x0 , we get
1 1 1 1
lim D g0 .x0 / :
x!x0 x x0 g.x/ g.x0 / g.x0 /2
304 9 Limits and Derivatives
For the general case, we use the particular case above, together with the result
of (b):
0 0
f 1 0 1 1
.x0 / D f .x0 / D f 0 .x0 / .x0 / C f .x0 / .x0 /
g g g g
f 0 .x0 / g0 .x0 / f 0 .x0 /g.x0 / f .x0 /g0 .x0 /
D f .x0 / D :
g.x0 / g.x0 /2 g.x0 /2
t
u
Example 9.33 Let f W R ! R be the polynomial function given by
1
arctan0 y D
1 C y2
for every y 2 R.
Proof First of all, observe that tan0 x D sec2 x ¤ 0 for every x 2 2 ; 2 . Hence,
it follows from Theorem 9.28 that arctan is differentiable. Moreover, if y 2 R and
x D arctan y, then y D tan x and, thanks to (9.15) and the result of the former
example, we get
9.4 Computing Derivatives 305
1 1 1 1
arctan0 y D D D D :
tan0 x sec2 x 1 C tan2 x 1 C y2
t
u
The most important of all differentiation rules is the chain rule. Loosely
speaking, it establishes the differentiability of the composite of two differentiable
functions, and shows how to compute such a derivative.
More precisely, let I and J be intervals and g W I ! J and f W J ! I be given
functions, with g differentiable at x0 2 I and f differentiable at y0 D g.x0 / 2 J.
Suppose that, in some neighborhood I \ .x0 r; x0 C r/ of x0 , equation g.x/ D y0
has x0 as its only root. Then, for x 2 I \ .x0 r; x0 C r/, we can safely write
As we shall see in Theorem 9.19, this formula continues to hold in the general
case, albeit the proof is much more involving, for, g.x/ D y0 can have infinitely
many solutions in any neighborhood of x0 (for instance, think of g as the function f
of Problem 11, and solve the equation g.x/ D 0). Hence, we need to develop some
preliminaries.
Let I R be an open interval and f W I ! R be differentiable at x0 2 I. Set
I x0 D fh 2 RI x0 C h 2 Ig;
(This last computation, together with r.0/ D 0, actually shows that r is differentiable
at 0 with r0 .0/ D 0. Nevertheless, we won’t use this fact here.)
Conversely, we have the following auxiliary result.
Lemma 9.36 Let I R be an open interval, f W I ! R be a given function and
x0 2 I. If there exists a real number L such that the function r W I x0 ! R, given
by r.h/ D f .x0 C h/ f .x0 / Lh, satisfies the condition limh!0 r.h/
h
D 0, then f is
differentiable at x0 , with f 0 .x0 / D L.
r.h/ f .x0 Ch/f .x0 / r.h/
Proof Since h
D h
L, the condition limh!0 h
D 0 is equivalent to
f .x0 C h/ f .x0 /
lim D L:
h!0 h t
u
Let I be an open interval and f W I ! R be a function continuous at x0 2 I.
For every L 2 R, the graph of the affine function h 7! f .x0 / C Lh passes
through .x0 ; f .x0 //. This being said, it is frequently useful to rephrase the previous
discussion by saying that f is differentiable at x0 if and only if f admits a best affine
approximation in a neighborhood of x0 , in the sense of Lemma 9.36.
In other words, letting f W I ! R be differentiable at x0 2 I, we have
with R W I x0 ! R given by
r.h/
R.h/ D h ; if h ¤ 0
:
0; if h D 0
where t.h/ D g0 .x0 /h C r.h/h. Hence, for h 2 I x0 , it follows from this and (9.20)
that
Finally, the discussion at the next to last paragraph before the statement of
the chain rule guarantees that, in order to show that f ı g is differentiable at x0
and that (9.19) is true, it suffices to prove that R is continuous at I x0 , with
R.0/ D 0. But this is straightforward from the chain rule for continuous functions
(cf. Proposition 8.12), together with the fact that r, s and t are continuous, with
r.0/ D s.0/ D t.0/ D 0. t
u
Example 9.38 The chain rule allows us to compute the derivative of the cosine
function from that of the sine function. Indeed, since cos x D .sin ıg/.x/, where
g W R ! R is given by g.x/ D 2 x, it follows from the chain rule that
cos0 x D .sin ıg/0 .x/ D sin0 g.x/ g0 .x/ D cos g.x/ .1/
D cos x D sin x:
2
p
Example 9.39 Let R > 0 and f W .R; R/ ! R be given by f .x/ D Rp2 x2 .
Define g W .0; C1/ ! R and h W .R; R/ ! .0; C1/ by letting g.x/ D x and
1
h.x/ D R2 x2 , so that f D g ı h. Since g0 .y/ D 2p y
for y > 0, the chain rule gives,
for x 2 .R; R/,
1 x
f 0 .x/ D .g ı h/0 .x/ D g0 .h.x//h0 .x/ D p .2x/ D p :
2 h.x/ R x2
2
A faster way of using the chain rule to compute the derivative of f at x 2 .R; R/
is this: first write f .x/2 Cx2 D R2 , then differentiate both sides with respect to x (and
with the aid of the chain rule) to get 2f .x/f 0 .x/ C 2x D 0; hence, f 0 .x/ D f .x/
x
D
p 2 2 .
x
R x
Example 9.40 We can use the chain rule, together with the formula for differen-
tiation of a product, to establish the differentiation formula for a quotient. To this
end, let f ; g W I ! R be differentiable in I, with g.x/ ¤ 0 for every x 2 I. Letting
W R n f0g ! R denote the inverse proportionality function, so that .x/ D x1 for
every x ¤ 0, it follows from item (c) of Example 9.21 that is differentiable, with
f .x/
0 .x/ D x2 for every x 2 R n f0g. Now, notice that g.x/ D f .x/ . ı g/.x/ for
every x 2 I, so that the formula for differentiation of a product and the chain rule
give
0
f
.x/ D f 0 .x/ . ı g/.x/ C f .x/ . ı g/0 .x/
g
1
D f 0 .x/ C f .x/ 0 .g.x//g0 .x/
g.x/
f 0 .x/
D C f .x/.g.x/2 /g0 .x/
g.x/
f 0 .x/g.x/ f .x/g0 .x/
D :
g.x/2
9.4 Computing Derivatives 309
The coming example is considerably more sophisticated than the previous ones.
Example 9.41 Prove that f W R ! R, given by f .x/ D sin.x2 /, is not periodic.
Proof First of all, note that if f W R ! R is continuously differentiable and periodic,
then its derivative f 0 W R ! R, besides being continuous, is also periodic. Indeed,
if f .x/ D f .x C p/ for some real p > 0 and every x 2 R, then the chain rule gives
f 0 .x/ D f 0 .x C p/ for every x. Therefore,
Im.f 0 / D Im.fjŒ0;p
0
/;
and the continuity of f 0 , together with Corollary 8.30, guarantees that f 0 is a bounded
function.
Now, letting f be the given function and g W R ! R be given by g.x/ D x2 , we
have f .x/ D .sinıg/.x/, so that the chain rule furnishes
Hence, if f .x/ D sin.x2 / were periodic, the discussion at the preceding paragraph
would guarantee that f 0 would be bounded. Since this is obviously false, we have
reached a contradiction. t
u
The following corollary of the chain rule will be of interest later.
Corollary 9.42 Let I and J be open intervals and g W I ! J and f W J ! R given
functions. If g is n times differentiable at x0 2 I and f is n times differentiable at
g.x0 / 2 J, then f ı g W I ! R is n times differentiable at x0 .
Proof We make induction on n, relying on the chain rule for the case n D 1. Assume
the corollary to be true when n D k, and let f and g be k C 1 times differentiable
at x0 .
In a neighborhood of x0 , we have .f ı g/0 D .f 0 ı g/g0 , with f 0 and g0 being k
times differentiable at that point. The induction hypothesis guarantees that f 0 ı g is k
times differentiable at x0 , from where Problem 3, page 300 (applied to f 0 ı g and g0 )
assures that .f ı g/0 D .f 0 ı g/g0 is also k times differentiable at x0 . Therefore, f ı g
is k C 1 times differentiable at x0 . t
u
1. Compute the derivatives of the given functions, assuming in each case that the
function and its derivative are defined in their maximal domains:
f 0 .x/ X 1
n
D :
f .x/ jD1
x ˛j
2xC1
6. If f W R n f3g ! R is given by f .x/ D x3 , compute f .5/ .0/.
7. Let f W Œ0; C1/ ! R be given by
s r q
p
f .x/ D xC xCC xC x C 1;
pk .x/
8. For k 2 N, show that arctan.k/ .x/ D .1Cx 2 /k , where pk is a polynomial of degree
(c) If P is one of the intersection points of item (b), then the straight lines
tangent to E and H at P are perpendicular.
11. Do the following items:
(a) If f W R ! R is given by
x2 sin 1x ; if x ¤ 0
f .x/ D ;
0; if x D 0
X
n
f .x/ D aj cos.jx/
jD1
15. * Prove the following version of l’Hôpital’s rule6 : let f ; g W Œa; b/ ! R be two
functions differentiable at a and such that f .a/ D g.a/ D 0. If g ¤ 0 in .a; b/
and g0 .a/ ¤ 0, show that
f .x/ f 0 .a/
lim D 0 :
x!a g.x/ g .a/
Let f W Œa; b ! R be continuous in Œa; b, differentiable in .a; b/ and such that
f .a/ D f .b/ D 0. In addition, suppose that f is nonzero in at least one point of the
interval .a; b/. Arguing heuristically, translate the horizontal axis parallel to itself,
until it becomes tangent to the graph of f at a point .c; f .c// (cf. Fig. 9.7). Then, on
the one hand, such a tangent line must have slope f 0 .c/; on the other, being parallel
to the horizontal axis, its slope must be equal to 0, so that f 0 .c/ D 0.
The first result of this section, which is known in the literature as Rôlle’s
theorem,7 puts the discussion of the previous paragraph in solid grounds.
Lemma 9.43 (Rôlle) Let f W Œa; b ! R be continuous in Œa; b and differentiable
in .a; b/. If f .a/ D f .b/ D 0, then there exists c 2 .a; b/ such that f 0 .c/ D 0.
Proof Let c; d 2 Œa; b be the points in which f attains its minimum and maximum
values, respectively. If c; d 2 fa; bg, then for every x 2 Œa; b we have 0 D f .c/
f .x/ f .d/ D 0. Therefore, f vanishes identically in Œa; b and there is nothing left
to do.
Else, suppose that c 2 .a; b/ (the case of d 2 .a; b/ can be dealt with
analogously). Since f .x/ f .c/ for every x 2 Œa; b, we conclude that
a c x
b
6
After Guillaume F. Antoine, Marquis de l’Hôpital, French mathematician of the XVII century. A
more refined version of l’Hôpital’s rule will be presented at Proposition 9.50.
7
After Michel Rôlle, French mathematician of the XVII century.
9.5 Rôlle’s Theorem and Applications 313
Thus, f 0 .c/ D 0. t
u
The next example shows a typical application of Rôlle’s theorem.
Example 9.44 Given a; b; c 2 R, show that the equation 4ax3 C 3bx2 C 2cx D
a C b C c has at least one real root between 0 and 1.
Proof First of all, note that a standard application of the IVT is not conclusive,
for, if f .x/ D 4ax3 C 3bx2 C 2cx .a C b C c/, then f .0/ D .a C b C c/ and
f .1/ D 3a C 2b C c, so that f .0/ and f .1/ can have equal signs (for instance, for
a D 1 and b D c D 23 ).
On the other hand, if f .x/ D ax4 Cbx3 Ccx2 .aCbCc/x, then f .0/ D f .1/ D 0.
Hence, Rôlle’s theorem assures the existence of x0 2 .0; 1/ such that f 0 .x0 / D 0.
Now, note that this is the same as saying that x0 is a root of the given equation. u
t
Let’s (heuristically) analyse a situation more general than that of Rôlle’s theorem.
More precisely, let’s consider again a function f W Œa; b ! R, continuous in Œa; b
and differentiable in .a; b/, but such that f .a/ and f .b/ can assume any values
whatsoever. The slope of the secant line to the graph of f passing through .a; f .a//
and .b; f .b// equals f .b/f .a/
ba . We translate it parallel to itself until it becomes tangent
to the graph at a point .c; f .c//, with c 2 .a; b/ (cf. Fig. 9.8). Then, on the one hand,
the slope of this tangent line equals f 0 .c/; on the other, since it is parallel to the
original secant line, its slope equals that of the secant, so that
f .b/ f .a/
f 0 .c/ D :
ba
This result is the content of the mean value theorem (MVT, shortly) of
Lagrange, and it is the content of the coming result.
Theorem 9.45 (Lagrange) If f W Œa; b ! R is continuous in Œa; b and
differentiable in .a; b/, then there exists c 2 .a; b/ such that
f .b/ f .a/
D f 0 .c/:
ba
f (a)
f (c)
a c b x
for every x 2 Œa; b. (Note that the function of x between parentheses is precisely the
one that defines the secant to the graph of f passing through .a; f .a// and .b; f .b//.)
Clearly, g is continuous in Œa; b and differentiable in .a; b/, with g.a/ D g.b/ D 0.
Hence, by Rôlle’s theorem, there exists c 2 .a; b/ such that g0 .c/ D 0. Now, an easy
computation gives
0 0 f .a/ f .b/
g .x/ D f .x/ C ;
ba ba
so that
f .b/ f .a/
0 D g0 .c/ D f 0 .c/ :
ba t
u
Example 9.46 For x; y 2 2 ; 2 , show that j tan x tan yj jx yj.
Proof If x D y, there is nothing to do. Otherwise, it follows from the MVT of
Lagrange the existence of a real c between x and y, such that
1
tan x tan y D .tan c/.x y/ D .x y/:
cos2 c
1
j tan x tan yj D jx yj jx yj:
cos2 c t
u
The following corollary sets an important consequence of Lagrange’s MVT.
Corollary 9.47 Let I R be an interval and f ; g W I ! R be two differentiable
functions. If f 0 D g0 , then f g is constant.
9.5 Rôlle’s Theorem and Applications 315
The above inequalities can be seen as a linear system of two equations in the
unknowns f .x/ and g.x/. We can easily solve it to get f .x/ D f .0/ cos x C g.0/ sin x
and g.x/ D g.0/ cos x f .0/ sin x. t
u
In what comes next, we discuss an important generalization of Lagrange’s MVT
to two given functions, known as Cauchy’s mean value theorem. Note that (9.23)
reduces to the MVT of Lagrange if g.x/ D x.
Theorem 9.49 (Cauchy) If f ; g W Œa; b ! R are continuous in Œa; b and
differentiable in .a; b/, then there exists c 2 .a; b/ such that
for every x 2 Œa; b. It is immediate that h is continuous in Œa; b and differentiable
in .a; b/, so that, by Lagrange’s MVT, there exists c 2 .a; b/ satisfying
h.b/ h.a/
h0 .c/ D :
ba
316 9 Limits and Derivatives
Now, since h.a/ D h.b/ D f .b/g.a/ f .a/g.b/, we have h0 .c/ D 0 and it suffices
to observe that
f .x/
lim :
x!a g.x/
f 0 .x/ f .x/
lim D L ) lim D L:
x!a g0 .x/ x!a g.x/
Proof Start by continuously extending f and g to Œa; b/, setting f .a/ D g.a/ D 0.
For x 2 .a; b/, it follows from Cauchy’s MVT that
for some c.x/ 2 .a; x/; in particular, we clearly have limx!a c.x/ D a.
0
Now, if limx!a gf 0.x/
.x/ D L, a slight modification in the hint given to Problem 10,
f 0 .c.x//
page 291, guarantees that limx!a g0 .c.x// D L. Hence, it readily follows from (9.24)
f .x/
that limx!a g.x/ D L. t
u
Example 9.51 We use l’Hôpital’s rule for 00 to compute limx!0 1cos p x . (Note that
p x
x 7! x is not differentiable at x D 0, so that we cannot usep the result of
Problem 15, page 311). Indeed, both f .x/ D 1 cos x and g.x/ D x (for x > 0)
satisfy the hypotheses of the previous proposition and are such that
f 0 .x/ sin x p
lim D lim p D lim 2 x sin x D 0
x!0 g0 .x/ x!0 1=.2 x/ x!0
f .x/
Therefore, limx!0 g.x/
D 0.
9.5 Rôlle’s Theorem and Applications 317
for every x 2 R.
2. Let > 0 and f W R ! R be twice differentiable and such that f 00.x/ C f .x/ D
0 for every x 2 R. Prove that
p f 0 .0/ p
f .x/ D f .0/ cos. x/ C p sin. x/;
for every x 2 R.
3. * Let I be an interval and f W I ! R be a twice differentiable function, such
that f 00 is constant. Prove that f either vanishes identically or is a polynomial
function of degree at most 2. More precisely, if x0 2 I, show that
f 00 .x0 /
f .x/ D f .x0 / C f 0 .x0 /.x x0 / C .x x0 /2 :
2
4. (OIMU) Find all functions f W R ! R such that .f .x/ f .y//2 jx yj3 , for
every x; y 2 R.
5. Let I R be an interval and c 2 I. Let f W I ! R be a function continuous in I
and differentiable in I n fcg. If limx!c f 0 .x/ D l, prove that f is differentiable in
c, with f 0 .c/ D l.
6. (Putnam) Let a0 ; a1 ; : : : ; an be real numbers such that a10 C a21 C C nC1
an
D 0.
Show that the polynomial function a0 C a1 x C C an x has at least one real
n
root.
7. Show that the equation x2 D x sin x C cos x has exactly two real roots.
8. Redo Problem 8, page 300, this time under the assumption that f W I ! R is
continuously differentiable in I.
9. In each item below, compute the given limit:
p p
1cos x cos x 3 cos x
(a) limx!0 2
xp
. (c) limx!0 2 .
x
1 3 cos x sin2 .x/
(b) limx!0 x2
. (d) limx!1 .x3 C1/2 .
10. Let f ; g W .0; 2 / ! R be given by f .x/ D x3=2 sin 1x and g.x/ D x1=3 cos x C 1.
f .x/
Compute, if it exists, limx!0 g.x/ .
11. If f W .a; b/ ! R is twice differentiable, show that, for every x0 2 .a; b/,
12. Let f W Œa; b/ ! R be twice continuously differentiable in Œa; b/, with f .a/ D 0
and f 00.a/ ¤ 0.
(a) Show that there exists c 2 .a; b/ such that f .x/ .x a/f 0 .x/ ¤ 0, for every
x 2 .a; c/.
.xa/f .x/ 0 00
(b) Compute limx!a f .x/.xa/f 0 .x/ in terms of f .a/, f .a/ and f .a/ (item (a)
assures that the denominator doesn’t vanish in the interval .a; c/).
13. * Prove the following version of l’Hôpital’s rule for indeterminacies 00 at C1:
let a > 0 and f ; g W .a; C1/ ! R be differentiable functions such that g; g0 ¤ 0
in .a; C1/. If limx!C1 f .x/ D limx!C1 g.x/ D 0, then
f 0 .x/ f .x/
lim D L ) lim D L:
x!C1 g0 .x/ x!C1 g.x/
f 0 .x/ f .x/
lim 0
D L ) lim D L:
x!a g .x/ x!a g.x/
16. The differentiable function f W R ! R is such that jf 0 .x/j c < 1 for every real
x, where c is a positive real constant. Show that there exists only one x0 2 R
such that f .x0 / D x0 .
17. * Let I R be an open interval, x0 2 I and f W I ! R a continuous function,
which is differentiable in I n fx0 g. If there exists L D limx!x0 f 0 .x/, prove that f
is differentiable at x0 , with f 0 .x0 / D L.
18. (Romania) Prove that there does not exist a differentiable function f W
.0; C1/ ! .0; C1/ such that f .x/2 f .x C y/.f .x/ C y/ for every
x; y 2 .0; C1/.
19. (OBMU) Find all differentiable functions f W R ! R such that f .0/ D 0 and
jf 0 .x/j jf .x/j for every x 2 R.
In all that follows, I denotes an interval, whose interior is the interval obtained
by excluding its endpoints, if they exist. Thus, if I D Œa; b, Œa; b/, .a; b or .a; b/,
then the interior of I is the open interval .a; b/; analogously, if I D Œa; C1/ or
.a; C1/, its interior is .a; C1/, and if I D .1; b or .1; b/, its interior is
.1; b/.
Broadening the situation of the first paragraph above, our purpose in this section
is to solve the problem of finding extreme points and values of functions f W I ! R
which are continuous in I and differentiable in the interior of I. To this end, we begin
with the following
Definition 9.52 Given a function f W I ! R, we say that x0 2 I is a local maximum
(resp. local minimum) point for f if there exists ı > 0 such that f .x0 / f .x/ (resp.
f .x0 / f .x/), for every x 2 I \ .x0 ı; x0 C ı/.
In Fig. 9.9, points x0 , x00 and x000 are of local minimum for the function
f W .a; b/ ! R whose graph is depicted. Observe that, in a neighborhood of each
of these points, the values of f are no less than those it assumes at x0 , x00 and x000
(as indicated by the three dashed horizontal line segments). Also, x00 is the only
global minimum point of f (i.e., a point in which f attains the minimum possible
value in .a; b/), whereas x0 and x000 are not global minimum points. Finally, notice
that f has also two maximum local points (identify them!), albeit none of which is
a global maximum point.
Generically, local maximum or minimum points of f W I ! R are called its
extreme points, or simply its extrema. If I is an open interval and f W I ! R is
differentiable, we shall show below that its extrema (if any) can be found among the
points where f 0 vanishes. Since these vanishing points will play quite an important
role in subsequent discussions, it is worth to start by naming them.
Definition 9.53 Let I R be an open interval and f W I ! R a differentiable
function. We say that x0 2 I is a critical point of f if f 0 .x0 / D 0.
Gf
x0
a x0 x0 b x
f .x/ f .x0 /
f 0 .x0 / D lim 0:
x!x0 C x x0
f .x/ f .x0 /
f 0 .x0 / D lim 0:
x!x0 x x0
Therefore, f 0 .x0 / D 0: t
u
As a corollary to the first derivative test, we have the following search criterion
for extreme points.
Corollary 9.55 Let f W I ! R be continuous in I and differentiable in the interior
of I. If f attains a global (maximum or minimum) value in I, then the corresponding
extreme point is either an endpoint of I or a critical point of f .
Proof Suppose that f attains a global minimum value in I (the case of a global
maximum value can be treated analogously). Let x0 2 I be such that f .x0 / D
minff .x/I x 2 Ig. If x0 is an endpoint of I, there is nothing left to do. Otherwise,
x0 belongs to the interior of I; however, since the interior de I is an open interval,
the first derivative test assures that x0 is a critical point of f . t
u
The coming example makes it clear that the above corollary solves the problem
of finding the global extreme values of a continuous function f W Œa; b ! R,
differentiable in .a; b/, as long as we know how to find the roots of the equation
f 0 .x/ D 0.
9.6 The First Variation of a Function 321
Our next result, a direct consequence of Lagrange’s MVT, teaches us how to get
the monotonicity intervals of a differentiable function. From now on, we shall refer
to it as the study of the first variation of a differentiable function.8
Proposition 9.57 If f W I ! R is continuous in I and differentiable in the interior
of I, then:
(a) f 0 0 in the interior of I if and only if f is nondecreasing in I.
(b) If f 0 > 0 in the interior of I, then f is increasing in I, but the converse is not
necessarily true.
(c) f 0 0 in the interior of I if and only if f is nonincreasing in I.
(d) If f 0 < 0 in the interior of I, then f is decreasing in I, but the converse is not
necessarily true.
Proof Let’s prove only items (a) and (b), the proofs of items (c) and (d) being
entirely analogous.
(a) Suppose first that f is nondecreasing in I. For a given x0 in the interior of I,
take ı > 0 such that .x0 ı; x0 C ı/ I. If x0 < x < x0 C ı, then x 2 I and
f .x/ f .x0 / 0. Hence, f .x/f
xx0
.x0 /
0 and, making x ! x0 C, we get
f .x/ f .x0 /
f 0 .x0 / D lim 0:
x!x0 C x x0
f .b/ f .a/
D f 0 .c/ 0:
ba
8
This terminology alludes to the key role of the first derivative in this result, as well as to the fact
that, classically, the first derivative of a (differentiable) function was called its first variation.
322 9 Limits and Derivatives
f .b/ f .a/
D f 0 .c/ > 0:
ba
2x.x C 1/ .x2 C 3/ 1 x2 C 2x 3
f 0 .x/ D D ;
.x C 1/2 .x C 1/2
so that f 0 .x/ D 0 if and only if x D 1 (recall that we must have x 0).
On the other hand, since x2 C 2x 3 D .x C 3/.x 1/, we conclude that
0
f is negative in the interval .0; 1/ and positive in the interval .1; C1/. Hence,
Proposition 9.57 says that f decreases in Œ0; 1 and increases in Œ1; C1/, so that it
really attains a global minimum, at x D 1. Therefore, the minimum value of f is
f .1/ D 2. t
u
We now apply the first variation of functions to give a third proof of the inequality
between the arithmetic and geometric means.
Example 9.59 Given an integer n > 1 and positive reals a1 ; a2 ; : : : ; an we have
a1 C a2 C C an p
n a1 a2 : : : an ;
n
with equality if and only if a1 D a2 D D an .
Proof Let’s make induction on n > 1, relying in the discussion that led to (5.2) for
the case n D 2. Given an integer k > 2, suppose we have established the inequality
for k 1 arbitrary positive real numbers, with equality if and only if they are all
equal.
Given k positive reals a1 ; : : : ; ak1 ; ak , let f W .0; C1/ ! R be defined for
x > 0 by
p
f .x/ D a1 C C ak1 C x k k a1 : : : ak1 x:
We shall show that f .x/ 0 for every x > 0, with equality if and only if x D a1
and a1 D D ak1 ; in particular, it will follow that f .ak / 0, with equality if and
only if a1 D D ak1 D ak , as wished. To this end, note that f is differentiable,
with
9.6 The First Variation of a Function 323
p 1
f 0 .x/ D 1 k
a1 : : : ak1 x k 1 :
p
Hence, letting x0 D k1 a1 : : : ak1 , we have f 0 < 0 in .0; x0 /, f 0 .x0 / D 0 and f 0 > 0
in .x0 ; C1/, so that Proposition 9.57 assures f to decrease in .0; x0 and increase in
Œx0 ; C1/. Therefore, f attains a global minimum (only) at x D x0 .
Finally, a straightforward substitution gives
p
f .x0 / D a1 C C ak1 .k 1/ k1 a1 : : : ak1 ; (9.25)
f .x/ f .x0 / 0
for every x > 0, with equality if and only if x D x0 and f .x0 / D 0. It now suffices
to observe that, thanks to (9.25) and the induction hypothesis, we have f .x0 / D 0 if
and only if a1 D : : : D ak1 . t
u
We shall sometimes use the first variation in the form of the following
Corollary 9.60 Let I be an open interval, f W I ! R be twice differentiable and
x0 2 I be a critical point of f . If f 00 > 0 (resp. f 00 < 0) in I, then x0 is the only global
minimum (resp. maximum) point of f . In particular, x0 is the only critical point of f .
Proof Suppose f 00 > 0 in I (the case f 00 < 0 in I is totally analogous). Since f 00 D
.f 0 /0 , item (b) of Proposition 9.57 (applied to f 0 , in place of f ) guarantees that f 0
increases in I. However, since f 0 .x0 / D 0, it follows that f 0 .x/ < 0 for x < x0 and
f 0 .x/ > 0 for x > x0 (in particular, x0 is the only critical point of f ). Then, items
(b) and (d) of Proposition 9.57 (this time applied to f ) guarantee that f decreases in
I\.1; x0 and increases in I\Œx0 ; C1/. Therefore, x0 is the only global minimum
point of f . t
u
With the aid of the previous corollary, we can given an interesting application of
the first variation to Euclidean Geometry.
Example 9.61 We are given an angle †AOB such that Ab
OB < 90ı (cf. Fig. 9.10),
! !
as well as a point P inside it. Show how to choose points X 2 OA and Y 2 OB such
that X, P and Y are collinear and XY has the minimum possible length.
O B
Y
324 9 Limits and Derivatives
Proof In what follows, we shall rely on [4] for the necessary background in
Euclidean Geometry.
Taking angles in radians, let Ab
OP D ˛, Bb
OP D ˇ and Ob XY D
, so that ˛
!
and ˇ are known and
is variable (i.e., depends on the position of X along OA).
Since the sum of the internal angles of triangle OXY equals radians, we have
0 <
< ˛ ˇ and Ob YX D ˛ ˇ
.
Applying the Sine Law (cf. Sect. 7.3 of [4]) to triangles XOP and YOP, we get
PX sin ˛ PY sin ˇ
D and D ;
OP sin
OP sin.˛ C ˇ C
/
so that
sin ˛ sin ˇ
XY D PX C PY D OP C :
sin
sin.˛ C ˇ C
/
sin ˛ sin ˇ
f .
/ D C ;
sin
sin.˛ C ˇ C
/
sin ˛ cos
sin ˇ cos.˛ C ˇ C
/
f 0 .
/ D 2
:
sin
sin2 .˛ C ˇ C
/
lim f 0 .
/ D 1 and lim f 0 .
/ D C1:
!0
!
since ˛; ˇ;
; ˛ C ˇ C
2 .0; /. Hence, Corollary 9.60 says that
0 is the only
global minimum point of f . t
u
9.6 The First Variation of a Function 325
We finish this section by applying Proposition 9.57 to prove the following result,
known as Darboux’s theorem.9
Theorem 9.62 (Darboux) If I R is an interval and f W I ! R is a differentiable
function, then the function f 0 W I ! R, even if discontinuous, satisfies the mean
value property.
Proof Let a < b in I and d be a real number between f .a/ and f .b/. We wish to
establish the existence of c 2 Œa; b such that f .c/ D d. To this end, we consider two
cases separately:
(i) f 0 .a/ < 0 < f 0 .b/ (or vice-versa): if f is not injective in Œa; b, then there exist
˛ < ˇ in Œa; b such that f .˛/ D f .ˇ/. By Lagrange’s MVT, there exists c 2
.˛; ˇ/ (hence, c 2 .a; b/) such that f 0 .c/ D f .ˇ/fˇ˛
.˛/
D 0. If f is injective in
Œa; b, it follows from Theorem 8.35 that f is monotonic in Œa; b. Therefore, the
previous proposition gives f 0 0 in Œa; b or f 0 0 in Œa; b, which is not the
case.
(ii) f 0 .a/ < d < f 0 .b/ (or vice-versa): let g W I ! R be given by g.x/ D f .x/ dx,
so that g is differentiable, with g0 .x/ D f 0 .x/ d for every x 2 I. Hence,
g0 .a/ < 0 < g0 .b/ or vice-versa, and item (i) assures the existence of c 2 .a; b/
such that g0 .c/ D 0. This is the same as f 0 .c/ D d. t
u
9
After Jean-Gaston Darboux, French mathematician of the XIX and XX centuries.
326 9 Limits and Derivatives
Pb
QR D 90ı . Find the position of Q along side AD such that the area of triangle
PQR is as large as possible.
6. Let f W .a; b/ ! R be a differentiable function and P.x0 ; y0 / be a point not
belonging to the graph of f . If there exists A 2 Gf such that
AP D minf A0 PI A0 2 Gf g;
!
we say that AP is the distance from P to Gf . In such a case, prove that AP ?r,
where r denotes the tangent line to Gf at A.
7. Let f W .a; b/ ! R and g W .c; d/ ! R be differentiable functions whose graphs
do not intersect each other. If there exist points A.˛; f .˛// and B.ˇ; f .ˇ// such
that
we say that AB is the distance between the graphs of f and g. In such a case,
Show that f 0 .˛/ D g0 .ˇ/.
Given differentiable functions f W .a; b/ ! R and g W .c; d/ ! R whose
graphs do not intersect, the result of the previous problem assures that the
problem of finding points A 2 Gf and B 2 Gg satisfying (9.26) is equivalent
to that of minimizing F W .a; b/ .c; d/ ! R given by
H B C
9.6 The First Variation of a Function 327
B
a
b
r
A P B
B
a
b
A l B r
Along this section we continue to study how the behavior of the derivatives of a
function influence the shape of its graph. As before, in all that follows I R denotes
an interval.
Definition 9.63 A function f W I ! R is said to be convex if, for all a; b 2 I and
t 2 Œ0; 1, we have
As t varies from 0 to 1, basic Analytic Geometry guarantees that the points of the
form .1 t/.a; f .a// C t.b; f .b// trace out the line segment secant with endpoints
.a; f .a// and .b; f .b//. Hence, f W I ! R is convex if and only if, for all a < b in
I, the portion of the graph of f between lines x D a and x D b does not intersect
the open half-plane situated above the straight line passing through .a; f .a// and
.b; f .b//. Yet in another way, letting
for all a; b 2 I and t 2 Œ0; 1. Accordingly, f W I ! R is strictly concave if and only
if f ..1 t/a C tb// > .1 t/f .a/ C tf .b/ for all distinct a; b 2 I and every t 2 .0; 1/.
Properties of (strictly) concave functions are easily derived from those of
(strictly) convex ones, just changing f by f when necessary. For this reason, in
all that follows we shall restrict ourselves to the analysis of convex and strictly
convex functions, leaving to the reader the task of adapting it to concave and strictly
concave ones.
9.7 The Second Variation of a Function 329
Definition 9.63 guarantees that every strictly convex function is convex. Never-
theless, the converse is not true in general, as the coming example shows.
Example 9.65 Every affine function is convex, albeit not strictly convex. Indeed, if
f W R ! R is given by f .x/ D Ax C B, with A; B 2 R, it is immediate to verify that
f ..1 t/a C tb/ D .1 t/f .a/ C tf .b/, for all a; b 2 R and every t 2 Œ0; 1.
In Corollary 9.70 we will show that, if I is an open interval, then a twice
differentiable f W I ! R is convex if and only if f 00 .x/ 0 for every x 2 I. We
will also show that, if f 00 .x/ > 0 for every x 2 I, then f is strictly convex. By
assuming the truth of these results for the time being, we now list some examples
of convex and concave functions (note that when we sketched the graphs of these
functions we implicitly assumed that they were indeed convex or concave, according
to the case at hand).
Example 9.66
(a) The restriction of the inverse proportionality function to the set of positive reals
is strictly convex. Indeed, letting f .x/ D 1x for x > 0, we have f 00 .x/ D x23 > 0.
(b) Letting n > 1 be an integer, the function f W .0; C1/ ! R given by f .x/ D xn
is also strictly convex, for, f 00 .x/ D n.n 1/xn2 > 0.
(c) The sine function is strictly concave in the interval .0; /, since sin00 x D
sin x < 0 in .0; /.
(d) Since tan00 x D 2 tan x sec x > 0 in 0; 2 , the tangent function is strictly convex
in this interval.
Remark 9.67 If f W I ! R is convex, then, letting t D 12 in Definition 9.63, we
f .a/Cf .b/
get f . aCb
2 / 2 for all a; b 2 I. Conversely, if f is continuous and such that
f .a/Cf .b/
f. 2 /
aCb
2
for all a; b 2 I, it is possible to show (see Problem 8) that f
is convex. By the same token, it is also possible to show (see Problem 9) that a
continuous f W I ! R is strictly convex if f . aCb 2
/ < f .a/Cf
2
.b/
for all distinct a; b 2 I.
In turn, these remarks allow us to use the elementary inequalities we had studied to
establish the (strictly) convex or concave character of several common functions. In
this respect, take a look at Problem 11.
Back to the development of the theory, let f W I ! R be a convex function. Then,
given a < b in I and 0 < t < 1, we have
bx xa
f .x/ f .a/ C f .b/:
ba ba
330 9 Limits and Derivatives
f (b)
a (1 − t)a + tb b
for all a < x < b in I, then .b x/.f .x/ f .a// .x a/.f .b/ f .x// or, which is
the same,
In what follows, we use the previous discussion to show that a convex function,
defined in an open interval, is continuous.
Proposition 9.68 If I is an open interval and f W I ! R is convex, then f is
continuous.
Proof Fix x0 2 I and a < x0 < b also in I (this is possible due to the openness
of I). For x 2 .x0 ; b/, the first inequality in (9.28), with x0 in place of a, gives
f .x/f .x0 /
xx0 f .b/f .x0 /
bx0 ; in turn, the second inequality in (9.28), with x0 in place of x
f .x/f .x0 / f .x0 /f .a/
and x in place of b, gives xx0 x0 a . By combining these two inequalities,
we get
f .x0 / f .a/ f .b/ f .x0 /
.x x0 / f .x/ f .x0 / .x x0 /:
x0 a b x0
f .b/ f .a/
f 0 .a/ f 0 .b/:
ba
Hence, f 0 is nondecreasing in I.
If f is strictly convex, we refine the above reasoning bay taking c 2 .a; b/ and
x 2 .a; c/, y 2 .c; b/. Then, (9.28) and the strict convexity of f give
Proof Item (a) follows immediately from item (a) of the previous result, together
with item (a) of Proposition 9.57 (applied to f 0 ).
In what concerns (b), if f 00 > 0 in I, then item (b) of Proposition 9.57 (applied
once more to f 0 ) guarantees that f 0 increases in I. Therefore, item (b) of the previous
theorem assures that f is strictly convex. t
u
Let us see a simple application of this result.
p p
Example 9.71 Let f W R n f 2; 2g ! R be given by f .x/ D 2x x
2 . Find the
intervals of the real line in which f is strictly convex or strictly concave.
2Cx2
Solution Computing the first and second derivatives of f , we get f 0 .x/ D .2x2 /2
2x
and f 00 .x/ D .2x2 /3 .6Cx2 /
. Hence,
8 8 p
< x > 0 and 2 x2 > 0 < 0<x< 2
x
f 00 .x/ > 0 , >0, or , or p :
2 x2 : :
x < 0 and 2 x2 < 0 x< 2
p
p conclude that f is strictly convex in each of thepintervals .1;
We readily p 2/
and .0; 2/, and strictly concave in each of the intervals . 2; 0/ and . 2; C1/.
t
u
The former corollary is usually referred to as the study of the second variation
of a twice differentiable function. We now wish to look at the points in which the
graph of f changes from being convex to being concave, or vice-versa. Before that,
it is useful to have the following definition at our disposal.
Definition 9.72 Let I be an open interval and f W I ! R be a continuous function.
A point x0 2 I is said to be an inflection point of f if there exists ı > 0 such that f
is convex in I \ .x0 ı; x0 / and concave in I \ .x0 ; x0 C ı/, or vice-versa.
The coming corollary gives a necessary condition to be satisfied by the inflection
points of twice differentiable functions.
Corollary 9.73 Let I be an open interval and f W I ! R be a twice differentiable
function. If x0 2 I is an inflection point of f , then f 00 .x0 / D 0.
Proof Suppose f is convex in .x0 ı; x0 / and concave in .x0 ; x0 C ı/, with ı > 0 so
small that .x0 ı; x0 Cı/ I (the other case can be dealt with in an analogous way).
Then, Corollary 9.70 gives f 00 .x/ 0 in .x0 ı; x0 / and f 00 .x/ 0 in .x0 ; x0 C ı/.
Suppose f 00 .x0 / < 0, and fix an arbitrary a 2 .x0 ı; x0 /, so that f 00 .x0 / <
00
0 f 00 .a/. Since f 00 .x0 / < f .x 2
0/
< f 00 .a/, Darboux’s Theorem 9.62 assures the
00
existence of b 2 .a; x0 / such that f 00 .b/ D f .x 2
0/
< 0. However, since b 2 .a; x0 / )
00
b 2 .x0 ı; x0 /, we should also have f .b/ 0, which is a contradiction.
Analogously, we also reach a contradiction by supposing that f 00 .x0 / > 0. Hence,
the only possibility is that f 00 .x0 / D 0. t
u
334 9 Limits and Derivatives
We now state and prove Jensen’s inequality,10 a result that gives greater
flexibility to the applications of convex functions to problems of maxima and
minima.
Theorem 9.74 (Jensen) Let I R be an open interval and f W I ! R be a convex
function. If x1 ; x2 ; : : : ; xn 2 I and t1 ; t2 ; : : : ; tn 2 .0; 1/, with t1 C t2 C C tn D 1,
then t1 x1 C t2 x2 C C tn xn 2 I and
X
n
1 Xn
1
sj D tj D .1 tnC1 / D 1;
jD1
1 tnC1 jD1 1 tnC1
10
After Johan Jensen, Danish engineer and mathematician of the XIX and XX centuries.
9.7 The Second Variation of a Function 335
with equality if and only if y D xnC1 . On the other hand, also by induction
hypothesis we have
n a1 C a2 C C an
Df
a1 C a2 C C an n
f .a1 / C f .a2 / C C f .an /
n
1=a1 C 1=a2 C C 1=an
D
n
with equality if and only if a1 D a2 D D an . Now, it suffices to observe that
inequality
A2
An−1
A0 O A1
X
n1 X
n1
1
A.A0 A1 : : : An1 / D A.Ai OAiC1 / D R2 sin Ai b
OAiC1
iD1 iD1
2
1 2X
n1
D R sin ˛i :
2 iD1
Now, since the sine function is strictly concave in the interval Œ0; , it follows
from (9.31) that
!
X
n1
1 X
n1
sin ˛i .n 1/ sin ˛i D .n 1/ sin ;
iD1
n 1 iD1 n1
with equality if and only if ˛1 D D ˛n1 D n1 . Therefore, there is indeed only
one convex n-gon satisfying the prescribed conditions. t
u
X
n r p p
xi n x1 C C xn
p p ;
iD1
1 xi n1 n1
(viii) The computation of the limits limx!˙1 f .x/ (in case they make sense) allows
us to better understand the behavior of the graph of f for large values of jxj.
(ix) Plotting some other points along the graph can be quite helpful in sketching it.
In this section, we illustrate the application of the program just delineated to
sketching the graphs of some simple functions.
Example 9.79 Let D D Rnf 2 CkI k 2 Zg. Sketch the graph of the secant function
sec W D ! R.
Solution
(i) First of all, note that it suffices to sketch the graph of sec in . 2 ; 2 /[. 2 ; 3
2 /,
for, since the cosine function is periodic of period 2, so is the secant function.
On the other hand, since cos.x C / D cos x for every x 2 R, we have
sec.x C / D sec x for every x 2 D; hence, Problem 10, page 194,
guarantees that the portion of the graph of sec in the interval . 2 ; 3 2 / is
obtained from that in the interval . 2 ; 2 / by means of a reflection along
the horizontal axis, followed by a translation of units, parallel to that same
axis.
In view of the above, from now on we restrict the analysis of the graph of the
secant function to the interval . 2 ; 2 /.
(ii) and (iii) Since
sin x
sec0 x D .cos1 x/0 D .cos x/2 . sin x/ D D tan x sec x; (9.32)
cos2 x
Therefore, sec00 ¤ 0, and sec has no inflection points. Also, since sec is strictly
convex if and only if sec00 x > 0, which in turn happens if and only if cos x >
0, we conclude that sec is strictly convex in . 2 ; 2 /.
(vii) Since sec is periodic, it graph does not contain horizontal or oblique asymp-
totes. On the other hand, we easily get limx! 2 sec x D limx! 2 C sec x D
C1, so that the vertical lines x D ˙ 2 are asymptotes of the graph of sec.
(viii) Since sec is periodic and nonconstant, there does not exist limx!˙1 f .x/.
(ix) Computing cos x for x D 0; ˙ 12 , ˙ 6 , ˙ 4 , ˙ 3 and ˙ 11 , we plot points
p 12
p2
.0; 1/, ˙ 6 ; 3 , ˙ 4 ; 2 and ˙ 3 ; 2 on the graph of sec.
9.8 Sketching Graphs 341
y=1
− π2 0 π
2
π 3π
2
x
y = −1
1
− 32 2 1
−1 − 3 − 3 2
x
−1
31
− 27
35
− 27
43
− 27
− 74
(vii) Since f is defined in the whole real line, its graph does not possess vertical
asymptotes. On the other hand, since limx!˙1 f .x/ x D limx!˙1 .2x C 4x C
2
1
2 x / D C1, Problem 11, page 291, guarantees that the graph of f does not
have oblique asymptotes.
(viii) It follows from Example 9.17 that limx!˙1 f .x/ D ˙1.
(ix) Since f .0/ D 1, it follows from (viii) and the IVT that f has a positive root
˛. Computing f .x/ for x D 32 , 1, 13 , 0, 14 , 13 , 12 and 1, we plot points
. 32 ; 74 /, .1; 1/, . 13 ; 35 1 7 1 1 1 5
27 /, .0; 1/, . 4 ; 32 /, . 3 ; 27 /, . 2 ; 4 / and .1; 7/
on the graph of f . (Recall that, in item (vi), we had already get point . 23 ; 31 27
/
on the graph.) In particular, we conclude that ˛ 2 . 14 ; 13 /.
Finally, as in the previous example, we gather the above information to sketch
the graph of f in Fig. 9.17.
t
u
1
Example 9.81 Sketch the graph of f W R n f1g ! R, given by f .x/ D x2 C xC1
.
Solution
1 3
2x C4x C2x1 2
(ii) and (iii) Since f 0 .x/ D 2x .xC1/ 2 D .xC1/2
, we have f 0 .x/ D 0 ,
3 2
2x C 4x C 2x 1 D 0. Hence, the only critical point of f is the only positive
root ˛ of the polynomial function of the previous example, so that ˛ 2 . 14 ; 13 /.
Then, f 0 .x/ > 0 , 2x3 C 4x2 C 2x 1 > 0 , x > ˛, and we conclude that
f increases in .˛; C1/. Note that ˛ is a minimum local point of f .
2
(iv) and (v) Since f 00 .x/ D 2 C .xC1/ 3 , we have f strictly convex (resp. concave)
1 1
in an interval I if and only if xC1 > 1 (resp. xC1 < 1) in I. Therefore, f is
strictly convex in each of the intervals .1; 2/ and .1; C1/, and strictly
concave in .2; 1/. Moreover, its only inflection point is x D 2.
9.8 Sketching Graphs 343
(vi) Note that f .2/ D 3, and the tangent line to the graph of f at x D 2 has
slope f 0 .2/ D 5. Letting .0; c/ be the point in which such a tangent line
intersects the vertical axis, we have 0.2/ c3
D 5, so that c D 7.
(vii) Since x D 1 is the only point for which f .x/ is not defined, the vertical
line x D 1 is the only candidate for vertical asymptote of the graph. This
is indeed the case, for, limx!1 f .x/ D 1 and lim x!1C f .x/ D C1.
f .x/ 1
On the other hand, since limx!˙1 x D limx!˙1 x C x.xC1/ D ˙1,
Problem 11, page 291, assures that the graph has no oblique asymptotes.
(viii) It follows from Example 9.17 that limx!˙1 f .x/ D C1.
(ix) Computing f .x/ for x D 52 , 32 , 54 , 34 , 12 , 14 , 0, 14 , 12 , 1 and 2, we plot
points . 52 ; 67 3 7 5 39 3 73 1 9 1 67
12 /, . 2 ; 4 /, . 4 ; 16 /, . 4 ; 16 /, . 2 ; 4 /, . 4 ; 48 /, .0; 1/,
1 7 1 5 3 13
. 4 ; 32 /, . 2 ; 4 /, .1; 2 / and .2; 3 / on the graph of f . (Recall that, in item (vi),
we had already get point .2; 3/ on the graph.)
Once more, we collect all of the above to sketch the graph of f , as shown in
Fig. 9.18. Note that, for the purpose of a better qualitative picture, we adopted
different scales on the horizontal and vertical axes.
t
u
−2 −1 α x
344 9 Limits and Derivatives
1. Sketch the graph of the cosecant function, csc W R n fkI k 2 Zg ! R, such that
csc x D sin1 x .
2. In each of the items below, sketch the graph of the given function (defined in
its maximal domain). For each such function (and whenever pertinent), compute
explicitly or estimate the critical points, monotonicity intervals, inflection points,
tangent lines at inflection points, intervals of convexity or concavity, asymptotes
and behavior at infinity:
O Γ
f 00 .x0 /
kD D ;
R .1 C f 0 .x0 /2 /3=2
where
D jj 2 f˙1g is the sign of . (Observe that the bigger R, the closer jkj
is to 0. This fact reflects gives geometric intuition to the term curvature.)
p
(g) Let R > 0 and f W .R; R/ ! R be given by f .x/ D R2 x2 (so that the
graph of f is a semicircle of radius R). Show that the curvature radius of the
graph of f is constant and equal to R.
As an illustration, Fig. 9.19 brings the osculating circle to the graph of f .x/ D 1x ,
x > 0, at a point A.
4. The purpose of this problem is to sketch the graph of f W R ! R, given by
0; se x D 0
f .x/ D :
x sin 1x ; se x ¤ 0
This chapter completes the task of establishing the fundamentals of Calculus, this
time studying the operation of integration on functions. As we shall see in the next
section, in its most simple form this reduces to the computation of areas under
the graphs of nonnegative continuous functions f W Œa; b ! R, suggesting that
geometric intuition will play a strong role throughout. Among other byproducts of
the coming discussions, we prove the irrationality of and study the properties
of two of the most ubiquitous functions of Mathematics, the exponential and
logarithmic functions with base e.
Our story begins in the III century B.C., with the great Archimedes, who considered
(and solved) the problem of computing the area under an arc of parabola, by using
the method of exhaustion.
In modern notation and in a slightly more general situation, the heuristics behind
such a method is the following: given a nonnegative function f W Œa; b ! R (cf.
Fig. 10.1), one wants to compute the area of the region R of the cartesian plane,
situated under the graph of f and above the horizontal axis, so that
To this end, one starts by dividing the interval Œa; b into k equal intervals, with
the aid of the points a D t0 < t1 < < tk D b such that ti ti1 D ba k for
1 i k. Then (cf. Fig. 10.2), one considers the portion of R contained in the
vertical strip bounded by the straight lines x D ti1 and x D ti , and approximates its
area, from below, by that of the greatest rectangle contained therein and having the
Œti1 ; ti as one of its sides; accordingly, one approximates the area of that portion of
a b x
a ti−1 ti b x
R, from above, by that of the smallest rectangle containing it and having the interval
Œti1 ; ti as one of its sides.
One now approximates the whole area of R from below and from above,
computing, in each case, the sums of the areas of the k rectangles of the previous
paragraph. Assuming that f attains its minimum and maximum values in each
interval Œti1 ; ti (which is always the case if f is continuous), and letting mi and
Mi respectively denote such values (so that mi and Mi are the lengths of the heights
of the rectangles one is considering), one obtains the sums
X k
ba X
k
A.f I k/ WD mi .ti ti1 / D mi (10.1)
iD1
k iD1
and
X k
ba X
k
A.f I k/ WD Mi .ti ti1 / D Mi : (10.2)
iD1
k iD1
and hopes that A.f I k/ and A.f I k/ do approximate A.R/ better and better as k !
C1.
In order to illustrate the difficulties involved, let’s consider the more particular
situation of the function f W Œ0; b ! R given by f .x/ D xn , where n is a given
natural number. As before, letting k 2 N and 0 D t0 < t1 < < tk D b be
the partition of Œ0; b such that ti ti1 D bk for 1 i k, we have ti D bik for
0 i k. Also, since x 7! xn is increasing, it follows that mi D f .ti1 / D ti1
n
and
10.1 Some Heuristics II 349
k nC1 X
b X n b X bi n
k k
b
A.f I k/ D ti D D in (10.3)
k iD1 k iD1 k k iD1
and, analogously,
nC1 Xk
b
A.f I k/ D .i 1/n : (10.4)
k iD1
bnC1
A.f I k/ A.f I k/ D I (10.5)
k
Xk
.k 1/nC1 X n
k
.i 1/n < < i:
iD1
nC1 iD1
ˇ
ˇ bnC1 1 nC1 ˇˇ bnC1
ˇA.R/ 1 ˇ A.f I k/ A.f I k/ < :
nC1 k k
a x0 x b x
Let’s now return (again in a modern setting) to the general problem of the
computation of the area of the region under the graph of a given nonnegative
function f W Œa; b ! R (cf. Fig. 10.3).
For each real number x 2 Œa; b, let’s denote by A.x/ the area of the region of
the plane situated under the graph of f , above the horizontal axis and between the
vertical lines of abscissas a and x (with the convention that A.a/ D 0). For a fixed
x0 2 Œa; b/ and taking x0 < x b, we have
where A.Œx0 ; x/ denotes the area of the portion of the region just described, situated
between the vertical lines of abscissas x0 and x.
For small values of x x0 (when compared to the values of f along the interval
Œx0 ; x), it is reasonable to assume that a good approximation for the value of
A.Œx0 ; x/ is the area of the trapezoid with base lengths f .x0 / and f .x/ and having
the line segment Œx0 ; x as one of its legs (the gray trapezoid of Fig. 10.3). Since the
altitude of such a trapezoid equals x x0 , it follows from (10.7) that
f .x0 / C f .x/
A.x/ A.x0 / D A.Œx0 ; x/ Š .x x0 /
2
This section introduces the concept of integral for bounded functions f W Œa; b ! R
and establishes the integrability of two important classes of such functions, namely,
the continuous and the monotone ones. To this end, we begin by fixing some
notations.
A partition of an interval Œa; b is the choice of a finite subset
of Œa; b. Given such a partition and a bounded (but not necessarily nonnegative)
function f W Œa; b ! R, we shall systematically denote
and
Mj D supff .x/I xj1 x xj g WD sup f :
Œxj1 ;xj
Note that both mj and Mj are well defined, thanks to the boundedness of f .
352 10 Riemann’s Integral
The analogues of the lower and upper approximations for the area of the region
under the graph of f , discussed in the previous section, are the lower sum and the
upper sum of f with respect to P, which are respectively defined by
X
k X
k
s.f I P/ D mj .xj xj1 / and S.f I P/ D Mj .xj xj1 /:
jD1 jD1
supfs.f I P/I P is a partition of Œa; bg D inffS.f I P/I P is a partition of Œa; bg:
However, some care is needed, for, the inequality s.f I P/ S.f I P/ (which is valid
when we take the same partition P at both sides) does not necessarily imply a
relation between the supremum and infimum above. This is due to the fact that
there is no relation between the generic partition P that declares the set at the left
hand side above and the one that declares the set at the right hand side (the choice
of the same letter to represent them was just a matter of notational convenience).
Thus, before we proceed towards a coherent definition of integrable function and
integral, we ought to compare s.f I P/ and S.f I Q/, for two generic partitions P and
Q of Œa; b. We do that in the coming lemma.
Lemma 10.1 Let f W Œa; b ! R be a bounded function. Given partitions P and Q
of Œa; b such that P Q, we have
X
i1
S.f I Q/ D Mj .xj xj1 / C . sup f /.x0 xi1 /
jD1 Œxi1 ;x0
X
k
0
C sup f .xi x / C Mj .xj xj1 /: (10.13)
Œx0 ;xi jDiC1
so that
sup f .x0 xi1 / C sup f .xi x0 / Mi .x0 xi1 / C Mi .xi x0 /
Œxi1 ;x0 Œx0 ;xi
D Mi .xi xi1 /:
X
i1 X
k
S.f I Q/ Mj .xj xj1 / C Mi .xi xi1 / C Mj .xj xj1 /
jD1 jDiC1
X
k
D Mj .xj xj1 / D S.f I P/:
jD1
Finally, the proof of the inequality involving lower sums is completely analogue
and will be left to the reader (see Problem 1). t
u
As an immediate consequence of the previous lemma, given a bounded function
f W Œa; b ! R and (any) partitions P and Q of Œa; b, we have
In particular, given a partition Q of Œa; b, the above inequalities guarantee that
S.f I Q/ us an upper bound for the set fs.f I P/I P is a partition of Œa; bg of the lower
sums of f . Hence, Theorem 7.4 guarantees that such a set does have a supremum,
which will be denoted supP s.f I P/:
On the other hand, since the supremum of a set bounded from above is the smallest
of its upper bounds (one of which is S.f I Q/), we conclude that
However, since the partition Q was arbitrarily chosen, we conclude that this last
inequality is valid for every partition Q of Œa; b. To put in another way, the real
number supP s.f I P/ is a lower bound for fS.f I Q/I Q is a partition of Œa; bg, the set
of the upper sums of f . Hence, it follows from Problem 6, page 206, that such a set
has an infimum, which will be denoted by infQ S.f I Q/:
Finally, since the infimum of a set bounded from below is the largest of its lower
bounds (one of which is supP s.f I P/), it follows that
Fixed a partition P D f0 D x0 < x1 < x2 < < xk D 1g of Œ0; 1, Problem 4, page
206, assures that all of the intervals Œxj1 ; xj contain rational and irrational numbers.
Therefore, for 1 j k we have mj D 0 and Mj D 1, so that
X
k
s.f I P/ D mj .xj xj1 / D 0
jD1
1
When we look at large to Riemann’s legacy to Geometry and Analysis, we easily come to the
conclusion that they largely surpass the adequate formalization of the concept of integral that bears
his name. (As the attentive reader has certainly noticed, this chapter is entitled Riemann’s Integral!)
Indeed, Semi-Riemannian Geometry gave Albert Einstein the adequate theoretical apparatus for the
development of General Relativity. Other of Riemann’s creations, known today as Riemann’s zeta
function, has shown to be an ubiquitous object in Number Theory.
10.2 The Concept of Integral 355
and
X
k X
k
S.f I P/ D Mj .xj xj1 / D .xj xj1 / D 1:
jD1 jD1
Then,
and f is nonintegrable.
In order to discuss the integrability of functions in a more transparent way, it is
worth to recast the definition of integrable function as in the coming result, which is
known as Cauchy’s integrability criterion.
Theorem 10.4 (Cauchy) A bounded function f W Œa; b ! R is integrable if and
only if the following condition is satisfied: for every > 0, there exists a partition
P of Œa; b such that
Proof Firstly, suppose that f is integrable, so that supP s.f I P/ D infP S.f I P/. Let
this common value be denoted by I, and let > 0 be given. Since I D supP s.f I P/
is the least upper bound for the set of lower sums of f and I 2 < I, there exists a
partition P1 of Œa; b such that I 2 < s.f I P1 /. Analogously, since I D infP S.f I P/
is the greatest lower bound for the set of upper sums of f and I C 2 > I, there exists
a partition P2 of Œa; b such that S.f I P2 / < I C 2 . Letting P D P1 [ P2 , it follows
from (10.14) that
S.f I P / s.f I P / S.f I P2 / s.f I P1 / < I C I D :
2 2
Conversely, assume that the stated condition is satisfied. Then, given > 0, and
taking a partition P as in the statement, it follows from (10.15) that
However, since > 0 was arbitrarily chosen, it follows that infP S.f I P/
supP s.f I P/ D 0, as wished. t
u
We can finally establish the integrability of monotone and continuous functions.
Theorem 10.5 Every monotone function f W Œa; b ! R is integrable.
Proof Suppose f to be nondecreasing, the case of a nonincreasing function being
totally analogous. Then, the image of contained in the interval Œf .a/; f .b/, so that it
356 10 Riemann’s Integral
X
k X
k
S.f I P/ s.f I P/ D Mj .xj xj1 / mj .xj xj1 /
jD1 jD1
X
k
D .f .xj / f .xj1 //.xj xj1 /:
jD1
Letting ı D maxfxj xj1 I 1 j kg, the last equality above and (3.15) give
X
k
S.f I P/ s.f I P/ < .f .xj / f .xj1 //ı D ı.f .b/ f .a//:
jD1
Therefore, it suffices to choose P such that ı.f .b/ f .a// < , which happens
whenever ı < f .b/f.a/C1 , for instance. t
u
Theorem 10.6 Every continuous function f W Œa; b ! R is integrable.
Proof Corollary 8.25 guarantees that such an f is bounded. Hence, again by
Cauchy’s integrability criterion and given > 0, in order to establish the
integrability of f it suffices to find a partition P D fa D x0 < x1 < < xk D bg of
Œa; b such that (10.16) is satisfied for P D P.
As in the proof of the previous of the previous result,
X
k X
k
S.f I P/ s.f I P/ D Mj .xj xj1 / mj .xj xj1 /
jD1 jD1
(10.17)
X
k
D .Mj mj /.xj xj1 /:
jD1
X
k
0 .xj xj1 / D 0 .b a/:
jD1
Therefore, in order to get S.f I P/ s.f I P/ < it suffices to start with 0 D 2.ba/
.
t
u
As anticipated in the end of the previous section, if f W Œa; b ! R is an integrable
function, we let
Z b
f .x/dx (10.19)
a
denote the integral of f on the interval Œa; b. Note also that it does not matter
Rb
whether we denote the integral of f on Œa; b as above or writing a f .t/dt. Indeed,
such a change in notation is equivalent to changing the notation for the independent
variable of f , which certainly does not alter the value of the integral.
In spite of what we have done so far, unfortunately we do not have yet a general
procedure for computing integrals of (integrable) specific functions f W Œa; b ! R.
We shall remedy this situation in Sect. 10.5. However, see Problems 2, 4, 5 and 10.
2 p p 1 p p 1 p p
.u u v v/ .u v/. u v/ < . u C v/.u v/
3 3 2
and
1 p p 2 p p
. u C v/.u v/ < .u u v v/:
2 3
R1 1
7. We shall compute 0 1Cx 2 dx D 4 in Example 10.42. Use this fact to show that,
1 1 1
if an D n n2 C1 2 C n2 C22 C C n2 Cn2 , then the sequence .an /n1 converges
to 4 .
8. Let f W Œa; b ! R be a bounded function and ˛ > 0 be given. If for each k 2 N
there exists a partition Pk of Œa; b satisfying S.f I Pk / s.f I Pk / < ˛k , show that
f is integrable, with
Z b
f .x/dx D lim s.f I Pk / D lim S.f I Pk /:
a k!C1 k!C1
For the coming problem, the reader might find it convenient to read again
the statement of Problem 3, page 262.
9. Let f W Œa; b ! R be a Lipschitz function, with Lipschitz constant c, and let
Pk D fa D x0 < x1 < < xk D bg be a uniform partition of interval Œa; b.
2
(a) Show that S.f I Pk / s.f I Pk / < c.ba/
k .
(b) For 1 j k, choose a real number kj 2 Œxj1 ; xj and define the sum
P
†.f I Pk I k / by †.f I Pk I k / D kjD1 f .kj /.xj xj1 /. Prove that
Z b
f .x/dx D lim †.f I Pk I k /:
a k!C1
and
.k1/h
X
k cos a C 2
sin .kC1/h
2
cos.a C jh/ D
jD0
sin h2
Rb Rb
(b) Show that a sin x dx D cos a cos b and a cos x dx D sin b sin a.
11. Let f W Œ0; 1 ! R be given by
0; if x D 0 or x … Q
f .x/ D 1 :
n
; if x D mn with m; n 2 N and gcd.m; n/ D 1
ˇXk ˇ
ˇ ˇ
maxfjxj xj1 jI 1 j kg < ı ) ˇ f .j /.xj xj1 / I ˇ < : (10.20)
jD1
X
k
jPj D maxfjxj xj1 jI 1 j kg and †.f I PI / D f .j /.xj xj1 /;
jD1
we summarize (10.20) by saying that I is the limit of the Riemann sums †.f I PI /,
when jPj ! 0 and for every choice D .j / of intermediate points relative to P.
Moreover, in this case we simply write
The following result shows that the above definition of integral coincides with the
one we have adopted in the previous section. The proof of it can surely be omitted
on a first reading.
Theorem 10.7 (Riemann ) A bounded function f W Œa; b ! R is integrable (in
the sense of the previous section) if and only if the limit limjPj!0 †.f I PI / of its
Riemann sums exists, does not depending on the chosen intermediate points of .
Moreover, in this case one has
Z b
f .x/dx D lim †.f I PI /:
a jPj!0
10.3 Riemann’s Theorem and Some Remarks 361
Rb
Proof We first assume that f is integrable in Œa; b, with a f .x/dx D I.
Given > 0, we want to find ı > 0 such that, for every partition P of Œa; b, it
is the case that jPj < ı ) j†.f I PI / Ij < , for every choice of intermediate
points. To this end, start by choosing (by Cauchy’s integrability criterion) a partition
P0 D fa D x0 < x1 < : : : < xk D bg of Œa; b such that S.f I P0 / s.f I P0 / < 4 .
For a general partition P D fa D y0 < y1 < : : : < yl D bg of Œa; b, there are two
different kinds of intervals .yi1 ; yi /: those which contain at least one of the xj ’s and
those which do not.
If we choose 0 < ı < 12 jP0 j, then jPj < ı implies yi yi1 < 12 .xj xj1 / for all
1 i l, 1 j k. Therefore, each interval .yi1 ; yi / contains at most one of the
xj ’s and each interval .xj1 ; xi / contains at least three of the yi ’s. This way, we can
write
We now let M D supŒa;b jf j and estimate each of the summands at the right hand
side above.
(i) Estimating j†0 .f I PI /j: the restriction imposed on ı forces †0 .f I PI / to have
at most k1 summands (at most one for each of x1 , . . . , xk1 ). Since jf .i /j M
and yi yi1 < ı, we get
j†0 .f I PI /j D j†0 f .i /.yi yi1 /j †0 jf .i /j.yi yi1 / < Mı.k 1/:
so we are left to estimating j†00 .f I PI / s.f I P0 /j. To this end, first recall that
if 1 i l is one of the indices that fall into †00 , then .yi1 ; yi / does not
contain any of the xj ’s, but there is exactly one index 1 j k such that
Œyi1 ; yi Œxj1 ; xj . Hence, we can write
X
k
j†00 .f I PI / s.f I P0 /j j†00j .f I PI / mj .xj xj1 /j:
jD1
In order to deal with this last sum, fix 1 j k and let 1 i l be such that
Œyi1 ; yi Œxj1 ; xj . Recall that our choice of ı forces Œxj1 ; xj to contain at least
another point of P. Therefore, if rj i 1 and sj i are respectively the least and
greatest indices such that .yrj ; ysj / .xj1 ; xj /, then sj rj 2 and we have
yrj 1 < xj1 yrj < : : : < yi1 < yi < : : : < ysj xj < ysj C1 :
This way, and letting Sj D j†00j .f I PI / mj .xj xj1 /j and Mj D supŒxj1 ;xj f , we
estimate Sj by writing
ˇ X
sj
X
sj
ˇ
Sj ˇ f .t /.yt yt1 / mj .yrj xj1 / mj .yt yt1 /
tDrj C1 tDrj C1
ˇ
ˇ
mj .xj ysj /ˇ
X
sj
jf .t / mj j.yt yt1 / C jmj j.yrj xj1 / C jmj j.xj ysj /
tDrj C1
X
sj
.Mj mj /.yt yt1 / C 2jmj jı
tDrj C1
where we used the facts that jf .t / mj j Mj mj , 0 < yrj xj1 ; xj ysj < ı and
ysj yrj 1 xj xj1 .
Hence, we get from (10.22) that
X
k X
k
j†00 .f I PI / s.f I P0 /j Sj .Mj mj /.xj xj1 / C 2Mı
jD1 jD1
D S.f I P0 / s.f I P0 / C 2Mık < C 2Mık:
4
10.3 Riemann’s Theorem and Some Remarks 363
Finally, gathering together (10.21), (10.22) and the estimates of items (i) and (ii),
we arrive at
j†.f I PI / Ij < C Mı.3k 1/:
2
Therefore, if we further ask that 0 < ı < 2M.3k1/
, we get
n1 o
jPj < min jP0 j; ) j†.f I PI / Ij < ;
2 2M.3k 1/
as we wished to show.
The proof of the converse is left to the reader as an exercise (see Problem 9). u
t
In spite of the fact that the introduction of the integral by means of Riemann
sums was the actual way the theory historically evolved, the approach by means of
lower and upper sums makes it technically much easier to deal with the material of
the next section. Nevertheless, as we shall see right now, the characterization of the
integral by Riemann’s theorem allows us to deal with the heuristic discussion of the
previous section (concerning area computations) in a mathematically precise way.
Given a bounded and nonnegative function f W Œa; b ! R, let R denote the
region under the graph of f (and above the horizontal axis and between the vertical
lines x D a and x D b):
Let Pk D fa D x0 < x1 < < xk D bg be a uniform partition of Œa; b, i.e., such
that xj xj1 D ba k
for 1 j k.
If f is continuous, then Theorem 8.26 guarantees that, for 1 j k, there exist
kj ; kj0 2 Œxj1 ; xj such that
Hence, the lower and upper approximations A.f I k/ and A.f I k/ for the area of R
(respectively defined by (10.1) and (10.2), page 348)) respectively coincide with the
Riemann sums †.f I Pk I k / and †.f I Pk I k0 / (here, note that k D .kj / and k0 D
.kj0 /. However, since jPk j D 1k , it follows from Riemann’s theorem that
Z b
f .x/dx D lim †.f I Pk I k / D lim A.f I k/
a k!C1 k!C1
and, analogously,
Z b
f .x/dx D lim A.f I k/:
a k!C1
Rb
Thus, a f .x/dx is the only reasonable value for the area of R.
364 10 Riemann’s Integral
On the other hand, if f is merely integrable (albeit also nonnegative), then, except
for changing the minimum and maximum values of f along Œxj1 ; xj by its infimum
and supremum along such an interval, A.f I k/ and A.f I k/ also reduce to the lower
and upper sums s.f I Pk / and S.f I Pk /, respectively.
We sum up the above discussion in the following
Definition 10.8 Let f W Œa; b ! R be a nonnegative integrable function. We define
the area of the region R under its graph as
Z b
A.R/ D f .x/dx:
a
In the next section, the above definition will give heuristic arguments towards the
validity of some of the operational properties of Riemann’s integral.
II. LEBESGUE’S THEOREM AND INTEGRAL. In Theorem 10.6, we proved
that every continuous function f W Œa; b ! R is integrable. We also saw, in
Example 10.3, that the Dirichlet function (which, by Problem 11, page 255, is
discontinuous at every point of the interval Œ0; 1) is not integrable. On the other
hand, we shall prove in the next section (cf. Proposition 10.19) that a piecewise
continuous function f W Œa; b ! R (i.e., a function with a finite number of points
of discontinuity x0 2 Œa; b, such that for all of them the lateral limits limx!x0 ˙ f .x/
do exist) is still integrable. A more dramatic example is given by the function of
Problem 11, page 359, which is integrable but has an infinite number of points of
discontinuity.
The last paragraph suggests that there might be a more intimate relation between
the integrability of a bounded function f W Œa; b ! R and the size of the set Df of
its points of discontinuity. This is indeed the case, and a precise formulation of such
a connection requires the concept of a set of measure zero, which we now give.
Definition 10.9 We say that X R is a set of (Lebesgue2 ) measure zero, or a
null set if, given > 0, there exist open intervals I1 , I2 , I3 , . . . such that
[ X
X Ij and jIj j < ;
j1 j1
2
After Henri Lebesgue, French mathematician of the XX century.
10.3 Riemann’s Theorem and Some Remarks 365
We shall not prove Lebesgue’s theorem here, and refer the interested reader to
[1, 9] or [20]. The reason is that, apart from the Problem 4 (which will be obtained
by more elementary methods in the coming section) and Problem 5, page 468
(for which we present two proofs, one of which does not make use of Lebesgue’s
theorem), we will not use it anywhere else in the book. Nevertheless, it is important
to realize that it gives a unified explanation for the status, with respect to integration,
of all of the functions of the next to last paragraph. Indeed, since the set of points of
discontinuity of a continuous (resp. piecewise continuous) function is empty (resp.
finite), hence of measure zero, such functions are integrable. On the other hand, it
is possible to prove that Q is a null set, while Œ0; 1 is not (for the case of Q, see
Problem 2); this explains why the function of Problem 11, page 359 is integrable
(for, its set of points of discontinuity is Q \ Œ0; 1, which—being a subset of Q—is
certainly a null set), as well as why the Dirichlet function is not integrable (for, its
set of points of discontinuity is the whole interval Œ0; 1).
Let I denote the set of irrational numbers. It is possible to prove that the set
I \ Œa; b, of the irrational numbers of the closed interval Œa; b, is not a null set.
Hence, Lebesgue’s theorem assures that a function f W Œa; b ! R whose set of
points of discontinuity is I \ Œa; b is not integrable. However, as we stressed before,
such a function does not exist (a proof can be found in [1], for instance).
Lebesgue’s contributions to integration theory go far beyond Theorem 10.10. At
the dawn of the XX century, he introduced a much more refined and flexible concept
of integral in his doctor’s thesis. In order to get a glimpse of it, we shall first of all
need the following
Definition 10.11 The characteristic function of a set A R is the function XA W
Œa; b ! R such that
1; if x 2 A
XA .x/ D :
0; if x … A
X
k X
k
fP; D mj XŒxj1 ;xj and fP;C D Mj XŒxj1 ;xj ;
jD1 jD1
where mj D infŒxj1 ;xj f and Mj D supŒxj1 ;xj f . In the next section, we shall show
(cf. Example 10.20) that such functions are Riemann integrable, with
Z b X
k Z b X
k
fP; .x/dx D mj .xj xj1 / and fP;C .x/dx D Mj .xj xj1 /:
a jD1 a jD1
366 10 Riemann’s Integral
moreover, in this case the integral of f coincides with this common value.
On the contrary, in the realm of Lebesgue’s integral, we start with a partition
Q D f0 D y0 < y1 < < yl D Lg of the codomain Œ0; L of f . Then, for
1 j k 1 we take the inverse image Aj of the interval Œyj1 ; yj / by f , which is
defined by
If these sets are not too complicated (in a sense that doesn’t interest us at this
moment), one can show that it is possible to associate to Aj a nonnegative real
number m.Aj /, which we call the Lebesgue measure of Aj . Moreover, this is done
in such a way that if Aj is an interval then m.Aj / coincides with the length of Aj ; in
this case, we say that Aj is a (Lebesgue) measurable set.
Assuming that all of the Aj ’s are measurable, we consider the simple function
X
k
fQ D yj1 XAj
jD1
R
and define its Lebesgue integral, denoted Œa;b fQ , by
Z X
k
fQ D yj1 m.Aj /: (10.23)
Œa;b jD1
Finally, if for every partition Q of Œ0; L the resulting sets Aj are measurable,
then we say that f is a Rmeasurable function; if this is so, we define the Lebesgue
integral of f , denoted Œa;b f , by
Z Z
f D sup fQ I Q is a partition of Œ0; L : (10.24)
Œa;b Q
Although we have “defined” the Lebesgue integral just for nonnegative bounded
functions, it is possible to consider the notion of Lebesgue integral in the context
of bounded measurable (but not necessarily nonnegative) functions f W Œa; b !
R. (Actually, we could consider the even more general case of an unbounded
measurable function, but this will not play a role here).
10.3 Riemann’s Theorem and Some Remarks 367
This section is devoted to the derivation of some useful operational properties for
the Riemann integral. In spite of the fact that such properties will be extensively
used along the rest of the book, their proofs can be omitted in a first reading, with
essentially no loss of continuity.
Along all of this section, given bounded functions f ; g W Œa; b ! R and a
partition P D fa D x0 < x1 < < xk D bg of Œa; b, we shall denote the
infimums of f and g in Œxj1 ; xj by mj .f / and mj .g/, respectively; accordingly, Mj .f /
and Mj .g/ will denote the supremums of f and g in the same interval Œxj1 ; xj , also
respectively.
If f ; g W Œa; b ! R are nonnegative integrable functions, with f g, then the
regions Rf e Rg , respectively situated under the graphs of f and g, are such that
Rf Rg ; we therefore expect that A.Rf / A.Rg /. This is indeed the case, and
we usually refer to it by saying that the Riemann integral is monotonic. The general
case is as follows.
Proposition 10.12 If f ; g W Œa; b ! R are integrable functions such that f g,
Rb Rb
then a f .x/dx a g.x/dx.
Proof Letting P D fa D x0 < x1 < < xk D bg denote a partition of Œa; b, it
follows at once from f g that Mj .f / Mj .g/ for 1 j k. Hence,
X
k X
k
S.f I P/ D Mj .f /.xj xj1 / Mj .g/.xj xj1 / D S.gI P/:
jD1 jD1
Rb
However, since a f .x/dx S.f I P/, we conclude that
Z b
f .x/dx S.gI P/;
a
10.4 Operating with Integrable Functions 369
Rb
for every partition P of Œa; b. Thus, a f .x/dx is a lower bound for the set of the
upper sums S.gI P/, so that
Z b Z b
f .x/dx inf S.gI P/ D g.x/dx:
a P a
t
u
For what comes next, we consider again a nonnegative integrable function f W
Œa; b ! R and a positive real number c. Problem 10, page 194, guarantees that
the region Rcf under the graph of the function cf can be obtained from the region
Rf under the graph of f by vertically stretching Rf by a factor c; we thus hope that
A.Rcf / D cA.Rf /. On the other hand if g W Œa; b ! R is another integrable and
nonnegative function, then, for each x0 2 Œa; b, the segment of the vertical line
x D x0 contained in Rf Cg (the region under the graph of f C g) has length equal
to the sum of the lengths of the segments of such a line which are contained in Rf
and Rg ; thus suggests that Rf Cg can be obtained by glueing Rg right above Rf and,
hence, that we should have A.Rf Cg / D A.Rf / C A.Rg /.
The next result shows that the two properties of the Riemann integral hinted by
the heuristic reasonings above are actually true. From now on, we shall refer to these
properties by saying that the Riemann integral is respectively linear and additive.
Proposition 10.13 Let f ; g W Œa; b ! R be integrable, and let c 2 R. Then:
Rb Rb
(a) cf W Œa; b ! R is integrable, with a cf .x/dx D c a f .x/dx.
Rb Rb Rb
(b) f Cg W Œa; b ! R is integrable, with a .f .x/Cg.x//dx D a f .x/dxC a g.x/dx.
Proof (a) As in the proof of the previous proposition, let P D fa D x0 < x1 <
< xk D bg be a partition of Œa; b.
If c > 0, it is immediate to verify (see Problem 12, page 207) that
X
k X
k
s.cf I P/ D mj .cf /.xj xj1 / D c mj .f /.xj xj1 / D c s.f I P/
jD1 jD1
and, analogously, S.cf I P/ D c S.f I P/. The integrability of f and the elementary
properties of the concepts of supremum and infimum (see Problem 12, page 207,
again) give us
Z b
sup s.cf I P/ D sup.c s.f I P// D c sup s.f I P/ D c f .x/dx
P P P a
370 10 Riemann’s Integral
and
Z b
inf S.cf I P/ D inf.c S.f I P// D c inf s.f I P/ D c f .x/dx:
P P P a
If c < 0, the deduction of the two relations above is entirely analogous; one just
needs to observe that mj .cf / D c Mj .f /, Mj .cf / D c mj .f /, s.cf I P/ D c S.f I P/ and
S.cf I P/ D c s.f I P/, so that
Z b
sup s.cf I P/ D sup.c S.f I P// D c inf S.f I P/ D c f .x/dx
P P P a
Rb
and, in the same way, infP S.cf I P/ D c a f .x/dx.
Rb
In any case, we have infP S.cf I P/ D supP s.cf I P/ D c a f .x/dx. Therefore, cf
is integrable, with
Z b Z b
cf .x/dx D inf S.cf I P/ D c f .x/dx:
a P a
(b) Let P and Q be partitions of Œa; b such that P [ Q D fa D x0 < x1 < <
xk D bg. It readily follows from Problem 13, page 207, that
X
k
S.f C gI P [ Q/ D Mj .f C g/.xj xj1 /
jD1
X
k
.Mj .f / C Mj .g//.xj xj1 /
jD1
D S.f I P [ Q/ C S.gI P [ Q/
S.f I P/ C S.gI Q/:
However, since infR S.f C gI R/ S.f C gI P [ Q/, the inequalities above give
In the last inequality above, taking the infimum over all partitions P and Q of
Œa; b and invoking Problem 13, page 207, again, we get
Finally, since supR s.f C gI R/ infR S.f C gI R/, it follows from (10.25)
and (10.26) that
Z b Z b
inf S.f C gI R/ D sup s.f C gI R/ D f .x/dx C g.x/dx:
R R a a
Rb Rb Rb
Thus, f C g is integrable, with a .f .x/ C g.x//dx D a f .x/dx C a g.x/dx. t
u
An obvious corollary of the previous proposition is that, if f ; g W Œa; b ! R are
integrable functions, then f g W Œa; b ! R is also integrable, with
Z b Z b Z b
.f .x/ g.x//dx D f .x/dx g.x/dx: (10.27)
a a a
Indeed, item (a) guarantees the integrability of f , whereas item (b) guarantees that
of f g D f C .g/. Relation (10.27) now follows from the formulas of items (a)
and (b) of the previous proposition:
Z b Z b Z b
.f .x/ g.x//dx D f .x/dx C .g.x//dx
a a a
Z b Z b
D f .x/dx C .1/ g.x/dx:
a a
(For another proof of (10.28) we refer the reader to Problem 7.) Analogously,
Z b Z b Z b Z b
f .x/dx D .f /.x/dx j.f /.x/jdx D jf .x/jdx;
a a a a
so that
Z b Z b Z b
jf .x/jdx f .x/dx jf .x/jdx;
a a a
Mj .fC / mj .fC / Mj .f / mj .f /:
Mj .fC / mj .fC / D 0 Mj .f / mj .f /:
X
k
S.fC I P/ s.fC I P/ D .Mj .fC / mj .fC //.xj xj1 /
jD1
X
k
.Mj .f / mj .f //.xj xj1 /
jD1
Now, since f is integrable, Cauchy’s criterion for integrability assures that, given
> 0, we can choose the partition P in such a way that S.f I P/ s.f I P/ < .
However, this being the case, it follows from the last inequality above that S.fC I P/
s.fC I P/ < . Therefore, again by Cauchy’s criterion, we conclude that fC is
integrable. t
u
Example 10.15 Given a function f W Œa; b ! R, it may well happen that jf j is
integrable but f is not. The classical example is furnished by f W Œ0; 1 ! R such that
1; if x … Q
f .x/ D :
1; if x 2 Q
letting R, RjŒa;c and RjŒb;c denote the regions of the cartesian plane respectively
under the graphs of f and of its restrictions to Œa; c and Œc; b, we have R D RjŒa;c [
RjŒb;c , such that RjŒa;c and RjŒb;c have no interior points in common; therefore, one
expects that A.R/ D A.RjŒa;c / C A.RjŒb;c /.
Proposition 10.16 Let a function f W Œa; b ! R and a real number c 2 .a; b/ be
given. If the restrictions of f to the intervals Œa; c and Œc; b are both integrable, then
f is also integrable, with
Z b Z c Z b
f .x/dx D f .x/dx C f .x/dx:
a a c
Proof Let fjŒa;c and fjŒc;b denote the restrictions of f to the intervals Œa; c and Œc; b,
respectively. Given > 0, the integrabilities of fjŒa;c and fjŒc;b assure, by means of
the Cauchy criterion, the existence of a partition P of Œa; c and Q of Œc; b such that
S.fjŒa;c I P/ s.fjŒa;c I P/ < and S.fjŒc;b I Q/ s.fjŒc;b I Q/ < :
2 2
If R D P [ Q, then R is a partition of Œa; b and it is immediate that
Hence,
Therefore, jj < . Finally, since > 0 was chosen arbitrarily, we conclude that
D 0. t
u
10.4 Operating with Integrable Functions 375
The previous proposition shall let us present yet another class of examples of
integrable functions. However, before we can do so, we need the following auxiliary
result.
Lemma 10.17 If f ; g W Œa; b ! R are such that f is integrable and f D g in .a; b/,
Rb Rb
then g is integrable and a f .x/dx D a g.x/dx.
Proof Let’s consider the case in which f D g in Œa; b/, leaving the general case to
the reader (cf. Problem 8).
If P D fa D x0 < x1 < < xk D bg is a partition of Œa; b, the coincidence of f
and g in Œx0 ; xk1 furnishes
ˇXk ˇ
ˇ ˇ
js.f I P/ s.gI P/j D ˇ inf f inf g .xj xj1 /ˇ
Œxj1 ;xj Œxj1 ;xj
jD1
ˇ ˇ
ˇ ˇ
D ˇ inf f inf g .xk xk1 /ˇ
Œxk1 ;xk Œxk1 ;xk
analogously,
It follows from the computations above and the triangle inequality that
For the equality of the integrals of f and g, note that (again by the computations
above)
Z b
f .x/dx S.f I P/ D .S.f I P/ S.gI P// C S.gI P/
a
However, since > 0 was chosen arbitrarily, it follows from the above that
Z b Z b
f .x/dx g.x/dx 0:
a a
Finally, changing the roles of f and g in the reasoning above (which is perfectly
valid, for we have already established the integrability of g), we get the opposite
inequality between the integrals, thus showing that they are equal. t
u
We now need the coming
Definition 10.18 A function f W Œa; b ! R is piecewise continuous if there exist
real numbers a D x0 < x1 < < xk D b such that f is continuous in the
interval .xj1 ; xj /, for 1 j k, and the lateral limits limx!aC f .x/, limx!b f .x/
and limx!xj ˙ f .x/ exist, for 1 j < k.
Figure 10.4 sketches the graph of a piecewise continuous function f W Œa; b ! R,
which is discontinuous in exactly three points.
Proposition 10.19 If f W Œa; b ! R is piecewise continuous, then f is inte-
grable, with
Z b k Z
X xj
f .x/dx D f .x/dx: (10.30)
a jD1 xj1
10.4 Operating with Integrable Functions 377
a x1 x2 b x
X
k
f .x/ D cj XŒxj1 ;xj ;
jD1
Z b k Z
X xj
f .x/dx D f .x/dx:
a jD1 xj1
Our next result establishes the converse of Proposition 10.16 and will be of
crucial importance for the actual computation of integrals, in the next section.
Proposition 10.21 Let f W Œa; b ! R be integrable and a < c < b. Then, the
restrictions of f to the intervals Œa; c and Œc; b (which, whenever no danger of
confusion appears, will be denoted simply by f ) are also integrable, with
Z b Z c Z b
f .x/dx D f .x/dx C f .x/dx: (10.31)
a a c
Proof It suffices to prove the integrability of the restrictions of f to Œa; c and Œc; b.
Indeed, once we have done this, (10.31) will follow from Proposition 10.16.
For what is left to do, given > 0, Cauchy’s criterion assures the existence of a
partition R of Œa; b such that S.f I R/ s.f I R/ < . Letting R0 D R [ fcg, it follows
from Lemma 10.1 that S.f I R0 / s.f I R0 / S.f I R/ s.f I R/ < ; therefore, we can
suppose from the very beginning that c 2 R.
Let fjŒa;c denote the restriction of f to the interval Œa; c. If R D fa D x0 < x1 <
< xl D c < < xk D bg and P D fa D x0 < x1 < < xl D cg, then P is a
partition of Œa; c, such that
X
l
S.fjŒa;c I P/ s.fjŒa;c I P/ D .Mj .f / mj .f //.xj xj1 /
jD1
X
k
.Mj .f / mj .f //.xj xj1 /
jD1
(quite similarly as in the above reasoning) that f 2 is integrable if .f C c/2 has this
property. It then suffices to show that, if f is integrable and nonnegative, then f 2 is
also integrable.
For what is left to do, assume f to be integrable and nonnegative, so that f 2 is
surely bounded. Now, given a partition P D fa D x0 < x1 < < xk D bg of Œa; b,
it follows from Problem 14, page 207, that
X
k
S.f 2 I P/ s.f 2 I P/ D .Mj2 m2j /.xj xj1 /
jD1
X
k
D .Mj C mj /.Mj mj /.xj xj1 /
jD1
X
k
2 sup f .Mj mj /.xj xj1 /
Œa;b jD1
Let > 0 be given. By invoking Cauchy’s criterion again, and thanks to the
integrability of f , we can suppose that S.f I P/ s.f I P/ < 2.sup f /C1 . Therefore,
Œa;b
the above computations give
S.f 2 I P/ s.f 2 I P/ < 2.sup f / < :
Œa;b 2.supŒa;b f / C 1
RC 2
f D f.x; y/ 2 R I a x b and 0 y f .x/g
380 10 Riemann’s Integral
and
R 2
f D f.x; y/ 2 R I a x b and f .x/ y 0g;
prove that
Z b
f .x/dx D A.RC
f / A.Rf /:
a
For the next problem, given continuous functions f ; g W Œa; b ! R, such that
f .x/ g.x/ for every x 2 Œa; b, let
be the portion of the cartesian plane situated between the graphs of f and g. Let
the area of Rfg be defined by
Z b
A.Rfg / D .g.x/ f .x//dx:
a
Moreover, if f and g are continuous, show that equality happens if and only if
there exists 2 R such that f .x/ D g.x/ for every x 2 Œa; b.
13. * If f W Œa; b ! R is a continuous and convex function, prove that
Z b
f .a/ C f .b/ 1
f .x/dx:
2 ba a
The main purpose of this section is to turn (10.11) into a theorem. Along all that
follows, unless explicitly stated otherwise, we shall let I be an interval of the real
line and f W I ! R be integrable in each interval Œa; b I.
For c 2 I and Œa; b I, we set
Z c Z a Z b
f .x/dx D 0 and f .x/dx D f .x/dx:
c b a
With these conventions at hand, and applying the second part of Proposition 10.21,
it is immediate to verify that
Z b Z c Z b
f .x/dx D f .x/dx C f .x/dx; (10.32)
a a c
for every a; b; c 2 I (i.e., not only when a < b and c 2 .a; b/).
This being said, the following definition will be of paramount importance for
what is to come.
Definition 10.23 Let I R be an interval and f W I ! R be integrable in each
interval Œa; b I. For a fixed c 2 I, the indefinite integral of f based at c is the
function F W I ! R, given by
Z x
F.x/ D f .t/dt:
c
for all a; b 2 I.
382 10 Riemann’s Integral
We can now state and prove the major result of this section, which is known in
mathematical literature as the fundamental theorem of Calculus (FTC).
Theorem 10.24 (FTC) Let I R be an interval, let f W I ! R be integrable in
each interval Œa; b I and F W I ! R be the indefinite integral of f based at c 2 I.
If f is continuous at x0 2 I, then F is differentiable at x0 , with F 0 .x0 / D f .x0 /.
Proof For x 2 I n fx0 g and successively applying (10.33), Problem 2, page 357, the
additivity of the integral and the triangle inequality for integrals, we get
ˇ F.x/ F.x / ˇ ˇ 1 Z x ˇ
ˇ 0 ˇ ˇ ˇ
ˇ f .x0 /ˇ D ˇ f .t/dt f .x0 /ˇ
x x0 x x0 x 0
Z
1 ˇˇ x ˇ
ˇ
D ˇ .f .t/ f .x0 //dtˇ
jx x0 j x0
Z
1 ˇˇ x ˇ
ˇ
ˇ jf .t/ f .x0 /jdtˇ:
jx x0 j x0
Hence, for 0 < jx x0 j < ı, we have jt x0 j < ı for every t belonging to the
interval with endpoints x0 and x, so that jf .t/ f .x0 /j < . It thus follows from
the monotonicity of the integral, and once more from the result of Problem 2, page
357, that
ˇZ x ˇ ˇZ x ˇ
ˇ ˇ ˇ ˇ
ˇ jf .t/ f .x0 /jdtˇ ˇ dtˇ D jx x0 j:
x0 x0
Finally, the above computations assure that, for x 2 I such that 0 < jx x0 j < ı,
we have
ˇ F.x/ F.x / ˇ 1
ˇ 0 ˇ
ˇ f .x0 /ˇ jx x0 j D :
x x0 jx x0 j
In terms of the previous definition, the FTC guarantees that each indefinite
integral of a continuous function f W I ! R is a primitive of f in I. Indeed, letting
F W I ! R be the indefinite integral of F based at c 2 I, it follows from the FTC
and the continuity of f in I that F 0 .x/ D f .x/ for every x 2 I. Yet in another way, we
have that:
Z x
d
f continuous ) f .t/dt D f .x/: (10.34)
dx c
The coming result assures that, even for functions which are merely integrable,
there are no other possible primitives. In other words, it guarantees that, even if f is
just an integrable function, there are no other possible solutions for (10.37).
Theorem 10.26 Let I R be an interval and f W I ! R be integrable in each
interval Œa; b I. If F W I ! R is a primitive for f , then, for a given x0 2 I, we
have
Z x
F.x/ D F.x0 / C f .t/dt (10.35)
x0
for every x 2 I.
Proof Suppose x > x0 (the case x D x0 is trivial and R x the case x R<x x0 can be dealt
with in an analogous way, taking into account that x0 f .t/dt D x 0 f .t/dt).
Let P D fx0 < x1 < < xk D xg be a partition of Œx0 ; x. Since F is
differentiable in I, it is continuous in each interval Œa; b I. Hence, Lagrange’s
MVT gives j 2 .xj1 ; xj / such that
X
k
F.x/ F.x0 / D .F.xj / F.xj1 //
jD1
X
k
D f .j /.xj xj1 /
jD1
D †.f I PI /:
Z b ˇxDb X n
ˇ aj
f .t/dt D F.x/ˇ D .bjC1 ajC1 /:
a xDa
jD0
j C 1
10.5 The Fundamental Theorem of Calculus 385
R R R
Example 10.29 Compute 0 sin x dx, 0 sin2 x dx and 0 cos2 x dx.
Solution Since cos is a primitive for sin, the previous corollary gives
Z ˇxD
ˇ
sin x dx D cos xˇ D cos C cos 0 D 2:
0 xD0
t
u
Another way of rephrasing (10.34), which is quite important for the theory of
ordinary differential equations3 is collected in the following result.
Proposition 10.30 Let I R be an interval and f W I ! R be a continuous
function. Given x0 2 I and y0 2 R, the initial value problem
y0 D f .x/
(10.37)
y.x0 / D y0
3
In spite of being quite an interesting and important subject, we shall not have occasion to give
a systematic development of the theory of ODE’s in these notes. For the interested reader, we
recommend the outstanding book of professor George Simmons, [23].
386 10 Riemann’s Integral
Proof Firstly, note that the integrals in both sides of (10.39) are well defined. Indeed,
f and g, being differentiable, are continuous; on the other hand, since f 0 and g0 are
integrable, Proposition 10.22 guarantees that f 0 g and fg0 are also integrable.
Now, since .fg/0 D f 0 g C fg0 , Corollary 10.27 gives
Z b Z b ˇxDb
ˇ
.f 0 .x/g.x/ C f .x/g0 .x//dx D .fg/0 .x/dx D f .x/g.x/ˇ :
a a xDa
Finally, in order to obtain (10.39), it suffices to apply the additivity of the integral to
the left hand side of the equality above. t
u
R R 2
Example 10.32 Compute 0 x sin x dx and 0 x cos x dx.
Solution For the first integral, letting f .x/ D cos x and g.x/ D x in (10.39), we
get
Z Z ˇxD Z
0 ˇ
x sin x dx D x. cos x/dx D x cos xˇ . cos x/dx
0 0 xD0 0
Z ˇxD
ˇ
DC cos x dx D C sin xˇ D :
0 xD0
For the second, letting f .x/ D sin x and g.x/ D x2 in (10.39), we obtain
Z Z ˇxD Z
ˇ
x2 cos x dx D x2 sin0 x dx D x2 sin xˇ 2x sin x dx
0 0 xD0 0
Z
D 2 x sin x dx D 2;
0
where we used the result of the first part in the last equality above. t
u
10.5 The Fundamental Theorem of Calculus 387
D .n 1/In2 .n 1/In :
R =2 ˇxD
ˇ 2
(b) It’s immediate that I0 D 0 dx D xˇ D 2
. By induction hypothesis,
xD0
.2m/Š
suppose that I2m D .2m mŠ/2
2
for some integer m 0. Letting n D 2m C 2 in
the recurrence relation of item (a), we get
2m C 1 2m C 1 .2m/Š
I2mC2 D I2m D
2m C 2 2m C 2 .2m mŠ/2 2
.2m C 2/.2m C 1/ .2m/Š
D m 2
.2.m C 1//2 .2 mŠ/ 2
.2m C 2/Š
D :
.2mC1 .m C 1/Š/2 2
Loosely speaking, Lambert’s ingenious idea was to use the fact that 2 is rational
to try to choose g in such a way that, on the one hand, g.0/; g.1/ 2 Z, while, on the
other,
ˇZ 1 ˇ
ˇ ˇ
ˇ .g00 .x/ C 2 g.x// sin.x/dxˇ < 1:
ˇ ˇ
0
an
g00 .x/ C 2 g.x/ D bn 2nC2 f .x/ D f .x/I
b
Hence, to reach the desired contradiction, f must be also such that
ˇZ ˇ
an ˇˇ 1 ˇ
f .x/ sin.x/dxˇˇ < 1: (10.43)
b ˇ 0
but we have no hints on the size of a, in order to get (10.43) it is natural to choose f
in such a way that
C
maxfjf .x/jI 0 x 1g ;
nŠ
for some positive constant C. Once we have done that, we will have
ˇZ ˇ
an ˇˇ 1 ˇ n
ˇ a ;
ˇ f .x/ sin.x/dx ˇ
b 0 bnŠ
We now have enough clues to search for f , and a moment’s thought shows
1 n
f .x/ D x .1 x/n
nŠ
to be a strong candidate. Indeed, for such an f we obviously have jf .x/j 1 for
0 x 1, so that it suffices to check that f .j/ .0/; f .j/ .1/ 2 Z for every j 2 ZC .
What is left to do is this is relatively easy. First of all, it follows from the binomial
formula that
2n
1 X k
f .x/ D ak x ;
nŠ kDn
2n
1 X
f .j/ .x/ D k.k 1/ : : : .k j C 1/ak xkj
nŠ kDj
X2n
jŠ kŠ
D ak xkj
kDj
nŠ .k j/ŠjŠ
2n
!
X k
D j.j 1/ : : : .n C 1/ ak xkj ;
kDj
j
4. In each of the following items, solve the indicated initial value problem in the
largest possible interval I R:
p
y0 D x y0 D sec2 x
(a) . (b) .
y.1/ D 2 y. 6 / D 1
12. (Romania) Let F be the set of continuous functions f W Œ0; ! R such that
Z Z
f .x/ sin xdx D f .x/ cos xdx D 1:
0 0
R
Compute inff 2F 0 f .x/2 dx.
R =2
13. * Let n 2 ZC and In D 0 .cos x/n dx. The purpose of this problem is to show
I
that limn!C1 nC1In D 1. To this end, do the following items:
392 10 Riemann’s Integral
Continuing with the development of the theory, we shall now discuss the change
of variables formula. As the subsequent discussion will make it evident, such a
formula will provide another useful integration technique.
Theorem 10.37 Let f W Œa; b ! R be continuous. If g W Œc; d ! Œa; b is
differentiable and g0 is integrable, then
Z g.d/ Z d
f .t/dt D f .g.s//g0 .s/ds: (10.44)
g.c/ c
Proof In order to compute the integral at the left hand side, we start by taking a
primitive F W Œa; b ! R for f , whose existence follows from the FTC. Then, thanks
to Corollary 10.27, we get
Z g.d/ Z g.d/ ˇxDg.d/
ˇ
f .t/dt D F 0 .t/dt D F.x/ˇ
g.c/ g.c/ xDg.c/
ˇsDd
ˇ
D F.g.d// F.g.c// D .F ı g/.s/ˇ :
sDc
Now, since F and g are differentiable, the chain rule guarantees that the function
F ı g W Œc; d ! R is also differentiable, with
On the other hand, since f and g are continuous, we have f ı g continuous, thus,
integrable; however, since g0 is integrable, it follows from Proposition 10.22 that
.f ı g/g0 integrable. Hence, applying Corollary 10.27 once more, we have
ˇsDd Z d Z d
ˇ 0
.F ı g/.s/ˇ D .F ı g/ .s/ds D f .g.s//g0 .s/ds:
sDc c c
t
u
10.6 The Change of Variables Formula 393
Remark 10.38 For a variant of the previous result, applicable to the case of a
piecewise continuous f , see Problem 6.
The corollary below isolates a simple, yet quite useful, particular case of the
change of variables formula.
Corollary 10.39 If f W Œa; b ! R is continuous and 2 R , then
Z b Z
1 b
f .s/ ds D f .t/ dt: (10.45)
a a
Proof For > 0 it suffices to let g W Œa; b ! Œa; b be given by g.s/ D s in
Theorem 10.37. Indeed, with such a choice we get
Z b Z g.b/ Z b Z b
f .t/ dt D f .t/ dt D f .g.s//g0 .s/ ds D f .s/ ds:
a g.a/ a a
For < 0, let g W Œa; b ! Œb; a be given by g.s/ D s. Then, arguing as in
Ra Rb R a R b
the previous case and recalling that b D a and b D a , we obtain
Z b Z a Z g.a/
f .t/ dt D f .t/ dt D f .t/ dt
a b g.b/
Z a Z b
D f .g.s//g0 .s/ ds D f .s/ ds:
b a
t
u
We shall sometimes refer to (10.44) as the formula for integration by substitu-
tion. In what follows, we give some heuristics on such an expression.
Rb
Given a continuous function f W Œa; b ! R, in order to compute a f .t/dt with
the aid of the FTC we need to obtain a primitive for F. In this sense, and after a
careful examination of the expression of f .t/, we may eventually notice that it will
be quite simplified if we perform an adequate substitution of variables, changing t
by an expression g.s/, depending on a new variable s.
However, since f is defined in the interval Œa; b, for such a substitution to make
sense it is necessary that g.s/ belongs to the interval Œa; b when s varies in some
interval Œc; d. In general, the best way of identifying such an interval Œc; d is solving
inequalities a g.s/ b, thus finding the appropriate values of c and d. Then, by
formally differentiating the equality t D g.s/, we get dt D g0 .s/ ds, so that the
original integral must be corrected by the factor g0 .s/ before it can be transformed
into a new integral.
In short, the execution of both passages described below give us the formula for
integration by substitution in any particular case we encounter:
Z b substitting Z ‹ solving Z d
tDg.s/ 0 ag.s/b
f .t/ dt HHHHH f .g.s//g .s/ ds HHHHHHH f .g.s//g0 .s/ ds:
a ‹ c
394 10 Riemann’s Integral
Finally, we observe that for the last integral above to make sense, the function
s 7! f .g.s//g0 .s/ must be integrable. Hence (as we saw in the proof of the theorem)
it is natural to suppose that g is differentiable and g0 is integrable in Œc; d.
Let’s give a concrete example, showing how the informal discussion of the two
previous paragraphs simplifies the use of the change of variables formula in the
actual computation of integrals.
R 2p
Example 10.40 Compute p t sin.t2 / dt.
p
Solution Performing the substitution t D s, we get
Z 2p Z ‹ Z
2 p 1 1 ‹
p
t sin.t / dt D s.sin s/ p ds D sin s ds;
‹ 2 s 2 ‹
1
p
where the correction factor 2 s
was obtained by formally differentiating equality
p 1
t D s, to get dt D 2p ds.
s
p p
p In order
p to correct
p the integration interval Œ ; 2 , we solve the inequalities
s 2 , thus getting s 4. Hence,
Z 2p Z ˇsD4
2 1 4 1 ˇ
p
t sin.t / dt D sin s ds D . cos s/ ˇ D 1;
2 2 sD
The correct endpoints of the last integral can be easily found once we note that
0 2 y 2 if and only if 0 y 2 . Nevertheless, one has to be careful, for,
x D 0 ) y D 2 and x D 2 ) y D 0. Hence, we get
Z =2 Z 0 Z =2
.sin x/n dx D .cos y/n dy D .cos y/n dy D In :
0 =2 0
t
u
10.6 The Change of Variables Formula 395
From a more formal viewpoint, the last substitution above reduces to using g W
Œ0; 2 ! Œ0; 2 such that g.x/ D 2 x. Since g.0/ D 2 , g. 2 / D 0 and g0 .x/ D x,
we have
Z =2 Z g.0/ Z 0
.sin x/n dx D .sin x/n dx D .sin g.x//n g0 .x/dx
0 g. 2 /
2
Z 0 Z
2
D .sin.=2 x// dx D n
.cos x/n dx:
2 0
Depending on the algebraic expression that defines f .x/ in terms of x, the last
integral above may be much simpler to compute that the original one.
Although the discussion on the previous paragraph has a number of useful
variations (depending on the more adequate trigonometric function to be used),
we shall not try to list them all. Instead, we collect two relevant examples of
trigonometric substitutions.
R1 1
Example 10.42 Compute 0 1Cx 2 dx.
1
Proof Recalling, from Example 9.35, that d
dx
arctan x D 1Cx2
, we may use the FTC
to get
Z 1 ˇxD1
1 ˇ
2
dx D arctan x ˇ D arctan 1 D :
0 1Cx xD0 4
t
u
396 10 Riemann’s Integral
Example 10.43 Let be the circle with center O and radius R, and choose a
cartesianp coordinate system such that O.0; 0/. If f W ŒR; R ! R is given by
f .x/ D R2 x2 , then the region Rf under the graph of f (and above the horizontal
axis is the half-disk bounded by and lying in the upper half-plane. Then,
Z R Z R p
A.Rf / D f .x/dx D R2 x2 dx:
R R
an equality that can be easily remembered from (10.46), with f in place of g and
x 7! xr1 in place of f : on the one hand, the FTC gives us
Z f .x/
tr ˇˇf .x/ 1
tr1 dt D ˇ C C D f .x/r C CI
r r
We conclude this section by observing that, apart from the problems collected
below, we shall see some other relevant examples of integration by trigonometric
substitution in the coming sections.
2. In each of the items below, compute the given primitives, together with their
maximal domains of definition:
R p
(a) x2 p1 C x3 dx.
R
(b) x5 p1 C x3 dx.
R 3
(c) R x p 1 x2 dx.
(d) cos x dx.
3. Let f W Œa; a ! R be a continuous function. Show that:
Z a Ra
2 0 f .x/dx; if f is even
f .x/dx D :
a 0; if f is odd:
R1 sin.x/ R 1
2 cos.x/
4. Compute 1 1Cx2 dx and 12 1Cx3
dx.
398 10 Riemann’s Integral
For the coming example the reader might find it useful to review the basics on
ellipses, in Chap. 6 of [4], for instance.
5. Prove that the area of an ellipse of principal axes of lengths 2a and 2b equals
ab.
6. Prove the following version of Theorem 10.37: if f W Œa; b ! R is piecewise
continuous and g W Œc; d ! Œa; b is increasing (resp. decreasing) and
differentiable, with integrable derivative, then
Z g.d/ Z d
f .t/dt D f .g.s//g0 .s/ds:
g.c/ c
This section brings a fairly standard presentation of the main properties of two of
the most important functions of Mathematics, namely, the natural logarithm and the
exponential functions, postponing to the next section some interesting and important
applications of them.
Definition 10.46 The natural logarithm function log W .0; C1/ ! R is
defined by
Z x
1
log x D dt:
1 t
10.7 Logarithms and Exponentials 399
y
1
y= t
y=t
O 1 x t
It readily follows from the above definition that log 1 D 0, log x > 0 if x > 1
and log x < 0 if 0 < x < 1. Moreover, letting f W .0; C1/ ! R denote the inverse
proportionality function, so that f .x/ D 1x for every x > 0, then, in terms of the
area of the region of the first quadrant situated under the graph of f (cf. Fig. 10.5),
we have
8
< A.fjŒ1;x /; if x > 1
log x D 0; if x D 1 :
:
A.fjŒx;1 /; if x < 1
The FTC assures that log is a differentiable function, with
1
log0 x D :
x
Therefore, log is infinitely differentiable and, since log0 > 0, the study of the first
variation of functions shows that log is increasing. Also, since
1
log00 x D < 0;
x2
it follows from Corollary 9.70 that log is strictly concave.
The coming result and their corollaries bring more important properties of the
natural logarithm.
Proposition 10.47 For x; y > 0, we have
Proof For the first part, the additivity of the integral gives
Z xy Z x Z xy Z xy
1 1 1 1
log xy D dt D dt C dt D log x C dt:
1 t 1 t x t x t
Now, if g W Œ1; y ! Œx; xy is given by g.t/ D xt, the change of variables
Theorem 10.37 furnishes
Z xy Z g.y/ Z y Z y
1 1 1 0 1
dt D dt D g .s/ds D xds
x t g.1/ t 1 g.s/ 1 xs
Z y
1
D ds D log y:
1 s
t
u
Corollary 10.48 For x > 0 and r 2 Q, we have log xr D r log x.
Proof Applying (10.48) with y D x we get log x2 D 2 log x. If we assume that
log xk D k log x for every x > 0 and some k 2 N, then letting y D xk in (10.48)
we get
1
log xr D log xm=n D log.x1=n /m D m log x1=n D m log x D r log x:
n
In order to extend this last equality for negative rational exponents, start by letting
y D x1 in (10.48) to get
t
u
10.7 Logarithms and Exponentials 401
1
lim an D lim .1=a/n D 0 and lim an D lim D C1:
n!C1 n!C1 n!C1 n!C1 an
The limits above, together with the fact that log is an increasing function,
guarantee that limx!0C log x D 1 and limx!C1 log x D C1. Hence, it follows
from Problem 7, page 290, that Im .f / D R. t
u
We now relate the number e, defined in Example 7.41, to the natural logarithm
function.
Theorem 10.50 log e D 1.
Proof First of all, note (cf. Fig. 10.6, where we sketched the portion of the graph of
t 7! 1t from t D 1 to t D 1 C 1n ) that
Z 1C 1 Z 1C 1 ˇtD1C 1
1 n 1 n
ˇ n 1 1
log 1 C D dt < dt D tˇ D 1C 1 D
n 1 t 1 tD1 n n
and, analogously
Z 1C 1 Z 1C 1
1 n 1 n 1 n 1 1
log 1 C D dt > 1
dt D D :
n 1 t 1 1C n
nC1 n nC1
Hence,
1 1 1
< log 1 C < ; (10.49)
nC1 n n
1 n
n+1
1 1
1+ n
402 10 Riemann’s Integral
n
so that nC1 < n log 1 C 1n < 1. The squeezing theorem (Problem 6, page 218)
then gives limn!C1 n log 1 C 1n D 1 or, which is the same,
1 n
lim log 1 C D 1:
n!C1 n
n
Now, since (according to Theorem 7.42) e D limn!C1 1 C 1n and log is
continuous, we have
1 n 1 n
log e D log lim 1 C D lim log 1 C D 1:
n!C1 n n!C1 n
t
u
Gathering together the information obtained so far on the natural logarithm, we
sketch its graph on Fig. 10.7.
The sketch above of the graph of the logarithm function suggests that it grows
relatively slowly. The next result quantifies this suspicion.
log x
Theorem 10.51 For an integer n > 1, we have limx!C1 p
n x D 0.
1 1 1
Proof Firstly, observe that t > t1 2n for t > 1. Hence, t < 1 for t > 1, so that,
t1 2n
for x > 1,
Z Z Z
x
1 x
1 x
1
log x D dt < 1
dt D t 2n 1 dt
1 t 1 t 1 2n1
1 ˇtDx p p
t 2n ˇ
D ˇ D 2n. 2n x 1/ < 2n 2n x:
1=2n tD1
y
y=x
log
O 1 e x
2n
Now, since limx!C1 p
2n x
D 0, the squeezing theorem (cf. Problem 6, page 218)
log x
guarantees that p
n x ! 0 when x ! C1. t
u
Since the natural logarithm function log W .0; C1/ ! R is a continuous and
increasing bijection, we can consider its inverse
called the exponential function. Theorem 8.35 and Problem 8, page 176, guarantee
that exp is also a continuous and increasing bijection. Moreover, for given x; y 2 R,
with x > 0, we have
log x D y , x D exp.y/I
1 1
exp0 .y/ D D D x D exp.y/:
log0 x 1=x
In short,
exp0 D exp;
which in particular shows that exp is infinitely differentiable. Also, since exp00 D
exp > 0, it follows from Corollary 9.70 that exp is a strictly convex function.
The coming result translates, for the exponential function, the properties of the
natural logarithm expressed in Proposition 10.47 and Corollary 10.48.
Proposition 10.52 For x; y 2 R, we have:
(a) exp.x C y/ D exp.x/ exp.y/.
(b) exp.r/ D er , if r 2 Q.
Proof
(a) Letting a D exp.x/ and b D exp.y/, we have a; b > 0 and log a D x, log b D y.
Hence,
(b) Since log and exp are inverses of each other and log e D 1, it follows from
Proposition 10.47 that
exp.x/ D ex ;
stressing that the right hand side coincides with the usual meaning of ex when x 2 Q,
and defines ex when x … Q. In this new notation, item (a) of the previous result is
written as
exCy D ex ey ;
for all x; y 2 R.
More generally, the exponential function allows us to rigorously define what
is meant by the power ax , for positive a (and real x). Indeed, in the notation just
introduced, we set
ax D ex log a ; (10.51)
so that
and
for all x; y 2 R.
Notice that if r 2 Q, say r D m
n, with m 2 Z and n 2 N, Corollary 10.48 gives
p
n p
ar D e n log a D elog
m
am
D am I
n
in other words, for x D r 2 Q definition (10.51) matches with the meaning we have
attributed to ar up to now. In particular, we have a0 D 1 and a1 D a.
10.7 Logarithms and Exponentials 405
Observe that the requisite of f .x/ being positive gives sense to this definition. Also,
the chain rule applied to the right hand side of the equality above assures that the
function x 2 I 7! f .x/g.x/ is differentiable whenever f and g are so. In this respect,
see Problem 9.
We can now state and prove a useful extension of Theorem 10.50.
x
Theorem 10.53 e D limx!C1 1 C 1x .
Proof A reasoning analogous to that in the proof of Theorem 10.50 gives, for x > 0,
1 1 1
< log 1 C < : (10.54)
xC1 x x
x
It thus follows from (10.52) that log 1 C 1x D x log.1 C 1x /.
Multiplying both inequalities above by x > 0, we get
x 1 x
< log 1 C < 1;
xC1 x
p.x/ X m
xk
x
D ak x ;
e kD0
e
xn
it suffices to show that limx!C1 ex
D 0 for every n 2 N. To this end, letting y D ex
we have x D log y and, then,
n
xn .log y/n log y
D D p :
ex y n y
t
u
We close this section by deriving asymptotic estimates on the size of log n and
nŠ. In other words, we estimate the sizes of log n and nŠ for large values of n. Our
first result is due to Euler.4
Theorem 10.55 (Euler) Given n 2 N, we have
!
X
n
1
lim log n D ; (10.55)
n!C1
kD1
k
4
The Swiss mathematician Leonhard Euler is one of the most prolific mathematicians of all times,
even if we stick only to relevant Mathematics. His contributions impressively vary from Geometry
to Combinatorics, passing through Number Theory, Calculus and Physics. In each of these areas
of knowledge there is at least one, if not several, celebrated Euler’s theorem.
10.7 Logarithms and Exponentials 407
X
n
1 X
n
1 X
n1
0< log n D .log.k C 1/ log k/
kD1
k kD1
k kD1
n1
X
1 1 1
D log 1 C C :
kD1
k k n
1
Now, since n
! 0 when n ! C1, it suffices to prove that the series
X 1
1
log 1 C (10.56)
k1
k k
n1
X Xn1
1 1 1 1 1
log 1 C < D1 :
kD1
k k kD1
k kC1 n
X
n
1
log n Š ; (10.57)
kD1
k
nŠ 1 n
n < 1 C Š e:
n=2 n
408 10 Riemann’s Integral
On the other hand, since n C 1 2.n 1/ < 3.n 2/ < for n 3, we get
an an
an bn , lim D 1 and an D o.bn / , lim D 0:
n!C1 bn n!C1 bn
bn ˛an
lim D 0;
n!C1 an
bn ˛an
so that an D o.1/; consequently,
bn ˛an
bn D an C .bn an / D ˛ C an D .˛ C o.1//an : (10.58)
an
We are finally in position to prove the following result, due to the Scottish
mathematician of the XVIII century James Stirling and known in mathematical
literature as Stirling’s formula.
p n
Theorem 10.56 (Stirling) nŠ 2n ne .
n n p
D p1 or, which is the same, that
n e n
Proof We have to show that limn!C1 nŠ 2
p !
nn en 2n 1
lim log D log p :
n!C1 nŠ 2
Firstly, we shall show that the limit at the left hand side above actually exists. To
this end, note that
p
nn en n nŠ 1 .nŠ/2
log D n log n n log p D .n log n n/ log ;
nŠ n 2 n
10.7 Logarithms and Exponentials 409
with
X
n1
n log n n D Œ..k C 1/ log.k C 1/ .k C 1// .k log k k/
kD1
n1 Z
X kC1
D log x dx 1
kD1 k
1 .nŠ/2 1
log D .log.1 2/ C log.2 3/ C C log..n 1/n//
2 n 2
X
n1
log k C log.k C 1/
D :
kD1
2
Hence,
p Xn1 Z kC1
nn en n log k C log.k C 1/
log D log x dx 1: (10.59)
nŠ kD1 k 2
R kC1
For k 1, let ak D k log x dx log kClog.kC1/
2 . Problem 13 (with a D k,
b D k C 1 and f D log, which is concave) furnishes
Z kC1
log k C log.k C 1/
ak D log x dx > 0:
k 2
On the other hand, by successively applying the results of Problems 4 and 20, we get
Z kC1
1 1
log k C log x dx D log k C .k C1/ log.k C1/Ck log k C1 > 0:
2 k 2
Therefore,
1 log k C log.k C 1/
ak < log k C
2 2
2
1 k C 12 1 1
D log D log 1 C
2 k.k C 1/ 2 4k.k C 1/
1 1 1 1 1
< D ;
2 4k.k C 1/ 8 k kC1
P
We conclude that k1 ak is a series of positive terms, such that
X
n1 X
n1
1 1 1 1
ak < D :
kD1 kD1
8 k kC1 8
P
Thus, the comparison test assures that k1 ak does converge, and it follows
from (10.59) that there exists
p X
n1 X
nn en n
lim log D lim ak 1 D ak 1:
n!C1 nŠ n!C1
kD1 k1
P
For the actual computation of the desired limit, let c > 0 be such that k1 ak
1 D log c. Since
p
nn en n
lim log D log c;
n!C1 nŠ
we have
p
nn en n
c De D exp lim log
log c
n!C1 nŠ
n n p
n e n
D lim exp log
n!C1 nŠ
n n p
ne n
D lim :
n!C1 nŠ
p
It follows from (10.58) (with an D nn en n, bn D nŠ and ˛ D 1c ) that
1 p
nŠ D C o.1/ nn en n:
c
R =2
Now recall that, according to Example 10.35, if In D 0 .cos x/n dx for n 2 N, then
I2k ..2k/Š/2 2k
D
I2k1 24k .kŠ/4 2
h p i2
1 2k 2k
c C o.1/ .2k/ e 2k 2k
D h p i4 2
24k 1c C o.1/ kk ek k
2 k 2
D 2 ! 2c :
1
c C o.1/
k
However, Problem 13, page 391 showed that I2k1 I2k
! 1. Therefore, the above
computations give 2c D 1 or, which is the same, c D p1 .
2
t
u
2
Example 10.57 For n 2 N, the binomial number 2n n
is the largest of the 2n C 1
coefficients of the binomial expansion of .x C y/2n . This gives
! 2n
!
2n X 2n
.2n C 1/ > D 22n ;
n kD0
k
22n
so that 2n
n
> 2nC1 . Stirling’s formula considerably improves this estimate, for, with
the aid of it, we get
! p 2n
2n .2n/Š 2 2n 2ne 22n
D p 2 2n
D p :
n .nŠ/2 n
2n ne
log x 1
loga x D and log0a x D ; (10.60)
log a x log a
and, hence,
Z
1
sec3 t dt D .sec t tan t log j sec t tan tj/ C C:
2
16. * The hyperbolic sine and hyperbolic cosine, respectively denoted sinh; cosh W
R ! R, are defined by
ex ex ex C ex
sinh x D and cosh x D :
2 2
10.7 Logarithms and Exponentials 413
y y
x x
The next problem explains the names hyperbolic sine and hyperbolic cosine.
To put it in context, recall (cf. Fig. 10.8, left) that the area of the circular sector
of radius 1, contained in the first quadrant of the cartesian plane, centered at
the origin and bounded by the radii that join the origin to the points .1; 0/ and
.cos t; sin t/ equals 2t .
17. For all t 2 R, observe that the point .cosh t; sinh t/ belongs to the right branch
of the hyperbola x2 y2 D 1, as well as to the straight line y D .tanh t/x, where
tanh t D cosh
sinh t
t
. For a fixed t > 0, let A be the area of the bounded portion of the
cartesian plane bounded by the horizontal axis, the right branch of x2 y2 D 1
and the straight line y D .tanh t/x (cf. Fig. 10.8, right). Show that A D 2t .
18. Given > 0, show that limx!0C x log x D 0.
5
For the reader’s curiosity, the surface of revolution generated by the rotation of the catenary along
the horizontal axis is an example of a minimal surface, in this case a catenoid. Minimal surfaces
have very interesting properties, some of which can be found in the excellent book of professor
M. P. do Carmo [8].
414 10 Riemann’s Integral
X
n
1 1
0 < log n < :
kD1
k n
10.8 Miscellaneous
This section gathers some interesting applications of the natural logarithm and
exponential function. We start by examining three examples which show that the
fact that we now have logarithms and exponentials at our disposal gives greater
flexibility to the material discussed so far.
Example 10.58 (Thailand) Find all functions f W N ! R satisfying the following
conditions:
(a) f is increasing.
(b) f .1/ D 1 and f .2/ D 4.
(c) f .xy/ D f .x/f .y/ for all x; y 2 N.
Solution An easy induction gives f .2k / D 22k , for every k 2 N.
Take an integer a > 2 and suppose, for the sake of contradiction, that f .a/
a2 1. Since Q is dense in R, we can choose m; n 2 N such that
log 22 m log 22
< < ;
log a2 n log.a2 1/
so that 2n < am and .a2 1/m < 22n . Hence, successive applications of (a) and (c)
give us
which is a contradiction.
A similar reasoning shows that f .a/ a2 C 1 also leads to a contradiction, so
that the only left possibility is f .a/ D a2 . u
t
10.8 Miscellaneous 415
The next example shows how to use logarithms to guarantee the existence of
powers of a fixed (and almost arbitrary) given basis beginning (from the left) by
a prescribed set of algarisms. At this point, the reader might find it helpful to
review the material of Sect. 7.3, specially Corollary 7.33, as well as the statement of
Problem 1, page 411.
Example 10.59 Let a > 1 be an integer which is not a power of 10, and s be a
finite sequence of algarisms, the first of which is nonzero. Prove that there exists a
power of a whose decimal representation starts, from the left, with the sequence s
of algarisms.
Proof We want to show that there exist m; n 2 N for which
s 10m an .s C 1/ 10m :
Taking logarithms in base 10, this is the same as showing that there exist naturals m
and n for which log10 s C m n log10 a log10 .s C 1/ C m or, which is the same,
If we show that log10 a is irrational, the proof will be finished thanks to Corol-
lary 7.33 of Kronecker’s lemma.
Suppose, by contradiction, that log10 a 2 Q, say log10 a D pq , with p; q 2 N
relatively prime. Then aq D 10p , and the Fundamental Theorem of Arithmetic (cf.
introduction to Chap. 1), allow us to write a D 2k 5l b, for some nonnegative integers
k; l and some natural b relatively prime with 10. But these two facts give the equality
2kq 5lq bq D 2p 5p , from which we deduce that b D 1, kq D p and lq D p. Thus, we
conclude that k D l, which would give us a D 2k 5k D 10k , a contradiction. t
u
We now refine the kind of argument of the previous example to establish the best
possible converse of Problem 16, page 11.
Example 10.60 Prove that there exist infinitely many naturals n such that the
decimal representations of 2n and 5n start (at left) with a 3.
Proof If 2n and 5n are to begin with a 3, it is necessary and sufficient that there exist
naturals k and l such that 3 10k < 2n < 4 10k and 3 10l < 5n < 4 10l or, taking
logarithms in base 10, that
log10 3 < k C n log10 2 < log10 4 and log10 3 < l C n log10 5 < log10 4:
In turn, this is equivalent to asking that the fractional parts6 of both n log10 2 and
n log10 5 belong to the interval .log 3; log 4/.
6
Recall that the fractional part of x 2 R, denoted fxg, is defined as fxg D x bxc, and belongs to
the interval Œ0; 1/.
416 10 Riemann’s Integral
Since n log10 2 C n log10 5 D n and fng D 0, such fractional parts are at the same
distance from 12 . Finally, since 12 2 .log10 3; log10 4/, we conclude that it suffices to
show that there exist infinitely many naturals k and n such that
p x 1 C x2 C C xn
n
x1 x2 : : : xn ;
n
with equality if and only if x1 D x2 D D xn .
Essentially the same idea as that of the proof of the previous example allows one
to obtain the important Young’ inequality.7
7
After William H. Young, English mathematician of the XIX and XX centuries.
10.8 Miscellaneous 417
1 1
Proposition 10.62 (Young ) Let p and q be positive reals such that p C q D 1.
Given a; b > 0, we have
ap bq
ab C ; (10.63)
p q
ap bq
C ab;
p q
!1=p !1=q
X
n X
n
p
X
n
q
ai bi ai bi ;
iD1 iD1 iD1
8
After Otto Hölder, German mathematician of the XIX and XX centuries.
418 10 Riemann’s Integral
X
n Xn
ai bi
ai bi AB , 1:
iD1 iD1
A B
Now, letting xi D ai
A
and yi D bi
B
, we have
X
n
p 1 X p
n Xn
q 1 X q
n
xi D a D 1; y D b D1
iD1
Ap iD1 i iD1
i
Bq iD1 i
Pn
and want to prove that iD1 xi yi 1. But this is straightforward from Young’s
inequality:
X n p
X q
1X p 1X q
n n n
x y
xi yi i
C i D x C y D 1:
iD1 iD1
p q p iD1 i q iD1 i
p q
To have equality, we must have xi D yi for 1 i n or, which is the same,
p q
ai bi
Ap D Bq for 1 i n. Yet in another way, we must have
p p
a1 a2 apn Ap
q D q D ::: D q D q:
b1 b2 bn B
Conversely, it’s immediate to verify that if the above condition is satisfied then we
do have equality. t
u
A version of Hölder’s inequality for integrals will be the object of Problem 8 of
the next section.
We finish this section by discussing two more elaborate applications of Jensen’s
inequality, the first of which appeared as a problem at the 2001 IMO. We divide it
into two examples.
Example 10.64 Sketch the graph of the function f W R ! R given by
1
f .x/ D p :
1 C 8ex
Solution Firstly, note that 0 < f .x/ < 1 for every x 2 R. Moreover, since
limx!C1 ex D C1 and limx!1 ex D 0, we have
It follows that the graph of f is entirely contained in the strip of the cartesian plane
bounded by the horizontal lines y D 0 and y D 1, and that these lines are asymptotes
10.8 Miscellaneous 419
to the graph. On the other hand, since f .0/ D 13 , the graph intersects the vertical axis
at the point 0; 13 .
Now, an immediate computation with the aid of the chain rule gives
4ex
f 0 .x/ D < 0;
.1 C 8ex /3=2
for every x 2 R, so that f decreases along the whole real line. Also, computing f 00
(again with the aid of the chain rule) we get
4ex
f 00 .x/ D .1 4ex /:
.1 C 8ex /5=2
Therefore, the sign of f 00 coincides with that of 4ex 1, so that f is strictly convex
in .2 log 2; C1/, strictly concave in .1; 2 log 2/ and has x0 D 2 log 2 as its
only inflection point.
Finally, noticing that 2 log 2 Š 1:38, f .2 log 2/ D p1 Š 0:58 and gathering
3
together the above information, we get Fig. 10.9 as a reasonably accurate sketch for
the graph of f .
t
u
We now come to the promised IMO problem.
Example 10.65 (IMO) Prove that, for every positive reals a, b and c, one has
a b c
p Cp Cp 1:
a2 C bc b2 C ac c2 C ab
Proof Note, first of all, that the expression at the left hand side of the inequality
equals
1 1 1
q Cq Cq :
1 C 8 abc2 1 C 8 bac2 1 C 8 ab
c2
y=1
0, 58
−1, 38 x
Letting bc
a2
D ex , ac
b2
D ey and ab
c2
D ez we have x C y C z D 0 and wish to
prove that
2
f .x/ C f .y/ C f .z/ f .x/ C f .y/ 2f .2 log 2/ D p > 1:
3
and
1 2
g.2 log 2/ D p C p > 1:
3 17
so that g0 .x/ < 0 if and only if 2 log 7 < x < 2 log 2. Hence,
and
x 2 .2 log 7; 2 log 2/ ) g.x/ > g.2 log 2/ > 1:
t
u
9
The subsequent discussion derives an inequality which widely generalizes that
between the arithmetic and geometric means and is known as the inequality of
power means. We start by defining what one means for such power means.
Definition 10.66 Given positive reals a1 ; a2 ; : : : ; an and ˛ 2 R, we define the ˛-th
power mean of a1 , a2 , . . . , an as the positive real number M˛ D M˛ .a1 ; : : : ; an /,
such that
8
< a˛1 Ca˛2 CCa˛n 1=˛
; if ˛ ¤ 0
M˛ D p n :
: n a a : : : a ; if ˛ D 0
1 2 n
with equality in any of the inequalities above if and only if all of the ai ’s are equal.
Proof We can assume, without loss of generality, that not all of the ai ’s are equal.
Let’s momentarily assume that (10.64) is true (with strict inequalities) for all
0 < ˛ < ˇ. Given reals x < y < 0, we have x > y > 0; since a11 ; : : : ; a1n are also
positive and not all equal, it follows from our assumptions that
1 1 1 1 1 1
min < My ;:::; < Mx ;:::; < max :
1in ai a1 an a1 an 1in ai
9
Another such remarkable generalization, due to Newton and McLaurin, is the object of Sect. 17:2
of [5].
422 10 Riemann’s Integral
Hence, we conclude that it suffices to analyse the cases 0 < ˛ < ˇ and 0 D ˛ < ˇ.
Also without loss of generality, assume that a1 D max1in fai g. Since not all of
the ai ’s are equal, we get
ˇ ˇ ˇ
!1=ˇ ˇ
!1=ˇ
a1 C a2 C C an na1
Mˇ .a1 ; : : : ; an / D < D a1 D max fai g
n n 1in
and, analogously, min1in fai g < M˛ . Thus, we are left to showing that
Letting bi D .ai =K/ˇ , we conclude that not all of the bi ’s are equal, but satisfy
ˇ ˇ ˇ
!
b1 C b2 C C bn 1 a1 C a2 C C an
D ˇ D 1I
n K n
˛=ˇ
moreover, since .ai =K/˛ D bi , we want to show that
To what is left to do, let f W .0; C1/ ! R be given by f .x/ D x˛=ˇ . Since
0 < ˛=ˇ < 1, two applications of the result of Problem 8, page 412, give
00 ˛ ˛
f .x/ D 1 x˛=ˇ2 < 0;
ˇ ˇ
f .˛/ D M˛ .a1 ; : : : ; an /;
To this end, the continuity of x 7! ex and x 7! log x assures that we need only
show that
However, since
a˛1 C a˛2 C C a˛n
lim log D log 1 D 0;
˛!0 n
˛ ˛
1 a1 C a˛2 C C a˛n d a1 C a˛2 C C a˛n
lim log D lim log
˛!0 ˛ n ˛!0 d˛ n
.log a1 /a˛1 C C .log an /a˛n
D lim
˛!0 a˛1 C C a˛n
log a1 C C log an
D
n
log.a1 a2 : : : an /
D :
n
1. Prove that equation a2 D 2a has exactly three real roots, one of which is
irrational.
2. (Israel) Find all real solutions of the system of equations
8 p
< x C log.x C px2 C 1/ D y
y C log.y C p y2 C 1/ D z :
:
z C log.z C z2 C 1/ D x
10. Let x, y and z be positive reals whose sum of squares q equals 8. Prove that the
2
sum of their cubes is greater than or equal to 16 3 .
11. Prove the following generalization of the inequality between the arithmetic
and geometric means, known as the weighted arithmetic-geometric means
inequality: given positive reals a1 , a2 , . . . , an and naturals k1 , k2 , . . . , kn such
that k11 C k12 C C k1n D 1, we have
ak11 a k2 a kn
C 2 C C n a1 a2 : : : an ;
k1 k2 kn
Yn
1 xj 2n 1 n
:
jD1
1 C xj 2n C 1
16. (IMO shortlist) Let a1 , a2 , . . . , an be real numbers in the interval 0; 12 , with
a1 C a2 C C an < 1. Prove that
a1 a2 : : : an Œ1 .a1 C a2 C C an / 1
nC1 :
.a1 C a2 C C an /.1 a1 /.1 a2 / : : : .1 an / n
426 10 Riemann’s Integral
17. Prove Minkowski’s inequality10 : given an integer k > 1 and positive reals a1 ,
a2 , . . . , an , b1 , b2 , . . . , bn , we have
v v v
u n u n u n
uX uX uX
t
k
.ai C bi /
k t
k
ai C t
k k
bki ;
iD1 iD1 iD1
The purpose of this section is to extend Riemann’s integral to functions which are
not necessarily bounded and are defined in arbitrary intervals.
If I R is an interval, then I is of one of the forms ŒA; B, .A; B/, ŒA; B/, .A; B,
.A; C1/, ŒA; C1/, .1; A/, .1; A or .1; C1/ D R, with A; B 2 R. In
order to avoid the need of examining each one of these cases separately, and for
general notational convenience, we:
(i) shall refer to the real numbers A, B, or to the symbols 1, C1 as the endpoints
of I (albeit 1 and C1 are just formal symbols, and not real numbers), which
will be generically denoted by ˛ and ˇ;
(ii) impose that 1 < A; B < C1, for all A; B 2 R.
Thus, if I has endpoints ˛ and ˇ, with ˛ < ˇ, then
For instance, if I has endpoints ˛ D A and ˇ D C1, we have .A; C1/ I and
I n .A; C1/ fA; C1g, so that I D .A; C1/ or ŒA; C1/.
Let I be a fixed interval and f W I ! R be a given function, such that f is
integrable in each interval Œa; b I. (In particular, if I D Œa; b, then we assume
that f is integrable in I.) This implies that f is bounded in each interval Œa; b I,
albeit it may well happen that f is not bounded in I. (Nevertheless, recall that if f is
continuous in I, then, even if f is unbounded in I, Weierstrass’ Theorem 8.26 assures
that it is bounded in each interval Œa; b I.)
Definition 10.68 Let I R be an interval with endpoints ˛ and ˇ, with ˛ < ˇ,
and f W I ! R be integrable in each interval Œa; b I. We say that f is integrable
10
After Hermann Minkowski, German mathematician of the XIX and XX centuries. Minkowski
made several seminal contributions to different branches of Mathematics. His geometric method
on Algebraic Number Theory proved to be quite fruitful for the development of that theory. Also,
Minkowski’s four dimensional space is the correct setting for the geometric description of special
relativity.
10.9 Improper Integration 427
do exist. This being the case, we define the improper integral of f in I, denoted
Rˇ
˛ f .t/ dt, by setting
Z ˇ Z x0 Z y
f .t/ dt D lim f .t/ dt C lim f .t/ dt;
˛ x!˛ x y!ˇ x0
Rˇ
and say that the improper integral ˛ f .t/ dt converges.
It is important to realize that the former definition does not depend on the chosen
x0 2 I. More precisely, suppose that x0 2 I is chosen so that the limits (10.67) do
exist, and let x00 be another point in I. Then,
Z x00 Z x0 Z x00
f .t/ dt D f .t/ dt C f .t/ dt;
x x x0
R x00
so that the limit limx!˛ x f .t/ dt also exists, with
Z x00 Z x0 Z x00
lim f .t/ dt D lim f .t/ dt C f .t/ dt:
x!˛ x x!˛ x x0
Ry
Analogously, limy!ˇ x00
f .t/ dt does exist, with
Z y Z x0 Z y
lim f .t/ dt D f .t/ dt C lim f .t/ dt:
y!ˇ x00 x00 y!ˇ x0
R x0 R x0
Therefore, since 0
x0 f .t/ dt C x00 f .t/ dt D 0, we get
Z x00 Z y Z x0 Z y
lim f .t/ dt C lim f .t/ dt D lim f .t/ dt C lim f .t/ dt:
x!˛ x y!ˇ x00 x!˛ x y!ˇ x0
On the other hand, in case I D Œa; b, we claim that the definition of integral given
in Chap. 10 coincides with the present one. In order to check that, let f W Œa; b ! R
be (bounded and) integrable in Œa; b, and fix x0 2 .a; b/. Then f is integrable in
Œa; x0 and, for x 2 .a; x0 /, we have
Z x0 Z x0 Z x
f .t/ dt D f .t/ dt f .t/ dt:
x a b
428 10 Riemann’s Integral
Letting M be an upper bound for jf j in Œa; b, it follows from the triangle inequality
for integrals and Proposition 10.12 that
ˇZ x ˇ Z x
ˇ ˇ
ˇ f .t/ dt ˇ jf .t/jdt M.x a/;
ˇ ˇ
a a
Rx
so that the squeezing principle gives limx!a a f .t/ dt D 0. Therefore,
Z x0 Z x0 Z x Z x0
lim f .t/ dt D f .t/ dt lim f .t/ dt D f .t/ dt:
x!a x a x!a a a
Since f is also integrable in Œx0 ; b, a similar reasoning allows us to conclude that
Z y Z b Z b Z b
lim f .t/ dt D f .t/ dt lim f .t/ dt D f .t/ dt:
y!b x x0 y!b y x0
0
Thanks to the above discussion, we can restrict the study of improper integrals
to the cases in which I D .A; B, ŒA; B/, .1; B or ŒA; C1/, with A; B 2 R. Along
the rest of this section, we stick to the cases I D ŒA; B/ and ŒA; C1/, leaving to the
reader the task of formulating the analogous results when I D .A; B or .1; B.
Let’s start by looking at two examples.
R C1
Example 10.69 The improper integral 0 et dt does converge. More precisely,
R C1 t
we have 0 e dt D 1.
Proof For x > 0, it follows from the FTC that
Z x ˇtDx
ˇ
et dt D et ˇ D 1 ex :
0 tD0
R C1
Now, Theorem 10.54 gives limx!C1 ex D 0, so that 0 et dt D 1. t
u
R C1 ˛
Example 10.70 The improper integral 1 t dt converges if and only if ˛ < 1.
Proof For x > 1 and ˛ ¤ 1, note that
Z x
1 ˛C1 ˇˇtDx 1
t˛ dt D t ˇ D .x˛C1 1/: (10.68)
1 ˛C1 tD1 ˛C1
Now, since x˛C1 D e.˛C1/ log x , we have (again by Theorem 10.54) that
0; if ˛ C 1 < 0
lim x˛C1 D :
x!C1 C1; if ˛ C 1 > 0
10.9 Improper Integration 429
Hence, the right hand side of (10.68) has a (finite) limit when x ! C1 if and only
if ˛ C 1 < 0, i.e., if and only if ˛ < 1.
Finally, if ˛ D 1, then
Z x ˇtDx
ˇ
t1 dt D log tˇ D log x ! C1
1 tD1
when x ! C1. t
u
Also with respect to the previous example, since
x˛
lim D lim x˛C1 D 0
x!C1 x1 x!C1
R C1
if and only if ˛ < 1, we use to say that 1 t˛ dt converges if and only if x˛
decays to zero more rapidly than x1 when x ! C1. See also Problem 3, which
R1
deals with the case 0 t˛ dt.
The coming theorem is the fundamental result on the convergence of improper
integrals. Note that it is the analogue, for improper integrals, of Theorem 7.27. For
its statement, as well as for remark just following its proof, the reader may find it
convenient to read again Problem 8, page 218.
Theorem 10.71 Let f W ŒA; C1/ ! R be integrable in each interval Œa; b
ŒA; C1/. The following assertions are equivalent:
R C1
(a) A f .t/ dt converges. ˇR ˇ
ˇ x ˇ
(b) For every > 0, there exists M > A such that x1 ; x2 > M ) ˇ x12 f .t/ dtˇ < .
(c) There exists L 2 R such that, for every sequence R x .xn /n1 in ŒA; C1/ and
satisfying limn!C1 xn D C1, one has limn!C1 A n f .t/ dt D L.
Rx
Proof Let F W ŒA; C1/ ! R be defined by F.x/ D A f .t/ dt.
R C1 R C1
(a) ) (b): assume that A f .t/ dt converges, with A f .t/ dt D L. By
definition, we have limx!C1 F.x/ D L, so that, given > 0, there exists M > 0 for
which
x > M ) jF.x/ Lj < :
2
Hence, for x1 ; x2 > M, triangle inequality gives
ˇZ ˇ
ˇ x2 ˇ
ˇ f .t/ dtˇˇ D jF.x2 / F.x1 /j
ˇ
x1
jF.x2 / Lj C jL F.x1 /j
< C D :
2 2
430 10 Riemann’s Integral
(b) ) (c): assume that condition (b) holds, and let .xn /n1 be a sequence such
that xn A for every n 1 and limn!C1 xn D C1. Given > 0, take M as in the
statement and n0 2 N such that n > n0 ) xn > M. Then, for m; n > n0 , we have
xm ; xn > M, so that
ˇZ xm ˇ
ˇ ˇ
ˇ
jF.xm / F.xn /j D ˇ f .t/ dtˇˇ < :
xn
in the previous paragraph (with .x0n /n1 in place of .xn /n1 ) assures the existence of
L0 D limn!C1 F.x0n /. However, if it was L0 ¤ L, then the sequence .x00n /n1 , given
for k 1 by x002k1 D x2k1 and x002k D x02k , would also converge to C1 and be
such that
Since L ¤ L0 , the limit limn!C1 F.x00n / would not exist, which contradicts the
argument of the previous paragraph (this time with .x00n /n1 in place of .xn /n1 ).
(c) ) (a): suppose that there exists L 2 R such that, for every sequence
.xn /n1 in ŒA; C1/ with limn!C1 xn D C1, we have limn!C1 F.xn / D L. If
limx!C1 F.x/ does not exist or is different from L, then there exist > 0 and, for
every B > A, a real number xB > B such that jF.xB / Lj . Taking B successively
equal to A C 1, A C 2, A C 3, . . . , we would get a sequence x1 , x2 , x3 , . . . satisfying
xn > A C n and jF.xn / Lj for every n 1. Therefore, limn!C1 xn D C1
but .F.xn //n1 does not converge to L, which is an absurd. t
u
Remark 10.72 It is immediate to adapt the previous result to the case of a function
f W ŒA; B/ ! R which is integrable in each interval ŒA; b with A < b < B.
More precisely, and as the reader can easily verify, the following assertions are
equivalent:
RB
(a) A f .t/ dt converges.
ˇR every ˇ > 0, there exists ı > 0 such that B ı < x1 ; x2 < B )
(b) For
ˇ x2 ˇ
ˇ x1 f .t/ dtˇ < .
(c) There exists L 2 R such that, for everyR sequence .xn /n1 in ŒA; B/ and satisfying
x
limn!C1 xn D B, one has limn!C1 A n f .t/ dt D L.
We now establish an important consequence of the previous theorem, which is
known in mathematical literature as the comparison test for improper integrals.
Proposition 10.73 Let f ; g W ŒA; C1/ ! R be such that jf .x/j g.x/ for every
x 2 ŒA; C1/. If f and g are integrable in every interval Œa; b ŒA; C1/ and
R C1 R C1
A g.t/ dt converges, then A f .t/ dt also converges and
ˇZ C1 ˇ Z C1
ˇ ˇ
ˇ f .t/ dtˇˇ g.t/ dt:
ˇ
A A
10.9 Improper Integration 431
R C1
Proof Firstly, notice that g.x/ 0 for every x A, so that A g.t/ dt 0. Then,
R C1
given > 0, the convergence of A g.t/ dt, together with Theorem 10.71, assures
the existence of M > A such that
Z x2
x 1 ; x2 > M ) g.t/ dt < :
x1
Now, the triangle inequality for integrals, together with Proposition 10.12,
guarantees that, also for x1 ; x2 > M, we have
ˇZ ˇ Z Z
ˇ x2 ˇ x2 x2
ˇ f .t/ dtˇˇ jf .t/jdt g.t/ dt < :
ˇ
x1 x1 x1
Then, once more from Theorem 10.71, we get the convergence of the integrals
R C1 R C1
A f .t/ dt and A jf .t/j dt.
Finally, it follows from what we did above, together with the properties of limits
of functions (cf. Sect. 9.2) that
ˇZ ˇ ˇ Z x ˇ ˇZ x ˇ
ˇ C1 ˇ ˇ ˇ ˇ ˇ
ˇ ˇ ˇ
f .t/ dtˇ D ˇ lim ˇ ˇ
f .t/ dtˇ D lim ˇ f .t/ dtˇˇ
ˇ x!C1 A x!C1 A
A
Z x Z x
lim jf .t/jdt lim g.t/ dt
x!C1 A x!C1 A
Z C1
D g.t/ dt:
A
t
u
Remark 10.74 As in the previous remark, the former proposition can be easily
adapted to deal with the case I D ŒA; B/. More precisely, if f ; g W ŒA; B/ ! R
are integrable in each interval ŒA; b with A < b < B, and such that jf .x/j g.x/
RB RB
for every x 2 ŒA; B/, then the convergence of A g.t/ dt implies that of A f .t/ dt;
moreover, one also has
ˇZ ˇ Z
ˇ B ˇ B
ˇ f .t/ dtˇˇ g.t/ dt:
ˇ
A A
To this end, start by observing that the fundamental trigonometric limit (cf.
Lemma 9.11) assures that the function t 7! sint t extends continuously to 0.
R C1 sen t
Therefore, it suffices to establish the convergence of 1 t dt.
To what is left, let x > 1. Integrating by parts, we get
Z x Z x
sin t cos t ˇˇtDx cos t
dt D ˇ dt
1 t t tD1 1 t2
Z x
cos x cos t
D cos 1 dt:
x 1 t2
R C1 sin t
Since j cos xj 1 for every x, the computations above guarantee that 1 dt
R C1 cos t t
converges if and only if 1 t2
dt converges. Hence, we have reduced our work
to the convergence of this last improper integral. ˇ xˇ
Finally, it is enough to apply the comparison test, noticing that ˇ cos x2
ˇ 12 and
R C1 1 x
(according to Example 10.70) 1 t 2 dt does converge.
Problem 10, page 436, brings another proof of the convergence of Dirichlet’s
integral. On the other hand, it is possible to show (cf. [10, 20] or [27], for
instance) that
Z C1
sin t
dt D :
0 t 2
In turn, this computation is of fundamental importance to the theory of Fourier
series.11 Figure 10.10 sketches the graph of x 7! sinx x .
Back to the development of the theory, Proposition 10.73 assures that if f W
R C1
ŒA; C1/ ! R is integrable in each interval Œa; b ŒA; C1/ and A jf .t/j dt
R C1
converges, then A f .t/ dt also converges and
ˇZ C1 ˇ Z C1
ˇ ˇ
ˇ f .t/ dtˇˇ jf .t/j dt (10.69)
ˇ
A A
(recall that jf j is also integrable in each interval Œa; b ŒA; C1/, by Proposi-
R C1
tion 10.14). In such a case, we say that A f .t/ dt is absolutely convergent. Not
every convergent improper integral is absolutely convergent; actually, Problem 11,
R C1 sin t
page 436, shows that 0 t
dt is not absolutely convergent.
We finish this section by showing that the link between the theory of convergent
series and convergent improper integrals is actually deeper. More precisely, the
coming result establishes an equivalence between the convergence of the integral
of a positive function and that of a certain series. For this reason, it is known as the
integral test for absolute convergence of series.
11
After Jean-Baptiste-Joseph Fourier, French engineer and mathematician of the XIX century. For
more on Fourier see the footnote on page 469. A glimpse on the theory of Fourier Series is the
content of Problems 11 to 17, page 468.
10.9 Improper Integration 433
1.0
0.5
−0.5
Proof Let g; h W Œn0 ; C1/ ! R be the functions given, in the interval Œk; k C 1/ (for
each k n0 ), by g.x/ D f .k C 1/ and h.x/ D f .k/.
The conditions on f guarantee that 0 g.x/ f .x/ h.x/ for every x n0 .
Hence, the comparison test for improper integrals assures that
Z C1 Z C1
f .t/ dt converges ) g.t/ dt converges
n0 n0
and
Z C1 Z C1
h.t/ dt converges ) f .t/ dt converges:
n0 n0
Z n X
n Z n X
n1
g.t/ dt D f .k/ and h.t/ dt D f .k/:
n0 kDn0 C1 n0 kDn0
434 10 Riemann’s Integral
Finally, let’s apply equivalence (a) , (c) of Theorem 10.71, with xk D k for
every k n0 :
R C1 R C1
(i) If n0 f .t/ dt converges, then n0 g.t/ dt converges and, hence
Z C1 Z n X
n X
g.t/ dt D lim g.t/ dt D lim f .k/ D f .k/:
n0 n!C1 n0 n!C1
kDn0 C1 kn0 C1
P P P
Therefore,
P kn0 f .k/ converges, with kn0 f .k/ D f .n0 / C kn0 C1 f .k/.
(ii) If kn0 f .k/ converges, then
Z C1 Z n X
n1 X
h.t/ dt D lim h.t/ dt D lim f .k/ D f .k/;
n0 n!C1 n n!C1
0 kDn0 kn0
R C1
so that h.t/ dt converges. Hence, what we did before assures the conver-
R C1 n0
gence of n0 f .t/ dt.
t
u
Example 10.77 The integral test allows us to easily establish
P the divergence of the
harmonic series, as well as the convergence of the series k1 k1r , for every real
number r > 1. For the harmonic series, letting f W Œ1; C1/ ! R be given by
f .x/ D 1x , we have
Z C1 Z
1 x
1
dt D lim dt D lim log x D C1:
1 t x!C1 1 t x!C1
1
Analogously, taking f W Œ1; C1/ ! R given by f .x/ D xr
, with r > 1, we have
Z C1 Z
1 x
1 1 1 1
dt D lim dt D lim 1 D :
1 tr x!C1 1 t r x!C1 1r xr1 r1
To what is left, note that the substitution of variable s D log t (cf. Problem 2,
page 435) gives
Z C1 Z C1 Z C1
1 1
f .t/ dt D dt D ds
2 2 t log t log 2 s
D lim .log s log 2/ D C1:
s!C1
R C1 P 1
Hence, 2 f .t/ dt diverges, so that k2 k log k also diverges. t
u
R R1 1
2 (c) dt.
(a) sec t dt.
R C1
2 R0C1
log t
1
(b) 0 et sin t dt. (d) 1 1Ct2 dt.
R1
3. * Show that 0 t˛ dt converges if and only if ˛ > 1. Then, for ˛ < 0, conclude
R1 ˛
that 0 t dt converges if and only if x˛ grows to C1 more slowly than x1 as
x ! 0C. R1 R1
4. (Berkeley) Prove that both integrals 0 cos.x2 /dx and 0 sin.x2 /dx converge.
5. (Berkeley—adapted) Show that:
R R
(a) R0 log.sin x/dx andR 0 log j cos xjdx converge.
(b) 0 log.sin x/dx D 0 log j cos xjdx D log 2.
6. Examine the convergence of the following series, where ˛ > 0 is a given real
number:
P 1
(a) ˛.
Pk2 .log1k/
(b) ˛.
Pk2 k.log k/ 1
(c) k2 k.log k/.log log k/˛ .
436 10 Riemann’s Integral
R C1 log t
7. Show that 0 1Ct2
dt converges and compute its value.
The next problem extends the results of Problem 12, page 381, and of
Theorem 10.63.
8. * Let I R be an interval with endpoints ˛ and ˇ, f ; g W I ! .0; C1/ be
integrable functions and p; q > 0 such that 1p C 1q D 1.
Rˇ Rˇ Rˇ
(a) If ˛ f .t/p dt D ˛ g.t/q dt D 1, show that ˛ f .t/g.t/ dt 1.
(b) Deduce Hölder’s inequality for integrals:
Z Z !1=p Z !1=q
ˇ ˇ ˇ
f .t/g.t/ dt f .t/ dt
p q
g.t/ dt :
˛ ˛ ˛
9. In the proof of
P Theorem 10.55, we saw that in order to guarantee the existence
n 1
of limn!C1 kD1 k log n it suffices to assure the convergence of the series
X 1
log.k C 1/ C log k :
k1
k
1
To this end, let f W Œ1; C1/ ! R be given by f .x/ D x log.x C 1/ C log x for
x 1, and do the following items:
(a) Show that f is decreasing, with limx!C1 f .x/ D 0.
(b) Apply the integral test to establish the convergence of the series.
10. The purpose of this problem is to give another proof of the convergence of the
R C1 sin t
improper integral 0 t
dt. To this end, do the following items:
P
(a) Prove that it suffices to show the convergence of the series k1 sink k .
(b) In turn, use Abel’s criterion (cf. Problem 18, page 243) to show that this
P to establishing the boundedness of the sequence .sn /n1 , such that
amounts
sn D nkD1 sin k for n 2 N.
sin n
sin nC1
(c) Apply the result of Problem 10, page 359, to find sn D 2
sin 12
2
and,
1
hence, jsn j sin 12
.
We recall that our initial motivation for the study of the integral was geometric,
namely, to define and compute the area of the region of the cartesian plane under the
graph of a given nonnegative (integrable) function. We could also use the concept
of integral to deal with the following geometric problems:
(I) How to define and compute the length of the graph of f W Œa; b ! R?
(II) In case f is positive, how to define and compute the volume of the solid of
revolution generated by the rotation of the region under the graph of f around
the horizontal axis?
(III) Yet in the case of a positive f , how could one define and compute the area of
the surface of revolution generated by the rotation of the graph of f around the
horizontal axis?
In this section we analyse Problem (I), leaving the analysis of Problems (II) and
(III) to [4]. As a byproduct of our discussion, we will present a thorough definition
and computation of the length of a circle.
Then, let f W Œa; b ! R be continuous. Fix (cf. Fig. 10.11) an interval Œc; d
.a; b/, a partition P D fc D x0 < x1 < < xk D dg of Œc; d and let xj D xj xj1
for 1 j k.
For jPj sufficiently small, it is a reasonable guess to think of the line segment
joining points .xj1 ; f .xj1 // and .xj ; f .xj // as a good approximation for the portion
of the graph of f situated between the vertical lines x D xj1 and x D xj . If this is
so, it is also reasonable to assume that the length `j of such a segment should serve
as a good approximation for the length of this portion of the graph of f , whichever
way we define it.
12
For a sligthly different proof, see Chap. 9 of [5].
438 10 Riemann’s Integral
(xj−1 , f (xj−1 ))
(xj , f (xj ))
x
c = x0 xj−1 xj d = xk
This being said, the formula for the distance between two points of the cartesian
plane gives
q
`j D .xj xj1 /2 C .f .xj / f .xj1 //2
s
f .xj / f .xj1 / 2
D 1C xj :
xj xj1
Now, if f is continuously differentiable in .a; b/, then Lagrange’s MVT guarantees
f .x /f .x /
the existence of j 2 .xj1 ; xj / such that xj j xj1j1 D f 0 .j /; this way,
q q
`j Š 1 C f 0 .j /2 xj D 1 C f 0 .j /2 .xj xj1 /:
k q
X
1 C f 0 .j /2 .xj xj1 / (10.70)
jD1
as jPj ! 0.
10.10 Two Important Applications 439
Rdp
The discussion above allows us to conclude that c 1 C f 0 .x/2 dx is the only
reasonable way of defining the length of the graph of the restriction of f to the
interval Œc; d. If we now notice that the union of the graphs of such restrictions (as
Œc; d varies over all bounded and closed subintervals of .a; b/) coincides with the
graph of f but the points .a; f .a// and .b; f .b//, we are naturally led to state the
following
Definition 10.79 Let f W Œa; b ! R be continuous in Œa; b and continuously
differentiable in .a; b/. If the improper integral
Z b p
`D 1 C f 0 .x/2 dx; (10.71)
a
does exist, then we say that the graph of f is rectifiable and has length `.
Yet with respect to the former definition, recall that
Z b p Z b p
1 C f 0 .x/2 dx D lim 1 C f 0 .x/2 dx:
a !0 aC
Therefore, if f is continuously differentiable in Œa; b, then Problem 11, page 380,
assures that (10.71) holds in the ordinary sense (i.e., not as an improper integral).
Example 10.80 Let be a circle of center O and radius R, and choose a cartesian
system
p of coordinates such that O.0; 0/. If f W ŒR; R ! R is given by f .x/ D
R2 x2 , then the graph of f is the semicircle of situated on the upper half-
plane of the cartesian plane. Note that f is continuous in ŒR; R and continuously
differentiable in .R; R/. Hence, according to (10.71), the length of equals twice
the value of the improper integral
Z R p
1 C f 0 .x/2 dx;
R
provided it exists.
To see that this is indeed the case, first note that
s
p 2
x R
1 C f 0 .x/2 D 1C p D p :
R 2 x2 R x2
2
Now, performing the trigonometric substitution x D R cos t and observing that RC
x R , arccos.1 C R / t arccos.1 R /, we get
440 10 Riemann’s Integral
Z R p Z R
R
1 C f 0 .x/2 dx D p dx
RC RC R x2
2
Z arccos.1 R /
R
D p .R sin t/dt
arccos.1C R / R2 R2 cos2 t
Z arccos.1 R /
D R dt
arccos.1C R /
D R arccos 1 C arccos 1 :
R R
Finally, since the arccos function arccos W Œ1; 1 ! Œ0; is continuous, we obtain
Z
R p
lim 1 C f 0 .x/2 dx D lim R arccos 1 C arccos 1
!0 RC !0 R R
D R .arccos.1/ arccos.1// D R:
Finally, since essentially the same argument holds for f , we conclude (as
expected) that the length of equals 2R.
In spite of the former example, the integral in (10.71) is generally difficult
(or even impossible—cf. Problem 2) to compute exactly. Let’s see an illustrative
example.
Example 10.81 Given b > 0, let’s compute the length ` of the portion of the
parabola y D x2 situated between the points .0; 0/ and .b; b2 /.
With f .x/ D x2 in (10.71) we get
Z b p
`D 1 C 4x2 dx:
0
p
p If we set ˛ D arctan.2b/, then tan ˛ D 2b and, hence, sec ˛ D 1 C tan2 ˛ D
1 C 4b2 . Therefore,
1 ˇ˛
ˇ
` D .sec t tan t log j sec t tan tj/ C C ˇ
2 0
p 1 p
D b 1 C 4b2 log. 1 C 4b2 2b/:
2
We now turn to a completely different topic, using Proposition 10.73 to introduce
and discuss some of the amazing properties of the Gamma function.
The Gamma function is the function W .0; C1/ ! .0; C1/ defined by
Z C1
.x/ D et tx1 dt:
0
Since
Z C1 Z 1 Z C1
et tx1 dt D et tx1 dt C et tx1 dt;
0 0 1
in order to see that .x/ is well defined it suffices to establish the convergence of
both improper integrals at the right hand side above.
To what is left to do, initially observe that, since x > 0, Problem 3, page 435,
R1
guarantees the convergence of 0 tx1 dt. Since 0 < et tx1 < tx1 for 0 < t 1,
Remark 10.74 (modified in the obvious way to be applied to the interval .0; 1)
R1
assures the convergence of 0 et tx1 dt.
On the other hand, for a fixed x > 0, we claim that et tx1 Cet=2 for t 1,
where C is a positive constant that only depends on x. Indeed, letting n > x C 1 be
a natural number, it follows from Theorem 10.54 that
tn
lim D 0:
t!C1 et=2
tn
Therefore, there exists C > 0 such that et=2
C for every t 1. However, if this is
so, then
Physics and Analytic Number Theory. Roughly, one can say that this is due to the
fact that it extends the notion of factorial of natural numbers to the positive reals.
In order to check the above claim, let’s start by noting that Example 10.69 gives
Z C1
.1/ D et dt D 1:
0
On the other hand, for x > 0 the integration by parts formula furnishes
Z 1 Z s
.x/ D lim et tx1 dt C lim et tx1 dt
s!0 s s!C1 1
Z
1 t x ˇˇtD1 1 1 t x
D lim e tˇ C e t dt
s!0 x tDs x s
ˇ Z
1 t x ˇtDs 1 s t x
C lim e tˇ C e t dt
s!C1 x tD1 x 1
Z Z
1 1 t x 1 C1 t x 1
D e t dt C e t dt C lim es sx :
x 0 x 1 s!C1 x
Then,
.x C 1/ D x .x/
for every x > 0, and an easy induction starting from .1/ D 1 gives .n/ D .n1/Š
for every n 2 N.
We now show that is continuous and log W .0; C1/ ! R is convex in
.0; C1/. Actually, since
D elog
for all x; y > 0 and 0 < s < 1. Letting s D 1q and 1 s D 1p , we have p; q > 0 such
that 1p C 1q D 1. Then, the fact that log is an increasing function that takes products
10.10 Two Important Applications 443
into sums shows that the validity of the above inequality is equivalent to that of
x y
C .x/1=p .y/1=q ; (10.73)
p q
x1 y1
Therefore, letting f .t/ D et=p t p , g.t/ D et=q t q and applying Hölder’s
inequality (cf. Problem 8, page 436), we get
Z C1 Z C1 1=p Z C1 1=q
x y
C D f .t/g.t/dt f .t/p dt g.t/q dt
p q 0 0 0
Z C1 1=p Z C1 1=q
D et tx1 dt et ty1 dt
0 0
D .x/1=p .y/1=q :
Remark 10.82 One can actually show that is infinitely differentiable, and that
the expression for its kth derivative can be computed by differentiating under the
integral that defines . In particular, we must have
Z C1
0
.x/ D et tx1 log t dt:
0
The convergence of the improper integral at the right hand side above is the object of
Problem 5, and the proof of the continuous differentiability of the Gamma function,
together with the above formula for 0 .x/, will be the object of Problems 18 and 19,
page 471.
1. For x0 ¤ 0, compute the length of the portion of the catenary situated between
the points of abscissas 0 and x0 .
2. Let E be an ellipse of major axis AA0 and minor axis BB0 , with AA0 D 2a and
BB0 D 2b. If we choose a cartesian system in which A0 .a; 0/ and B0 .0; b/, it is
2 2
well known that E has equation ax2 C by2 D 1. In this respect, do the following
items:
444 10 Riemann’s Integral
p
(a) If f W Œa; a ! R is given by f .x/ D ba a2 x2 , show that the graph of f
is the portion of E situated in the upper halfplane. Then, conclude that the
length of E is given by twice the value of the improper integral
Z s
a
b 2 x2
1C dx:
a a2 .a2 x2 /
p
(b) Let c D a2 b2 and x D a cos t, with 0 t . Use the formula of
integration by substitution to show that the length of E is given by
Z
r c 2
2b 1C sin2 t dt: (10.74)
0 b
The definite integral of item (b) cannot be computed exactly. More generally, it
is possible to show (albeit this is well beyond the scope of these notes) that the
indefinite integral
Z p
1 C sin2 t dt; (10.75)
P
Γ
A
t
O t T π 2π x
13
A proof of this fact can be found in [12].
10.10 Two Important Applications 445
and such that is in the upper half-plane. Show that, after rolls in the
positive direction (with respect to the horizontal axis) by t radians starting
from O, this point will be at position P.x.t/; y.t//, with x.t/ D t sin t and
y.t/ D 1 cos t.
(b) Note that function t 7! x.t/, from Œ0; 2 into itself, is a differentiable
bijection with inverse x 7! t.x/ differentiable in .0; 2/. Therefore, if f W
Œ0; 2 ! R is the composite f .x/ D y.t.x//, the graph of f is the portion
of the cycloid situated in the vertical strip of the cartesian plane defined by
the inequalities 0 x 2. Such a portion is called a step of the cycloid.
Compute the length of it.
R C1 2 p
4. It is possible to show that 0 es ds D 2 (see, for instance, Chap. 8 of
p
[20]). Use this fact to compute . 12 / D .
R C1
5. Show that, for x > 0, the integral 0 et tx1 log t dt is absolutely convergent.
6. Let k > 1 and n1 ; n2 : : : ; nk be naturals such that k j .n1 C n2 C C nk /. Prove
that
p
n1 C n2 C C nk
Š k n1 Šn2 Š : : : nk Š:
k
2
Chapter 11
Series of Functions
This last chapter presents the adequate notion of convergence for sequences and
series of functions, with an emphasis on the study of power series. We start by
examining a power series naturally attached to an infinitely differentiable function,
called its Taylor series. Then, in Sects. 11.2 and 11.3 we discuss the basic results on
uniformly convergent series of functions and on power series. Finally, in Sect. 11.4,
we present some applications of the theory to sequences defined by linear recurrence
relations.
X
n1 .k/
f .x0 / f .n/ .c/
f .x/ D .x x0 /k C .x x0 /n : (11.1)
kD0
kŠ nŠ
1
After Brook Taylor, English mathematician of the XVIII century.
X
n1 .k/
f .t/ ˛
g.t/ D f .x/ .x t/k .x t/n ;
kD0
kŠ nŠ
with ˛ 2 R chosen in such a way that g.x0 / D 0 (evidently, such a choice is always
possible).
Since f is n times differentiable in I and Œx0 ; x I, several applications of the
results of Problem 3, page 300, guarantee that g is differentiable in Œx0 ; x. Since
g.x0 / D g.x/ D 0, Rôlle’s theorem assures the existence of c 2 .x0 ; x/ such that
g0 .c/ D 0. On the other hand, a simple computation furnishes
˛ f .n/ .t/
g0 .t/ D .x t/n1 ;
.n 1/Š
so that for c 2 .x0 ; x/ we have g0 .c/ D 0 if and only if ˛ D f .n/ .c/. Hence,
X
n1 .k/
f .x0 / f .n/ .c/
0 D g.x0 / D f .x/ .x x0 /k .x x0 /n ;
kD0
kŠ nŠ
as we wished to show.
Now, let x < x0 and J D x0 C x I D fx0 C x tI t 2 Ig. Then J is an interval and
I D x0 CxJ, so that we can define h W J ! R by h.s/ D f .x0 Cxs/. Corollary 9.42
assures that h is n times differentiable in J, with h.k/.s/ D .1/k f .k/ .x0 C x s/ for
0 k n. Hence, by applying (11.1) to h and to s0 < s in J, we obtain
X
n1 .k/
h .s0 / h.n/ .c/
h.s/ D .s s0 /k C .s s0 /n
kD0
kŠ nŠ
X
n1
.1/k f .k/ .x0 C x s0 / .1/n f .n/ .c/
f .x0 C x s/ D .s s0 /k C .s s0 /n
kD0
kŠ nŠ
for some c 2 .s0 ; s/. Finally, letting s0 D x and s D x0 (so that s0 < s), we get (11.1)
for x < x0 . t
u
If f W I ! R is a twice differentiable function, then, for distinct x0 ; x 2 I, Taylor’s
formula with Lagrange remainder guarantees that
00
f .c/
f .x/ D f .x0 / C f 0 .x0 /.x x0 / C .x x0 /2 (11.2)
2
for some c between x0 and x.
11.1 Taylor Series 449
a = x0 xk−1 xk b = xn
Letting denote the expression at the right hand side of the stated inequality, the
additivity of the integral, together with the fact that xk xk1 D ba
n for 1 k n,
gives
ˇ n Z ˇˇ
ˇX xk
1
ˇ ˇ
Dˇ f .x/ dx .f .xk / C f .xk1 //.xk xk1 / ˇ
ˇ xk1 2 ˇ
kD1
ˇ n ˇ (11.3)
ˇX ˇ X n
ˇ ˇ
Dˇ gk .xk /ˇ jgk .xk /j;
ˇ ˇ
kD1 kD1
1 1
g0k .x/ D f .x/ f 0 .x/.x xk1 / .f .x/ C f .xk1 //
2 2
450 11 Series of Functions
and
1
g00k .x/ D f 00 .x/.x xk1 /;
2
so that g0k .xk1 / D g00k .xk1 / D 0. Hence, Taylor’s formula with Lagrange remainder
guarantees the existence of ck 2 .xk1 ; xk / such that
g00k .ck /
gk .xk / D gk .xk1 / C g0k .xk1 /.xk xk1 / C .xk xk1 /2
2
1
D f 00 .ck /.ck xk1 /.xk xk1 /2 :
4
1 00
jgk .xk /j D jf .ck /jjck xk1 j.xk xk1 /2
4
1 00 .b a/3
jf .ck /jjxk xk1 j3 D jf 00 .ck /j :
4 4n3
Finally, by substituting this last expression in estimate (11.3) for , we get
X
n X
n
.b a/3 .b a/3 X 00
n
jgk .xk /j jf 00 .ck /j D jf .ck /j
kD1 kD1
4n3 4n3 kD1
.b a/3 .b a/3
3
n sup jf 00 j D 2
sup jf 00 j:
4n Œa;b 4n Œa;b
t
u
In words, the result of the previous example gives a numerical approximation,
with controlled error, for the computation of an integral. As the discussion of the
computation of the length of an ellipse shows (cf. Problem 2, page 443), sometimes
this is the best one can get. In this respect, see also Problem 3 of this section.
For more on the use of Calculus methods to get useful numerical approximations,
see Chap. 15 of [2] or Part B of [7].
Back to the development of the theory, let f W I ! R be infinitely differentiable.
For x0 2 I, we say that
X f .k/ .x0 /
.x x0 /k
k0
kŠ
X f .k/ .x0 /
f .x/ D .x x0 /k
k0
kŠ
for every x 2 I.
Proof Fixed x 2 I and n 2 N, it follows from Taylor formula with Lagrange
remainder that
X
n1 .k/
f .x0 / f .n/ .c/
f .x/ D .x x0 /k C .x x0 /n (11.4)
kD0
kŠ nŠ
where we used the result of Problema 3, page 218, in the last passage above.
Therefore, it suffices to let n ! C1 in (11.4) to get the desired result. t
u
Example 11.4 Since j sin.2j/ xj D j sin xj 1 and j sin.2j1/ xj D j cos xj 1, the
previous proposition gives
X sin.k/ 0 X .1/j1
sin x D xk D x2j1
k0
kŠ j1
.2j 1/Š
X .1/j
cos x D x2j ;
j0
.2j/Š
for every x 2 R. In Sect. 11.5, these series expansions will be the departure points
for the rigorous construction of the sine and cosine functions.
We now apply Taylor formula with Lagrange remainder to show that the Taylor
series of the exponential function converges to such a function along the whole real
line.
Theorem 11.5 For x 2 R, one has
X 1
ex D xk : (11.5)
k0
kŠ
452 11 Series of Functions
X
n1 k
x xn
ex D C ec ; (11.6)
kD0
kŠ nŠ
for some c between 0 and x. Now, using the fact that the exponential function is
increasing, together with the result of Problem 3, page 218, we obtain
ˇ c ˇ
ˇ e nˇ
ˇ x ˇ D ec jxj maxfe0 ; ex g jxj !
n n
n
ˇ nŠ ˇ 0:
nŠ nŠ
n
Hence, the squeezing principle gives limn!C1 ec xnŠ D 0. Finally, letting n ! C1
in (11.6), we get the desired result. t
u
Note that the previous result largely generalizes Example 7.41. For future
reference, we observe that changing x by ax in (11.5), we obtain
X ak
eax D xk (11.7)
k0
kŠ
for every x 2 R.
.n/
Sometimes, estimates of the error f nŠ.c/ .x x0 /n in (11.1) as simple as those
we did above do not suffice to guarantee that the Taylor series of an infinitely
differentiable function converges to such a function at all points it could do. We
now look at such an example.
Example 11.6 Function x 7! log.1 C x/ is defined in the whole interval .1; 1/
and is infinitely differentiable there. Letting f .x/ D log.1 C x/ for jxj < 1, it’s
immediate to verify that f .k/ .x/ D .1/.1Cx/.k1/Š
k1
k for every k 2 N. Hence, f .k/ .0/ D
.1/k1 .k 1/Š, and it immediately follows that the Taylor series of log.1 C x/
centered at 0 is given by
X .1/k1
xk : (11.8)
k1
k
ˇ ˇ P
ˇ ˇ
Since ˇ .1/k xk ˇ jxjk and k1 jxjk converges when jxj < 1, the comparison
k1
test for series assures that (11.8) converges whenever x 2 .1; 1/. Nevertheless,
if we try to estimate the error in (11.1) as we did before, we will only be able to
conclude that
X .1/k1
log.1 C x/ D xk
k1
k
11.1 Taylor Series 453
for x 2 Œ 12 ; 1/. Indeed, it follows from (11.1) that, for 0 < jxj < 1, we have
X
n1
.1/k1 .1/n1
log.1 C x/ D xk C xn
kD0
k n.1 C c/n
If 1 < x < 0, then 1 < x < c < 0, so that 1 C c > 1 C x > 0. Therefore,
ˇ ˇ
ˇ .1/n1 ˇ 1 jxjn 1 jxjn 1 jxj n
ˇ x nˇ
D D :
ˇ n.1 C c/n ˇ n .1 C c/n n .1 C x/n n 1Cx
jxj
For 1 < x < 0, one easily concludes that 1Cx
< 1 , 12 x < 0; yet in this
interval, the last estimate above gives
ˇ ˇ
ˇ .1/n1 ˇ 1 n
ˇ nˇ
ˇ n.1 C c/n x ˇ n ! 0:
jxj
On the other hand, for 1 < x < 12 we have 1Cx
> 1, so that
n
1 jxj n
! C1I
n 1Cx
ˇ ˇ
ˇ ˇ
hence, in the interval 1; 12 we are not able to conclude that ˇ .1/
n1
n.1Cc/n
xn ˇ ! 0
when n ! C1.
We shall remedy situations like that of the previous example in Sect. 11.3. For the
time being, let us show that it may well happen that the Taylor series of an infinitely
differentiable function f W R ! R, centered at some x0 2 R, does not converge to f
in any open interval centered at x0 .
Example 11.7 Let f W R ! R be given by
e1=x ; if x > 0
f .x/ D :
0; if x 0
for every x > 0. Let’s check this by induction on n 0, the case n D 0 being
obvious. By induction hypothesis, assume that for some integer k 0 there exists
a polynomial pk such that f .k/ .x/ D pk .1=x/e1=x for every x > 0. Then, the chain
rule gives, for x > 0,
.kC1/ 1 1 0 1
f .x/ D 2 pk pk e1=x D pkC1 .1=x/e1=x ;
x x x
with pkC1 .x/ D x2 .pk .x/ p0k .x//, so that pkC1 is again a polynomial.
Now that we have (11.9), let’s make one more induction to prove that f is
infinitely differentiable at 0, with f .n/ .0/ D 0 for every n 2 ZC . By induction
hypothesis, assume that f is k times differentiable at 0, with f .k/ .0/ D 0. Since f is
infinitely differentiable in Rnf0g, in order to show that f is kC1 times differentiable
at 0, with f .kC1/ .0/ D 0, it suffices to show that f .kC1/ .0/ D 0. To this end, note that
for h > 0 we have
where qk .x/ D xpk .x/, a polynomial; then, again by Theorem 10.54, we have
f .n/ .x0 / > 0 (resp. f .n/ .x0 / < 0), show that x0 is a point of strict local minimum
(resp. maximum) for f , i.e.,
x 2 I n fx0 g sufficiently close to x0 ) f .x/ > f .x0 / .resp.; f .x/ < f .x0 //:
3. Let `.E/ be the length of the ellipse E of major axis 2a, minor axis 2b and focal
distance 2c. Show that, for every n 2 N, one has
ˇ 0 s 1ˇ
ˇ X ˇ
ˇ n1
j Aˇˇ 2. C 1/b 3
ˇ`.E/ 2b @1 C 1 C 2 sin2
ˇ ˇ n2
;
ˇ n jD1
n ˇ
where D bc .
4. Obtain the Taylor series of the functions sinh and cosh. In each case, show that it
converges to the corresponding function in the whole real line.
5. Let I be an interval and f W I ! R be n times differentiable and such that f .n/ is
constant. Prove that f either vanishes identically or is a polynomial of degree at
most n. More precisely, if x; x0 2 I, show that
X
n
f .j/ .x0 /
f .x/ D .x x0 /j :
jD0
jŠ
for every n 1.
Yet with respect to the previous problem, note that Problem 10, page 219,
guarantees that n .d c/2 ! 0 as n ! C1. Hence, the result of item (iii)
n
above estimates the error with which an approaches ˛. Note also that we can
get results similar to those of the previous problem by assuming that f 0 > 0 and
456 11 Series of Functions
f 00 < 0 in .a; b/; it suffices to start with a1 2 Œa; ˛/. The same is true if we assume
that f 0 < 0 and f 00 < 0 (resp. f 00 > 0) in .a; b/; it suffices to start with a1 2 .˛; b
(resp. a1 2 Œa; ˛/).
8. The polynomial f .x/ D x3 2x 5 is such that f .2/ D 1 and f . 52 / D 92 .
Therefore, the IVT guarantees the existence of a root ˛ 2 .2; 52 / for f . Apply the
results of the previous problem to estimate ˛ with five correct decimal places.
The material on this section is a prelude to the study of power series and extends, to
sequences and series of functions, some concepts and results of Sects. 7.2 and 7.4.
We start by defining the concept of limit of a sequence of functions.
In all that follows, unless stated otherwise, I denotes an interval.
Definition 11.8 For each n 2 N, let a function fn W I ! R be given. If
limn!C1 fn .x/ exists for each x 2 I, we define the pointwise limit of the sequence
of functions .fn /n1 as the function f W I ! R such that
for every x 2 I. In this case, we also say that .fn /n1 is a pointwise convergent
sequence of functions, or that .fn /n1 converges pointwise to f .
The pointwise limit of a sequence fn W I ! R of functions, if exists, is unique.
Indeed, if .fn /n1 converges pointwise to f W I ! R, it follows from (11.10) that
for each x 2 I the sequence of real numbers .fn .x//n1 converges to f .x/; hence,
the uniqueness of the limit of a convergent sequence of reals (cf. Proposition 7.14)
assures that there exists only one possible value for f .x/.
Example 11.9 We collect here two examples illustrating the fact that the pointwise
limit of a pointwise convergent sequence of functions is not necessarily a well
behaved function.
(a) For each n 2 N, let fn W Œ0; 1 ! R be such that fn .x/ D xn for every x 2 Œ0; 1. It
follows from Example 7.12 that the pointwise limit of .fn /n1 exists and is the
function f W Œ0; 1 ! R such that
0; if x 2 Œ0; 1/
f .x/ D :
1; if x D 1
(b) For each n 2 N, let fn W Œ0; 1 ! R be given by fn .x/ D nx.1 x2 /n for every
x 2 Œ0; 1. Since 0 < x < 1 ) 0 < 1 x2 < 1, Problem 4, page 218, assures
that limn!C1 nx.1 x2 /n D 0 if 0 < x < 1. Also, since fn .0/ D fn .1/ D 0
for every n 2 N, we conclude that the pointwise limit of the sequence .fn /n1
exists and equals the function f W Œ0; 1 ! R that vanishes identically in Œ0; 1;
R1
in particular, f is integrable, with 0 f .x/dx D 0. Now, note that the FTC gives
Z 1
n.1 x2 /nC1 ˇˇxD1 n
fn .x/dx D ˇ D :
0 2.n C 1/ xD0 2.n C 1/
R1
Hence, limn!C1 0 fn .x/dx D limn!C1 n
2.nC1/ D 12 , so that
Z 1 Z 1
lim fn .x/dx ¤ lim fn .x/dx:
n!C1 0 0 n!C1
x 2 R. In view of this inequality and given > 0, choose n0 2 N such that 21n <
for n n0 . Then, for n n0 we have jfn .x/ 0j < for every x 2 R, as wished. u t
The coming corollary is an immediate consequence of Definition 11.10.
Corollary 11.12 If a sequence .fn /n1 of functions fn W I ! R converges uniformly
to f W I ! R, then .fn /n1 converges pointwise to f .
The concept of uniform convergence has the following geometric interpretation:
since
f−
fn
a b x
jf .x/ f .x0 /j jf .x/ fn .x/j C jfn .x/ fn .x0 /j C jfn .x0 / f .x0 /j:
Now, for a given > 0, the uniform convergence of .fn /n1 to f guarantees the
existence of n0 2 N such that n n0 ) jfn .x/ f .x/j < 4 for every x 2 I. Hence,
by writing the previous inequality for n D n0 we get
jf .x/ f .x0 /j 2 C jfn0 .x/ fn0 .x0 /j D C jfn0 .x/ fn0 .x0 /j:
4 2
On the other hand, the continuity of fn0 assures the existence of ı > 0 such that
x 2 I; jx x0 j < ı ) jfn0 .x/ fn0 .x0 /j < :
2
Therefore, for x 2 I such that jx x0 j < ı, we have
jf .x/ f .x0 /j C jfn0 .x/ fn0 .x0 /j < C D ;
2 2 2
so that f is continuous at x0 .
Finally, since x0 was chosen arbitrarily in I, we conclude that f is continuous in I.
t
u
11.2 Series of Functions 459
Proof For x 2 Œa; b, it follows from the triangle inequality for integrals (cf.
Proposition 10.14) that
ˇZ x ˇ Z x
ˇ ˇ
jgn .x/ g.x/j D ˇ .fn .t/ f .t//dtˇˇ
ˇ jfn .t/ f .t/jdt:
a a
Now, given > 0, the uniform convergence of .fn /n1 to f gives n0 2 N such that
n n0 ) jfn .t/ f .t/j
ba
for every t 2 Œa; b. Hence, for n n0 and x 2 Œa; b, the above inequalities allow us
to estimate
Z x
jgn .x/ g.x/j dt D .x a/ .b a/ D :
a b a b a b a
For the second part, recall that uniform convergence implies pointwise conver-
gence. Therefore, gn .b/ ! g.b/ as n ! C1, which is the same as (11.12). t
u
Problem 5 extends the above result to the realm of integrable functions. Actually,
the following important remark holds true.
Remark 11.15 In the notations of the previous result, a much more general result
is available. In order to state it properly, let I R be an interval with endpoints
˛ and ˇ (in the sense of Sect. 10.9) and fn W I ! R be given integrable functions
(in the improper sense, if I or f is unbounded), with .fn /n1 converging pointwise
to an integrable function f W I ! R. If there exists an integrable function F W I !
Œ0; C1/ such that
jfn .x/j L; 8 x 2 I; n 2 N:
On the other hand, item (b) of Example 11.9 shows that such a condition is necessary
for Lebesgue’s DCT to hold true.
Finally, Problem 4 shows that proving (11.12) in a particular case and without
the aid of Lebesgue’s DCT can be a somewhat difficult task. Problem 18 asks you
to prove a very important consequence of Lebesgue’s DCT.
We continue our study of uniform convergence by presenting the famous
Weierstrass approximation theorem, that states that every continuous function
defined in a closed and bounded interval can be uniformly approximated by a
sequence of polynomials.
Theorem 11.16 (Weierstrass) Every continuous function f W Œa; b ! R is the
uniform limit of a sequence of polynomial functions pn W Œa; b ! R.
Before we jump into the proof, it is worth to do some heuristics to motivate it.
For n 2 N, the binomial theorem gives
!
n Xn
n k
f .x/ D f .x/ x C .1 x/ D f .x/ x .1 x/nk
kD0
k
for large n, and the right hand side is a polynomial function. This is precisely what
we shall prove.
Proof To make the forthcoming computations a little simpler, assume first that a D
0 and b D 1, and let
!
Xn k n
pn .x/ D f xk .1 x/nk :
kD0
n k
11.2 Series of Functions 461
Since f is uniformly continuous in Œ0; 1, given > 0 we can choose ı > 0
such that jf .x/ f .y/j < whenever x; y 2 Œ0; 1 satisfy jx yj < ı. Letting
M D maxfjf .t/jI t 2 Œ0; 1g, we have for a fixed x 2 Œ0; 1 that
!
ˇX n k n ˇ
ˇ ˇ
jf .x/ pn .x/j D ˇ f .x/ f xk .1 x/nk ˇ
kD0
n k
!
X ˇˇ k ˇ n
ˇ
ˇf .x/ f ˇ xk .1 x/nk
0kn
n k
jx nk j<ı
!
X ˇˇ k ˇ n
ˇ
C ˇf .x/ f ˇ xk .1 x/nk
0kn
n k
jx nk jı
! !
X n k X n k
x .1 x/ nk
C 2M x .1 x/nk :
0kn
k 0kn
k
jx nk j<ı jx kn jı
In the last line above, the first summand does not exceed
!
Xn
n k n
x .1 x/nk D x C .1 x/ D :
kD0
k
On the other hand, the difficulty in estimating the second summand lies in
estimating how many integers 0 k n satisfy the inequality jx nk j ı. We
2
overcome this by inserting the factor ı12 x nk 1 under the † sign to get
! n
!
X n k 2M X k 2 n
2M x .1 x/nk 2 x xk .1 x/nk :
0kn
k ı kD0 n k
jx nk jı
Letting
n
!
X k 2 n k
SD x x .1 x/nk ;
kD0
n k
we proved that
2M
jf .x/ pn .x/j C S (11.14)
ı2
2 2k
k 2
Substituting x nk D x2 n
x C n
we compute
! !
Xn
n k X n
k n k
2
S Dx x .1 x/ nk
2x x .1 x/nk
kD0
k kD0
n k
!
X k2 n
n
C 2 k
xk .1 x/nk
kD0
n
! !
X n
n 1 k1 X n
k n1 k
2 2
D x 2x x .1 x/ nk
C x .1 x/nk
kD1
k 1 kD1
n k 1
! !
X k1 n1
n Xn
1 n1 k
2 2
D x 2x C x .1 x/
k nk
C x .1 x/nk
kD1
n k 1 kD1
n k 1
!
n 1 X n
k1 n1 k x
D x2 C x .1 x/nk C
n kD1
n1 k1 n
!
n 1 X n
n 2 k2 x
D x2 C x2 x .1 x/nk C
n kD2
k2 n
n 1 x 1
D x2 C x2 C D x x2 ;
n n n
1
so that S 4n on the interval Œ0; 1.
Therefore, back to (11.14) we get
M
jf .x/ pn .x/j C ;
2nı 2
for every x 2 Œ0; 1. Now, it suffices to see that this last expression is less than 2,
provided n > 2ıM
2.
For the general case,let f W Œa; b ! R be a continuous function. Let g W Œ0; 1 !
R be given by g.x/ D f .1 x/a C xb , so that g is also continuous. What we did so
far guarantees the existence of a sequence .qn /n1 of polynomial functions such that
n ya
qn ! g uniformly on Œ0; 1. If pn .y/ D qn ba , then pn is obviously a polynomial
function and, for y 2 Œa; b,
ˇ y a y a ˇ
ˇ ˇ
jf .y/ pn .y/j D ˇg qn ˇ:
ba ba
n
Thus, pn ! f uniformly on Œa; b. t
u
11.2 Series of Functions 463
for x 2 Œa; b, we get a polynomial function satisfying pn .a/ D f .a/ and p0n D qn for
every n 2 N.
Hence, we can use the FTC again to write
Z Z Z
x x x
f .x/ pn .x/ D f 0 .t/dt qn .t/dt D f 0 .t/ qn .t/ dt
a a a
P
and note that f .x/ D k1 fk .x/ for every x 2 I.
464 11 Series of Functions
t
u
The former corollary will only be useful if we have an efficient way of finding
out, in cases of interest, whether or not a given series of functions is uniformly
convergent. The coming result, known in mathematical literature as Weierstrass M-
test, provides a simple sufficient condition for the uniform convergence of a series
of functions. Note that the “M” stands for majorization.
P
Theorem 11.20 (Weierstrass M-Test) Let k1 fk be a series of functions defined
in an interval I and satisfying the following conditions:
(a) For each k P 1, there exists Mk > 0 such that jfk .x/j Mk for every x 2 I.
(b) The series k1 Mk converges.
P
Then, the series of functions k1 fk converges
P uniformly in I. In particular, if all of
the fk ’s are continuous in I, then so is k1 fk .
P
Proof Given x 2 I, since jfk .x/j Mk for every n 1 and k1 Mk converges,
the comparison test for series of real
P numbers (cf. Proposition 7.44) guarantees the
absolute convergence of the series k1 fk .x/. Hence, we get a well defined function
f W I ! R such that
X
f .x/ D fk .x/; 8 x 2 I:
k1
11.2 Series of Functions 465
P P
In order to establish the uniform convergence of k1 fk , let sn D nkD1 fk . For
x 2 I and n 2 N, we have
ˇ ˇ
ˇX ˇ X X
ˇ ˇ
jf .x/ sn .x/j D ˇ fk .x/ˇ jfk .x/j Mk : (11.15)
ˇ ˇ
k>n k>n k>n
Pn P
Now, given > 0, the convergence of the sequence kD1 Mk n1 to k1 Mk
assures the existence of n0 2 N such that
ˇ ˇ
ˇ ˇ
ˇX Xn
ˇ
ˇ M M ˇ
k ˇ < ; 8 n n0 :
ˇ k
ˇ k1 kD1 ˇ
ˇP ˇ
Hence, ˇ k>n Mk ˇ < for every n n0 .
Back to (11.15) we conclude that
X
jf .x/ sn .x/j Mk <
k>n
for n n0 and every x 2PI. Therefore, .sn /n1 converges uniformly to f , and this is
the same as saying that k1 fk converges uniformly to f .
The last part follows at once from Theorem 11.13. t
u
Example 11.21
P The Weierstrass M-test can be used to show that the series of
functions k1 k12 sin.kx/ converges uniformly in R. Indeed, on the one hand, since
j sin.kx/j 1 for every k 1 and x 2 R, we get
ˇ ˇ
ˇ1 ˇ
ˇ sin.kx/ˇ 1
ˇ k2 ˇ k2
P 1
for every k 1 and x 2 R. On the other, just note that the series k1 k2 is
convergent.
Accidentally,
P the previous example also shows that, given a uniformly
P convergent
series k1 fk of differentiable functions fn W I ! R, the series k1 fk0 of their
derivatives is not necessarily convergent, even pointwise. Indeed,
X d 1 X
1
2
sin.kx/ D cos.kx/;
k1
dx k k1
k
which doesn’t converge pointwise in any real x of the form x D 2`, with ` 2 Z.
We finish this section by using the material developed here to give an example of
a continuous function which is not differentiable at any point. For what comes next,
the reader might find it helpful to read the statement of Example 8.23 again. Up to
details, we follow the discussion in Sect. 9.7 of [9].
466 11 Series of Functions
Example 11.22 For y 2 R, let d.y/ denote the distance from x to the nearest integer.
If f W R ! R is given by
X 9 k
f .x/ D d.10k x/;
k1
10
This will contradict an obvious slight modification of Problem 8, page 300 (by
allowing an x < bn , instead of an < x < bn , there).
Given x > 0 and an integer n 2, let mn 2 N be such that mn 10n x <
mn C 1. If an D 10n mn and bn D 10n .mn C 1/, we obviously have an x < bn
n
ˇand bkn an D k10 ˇ. If k n is also integer, then 10 an ; 10 bn 2 N, so that
k k
ˇd.10 bn / d.10 an /ˇ D 0; on the other hand, if k < n, then Example 8.23 gives
ˇ ˇ
ˇd.10k bn / d.10k an /ˇ 10kn : (11.16)
Therefore,
n1
X 9 k
f .bn / f .an / D d.10k bn / d.10k an / : (11.17)
kD0
10
n C1
Since 10n1 an D m10n and 10n1 bn D m ˇ 10 , 1
by separately considering
ˇ the cases
0 < x < 1 and x 1 one easily sees that ˇd.10 bn d.101 an /ˇ D 10 1
. Therefore,
the triangle inequality in (11.17), together with (11.16), gives
11.2 Series of Functions 467
1 9 n1 ˇˇ X 9 k ˇˇ
n2
jf .bn / f .an /j ˇ d.10k bn / d.10k an / ˇ
10 10 kD0
10
9n1 X 9 k ˇˇ ˇ
n2
d.10k bn / d.10k an /ˇ
10 n
kD0
10
9n1 X 9 k 1 X k
n2 n2
9n1
10 kn
D 9
10n kD0
10 10n 10n kD0
7 9n1 C 1 7 9n1 C 1
D D jbn an j:
8 10n 8
Thus,
ˇ f .b / f .a / ˇ 7 9n1 C 1 n
ˇ n n ˇ
ˇ ˇ ! C1:
bn an 8
t
u
d X X
fk .x/ D fk0 .x/
dx k1 k1
exists, for each k 2 ZC . Prove that, for any given continuous function f W
Œ0; 1 ! R, the limit
Z 1
lim f .x/'n .x/dx
n!C1 0
also exists. P
10. With respect to the series of functions k1 1k sin.kx/, do the following items:
(a) Use Abel’s criterion (cf. Problem 18, page 243), together with the discus-
sion of items (b) and (c) of Problem 20, page 244, to show that the given
series converges pointwise in the interval .0; 2/.
(b) Revisit the proof of Abel’s criterion, as sketched in the hints given to
Problem 18, page 243, to show that Abel’s identity guarantees that the
convergence of item (a) is uniform in every interval of the form Œı; 2 ı,
for 0 < ı < .
11.2 Series of Functions 469
a0 .f / X
C .ak .f / cos.kx/ C bk .f / sin.kx// ;
2 k1
where
Z Z
1 1
ak .f / D f .x/ cos.kx/dx and bk .f / D f .x/ sin.kx/dx
1 1
ak .f / D bk .f 0 / and bk .f / D ak .f 0 /:
k k
1 1
ak .f / D ak .f 00 / and bk .f / D bk .f 00 /:
k2 k2
Then, conclude that the Fourier series of f converges uniformly in R and,
hence, defines a continuous function in R.
(c) Yet assuming f to be twice continuously differentiable, let g W R ! R be
the continuous function given by
a0 .f / X
g.x/ D C .ak .f / cos.kx/ C bk .f / sin.kx// :
2 k1
2
Fourier’s studies on heat conduction, collected in his famous book Théorie Analytique de la
Chaleur, laid the groundwork for modern Mathematical Physics, which in turn has greatly
influenced the development of the theory of Partial Differential Equations.
3
In spite of being known by the name of Fourier, such a result is actually due to Dirichlet.
470 11 Series of Functions
1
lim f .x/ C lim f .x/ :
2 x!x 0 x!x 0 C
1 1 1
D 1 C C :
4 3 5 7
14. (Miklós-Schweitzer—adapted) Let p > 0 be a real number and f ; g W R ! R
be continuous functions, with g being periodic of period p.
(a) Prove that
n Z
X
kp
kp Z p 1X n kp
n
f g.nx/dx D g.y/dy f
.k1/p n n kD1 n
kD1 n 0
Z
p Z
p
n 1
! f .x/dx g.y/dy :
p 0 0
4
We shall give a self-contained proof of such a formula in Problem 9, page 484.
5
After Lipót Fejér, Hungarian mathematician of the XX century.
11.2 Series of Functions 471
a0 X
n
Sn f .x/ D C ak cos.kx/ C bk sin.kx/
2 kD1
then F is continuous.
(b) Let the function x 7! f .x; t/ be continuously differentiable for each t 2 J,
@f
and let @x denote its derivative. Assume that for every x0 2 I and every
sequence .xn /n1 in I converging to x0 , there exist integrable
ˇ functions
ˇ gW
ˇ @f ˇ
J ! R and G W J ! R such that jf .xn ; t/j g.t/ and ˇ @x .xn ; t/ˇ G.t/
@f
for every n 1 and every t 2 J. Then, t 7! f .x; t/ and t 7! @x .x; t/
are integrable for each x 2 I and F W I ! R, defined as in item (a), is
continuously differentiable in I, with
Z ˇ
@f
F 0 .x/ D .x; t/dt: (11.18)
˛ @x
6
I.e., such that both t 7! f .x; t/ and x 7! f .x; t/ are continuous functions.
472 11 Series of Functions
where a0 ; a1 ; a2 ; : : : are given real numbers. In this case, as it happens with number
series, we say that an .x x0 /n is the n-th term of the power series and that an is the
n-th coefficient of it.
In this section we develop the basic aspects of the theory of power series, starting
with the following central result.
P
Theorem 11.23 Given a power series k0 ak .x x0 /k , there exists 0 R C1
such that the series:
(a) Converges absolutely in the interval .x0 R; x0 C R/ and diverges in
R n Œx0 R; x0 C R.
(b) Converges uniformly in the interval Œx0 r; x0 C r, 8 0 < r < R.
P
Proof Firstly, note that the series k0 ak .x x0 /k converges absolutely in the
P
interval .x0 R; x0 C R/ if and only if the series k0 ak xk converges absolutely
P
in the interval .R; R/. Accordingly, k0 ak .x x0 /k converges uniformly in
P
Œx0 r; x0 C r if and only if k0 ak xk converges uniformly in Œr; r. Therefore,
we can assume that x0 D 0.
Let’s first deal with item (a).
P
Claim 4 if k0 ak xk converges at x D ˛ ¤ 0, then it converges absolutely at any
x 2 .˛; ˛/.
Indeed, for such an x, we have
X X ˇ x ˇk
ˇ ˇ
jak xk j D jak ˛ k j ˇ ˇ : (11.20)
k0 k0
˛
P
Since ak ˛ k converges, Proposition 7.36 guarantees the existence of n0 2 N
k0
ˇ ˇk
such that n n0 ) jan ˛ n j < 1; therefore, it follows from (11.20) that jak xk j ˇ ˛x ˇ
11.3 Power Series 473
ˇ x ˇk P ˇ x ˇk ˇ ˇ
ˇ ˇ for k n0 . Since the geometric series ˇ ˇ is convergent (for, ˇ x ˇ < 1),
˛ kn0 ˛ P ˛
the comparison test assures that the same holds true for kn0 jak xk j and, thus, for
P
k0 jak x j.
k
P
Claim 5 if k0 ak xk diverges at x D ˇ ¤ 0, then it also diverges at any x 2 R
such that jxj > jˇj. P
For such an x, if k0 ak xk converged, the previous claim would assure the
P
(absolute) convergence of the series k0 ak ˇ k , which is an absurd.
P
Claims 4 and 5 guarantee that, with respect to k0 ak xk , one of the three
following possibilities does happen: (i) it converges only at x D 0; (ii) it is absolutely
convergent at any x 2 R; (iii) there exist ˛; ˇ ¤ 0 such that j˛j < jˇj and the series
converges absolutely when jxj < ˛ and diverges when jxj > jˇj.
If either (i) or (ii) happens, there is nothing left to do. If (iii) happens, let
X
R D supfr > 0I ak uk converges absolutely when juj < rg:
k0
P
We claim that k0 ak xk converges absolutely when jxj < R and diverges when
jxj > R. To this end, let’s look at two separate cases:
P
(i) If jxj < R, take r such that jxj < r < R and k0 ak uk converges absolutely
P
when juj < r. Then, in particular k0 ak xk is absolutely convergent.
P
(ii) If jxj > R and k0 ak xk converged, then, taking RQ satisfying R < RQ < jxj,
P k
it would follow from Claim 4 that k0 ak u converges absolutely when
P
Q But this obviously contradicts the definition of R. Therefore, k0 ak xk
juj < R.
diverges.
P
For item (b), given 0 < r < R, it follows from what we did above that k0 jak rk j
converges. Hence, letting Mk D jak rk j, we have,P for jxj r, that jak xk j jak rk j D
Mk . Therefore, Weierstrass M-test assures that k0 ak xk converges uniformly in
the interval Œr; r. t
u
In the notations of the previous result, we say that 0 R C1 is the radius
of convergence of the power series (11.19), and .x0 R; x0 C R/ is its interval of
convergence.
Problems 10 and 11 give a general formula for the radius of convergence in
terms of the coefficients of a given power series. For the time being, the following
consequence of the previous result allows us to easily compute it in a number of
interesting examples. At this point, we suggest that the reader runs through the proof
of the ratio test (cf. Proposition 7.51) once more, just to note that it remains valid if
(in the corresponding notations) l D C1.
Corollary 11.24 Let .an /n0 be a sequence
ˇ ˇ of nonzero real numbers. IfPthere exists
ˇ an ˇ
0 R C1 such that limn!C1 ˇ anC1 ˇ D R, then the power series k0 ak .x
x0 /k has radius of convergence equal to R.
474 11 Series of Functions
Proof Since
ˇ ˇ
ˇ anC1 .x x0 /nC1 ˇ jx x0 j n jx x0 j
ˇ ˇ ˇ ˇ !
ˇ a .x x /n ˇ D ˇ an ˇ R
;
n 0 ˇ anC1 ˇ
P
the ratio test assures that k0 ak .x x0 /k converges absolutely if jxx
R
0j
< 1 and
jxx0 j jxx0 j
diverges if R > 1. Since R < 1 if and only if jx x0 j < R, the previous
result guarantees that R is precisely the radius of convergence of the given power
series. t
u
Example 11.25
P ˇ ˇ
ˇ 1=n ˇ
(a) The power series k1 1k xk has radius of convergence 1, for ˇ 1=.nC1/ ˇ D nC1
n
!
1 as n ! C1. ˇ ˇ
P ˇ nŠ ˇ 1
(b) The series k1 kŠxk has radius of convergence 0, for ˇ .nC1/Š ˇ D nC1 ! 0.
P ˇ ˇ
1 ˇ 1=nŠ ˇ
(c) The series k1 kŠ .x 2/ has radius of convergence C1, for ˇ 1=.nC1/Š ˇ D
k
1 ˇ ˛ ˇ 1
has radius of convergence j˛j , for ˇ ˛nC1 ˇ D j˛j for every n 0. By
1
Proposition 7.38, for jxj < j˛j
we have
X 1
.˛x/k D : (11.21)
k0
1 ˛x
P
Theorem 11.27 If the power series k0 ak .x x0 /k has radius of convergence
R > 0, then:
P
(a) The function f W .x0 R; x0 C R/ ! R, given by f .x/ D k0 ak .x x0 /k , is
infinitely differentiable. P
(b) For every n 2 N, we have f .n/ .x/ D kn .kn/Š
kŠ
ak .x x0 /kn for every x 2
.n/
.x0 R; x0 C R/, and the series defining f also has radius of convergence
equal to R.
Proof As in the proof of Theorem 11.23 we can assume, without any loss of gen-
We shall first prove that, letting RQ be thePradius of convergence
erality, that x0 D 0. P
of the power series k1 kak xk1 , then RQ D R and f 0 .x/ D k1 kak xk1 for every
x 2 .R; R/.
Note first that jkak xk j Pjak xk j for every integer k 1 and every x 2 R.
Therefore, if the power series k1 kak xk1 converges absolutely at some x, then the
P P
same holds true for the power series k1 kak xk and, thus, for k0 ak xk . Hence,
RQ R.
Now, for 0 < x < R and 0 < h < R x, we have
0 1
f .x C h/ f .x/ 1 @X X X .x C h/k xk
D ak .x C h/k a k xk A D ak :
h h k1 k1 k1
h
On the other hand, Lagrange’s MVT gives, for each k 1, some ck 2 .x; x C h/
such that .xCh/h x D kck1
k k
k . Therefore,
f .x C h/ f .x/ X
D k ;
kak ck1
h k1
P
so that the power series k1 kak ck1 converges absolutely. Since jkak xk1 j
k P
jkak ck1
k j, the comparison test guarantees that k0 kak x
k1
also converges abso-
lutely. Analogously, such a series converges absolutely if R < x < 0, so that
RQ R.
P Our discussion up to this point assures that g W .R; R/ ! R, given by g.x/ D
k1
k0 ka k x , is a well defined function. By item (b) of the previous proposition,
we have
Z x X kak X
g.t/dt D xk D ak xk D f .x/ a0 :
0 k1
k k1
X kŠ
f .mC1/ .x/ D .k m/ak .x x0 /km1
kmC1
.k m/Š
X kŠ
D ak .x x0 /k.mC1/ :
kmC1
.k .m C 1//Š
This completes the inductive argument and, thus, the proof of (b). t
u
In the notations of item (b) of the previous result, we observe that the power
series expansion of f .n/ is obtained by termwise differentiating, n times, the power
dn
series expansion of f . Indeed, letting dx n denote the n-th derivative of a function (if
it exists), an immediate computation gives
dn kŠ
.x x0 /k D .x x0 /kn :
dx n .k n/Š
We now collect two useful consequences of the previous theorem, the first of
which refines the analysis of item (b) of Proposition 11.26.
P
Corollary 11.28 If the power series k0 ak .x x0 /k has radius ofPconvergence
R > 0, then the radius of convergence of the integrated power series k0 kC1 ak
.x
x0 /kC1 is also equal to R.
P d P
Proof Since k0 ak .x x0 /k D dx k0 kC1 .x x0 /
ak kC1
, item (b) of the previous
theorem guarantees that both of the given series have the same radius of conver-
gence. t
u
P
Corollary 11.29 Assume that the power series k0 ak .x x0 /k has radius of
convergence R > 0, and let f W .x0 R; x0 C R/ ! R be given by f .x/ D
P
k0 ak .x x0 / . Then:
k
.n/
an D f nŠ.x0 / , for every n 0.
(a) P
k0 ak .x x0 / is the Taylor series of f .
k
(b)
Proof Item (b) follows immediately from (a). For (a), item (b) of the previous result
.n/
gives f .n/ .x0 / D .nn/Š
nŠ
an D nŠan ; therefore, an D f nŠ.x0 / . t
u
We shall now see that the results presented so far allow us to reobtain, by means
of a unified approach, some of the Taylor expansions discussed in Sect. 11.1. In this
respect, see also Problem 3.
11.3 Power Series 477
1 P
Example 11.30 Item (e) of Example 11.25 assures that 1Cx D k0 .1/k xk , with
radius of convergence equal to 1. Hence, item (b) of Proposition 11.26 gives, for
jxj < 1,
Z X Z X .1/k
x
1 x
log.1 C x/ D dt D .1/k tk dt D xkC1 : (11.22)
0 1Ct k0 0 k0
kC1
1
D 1 x2 C x4 x 6 C
1 C x2
for jxj < 1. Therefore, again by item (b) of Proposition 11.26, we have, for jxj < 1,
Z X Z
x
1 x
arctan x D dt D .1/k t2k dt
0 1 C t2 k0 0
x3 x5 x7
D x C C : (11.23)
3 5 7
Problems 8 and 9 will show that (11.22) and (11.23) remain true for x D 1.
Example
P 11.32 It follows easily from Corollary 11.24 P that the power series
1 k 1 k
k0 kŠ x converges in the whole real line. If f .x/ D k0 kŠ x for x 2 R,
then item (b) of Theorem 11.27 furnishes
X 1 X 1
f 0 .x/ D xk1 D xk D f .x/:
k1
.k 1/Š k0
kŠ
Therefore,
d x
.e f .x// D ex .f .x/ C f 0 .x// D 0;
dx
so that x 7! ex f .x/ is a constant function. Finally, since f .0/ D 1, we get ex f .x/ D
1 or, which is the same, f .x/ D ex .
We now apply Theorem 11.27 to show that, for any ˛ 2 R n f0g and jxj < 1, one
can write .1 C x/˛ as a convergent power series. To this end, given
˛ 2 R and
an
integer n 0, we define the generalized binomial number ˛n by letting ˛0 D 1
and, for n 1,
!
˛ ˛.˛ 1/.˛ 2/ : : : .˛ n C 1/
D : (11.24)
n nŠ
478 11 Series of Functions
t
u
The coming result is known in mathematical literature as the binomial series
theorem, or simply as the binomial theorem, and is due to Newton. Notice
that (11.25) generalizes (4.11), for ˛k D 0 if ˛ 2 N and k > ˛.
Theorem 11.34 (Newton) For ˛ ¤ 0 and jxj < 1, we have
!
X ˛
˛
.1 C x/ D xk : (11.25)
k0
k
P
Corollary 11.24 assures that k0 ˛k xk has radius of convergence equal to 1.
Hence,
P ˛by k Theorem 11.27, the function f W .1; 1/ ! R given by f .x/ D
k0 n x is differentiable. Also from that result,
! !
X ˛ k1 X ˛ 1 k1
0
f .x/ D k x D ˛ x ; (11.26)
k1
k k1
k1
where we used item (c) of the previous lemma in the last equality above. Therefore,
!
X ˛1
0
.1 C x/f .x/ D ˛.1 C x/ xk1
k1
k 1
0 ! ! 1
X ˛1 X ˛1
D ˛@ xk1 C xk A
k1
k 1 k1
k 1
0 ! ! 1
X ˛1 X ˛1
D ˛ @1 C xk1 C xk1 A
k2
k 1 k2
k 2
0 ! 1 !
X ˛ X ˛
D ˛ @1 C x k1 A
D˛ xk
k2
k1 k0
k
D ˛f .x/:
Therefore, g is constant in .1; 1/. Since g.0/ D 1, we get .1 C x/˛ f .x/ D 1 for
jxj < 1, as wished. P
For the general case, suppose that .1 C x/˛ D k0 ˛k xk for some ˛ ¤ 0 and
every x 2 .1; 1/. Let us show that similar formulas hold true for ˛ 1 and ˛ C 1
(and every x 2 .1; 1/):
(a) Theorem 11.27, together with item (b) of Lemma 11.33, gives, for jxj < 1,
!
˛1 1 d ˛ 1 X ˛ k1
.1 C x/ D .1 C x/ D k x
˛ dx ˛ k1 k
! !
X ˛1 X ˛1
D x k1
D xk :
k1
k 1 k0
k
480 11 Series of Functions
By induction, we conclude that (11.25) is true for all ˛ ¤ 0 and jxj < 1. t
u
Corollary 11.35 Given ˛; ˇ ¤ 0, we have
!
X ˛
˛
.1 C ˇx/ D .ˇx/k
k0
k
1
for every x 2 R such that jxj < jˇj .
Proof It suffices to apply (11.25) with ˇx instead of x, noticing that jˇxj < 1 ,
1
jxj < jˇj . u
t
1
Example 11.36 As an application of the corollary above, note that for jxj < 2
we have
!
X 1=2
1=2
.1 2x/ D .2x/k ;
k0
k
with
!
1=2 1 1 1 1
D 1 ::: n C 1
n nŠ 2 2 2
.1/n 1 3 5 2n 1
D
nŠ 2 2 2 2
.1/n .2n/Š
D n
nŠ 2 2 4 .2n/
!
.1/n .2n/Š .1/n 2n
D D :
nŠ 2n 2n nŠ 4n n
1
Hence, for jxj < 2 we have
!
X 1 2k
1=2
.1 2x/ D xk :
k0
2 k k
11.3 Power Series 481
X
n
X
a k Rk xk C a k Rk ;
kD0 k>n
X
n
x 2 .R ı; R/ ) ak Rk xk < :
kD0
With such choices and in view of the above computations, we conclude that
X
n
x 2 .R ı; R/ ) 0 ak Rk f .x/ < 2:
kD0
t
u
Remark 11.38 In the notations of the statement of the previous lemma, Problem 7
shows that the assumption on the sign of the ak ’s can be dropped.
It is somewhat surprising that the converse of the previous lemma also holds true.
Such a result is known in mathematical literature as a tauberian theorem after the
work of Alfred Tauber, Austrian mathematician of the XIX and XX centuries. For
the sake of simplifying the notation, we assume that R D 1, leaving to the reader
the (easy!) task of dealing with the general case.
482 11 Series of Functions
P
Theorem 11.39 Let f .x/ D k
k0 ak x in Pthe interval .1; 1/, with ak 0 for
every
P k 0. If lim x!1 f .x/ does exist, then k0 ak converges and limx!1 f .x/ D
a
k0 k .
Proof The last part follows from the previous lemma. For what is left to do, note
that, for 0 x < 1,
X 2 X1
jC1
X
xf 0 .x/ D kak xk D kak xk
k1 j0 kD2j
X 2 X1
jC1
2 j a k x2
jC1 1
j0 kD2j
X 2 X1 jC1
jC1
D 2 j
ak x2 1 :
j0 kD2j
Z x X
l X1 Z x jC1
2jC1
0
tf .t/dt 2 j
ak t2 1 dt
0 0
jD0 kD2j
Xl 2jC1
X1 x2jC1
D 2 j
ak jC1
jD0 j
2
kD2
1 X X 2jC1
l 2 1 jC1
D ak x :
2 jD0 j kD2
l 2X
X 1 jC1
jC1
ak x2 2a
jD0 kD2j
2lC1
X1 l 2X
X 1 jC1
ak D ak 2a:
kD1 jD0 kD2j
11.3 Power Series 483
P
Finally, since the sequence of partial sums of the series k0 ak is nondecreas-
ing, this last inequality gives
X
ak a0 C 2a:
k0
t
u
Example 11.40 By applying the formula of Problem 11 for the radius of conver-
gence,
P we conclude that .1; 1/ is the interval of convergence of the power series
2j
j0 x . Alternatively, we can observe that, for jxj < 1,
X X
x2
j
0 jxjk < C1:
j0 k1
On the other hand, since the sequence of the coefficients of this series clearly
diverges, the tauberian theorem assures that
X
x2 D C1:
j
lim
x!1
j1
P
x2 as x ! 1, see
j
For a more refined estimate on the growth of j1
Problem 12.
3. Use the approach of Example 11.32, together with the result of Example 9.48,
to show that
X .1/k1 X .1/k
sin x D x2k1 and cos x D x2k :
k0
.2k 1/Š k0
.2k/Š
6. The formula of Example 11.30 can be modified in order to compute log a for
every a > 0. To this end, do the following items:
(a) Show that, for x 2 .1; 1/, one has
1 1Cx x3 x5 x7
log DxC C C C :
2 1x 3 5 7
X
m
n X
m
ak x D rn y rmC1 y C
k m
rk yk yk1 :
kDn kDnC1
ˇXm ˇ
ˇ ˇ
ˇ ak xk ˇ jrn j C jrm j C sup jrk j yn ym :
kDn k>n
P
(c) Show that k0 ak xk converges uniformly on the interval Œ0; R. Then,
conclude that (11.27) is true.
8. Use the result of the previous problem to show that
1 1 1
log 2 D 1 C C :
2 3 4
11.3 Power Series 485
9. * Use the material of this section to give a self-contained proof of the Leibniz
formula for :
1 1 1
D 1 C C :
4 3 5 7
For the next two problems, we extend the notion of supremum by saying that
a sequence .ak /k0 has supremum C1 if it is unbounded from above; in this
case, we write
supfak I k 0g D C1:
Accordingly, we extend the notion of infimum by saying that a sequence .ak /k0
has infimum 1 if it is unbounded from below; in this case, we write
inffak I k 0g D 1:
10. Let .ak /k0 be a sequence of real numbers. Define its limit superior, denoted
lim sup ak , by letting
(a) Prove that lim sup ak is a well defined concept and that, if sj D
supfaj ; ajC1 ; : : :g 2 R [ fC1g, then lim sup ak D inffsj I j 0g 2
R [ f˙1g.
(b) If lim sup ak 2 R, prove that it is the only real number M satisfying the two
following conditions:
i. Given > 0, there exists infinitely many n 2 N such that an > M .
ii. Given > 0, there exists at most finitely many n 2 N such that an
M C .
P
11. Let k0 ak .x x0 /k be a power series with radius of convergence R. The
purpose of this problem is to prove that
1
RD p ;
lim sup k
jak j
X X r k
jxj r ) jak xk j < C1:
kk0 kk0
s
486 11 Series of Functions
(b) Fix jxj > R and show that there exist infinitely
P many indices k 2 N such
that jak xk j > 1. Conclude that the series k0 ak xk diverges.
(c) Conclude that R, defined as above, is indeed the convergence radius of the
given series.
P
12. As we saw in Example 11.40, the power series j1 x2 defines a differentiable
j
In this section, we briefly discuss the use of power series in Algebra and differential
equations. For further applications of power series to Algebra and Combinatorics,
we refer the reader to [5].
We begin with the following
Definition 11.41 The (ordinary) generating function7 of a sequence of real
numbers .an /n0 is the power series
X
a k xk : (11.28)
k0
The previous definition suggests that the main difference between the theory of
power series and the method of generating function lies in a change of point of
view. In the first case, we are primarily interested in examining the properties of the
function f defined by the power series at the right hand side of (11.28); in the second
(as we shall see next), we want to use the properties of f to infer conclusions about
the terms of the sequence .an /n0 .
Let’s illustrate this by revisiting Problem 20, page 220 with the aid of generating
functions.
Example 11.42 (TT) The set of naturals is partitioned into m disjoint, infinite and
nonconstant arithmetic progressions, of common ratios d1 , d2 , . . . , dm . Prove that
1 1 1
C CC D 1:
d1 d2 dm
P
Proof If f .x/ D k1 xk and ai is the initial term of the i-th AP (that of common
ratio di ), then the given condition, together with item (e) of Example 11.25, gives
7
In opposition to exponential generating functions, cf. Chap. 3 of [5], for instance.
11.4 Some Applications 487
X
m X
m
xai
f .x/ D .xai C xai Cdi C xai C2di C / D :
iD1 iD1
1 xdi
X
m
xai
xD (11.29)
iD1
1CxC x2 C C xdi 1
for jxj < 1. Now, both sides of this last equality define functions continuous in Œ0; 1
and which coincide in Œ0; 1/; hence,Psuch functions coincide for x D 1 too, so that,
1
letting x D 1 in (11.29), we obtain m iD1 di D 1. t
u
The use of generating functions is particularly useful in the study of sequences
.an /n0 satisfying a given linear recurrence relation. Here, as in Sect. 3.2, we shall
treat the cases of (certain) linear recurrence relations of orders 2 and 3, postponing
the analysis of the general case to Chap. 20 of [5]. P
The idea is to consider the generating function k0 ak xk corresponding to the
given sequence and, then, follow the various stages below, which comprise a sort
receipt for several similar problems:
I. Use the initial terms of the sequence, as well as the recurrence relation it
satisfies, to conclude that the given generating function converges at some
interval of the form .r; r/.
II. Again with the aid of the initial terms and the given recurrence relation,
perform
P appropriate (generally algebraic) operations with the equality f .x/ D
k0 ak x to get a formula for f .x/.
k
Example 11.44 Let .Fn /n1 be the Fibonacci sequence, so that F1 D 1, F2 D 1 and
FkC2 D FkC1 C Fk for every integer k 1. Compute Fn as a function of n.
P
Solution Let f .x/ D k
k1 Fk x be the generating function corresponding to the
Fibonacci sequence, and let’s run through the previously listed stages I to IV.
Step I: note first that an easy induction gives Fn P2n for every n 1.
Hence, Lemma 11.43 guarantees that the power series k1 Fk xk converges in
the interval 12 ; 12 .
Step II: for x 2 . 12 ; 12 /, we can write
X
f .x/ D F1 x C F2 x2 C F k xk
k3
X
D x C x2 C .Fk1 C Fk2 /xk
k3
X X
D x C x2 C x Fk1 xk1 C x2 Fk2 xk2
k3 k3
x
f .x/ D :
1 x x2
1 1
Now, developing 1˛x and 1ˇx as geometric series, we obtain
0 1
1 @X X X ˛k ˇk
kA
f .x/ D p .˛x/
k
.ˇx/ D p xk ;
5 k0 k0 k1
5
11.4 Some Applications 489
n o
1 1 1
as long as jxj < min 2 ; j˛j ; jˇj D 12 . Then,
X X ˛k ˇk
Fk x D k
p xk ;
k1 k1
5
for every x 2 . 12 ; 12 /.
Step IV: finally, Corollary 11.29 assures that the power series expansion of f is
unique, so that the last equality above, together with the initial definition of f ,
gives Fn D ˛ pˇ for every n 1.
n n
5
t
u
Proceeding similarly to the above example, we now use generating functions to
give another proof of Theorem 3.16 (see also Problem 6).
Theorem 11.45 Given u; v 2 R, with v ¤ 0, let .an /n1 be such that akC2 D
uakC1 C vak for every k 1. Moreover, assume that the characteristic equation
x2 ux v D 0 has real roots ˛ and ˇ.
(a) If ˛ ¤ ˇ, then an D A˛ n1 C Bˇ n1 for every n 1, where A and B are the
A C B D a1
solutions of the linear system .
˛A C ˇB D a2
(b) If ˛ D ˇ, then an D.A C Bn/˛ n1 for n 1, where A and B are the solutions
A C B D a1
of the linear system .
.A C 2B/˛ D a2
Proof We start by showing that there exists q > 0 such that jan j qn for every
n 1. Indeed, assuming that jak j qk and jakC1 j qkC1 , it follows from the
triangle inequality that
so that jakC2 j qkC2 provided jujqkC1 C jvjqk qkC2 or, which is the same,
jujqCjvj q2 . Since the greatest 2
root of the second degree equation x jujxjvj D
1
p
2 2
0 is x0 D 2 juj C u C 4jvj , we get jujq C jvj q whenever q > x0 . Hence,
2
jan j qn for every n 1, as longp as ja1 j q, ja2 j q and q > x0 , for which it
suffices to choose q > maxfja1 j; ja2 j; x0 g.
Fix such a q, so that jan j qn for every n 1. Lemma11.43 guarantees that
P
the generating function k1 ak xk converges in the interval 1q ; 1q , thus defining
P
f W 1q ; 1q ! R by f .x/ D k1 ak xk .
In order to get a simpler expression for f .x/, we argue as in the previous example,
using the recurrence relation satisfied by the sequence .ak /k0 of the coefficients
of f :
490 11 Series of Functions
X X
f .x/ D a k xk D a 1 x C a 2 x2 C a k xk
k1 k3
X
D a 1 x C a 2 x2 C .uak1 C vak2 /xk
k3
X X
2
D a1 x C a2 x C ux ak1 xk1 C vx2 ak2 xk2
k3 k3
P
Finally, comparing this last expression with the fact that f .x/ D k1 ak xk , we
obtain ak D A˛ k1 C Bˇ k1 for every k 1.
2
2 ua1 /x
(b) If ˛ D ˇ, then 2˛ D u, ˛ 2 D v and f .x/ D a1 xC.a .1˛x/2
. This time we
try to decompose the right hand side of (11.30) in partial fractions by writing
8
A general theorem on the existence of partial fraction decomposition for quotients of polynomials
will be seen in [5]. For the time being, we shall content ourselves to describe what such a theorem
says in the simple cases we consider here.
11.4 Some Applications 491
f .x/ D 1˛x
Ax
C .1˛x/
Bx
2 , for some A; B 2 R. Obviously, such A and B must
A C B D a1
satisfy , and again this linear system has a unique solution,
˛A D ua1 a2
for v ¤ 0 ) ˛ ¤ 0.
1 1
Expanding 1˛x and .1˛x/2
i power series (for the second fraction with the aid
of Problem 4), we get
X X
f .x/ D Ax .˛x/k C Bx .k C 1/.˛x/k
k0 k0
X
D .A C B.k C 1//˛ k xkC1
k0
X
D .A C Bk/˛ k1 xk :
k1
Finally, arguing as in the previous case we get ak D .A C Bk/˛ k1 for every integer
k 1. t
u
Generating functions can also be used to deal with linear recurrence relations
with nonconstant coefficients. To present such an example, we first need the
following result.
P 11.46 If f ; g W .x0 R; xP
Proposition 0 C R/ ! R have power series expansions
f .x/ D k0 ak .x x0 /k and g.x/ D l0 bl .x x0 /l , thenPthe function fg W .x0
R; x0 C
P R/ ! R hasPpower series expansion .fg/.x/ D n0 cn .x x0 / , with
n
n
cn D kClDn ak bl D kD0 ak bnk .
Proof Theorem 11.23 assures that the power series defining f and g converge
absolutely in the interval .x0 R; x0 C R/. Hence, by Theorem 7.53 and for every
x 2 .x0 R; x0 C R/, we have
0 10 1
X X
f .x/g.x/ D @ ak .x x0 /k A @ bl .x x0 /l A
k0 l0
!
X X
D ak .x x0 / bl .x x0 /
k l
n0 kClDn
!
X X
D ak bl .x x0 /n
n0 kClDn
X
D cn .x x0 /n :
n0
492 11 Series of Functions
Finally, note that Theorem 7.53 guarantees that the convergence of this last series is
also absolute in the interval .x0 R; x0 C R/. t
u
Example 11.47 Let .an /n0 be the sequence such that a0 D 1, a1 D 1 and
ak1
ak D C 2ak2
k
for every integer k 2. Compute an as a function of n.
P
Solution Denoting by f .x/ D k
k0 ak x the generating function of the given
sequence, let’s once again run through the previously described stages I to IV.
Step I: once more we shall try to apply the comparison test for series, in the
form of Lemma 11.43. Assuming that jak2 j ˛ k2 , jak1 j ˛ k1 , the triangle
inequality gives
jak1 j ˛ k1
jak j C 2jak2 j C 2˛ k2 I
k k
˛ k1
hence, we shall have jak j ˛ k provided k
C 2˛ k2 ˛ k or, which is the
same,
˛
C 2 ˛2 :
k
Taking into account that a1 C a0 D 0, we get f 0 .x/ D .4x 1/f .x/ C 2xf 0 .x/ or,
which is the same,
In order to solve such a differential equation, first note that f is positive in the
interval .r; r/, for some 0 < r 12 (for, f .0/ D a0 D 1 > 0 and f continuous
imply f positive in some neighborhood of 0). Therefore, for jxj < r we can write
f 0 .x/ 4x 1 1
D D 2
f .x/ 2x 1 2x 1
Step III: note that the power series expansion of e2x is given by (11.7), with
a D 2, and holds in the whole real line:
X .2/k
e2x D xk :
k0
kŠ
Hence, it follows from Example 11.36 and Proposition 11.46 that, for f given as
in (11.31) and jxj < r 12 , we have
0 10 ! 1
X .2/k X 1 2l
f .x/ D @ xk A @ xl A
k0
kŠ l0
2 l l
!!
X X .2/k 1 2l
D l xn
n0 kClDn
kŠ 2 l
!!
X X n
.2/nl 1 2l
D l xn
n0 lD0
.n l/Š 2 l
!!
X Xn
.1/ nl
2l
D 2n xn :
n0 lD0
4 l .n l/Š l
494 11 Series of Functions
P
Step IV: comparing the last expression above with f .x/ D n0 an xn , it follows
once more from the uniqueness of power series expansions that
!
Xn
.1/nl 2l
an D 2 n
:
lD0
4l .n l/Š l
t
u
All of the above results can be seen as sorts of discrete versions of the following
result on ordinary differential equations:
Theorem 11.48 Let x0 2 R, R > 0 and p; q W .x0 R; x0 CR/ ! R be given by their
Taylor series centered at x0 . For ˛; ˇ 2 R, there exist a unique twice differentiable
function f W .x0 R; x0 C R/ ! R such that
00
f C pf 0 C qf D 0
: (11.32)
f .x0 / D ˛; f 0 .x0 / D ˇ
Moreover, f is actually infinitely differentiable and given by its Taylor series
centered at x0 .
Interesting (and important) applications of the above result can be found, for
instance, in Chap. 6 of [2], or in Chap. VI of [19]. We now discuss a particular case
of it, referring to Problem 11.48 for a proof of the general case. See, also, Problem 4,
page 510.
Given real numbers a and b and a twice differentiable function f W R ! R
such that
f 00 C af 0 C bf D 0; (11.33)
2 C a C b D 0 (11.34)
(b) f1 .x/ D e˛x and f2 .x/ D eˇx if > 0 and ˛ andpˇ are the roots of (11.34).
px
ax
and f2 .x/ D e 2 sin x
ax
(c) f1 .x/ D e 2 cos 2 2 if < 0.
In particular, for given values of f .0/ and f 0 .0/, (11.33) has exactly one solution.
11.4 Some Applications 495
Proof First of all, writing f 00 D af 0 bf it is immediate that any solution of (11.33)
is infinitely differentiable.
Claim 1 f is given by its Taylor series in the whole real line.
Fix M > 0 and choose C > 0 such that jf j C and jf 0 j C2 in ŒM; M. Since
f .kC2/ C af .kC1/ C bf .k/ D 0 for every k 2 ZC , we get
X
n1 .k/
f .0/ f .n/ .c/ n
f .x/ D xk C x ; (11.35)
kD0
kŠ nŠ
P
and, analogously, f 00 .x/ D Substituting these expressions into f 00 C
akC2 k
k0 kŠ
x.
af 0 C bf D 0, we find
X xk
.akC2 C aakC1 C bak / D 0;
k0
kŠ
496 11 Series of Functions
and the uniqueness of power series expansions gives akC2 C aakC1 C bak D 0 for
every k 0. However, since a0 D f .0/ and a1 D f 0 .0/, the sequence .ak /k0 , and
hence the function f , is completely determined.
Finally, to finish the proof it suffices to check, in each of the cases (a), (b) and
(c), that: (i) the given functions f1 and f2 satisfy (11.33); (ii) for any given values for
f .0/ and f 0 .0/, one is able to find real values for A and B such that
Once this has been done, the uniqueness of solution, as established in Claim 2,
allows us to conclude that f D Af1 C Bf2 .
Checking items (i) and (ii) in each of (a), (b) and (c) is actually quite simple. For
item (c), for instance, item (ii) amounts to showing that the system of equations
p
a
A D f .0/; A C B D f 0 .0/
2 2
always has a solution, which is immediate. As for (i), one just needs to compute fi0
and fi00 for i D 1; 2, then showing that fi00 Cafi0 Cbfi D 0. We leave this as an exercise
to the reader. t
u
Remark 11.50 Note that the prescription of the values of f .0/ and f 0 .0/, instead of
f .x0 / and f 0 .x0 / for some x0 2 R, is irrelevant. Indeed, letting g.x/ D f .x C x0 /, we
have g.0/ D f .x0 /, g0 .0/ D f 0 .x0 / and g00 C ag0 C bg D 0, so that we can apply the
previous result to find g and then find f .x/ D g.x x0 /.
Example 11.51 Find all differentiable functions f ; g W R ! R such that f .0/ D 0,
g.0/ D 1, f 0 D f C g and g0 D 2f .
Solution Since f 0 D f C g, which is a differentiable function, we conclude that f is
twice differentiable. Then f 00 D f 0 C g0 D f 0 C 2f or, which is the same,
f 00 f 0 2f D 0:
Now, the second degree equation 2 2 D 0 has real roots 2 and 1, so that
the previous theorem gives f .x/ D Ae2x C Bex for every real x. Since f .0/ D 0 and
f 0 .0/ D f .0/ C g.0/ D 1, we get A C B D f .0/ D 0 and 2A B D f 0 .0/ D 1.
Solving such a system of equations, it follows that A D 13 and B D 13 , so that
f .x/ D 13 e2x ex and g.x/ D f 0 .x/ f .x/ D 23 e2x C 2ex . t
u
11.4 Some Applications 497
a1 C a2 C C ak D m;
X
cn D ai1 ai2 : : : aik
and the above sum extends over all k-tuples .i1 ; : : : ; ik / of nonnegative integers
satisfying i1 C C ik D n.
4. In the notations of the discussion of Example 11.42, prove that
a1 a2 am mC1
C CC D :
d1 d2 dm 2
1 2x C 2x2 A B C
D C C :
.1 x/2 .1 2x/ .1 x/2 1x 1 2x
(d) Expand each of the functions of the right hand side above in power series
to conclude that an D 2nC1 .n C 1/ for n 0.
8. Let .an /n0 be given by a0 D 1, a1 D 3, a2 D 5 and akC3 D akC2 C 12 akC1
1
a for every integer k 0. The purpose of this problem is to show that .an /n0
2 k
converges and to compute the corresponding limit. To this end, do the following
items:
498 11 Series of Functions
X
B C .1/k1 C k
f .x/ D 9 p x:
k0 . 2/kC1
BC.1/n1 C n
(d) Conclude that an D 9 p for n 0, and hence that an ! 9.
. 2/nC1
x3 x6 x9
u.x/ D 1 C C C C
3Š 6Š 9Š
x4 x7 x10
v.x/ D x C C C C
4Š 7Š 10Š
x2 x5 x8 x11
w.x/ D C C C C
2Š 5Š 8Š 11Š
Prove that, for every real x, one has
10. The purpose of this problem is to give a proof of Theorem 11.48. To this end,
let
X X
p.x/ D bk .x x0 /k and q.x/ D ck .x x0 /k
k0 k0
X
k
.k C 2/.k C 1/akC2 C .j C 1/bkj ajC1 C ckj aj D 0
jD0
X
k
.k C 2/.k C 1/AkC2 .j C 1/jbkj jrkjC1 kj AjC1
jD0
X
k
C jckj jrkj kjC2 Aj :
jD0
(c) Let B; C > 0 such that jbk jrk B and jck jrk C for every k 0. If
M D maxfBr; Cg, show that
X
kC1
.k C 2/.k C 1/AkC2 M .j C 1/kC1j Aj :
jD0
X
kC1
.k C 2/.k C 1/AQ kC2 D M .j C 1/kC1j AQ j :
jD0
In this section, we show that a function defined by a power series is analytic, i.e., is
given by a convergent power series around each point of its interval of convergence.
We also study the set of zeros of an analytic function and use the results we get to
give purely analytic derivations of the properties of the sine and cosine functions.
Our purpose in doing so is to show the reader that all of the arguments we
have done so far concerning these functions are actually independent of any
500 11 Series of Functions
XX XX
ajk D ajk : (11.36)
j0 k0 k0 j0
X
n X
m X
m X
n X
m X XX
jajk j D jajk j jajk j jajk j < C1:
kD0 jD0 jD0 kD0 jD0 k0 j0 k0
X
n X XX
jajk j jajk j < C1;
kD0 j0 j0 k0
11.5 A Glimpse on Analytic Functions 501
P
for every n 2 N. Therefore, j0 jajk j < C1 for every k 0 and, letting n ! C1
in the above inequality, we get
XX XX
jajk j jajk j < C1:
k0 j0 j0 k0
P P
Thus, k0 j0 jajk j also converges. P
Now we can prove (a) and (b), except for (11.36): since j0 jajk j converges for
P ˇP ˇ
ˇ ˇ
every k 0, it is also true that j0 ajk converges for every k 0, with ˇ j0 ajk ˇ
P
j0 jajk j. Summing this inequality over k 0, we get
X ˇˇ X ˇˇ X X
ˇ ajk ˇ jajk j < C1:
k0 j0 k0 j0
P P
Therefore, k0 j0 ajk is absolutely convergent, hence convergent. P Arguing in
an
P P entirely analogous way, we conclude that both of the series k0 ajk and
a are (absolutely) convergent.
k0 j0 jk
ˇP P P P ˇ
ˇ ˇ
In order to prove (11.36), let D ˇ j0 k0 ajk k0 j0 ajk ˇ. Then, for
m; n 2 N, we have
ˇXm X XX X
n X X X ˇˇ
ˇ
Dˇ ajk C ajk ajk ajk ˇ
jD0 k0 j>m k0 kD0 j0 k>n j0
ˇXm X X
n X ˇ ˇXX ˇ ˇXX ˇ
ˇ ˇ ˇ ˇ ˇ ˇ
ˇ ajk ajk ˇ C ˇ ajk ˇ C ˇ ajk ˇ:
jD0 k0 kD0 j0 j>m k0 k>n j0
P P P P
Let > 0 be given. Since j0 k0 jajk j and k0 j0 jajk j converge, we
can choose m0 ; n0 2 N such that
XX XX
m m0 ) jajk j < and n n0 ) jajk j < :
j>m k0 k>n j0
ˇP P ˇ ˇP P ˇ
ˇ ˇ ˇ ˇ
In turn, this gives ˇ j>m k0 ajk ˇ < and ˇ k>n j0 ajk ˇ < , so that, with
m m0 and n n0 , we get in the above estimate for
502 11 Series of Functions
ˇXm X X
n X ˇ ˇXX
m XX
n ˇ
ˇ ˇ ˇ ˇ
ˇ ajk ajk ˇ C 2 D ˇ ajk ajk ˇ C 2
jD0 k0 kD0 j0 k0 jD0 j0 kD0
ˇXn X
m X
m X
n ˇ ˇXX m ˇ ˇXX n ˇ
ˇ ˇ ˇ ˇ ˇ ˇ
ˇ ajk ajk ˇ C ˇ ajk ˇ C ˇ ajk ˇ C 2
kD0 jD0 jD0 kD0 k>n jD0 j>m kD0
XX
m XX
n XX XX
jajk j C jajk j C 2 jajk j C jajk j C 2 < 4:
k>n jD0 j>m kD0 k>n j0 j>m k0
t
u
We are now in position to prove the following
Theorem 11.54 If f W .x0 R; x0 C R/ ! R is given by the power series f .x/ D
P
k0 ak .xx0 / , then f is analytic. More precisely, for a given y0 2 .x0 R; x0 CR/,
k
P
there exist b0 ; b1 ; b2 ; : : : 2 R such that f .x/ D k0 bk .x y0 /k in .y0 r; y0 C r/,
where r D R jy0 x0 j > 0.
Proof First of all, note that .y0 r; y0 C r/ .x0 R; x0 C R/. Now, for x 2
.y0 r; y0 C r/, we can write
X X k
f .x/ D ak .x x0 /k D ak .x y0 / C .y0 x0 /
k0 k0
!
X Xk
k
D ak .y0 x0 /kj .x y0 /j :
k0 jD0
j
If we can change the order of the sums in the last expression above, we will get
!
XX k
f .x/ D ak .y0 x0 /kj .x y0 /j :
j0 kj
j
P k
Letting bj D kj ak j .y0 x0 /kj , we will have
X
f .x/ D bj .x y0 /j ;
j0
P
which is finite, for jy0 x0 j C jx y0 j < jy0 x0 j C r D R and k0 ak uk converges
absolutely for u 2 .R; R/. t
u
We show next that the set of zeros of an analytic function f W I ! R do not
accumulate on I. To this end, we first need a formal definition of this concept.
Definition 11.55 Given X R and a 2 R, we say that a is an accumulation point
of X if there exists a sequence .an /n1 of pairwise distinct elements of X such that
n
an ! a.
Theorem 11.56 Let I R be an open interval, f W I ! R be an analytic function
and Z D fx 2 II f .x/ D 0g. The following conditions are equivalent:
(a) Z has no accumulation point in I.
(b) There exists x0 2 I such that f .k/ .x0 / D 0 for every integer k 0.
(c) f vanishes identically.
Proof (a) ) (b): let x0 2 I be an accumulation point of Z, say x0 D limn!C1 xn ,
where .xn /n1 is a sequence of pairwise distinct elements of Z. Since f .xn / D 0 for
every n 1, the continuity of f gives f .x0 / D 0.
By contradiction, assume that there exists an integer k0 1 such that
f .k0 / .x0 / ¤ 0. Without loss of generality, we can assume that k0 is minimum
with such a property. By the analyticity of f , we can choose R > 0 such that
.x0 R; x0 C R/ I and, in such an interval,
P f .k/ .x0 /
If g.x/ D kk0 kŠ .x x0 /
kk0
, then the series which defines g converges in
f .x/
.x0 R; x0 C R/ n fx0 g (for g.x/ D .xx k in such an interval). Since it obviously
0/ 0
converges at x0 , we conclude that g is continuous in .x0 R; x0 C R/, with g.x0 / D
f .k0 / .x0 /
k0 Š ¤ 0. By the sign preserving lemma for continuous functions, there exists
0 < ı R such that g ¤ 0 on .x0 ı; x0 C ı/. Since f .x/ D g.x/.x x0 /k0 in this
last interval, we conclude that f ¤ 0 in .x0 ı; x0 C ı/ n fx0 g. But this contradicts
the fact that xn 2 .x0 ı; x0 C ı/ n fx0 g for every sufficiently large n.
504 11 Series of Functions
(b) ) (c): let x0 2 I be such that f .k/ .x0 / D 0 for every integer k 0, and let
R be the radius of convergence of the power series defining f around x0 . Then, for
x 2 .x0 R; x0 C R/ we have
X f .k/ .x0 /
f .x/ D .x x0 /k D 0;
kk
kŠ
0
and cos0 D sin, then they are infinitely differentiable and ultimately must be given
by their Taylor series centered at 0, as in Example 11.4.
The key to free Trigonometry from Euclidean Geometry is to turn this around,
defining sin and cos by means of those power series. Therefore, we start by letting
X .1/j1 X .1/j
f .x/ D x2j1 and g.x/ D x2j (11.37)
j1
.2j 1/Š j0
.2j/Š
for every x 2 R.
The analytic foundations of Trigonometry are the content of the four coming
results.
Proposition 11.59 With notations as above, we have that:
(a) f and g are well defined and analytical, with f odd and g even.
(b) f .4k/ D f , f .4kC1/ D g, f .4kC2/ D f , f .4kC3/ D g and g.4k/ D g, g.4kC1/ D f ,
g.4kC2/ D g, g.4kC3/ D f , for every integer k 0.
(c) f and g are nonconstant and such that f .x/2 C g.x/2 D 1 for every x 2 R.
(d) f .x C y/ D f .x/g.y/ C f .y/g.x/ and f .x y/ D f .x/g.y/ f .y/g.x/ for every
x; y 2 R. In particular, f .2x/ D 2f .x/g.x/ for every x 2 R.
P
Proof (a) The convergence of k0 jxjkŠ , together with the comparison test for
k
series, assures the absolute convergence, in the whole real line, of both power
series in (11.37). Then, Theorem 11.54 shows that they are analytical functions.
Now, the oddness of f and the evenness of g follow from their very definitions.
(b) Theorem 11.27 gives
X .1/j1 X .1/i
f 0 .x/ D x2j2 D x2i D g.x/I
j1
.2j 2/Š i0
.2i/Š
Since f 0 .y/ D g.y/, the uniqueness part of the proof of Theorem 11.49 (which
did not depend on any properties of sin and cos) guarantees that h D l.
506 11 Series of Functions
Now, the first formula, together with the fact that f is odd and g is even, gives
f .x y/ D f .x/g.y/ C f .y/g.x/ D f .x/g.y/ f .y/g.x/:
Finally, the expression for f .2x/ also follows from the first formula, once we let
y D x. u
t
Proposition 11.60 The function f has a smallest positive zero.
Proof Start by observing that, with item (b) of the above proposition at our disposal,
the results of Problem 4, page 325 remain true for f and g in place of sin and cos,
respectively, so that
x3 x2
x f .x/ x and g.x/ 1
3Š 2Š
for every x 0. In turn, this gives
x3 x5 x2 x4
f .x/ x C and g.x/ 1 C
3Š 5Š 2Š 4Š
for every x 0. For instance, for x 0 one has
d x2 x4 x3
g.x/ 1 C D f .x/ C x 0;
dx 2Š 4Š 3Š
2 4
so that g.x/ 1 C x2Š x4Š g.0/ 1 D 0.
3 p
Now, the maximum of x x3Š for x 0 is attained at x D 2 (check it!).
Therefore, it follows from the above that
p 3 p
p p 2 2 2
f . 2/ 2 D :
3Š 3
x2 4 p
On the other hand, the minimum of 1 C x for x 0 is attained at x D 10
2
2Š 4Š
x4
(check it too!). Since f .x/ x 1 x2Š C 4Š for x 0, this gives
p 2 p 4 p
p p 10 10 10
f . 10/ 10 1 C D :
2Š 4Š 6
p
2 2
p
10
p p p
Since 3 > the IVT (applied to the intervals Œ0; 2 and Œ 2; 10
6 ,
p p p
assures the existence of 0 < ˛ < 2 < ˇ < 10 such that f .˛/ D f .ˇ/ D 610 .
Then, the MVT of Lagrange gives 2 .˛; ˇ/ such that f 0 . / D 0.
It follows from the last part of item (d) of the previous proposition that
f .2 / D 2f . /g. / D 2f . /f 0 . / D 0;
11.5 A Glimpse on Analytic Functions 507
so that f has a positive root. Since f doesn’t vanish identically, Corollary 11.58
shows that it has a smallest positive zero, as wished. t
u
From now on, we let denote the smallest positive zero of f . The last part of
item (d) of Proposition 11.59 gives
0 D f ./ D 2f
g ;
2 2
and since f is positive on .0; / this shows that g 2 D 0. Item (c) of Proposi-
tion 11.59 allows us to write
r 2
0<f D ˙ 1g D ˙1;
2 2
so that f 2 D 1. Then, item (d) of Proposition 11.59 gives
f x Df g.x/ f .x/g D g.x/: (11.38)
2 2 2
We are finally in position to go one step further.
Proposition 11.61 With notations as above, we have that:
(a) g W Œ0; ! Œ1; 1 is a decreasing bijection with continuous inverse.
(b) g1 W .1; 1/ ! .0; / is differentiable.
(c) Im.f / D Im.g/ D Œ1; 1.
(d) f .x C / D f .x/ and g.x C / D g.x/.
(e) f and g are periodic, with period 2.
Proof (a) Since g0 D f < 0 in .0; /, we conclude that g is decreasing in Œ0; .
Since
p
1 D g.0/ > g./ D ˙ 1 f ./2 D ˙1;
we conclude that g./ D 1. The rest follows from Theorem 8.35.
(b) This follows directly from Theorem 9.28. Indeed, since f ¤ 0 in .0; /, we get
for x 2 .0; / that g0 .x/ D f .x/ ¤ 0.
(c) Item (c) of Proposition 11.59 gives Im.g/ Œ1; 1. Then, item (a) above gives
Im.g/ D Œ1; 1, from where (11.38) gives Im.f / D Œ1; 1.
(d) An application of item (d) of Proposition 11.59, together with g./ D 1, gives
Therefore,
3
f Df C D f D 1
2 2 2
508 11 Series of Functions
and, hence,
3 r 3 2
g D ˙ 1f D 0:
2 2
In turn, (11.38) gives
g.x C / D f x D f x C
2 2
D f .x/g f g.x/ D g.x/:
2 2
(e) Firstly, two applications of (d) give
f .x C 2/ D f .x C / C D f .x C / D f .x/
f .p / D f .p/g./ f ./g.p/ D 0;
For the uniqueness part, assume, for the sake of contradiction, that there exist
0
1 <
2 < 2 such that .x0 ; y0 / D .g.
1 /; f .
1 // D .g.
2 /; f .
2 //. Then,
f .
2
1 / D f .
2 /g.
1 / f .
1 /g.
2 / D 0;
with 0 <
2
1 < 2. Since is the smallest positive zero of f and the next one
is 2, this gives
2
1 D . However, if this is so, then
x0 D g.
2 / D g.
1 C / D g.
1 / D x0
;
y0 D f .
2 / D f .
1 C / D f .
1 / D y0
X2k 2kC1
X .1/j1
.1/j1 2j1
x sin x x2j1
jD0
.2j 1/Š jD0
.2j 1/Š
and
2k1
X 2k
X .1/j
.1/j 2j
x cos x x2j :
jD0
.2j/Š jD0
.2j/Š
2. One can prove that if I; J R are open intervals and g W I ! J and f W J ! R are
analytic, then so is f ı g W I ! R. Use this fact to show that log W .0; C1/ ! R
is analytic and to find its Taylor series expansion around x0 > 0.
510 11 Series of Functions
9
After S. N. Bernstein, Russian mathematician of the XX century.
Bibliography
26. M.B.W. Tent, The Prince of Mathematics: Carl Friedrich Gauss (A.K. Peters Ltd, Wellesley,
2006)
27. R. Wheeden, A. Zygmund, Measure and Integral: An Introduction to Real Analysis
(Chapman & Hall, New York, 1977)
Appendix A
Glossary
Problems tagged with a country’s name refer to any round of the corresponding
national mathematical olympiad. For example a problem tagged “Brazil” means that
it appeared in some round of some edition of the Brazilian Mathematical Olympiad.
Problems proposed in other mathematical competitions, or which appeared in
mathematical journals, are tagged with a specific set of initials, as listed below:
1. APMO: Asian-Pacific Mathematical Olympiad.
2. Austrian-Polish: Austrian-Polish Mathematical Olympiad.
3. BMO: Balkan Mathematical Olympiad.
4. Berkeley: Berkeley Preliminary Examination.
5. Baltic Way: Baltic Way Mathematical Contest.
6. Crux: Crux Mathematicorum, a mathematical journal of the Canadian Mathe-
matical Society.
7. EKMC: Eötvös-Kürschák Mathematics Competition (Hungary).
8. IMO: International Mathematical Olympiad.
9. IMO shortlist: problem proposed to the IMO, though not used.
10. Israel-Hungary: Binational Mathematical Competition Israel-Hungary.
11. Miklós-Schweitzer: The Miklós-Schweitzer Mathematics Competition
(Hungary).
12. NMC: Nordic Mathematical Contest.
13. OCM: State of Ceará Mathematical Olympiad.
14. OCS: South Cone Mathematical Olympiad.
15. OBMU: Brazilian Mathematical Olympiad for University Students.
16. OIM: Iberoamerican Mathematical Olympiad.
17. OIM shortlist: problem proposed to the OIM, though not used.
18. OIMU: Iberoamerican Mathematical Olympiad for University Students.
19. Putnam: The William Lowell Mathematics Competition.
20. TT: The Tournament of the Towns.
Section 1.1
Section 1.2
1. For items (b) and (c), use (7’) and (a); item (g) follows from items (b) and (c).
2. with equality if and only if a D b D c D 0. Adapt the proof of Corollary 1.5.
3. For item (a), observe that .rs/n D .rs/ .rs/ D .„ƒ‚…
r r / .„ƒ‚…
s s / D rn sn . For
„ ƒ‚ …
n n n
the other items, apply analogous arguments.
4. Consider separately the cases m < n, m D n and m > n. In each case, make
judicious use of the property of item (b) of the last problem.
If b D 2k 5l , with k; l 2 ZC , and n D maxfk; lg, then ab D a2 10n5 .
nk nl
5.
6. If n D 2k, with k 2 N, write xn D .x2 /k and, then, apply item (g) of
Proposition 1.2. The case of an odd n can be treated similarly.
7. Start by showing that 12 13 C 14 15 > 15 .
Compute the difference aC1 bC1 bC2 .
aC2
8.
9. Letting S denote the sum of the ten given numbers, show that 4S can be written
as the sum of ten real numbers, each of which equals the sum of four of the ten
given numbers.
10. Argue as in Example 1.4, using the fact that 31 < 32 D 25 and 17 > 16D 24 .
n
11. For items (a) and (b), begin by observing that an < bn if and only if ab < 1;
then, apply the result of Corollary 1.3.
For item (c), begin by noticing that
a n b n
an C bn < .a C b/n if and only if aCb C aCb < 1; then, apply the result
of Corollary 1.3.
12. Use the fact that a3 < ca2 and b3 < cb 2
1 and, then,
n apply
Pythagoras’ theorem.
n n
13. The given inequality is equivalent to 7 C 47 C 97 2. This last one
holds trivially for n D 1 and n D 2; for n 3, apply item (b) of Corollary 1.3
n 3
to conclude that 97 97 > 2.
14. Start by observing that, if a 4, then 0 < 1a C 1b C 1c 14 C 14 C 14 < 1, so that
1
a
C 1b C 1c cannot be an integer.
15. If a > 5 is an integer, show that 4.a 4/ > a; from this, conclude that it is not
worth to have summands greater that 5. Then, by performing similar changes
of summands, discard all of those which are equal to 4 or 5. Finally, show that
is more advantageous to have more summands equal to 3 than summands equal
to 2.
16. If we let a be the common leftmost digit, show that there exist nonnegative
integers k and l such that a10k < 2n < .aC1/10k and a10l < 5n < .aC1/10l ;
then, multiply these inequalities.
17. Suppose that all three inequalities are true; the second one can be written as
.a d/.c b/ < 0, so that the first one gives a d < 0 < c b. Now, use the
third inequality to get ad.c b/ < bc.a d/ and, then, arrive at a contradiction.
B Hints and Solutions 517
Section 1.3
p p p
2. For item (a), for instance, show that . n xy/n D xy and . n x n y/n D xy; then, use
the definition of the n-th root.
3. By contraposition,1 show that, if b ¤ 0, then r is rational.
4. Reduce this problem to the previous one.
5. Argue by contraposition. p p p
6. If ab ¤ 0, compute .a C b 2/2 D .c 3/2 to conclude that 2 should be
rational, which is an absurd. Therefore, ab D 0 and, thus, a D 0 or b D 0. Then,
apply the result of Problem 3. p
7. By the sake of contradiction, write 2 D ab , with a and b relatively prime natural
numbers, so that 2b2 D a2 . Then, successively show that a is even and b is even,
thus reaching a contradiction.
8. Adapt, to the present case, the hint given to the previous problem.
Section 2.1
1. For item (a), it suffices to develop the right hand side to get
Finally, item (c) is also obtained by expanding the right hand side, and we leave
this to the reader.
1
The standard ways to prove a proposition of the form A ) B (i.e., If A, then B) are either directly,
by contraposition or by contradiction. In the first case, we assume the validity of assertion A and
deduce the validity of assertion B directly; in the second case, we assume that assertion B is false
and deduce, directly, that assertion A is also false; finally, in the third case, we assume that assertion
A is true and assertion B is false and, from this, directly deduce a contradiction (i.e., deduce that an
assertion that is obviously false should be true, which is something impossible to happen). For a
detailed discussion on the fundamentals of Logic and methods of proof, we refer the reader to [21].
518 B Hints and Solutions
m n p m2 C n2 C p2
C C D
np mp mn mnp
.m C n C p/2 2.mn C mp C np/
D :
mnp
3. The statement of the problem, together with item (a) of Proposition 2.1, gives
us
2
b 1b 2
D , .b.1 a//2 .a.1 b//2 D 0
a 1a
, .b.1 a/ a.1 b//.b.1 a/ C a.1 b// D 0
, .b a/.a C b 2ab/ D 0:
1 1 aCb 2ab
C D D D 2:
a b ab ab
4. Successively applying items (b) and (a) of Proposition 2.1, we get
2
2
1 xy 1 yx
p p p D p p
. x y/2 C 2 xy .x 2 xy C y/ C 2 xy
x2 y2 .x y/.x C y/ xy
D 2
D 2
D 2 :
x .x C y/ x .x C y/ x
.x3 C y3 C z3 /2 .x3 y3 z3 /2
D
yCz
Œ.x3 C y3 C z3 / .x3 y3 z3 /Œ.x3 C y3 C z3 / C .x3 y3 z3 /
D
yCz
2.y3 C z3 / 2x3
D
yCz
2.y C z/.y2 yz C z2 / 2x3
D
yCz
D 4x3 .y2 yz C z2 /:
B Hints and Solutions 519
1 1
6. Since ab D 1 , a Db, b D a, we have
a 1a b C 1b .a b/ .b C a/ a2 b2
D D D 1;
a2 b2 a2 b2 a2 b2
where we have used item (a) of Proposition 2.1 in the next to last equality above.
7. We have .y x/.y C x/ D 192 , with y x and y C x integers such that
0 < y x < y C x. Therefore, the only possible choice is to have y x D 1 and
y C x D 361, so that y D 181 and x D 180.
8. A judicious application of item (b) of Proposition 2.1 gives
13.ab/2 a6 Cb6
so that a6 Cb6
D a6 Cb6
D 1.
10. Start by observing that
p p
As for (b), we apply item (c) of Proposition 2.1, with 3 x and 3 y in place of
p p3 p
x and y, respectively, and observe that . 3 x/2 D x2 and . 3 x/3 D x (and
analogously for y). This way, we get
p
3 2 p p
1 x 3 xy C 3 y2
p p D p p p p p
3
x˙ 3 y 3
. 3 x ˙ 3 y/. x2 3 xy C 3 y2 /
p3 2 p p
x 3 xy C 3 y2
D :
xy
and, hence,
p p p
1 2 3 3C 3 9
2
p p D p
2C 3 3 3C2 2
p p p p
2 2 3 3C 3 9 32 2
D p p
3C2 2 32 2
p p p 3
p
3
D .3 2 2/.2 2 3 C 9/:
x2 y2 z2 x2 y2 z2
C C D C C D 3:
.y C z/2 .x C z/2 .x C y/2 .x/2 .y/2 .z/2
x3 y3 z3
Analogously, .yCz/3
C .xCz/3
C .xCy/3
D 3.
17. Applying item (a) of Proposition 2.1 several times, we get
With respect to (b), observe that the changes of sign make sense precisely
because n is odd:
19. More generally, let us factorise x4n C 4y4n , where n 2 N, inspiring ourselves in
the formula for .a C b/2 :
6 j .a C b C c/ , 6 j .a C b C c/3
, 6 j Œa3 C b3 C c3 C 3.a C b/.a C c/.b C c/
, 6 j .a3 C b3 C c3 /:
21. We present two different solutions. For the first one, just note that
a C b C c D 0 ) .a C b/3 D .c/3
) a3 C b3 C 3ab.a C b/ D c3
) a3 C b3 C 3ab.c/ D c3
) a3 C b3 C c3 3abc D 0:
a C b C c D 0 ) .a C b C c/3 D 0
) a3 C b3 C c3 C 3.a C b/.a C c/.b C c/ D 0:
p ! p !
p aC a2 b a a2 b
a˙ b D C
2 2
v ! !
u p p
u a C a2 b a a 2b
˙ 2t
2 2
s p
a2 . a2 b/2
D a˙2 :
4
1 1 1
xy C xz C yz D C C
.b c/.c a/ .b c/.a b/ .c a/.a b/
1
D Œ.a b/ C .c a/ C .b c/ D 0:
.b c/.c a/.a b/
and, hence,
p p
p
3 . 3 a C 3 b/3 .a C b/
ab D p p 2 Q:
3. 3 a C 3 b/
524 B Hints and Solutions
p p
3 ab ab
3
a bD p p3
p
3
D p p p 2 Q:
2
. a/ C ab C . b/
3 2 . a C 3 b/2 3 ab
3
Finally,
p 1 p p
3 p p
3
3
aD Œ. 3 a C b/ C . 3 a b/ 2 Q
2
p
and, analogously, 3 b 2 Q.
26. Raise both members of the equality a C c C d D .b C e C f / to the third power
and apply the formula of Example 2.7 to get
Finally, multiplying the two relations thus obtained and cancelling out the
common factor .c C d/.e C f /, we arrive at the desired relation.
27. Start by observing that
Section 2.2
1. For the first inequality, interpret jx aj as the distance from x to a in the real line
(alternatively, apply the definition of modulus, separately considering the cases
x a 0 and x a < 0). For the other three inequalities, adapt the hint given to
the analysis of the first one.
2. Separately analyse the cases (i) x; y 0, (ii) x 0 > y, (iii) y 0 > x and (iv)
x; y < 0, showing that equality occurs in all of them.
3. For item (a), consider the cases x 0 and x < 0. For item (b), argue as in
Example 2.9.
4. Separately analyse cases x < 0, 0 < x < 1 and x > 1. The solution set is
.1; 0/ [ .1; C1/.
5. Since jyj y and j yj D jyj for all real y, we have jx aj x a and
jx bj D jb xj b x. Therefore, in order for the equation to have a real root
x, we must necessarily have
c D jx aj C jx bj .x a/ C .b x/ D b a:
For the second part, show that every x 2 .n; n C 1/ is a root of the equation.
Section 2.3
2
p of variables y D x C 18x transform
for item (c), we note that the substitution
the given equação into y C 30 D 2 y C 45. Squaring both sides, we easily
arrive at y2 C 56y C 720 D 0, an equation whose roots are p y D 36 and
y D 20. From these, only y D 20 serves us, for y C 30 D 2 y C 45 gives
y C 30 0 and ypC 45 0, i.e., y 30. Therefore, x2 C 18x C 20 D 0 and,
then, x D 9 ˙ 61. p
3. Start by observing that, in order for 2x 1 to have a meaning in the set of
reals, we must have x 12 . By a similar reason, it must be the case that x
p
2x 1. However, since
p
x 2x 1 , x2 2x 1 , x2 2x C 1 0;
which is always true, the conditions for the existence of the square roots in the
given equation reduce to x 12 . Now,
q q 2
2
p p
A D x C 2x 1 C x 2x 1
q p p
D 2x C 2 .x C 2x 1/.x 2x 1/
p
D 2x C 2 x2 .2x 1/ D 2x C 2jx 1j:
We shall consider here only item (b), leaving the other two items as exercises to
the reader. If A D 1, we have to solve the equation 2x C 2jx 1j D 1, under the
condition x 12 . If x 1, that equation reduces to 2x C 2.x 1/ D 1, so that
x D 34 (which does not satisfy the condition x 1). If 12 x < 1, the equation
reduces to 2x C 2.1 x/ D 1, an equality which is false for every x 2 12 ; 1 .
Therefore, there are no solutions in item (b).
4. Let ax2 C bx C c D 0 be the equation of the first quiz, a0 x2 C bx C c D 0 be
that of the second one and ax2 C bx C c0 D 0 be that of the third one. Since
the roots of the second quiz are 2 and 3, we have ab0 D 5 and ac0 D 6, so
that bc D 56 ; on the other hand, since the roots of the third quiz are 2 and 7,
we have ba D 5. Therefore, the second degree equation of the first quiz
has the form ax2 C 5ax 6a D 0, and, thus, has the same roots as those of
x2 C 5x 6 D 0, i.e., 6 and 1.
5. It follows from Proposition 2.16 that a C b D a and ab D b. The second
equality gives a D 1, and substituting this value into the first equality we get
b D 2.
6. Again by Proposition 2.16 (applied to both equations), we have a D ˛2 C
ˇ 2 D .˛ C ˇ/2 2˛ˇ D 13 2 9 D 5 and b D ˛ 2 ˇ 2 D .˛ˇ/2 D 92 D 81.
Therefore, a C b D 86.
528 B Hints and Solutions
S D u3 C v 3 D .u C v/.u2 uv C v 2 /
D Œ.u C v/2 3uv
D Œ.1/2 3.1/ D 4
Now, it follows from (a) that the equalities in the statement of the problem can
be rewritten as
to get
x3 C ax2 C bx C c D .x ˛/.x2 C b0 x C c0 /
to get the equalities b0 ˛ D a, c0 ˛b0 D b and ˛c0 D c. Now, use the first
and third relations to get b0 D a C ˛ and c0 D ˛c , showing, then, that these
values for b0 and c0 also satisfy the relation c0 ˛b0 D b (at this step you shall
p
need to use the fact that ˛ 3 Cpa˛ 2 C b˛ Cpc D 0.p
3 3
18. For item (a), let a D 2 C 5 and b D 2 5, so that ˛ D a C b. Then,
use the fact that
˛ 3 D a3 C b3 C 3ab.a C b/ D a3 C b3 C 3ab˛;
530 B Hints and Solutions
together with the relation ab D 1. For (b), use the result of the previous
problem to show that x3 C 3x 4 D .x 1/.x2 C x C 4/, so that x D 1 is the
only real root of x3 C 3x 4 D 0.
19. Use the result of Problem 21 to compare the coefficients of a3 x3 C a2 x2 C a1 x C
a0 D 0 and a3 .x x1 /.x x1 /.x x3 /.
20. In order to compute the value of the first sum of powers, use relations (2.21),
together with the identity ˛ 2 C ˇ 2 C 2 D .˛ C ˇ C /2 2.˛ˇ C ˛ C ˇ /;
for the other two sums of powers, start by observing that, since ˛ is a root of
the equation, we have ˛ 3 D 3˛ 1, and analogously for ˇ and . Then add the
left and right hand sides of these three identities.
21. Expand the products in .y C d/3 C a.y C d/2 C b.y C d/ C c and impose that
the coefficient of y2 equals 0.
23. It suffices to see that
3 1 1 2 1
x C 3 D xC x 1C 2
x x x
!
1 1 2
D xC xC 3 :
x x
with the second equation, if necessary, we can assume from the beginning that
a > 0 > b. Then, letting c d be the roots of x2 C ax C b D 0, we get
c C d D a < 0 and cd D b < 0, so that c > 0 > d. For (b), we have
D m2 4n and 0 D p2 4q. Compute
p !2 p !
0 m C m
D 4
2 2
1 2 p p
D m 2m C C 8m C 8
4
and notice that, since m > 0,
p p
0 < , m2 2m 4 C 8 3
p p p
, <m< 3 8 ,m< 3 8
, .m C 8/2 < 9.m2 4n/
, 2.m 1/2 > 9.n C 2/:
For (c), note that the process cannot continue indefinitely if the inequality 0 <
always holds. Then, show that at some point we reach a situation in which
n D 1 or m D 1, n D 2. If n D 1, conclude that the roots of x2 Cmx1 D 0
should be 1 and 1, so that m should be equal to 0, which is not the case. Then,
we are left to m D 1 and n D 2. In this case, argue backwards to show that
a D 1, b D 2.
Section 2.4
Sections 2.5
Hence, there are two possibilities: .z; x/ D .a; 1/ or .z; x/ D .1; a/. Those give
rise to the systems
x C y D a1 xCyD 1
and ;
xD1 x D a1
Then, apply the result of Lemma 2.14 to the second degree trinomial ax2 C
.b 1/x C c and analyse separately itens (a) and (b).
5. Suppose, without loss of generality, that x y. Then, use the equations of the
system to conclude that x D y D z. Finally, note that x D 1 is a root of the third
degree polynomial equation x3 2xC1 D 0, and apply the result of Problem 21,
page 43.
6. Write the left hand side as a sum of squares and apply Lemma 2.31, with n D 3.
7. Substituting y D 14 .a z 3x/ into the first equation, we get
1
x2 C .a z 3x/2 D 4z
16
or, which is the same,
For this last equation to have a single solution, the discriminant of the second
degree trinomial (in x) of the left hand side must be equal to 0, i.e., we should
have
or, which is the same, .a z/2 D 100z. If this is so, then (from the second
3
degree equation in x above) x D 25 .a z/ and (from the second equation of the
4
system) y D 25 .a z/. Therefore, the system has a single solution if and only if
the equation (in z) .a z/2 D 100z has a single solution. Since it is equivalent
to z2 2z.a C 50/ C a2 D 0, its discriminant must also be equal to 0, i.e., we
should have .a C 50/2 D a2 . Thus, a D 25.
8. Start by writing
3 p 3 p
xC 2 x C 2 D .x 1/ C 1 C 2 xC2
x1 x1
xC2 p
D .x 1/ C 2 xC2
x1
r !2
p xC2
D x1
x1
Then, apply Lemma 2.31 to conclude that it suffices to solve the system of
equations
r s r
p xC2 p yC2 p zC2
x1 D y1 D z1 D 0;
x1 y1 z1
p
so that x D y D z D 3C2 13 .
p p
9. Letting a D 3 x C 5 and b D 3 4 x, we get a C b D 3 and a3 C b3 D 9.
Hence
1
.x C 1/2 C D y2 2;
.x C 1/2
1
xC1C D 3 or 1;
xC1
1 1 C 4x2 1 u2
vD D 2
D 2 C1D C 1:
y 4x 4x 4
Transforming the other two equations in analogous ways, we get the system
below, which is equivalent to the original one:
13. Add the equations of the system to conclude that it suffices to solve the system
of equations
8
ˆ
ˆ a C b C c D 3d
<
b C c C d D 3a
:
ˆ c C d C a D 3b
:̂
d C a C b D 3c
4u2 u4 D 4 .4 4u2 C u4 / D 4 .2 u2 /2 :
Hence,
p p p
6 D 4x2 x4 3 C 4y2 y4 C 4z2 z4 C 5
p p p
D 1 .2 x2 /2 C 4 .2 y2 /2 C 9 .2 x2 /2
p p p
1 C 4 C 9 D 1 C 2 C 3;
and it follows
p from thepprevious problem
p that 2 x2 D 2 y2 D 2 z2 D 0,
i.e., x D ˙ 2, y D ˙ 2, z D ˙ 2.
16. If x 4, write 5x 4 4
4 x D xC 4 x to conclude that z 4; then, argue similarly
x
4 4 4 x y z
C C D C C ;
x y z 4 4 4
2 p
2y D x C 2 2;
x
536 B Hints and Solutions
p p
so that y 2; analogously, x; z 2. Adding the three equations of the
system, we get
2 2 2
xCyCzD C C :
x y z
2
p
Finally, note that u u
for u 2, and use the result of Problem 14.
Section 3.1
2. For item (a), let an D n, for odd n 1, and an D .1/n1 n, for even n 1.
For (b), let an D nC1 n
, for n 1. Finally, for item (c), let an D n, for odd n 1,
1
and an D n for even n 1.
3. It suffices to note that bkC1 D akC1 1
D 3aak C1
k
D 3 C a1k D 3 C bk .
4. For item (a), we can write a1 D a2 D a3 D 1 and ak D ak1 C ak2 C ak3 , for
k 4. For item (b), we can write a1 D 1 and akC1 D 2ak , for k 1.
5. Observe that the entries in the n-th line form and AP of initial term n, common
difference 2 and 2n 1 terms.
6. Write 11 : : :…
„ ƒ‚ 1 as a sum of powers of 10, then use the formula for the sum of the
n
terms of a GP.
7. Use the result of the previous problem.
8. It suffices to show that bkC1 bk does not depend on k; the common value we
shall obtain will be the common difference of the AP .bk /k1 . By definition, we
have
11. The formula for the general term of AP’s gives us the relations ˛ D ap D
a1 C .p 1/r, ˇ D aq D a1 C .q 1/r and apCq D a1 C .p C q 1/r, where r
is the common difference of the AP. By viewing the first two relations above as
a linear system of equations in a1 and r, express them in terms of ˛, ˇ, p and q;
then, substitute these expressions for a1 and r into the formula for apCq .
B Hints and Solutions 537
12. Letting 2n 1 denote the smallest of such integers and k their quantity, the
formula for the general term of an AP assures that the biggest of the numbers
equals .2n 1/ C 2.k 1/ D 2n C 2k 3. Therefore, their sum is equal to
1 3 3
2 Œ.2n 1/ C .2n C 2k 3/k D 7 , so that .2n C k 2/k D 7 . Now, since
2
2n C k 2 k > 1, the only possibility is 2n C k 2 D 7 and k D 7, so that
n D 22.
13. Let r be the common difference of the AP and let a2 D m, so that r 2 N and
m > 1. Then, amkC2 D a2 C ..mk C 2/ 2/r D m C mkr D m.1 C kr/, which
is clearly a composite number.
14. Item (b) follows immediately from (a). For item (a), separately consider the
cases n even and n odd. p p p
15. By contradiction, suppose that am D 2, an D 3 and ap D 5, where
.ak /k1 is an AP and m, n and p are pairwise distinct natural numbers. Then,
use the formula for the general term of AP’s to compute nmpm .
16. Apply, to the left hand side, the formula for the general term of GP’s.
17. Adapt, to the present case, the solution given to the Example 3.14.
18. Adapt, to the present case, the solution given to the Example 3.14.
19. Suppose that there exists such an AP .ak /k1 , of common difference r, i.e.,
such that ar1 … Q but a2n D am ap , for some distinct indices m, n and p. Use the
formula for the general term of an AP to reach a contradiction.
20. Show that the sequence .bk /k1 , such that bk D ak 1, is a GP.
22. Suppose that .ak /k1 is a GP of common ratio q, such that am D 2, an D 3 and
ap D 5. Then, a1 qm1 D 2, a1 qn1 D 3 and a1 qp1 D 5, so that qmn D 23 and
qmp D 25 . Therefore,
mp mn
2 .mn/.mp/ 2
Dq D
3 5
or, which is the same, 2np 5mn D 3mp . Now, get a contradiction.
Section 3.2
For item (b), it follows from (a) that 2bnC2 bnC1pD 3 D 2bnC1 bn and, thus,
2bnC2 3bnC1 C bn D 0. Since b1 D 5 and b2 D 1 C 24a2 D 4, Theorem 3.16
guarantees that bn D 3 C 24n . At last, for item (c), it follows from items (a) and
(b) that
!
1 2 1 4 2
an D .b 1/ D 3C n 1
24 n 24 2
1 4 4
D 2C n 4C n
24 2 2
1 1 1
D 1 C n1 1C n :
3 2 2
Sections 3.3
1. If .ak /k1 is a second order AP, then .akC1 ak /k1 is a non constant AP, so
that there exists r ¤ 0 for which .akC2 akC1 / .akC1 ak / D r, for every
k 1. In particular, for every k 1, we have akC2 2akC1 C ak ¤ 0 and
8. For item (a), it suffices to observe that .2k C 1/3 .2k 1/3 D 24k2 C 2 D
16k2 C 8k2 C 2 D 16k2 C .2k 1/2 C .2k C 1/2 . For item (b), start by using
the formula for telescoping sums to write
X
n
.2n C 1/3 2 D Œ.2k C 1/3 .2k 1/3 C .33 2/
kD2
X
n
D Œ.4k/2 C .2k C 1/2 C .2k 1/2 C 25
kD2
X
n
D Œ.4k/2 C .2k C 1/2 C .2k 1/2 C 42 C 32 :
kD2
k2 C k C 1 1 1 1
D1C D1C
k.k C 1/ k.k C 1/ k kC1
and use the formula for telescoping sums or the result of Problem 4.
Fk
D F1k FkC2
1
F F
11. Write Fk FkC1
kC2
D FkC2
k FkC2
and, then, use the formula for telescoping
sums.
12. Rationalise the fraction pak C1pakC1 —see Problem 12, page 25—and, then, use
the formula for telescoping sums.
13. For item (a), just note that x4 C x2 C 1 D .x4 C 2x2 C 1/ x2 D .x2 C 1/2 x2 D
.x2 C 1 x/.x2 C 1 C x/. For (b), use the result of (a) to write
k k
D 2
k4 2
Ck C1 .k C k C 1/.k2 k C 1/
1 1 1
D :
2 k2 k C 1 k2 C k C 1
15. Adapt, to the present case, the hint given to the previous problem.
16. Initially, observe that
Yn Yn n
1 1 Y 1
1 2 D 1 1C
jD2
j jD2
j jD2 j
n
Y n
j1 Y jC1
D :
jD2
j jD2
j
Yn Yn n
1 aj1 Y ajC1
1 2 D
jD2
j jD2
aj jD2
aj
101
X x3k
SD :
kD0
.1 xk /3 C x3k
Since 1 xk D 1 k
101
D x101k , we get
101
X x3k
SD 3
x
kD0 101k
C x3k
Hence, S D 51.
18. StartQby factorising both the numerator and denominator of the fraction under
the notation. If ak D k2 k C 1, show that akC1 D k2 C k C 1 and, then, use
the formula for telescoping products.
B Hints and Solutions 541
19. We first show that n must be even. To this end, suppose that A D A1 [ [ An ,
with A1 , . . . , An satisfying the conditions of the problem. For 1 k n, let
xk 2 Ak be the Pelement of Ak that equals the arithmetic mean of the other three.
Then 4xk D x2Ak x, and it follows that
X
n X
n X X
4 xk D xD x D 1 C 2 C C 4n D 2n.4n C 1/:
kD1 kD1 x2Ak x2A
Hence, n.4n C 1/ must be even and, since 4n C 1 is odd, it follows that n must
be even. Now, let n D 2k, with k 2 N, and write
where
X
10000
p p 10000
X 1 X p
10000
p
kC1 k < p < k k1 :
kD2 kD2
k kD2
However,
X
10000
p p p p
k C 1 k D 2 10001 2 2
kD2
and
X
10000
p p p
k k1 D2 10000 1 D 198:
kD2
542 B Hints and Solutions
p p
Hence, if we show that 2 10001 2 2 > 197, it will come that the greatest
integer which is less than or equal to S equals 198. To what is left to do, just see
that
p p p 2 p 2
2 10001 > 2 2 C 197 , 2 10001 > 2 2 C 197
p
, 40004 > 788 2 C 38817
p
, 788 2 < 1187 , 7882 2 < 11872
, 1241888 < 1408969;
X
pC1
.k C 1/ pC1
k pC1
D .k C 1/pC1j kj1
jD1
X
pC1
< .k C 1/pC1j .k C 1/j1
jD1
D .p C 1/.k C 1/p :
Analogously,
X
pC1
.k C 1/ pC1
k pC1
D .k C 1/pC1j kj1
jD1
X
pC1
> kpC1j kj1 D .p C 1/kp :
jD1
For item (b), it follows from (a) and from the formula for telescoping sums that
X
n1 X
n1
.k C 1/pC1 kpC1 npC1 1 npC1
k <
p
D <
kD1 kD1
pC1 pC1 pC1
and
X
n X
n
kpC1 .k 1/pC1 npC1
kp > D :
kD1 kD1
pC1 pC1
B Hints and Solutions 543
Section 4.1
X
k 2 2
k.k C 1/ .k C 1/.k C 2/
j3 D C .k C 1/3 D :
jD1
2 2
1
.k C 1/.h.k C 1/ h.k// D .k C 1/ D 1:
kC1
X
k
1
j.j C 1/ D .k 1/k.k C 1/ C k.k C 1/
jD1
3
1
D k.k C 1/.k C 2/:
3
6. It follows from the induction hypothesis that
X
kC1
1
.2j 1/2 D .2k 1/2k.2k C 1/ C .2k C 1/2
jD1
6
1
D .2k C 1/2.k C 1/.2k C 3/:
6
544 B Hints and Solutions
D 3k qŒ..43 /2 1/ C .43
k1 k1
1/ C 3
3k1 3k1
D 3k qŒ.4 1/.4 C 1/ C 3k q C 3
D 3k qŒ3k q.43
k1
C 1/ C 3k q C 3
n n : : : nk C n1 n2 : : : nk C C n1 n2 : : : nk D n1 n2 : : : nk nkC1 :
„1 2 ƒ‚ …
nkC1 times
12. By induction hypothesis, suppose that any set with k elements has exactly 2k
subsets. If A D fa1 ; a2 ; : : : ; ak ; akC1 g is a set with k C 1 elements. It has two
distinct kinds of subsets: those which are subsets of B D fa1 ; a2 ; : : : ; ak g and
those which contain akC1 . By the induction hypothesis, A has exactly 2k subsets
of the first type. On the other hand, the subsets of the second type are of the form
B0 [ fakC1g, where B0 is a subset of B. Again by induction hypothesis, there are
also 2k subsets of this second type. Hence, A has 2k C 2k D 2kC1 subsets.
13. To the first case, by using the induction hypothesis (strong induction!), we get
ukC2 D rukC1 suk D rakC1 sak D akC2 . The second case is completely
analogous.
B Hints and Solutions 545
k X 1 k X 1 k1
1 k1 1
D D C D C
a1 akC1 aa
jD1 j jC1
aa
jD1 j jC1
ak akC1 a1 ak ak akC1
For (b), it follows from the induction hypothesis and the result of (a) that
X
kC1
1 Xk
1 1 1 1
D C D2 C
jD1
a j jD1
a j a kC1 a 1 a 2 : : : a k a kC1
akC1 a1 a2 : : : ak 1
D2 D2 :
a1 a2 : : : ak akC1 a1 a2 : : : ak akC1
2 k
show that k2k .k C 1/kC1 or, which is the same, kC1 k
k C 1. Such an
inequality is true for every integer k 3, for we have
k k
k2 1
D k1C > .k 1/k .k 1/3 k C 1:
kC1 kC1
17. For n 2 Z, let an D xn C xn . We first show that an 2 Z for every n 2 N. The
hypothesis of the problem gives a1 2 Z, so that
1 1 2
a 2 D x2 C D x C 2 D a21 2 2 Z:
x2 x
1
akC1 D xkC1 C kC1
x
1 1 1
D x C k
k
xC x k1
C k1
x x x
D ak a1 ak1 2 Z;
and the induction step is complete. To what was left to show, it suffices to see
that a0 D 2 and, if n < 0 is an integer, then an D an 2 Z.
18. It follows from induction hypothesis that
xnC1 Cxn1
19. It suffices to prove that the quotient xn
does not depend on n. This is
immediate by induction, for
Items (c), (d) and (e) can be dealt with in an analogous way.
21. We make induction on n 1. For n D 1, the relation in the statement of the
problem is simply the recursive definition of FmC1 . In general, it follows from
the induction hypothesis that
22. In both items, make induction on n, for a fixed m 1. For the initial cases
2FmC1 D Fm C Lm and 2LmC1 D 5Fm C Lm , you will also have to make
inductive arguments.
B Hints and Solutions 547
that
13 C 23 C C k3 C a3kC1 D .1 C 2 C C k C akC1 /2 :
Performing the computations at the right hand side, we arrive at a3kC1 D a2kC1 C
k.k C1/akC1 or, which is the same, a2kC1 akC1 k.k C1/ D 0. Since akC1 > 0,
it immediately
p follows that akC1 D k C 1.
26. If xk > a, then
p 1 p a 1 p
xkC1 a D xk 2 a C D .xk a/2 > 0:
2 xk 2xk
p
To the other inequality, we start by observing that, if x y > a, then x C ax
y C ay (*); in fact,
548 B Hints and Solutions
a a 1 1 .x y/
xC yC D .x y/ C a D .xy a/ 0;
x y x y xy
p 2 p 1
for xy a D a. Now, if xk a C ,
then, from (*), we get
2k1
!
1 a 1 p 1 a
xkC1 D xk C a C k1 C p 1
2 xk 2 2 a C 2k1
1 p 1 a p 1
< a C k1 C p D aC k:
2 2 a 2
27. If k 6 and a1 , a2 , . . . , ak are positive integers such that
1 1 1
2
C 2 C C 2 D 1;
a1 a2 ak
then
1 1 1 1 1 1 3 1
C C C 2
C 2
CC 2
D C D 1:
4 4 4 .2a1 / .2a2 / .2ak / 4 4
Hence, by a simple variant of the strong form of the principle of mathematical
induction, it suffices to consider the cases k D 6, k D 7 and k D 8. Before we
look at them, note that
1 1 1 1 1 1 1 1
C C D and C C C D 1:
9 9 36 4 4 4 4 4
Therefore, for k D 6, we have 14 C 14 C 14 C 19 C 19 C 36 1
D 1; for k D
1 1 1 1 1 1 1
7, we have 16 C 16 C 16 C 16 C 9 C 9 C 36 D 1; for k D 8, we have
1
4
C 14 C 19 C 19 C 36
1
C 19 C 19 C 36
1
D 1.
28. Argue as in the proof of Example 4.12.
29. For the existence of a representation for n, if .k C 1/Š n < .k C 2/Š, let akC1
be the greatest natural number such that akC1 .k C 1/Š n < .akC1 C 1/.k C 1/Š.
Then, the inequalities akC1 .k C1/Š n < .k C2/Š D .k C2/ .k C1/Š guarantee
that akC1 k C 1; on the other hand, since
(a) m D FnC1 : there are two subcases. First, suppose that FnC1 D Ft1 C Ft2 C
C Ftr , with r 1, 1 < t1 < t2 < < tr < n and tjC1 tj > 1 for
1 j < r. Then,
Section 4.2
2n1
1. It suffices to note that 2n 2n 2n1
n D n n1 D 2 n1 .
2. Adapt, to the present
n case, the idea of the solution
of Example 4.18.
3. Compare nk to kC1 using the definition of nj .
4. Make a proof by induction on n.
5. Apply Stiefel’s relation to the numerator of the fraction and, then, use telescoping
sums.
6. Make a proof by induction, applying Stiefel’s relation in the induction step.
7. Let an be defined by the expression at the right hand side of the equality we wish
to establish. Show that a1 D 1, a2 D 1 and (using Stiefel’s relation) anC2 D
anC1 C an , for every integer n 1.
8. Make induction on n; more precisely, show that, for each n 2 N, the set
f1; 2; : : : ; ng has nk subsets of k elements each, for every 0 k n. For the
induction step, note that each of the subsets of k elements of f1; 2; : : : ; n C 1g is
of one of two possible types: those not containing n C 1 – thus, being subsets of k
elements of f1; 2; : : : ; ng – and those containing n C 1 – thus, being equal to sets
of the type A [ fn C 1g, where A is a subset of k 1 elements of f1; 2; : : : ; ng.
Section 4.3
1 65 1
Analogously, 65 k > , k 16, so that the maximal term is
65 1 k 3 kC1 3kC1
16 316 .
6. The given condition is equivalent to a2 C b2 C ab D c2 C d2 C cd. On the other
hand, it follows from the binomial expansion that a4 Cb4 C.aCb/4 D c4 Cd 4 C
.cCd/4 if and only if a4 Cb4 C2a3 bC3a2 b2 C2ab3 D c4 Cd 4 C2c3 d C3c2 d2 C
2cd3 . Finally, observe that a4 C b4 C 2a3 b C 3a2 b2 C 2ab3 D .a2 C b2 C ab/2
and, analogously,c4 C d4 C 2c3 d C 3c2 d2 C 2cd3 D .c2 C d2 C cd/2 .
7. Use the result of Problem 3, page 103—with 2n in place of n—together with
the theorem on the lines of the Pascal triangle.
8. Adapt, to the present case, the idea of the solution of Example 4.27.
9. It suffices to see that
! !
Xn
n j X
n
n 1 j1
ja D an a D an.1 C a/n1
jD0
j jD1
j 1
and
! ! ! !
Xn
n 2 Xn
n1 X n
n1 Xn
n1
j D n jD n .j 1/ C n
jD0
j jD1
j1 jD1
j1 jD1
j1
! !
Xn
n2 Xn
n1
D n.n 1/ Cn
jD2
j 2 jD1
j1
10. Use the formula for the general term of an AP, together with the result of the
previous problem.
11. Write 1q D 1 C a, with a > 0, and use the fact that .1 C a/n 1 C na.
p p p
12. Making ˛ D a C n b, we get .˛ a/n D b. Now, arguing as in the proof
of the Example 4.23, we conclude that
p n
.˛ a/ D .˛ n C an2 ˛ n2 C an4 ˛ n4 C /C
p
C .an1 ˛ n1 C an3 ˛ n3 C / a;
13. First of all, let’s find out which is the (constant) value of the given expression.
To this end, substituting x D 2 and y D z D 1 (observe that xCyCz D 0), we
x5 Cy5 Cz5
get x5 C y5 C z5 D 30 and xyz.xy C yz C zx/ D 6, so that xyz.xyCyzCzx/ D 5.
x Cy Cz 5 5 5
Therefore, we ought to prove that xyz.xyCyzCzx/ D 5, for all nonzero reals x, y
and z such that x C y C z D 0. To this end, note that
Now, observe that nl nlk D .nkl/ŠkŠlŠ
nŠ
, so that it suffices to make j D n k l
to get the formula in the statement of the problem.
10 10 j k 1 l
16. The general term in the expansion of 1 C x C 6x is j;k;l 1x x D
10 kl
j;k;l
x , with j C k C l D 10. Therefore, we should have k C l D 10 j
and k l D 0, so that k D l D 5 2j and, hence, j must be even. Now, make
equal to 0, 2, 4, 6, 8 and 10 and add all of the corresponding
j successively
summands j;5 10 j j .
2 ;5 2 P n
17. Use the trinomial expansion formula to get jCkClDn j;k;l D .1 C 1 C 1/n and
P
jCkClDn .1/ j;k;l D .1 C 1 1/ .
l n n
18. Make induction on n. To the induction step you will need to use Stiefel’s
relation, togetherPwith the fact that FkC2 D FkC1 C Fk , for every k 1.
19. Let’s show that 995 k 1991 1991k
kD0 .1/ 1991k k
D 1. To this end, start by writing
995
!
X 1991 1991 k
.1/ k
D
kD0
1991 k k
995
! 995 !
X X
k 1991 k k 1991 k
D .1/ C .1/ k
kD0
k kD0
1991 k k
995
! 995
!
X X
k 1991 k 1990 k
D .1/ C .1/k :
kD0
k kD0
k1
P
Now, for n 2 N, define Sn D k0 .1/k nk by setting nk D 0 for k >
k k Pn2
n k, so that S0 D S1 D 1. Use the columns’ theorem to get mD0 Sm D
P P n2
k0 .1/ k n1k
kC1
D 1 S n or, which is the same, S n D 1 mD0 m From
S .
this last relation, prove that SnC1 D Sn Sn1 for every n 2 ZC , and compute
554 B Hints and Solutions
995
!
X 1991 1991 k
.1/k
D S1991 S1989
kD0
1991 k k
D S5 S3 D 0 .1/ D 1:
Section 5.1
Then, use the inequality between the arithmetic and geometric means for six
numbers.
8. By adequately grouping in pairs the summands at the left hand side and
applying inequality (5.7), between the arithmetic and harmonic means, we get
aCc cCa 1 1 4
C D .a C c/ C .a C c/
aCb cCd aCb cCd .a C b/ C .c C d/
and
bCd dCb 1 1 4
C D .b C d/ C .b C d/ :
bCc dCa bCc aCd .b C c/ C .a C d/
In order to prove this last one, it suffices to apply inequality (5.1) n times,
adding the n inequalities thus obtained.
10. It follows from (5.4) that
p
4
3a4 C b4 D a4 C a4 C a4 C b4 4 a4 a4 a4 b4 D 4ja3bj 4a3 b:
11. We’ve already shown that a2 C b2 C c2 ab C bc C ca, for every positive reals
a, b and c. Multiply both sides of this inequality by a C b C c to get the desired
inequality.
12. Since .ab C bc C ca/2 D .ab/2 C .bc/2 C .ca/2 C 2abc.a C b C c/, it suffices
to show that
16. Apply the inequality between the arithmetic and geometric means.
17. Change ai by ai and, then, apply the inequality between the arithmetic and
geometric means.
18. For item (a), it follows from the inequality between the arithmetic and
geometric means that
n n
1C2CCn nC1
nŠ D 1 2 n D :
n 2
For item (b), argue as in (a), applying the inequality between the arithmetic and
geometric means to the numbers 12 , 22 , . . . , n2 and, then, using the result of
Example 4.18.
19. For 1 k m 1, apply the inequality between the arithmetic and geometric
means to x C k D x C 1„ C ƒ‚ C…
1. Then, multiply the m 1 inequalities thus
k
obtained.
P
20. Since njD1 .S xj / D .n 1/S, the inequality in the statement of the problem
is equivalent to
0 10 1
Xn X
n
1 A
@ .S xj /A @ n2 :
jD1 jD1
S xj
with pi D k1 : : : ki1b ki kiC1 : : : kn , the hat over ki indicating that the product
contains all of k1 , k2 , . . . , kn , with the exception of ki . Then, we can write
Now, expand the summands of the numerator in such a way that each aki i appears
pi times and apply the inequality between the arithmetic and geometric means
to get
1
Œ.ak1 C C ak11 / C C .aknn C C aknn /
k1 k2 : : : kn „1 ƒ‚ … „ ƒ‚ …
p1 vezes pn times
1
.a1 /k1 /p1 : : : .an /kn / pn k 1 k2 :::kn
Y
n pi ki
k k :::kn
D ai 1 2 D a1 a2 : : : an :
iD1
1 1 j2
CC :
aj aj1 a1 a0 aj
Then, add the above inequalities for 1 j n and group equal terms to get
the desired inequality. Finally, conclude that equality holds if and only if the
sequence .ak /k1 is an AP.
24. Substituting a, b and c respectively by 6x6 , 6y6 and 6z6 , show that it suffices to
prove that
To this end, write the left hand side expression as a sum of seven other
expressions, chosen in such a way that, when applying the inequality between
the arithmetic and geometric means to each one of them, we get the seven
summands of the right hand side. For instance:
Section 5.2
p p
2 2 2 2
1. Byp Cauchy’ inequality, we have 12144D 3x C 4y 3 C 4 x C y D
5 x2 C y2 . Therefore, x2 C y2 25 , with equality holding if and only if
y 36
3 D 4 . However, since 3x C 4y D 12, equality holds if and only if x D 25 and
x
48
y D 25 .
2. Adapt the discussion of the case n D 3, presented in the text.
3. Making b1 D b2 D D bn D 1 in Cauchy’s inequality, we get
q p
a1 C a2 C C an a21 C a22 C C a2n n;
with equality if and only if there exists a nonzero real number such that a11 D
a2
1
D D a1n D , i.e., if and only if a1 , a2 , . . . , an are all equal. Dividing by
n both sides of the above inequality, we get the decided inequality.
4. Apply the inequality between the quadratic and arithmetic means (see last
problem) to the numerator of the fraction at each summand above. Then, add
the results thus obtained.
5. We want to prove that
0 10 1
X
n Xn
1A
@ xj A @ n2 ;
jD1
x
jD1 j
p
for all positive reals x1 , x2 , . . . , xn . To this end, make aj D xj , bj D p1xj and
apply Cauchy’s inequality.
6. Multiply both psides by ap C b C c C dpand applyp Cauchy’s inequality for n D 4,
with a1 D a, a2 D b, a3 D c, a4 D d and b1 D p1a , b2 D p1b ,
2 p4 . Alternatively, apply (5.7), writing c 4 1 1
b3 D p
c
, b4 D d
D 2c C 2c , c D c=2 C c=2 ,
16 1 1 1 1
d D d4 C d4 C d
4 C 4 and d D d=4 C
d
C d=4 C d=4 .
pd=4
p p
7. For the left hand inequality, let S D 2x C 3 C 2y C 3 C 2z C 3. Applying
the inequality between the arithmetic and geometric means for three numbers,
we get
p
S 3 6 .2x C 3/.2y C 3/.2z C 3/
p
D 3 6 8xyz C 12.xy C xz C yz/ C 18.x C y C z/ C 27
p6
p
3
> 3 18 3 C 27 D 3 9:
For the right hand inequality, apply the inequality between thepquadratic and
arithmetic
p means forpthree numbers (cf. Problem 3), with a1 D 2x C 3, a2 D
2y C 3 and a3 D 2z C 3, and use the fact that x C y C z D 3.
B Hints and Solutions 559
10. Since
we get
X
n
a2k Xn Xn
b2k Xn
b2k
D .ak bk / C D :
kD1
ak C bk kD1
a C bk
kD1 k
a C bk
kD1 k
Hence,
X
n
a2k 1X
n X a2 C b2
n
1X
n
ak , k k
.ak C bk /;
kD1
ak C bk 2 kD1 a C bk
kD1 k
2 kD1
Cy 2 2
and it suffices to show that xxCy xCy2 , for all positive reals x and y. This
follows immediately from the inequality between the quadratic and arithmetic
means (cf. Problem 3).
11. Start by letting anC1 D a1 . Now, for 1 j n, use Cauchy’s inequality to
write
!2 !
p aj a2j
.1 C aj /2 D 1 1 C ajC1 p .1 C ajC1 / 1 C ;
ajC1 ajC1
so that
a2j .1 C aj /2
1C :
ajC1 1 C ajC1
Then, multiply the n inequalities obtained in this way to reach the desired one.
12. Perform Ravi’s transformation (according to the discussion that precedes the
statement of Problem 6, page 120) to show that the given inequality is
equivalent to
560 B Hints and Solutions
p p p p p p
2x C 2y C 2z y C z C x C z C x C y;
and
v ! n !
u n
X
n u X X
2 2
ak bk t 4
ak 4
bk
kD1 kD1 kD1
Now, check that the inequality between the quadratic and arithmetic means (cf.
Problem 3) gives
v !
u
X
n u Xn
2 t
ck n 4
ck :
kD1 kD1
Finally, use the inequality between the arithmetic and geometric means to
conclude the proof.
Section 5.3
m
2. Write am D mm 1 C am m
and apply Bernoulli’s inequality. Do the same with
an and, then, add the results.
3. The desired inequality is equivalent to
a6 b6 c6
C C ab C ac C bc:
b 2 c2 a 2 c2 a2 b2
1
x41 C x42 C C x41994 .x3 C x32 C C x31994 /.x1 C x2 C C x1994 /:
1994 1
Now, use the equations of the system.
5. Apply Chebyshev’s inequality to each summand of the left hand side.
6. Let’s show that, for every n > 1, one has
! !
X
n X
n X
n
i a i b i i a i i b i 0:
iD1 iD1 iD1
To this end, start by observing that the left hand side equals
X
n X
n
i .1 C C b
i C C n /ai bi i j a i b j ; (B.2)
iD1 i;jD1
i¤j
where b i indicates that the summand i is omitted. Then, notice that, to each
pair .i; j/ with i < j, the sum
Equality holds when all of the ai ’s or all of the bi ’s are equal. Finally, the usual
Chebyshev’s inequality corresponds to the particular case in which i D 1n , for
1 i n.
7. Apply Chebyshev’s inequality to conclude that the expression at the left hand
side is greater than or equal to
1 3 3 3 3 1 1 1 1
.a C b C c C d / C C C :
4 bCcCd aCcCd aCbCd aCbCc
Then, apply the result of Corollary 5.20, together with (5.7), to conclude that
the last expression above is greater than or equal to
3
aCbCcCd 42 .a C b C c C d/2
D :
4 3.a C b C c C d/ 12
Finally, observe that the condition given in the statement of the problem is
equivalent to .aCc/.bCd/ D 1, and apply the inequality between the arithmetic
and geometric means.
8. By symmetry we can suppose, without loss of generality, that a b c. This
1 1 1
way, we get bCc aCc aCb . Applying Chebyshev’s inequality and (5.7), we
obtain
an bn cn 1 1 1 1
C C C C .an C bn C cn /
bCc aCc aCb 3 bCc aCc aCb
3
.an C bn C cn /:
2.a C b C c/
1
a n C b n C cn .a C b C c/.an1 C bn1 C cn1 /;
3
which completes the proof.
9. Suppose, with no loss of generalidade, that x y z. Use Chebyshev’s
inequality to conclude that the expression in the statement of the problem is
greater than or equal to
x3 C y3 C z3 .x C 1/ C .y C 1/ C .z C 1/
:
3 .x C 1/.y C 1/.z C 1/
B Hints and Solutions 563
Apply Chebyshev’s inequality to the first factor and the inequality between the
arithmetic and geometric means to the second factor to conclude that the last
3t3 1
expression above is greater than or equal to .tC1/ 2 , where t D 3 .x C y C z/.
Finally, use again the inequality between the arithmetic and geometric means
3t3 3s3
to conclude that t 1, and prove that t s ) .tC1/ 2 .sC1/2 .
10. In order to get the first inequality, apply Chebyshev’s inequality to the first
sum, followed by the result of Corollary 5.8. For the second inequality, use the
inequality between the arithmetic and geometric means.
11. Suppose, with no loss of generalidade, that a1 a2 a2n . Arrange the
given numbers in n pairs .b1 ; c1 /, .b2 ; c2 /, . . . , .bn ; cn /, with b1 b2 bn
and bj cj , for each j. We wish to maximize
S D b 1 c1 C b 2 c2 C C b n cn :
S0 D S bi ci bj cj C bi cj C bj ci D S C .bi bj /.cj ci / S:
b 1 c1 b 2 c2 b n cn ;
AM .x1 ; : : : ; xn / GM .x1 ; : : : ; xn /;
564 B Hints and Solutions
and
and, therefore
and
and, hence,
14. Make a proof by induction on n > 1. For the induction step, it suffices to prove
that
!
a2n a2nC1 a2
1C 1C 1 C n .1 C anC1 /
anC1 a1 a1
1 1 a2
a2nC1 C a2n anC1 C n :
a1 anC1 a1
and suppose that the expression attains its greatest value2 for some sequence
.x1 ; x2 ; : : : ; xn / such that none of the xi ’s is equal to 0. Substituting xn1 by 0
and x1 by x1 C xn1 , we obtain a new expression, which is less than or equal to
the original one, i.e., which is such that
2
Here, we’re tacitly assuming that there exists a sequence .x1 ; x2 ; : : : ; xn / such that x1 C x2 C C
xn D 1 and E.x1 ; x2 ; : : : ; xn / is maximal. Although this could be rigorously proved, such a proof is
beyond the scope of these notes, so that we shall not present it here.
566 B Hints and Solutions
X
n
xi .xiC1 C xiC3 / 0;
iD1
X
n
aj .j 1/ a1 min f1 1; 1 C 2 2; : : : ; 1 C C n ng:
jD1
Finally, use the condition given in the statement of the problem, together with
the inequality between the arithmetic and geometric means to show that 1 C
C k k, for 1 k n.
Section 6.1
2 12 j
3. For item (a), just see that f .1/ D f 11 D j112 C1 2 D 0, f .10/ D f 101
D
j102 12 j 99
24 2 j22 32 j 5 2 2
D 101 and f 36 D f 3 D 22 C32 D 13 . For item (b), if ja b j D 55
102 C12
and a2 C b2 D 73, then .a; b/ D .8; 3/ or .3; 8/, so that f 38 D f 83 D 55 73
.
If ja2 b2 j D 32 and a2 C b2 D 257, then 2a2 D 289 or 2a2 D 225 and,
hence, a … N. However, setting a2 b2 D 32k and a2 C b2 D 257k, with
k 2 N, we get 2a2 D 289k and 2b2 D 225k, so that k D 2 gives a D 17 and
b D 15 (analogously,
15 by
17letting
b2 a2 D 32k, we obtain a D 15 and b D 17).
32
Therefore, f 17 D f 15 D 257 . Finally, if a2 b2 D 101k and a2 Cb2 D 89k,
for some k 2 N, then a2 D 95k and b2 D 6k, so that .ab/2 D 570k2 ; if
b2 a2 D 101k and a2 C b2 D 89k, we conclude, analogously, that .ab/2 D
570k2 . In any case we reach ap contradiction, for, .ab/2 > 0 > 570k2 . p
5. Start by making x D 1 and y D 2 in the given relation, to compute f .1C 2/.
6. One has to prove that f .akC1 / D f .ak / C f .r/, for every k 1. To this end, let
x D ak and y D r, so that x C y D akC1 .
7. Adapt, to the present case, the hint given to the previous problem.
8. In the case of gf , notice that one has to shrink the common domain X of f and g,
to avoid those x 2 X for which g.x/ D 0.
9. The image is the set Z of integers.
10. Start by showing that fxg 2 Œ0; 1/, for every x 2 R. Then, show that the image
of fg is precisely the interval Œ0; 1/.
11. If an denotes the n-th natural number which is not a perfect square, then there
exist positive integers s and t, such that 1 s 2t and an D t2 Cs; in particular,
t2 is the greatest perfect square which is less than or equal to an . Since there are
exactly an integers from 1 to an , exactly t of which are perfect squares (these
being 12 , 22 , . . . , t2 ), it follows that an is the .an t/-th non perfect square.
Hence, n D an t D t2 C s t, so that
r
1 1
tD nsC C :
4 2
p the1 other hand, by using the fact that s 2t, we get, as above, an > n C
On
n 2 . Therefore,
p 1 p 1
nC nC 1 < an < n C n C ;
2 2
p
and, thus, an D bn C n C 12 c.
568 B Hints and Solutions
12. For the first part of item (a), make x D y D 0; for the second, make y D x.
For item (b), use (a). For item (c), consider initially the case of m 2 N, by
induction; then, use (a) to get the case m < 0. Finally, apply the result of (c) to
get (d) and the results of (c) and (d) to get (e).
Section 6.2
" 2 #
b b 2
f .x/ < 0 , a xC <0, xC
2 2 <0
2a 4a 2a 4a
2 ˇ ˇ p p p
b ˇ bˇ b
, xC < 2 , ˇˇx C ˇˇ < , <xC <
2a 4a 2a 2a 2a 2a 2a
p p
b b C
, <x< :
2a 2a
b 2
b 2
so that af .x0 / < 0 if and only if x0 C 2a 4a2 < 0. Now, since x0 C 2a
b 2
0, we must have 0 x0 C 2a < 4a2 and, hence, > 0. For the second part,
use the result of the previous problem.
9. If a rectangle of perimeter 2p has dimensions x and y, then x C y D p. Show
that its area depends on x according to the quadratic function f .x/ D x2 C px,
and apply the result of Proposition 6.25.
10. Letting l be the width and h be the height of the truck (measured in meters),
it is immediate that the load volume of each truck is 18lhm3 . Hence, we have
to maximize the product lh, subjected to the condition that the trucks can enter
the tunnel using the correct lanes. Since the cross section of the trucks are l h
rectangles, the common length d of their diagonals p should satisfy d 5m.
However, by Pythagoras’ Theorem, we have d D l2 C h2 and, hence,
p p p
lh D l d2 l2 l 25 l2 D 25l2 l4 :
P Pk
Now, show that every x 2 .˛k ; ˛kC1 / satisfies f .x/ D 2k jDkC1 ˛j jD1 ˛j .
Finally, the case of an odd n can be dealt with in an analogous way.
570 B Hints and Solutions
13. For item (a), apply the inequality between the arithmetic and geometric means
to the denominator of the defining formula for f .x/.For item (b), write f .x/ D
2 C x25x
C1
and, then, proceed as in (a). Finally, for item (c), apply the inequality
between the arithmetic and geometric means to get
1
x3 .1 x3 /3 D 3x3 .1 x3 /3
3
4
1 3x3 C .1 x3 / C .1 x3 / C .1 x3 /
:
3 4
.xC10/.xC2/
14. For item (a), write xC1
D ..xC1/C9/..xC1/C1/
xC1
9
D .x C 1/ C xC1 C 10; for
3 3
3
x D 2x C 2x ; for (c), write x D 2 C 2 ; finally, for
a a a x x
item (b), write (d), write
6x D 3x C 3x.
15. Start by writing x4 C y4 D .x2 C y2 /2 2.xy/2 .2 C xy/2 2.xy/2 , and look
at this last expression as a quadratic function of z D xy.
16. First of all, prove that the function f W R ! R, given by f .x/ D x3 C 2x, is
increasing. Then, observe that the equations of the given system can be written
as y D f .x/, z D f .y/, t D f .z/ and x D f .t/. Finally, suppose that x y and
use fact that f is increasing to conclude that x y z t x.
17. For fixed m; n 2 N, let Ik D Œ.kmnC1/n; .kmnC1/m. Since ..kC1/mnC1/n <
2 Cnm
.kmn C 1/m for k > mn mn.mn/
, there exists k0 2 N such that
[
Ik D Œ.k0 mn C 1/n; C1/:
kk0
Since f has infinitely many strangulation points, there exist natural numbers
k k0 and p such that p is a strangulation point of f and p 2 Ik . Then, f ..kmn C
1/n/ < f .p/ < f ..kmn C 1/m/. However, kmn C 1 is relatively prime to m and
n, so that f .kmn C 1/ C f .n/ < f .kmn C 1/ C f .m/ and, hence, f .n/ < f .m/.
Section 6.3
1. For every z 2 R, we have jzj > z if and only if z < 0. Hence, we want to find
the set of x 2 R for which g.f .x// < 0. Make y D f .x/, and apply the result of
Problem 7, page 161, to conclude that 12 < y < 12 . Now, recall that y D f .x/
to show that 12 < x 72 < 12 .
2. Start by observing that .f ı g/.x/ D f .g.x// D 2g.x/ C 7 and, hence, that
2g.x/ C 7 D x2 2x C 3.
2
3. Make g.x/ D y to obtain x D yC3 2 and, then, f .y/ D f .g.x// D 2x 4x C 1 D
2
2 yC32 4 yC32 C 1.
B Hints and Solutions 571
f .x/ C a
xCa
Ca .a C 1/x C .ab C a/
f .f .x// D D xCb
D :
f .x/ C b xCa
xCb
Cb .b C 1/x C .a C b2 /
4x2 4.x2 1/ 4
f .x/ D CkD C Ck
xC1 xC1 xC1
4 1
D 4.x 1/ C Ck D 4 xC1C C .k 8/:
xC1 xC1
1
Hence, letting g.x/ D x C 1 C xC1 , we have f .x/ D 4g.x/ C .k 8/.
Now, Problem 2, together with the fact that g is an odd function, implies that
Im .g/ D Rn.2; 2/, so that (by the result of Problem 5) Im .4g/ D Rn.8; 8/.
Finally, Problem 4 guarantees that Im .f / D R n .k 16; k/, and we must have
k 16 D L and k D L.
15. For item (a), start by showing that every x 2 R can be written as
x D kp C ˛, with k 2 Z and ˛ 2 Œ0; p/. Then, make induction on k to
572 B Hints and Solutions
that f .x/ D f .x kp/, for every k 2 Z. For item (b), first observe that
show
g x C jaj
p
D f .ax ˙ p/ D f .ax/ D g.x/, for every x 2 R; subsequently, note
that if p0 > 0 satisfies g.xCp0 / D g.x/ for every x 2 R, then f .axCap0 / D f .ax/
for every x 2 R. Finally, use this fact to deduce that jajp0 D jap0 j p.
14. In order to show that f is odd, start by observing that f .20 C x/ D f .10 C
.10 C x// D f .10 .10 C x// D f .x/ and, analogously, f .20 x/ D f .x/.
Then, use the fact that f .20 C x/ D f .20 x/. For the second part, compute
f .40 C x/ D f .20 C .20 C x// D f .20 C x/ D f q .x/.
17. Making g.x/ D f .x/ C 12 , show that g.x C a/ D 14 g.x/2 and, hence, that
g.x C 2a/ D g.x/, for every x 2 R.
18. If X D fx 2 ZI x > 100g, then Im.f / D f .X/ [ f .Z n X/. On the one hand,
f .X/ D fy 2 ZI y > 90g. Now, if 90 x 100, then 101 x C 11 111 and
f .x/ D f .f .x C 11// D f .x C 1/. Therefore, f .90/ D f .91/ D D f .100/ D
f .f .111// D f .101/ D 91. Use strong induction to show that f .90x/ D 91 for
every integer x 1. For the induction step, if f .90 x/ D 91 for every x < n,
observe that, for x D n, one has f .90 n/ D f .f .101 n// D f .91/ D 91.
Finally, conclude that Im.f / D f91; 92; 93; : : :g.
19. From f .f .n// D 4n C 1, we get f .4n C 1/ D f .f .f .n/// D 4f .n/ C 1. On the
other hand, it’s easy to prove by induction, with the aid of items (a) and (b), that
f .2k / D 2kC1 1. Consequently, again by item (c), we obtain f .2kC1 1/ D
f .f .2k // D 4 2k C 1. Due to these facts, and since 1993 D 4 498 C 1,
we successively compute f .1993/ D 4f .498/ C 1, f .498/ D 2f .249/ C 1,
f .249/ D f .4 62 C 1/ D 4f .62/ C 1, f .62/ D 2f .31/ C 1 and f .31/ D
f .25 1/ D 4 24 C 1 D 65.
20. Write N D A1 [ A2 [ A3 [ : : :, with A1 , A2 , . . . pairwise disjoint infinite sets.
Then, for each n 2 N, set f .x/ D n whenever x 2 An .
21. For item (a), use the assumptions on A and B to construct an injection from
A B to N N. For item (b), take f W Z N ! Q such that f .m; n/ D mn . The
first part of (c) now follows from (b) and Lemma 6.43. As for the second part,
let g W Q ! N be defined by letting g.r/ D minfn 2 NI f .n/ D rg.
22. For item (a), let B1 DA1 and, once we have defined B1 ; : : : ; Bk1 , for some
natural k, let Bk D Ak n [k1 jD1 Aj . For (b), since Bk Ak , we conclude that Bk is
S
finite or countably infinite. If fk W Bk ! N is injective, let f W k1 Bk ! N N
be given by f .x/ D .fk .x/; k/ if x 2 Bk and verify that f is injective. Finally, (c)
follows from (b) as a much easier particular case, arguing by contradiction and
taking A1 D I and A2 D Q.
23. Suppose, for the sake of contradiction, that F D fA1 ; A2 ; A3 ; : : :g. Define A D
fx1 ; x2 ; x3 ; : : :g N by choosing x1 … A1 , then x2 … A2 such that x2 > x1 C 1,
then x3 … A3 such that x3 > x2 C 1 etc. Since xk … Ak , we have A ¤ Ak . On the
other hand since x1 < x2 < x3 < , we conclude that A is infinite and N n A
contains the infinite set fx1 C 1; x2 C 1; x3 C 1; : : :g, hence is also infinite.
B Hints and Solutions 573
Section 6.4
1. For item (a), use Proposition 6.39. For item (b), suppose that g1 ; g2 W Y ! X
satisfy gi ı f D IdX and f ı gi D IdY , for i D 1; 2. Use Proposition 6.34 to get
g1 D g1 ı IdY D g1 ı .f ı g2 / D .g1 ı f / ı g2 D IdX ı g2 D g2 .
2. Use several times the result of Proposition 6.34, together with the fact that f 1 ı
f D IdX , f ı f 1 D IdY , g ı g1 D IdZ and g1 ı g D IdY .
3. Apply the result of Example 6.48.
4. See the hint to the next problem.
cxCd D y has, for every y ¤ c , the
5. For the first part, note that the equation axCb a
so that f .x1 / < f .x2 / and f is increasing, thus injective. For the second part,
let y 2 Œ0; C1/ be given. Since we are assuming that Im.f / D Œ0; C1/, there
exists x 2 Œ0; C1/ such that y D xn ; then, by the definition of the n-th root of
p p
y, we get x D n y, so that f 1 .y/ D n y.
7. Setting f .x/ D 2x, we have f 1 .x/ D 2x , so that .f C f 1 /.x/ D 5x 2
and .f
f 1 /.x/ D 3x2 . These last two functions are bijections from R to itself.
8. Suppose that f is increasing (the case of a decreasing f is entirely analogous).
Given y1 < y2 in J, take the elements x1 ; x2 2 I such that f .x1 / D y1 and
f .x2 / D y2 . If x1 D x2 , then y1 D f .x1 / D f .x2 / D y2 , which is not the case; if
x2 < x1 , it follows from the fact that f is increasing that y2 D f .x2 / < f .x1 / D
y1 , which is not the case either. Therefore, x1 < x2 or, which is the same,
f 1 .y1 / < f 1 .y2 /. Since the elements y1 < y2 of J were chosen arbitrarily, this
assures that f 1 is also increasing.
9. Show that, for fixed a; b 2 R, the p system of equations
p x3 D a and x f .y/ D b
1 3
admits the single solution x D a and y D f . a b/.
3
10. Start by showing that it suffices to prove that any two functions in G commute
in respect to the operation of composition of functions. To this end, use items
(a) and (b), together with the fact that the only function h 2 G of the form
h.x/ D x C a, for some a 2 R, is h D IdR .
11. Set g D f 1 and start by showing that what is needed to be proved is equivalent
to the existence of positive integers x < y < z, in arithmetic progression and
such that g.x/ < g.y/ < g.z/; then, show that we can suppose that x D 1. To
what is left to do, fix t 2 R and show that one cannot have g.t/ > g.2t 1/ >
g.4t 3/ > g.8t 7/ > > g.1/.
574 B Hints and Solutions
Section 6.5
1. If there existed x1 < x2 in I such that f .x1 / D g.x1 / and f .x2 / D g.x2 /, we
would have f .x1 / > f .x2 / D g.x2 / > g.x1 / D f .x1 /.
2. Initially, note that x D 4 is a solution. Then, look at both sides of the given
equation as functions from .0; C1/ to R and apply the result of the former
problem.
3. Letting g.x/ D f .x/ f .0/ in the given relation, conclude that it suffices to
consider the case which f .0/ D 0. Under this additional hypothesis, make
in
y D 0 to get f 2x D f .x/ 2 , for every x 2 Q. Then, use the given relation to
conclude that f .x C y/ D f .x/ C f .y/, for all x; y 2 Q, so that (according to the
first part of the solution of Example 6.56) f .x/ D f .1/x, for every x 2 Q.
4. Use the given relation to prove by induction that f .n/ D f .1/, for every n 2 N;
then, consider the case n < 0 by making y D x.
5. Make x D y D z D 0 to get f .0/ D 12 ; then, obtain f .1/ D 12 by an analogous
reasoning. Make y D z D 1 to conclude that f .x/ 12 , for every x 2 R. Finally,
make an analogous substitution to prove that f .x/ 12 , for every x 2 R.
6. In each of the intervals Œn; n C f .n/ and Œf .n/; 2f .n/ there are f .n/ C 1 naturals.
Use the increasing character of f , together with the given relation, to conclude
that f .n C k/ D f .n/ C k, for every 1 k f .n/. Now, fix natural numbers
k and n, with n > k. Use the fact that f .n/ n to show that f .n/ > k and, by
what was done above, that f .n/ D n 1 C f .1/.
7. Set x D a D 0 in (a) to get f .0/ D 0. Then, letting x D 0 in (a), show that
f .a/ D a for every a 2 Z. Now taking x 2 Œ0; 1/, use the fact that f .x/ 2 Z and
f .f .x// D 0 to conclude that f .x/ D 0. Finally, for a general x 2 R, change x
by fxg (the fractional part of x) and make a D bxc in (a) to obtain f .x/ D bxc.
8. Start by computing f .x C f .y C f .0/// in two different ways to get f .0/ D 0.
Then, deduce that f .f .y// D y for every y 2 Z, so that f is bijective. From this
last relation, compute f .f .f .x/// in two different ways to conclude that f is odd.
Change x by f .x/ in the given relation to obtain f .f .x/ C f .y// D f .f .x C y//
and, then, f .x/ C f .y/ D f .x C y/, for all x; y 2 Z. Finally, use induction to
get f .x/ D f .1/x, for every x 2 Z, thus arriving at the contradiction 1 D
f .f .1// D f .1/f .1/ 0.
9. Set k D 0 to conclude that f .1/ D 2; then, use induction to show that f .n/ D
n C 1, for every n 2 N. Now, use the given relation to show that, if f .1/ D a,
then a < 0 or a > 0 lead to contradictions. Finally, make another induction to
conclude that f .n/ D n C 1, for every n 2 N.
10. First of all, show that we can suppose that f has no fixed points. Then, write
A D B [ C, where B and C are disjoint sets such that f .x/ > x for every x 2 B
and f .x/ < x for every x 2 C. If P B has l elements and a is the common value
of jf .xj / xj j, conclude that 0 D kjD1 .f .xj / xj / D .2l k/a and, hence, that
a D 0, which is a contradiction.
B Hints and Solutions 575
11. Set x D y D 0 in the given relation to obtain f .0/ D 2f .0/f .a/ and, hence,
f .a/ D 12 . Set just y D 0 to get f .x/ D f .a x/ for every x 2 R, so that
f .x/ D f .a C x/, also for every x 2 R. Substitute y D a in the given relation
to get
1 1
f .x/ D f .x C a/ D f .x/f .0/ C f .a/f .a x/ D f .x/ C f .x/ D f .x/;
2 2
for every x 2 R. Now, let y D x in the given relation to find
1
D f .0/ D f .x/f .a C x/ C f .x/f .a x/
2
D f .x/f .x/ C f .x/f .x/ D 2f .x/2 :
Therefore, if we show that f .x/ 0 for every x 2 R, it will follow from the last
relation above that f .x/ D 12 for every x 2 R. To what is left to do, it suffices to
make y D x in the given relation, to obtain
Section 6.6
1. For the second part, show that the x-coordinates of the intersection points of the
graphs are the solutionsp
of the equation f .x/ D x. In our case, the only such
5C 5
solution is x D 2 .
2. If jf .x/j M for every x 2 I and .x0 ; y0 / 2 Gf , then jy0 j D jf .x0 /j M, so that
M y0 M. Hence, .x0 ; y0 / belongs to the horizontal strip of the cartesian
plane bounded by the parallel lines y D M and y D M.
3. Since the bisector of the odd quadrants of the cartesian plane is the set of points
.x; y/ for which x D y, we conclude that x0 2 I is a fixed point of f if and only if
f .x0 / D x0 , i.e., if and only if .x0 ; x0 / 2 Gf . Therefore, the fixed points of f are
exactly the abscissas of the points where the graph of f intersects that bisector.
In Fig. B.1, if f is increasing for x < x1 and decreasing for x > x3 , its fixed
points are x1 , x2 and x3 , which, in turn, are the abscissas of the points A, B and
C, respectively.
4. Such a point .x; y/ satisfies x 2 I and y D f .x/ D g.x/. Therefore, it suffices to
solve, for x 2 I, the equation f .x/ D g.x/.
5. Let’s prove item (a), the proof of (b) being totally analogous. If f is even, then
f .x/ D f .x/ for every x 2 I. Hence, for x 2 I, we have
However, since the points .x; y/ and .x; y/ are symmetric with respect to the
vertical axis, we conclude that the same is true of Gf .
6. Functions f2 and f4 are always nonnegative and vanish only at x D 0. Also,
it is clear that they are even functions, so that (by the result of the previous
problem) their graphs are symmetric with respect to the vertical axis of the
cartesian system. On the other hand, as jxj increases to C1, it’s evident that
y
C
B
x1
x2 x3 x
x2
x4
O x
the values of f2 and f4 become bigger and bigger, eventually surpassing any
predefined value. Finally,
justifying the fact that the graph of f4 is situated below (resp. above) the graph
of f2 in the interval .1; 1/ (resp. outside the interval Œ1; 1). See Fig. B.2.
Functions f3 and f5 are positive for x > 0, negative for x < 0 and vanish
at x D 0. Also, since they are odd functions, their graphs are symmetric with
respect to the origin of the cartesian plane. Now, as jxj increases, it is clear that
the values of jf3 j and jf5 j become bigger and bigger, eventually surpassing any
predefined value. Finally,
jxj < 1 ) jx5 j < jx3 j; jxj > 1 ) jx5 j > jx3 j and jxj D 1 ) jx5 j D jx3 j;
which justifies the fact that, in the interval .1; 1/ (resp. outside the interval
Œ1; 1), the graph of f5 is closer to (resp. further away from) the horizontal axis
than the graph of f3 . See Fig. B.3.
7. Note that f is the inverse of the function g W R ! R such that g.x/ D x3 , for
every x 2 R. Then, apply the result of Proposition 6.63.
8. Verify that the graph of the integer part function is the union of the sets Œn; n C
1/ fng, when n varies in Z.
9. For item (a), start by observing that, for x 2 Œ0; p/, the periodicity of f
guarantees that f .x C kp/ D f .x/, for every k 2 Z. Hence, .x; y/ 2 Gf , y D
f .x/ , y D f .xSC kp/ , .x C kp; y/ 2 Gf . Then, letting F D Gf \ .Œ0; p/ R/,
we have Gf D k2Z .F C k/, where F C k denotes the translation of F, of k
units, in the direction of the vertical axis. For item (b), note that fxg D x for
x 2 Œ0; 1/. Since fg is periodic of period 1, it follows from (a) that Fig. B.4
sketches the graph of fg.
B Hints and Solutions 579
y x3
x5
O x
... ...
−4 −3 −2 −1 0 1 2 3 x
10. For the item (a), it suffices to show that .x; y/ is on the graph of f if and only if
.x a; y/ is on the graph of g. Indeed, if .x; y/ belongs to the graph of f , then
f .x/ D y and, hence, g.x a/ D f ..x a/ C a/ D y, so that .x a; y/ belongs
to the graph of g; moreover, the converse statement can be established in an
analogous way. For item (b), it suffices to show that .x; y/ is on the graph of f
if and only if .x; y C a/ is on the graph of g. Actually, if .x; y/ belongs to the
graph of f , then f .x/ D y and, hence, g.x/ D f .x/ C a D y C a, so that .x; y C a/
belongs to the graph of g; as in the previous case, the converse statement can
be proved analogously. The other items can be similarly dealt with.
11. At a first glance, it may seem that we cannot use the results of the previous
problem, for, the domain of the function under scrutiny is not the set R of real
numbers and its expression doesn’t match any of the ones considered there.
Nevertheless, since
580 B Hints and Solutions
(0, 2)
(−1, 0)
O x
2x
Fig. B.5 Graph of x 7! xC1
2x 2x C 2 2 2
D D 2 ;
xC1 xC1 xC1
equation x2 y2 D 4). By standard analytic geometry (see, for instance, Chapter
2 2 p
6 of [4]), we obtain the curve of equation xy
p xCyp D 2 2 or, which
2 2
is the same, xy D 1.
Section 6.7
1. For the first part of item (a), let A D .1; 0/ and P be the point on the
_
trigonometric circle such that AP D x (of course, in the counterclockwise
direction). Since sin x is the ordinate of P, if x1 ; x2 2 2 ; 2 , with x1 < x2 ,
have sin x1 < sin x2 , so that the restriction of the sine function to the
we clearly
interval 2 ; 2 is increasing and, as such, injective. Since this restriction has
image Œ1; 1, the arcsin function is well defined and, by Problem 8, page 176,
is also increasing. The first part of item (b) can be dealt with in a similar way,
observing that cos x is the abscissa
of P. Finally, the first part of item (c) follows
from the fact that, for x 2 2 ; 2 , the real number tan x is the ordinate of the
!
intersection point of the half-line OA with thevertical
line passing through A.
2. For item (a), observe that cot x D tan x C 2 and use the results of
Problema 10, page 194. For items (b) to (d), use the result of Proposition 6.63.
3. For items (a) and (b), we refer the reader to the discussion of Example 6.68,
more precisely to Eq. (6.15); as for item (c), we suggest the reader to recall
items (a) and (e) of Problem 10, page 194.
p
4. Argue as in Example 6.68 to get f .x/ D a2 C b2 cos.x ˛/, where cos ˛ D
p a
2 2
and sin ˛ D p 2b 2 . Now, show that f ha period 2jj
.
a Cb a Cb
5. For item (a), start by using trigonometric identities to get f .x/ D 2 sin2 x C
2
2 sin x C 1. Then, observe that the quadratic y 7! 2y C 2yC
function 1 attains
its maximum value at y D 2 , increases on 1; 2 and decreases on 12 ; 1 . Item
1 1
(b) follows from the expression obtained above for f .x/, together with the fact
that the sine function has period 2. Finally, for item (c), sketch the graphs of
x 7! 2 sin x and x 7! cos.2x/ on the interval Œ0; 2 and in a single cartesian
coordinate system; then, add corresponding ordinates of the points of these two
graphs to get the desired sketch of the graph of f .
6. Use the result of Problem 4, page 193.p
7. Perform the change of variable x D 5 cos
, with 0
. Then, use
the discussion of Example 6.68 to conclude that f .x/ D 25 cos.
˛/, with ˛
satisfying cos ˛ D 35 and sin ˛ D 45 .
8. Suppose, by the sake of contradiction, that f is periodic, of period > 0. Then,
we must have f ./ D f .0/ D 2. Use this equality, together with the fact that
the cosine of every real number is at most 1, to get cos D cos.˛/ D 1.
Conclude that there must exist (nonzero) integers k and l for which D 2k
and ˛ D 2l, and arrive at a contradiction.
9. Initially, show that the equality f .x C 3/ D f .x/ implies the equality
582 B Hints and Solutions
5x 15 5x
.1/ sin
n
C D sin ;
n n n
Section 7.1
n
1 1
1. For item (a), observe that p n
1=x
D 1=x D x. For (b), argue in an analogous
way. Item (c) now follows immediately from the results of Problems p p6 and 8,
page 176. Finally, for item (d), Problem 6, page 176, gives m a > n a if and
only if an > am ; the rest follows from Corollary 1.3.
2. For item (a) suppose, by the sake of contradiction, that a > 0. Then, by the
Archimedian property of N, we can choose n 2 N such that n > 1a . This is
equivalent to a < 1n , which is an absurd. For item (b), choose n 2 N such that
n > cba .
3. For item (b), let r1 ; : : : ; rn 2 .a; b/ \ Q, with r1 ; : : : ; rn pairwise distinct. If
r D minfr1 ; : : : ; rn g, then r 2 .a; b/ \ Q, and item (a) gives a < aCr 2 < r.
Finally, argue similarly, using the second part of (a), to get infinitely many
irrational numbers in the interval .a; b/.
4. For item (a), suppose that we already have the truth of the result for a 0. If
b 0, then b 0 and, by item (a), there existe r 2 Q and ˛ 2 R n Q such
that r; ˛ 2 .b; a/. Hence, r; ˛ 2 .a; b/, with r 2 Q and ˛ 2 R n Q.
If a 0 < b, show that it suffices to apply, to the interval .0; b/, the result we
assumed to be known. For item (b),puse the Archimedian property of natural
numbers to get n 2 N such that n > ba2 . Finally, in what concerns (c), starts by
using the Archimedian
p
property to guarantee the existence of m 2 N for which
m 2
n > b (resp. n > b); continue by showing that, if m is the least possible
m
p
natural number fullfilling this requirement, then m > 1 and .m1/n
2
2 .a; b/.
5. Adapt the proof of items (b) and (c) of the previous problem to the present case.
6. Use the result of Example 7.2, together with Theorem 7.4.
7. Use the result of item (b) of Problemp 4. p
8. For a D k2 and b D k2 C 1, we have j a bj D pjabj p D p
2
1 p
2
< 2k1
.
aC b k C1C k
Since every element of X is positive, it follows that inf.X/ D 0.
9. Observe that, for k 2 N, every number of each of the forms 31k and 1 31k belong
to C. Now, use the fact that 31k < 1k , for every k 2 N.
B Hints and Solutions 583
Section 7.2
1
as n ! C1. Hence, fixed a real number ˛ such that jaj < ˛ < 1, there exists
q
n nk nk
n0 2 N for which n > n0 ) jajn < ˛, or jajn < ˛ . However, since ˛ n ! 0
n
nk
as n ! C1, the same happens with an
. Alternatively, let jaj D 1 C ˛, with
˛ > 0, and note that, for n > k,
! !
Xn
n j n
jaj D .1 C ˛/ D
n n
˛ > ˛kC1
jD0
j k C 1
n.n 1/ : : : .n k/ kC1
D ˛ :
.k C 1/Š
584 B Hints and Solutions
nk .k C 1/Š nk
<
jaj n ˛ kC1 n.n 1/ : : : .n k/
.k C 1/Š 1
D 1
;
˛ kC1
n.1 n / : : : .1 nk /
p an C b a C bn
n2 C an C b n D p D q :
n2 C an C b C n 1 C an C nb2 C 1
p
For item (c), observe that, if an D n
1 C qn , then an > 1 and, by item (a) of
Problem 18, page 26,
ann 1 qn
0 < an 1 D < :
ann1 C ann2 C C an C 1 n
p q
Finally, for (d), write n an C bn D a n 1 C . ba /n and use the result of the
previous item.
8. Adapt, to the present case, the reasoning that led to the proof of Proposi-
tion 7.18.
9. For the first part, write q D 1 C ˛, with ˛ > 0, and note that an D .1 C ˛/n
1 C n˛, so that an > M for n > M1 ˛
.
10. Let b D 1q , so that b > 1. Use the first part of the previous problem to get k 2 N
such that bk > a. Then, show that a2n > kn
n
for every sufficiently large n and,
from this, conclude that a q D bk b2n1kn ! 0 when n ! C1.
n 2n
19. First of all, use the inequality between pthe arithmetic and geometric means of
two numbers to conclude that akC1 a, for every k 2 N. Now, make k !
C1 in the given precurrence relation to conclude that, if .an /n1 converges, then
its limit equals a. To what is left to do, use the triangle inequality to obtain
ˇ ˇ
1ˇ a a ˇˇ
jakC1 ak j D ˇˇak ak1 C
2 ak ak1 ˇ
ˇ ˇ
1 ˇ a ˇˇ
D jak ak1 j ˇˇ1
2 ak ak1 ˇ
1
jak ak1 j;
2
for each k > 2. Finally, establish the convergence of .an /n1 from the result of
Example 7.28.
20. Let a1 , a2 , . . . , an be the initial terms of the AP’s, and say that the i-th AP is
that of initial term ai and common ratio di . Choose n 2 N greater than all of
the ai ’s, and let ki be the number of terms of the i-th AP belonging to the set
f1; 2; : : : ; ng. Then, on the one hand, k1 C k2 C C km D n; on the other,
ai C .ki 1/di n < ai C ki di . Use these relations to show that
Sm X 1
m
S
1C < < 1C ;
n d
iD1 i
n
P
m ai
where S D iD1 di . Then, make n ! C1.
21. Suppose such a function does exist, let k > 1 be an integer and xj D kj , for
0 j k. Given a real number x 2 Œxj ; xjC1 , show that d.f .x/; f .xj // k˛C1=2
c
.
Letting Cj denote the circle in the plane with center f .xj / and radius k˛C1=2 c
,
conclude that x 2 Œxj ; xjC1 ) f .x/ 2 Cj . Then, use this fact to show that
P c2
1 k1jD0 A.Cj / D k2˛ . Finally, use the fact that k > 1 can be chosen arbitrarily
and ˛ > 0 to reach a contradiction.
22. Start by proving that anC1 an C 1 and, hence, that an n for every n 1.
Now, use telescoping products to show that, for a fixed p 2 N, we have
!
anCp Y
p1
1
1< < 1C p
an jD0
a nCj
and, then,
!
anCp Y anCjC1 Y
p1 p1
1
1< < < 1C p :
an jD0
anCj jD0
anCj
!
1 1 1
lim 1 C p 1C p 1C p D 1:
n!C1 an anC1 anCp1
Now, if y D 1, conclude that what is asked to be proved follows from the limit
above. If y < 1, conclude from the limit above that we can choose np 2 N such
that n > np ) anCpan
> y. Use the fact that ak k for every k to show that we
1
can choose a natural number n0 such that p
1C an0 C1 < y x. Then, note that
n > n0 and p natural furnish
an an
p < y x:
anCp anCp C anCp
c1 C c2 C C cn D a 0 a n c n : (B.3)
cn
On the other hand, the assertion of the problem is equivalent to cn1 < 21=n for
infinitely many values of n 2 N. Arguing by contradiction, suppose that there
exists a natural N such that the opposite inequality is true for every n N.
Then, for n > N we have
1 1
CC 1n
D C 2.1C 2 CC n / ;
NC1 C NC2 1 1
cn cN 2
X
n
1 1 1 1 1 1 1
1C C C CC C CC k
jD1
j 2 3 4 7 2k1 2 1
1 C 1 C 1 C C 1 D k;
588 B Hints and Solutions
so that cn C 2k for 2k1 n < 2k . Let r 2 N be such that 2r1 N < 2r ,
and let m > r. Then,
Section 7.3
note that there are more numbers of the form fj˛g than intervals
2. For (a), k of the
p . For (b), take 1 s < t p C 1 such that fs˛g; ft˛g 2 p ; p ,
form pk ; kC1 kC1
write fs˛g D s˛ bs˛c,ft˛g D t˛ bt˛c and observe that 0 < jfs˛g ft˛gj < 1p .
3. Use Corollary 7.33 to inductively construct a sequence .ak /k1 such that ak D
mk C nk ˛, for integers mk and nk satisfying item (a), and such that jak lj < 1k ,
for every k 1.
!
4. For item (c), choose A and B in such that O … AB , ˛ 2 R n Q and make
! !
X D f.m1 C n1 ˛/OA C .m2 C n2 ˛/OBI m1 ; m2 ; n1 ; n2 2 Zg:
For item iii., conclude that there exist distinct lines r and s through O, such that
X \ r; X \ s ¤ ; and X \ r is not dense in r. Show that one can reconstruct X
from X \ r and X \ s, apply Kronecker’s lemma and make d D s.
5. For item (b), write n D ˛n C ˇn and take integer and fractionary parts. For (c),
argue by contradiction,
n o using the fact o ˛ and ˇ are irrational to show that one
n that
cannot have either ˛ D 1 ˛ or ˇ D 1 ˇ1 .
n 1 n
Section 7.4
1 1 1
1. Use que fact that anC1 D an an C1 , for every n 2 N.
Pn1 1
ak akC1 D a1 an , for
n1
2. Use the result of Problem 5, page 84, to show that kD1
every integer n 1. Then, make n ! C1, noting that an D a1 C .n 1/r.
B Hints and Solutions 589
Pn 2k1
3. Make Sn D kD1 ak
and show that
1 1 1 2n 1
.a 1/Sn D aSn Sn D a C 2 C 2 C C n1
a a a an
2a 1 1 2n 1
DaC n :
a1 a a an
Now, use the results of Example 7.12 and of Problem 4, page 218, to conclude
n 2
that Sn ! a1
a
C .a1/ 2.
p p p P P
1
4. Note that k C k 1 < 2k, so that k1 p p
2 > k1 p12k . Since
kC k2 1
p1 1
for n 1, use the comparison test to show that the given series
n n
diverges.
5. Show that n3 1000n2 > .n 500/3 for every sufficiently large natural n.
Then,
P use the comparison test, together with the convergence of the series
1
k>500 .k500/3=2 .
1
6. Let r > 0 be the common ratio of the AP. For item (a), we have ak D
1 1
a1 C.k1/r > 2.k1/r , provided k > r C 1. For item (b), we have
a1
1 1 1 1
D < k k1 :
a2k a1 C .2k 1/r .2 1/r 2 r
X1 XN
1 X 1 X 1
N N
< ;
x2A
x aD0
2a bD0 3b cD0 5c
for some natural number N. Then, make N ! C1 and use the formula for the
sum of a geometric series. P a
9. Use the comparison test to show that the series j1 10jj converges. Now, letting
x denote its sum, note that, for k n, we have
Xn
aj X aj X 9 1
x D D n:
jD1
10 j
jnC1
10 j
jnC1
10 j 10
10. For the first part, suppose that, for some x 2 .0; 1/, we have x D 0:a1 a2 a3 : : : D
0:b1 b2 b3 : : :, with an ¤ 0 for infinitely many values P of nakand, accordingly,
P
bn ¤ 0 for infinitely many values of n. Then, 10 a1
< k1 10 k D k1 10k
bk
P 9 b1 C1
10 C k2 10k D 10 , so that a1 b1 . Analogously, show that a1 b1 , so that
b1
test.
14. For item (a) (the proof p of item (b) is analogous), let q D lC1 2
2 .0; 1/ and take
n0 2 N such that an < q for n > n0 or, which is the same, an < qn para
n
n > n0 . Then, adapt the reasoning used at the proof of the ratio test.
15. Apply Theorem 7.53.
P
Start by observing that the series k0 .1/
k
16. kŠ
is absolutely convergent. Then, let
P .1/k P
a D k0 kŠ and use the fact that e D k0 kŠ1 , together with Theorem 7.53,
to get ae D 1.
P P n
17. For item (a), let sn D nkD1 ak and tn D nkD1 a'.k/ , so that s1 s2 s3 !
s, where s D supfsn I n 1g, and analogously for .tn /n1 . Given n 2 N, prove
that there exists m 2 N such that sn tm ; then, make n ! C1 to conclude
that s t. Change the roles of the two sequences to get the reverse inequality.
For item (b), note that jan j D aC
n C an , so that an P
C
D 12 jan j C an and a n D
1
C P
2
ja n j a n . For (c), apply the result of (b) to both a
k1 k and k1 k .
a
18. By the definition of convergent series, it suffices to show that the sequence
.xn /n1 , given for n 1 by xn D a1 b1 C a2 b2 C C an bn , is convergent. As
we know, this is the same as showing that it is a Cauchy sequence. So, let m
and n be natural numbers, with m > n. Let M > 0 be such that jsk j < M for
every k 1. It follows from Abel’s identity (5.19) (with the roles of ai and bi
exchanged) and from triangle inequality that
ˇ m1 ˇ
ˇX ˇ
jxm xn j D ˇ si .bi biC1 / C sm bm sn bn ˇ
iDn
X
m1
jsi j.bi biC1 / C jsm jbm C jsn jbn
iDn
20. For item (a), observe that 2 sin.a C jh/ sin h2 D cos.a C .j 1/h/ cos.a C jh/.
Now, sum from j D 0 to j D k and transform the result in a product. For item
P that it suffices to show that the sequence .sn /n1 , given for n 1 by
(b), note
sn D nkD1 sin k, is bounded. To this end, apply the result of item (a) to show
sin n
sin nC1 1
that sn D 2
sin 12
2
and, hence, that jsn j sin 12
.
21. Letting ˛n D An b
OAnC1 , we have sin ˛n D p1 . Now,
use the unit circle to show
n
P
that sin ˛ < ˛ for every ˛ 2 0; 2 . Finally, use the divergence of k1 p1k to
P
show that k1 ˛k also diverges.
22. Letting ˛n D 2ˇn , it suffices to show that ˇ1 Cˇ2 Cˇ3 C D 4 . Equivalently,
n
letting bn D tan.ˇ1 C C ˇn / for n 1, it suffices to show that que bn ! 1.
2 b C1
Use some Trigonometry to show that bnC1 D 2.nC1/ n
2.nC1/2 bn
, for every n 2 N.
Then, use induction to prove that bn D nC1 , for every n 1.
n
Section 8.1
1. For the second part of item (a) note that, if a < 0 < b, then
the image of the
interval Œa; b by f is the union 1; 1a [ f0g [ 1b ; C1 . For item (b), start
by computing the image of the interval 12 ; 0 by f .
2. Suppose that g is continuous, and let g.x0 / D c. For x 2 I n fx0 g, it follows
from triangle inequality that jc f .x0 /j jc g.x/j C jg.x/ f .x0/j D jg.x0 /
g.x/j C jf .x/ f .x0 /j. Now, use the fact that f and g are continuous to conclude
that jc f .x0 /j < , for every > 0.
3. For items (a) and (b), use the result of the previous problem. For items (c) and
(d), use the definition of continuity to conclude that there exists no value of c
that turns the given function into a continuous one – it may help to sketch the
graph of the function.
4. Start by observing that the set .x0 ı; x0 C ı/ .f .x0 / ; f .x0 / C / is an open
(i.e., with its boundary removed) rectangle centered at P0 .x0 ; f .x0 // and with
sides parallel to the axis. Then, recall that every disk centered at P0 contains an
open rectangle centered at P0 and with sides parallel to the axis, and vice-versa.
5. Note that, in D, the tangent function is the quotient of the continuous functions
sin and cos.
6. Use the chain rule for continuous functions.
7. Use the chain rule for continuous functions.
8. For item (a) adapt, to the present case, the proof
p of the continuity of the square
p
root function at x0 D 0. For (b), make y D n x, y0 D n x0 and apply item (a) of
Problem 18, page 26, to get
jyn yn0 j
jy y0 j D :
yn1 C yn2 y0 C C y0n1
592 B Hints and Solutions
Finally, for item (c), use the result of (b) to adapt, to the present case, the proof
of the continuity of the square root function at x0 .
9. Use the result of the previous problem, the chain rule and Example 8.10.
10. Use the chain rule to show that f is continuous at every x0 ¤ 0. Then, use the
definition of continuity, together with the fact that jx sin 1x j jxj, to show that f
is continuous at 0.
11. Fixed x0 2 Œ0; 1, take D 12 and assume, by contradiction, that f is continuous
at x0 . By the definition of continuity, there exists ı > 0 such that x 2 Œ0; 1
and jx x0 j < ı imply jf .x/ f .x0 /j < 12 . Now, separately consider the cases
x0 2 Q and x0 … Q, and use the result of Problem 4, page 206, to arrive at a
contradiction.
12. In order to show that f is discontinuous at every rational number of the
interval Œ0; 1, reason as in the hint to the previous problem, using the result
of Problem 4, page 206. If x0 2 Œ0; 1 is an irrational number and n0 2 N, show
that there exists ı > 0 such that every fractional representation of a rational
number of the interval .x0 ı; x0 C ı/ has denominator greater than n0 ; then,
conclude that jf .x/ f .x0 /j < n10 , for x 2 Œ0; 1 \ .x0 ı; x0 C ı/.
13. The first part follows from the fact that f .x0 / is an upper bound for the set on the
left hand side, whereas is a lower bound for the set on the right hand side. For
the second part, use the definition of continuity, together with Proposition 7.8.
14. That f is well defined follows Pfrom Problem 17, page 243, together with the
fact that the geometric series n1 21n converges. For item (b), show that f .x/
f .xm / 21m if x > xm , so that each xm is a point of discontinuity of f . Now, let
be given x … D and > 0. Take m0 2 N such that 21m < whenever m > m0 ;
then, there exists ı > 0 such that fx1 ; : : : ; xm0 g \ .x ı; x/ D ;. In other words,
xm 2 .x ı; x/ ) m > m0 , and this gives that, for y 2 .x ı; x/,
X 1 X 1 X 1 1
f .x/ f .y/ D D m < :
yx <x
2n 2m mm C1 2m 2 0
n xı<xm <x 0
Analogously, show that f .y/ f .x/ < if y 2 .x; x C ı/ and ı > 0 is sufficiently
small.
Section 8.2
1. First of all, recall that Problem 12, page 152, gives f .x/ D f .1/x for every
x 2 Q. Then, fix an irrational number x0 2 R and use the result of Problem 4,
n
page 206, to find a sequence .an /n1 in Q satisfying an ! x0 . Finally, apply
Theorem 8.16.
B Hints and Solutions 593
2. Start by using the second hypothesis to show that f is injective; to this end, take
distinct x; y 2 R and consider the divergent sequence .an /n1 such that a2k D x
and a2k1 D y, for every k 1. Finally, use the hypotheses on f to show that
f 1 transforms convergent sequences into convergent sequences.
3. Given > 0 and letting c > 0 be the Lipschitz constant of f , show that we can
take ı D c in the definition of uniform continuity.
4. Note that, for every ı > 0, there exist x; y 2 R such that jx yj < ı but
jf .x/ f .y/j 1; for instance, taking x > 0 and y D x C 2ı , it follows from the
n
binomial formula that jf .x/ f .y/j D x C 2ı xn n 2ı xn1 nı2 x, and it
suffices to take a sufficiently large x.
5. If f was uniformly continuous, to the given > 0 there would correspond
n
some ı > 0 as in (8.4). Now, since jan bn j ! 0, we could choose n 2 N
satisfying jan bn j < ı; hence, we should have jf .an / f .bn /j < , which is a
contradiction.
6. Use the result of the previous problem.
7. For item (c), consider g W Œa; b ! R given by g.x/ D Mf1 .x/ , for every x 2
Œa; b.
1
8. It suffices to let f .x/ D .xa/.xb/ , for x 2 .a; b/.
9. Initially, observe that the problem of computing the distance from P to Gf is
equivalent to the problemp of minimizing the continuous function g W .a; b/ !
R, given by g.x/ D .x x0 /2 C .f .x/ y0 /2 . Now, let Q.x0 ; f .x0 // and, for
0 < ı < b a, let Rı D .a C ı; f .a C ı// and Sı D .b ı; f .b ı//. Use
the stated condition to show that there exists ı0 > 0 such that, for 0 < ı < ı0 ,
we have Rı P; Sı P > PQ. Then, conclude that it suffices to minimize g in the
interval Œa C ı0 ; b ı0 . Finally, apply Weierstrass theorem to finish the proof.
10. For item (a), apply the triangle inequality to get
on the other, if ˇ 2 R is such that f .ˇ/ D ˇ, show that the inequalities j˛ˇj D
jf .˛/ f .ˇ/j cj˛ ˇj,
p together with the fact that 0 < c < 1, give ˛ D ˇ.
12. Setting f .x/ D 12 .x C x2 C 1/, we clearly have f .x/ < x for every real x.
Moreover,
594 B Hints and Solutions
ˇ ˇ
1 ˇˇ x 2 y2 ˇ
ˇ
f .x/ f .y/ D ˇx y C p p ˇ
2ˇ x C 1 C y C 1ˇ
2 2
ˇ ˇ
1 ˇ xCy ˇ
ˇ ˇ
D jx yj ˇ1 C p p ˇ
2 ˇ x C 1 C y C 1ˇ
2 2
!
1 jxj C jyj
jx yj 1 C p p
2 x2 C 1 C y2 C 1
< jx yj:
13. Use the sign-preserving lemma, together with Problem 5, page 206, and
Theorem 8.16.
14. Show that the stated condition gives f .0/ D f .m C n˛/, for all m; n 2 Z. Then,
use Corollary 7.32, together with Theorem 8.16, to prove that f is constant.
Section 8.3
1. We could only apply the IVT if the domain of f contained the interval Œ0; 2,
which is not the case.
2. Apply the IVT in each case. Sketching the graphs of the involved functions may
help.
3. Let y D ax C b be a non vertical line, so that a ¤ 0. It suffices to show that
the function g W R ! R, given by g.x/ D x3 sin x ax b, has infinitely
many zeros. To what is left to do, suppose that a > 0 (the case a < 0 can
be handled similarly); show that, for every sufficiently large k 2 N, we have
g.2k/ < 0 < g.2k C 2 /.
4. Look at the function f W Œ0; 1 ! Œ0; 1 such that f .x/ D xC.1x/2 sin .n2/x.
5. Adapt the argument in the proof of Bolzano’s theorem. More precisely, suppose
I D Œa; b (the other cases can be dealt with analogously), fix ˛ 2 X and let
A D fx 2 Œ˛; bI Œ˛; x Xg, so that A ¤ ;. If c D sup A, take a sequence
.an /n1 in A such that an ! c and use condition (ii) to show that c 2 A. If
c < b, use condition (i) to choose 0 < ı < b c (corresponding to x0 D c) to
conclude that c C 2ı 2 A, which is a contradiction. Finally, make an analogous
reasoning to show that Œa; x0 X.
6. Since x D 0 is not a solution, we can write the given equationqas f .x/ D 0,
P
where f W .0; C1/ ! R is the function given by f .x/ D niD1 x12 C axi n.
Thus, we needq to show that f has exactly one positive zero. To this end, since the
function x 7! x12 C axi is continuous and decreasing, and since f is a finite sum
of functions of this kind, we conclude that f is continuous and decreasing too,
which forces it to have at most one positive zero. On the other hand, f .1/ D
B Hints and Solutions 595
Pn p q p
1
iD1 1 C a i n > 0; also, x2
C axi < 1 if x > maxf 2; 2ai g, so that
p
f .x/ < 0 if x > maxf 2; 2a1 ; : : : ; 2an g. Now,
P apply the IVT.
7. Let f W Œ0; 1 ! R be given by f .x/ D 1n niD1 jx xi j. It suffices to guarantee
the existence of x 2 Œ0; 1 for which f .x/ D 12 . To this end, note that
1X 1X
n n
f .0/ D xi and f .1/ D 1 xi ;
n iD1 n iD1
15. Start by observing that, for x 2 .0; 1/, the function g.x/ D x.f .x/ f .0//
measures the area of the closed rectangle having .0; f .0// and .x; f .x// as
endpoints of a diagonal and sides parallel to the coordinate axis. Hence, there
exists 0 < x1 < 1 such that this rectangle has area n12 . Arguing in an analogous
way, we find rectangles R1 , . . . , Rk , each of which having area n12 and sides
parallel to the coordinate axis, such that R1 [ : : : [ Rk contains the graph of f
and is contained in the closed square Œ0; 1 Œ0; 1 of the cartesian plane. Finally,
use Cauchy-Schwarz inequality to show that k n.
16. Suppose, by the sake of contradiction, that there exist x0 ; y0 2 R for which
f .x0 / < f .y0 /; moreover, suppose that y0 2 .x0 ; x0 C f .x0 // (the other cases
can be treated analogously). Use the fact that f .x0 C f .x0 // D f .x0 / to show
that it is possible to choose n 2 N and c 2 R such that the line r of equation
x C ny D c leaves the point .y0 ; f .y0 // in a semiplane opposite to that of the
points .x0 ; f .x0 // and .x0 Cf .x0 /; f .x0 Cf .x0 // D .x0 ; f .x0 //. Now, show the IVT
guarantees the existence of real numbers a 2 .x0 ; y0 / and b 2 .y0 ; x0 C f .x0 //
such that the points A.a; f .a// and B.b; f .b// are on the line r, i.e., such that
a C nf .a/ D c and b C nf .b/ D c. Finally, conclude that f .c/ D f .a C nf .a// D
f .a/ and f .c/ D f .b C nf .b// D f .b/; hence, f .a/ D f .b/ and, then, a D b.
17. Making x D y D 0 in the given relation, we get f .0/ C g.0/ D h.0/. Thus,
letting f1 .x/ D f .x/ f .0/, g1 .x/ D g.x/ g.0/ and h1 .x/ D h.x/ h.0/, we
have
f1 .x C y3 / C g1 .x3 C y/ D h1 .xy/;
with f1 .0/ D g1 .0/ D h1 .0/ D 0. Hence, we can start by supposing that f .0/ D
g.0/ D h.0/ D 0. Making y D x3 in the given relation, we get g.x x9 / D
h.x4 / for every real x. Making x D y3 , we get f .y9 C y/ D h.y4 / for
every y 2 R; therefore, it follows that f .x x9 / D g.x x9 /. However, since
the image of the polynomial function x 7! x x9 is the whole set of reals (see
Example 8.28), we have f .x/ D g.x/ for every x 2 R. Then, the relation at the
statement of the problem reduces to
Performing the substitutions y D x3 and x D y3 in (B.4) and taking into
account that f .0/ D 0, we obtain, respectively,
so that f .x x9 / D f .x9 C x/. Using once more the fact that the image of
the polynomial function x 7! x x9 is R, it follows that f is an even function,
i.e., that f .x/ D f .x/ for every x 2 R. Now, letting y D 0 in (B.4), we get
f .x/ C f .x3 / D 0, so that f .x/ D f .x3 /. Back to (B.4), this relation gives us
B Hints and Solutions 597
has at least one real solution, then h.a/ D 0. Let us prove that, for every a 0,
this is indeed the case: if a D 0, we already have h.a/ D 0. If a ¤ 0, the the
second equation of the system is equivalent to y9 C 3y6 x C 3x2 y3 C y C 2x3 D 0.
Writing x D ay , we conclude that it suffices to guarantee the existence of a real
number y such that
To this end, let p.y/ D y12 C 3ay8 C .3a2 C 1/y4 C 2a3 . If a < 0, then p.0/ D
2a3 < 0 and p.a/ > 0, so that the IVT guarantees that p has at least one
real root. To finish, not that f .x x9 / D h.x4 /, so that f .x x9 / D 0 for
every x 2 R. Invoking once more the surjectivity of f , it follows that f D 0.
Therefore, (B.4) gives h D 0. Then, the functions f , g and h that satisfy the
stated conditions are the constant functions f D f .0/, g D g.0/ and h D h.0/,
such that f .0/ C g.0/ D h.0/.
Section 9.2
1. The idea is to adapt, to the present case, the proof of the uniqueness of limits of
sequences, given in Proposition 7.14. More precisely, suppose that L and M are
distinct reals and that f .x/ converges simultaneously to L and M when x ! x0 .
Take D 12 jL Mj > 0 and ı > 0 such that x 2 I and 0 < jx x0 j < ı give
jf .x/ Lj < and jf .x/ Mj < . Then, use the triangle inequality to arrive at
a contradiction.
2. Suppose we had L > M, and let D LM 2
> 0. Take ı > 0 such that x 2 I
and 0 < jx x0 j < ı give jf .x/ Lj < and jg.x/ Mj < . Use the triangle
inequality to arrive at a contradiction.
3. Make induction on n. The initial case n D 2 is exactly the content of items (a)
and (b) of Proposition 9.7.
4. For right-handed limits, for instance, copy the statement and proof of Proposi-
tion 9.7, changing, whenever convenient, lim by lim and 0 < jx x0 j < ı
x!x0 x!x0 C
(resp. 0 < jx x0 j < ı1 ; ı2 ) by x0 < x < x0 C ı (resp. x0 C ı1 , x0 C ı2 ).
598 B Hints and Solutions
5. Implication ) and the last part follow from the fact that each one of x0 ı <
x < x0 and x0 < x < x0 C ı 0 < jx x0 j < ı do imply 0 < jx x0 j < ı.
Conversely, suppose lim f .x/ D lim f .x/ D L, and let > 0 be given.
x!x0 C x!x0
Take ı1 ; ı2 > 0 such that each of x0 ı1 < x < x0 and x0 < x < x0 C ı imply
jf .x/ Lj < . Letting ı D minfı1 ; ı2 g > 0, conclude that 0 < jx x0 j < ı
implies jf .x/ Lj < .
6. Suppose that lim f .x/ D L > 0 and lim g.x/ D C1 (the other cases are
x!x0 x!x0
completely analogous), and let M > 0 be given. By the sign-preserving lemma,
there exists ı1 > 0 such that x 2 I; 0 < jx x0 j < ı1 ) f .x/ > L2 ; by the
definition of infinite limit, there exists ı2 > 0 such that x 2 I; 0 < jx x0 j <
ı1 ) g.x/ > 2M L
. Then, let ı D minfı1 ; ı2 g and conclude that for x 2 I
satisfying 0 < jx x0 j < ı we have f .x/g.x/ > M. For what is left, start by
observing that the analogous result for limits at infinity (at C1, for instance)
says that if lim f .x/ D L > 0 and lim g.x/ D C1, then lim f .x/g.x/ D
x!C1 x!C1 x!C1
˙1. For the proof of this fact, copy the former one, changing 0 < jx x0 j < ı
by x > A whenever convenient.
7. Suppose first that a; b; L; M 2 R (the cases in which a D 1, b D C1,
L D 1 or M D C1 are totally analogous). Given > 0, let ı > 0 be such
that a < x < a C ı and b ı < y < b respectively imply f .x/ < L C and
g.x/ > M . Then, apply the IVT to conclude that ŒL C ; M Im.f /.
8. In all items one has to use the fundamental trigonometric limit, together
with some algebra. For items (a) and (b), multiply both the numerator and
sin.2x/ 2 sin.2x/ 3x
denominator by 1 C cos x; for item (c), write sin.3x/ D 3 2x sin.3x/ ; for (d),
multiply both the numerator and denominator by 2 ; for (e), use
(d) and the fact
x 1y3
that cos 2 x D sin 2 2 ; finally, for (f), apply the identity 1 y D 1CyCy 2
p
with y D 3 cos x and, then, the result of item (b).
9. For the first part adapt, to the present case, the proof of Corollary 9.9. In order
to compute the limits, use the result of the first part, together with the fact that
j sin xj; j cos xj 1 for all x 2 R.
10. Given > 0, let ı > 0 be such that x 2 I and 0 < jx x0 j < ı imply
jf .x/ Lj < . Take n0 2 N such that n > n0 ) jan x0 j < ı. Then, for
n > n0 we have 0 < jan x0 j < ı (since an ¤ x0 for every n 1), so that
1
jf .an / Lj < . For the second part, compute f .an / and f .bn /, where an D 2n
1
and bn D 2nC=2 for n 1.
11. For the first case, if limx!C1 .f .x/ .ax C b// D 0, then
f .x/ .ax C b/ f .x/ b f .x/
0 D lim D lim a D lim a:
x!C1 x x!C1 x x x!C1 x
and
The second case (i.e., that of f W .1; A/ ! R) is entirely analogous the the
first one, so will be left to the reader.
12. In item (a), if y D ˛x C ˇ is an oblique asymptote when x ! C1, then
˛ D limx!C1 f .x/ x
D lim x!C1 1 C 1
x 2 D 1 and ˇ D limx!C1 .f .x/˛x/ D
limx!C1 1x D 0. Analogously, y D x is an oblique asymptote when x ! 1.
In item (b), if y D ˛x C ˇ is anqoblique asymptote when x ! C1, then
f .x/ a2
˛ D limx!C1 x D limx!C1 b
a 1 x2
D b
a and
b p 2
ˇ D lim .f .x/ ˛x/ D lim x a2 x
x!C1 x!C1 a
2
b a
D lim p D 0:
x!C1 a x2 a 2 C x
r q
q p p 1 C p1x
p p xC xC xx
xC xC x xD q p D r q :
p p
xC xC xC x 1 1
1C x C p 3 C1
x
f .nx/
14. We can suppose n > 2, so that > ff.2x/
f .x/ .x/
. On the other hand, if k 2 N is such
f .nx/ f .2k x/ Qk1 f .2jC1 x/
that 2k > n, then f .x/
< f .x/
D jD1 f .2j x/ . Hence,
f .nx/ Y f .2jC1 x/
k1
f .2x/
< < ;
f .x/ f .x/ jD1
f .2j x/
Section 9.3
g.x/g.x0 / f .x/f .x0 / h.x/h.x0 /
1. For item (a), observe that xx0 D c xx0 . For item (b), xx0 D
f .x/f .x0 /
xx0 .
g.x/g. c0 /
x
f .cx/f .x0 /
2. For x in the domain of g, we have x
x c0
D c cxx0
. Now, making
y D y.x/ D cx, we have y ! x0 if x ! c , so that
x0
g.x/ g xc0 f .y/ f .x0 /
limx D c lim D cf 0 .x0 /:
x! 0
c
x xc0 y!x0 y x0
3. Make induction on k.
4. For arcsin, take x 2 .1; 1/ and let y 2 . 2 ; 2 / be such that sin y D x. Since
sin0 y D cos y ¤ 0, it follows from Theorem 9.28 (with the roles of x and y
changed) that arcsin is differentiable at x, with
1 1 1 1
arcsin0 x D D Dp Dp :
sin0 y cos y 2
1 sin y 1 x2
The reasoning for arccos is completely analogous and will be left to the reader.
5. Use the results of the three previous problems.
6. For the second part of (a), note that f 0 .1/ D n and g0 .1/ D m, so that f 0 .1/ D
g0 .1/ only if m D n. For item (b), we have f .x0 / D f 1 .x0 / D y0 and f 0 .x0 / D
.f 1 /0 .x0 / (since the tangent lines to the graphs of f and f 1 at A must coincide).
Now, apply Theorem 9.28 to conclude that f 0 .x0 /f 0 .y0 / D 1.
7. Express Newton’ quotient of x 7! f .x/ at x0 in terms of Newton’s quotient of f
at x0 .
8. Start by writing
f .bn / f .an / f .bn / f .x0 / bn x0 f .x0 / f .an / x0 an
D C :
bn an b n x0 bn an x0 a n bn an
Then, use the triangle inequality to get
ˇ ˇ ˇ ˇ
ˇ f .bn / f .an / ˇ ˇ f .bn / f .x0 / ˇ b n x0
ˇ ˇ
Lˇ ˇˇ Lˇˇ C
ˇ b a b n x0 bn an
n n
ˇ ˇ
ˇ f .x0 / f .an / ˇ x0 a n
Cˇ ˇ Lˇˇ
x0 a n bn an
ˇ ˇ ˇ ˇ
ˇ f .bn / f .x0 / ˇ ˇ f .x0 / f .an / ˇ
ˇˇ Lˇˇ C ˇˇ Lˇˇ :
b n x0 x0 a n
B Hints and Solutions 601
9. For item (a), note that the tangent line to the graph of f at .an ; f .an // has
equation yf .an / D f 0 .an /.xan /. For (b), make f .x/ Dx2 a and use the result
a2n a 1
of item (b) of Problem 1 to conclude that anC1 D an 2a n
D 2
a n C a
an
.
10. For item (b), write Newton’s quotient of f in x0 as
p q
R2 x2 R2 x20 1 .R2 x2 / .R2 x20 /
D p q
x x0 x x0 R 2 x2 C R 2 x 2 0
x C x0
D p q I
R2 x2 C R2 x20
then, make x ! x0 . For (c), let r be the straight line that passes through A and
is perpendicular to OA. Elementary Analytic Geometry (cf. Chap. 6 of [4], for
!
instance) teaches us that the slope of r equals m1 , where m is the slope of OA .
Since m D f .xx00 / , it follows that the slope of r equals f .xx00 / . Now, (b) gives
f .xx00 / D f 0 .x0 /, as we wished to show.
11. For the first part, since .x0 ; y0 / belongs to the graph of f , we must have y0 D
f .x0 /. Now, since r is tangent to the graph of f at .x0 ; y0 /, its slope must be equal
to f 0 .x0 /; on the other hand, since r passes through .a; b/, its slope must be also
equal to yx00 b a . For what is left to do, if a straight line passes through .1; 1/
and is tangent to the given parabola at .x0 ; y0 /, it follows from the first part and
from Problem 1 that
(
x2
y0 D 40 :
y0 .1/ D x20 .x0 1/
x2 p p
Hence, 40 C 1 D x20 .x0 1/, so that x0 D 1 ˙ 5 and y0 D 3˙2 5 .
12. Choose a cartesian coordinate system such that F.0; p2 / and d is the straight
line y D p2 . Since Q.x; y/ 2 P if and only if FQ D dist.QI d/, a simple
x2
computation gives y D 2p as the equation of the parabola. Now, letting
2
P.x0 ; p2 / and Q.x; 2p
x
/, use the result of the previous problem to conclude that
!
PQ is tangent to P if and only if
x2
d 1 2 2p
p2
x D
dx 2p x x0
or, which is the same (cf. Problem 1), x2 2x0 x p2 D 0. Letting ˛ and ˇ
2 2
be the roots of this second degree equation, we have A.˛; ˛2p / e B.ˇ; ˇ2p /. On
˛2
! . /
p
2 p2
the other hand, the slope of AF is m D 2p ˛0 2 D ˛ 2p˛ ; however, since
602 B Hints and Solutions
2x0 ˛
˛ 2 2x0 ˛ p2 D 0, it comes that m D 2p˛
D x0
p
. Analogously, the slope of
! ! !
BF also equals xp0 , so that AF D BF , and A, B and F are collinear.
p p
13. Since k … Q, we have f . k/ D 0. Now, letting x be an irrational number
p p
f .x/f . k/
different from k, we have x k D 0. It thus suffices to show that
p
p
f .x/ f . k/
lim
p
p D 0:
x! k x k
x2Q
A D fp=qI p; q 2 N; 0 < q ng :
Section 9.4
3. Apply the differentiation formulas for the sums, product and quotient of two
functions, together with the derivative computations of this and the last section.
For items (e) and (f), start by reviewingp the results of Problem 4, page
p 300.
4. If r D mn , with m; n 2 Z , then f .x/ D n xm . For x > 0, let g.x/ D n x, so that
g is differentiable (by Example 9.29). Since x 7! xm is also differentiable (for
x ¤ 0 if m < 0) and f .x/ D g.xm /, the chain rule guarantees the differentiability
of f , with f 0 .x/ D g0 .xm / mxm1 . Then, again by Example 9.29, we have
1 m
.xm /1=n1 mxm1 D xm. n 1/Cm1 D r x n 1 D rxr1 :
1 m
f 0 .x/ D
n n
5. For the first part, differentiate both sides of (9.22) and, then, divide both sides
of the resulting equality by f .x/. For the second part, differentiate both sides of
the equality f .x/ D a.x ˛1 /.x ˛2 / : : : .x ˛n / and, then, divide both sides
of the resulting equality by f .x/.
6. Straightforward computations give f 0 .x/ D 7.x 3/2 , f 00 .x/ D 14.x 3/3 ,
f .3/ .x/ D 42.x 3/4 , f .4/ .x/ D 168.x 3/5 and f .5/ .x/ D 840.x 3/6 .
Hence, f .5/ .0/ D 840 36 D 840 .
p 729
7. Set f1 .x/ D x C 1 and, for an integer n 2,
s r q
p
fn .x/ D xC x CC xC x C 1;
with n square root signs. For n 2, we have fn .x/2 D x C fn1 .x/. Hence,
it follows from the chain rule that 2fn .0/fn0 .0/ D 1 C fn10
.0/ or, which is the
same, 2fn .0/ D 1 C fn1 .0/ (since fn .0/ D 1). Now, since f10 .0/ D 12 , we can use
0 0
x2 y2 x2
10. For item (b), solve the system formed by the equations a2
C b2
D 1 and a02
y2
b02
D 1, concluding from (a) that it has exactly four distinct solutions. For
(c), analyse the case in which P.x0 ; y0 /, with x0 ; y0 > 0 (the remaining cases
can be dealt with in analogous ways). Show (using either the chain rule, implicit
differentiation or directly the result of Problem 2) that the slopes of the tangents
2 02
to E and H at P are respectively equal to ba2 xy00 and ba02 xy00 . Finally, conclude
2 02 x2
that it suffices to show that ba2 ba02 y02 D 1, and use the result of (b) to verify that
0
such an equality is indeed true.
11. For item (b), take g to be equal to the sum of the functions x 7! f .x xk /, where
f is as in item (a).
12. Adapt, to the present case, the discussion of Example 9.41.
13. Firstly, observe that a1 C 2a2 C C nan D f 0 .0/. Then, use the fact that
jf .x/j j sin xj, together with f .0/ D 0 and the fundamental trigonometric
limit, to get jf 0 .0/j 1.
14. Letting p > 0 be the period of f , show that f .k/ .p/ D f .k/ .0/ for every integer
k 0. Then, looking at these n equalities as a linear system of equations in the
unknowns cos.jp/, 1 j n, conclude that its only solution is cos.jp/ D 1 for
1 j n.
15. Firstly, verify that, for x 2 .a; b/, the discussion that precedes Lemma 9.36
allows us to write
r.x/ s.x/
with limx!a xa
and limx!a xa
D 0.
Section 9.5
xf .x/ d
.xf .x//
lim f .x/ D lim D lim dx
d
D lim .f .x/ C xf 0 .x//
x!C1 x!C1 x x!C1
dx x
x!C1
Finally, use the fact that g.x/ ! C1 as x ! a to find 0 < ı < such that, for
a < x < a C ı,
ˇ ˇ ˇ ˇ
ˇ ˇ ˇ ˇ
ˇ L g.x0 / f .x0 / ˇ ; ˇ L C g.x0 / f .x0 / ˇ < :
ˇ 2 g.x/ g.x/ ˇ ˇ 2 g.x/ g.x/ ˇ 2
16. Use Lagrange’s MVT, together with the result of Problem 11, page 263.
17. Take a sequence .xn /n1 in I n fx0 g such that xn ! x0 . If xn > x0 (resp. xn < x0 ),
then, applying Lagrange’s MVT to the interval Œx0 ; xn (resp. Œxn ; x0 ), conclude
that there exists yn 2 .x0 ; xn / (resp. yn 2 .xn ; x0 /) such that
f .xn / f .x0 /
f 0 .yn / D : (B.5)
x n x0
f .xn / f .x0 /
x n ! x0 ) ! L:
xn x 0
Argue by contradiction to show that this suffices to prove that limx!x0 f .x/f
xx0
.x0 /
Section 9.6
1. Consider the quadratic f .x/ D ax2 C bx C c, with a > 0 (the case a < 0 is
completely analogous). Since f 0 .x/ D 2ax C b and a > 0, we have f 0 .x/ < 0
B Hints and Solutions 607
product of the slopes of two perpendicular lines, none of which being vertical,
equals 1.
!
7. Just notice that, from the previous problem, AB is perpendicular to the tangents
to the graph of f at A and to the graph of g at B.
608 B Hints and Solutions
2
8. If A.a; a2 / and B.b; 1 .b 3/2 /, then AB D .a b/2 C .a2 1 C .b 3/2 /2 .
2
Condition f 0 .a/ D g0 .b/ gives us b D 3 a, so that AB D 4a4 12a C
4
10; hence, we are left to p minimize h.x/ D 4x 12x C 10. The test of the
first derivative gives a D q3 3=4 as the only minimum point of h, and an easy
p
computation gives AB D 10 9 3 3=4.
9. Let AH D h, BH D l, Hb AB D ˛ and Bb C D
. Apply some Trigonometry to
tan.˛C
/tan ˛
get tan
D 1Ctan.˛C
/tan ˛ D h22hl
C3l2 . Now, if f .l/ D h22hl
C3l2
for l > 0, compute
2 2
f 0 .l/ D 2h.h 3l /
.h2 C3l2 /2
and conclude that the critical point l D ph3 of f is a point of
global maximum.
10. Let
D Bb AD D Ab DC, so that 0 <
< 2 , AD D a C 2b cos
and h D
b sin
, where h is the height of ABCD. Hence, A.ABCD/ D 12 . AD C BC/h D
.a C b cos
/b sin
, and it suffices to maximize f W .0; 2 / ! R given by
f .
/ D .a C b cos
/ sin
. To what is left to do, use some Trigonometry to
get f 0 .
/ D 2b cos2
C a cos
b and f 00 .
/ D 2b sin.2
/ ab sin
. Now,
check that f 00 .
/ < 0 for every
2 .0; 2 / and that
D arccos p 2b
aC a2 C8b2 Ca
is the only critical point of f . Finally, apply Corollary 9.60 to f .
2
OP sin ˛ sin ˇ
11. In the notations of Example 9.61, we have PX PY D sin
sin.˛CˇC
/ . Since
2
OP sin ˛ sin ˇ is constant, we conclude that PX PY is minimum if and only
if f .
/ D sin
sin.˛ C ˇ C
/ is maximum, where f W .0; ˛ ˇ/ ! R. In
turn, the first derivative test assures that if a global maximum point for f exists
at all, it should be a critical point of f . However, since
f 0 .
/ D cos
sin.˛ C ˇ C
/ C sin
cos.˛ C ˇ C
/ D sin.˛ C ˇ C 2
/;
it follows that
is critical if and only if ˛ C ˇ C 2
D or, which is the
same, if and only if
D ˛ ˇ
, i.e., if and only if OXY is isosceles
with basis XY. It now suffices to show that, OXY being isosceles with basis
XY (or, which is the same, letting
D 12 . ˛ ˇ/ – which is the only
critical point of f ), then f does attain its maximum possible value. To this end,
just note that if 0 <
< 12 . ˛ ˇ/ then 0 < ˛ C ˇ C 2
< , so that
f 0 .
/ D sin.˛CˇC2
/ > 0; on the other hand, if 12 . ˛ˇ/ <
< ˛ˇ,
then < ˛ C ˇ C 2
< 2, so that f 0 .
/ D sin.˛ C ˇ C 2
/ < 0. Therefore,
it follows from Proposition 9.57 that f is increasing in .0; 12 . ˛ ˇ// and
decreasing in . 12 . ˛ ˇ/; ˛ ˇ/, so that
D 12 . ˛ ˇ/ is, indeed,
the only global maximum point of f .
12. In the notations of Fig. B.6, we shall prove that the position of P along r that
minimizes AP C BP is such that ˛ D ˇ.
To this end, start by observing that, had we drawn the point P to the left of
A, then we would have ˛ > 2 ; accordingly, had we drawn P to the right of B,
we would have ˇ > 2 . Therefore, ˛; ˇ 2 .0; /.
B Hints and Solutions 609
B
a
b
α β
r
A P B
p
If P D A0 , Pythagoras’ theorem gives 2 2
p AP C PB D a C0 b0 C c ; if P D B ,
0
a b
C D c; (B.6)
tan ˛ tan ˇ
values of AP C PB when P D A0 or B0 ).
13. Fix a cartesian system for which A.0; a/ and B.l; b/, and let P.x; 0/, with x 2 R.
Use basic Analytic Geometry to show that it suffices to prove that f W R ! R,
given by f .x/ D .x2 C a2 /..x l/2 C b2 /, has a single minimum global point,
610 B Hints and Solutions
p
provided l 2.a2 C b2 /. Compute f .x/ D x4 2lx3 C .a2 C b2 C l2 /x2
2a2 lx C a2 .b2 C l2 /, f 0 .x/ D 4x3 6lx2 C 2.a2 C b2 C l2 /x 2a2 l and f 00 .x/ D
12x2 12lx C 2.a2 C b2 C l2 /. Then, show that f 0 has a root in the interval .0; l/
and examine under what conditions we have f 00 .x/ 0 for every x 2 R.
Conclude that it suffices to show that f .x/ D .xC1/
nC1
14. xn , x > 0, attains its
minimum value at x D n. To this end, study the first variation of f .
p p
15. Firstly, show that y D 12 k. 1 C x2 x/ C 1k . 1 C x2 C x/ . Then, study the
first variation of f W R ! R given by f .x/ D x C y.
16. Use the inequality between the arithmetic and geometric means of two reals
to show that the given expression is greater that or equal to f .x C y/, where
f .t/ D t C 2t C t22 for t > 0. Then, study the first variation of f .
17. Letting ˛ D sin b A
2
and ˇ D sin b B
, show that it suffices to analyse the equality
2
23
f .˛/ D f .ˇ/, where f .x/ D .1x x
2 /24 , for x 2 .0; 1/. To this end, study the first
variation of f .
18. Use the MVT to guarantee the existence of c 2 .a; b/ such that f 0 .c/ D 1. If
f .c/ D c, it suffices to apply the MVT twice. If f .c/ < c, use the MVT to assure
the existence of ˛ 2 .0; c/ and ˇ 2 .c; 1/ such that f 0 .˛/ < 1 and f 0 .ˇ/ > 1;
then, apply Darboux’s theorem. If f .c/ > c, argue analogously.
19. Write f .x/ f 0 .x/ f 00 .x/ C f 000 .x/ D .f .x/ f 0 .x// dxd
.f 0 .x/ f 00 .x//. Then,
apply the following claim twice: if g is a polynomial function of positive degree,
such that g.x/ g0 .x/ 0 for every x 2 R, then g.x/ 0 for every x 2 R. In
order to prove this claim, start by using the result of Example 9.17 to show that
g has even degree and positive leading coefficient. Then, take (by the result of
Problem 10, page 263) x0 2 R such that g attains its minimum at x0 and apply
the first derivative test.
Section 9.7
1. For a < b in I and 0 < t < 1, it follows from the strictly convex character of g
that g..1 t/a C tb/ < .1 t/g.a/ C tg.b/. Now, since f is increasing, we have
.f ı g/..1 t/a C tb/ D f .g..1 t/a C tb// < f ..1 t/g.a/ C tg.b//. Finally,
use the fact that f is also strictly convex to show that the same is true of f ı g.
2. Firstly, Theorem 8.35 shows that f is either increasing or decreasing. Suppose
that f is increasing (the other case in totally analogous). For a < b in I and
0 < t < 1, we have f ..1 t/a C tb/ < .1 t/f .a/ C tf .b/. Let ˛ D f .a/ and
ˇ D f .b/, so that a D f 1 .˛/ and b D f 1 .ˇ/. Applying f 1 to both sides of
the above inequality and invoking the result of Problem 8, page 176, show that
.1 t/f 1 .˛/ C tf 1 .ˇ/ < f 1 ..1 t/˛ C tˇ/. From this, conclude that f 1 is
strictly concave.
3. Use the definition of strictly convex (resp. concave) function, along the lines of
the hints given to the two previous problems.
B Hints and Solutions 611
.x1/C1
p
4. Write f .x/ D p D 1x C p1 and show that each one of the
p1x 1
1x
functions t 7! t and t 7! p
t
, t 2 .0; 1/, is strictly convex. Then, use the
4x
result of the previous problem. Alternatively, compute f 00 .x/ D 4.1x/ 5=2 and
00 00
solve the inequality f .x/ > 0 (resp. f .x/ < 0).
5. Since f 00 .x/ D 6x C 2a, we have f 00 > 0 in . a3 ; C1/ and f 00 < 0 in .1; a3 /.
Hence, f is strictly convex in . a3 ; C1/ and strictly concave in .1; a3 /, so
that x D a3 is its only inflection point.
6. Computing derivative, we get f 00 .x/ D .2 x12 / sin 1x 2x cos 1x . Now, if n 2 N
1
and xn D n , then x1n D n, so that f 00 .xn / D 2n cos.n/ D .1/nC1 2n.
Therefore, f .xk /f 00 .xkC1 / < 0, and the IVT guarantees the existence of yn 2
00
.xnC1 ; xn / such that f 00 .yn / D 0 (notice that we need not invoke Darboux’s
Theorem 9.62, since f 00 is continuous in .0; p12 /). Verify that yn is an inflection
point of f .
7. Note that g is also twice differentiable, with g00 .x/ D 2f 0 .x/Cxf 00 .x/ 0. Then,
apply item (a) of Theorem 9.69.
8. Suppose that f is convex. Give x; y 2 I, let us first show that f ..1 t/x C ty//
.1 t/f .x/ C tf .y/ (*) for every dyadic rational number t 2 Œ0; 1 (cf. Problem 5,
page 206); to this end, let’s make induction on k 1. For k D 1, we have the
validity of (*) for t D 0, 12 and 1. For t D 34 , applying the convexity of f twice
we get
!
f xCy
C f .y/
x C 3y
xCy
2
Cy 2
f Df
4 2 2
f .x/Cf .y/
2
C f .y/ 1 3
D f .x/ C f .y/;
2 4 4
sCu sCu
f ..1 t/x C ty/ D f 1 xC y
2 2
..1 s/x C sy/ C ..1 u/x C uy/
Df
2
f ..1 s/x C sy/ C f ..1 u/x C uy/
2
1
Œ..1 s/f .x/ C sf .y// C ..1 u/f .x/ C uf .y//
2
sCu sCu
D 1 f .x/ C f .y/
2 2
D .1 t/f .x/ C tf .y/:
Now, since f is convex and twice differentiable, it follows from Corollary 9.70
that f 00 .d/ 0. Then, taking into account that .c x0 /.x x0 / > 0, the above
computations give
13. Suppose, to the contrary, that there exist natural numbers n1 < n2 < n3 <
such that .f .nk //k1 is an arithmetic progression. Use the strict convexity and
the increasing character of f to successively get, for k > 1,
nk1 C nkC1
f .nk / > f
2
Section 9.8
1. Since sin x D cos. 2 x/, we have csc x D sec. 2 x/. Now, use the discussion
of Example 9.79, together with the results of Problem 10, page 194.
614 B Hints and Solutions
2. We sketch hints to some of the items, letting the other ones to the reader. For
item (b), compute f 0 .x/ D 1 x12 and f 00 .x/ D x23 to conclude that f decreases
in .1; 1/, increases in each of .1; 1/ and .1; C1/, is strictly convex in
.0; C1/ and strictly concave in .1; 0/. Also, use the result of Problem 11,
page 291, to conclude that y D x is the only oblique asymptote of f . For item
2 2x.x2 3/
(f), compute f 0 .x/ D .x1x 00
2 C1/2 and f .x/ D .x2 C1/3 to conclude that x D ˙1 are
p
the only critical points of f , while x D 0 and x D ˙ 3 are its only inflection
points. Also, f 0 .x/ > 0 if and only if jxj < 1, so that f increases along the
interval .1; 1/ and decreases along each of .1; 1/ and .1; C1/; therefore,
x D 1 is a local minimum point and x D 1 is a local maximum point of f .
However, since f .1/ D 12 , f .1/ D 12 and limjxj!C1 f .x/ D 0, it follows that
x D 1 is the only global minimum point of f , while x D 1 is its only global
maximum point. Now, at the inflection point x D 0, the tangent to the graph is
(trigonometrically) at 4 from the horizontal axis, while at the inflection points
p
x D ˙ 3 such an angle equals arctan 18 . Note that f 00 .x/ > 0 if and only
p p
if x 2 . 3; 0/ [ . 3; C1/; therefore, f is strictly convex in each of these p
pAccordingly, f is strictly concave in each of the intervals .1; 3/
intervals.
and .0; 3/. Finally, the horizontal axis is the only horizontal asymptote of the
4 C1 12x3 .x4 C1/
graph. For item (h), compute f 0 .x/ D .x3x4 1/ 00
2 and f .x/ D .x4 1/3
to conclude
that f decreases along all of its domain, is strictly convex in each of the intervals
.1; 0/ and .1; C1/, and strictly concave in each of .1; 1/ and .0; 1/. Also,
x D 0 is its only inflection point, in which the tangent to the graph makes a
trigonometric angle of 4 with the horizontal axis. Finally, this axis is the only
horizontal asymptote of f , with limx!C1 f .x/ D limx!1 f .x/ D 0; lines x D
1 and x D 1 are the only vertical asymptotes, with limx!.1/ f .x/ D 1,
limx!.1/C f .x/ D C1, limx!1 f .x/ D 1 and limx!1C f .x/ D C1.
3. First of all, note that since f 0 .x0 / D g0 .x0 /, line r is tangent to the graph of g
!
(hence, to ) at A; thus, OA ?r. For item (a), if f 0 .x0 / D 0, then r is horizontal
!
and, then, OA is vertical. If f 0 .x0 / ¤ 0, basic Analytic Geometry assures that
!
the slope of OA equals f 0 .x1 0 / . For (c), it follows from .x ˛/2 C .g.x/
ˇ/2 D R2 and the chain rule that 2.x ˛/ C 2.g.x/ ˇ/g0 .x/ D 0; therefore,
2 C 2g0 .x/2 C 2.g.x/ ˇ/g00 .x/ D 0. Now, substitute x D x0 in the first equality
and use the result of the second part of (a) to get the first equality of (c); then,
substitute x D x0 in the second equality and use the fact that g.x0 / D f .x0 /
and g0 .x0 / D f 0 .x0 / to get the second equality of (c). For item (e), observe that
R2 D .x0 ˛/2 C .f .x0 / ˇ/2 and substitute the values of ˛ and ˇ computed
at (d). Finally, for the second part of (f), notice that condition f 0 .x0 / D g0 .x0 /
!
guarantees that r? AO.
4. For item (a), note that f is even. For (b), compute f 0 .x/ D sin 1x 1x cos 1x . For (c),
observe that sin y y cos y D 0 implies cos y ¤ 0 and, then, tan y D y. Now, use
the IVT (sketching the graphs of y 7! tan y and y 7! y may help) to guarantee
B Hints and Solutions 615
that the set of reals y > 0 such that tan y D y can be put in a sequence .yn /n1 ,
satisfying the given conditions. For item (d), note that
1.0
0.5
−3 −2 −1 1 2 3
−0.5
1
Fig. B.7 Sketching the graph of x 7! x sin x
616 B Hints and Solutions
Section 10.2
1. Adapt the proof of S.f I Q/ S.f I P/ to this case. You shall need to use the fact
that infŒxi1 ;x0 f ; infŒx0 ;xi f infŒxi1 ;xi f .
2. Take a partition P of Œa; b and show that s.f I P/ D S.f I P/ D c.b a/.
3. Read the proof of Theorem 10.5 again to check that lim .S.f I Pk /
k!C1
s.f I Pk // D 0. Then, use the fact that f is integrable, together with the
Rb
inequalities s.f I Pk / a f .x/dx S.f I Pk /, to show that the limits
Rb
lim S.f I Pk / and lim s.f I Pk // do exist and are equal to a f .x/dx.
k!C1 k!C1
4. Letting Pk D fa D x0 < x1 < < xk D bg be a uniform partition, we
have A.f I k/ D s.f I Pk / and A.f I k/ D S.f I Pk /, so that (10.6) gives s.f I Pk / <
nC1
bnC1
nC1
1 1k < S.f I Pk /. Now, make k ! C1.
5. For item (a), substitute u D x2 , v D y2 and write down the expressions involved
in terms of x and y. For item (b), let u D xj and v D xj1 in (a) and add the
inequalities thus obtained, observing that xj xj1 D ba k
. Finally, for (c) let
k ! C1 in the inequalities of (b) and use the result of Problema 3.
6. For 1 k n, write p1n pkCn 1
D 1n p 1k . Then, if f .x/ D pxC1 1
and
n C1
1 2
Pn D f0; n ; n ; : : : ; n ; 1g,
n1
note that an D s.f I Pn /. Finally, use the result of
Problem 3.
1
7. For 1 k n, let n n2 Ck 2 D 1n 1 1
2 . Then, take f .x/ D 1Cx2 ,
1C. nk /
Pn D f0; 1n ; 2n ; : : : ; n1
n
; 1g and note that an D s.f I Pn /. Finally, use the result of
Problem 3.
8. For the first part, use the given inequality to show that Cauchy’s integrability
criterion is satisfied. For the second part, conclude (also from the given
inequality) that limk!C1 s.f I Pk / and limk!C1 S.f I Pk / exist and are equal.
Rb
Then, use the fact that s.f I Pk / a f .x/dx S.f I Pk / to finish the proof.
9. For item (a), adapt (10.17) to the present case, noticing that, if Mj D f .x0j / and
mj D f .x00j /, then Mj mj D f .x0j / f .x00j / cjx0j x00j j c.xj xj1 / D c.ba/
k .
Rb
For item (b), start by showing that †.f I Pk I / and a f .x/dx both belong to the
closed interval with endpoints s.f I Pk / and S.f I Pk /.
10. For the first part of item (a), note that 2 sin.a C jh/ sin h2 D cos.a C .j 1/h/
cos.a C jh/, add such equalities from j D 0 to j D k and transform the result
thus obtained in product. For the second part, just note that cos x D sin 2 x .
For the first part of (b), Example 8.8 shows that the sine function is Lipschitz,
with Lipschitz constant c D 1. Let Pk D fa D x0 < x1 < < xk D bg be
a uniform partition of Œa; b and (in the notations of the previous problem) take
kj D xj for 1 j k. Letting h D ba k D xj xj1 for 1 j k, note that
.k1/h
X
k h sin a C 2 sin .kC1/h
2
†.sinI Pk I k / D h sin.a C jh/ D h sin a:
jD1
sin h2
B Hints and Solutions 617
Then, apply the fundamental trigonometric limit, together with the result of
item (b) of the previous problem, to conclude that
0 1
Z b h sin a C .k1/h
sin .kC1/h
2 2
sin x dx D lim @ h sin aA
a h!0 sin h2
" #
h
2 bCa h ba h
D 2 lim sin sin C
h!0 sin h
2
2 2 2 2
b C a b a
D 2 sin sin D cos a cos b:
2 2
For the second part, adapt the above argument.
11. Firstly, note that, by the density of the irrationals in the real line, we have
s.f I P/ D 0 for every partition P. Now, give > 0, fix n0 2 N such that
n0 > 2 . Then, if A D fx 2 Œ0; 1I x D mn ; with m; n 2 N and n < n0 g, show
that A is finite and use this fact to guarantee the existence of a partition P0 of
Œ0; 1, such that the sum of the lengths of the intervals of P0 which contain some
element of A is less than 2 . Finally, show that S.f I P0 / < .
Section 10.3
Then, fix s 2 Œ0; 1 and let t D g.s/ in the above inequality to get f .g.s// g.s/
R1
0 f .x/dx. Finally, use the fact that s 2 Œ0; 1 was chosen arbitrarily to conclude
that
Z 1 Z 1
.f .g.s// g.s//ds f .x/dx:
0 0
X
k
2
jf .j / f .j /j.xj xj1 / < : (B.7)
jD1
3
Now, let Mj D supŒxj1 ;xj f , mj D infŒxj1 ;xj f and choose sequences .jn /n1 and
n n
.jn /n1 in Œxj1 ; xj satisfying f .jn / ! Mj and f .jn / ! mj . Make j D jn
P
and j D jn in (B.7) to show that kjD1 .Mj mj /.xj xj1 / 23 .
Section 10.4
1. Modify the function of item (a) of Problem 11, page 311, taking, for instance
1
f .x/ D .x a/4=3 sin xa , if x 2 .a; b.
B Hints and Solutions 619
Z b k Z
X xj
f .x/dx D f .x/dx:
a jD1 xj1
R xj
Now, observe that, in this last expression, the sum of the summands xj1 f .x/dx
for which f > 0 in .xj1 ; xj / corresponds to A.RC
f /,
whereas the sum of those
summands for which f < 0 in .xj1 ; xj / corresponds to A.R f /.
3. The hypothesis gives g1 .x/ f1 .x/ D g2 .x/ f2 .x/ for every x 2 Œa; b.
4. Use Proposition 10.19, together with Problem 2, page 357, and Lemma 10.17.
5. For item (a) and for j 2 ZC , let fj W Œj; j C 1 ! R be the restriction of the
fractional part function to the interval Œj; j C 1. If P D f0 D x0 < x1 < <
xk D 1g is a partition of the interval Œ0; 1 and yj D xj C n for 0 j k (so that
Q D fn D y0 < y1 < < yk D n C 1g is a partition of Œn; n C 1), show that
s.f0 I P/ D s.fn I Q/ and S.f0 I P/ D S.fn I Q/. For item (b), use Proposition 10.19,
together with Problem 4, page 358, and Lemma 10.17.
6. First of all, show that it suffices to consider the case f D 0. In this case, if
there exists x0 2 Œa; b such that g.x0 / > 0, use Lemma 8.20 to show that
Rb
a g.x/dx > 0. Rb Rb Rb
7. Conclude first that a jf .x/jdx a f .x/dx a jf .x/jdx. Then, finish the
proof.
8. Adapt, to this more general case, the approach to the case of f D g in Œa; b/, as
presented in the text.
9. For the first part, use the result of Problem 4, page 358, together with
Proposition 10.21. For the second part, use the result of Proposition 10.13.
11. If M > 0 is such that jf .x/j M for every x 2 Œa; b, then
ˇZ Z ˇ ˇZ aC Z b ˇ
ˇ b b ˇ ˇ ˇ
ˇ f .x/dx f .x/dxˇˇ D ˇˇ f .x/dx C f .x/dxˇˇ
ˇ
a aC a b
ˇZ aC ˇ ˇZ b ˇ
ˇ ˇ ˇ ˇ
ˇˇ f .x/dxˇˇ C ˇˇ f .x/dxˇˇ
a b
Z aC Z b
jf .x/jdx C jf .x/jdx
a b
Z aC Z b
Mdx C Mdx D 2M:
a b
R R b
b
Therefore, lim!0 a f .x/dx aC f .x/dx D 0.
Rb
12. From a .f .x/ tg.x//2 dx 0 for every t 2 R, conclude that At2 2Bt C
Rb Rb
C 0 for every t 2 R, where A D a g.x/2 dx, B D a f .x/g.x/dx and C D
Rb 2 2
a f .x/ dx. Then, note that D 4.B AC/ 0. For the last part, use the
result of Problem 6.
620 B Hints and Solutions
13. First consider the case in which f is positive, and look at both f .a/Cf
2
.b/
.b a/
Rb
and a f .x/dx as the areas of certain regions in the cartesian plane. For the
general case, note that changing f by f C c amounts to adding c to both sides of
the inequality in the statement of the problem.
Section 10.5
1. For item (a), use that cos2 t D 12 .1 C cos.2t//; for (b), write cos t D 1 2 sin2 2t ;
for item (c), integrate by parts.
2. For item (a), integrate by parts; for (b), integrate by parts twice.
3. For the first integral, if m ¤ n write sin.mx/ sin.nx/ D 12 .cos.mn/xcos.mC
n/x/; if m D n, use the fact that sin2 nx D 12 .1 cos.2nx//. For the second
integral, argue in an analogous way.
4. In both cases, use p (10.38). Additionally, for item (a) note that x 7! 23 x3=2 is a
primitive for x 7! x in Œ0; C1/; for (b), note that x 7! tan x is a primitive for
x 7! sec2 x in . 2 ; 2 /.
5. Start by taking a primitive F for f and computing the given integral with the aid
of the FTC.R
6. If F.x/ D x f .t/dt, use the result of the previous problem to show that F 0
xCp
vanishes identically.
Rb
7. Let f .x/ D mx C n, compute a .mxn /dx with the aid of the FTC and, then,
check that the result coincides with the ordinary formulas for the area of a right
triangle (if f vanishes at x D a or x D b) or for the area of a right trapezoid (if
f is positive in Œa; b).
R y F be one such primitive and a < x < y < b, use the interpretation
8. Letting
of x f .t/ dt
as an area and the
monotonicity of f to compare the differences
F.y/ F xCy 2 and F xCy
2 F.x/. Then, apply the result of Problems 8
and 9, page 338. P Rx
9. Let f .x/ D .x C 1/n D nkD0 nk xk , x 2 R. Compute 0 f .t/dt in two different
ways to get
!
Xn
n xkC1 .x C 1/nC1 1
D :
kD0
k kC1 nC1
Then, apply Cauchy inequality for integrals (cf. Problem 12, page 381) to x 7!
xf .x/ and f 0 .
12. Firstly, use the computations in Example 10.29 to show that there exists a single
f0 2 F of the form f0 .x/ DRA cos x C B sin x, with A; B 2 R. Then, expand the
integrand in the inequality 0 .f .x/ f0 .x//2 dx 0 to get the desired result.
13. Since 0 cos x 1 for x 2 Œ0; 2 , we have .cos x/nC1 .cos x/n for every
R =2
InC1 .cos x/nC1 dx
integer n 0. Hence, In D 0
R =2 1 for every n 0. On the
0 .cos x/n dx
other hand, we saw in Example 10.35 that InC1 D n
nC1 In1 for every n 1.
n
InC1 nC1 In1 n2
Hence, 1 D In D n1 n2 1
In1
In2 In1
In2 , so that each one of the
n In2
sequences II2k1
2k2
I2k
and I2k1 is nondecreasing and bounded from above,
k1 k1
thus convergent. Letting `0 and `1 respectively denote their limits, we have
n 1
I nC1 In1
`0 ; `1 > 0. Now, again by (a), observe that nC1
In D In D nC1
n In
In1 .
Therefore, letting n D 2k ! C1 we get `1 D `1 0 or, which is the same,
InC1
`0 `1 D 1. However, since In 1, it easily follows that `0 ; `1 1. Thus,
1 D `0 `1 1 1 D 1,which gives us `0 D `1 D 1. Finally, since
both
sequences II2k1
2k2
and I
I2k
2k1
converge to 1, we conclude that I
In
n1
k1 k1 n1
also converges to 1.
Section 10.6
1. For item (a), use the substitution x D t 1, together with the fact that p1 t D
1=2
t1=2 D dtd t1=2 ; for (b), use the substitution x D sin t; for item (c), use the
p
substitution x D t.
2. For (a), make the substitution y D 1 C x3 ; for (b), integrate by parts and use the
result of (a); for item (c), make the substitution x D sin t and, then, integrate
by parts; finally, for (d), perform the substitution x D t2 and, then, integrate by
parts.
Ra R0 Ra
3. In both cases, write a f .x/dx D a f .x/dx C 0 f .x/dx and, for the first
integral, perform the substitution x D t.
4. For the first integral, note that the sine function is odd and apply the result of
the previous problem. For the second one, argue in a similar way.
5. Assume (without loss of generality) that O.0; 0/ is the center and A.a; 0/ and
A0 .a; 0/ are the endpoints of the major axis of the ellipse. It’s a well known fact
622 B Hints and Solutions
2 2
that, in this setting, the equation of the ellipse is given by ax2 C by2 D 1. Letting
q
2
f W Œa; a ! R be given by f .x/ D b 1 ax2 , show that the area of the ellipse
Ra
equals 2 a f .x/dx. Then, perform the trigonometric substitution x D a cos t,
with 0 t , and compute the integral.
6. Suppose that g is crescente (the case of a decreasing g in completely analogous).
Let t1 < t2 < < tk be the points of the interval Œg.c/; g.d/ in which f is
discontinuous, and c s1 < s2 < < sk d be such that g.sj / D tj for 1
j k. Let fj W Œtj ; tjC1 ! R be such that fj D f in .tj ; tjC1 /, fj .tj / D limt!tj C f .t/
and fj .tjC1 / D limt!tjC1 f .t/, so that fj is continuous. Use Lemma 10.17,
Rt Rs
together with Theorem 10.37, to get tj jC1 f .t/dt D sj jC1 f .g.s//g0 .s/ds for
R t1 Rs R g.d/
1 j < k and, analogously, g.c/ f .t/dt D c 1 f .g.s//g0 .s/ds and tk f .t/dt D
Rd 0
sk f .g.s//g .s/ds. Finally, add all such equalities to get the desired result.
7. First of all, note that f .a/ D a and f .b/ D b. Hence, integrating successively
by substitution and parts, we get
Z b Z f .b/ Z b
f 1 .x/ dx D f 1 .x/ dx D f 1 .f .x//f 0 .x/ dx
a f .a/ a
Z b ˇxDb Z b
0 ˇ
D xf .x/ dx D xf .x/ˇ f .x/ dx
a xDa a
Z b
D bf .b/ af .a/ f .x/ dx
a
Z b
D b2 a2 f .x/ dx:
a
8. Start
by observing that, since p has degree n, it is reciprocal if and only if
p 1x D p.x/
xn
for infinitely many nonzero real values of x (here we are using the
fact—to be proved in [5]—that two real polynomial functions which assume
the same values for every x 2 A, where A R is infinite, are actually
identical). Now, assume the equality of the statement holds for every x > 1.
Then, differentiating both sides with the aid of Problem 5, page 391, gives
p.x/
xn=2C1
p.1=x/
D .1=x/ n=2C1 x12 , which is the same as p 1x D p.x/
xn for every x > 1.
1
Conversely, assuming p to be reciprocal, the variable substitution t 7! t gives
Z Z Z 1
x
p.t/ 1=x
p.1=t/ 1 p.t/ n=21
dt D dt D t dt
1 t n=2C1
1 .1=t/n=2C1 t2 1=x t
n
Z 1
p.t/
D :
1=x tn=2C1
B Hints and Solutions 623
9. Since f 0 > 0, we conclude that f is increasing. Therefore, f .x/ > 1 for x > 1
and the FTC gives, also for x > 1,
Z Z
x
1 x
1
f .x/ 1 D dt < dt D arctan x < :
0 t C f .t/2
2
0 t2 C1 4 4
X Z g.x/
m1 Z g.x/
D f .t/dt D m f .t/dt:
kD0 x x
Section 10.7
f .x/ D x
5. Let log.1 C x/ and show that f is increasing. Then, write f .x/ D
x 1 log.1Cx/
x
and use L’Hôpital’s rule to get limx!0C f .x/ D 0.
p
6. It suffices to observe that, since log is increasing and xCy 2 xy for every
p log xClog y
x; y > 0, we have log 2 log xy D
xCy
2
, with equality if and only
if x D y.
7. Since f 00 .x/ D ex C6ax, we wish to find all a < 0 such that ex C6ax has a single
real root. Arguing geometrically, conclude that it suffices to find all a < 0 for
which the straight line y D 6ax is tangent to the graph of x 7! ex . To this end,
use the result of Problem 11, page 301 (with .a; b/ D .0; 0/), to concludeˇ that, if
d xˇ
.x0 ; e / is the point of tangency, then e D 6ax0 and 6a D dx e ˇ
x0 x0
D ex 0 .
xDx0
8. Recall that x˛ D e˛ log x and apply the chain rule.
9. Adapt, to the present case, the hint given to the previous problem to get h0 .x/ D
.f .x/g0 .x/ log f .x/ C f 0 .x/g.x//f .x/g.x/1 .
10. Study the first variation of f to conclude that x D aCba
is its only critical point.
Then, use Weierstrass’ Theorem 8.26 to conclude that such a point is its only
maximum point.p
11. Let a D b D 2 and analyse the possibilities ab rational and ab irrational,
observing, in this last case, that .ab /b 2 Q.
12. The continuity of the natural logarithm guarantees that it suffices to show that
a
lim x log 1 C Da
x!C1 x
1 and limx!C1 cosh x D C1. For (c), show that sinh0 x D cosh x > 0 and
sinh00 x D sinh x, with sinh x > 0 if and only if x > 0. In item (d), you should
get something similar to Fig. B.8. For item (e), letting k.x0 / denote the curvature
we wish to compute and f .x/ D cosh x, conclude from Problem 3, page 344,
1
that k.x0 / D .1Csinh 2 x /3=2 D cosh2 x .
cosh x0
0 0
B Hints and Solutions 625
cosh x
1 x 1 −x
2e 2e
sinh x
p
17. Since x2 y2 D 1 , y D ˙ x2 1, we have
Z 1 Z cosh t p
A D .tanh t/x dx C ..tanh t/x x2 1/dx
0 1
Z tp
1 1
D tanh t C .tanh t/ .cosh2 t 1/ cosh2 s 1 sinh s ds
2 2 0
Z
1 1 t t
D sinh t cosh t .cosh.2s/ 1/ds D :
2 2 0 2
X
n
1 X 1 1
log n D log 1 C
kD1
k kn
k k
X 1 1 1
< D :
kn
k kC1 n
20. Firstly, note that f .1/ D log.3e=8/ > 0, by Problem 8, page 242. Now, write
x
f .x/ D log xC1=2
xC1
log 1 C 1x C 1 to get limx!C1 f .x/ D 0, and conclude
that it suffices to show that f 0 .x/ < 0 for x 1. For what is left to do, compute
626 B Hints and Solutions
1
x 00 1 1
f 0 .x/ D xC1=2 C log xC1 , f .x/ D x2 Cx
x2 CxC1=4
and conclude that f 0 is
increasing, with limx!C1 f 0 .x/ D 0.
21. Apply Stirling’s formula.
22. Use Stirling’s formula to get
! r
.k C 1/n 1 kC1 1 kn
p .k C 1/n 1 C :
n 2n k k
Section 10.8
1. Note that a D 2 and a D 4 are two roots. To show that there are exactly three
real roots, start by sketching the graphs of x 7! x2 and x 7! 2x ; then, use the
IVT to show that there exists x0 2 .2; 0/ such that x20 D 2x0 . Finally, assume
that x0 is rational and derivea contradiction.
p
2. Firstly, note that f .x/ D log x C x2 C 1 is increasing in R, so that the given
system of equations can be written as
Assume, without any loss of generality, that x D maxfx; y; zg. Then, succes-
sively conclude that f .x/ D maxff .x/; f .y/; f .z/g, y z; x and
py D x. Arguing
in an analogous way, show that x D y D z and solve log.x C x2 C 1/ D 0.
3. By the sake of contradiction, differentiate the equality p.x/ D log x and, then,
use the results of Example 9.17.
4. Let k 2 ZC be such that 10k n < 10kC1 and conclude that k D blog10 nc.
5. We want that
.a1 a2 : : : am 00 : : :…
„ ƒ‚ 0/10 nk .a1 a2 : : : am 99 : : :…
„ ƒ‚ 9/10 :
l l
for some n; l 2 N. This is the same as 10l p nk < 10l .p C 1/, where
p D .a1 a2 : : : am /10 . Take l D kq and show that, in this case, the last
condition above p is equivalent to the existence of a natural pnumber n such that
p p
n 2 Œ10q k p; 10q k p C 1/, for whichpit suffices to have 10q . k p C 1 k p/ > 1
p
or, which is the same, q > log10 . k p C 1 k p/.
6. Start by showing the validity of the inequalities loga .loga b/ > logc .loga b/ and
logb .logc a/ > logc .logc a/. Then, apply item (e) of Problem 1, page 411.
7. Use the given functional equation to show that, if a0 D 0, b0 D 0, anC1 D
2an C 1 and bnC1 D 3bn C 5, then f .an / D bn for every n 0. Then, use the
material of Sect. 3.2 to conclude that f .2n 1/ D 52 .3n 1/ for every n 0.
Finally, solve for n the equation 2n 1 D x.
B Hints and Solutions 627
8. Start by showing that f .0/ D 0, 1 or 2. In the first two cases, show that f is
constant. In the third case, show that f .m/ D 2m whenever m is a power of 2.
Then, take n < m and suppose that f .m/ D f .n/; use induction to show that
f .m2 / D f .n2 / and, to reach a contradiction, choose a; b; c 2 N such that
k k
Finally, take together the two computations above to compute the desired limit.
10. Use the power means inequality.
11. Apply Jensen’s inequality to the natural logarithm, as in Example 10.61.
12. Apply Jensen’s inequality to f .x/ D x log x, x > 0.
1x
13. Apply Jensen’s inequality to f .x/ D log 1Cx , 0 < x < 1.
14. Apply Jensen’s inequality to f .x/ D log sinx x , x 2 .0; /.
1
15. If ai D exi , with xi 0, show that f .x/ D ex C1
, x 0, is strictly convex and
apply Jensen’s inequality.
16. Start by writing the given inequality as
Y
n
ai 1 s
nC1 ;
iD1
1 ai n 1s
X
n s
ai
log .n C 1/ log n C log :
iD1
1 ai 1s
628 B Hints and Solutions
If ai < 12 for 1 i n, apply Jensen’s inequality to f .x/ D log x
1x ,0<x<
1
2 ; subsequently, show that
s s
n log C .n C 1/ log n log :
ns 1s
Pn 1=k
17. Let S D iD1 .ai C b i /k and write
X
n X
n
S D
k
ai .ai C bi / k1
C bi .ai C bi /k1 :
iD1 iD1
1 1
Then, choose q > 0 such that q
D1 k
and apply Hölder’s inequality to each
of the sums above.
Section 10.9
1. For item (a), use the result of item (b) of Problem 15, page 412. For (b),
apply the comparison test for improper integrals, together with the fact that
jet sin tj et for every t 2 R. For item (c), note that log1 t ! 0 when t ! 0C
and apply the comparison test when t ! 1. Finally, item (d) follows from the
1
fact that arctan is a primitive for 1Ct 2.
R g.x/
2. Apply Theorem 10.37 to compute g.B/ f .t/ dt for x > B. Then, let x ! C1.
3. Adapt, to the present case, the argument of Example 10.70. Alternatively, apply
the change of variables theorem for improper integrals (previous problem) with
the variable substitution t 7! 1t , that maps .0; 1 bijectively to Œ1; C1/.
4. For a > 2 , the change of variables formula, together with the formula for
integration by parts, gives
Z a Z p
a
2
cos.x /dx D t1=2 cos t dt
2
ˇpa 1 Z pa
1=2 ˇ
Dt sin tˇ C t3=2 sin t dt
2
p Z pa
sin a 1 sin t
D p C dt:
4
a 2 t3=2
R C1 sin t
Now, argue as in Example 10.75 to show that t3=2
dt converges.
5. For the convergence of the integrals, note first that (cf. Problem 4, page 325)
3 p
sin x x x6 for x 0. Then, for 0 < x < 6,
B Hints and Solutions 629
x3 x2
0 log.sin x/ log x D log x C log 1 :
6 6
R R 2
Since both integrals 0 log x dx and 0 log 1 x6 dx converge, the same is
R 3
p
true of 0 log x x6 dx. Given 0 < ˛ < ˇ < 6, it follows from the above
inequalities that
ˇZ ˇ ˇ ˇZ ˇ x3 ˇˇ
ˇ ˇ ˇ
ˇ log.sin x/dxˇ ˇ log x dxˇ:
˛ ˛ 6
R
Therefore, 0 log.sin x/dx converges. Now, use the change of variables formula
to show that
Z Z
log.sin x/dx D log j cos xjdx;
0 0
R R
so that 0 log j cos xjdx also converges. Finally, let I D 0 log.sin x/dx and
using the change of variables formula again, we compute
Z 2 Z
2I D 2 log j sin xjdx D 2 log j sin.2y/jdy
0 0
Z
D2 log 2 sin yj cos yj dy
0
Z
D 2 log 2 C 2I C 2 log j cos yjdy:
0
R
Then, 0 log j cos yjdy D log 2.
6. Apply the integral test to each of the functions x 7! .log1x/˛ , x 7! x.log1 x/˛ and
1
x.log x/.log log x/˛
.
p
7. First recall that, pfor some A > 1, we have log t < t for t > A. Therefore,
p R C1
log t
1Ct2
< 1Ctt2 < t2t D t3=2 for t > A, and A t3=2 dt converges. Now, take
x > 1 and use the result of Problem 5, page 391, R x aslog
well as Corollary 9.47, to
t
show that f W .0; C1/ ! R, given by f .x/ D 1=x 1Ct 2 dt, is constant. Deduce
R x log t R 1=A log t
that 1Ct2
dt D 0 and, then, that 0 1Ct2 dt converges. Finally, compute
R C1 1=x log t
0 1Ct2
dt D 0.
8. For item (a), apply Young’s inequality (cf. Proposition 10.62) to a D f .t/ and
b D g.t/ and then integrate the result on I. For (b), start by observing that we
Rˇ Rˇ
can assume that ˛ f .t/p dt > 0 and ˛ g.t/q dt > 0. Then, let f1 D f =A and
R 1=p R 1=q
ˇ ˇ
g1 D g=B, where A D ˛ f .t/p dt and B D ˛ g.t/q dt . Finally, show
Rˇ R ˇ
that ˛ f1 .t/p dt D ˛ g1 .t/q dt D 1 and apply the result of item (a).
630 B Hints and Solutions
9. For item (a), on the one hand we have f 0 .x/ D x12 xC11
C 1x D x12
1
x.xC1/ < 0, so that
f is indeed decreasing; on the other, limx!C1 f .x/ D
limx!C1 1x log 1 C 1x D log 1 D 0. For (b), given x > 1 and using
the result of Problem 4, page 412, together with Theorem 10.53, we get
Z Z x
x
1
f .t/dt D log.t C 1/ C log t dt
1 1 t
ˇtDx ˇtDx
ˇ ˇ
D log tˇ ..t C 1/ log.t C 1/ .t C 1//ˇ
tD1 tD1
ˇtDx
ˇ
C .t log t t/ˇ
tD1
1 1 x
D log 1 C log 1 C C 2 log 2 C 1
x x
x!C1
! log 1 log e C 2 log 2 C 1 D 2 log 2:
10. For item (a) adapt, to the present case, the proof of the integral test.
11. For item (a), argue geometrically to show that if 2l 6 < n < nC1 < 2lC 6 ,
then 2l C 6 < n C 2 < 2l C 5 6
, and if 2l C 5
6
< n < n C 1 < 2l C 7
6
,
7 11
then 2l C 6 < n C 2 < 2l C 6 . For item (b), use (a) to conclude that
j sin nj
n C j sin.nC1/j
nC1 C j sin.nC2j
nC2 1
2.nC2/ . Then, use the divergence of the harmonic
series to show what is asked. Finally, for item (c), for each k 2 N show that
R .kC1/ ˇ sin t ˇ R ˇ ˇ
ˇ ˇ dt .kC1/ 6 ˇ sin t ˇ 1=2
dt .kC1/ 2 . Then, conclude what is
k t kC 6 t 6
asked again from the divergence of the harmonic series. (Why can’t one use the
integral test?)
12. For item (a), study the first variation of f .x/ D .1 x/e2x in the interval
Œ0; 12 . For (b), use the Fundamental Theorem of Arithmetic (cf. introduction
to Chap. 1). Item (c) follows from items (a) and (b), together with the fact that
1
1 C p1k C p12 C D 1 p1k . Finally, for item (d) use the divergence of the
k
harmonic series.
Section 10.10
p
1. Setting f .x/ D cosh x in (10.71), we haveˇ 1 C f 0 .x/2 D cosh x, so that the
Rx ˇxDx0
desired length equals 0 0 cosh x dx D sinh xˇ D sinh x0 .
xD0
2. Item (a) follows immediately from (10.71). For item (b), use the fact that
a2 D b2 C c2 , together with the fundamental relation of Trigonometry to write
B Hints and Solutions 631
3. For item (a), observe (cf. Fig. 10.12) that after has rolled to the right by t
radians starting from O, the point of tangency between and r will be T.t; 0/
and the center of will be A.t; 1/. On the other hand, angle †TOP, measured in
the counterclockwise sense, will be equal to t radians. Since BbAP D 32
t (in
the counterclockwise sense), it’s immediate that
3 3
x.t/ D t C cos t D t sin t; y.t/ D 1 C sin t D 1 cos t:
2 2
Z Z Z s
2 p 2 p 2
y0 .t/2 0
1 C f 0 .x/2 dx D 1 C f 0 .x.t//2 x0 .t/dt D 1C x .t/dt
0 0 0 x0 .t/2
Z 2 p p Z 2 p
D 0 2 0 2
x .t/ C y .t/ dt D 2 1 cos t dt:
0 0
For the computation of the last integral, see item (b) of Problem 1, page 397.
R C1
4. By definition, we have . 21 / D 0 et t1=2 dt. Now, perform the substitution
of variable s D t1=2 and apply the result of Problem 2, page 435.
5. We start by using the result of Problem 18, page 413, to conclude that there exists
x
A > 0 such that t 2 j log tj A for 0 < t 1. Then, observe that
Z 1 Z 1 Z 1
x x x
et tx1 j log tj dt D et t 2 1 t 2 j log tj dt < A et t 2 1 dt:
0 0 0
6. For the second part, let nj D j for 1 j k. For the first part, use the fact that
log is convex, together with Jensen’s inequality.
Section 11.1
1. Use version (11.2) of the Taylor’s formula of f centered at x0 , together with the
fact that f 00 0.
2. Assume f .n/ .x0 / > 0 (the other case is entirely analogous). Since f .n/ is
continuous and I is open, we can take an open interval J I, centered at x0 and
such that f .n/ .x/ > 0 for every x 2 J. For a fixed x 2 J n fx0 g, (11.1) guarantees
.n/
the existence of a real c between x0 and x such that f .x/ D f .x0 / C f nŠ.c/ .x x0 /n .
Conclude that p f .x/ > f .x0 /.
sin2 t/
3. If f .t/ D 1 C sin2 t for 0 t , then f 00 .t/ D .1.C2/ .1C sin2 t/3=2
. Now,
look at (10.74) and substitute the formula for f in both sides of the inequality
of Example 11.2, noticing that jf 00 .t/j . C 1/ for 0 t .
4. The Taylor series of the exponential function, together with the definition of
P P
sinh x, furnishes sinh x D 12 .ex ex / D 12 k0 1.1/
k
1
kŠ
xk D j1 .2j1/Š x2j1 .
Argue analogously to cosh x.
5. Apply the result of Proposition 11.3, with n C 1 in place of n.
6. Let p.x/ D am xm C am1 xm1 C C a1 x C a0 and assume am > 0 (the other case
can be dealt with in an analogous way). By Example 9.17, we can take A > 0
such that p.x/ > 0 for x > A. Then, for x > A, apply (11.6) with n D m C 1 to
get the desired result.
7. For item (a), since f .ˇ/; f 0 .ˇ/ > 0, we clearly have < ˇ; now, use the result
of Problem 1 to conclude that f . / 0, and hence ˛. For (b), use the result
of (a) to show that .an /n1 is decreasing and such that an 2 Œ˛; a1 for n 1, so
that there exists ` 2 Œ˛; a1 satisfying an ! ` as n ! C1; then, let n ! C1
in the recurrence relation defining .an /n1 to get ` D ˛. For (i), use (11.2) to get
f .˛/ D f .an / C f 0 .an /.˛ an / C 12 f 00 .n /.˛ an /2 for some n 2 .˛; an /. For the
equality in (ii), use again the recurrence relation of .an /n1 ; for the inequality, use
the fact that f 0 is increasing. Finally, for the first part of item (iii), use the result
of (ii); for the second part, make induction on n, noting that 0 an ˛ d c.
00
8. Verify that f 0 ; f 00 > 0 in Œ2; 52 and maxŒ2; 5 2ff 0 D 35 . In the notations of item
2
(iii) of the previous problem, conclude that we can let c D 2 and d D 52 to get
n 2a3 C5
0 an ˛ 35 212n for every n 1, where a1 D 52 and anC1 D 3an2 2 . Finally,
4 n
Section 11.2
jxj n 1
1. We saw in Example 11.11 that . 1Cx 2 / 2n for every x 2 R. Hence, letting
P
Mk D 21k and noting that k1 Mk < C1, it follows from Weierstrass M-test
P
that k1 . 1Cx
x
2 / converges uniformly in R. Now, Proposition 7.38 gives
k
X x
k x
1Cx2 x
D D :
k1
1 C x2 1 1Cx
x
2 x2 x C 1
2. In all cases we shall use Weierstrass ˇ 1 k ˇ M-test. For the Taylor series of the
ˇ ˇ ak
exponential function, we have kŠ x kŠ if jxj a. Therefore, letting
k P P ak
Mk D akŠ for k 0, we have k0 Mk D k0 kŠ D e < C1, and the
a
6. Start by using Lagrange’s MVT to get jfk .x/j jfk .x/ fk .x0 j C jfk .x0 /j
0
Mk jx x0 j C jfk .x0 /j Mk .b a/ C jfk.x0 /j. Then,
P apply the M-test with Mk D
Mk .b a/ C jfk .x0 /j in place of Mk to show that k1 fk converges P uniformly in
Œa; b to a function f W Œa; b ! R. The M-test also shows that k1 fk0 converges
0
uniformly in Œa; b to a function g W Œa; b ! R,ˇ and we claim that ˇ f D g. To
ˇ f .x/f .x0 / ˇ
this end, let x0 ; x 2 Œa; b, with x ¤ x0 , and S D ˇ xx0 g.x0 /ˇ. With the aid
of Corollary 11.19, write
ˇ X f .x/ f .x / ˇ ˇ 1 XZ x ˇ
ˇ k 0 ˇ ˇ ˇ
fk0 .t/dt g.x0 /ˇ
k
S Dˇ g.x0 /ˇ D ˇ
k1
x x 0 x x 0 k1 x0
ˇ 1 Z xX ˇ ˇ 1 Z x ˇˇ
ˇ ˇ ˇ
Dˇ fk0 .t/dt g.x0 /ˇ D ˇ g.t/ g.x0 / dtˇ
x x0 x0 k1 x x0 x 0
Z
1 ˇˇ x ˇˇ ˇ ˇˇ
ˇ g.t/ g.x0 /ˇdtˇ:
jx x0 j x0
Now, use the continuity of g to show that this last expression can be made less
than , provided x 2 Œa; b is chosen to be sufficiently close to x0 .
Rb
7. Use the given condition to conclude that a f .x/p.x/dx D 0 for every real
polynomial p. Then, use Weierstrass approximation theorem to find a sequence
n
.pn /n1 of polynomials such that pn ! f uniformly on Œa; b. Finally, conclude
Rb
from this fact that a f .x/2 dx D 0.
R1
8. If such an f did exist, we would have R 0 f .x/p.x/dx D p0 .0/, for every
polynomial function p. Now, if F.x/ D 0 f .t/dt, then F 0 D f and F.0/ D
x
R1
F.1/ D 0. Integrating by parts, we get 0 F.x/p0 .x/dx D p0 .0/, which is the
R1
same as 0 F.x/q.x/dx D q.0/, for every polynomial function q. Then, follow
the hint given to the previous problem to get
Z 1 Z 1
F.x/2 dx D lim F.x/qn .x/dx D lim qn .0/ D F.0/ D 0;
0 n!C1 0 n!C1
so that F, and hence f , vanishes identically. But this contradicts the fact that
R1
0 f .x/xdx D 1.
9. Given > 0, use Weierstrass approximation theorem to choose a polynomial p
R1
so that jf .x/p.x/j < for every x 2 Œ0; 1. Note that limn!C1 0 p.x/'n .x/dx
does exist does exist and estimate
ˇZ 1 Z 1 ˇ Z 1 ˇZ 1 Z 1 ˇ
ˇ ˇ ˇ ˇ
ˇ f 'm f 'n ˇ .'m C 'n / C ˇ p'm p'n ˇ:
0 0 0 0 0
R1 R1
Now, use the fact that n 7! 0 'n and n 7! 0 p'n converge, together with
R1
the computations above, to show that n 7! 0 f 'n is a Cauchy sequence.
B Hints and Solutions 635
10. For item (a) and for x 2 .0; 2/, use the result of item (a) of Problem 10, page
359, to get
X n sin .n1/x
2
sin .nC1/x
2
sin.kx/ D x :
kD1
sin 2
ˇP ˇ
Then, conclude that ˇ nkD1 sin.kx/ˇ sin1 x and apply Abel’s criterion. For item
2
(b), note that for 0 < ı < and x 2 Œı; 2 ı, one has j sin 2x j sin ı.
11. For the first parts of items (a) and (b), integrate by parts. For the second
part of (a), use the triangle inequality for integrals to get jak .f /j; jbk .f /j
1
R 2
2 0
jf .x/jdx for every k 1. For the second part of (b), use again the
triangle inequality for integrals, together with the M-test. Finally, for item (c)
use Corollary 11.19, together with Problem 3, page 390.
12. If ak .f / and bk .f / are as in the statement of the previous problem, compute
2
the integrals that define such numbers to get a0 .f / D 23 , ak D 4.1/
k
k2
and
bk D 0 for every P k 2 N. Then, use Fourier convergence theorem to write
f .x/ D a02.f / C k1 ak .f / cos.kx/. Finally, compute f ./ D 2 with the aid of
such a series.
13. Again in the notations of Problem 11, compute the appropriate integrals to find
2
a0 .f / D 1, an .f / D 0 for n 1, b2k .f / D 0 for k 1 and b2k1 .f / D .2k1/
for k 1. Then, evaluate the Fourier series of f at 2 , with the aid of Fourier’s
convergence theorem.
14. For item (a), apply the change of variables formula to g, followed by Riemann’s
Theorem 10.7. For (b), write
Z p n Z
X
kp
n
f .x/g.nx/dx D f .x/g.nx/dx:
.k1/p
0 kD1 n
Then, given > 0, use the uniform continuity of f to find n0 2 N such that, for
n > n0 ,
ˇ n Z kp ˇ Z p
ˇX n kp ˇ
ˇ ˇ
ˇ f .x/ f g.nx/dxˇ jg.nx/jdx p max jgj :
ˇ .k1/p n ˇ 0 Œ0;p
kD1 n
R 2
15. First of all, it follows from Problem 6, page 391, that ak .f / D 1 0 f .x/ cos
.kx/dx. Then, use the result of the previous problem. For bk .f /, argue in a
similar way.
16. Since ak .f / D ak .g/ and bk .f / D bk .g/ for every k, we have Sk f D Sk g for
every k, so that
1 X 1 X
n n
n f D Sk f .x/ D Sk g.x/ D
n g
n C 1 kD0 n C 1 kD0
n n
for every n 2 N. Now, Fejér’s theorem guarantees that
n f ! f and
n g ! g
(even) uniformly on R, so that
n f D
n g implies f D g.
636 B Hints and Solutions
Let Gn .t/ D @f @f
@x .n ; t/, G0 .t/ D @x .x0 ; t/, so that G0 is the pointwise limit of Gn .
Apply Lebesgue’s DCT to the Gn ’s and G0 to show that
Z ˇ
F.xn / F.x0 / n @f
! .x0 ; t/dt:
xn x 0 ˛ @x
B Hints and Solutions 637
18. Use the result of the previous problem, together with Problem 5, page 445. You
may wish to use the following facts: if .xn /n1 is a sequence of positive reals
that converges to the positive real x0 , then .xn /n1 is bounded; in turn, show that
this guarantees the existence of integrable functions g0 ; G0 W .0; 1 ! R and
g1 ; G1 W Œ1; C1/ ! R such that et txn 1 g0 .t/ and et txn 1 log t G0 .t/ in
.0; 1, whereas et txn 1 g1 .t/ and et txn 1 log t G1 .t/ in Œ1; C1/.
Section 11.3
1. For items (a) and (b), use the result of Corollary 11.24. For item (c), note that
the radius of convergence of the series is at most 1,P since it doesn’t converge
when x D 1. Then, note that for jxj < 1 we have k0 x2 < C1, so that
k
P k 2k
k0 .1/ x is absolutely convergent.
P P
2. For jxj < 1, note that k0 kŠ1 x2 < k0 kŠ1 < C1. On the other hand, for
k
hence, the general term of the series doesn’t go to 0 as k increases, and the
series diverges. Therefore, its radius of convergence is 1.
P P
If f .x/ D k1 .1/ x2k1 and g.x/ D k0 .1/
k1 k
3. .2k1/Š .2k/Š
x2k , then f and g are defined
in the whole real line, and Theorem 11.27 gives f 0 .x/ D g.x/ and g0 .x/ D
f .x/. However, since f .0/ D 0 and g.0/ D 1, it follows from Example 9.48
that f .x/ D sin x and g.x/ D cos x. P
1
4. Apply Theorem 11.27 k 1 times, starting from 1x D n0 xn . Alternatively,
expand .1 x/k with the aid of the binomial theorem.
5. Expand .1 x2 /1=2 in a power series with the aid of the binomial theorem.
Then, use item (b) of Proposition 11.26, together with the result of Problem 4,
page 300.
6. For item (a), observe that log 1Cx 1x
D log.1 C x/ log.1 x/ and apply the
result of Example 11.30. For item (c), we have 1Cx1x
D 3 for x D 12 . Substituting
x D 2 in the formula of item (a), note that 2 C 23 3 C 2515 C 2717 C 2919 C 211111
1 1 1
7. For item (a), adapt the idea of the proof of Abel’s identity (5.19). More
precisely, noting that ak D rk rkC1 , we have
X
m X
m X
m
a k xk D a k R k yk D .rk rkC1 /yk
kDn kDn kDn
X
m X
m X
m X
mC1
D rk yk rkC1 yk D rk yk rk yk1
kDn kDn kDn kDnC1
X
m X
m
D rn yn C rk yk rk yk1 rmC1 ym
kDnC1 kDnC1
X
m
D rn yn rmC1 ym C rk yk yk1 :
kDnC1
ˇXm ˇ X
m
ˇ ˇ
ˇ kˇ
ˇ a k ˇ
x jrn j C jrm j C jrk jˇyk yk1 ˇ
kDn kDnC1
X
m
jrn j C jrm j C sup jrk j yk1 yk
kDnC1 k>n
D jrn j C jrm j C sup jrk j yn ym :
k>n
The first part of item (c) now follows from the fact that, on the one hand, jyj 1
gives
ˇXm ˇ
ˇ ˇ
ˇ ak xk ˇ jrn j C jrm j C sup jrk j I
kDn k>n
n
on the other, rn ! 0. Finally, for the second part of (c), note that the first part
guarantees that f extends continuously to Œ0; R.
P
8. In Example 11.30, we saw that log.1 C x/ D k0 .1/
k
kC1 x
kC1
for jxj < 1. Since
P .1/ k
k0 kC1 converges (cf. Example 7.50), it suffices to use the result of the
previous problem.
9. Adapt the hint given to the previous problem, this time using (11.23). Alterna-
tively, start by showing that
for every real x. Then, integrate both sides of the above inequality from 0 a 1,
noticing that
ˇZ ˇ Z 1 Z 1
ˇ 1
.1/n x2n ˇˇ x2n 1
ˇ dx dx x2n dx D I
ˇ 1 C x2 ˇ 1 C x 2 2n C1
0 0 0
Moreover, since
2j x2 > x2 C x2 C1 C x2 C2 C C x2
j j j j jC1 1
;
we conclude that
X j X k x2
xf 0 .x/ > x2 C x2 C1 C x2 C2 C C x2 1 D
j j jC1
x D :
j1 k2
1x
Section 11.4
f .x/ D .1 C x C x2 C /.1 C x C x2 C / : : : .1 C x C x2 C / :
„ ƒ‚ …
k
1 P kCn1 n
Next, show that f .x/ D .1x/ k D n0 n x for jxj < 1, where in the
last equality we used the result of Problem 4, page 483, to get the power series
expansion of .1 x/k .
2. Proceeding as in the hint given to the previous problem, we want to compute
the coefficient of x20 in
Letting f .x/ denote such an expression when jxj < 1, conclude that f .x/ D
1
.x2 C x3 C x4 C C x9 / .1x/ 3 . Then, use once more the result of Problem 4,
page 483.
3. Make induction on k 2, the case k D 2 being given by Proposition 11.46.
4. Differentiate both sides of (11.29) and, then, make x D 1. Note that Theo-
rem 11.27 assures we can do this.
5. Firstly, use
P induction to show that an n2 for n 2. Secondly, letting
f .x/ D k
a x and noticing that the radius of convergence of the power
P k1 k
series k2 k2 xk is equal to 1, conclude that f is defined in the interval .1; 1/.
Now, use the given recurrence relation to show that f .x/ D 2x C xf .x/ C xg0 .x/,
with g.x/ D 1xx
for jxj < 1. From this point, get a closed formula for f .x/, valid
for jxj < 1, expand such a formula in power series and find ak in terms of k.
6. Adapt, to the present case, the proof given in this section to Theorem 11.45.
7. For item (b), compute
X X
f .x/ D 1 C a n xn D 1 C anC1 xnC1
n1 n0
X
D1C .2an C n/xnC1
n0
X X
D 1 C 2xf .x/ C .n C 2/xnC1 2 xnC1
n0 n0
2x
D 1 C 2xf .x/ C g0 .x/ ;
1x
2
with g.x/ D 1x x
. For item (c), reduce both sides to a common denominator
and, then, compare coefficients of the corresponding numerators. For item (d),
use the result of Problem 4, page 483.
8. Adapt, to the present case, the hints given to the items of the previous problem.
B Hints and Solutions 641
X
k
.k C 2/.k C 1/jakC2j .j C 1/jbkj jjajC1 j C jckj jjaj j :
jD0
X
k
.k C 2/.k C 1/AkC2 .j C 1/jbkj .x x0 /kj jAjC1 jx x0 j
jD0
X
k
ˇ ˇ
C ˇckj .x x0 /kj Aj ˇ jx x0 j2
jD0
Xk jx x j kj
0
.j C 1/jbkj jrkj jAjC1 r
jD0
r
X
k jx x j kjC1 jx x j
0 0
C jckj jrkj Aj :
jD0
r r
Item (c) now follows directly from (b). Item (d) follows from (c) and induction
on k. For second part of (e), use the recurrence relation of (d) to get
Section 11.5
kxk
.xx0 /k
0
in .0; 2x0 /.
642 B Hints and Solutions
3. Let n0 2 N be such that f .n/ .x/ 0 for every x 2 I and n n0 ; then, f .n/ is
nondecreasing for every n > n0 . Now, given x0 ; x 2 I, the Taylor formula for f
centered at x0 gives
X
n1 .k/
f .x0 / f .n/ .c/
f .x/ D .x x0 /k C .x x0 /n ;
kD0
kŠ nŠ
X
n1 .k/
f .x0 / f .n/ .x/ n
0 f .x/ .x x0 /k .x x0 /n ! 0:
kD0
kŠ nŠ
.k/
Conclude that g.k/ .a/ D fJ .a/ for every integer k 0 and then that g D fJ in
.a s; b/. Then, extend fJ to J [ .a s; a C s/. Finally, for (d), let J be the largest
open interval contained in I such that (11.39) has a solution in J. Use item (c) to
show that J D I.
Index
H Dirichlet’s, 431
Hölder improper, 427
inequality of, 417, 436 indefinite, 351, 381
Otto, 417 Lebesgue, 366
Harmonic linearity of the Riemann, 369
mean, 117 monotonicity of the Riemann, 368
series, 230 test, 432
Hyperbolic triangle inequality for the Riemann, 372
cosine, 412 Integration
sine, 412 by parts, 386
by substitution, 393
by trigonometric substitution, 395
Intermediate value theorem, 265
I Interval, 17
Identity, Abel’s, 137 bounded, 17
Image closed, 17
inverse, 366 closed on the left, 17
of an element, 143 closed on the right, 17
Improper integral, 427 endpoints of an, 17
change of variables for an, 435 interior of an, 319
Induction length of an, 17
hypothesis, 91 open, 17
principle of mathematical, 87, 90 open on the left, 17
proof by, 88, 90 open on the right, 17
step, 91 partition of an, 351
strong, 93 unbounded, 17
Inequality Inverse
between arithmetic and geometric means, additive, 5
127, 322 image, 366
between arithmetic and harmonic means, multiplicative, 5
117 Isometry, 224
for integrals, Cauchy, 381 IVT, 265
for integrals, triangle, 372
of Abel, 137
of Cauchy, 124 J
of Chebyshev, 129 Jacobi
of Hölder, 417, 436 Carl Gustav Jacob, 444
of Jensen, 334 function of, 444
of Minkowski, 426 Jensen
of power means, 421 inequality of, 334
of Young, 416 Johan, 334
rearrangement, 133
triangle, 30, 96, 126
weighted arithmetic-geometric means, 122, K
425 Kronecker
Infimum of a set, 203 lemma, 221
Initial value problem, 385 Leopold, 221
Injection, 166
Integer part, 152
Integral L
absolutely convergent, 432 l’Hôpital
additivity of the Riemann, 369 Marquis de, 312
convergence of an improper, 427 rule of, 312, 316
648 Index
P Progression
Parabola, 189 arithmetic, 62
axis of a, 190 geometric, 65
directrix of a, 189 Pythagoras
focus of a, 189 of Samos, 16
vertex of a, 190 theorem of, 16
Part
fractional, 152
integer, 152 Q
Partition Quadratic function, 153
norm of a, 360 canonical form of a, 161
of an interval, 351 Quotient, Newton’s, 292
uniform, 358, 363
Pascal
Blaise, 100 R
triangle of, 100 Rôlle
Perfect Michel, 312
cube, 13 theorem of, 312
power, 13 Radius
square, 13 of convergence, 473
Periodic of curvature, 344
function, 195 Rate of change, 276
sequence, 4 instantaneous, 277
Permutation, 133 Rational
Point dyadic, 206
accumulation, 503 function, 251
critical, 319 number, 1
extreme, 158, 319 Ravi’s transformation, 120
inflection, 333 Real
local maximum, 319 function, 147
local minimum, 319 line, 16
of a function, fixed, 179 number, 5
Polynomial, 20 variable, 19
coefficients of a, 20 Rearrangement, 133
equation of degree n, 34 Rearrangement inequality, 133
expression, 20 Rectifiable graph, 439
reciprocal, 44 Recurrence
Polynomials, division algorithm for, 43, 310 relation, 62
Positive half-line, 15 with constant coefficients, first order linear,
Postulate, 4 70
Power with constant coefficients, second order
nth, 9 linear, 70
perfect, 13 Recursive definition, 62
series, 472 Region under the graph, 363
Primitive of an integrable function, 382 Relation
Principle between two sets, 145
multiplicative, 96 on a set, 145
of Cavalieri, 380 order, 7
of counting, fundamental, 96 recurrence, 62
Problem, initial value, 385 Stifel’s, 99, 478
Product Representation
cartesian, 145 madic, 98
notation, 82 binary, 94
telescoping, 84 decimal, 4, 12, 243
650 Index
W Z
Weierstrass Zeckendorf
approximation theorem, 460 Édouard, 98
Karl, 213 theorem of, 98
M-test of, 464 Zero of a function, 301
theorem of, 215