NOTES ON RATIONAL AND REAL NUMBERS

FELIX LAZEBNIK

The notion of a number is as old as mathematics itself, and their developments

have been inseparable. Usually a new set of numbers included the old set, or, as
we often say, extended it. In this way, the set {1, 2, 3, ...} of natural numbers, has
been extended by adding zero to it, negative integers, rational numbers (fractions),
irrational numbers, complex numbers. The actual extensions happened not in the
order suggested by the previous sentence, and the history of the process is fascinat-
ing. The main motivation for this extension came from mathematics itself: having
a greater set of numbers allowed mathematicians to express themselves with better
precision and fewer words, i.e., with greater ease. Numbers form one of the most
important part of mathematical “language”, and in this regard, their development
is very similar to the development of live languages, where the vocabulary increases
mostly for convenience, rather than of necessity.
It is very hard to part with conveniences after getting used to them. Imagine the
world without electricity,
√ or even worse – the mathematics without real numbers.
Then objects like 5, or sin 10◦ , or log10 7 would cease to exist, as they do not
exist among rational numbers. The use of quadratic equations, or Trigonometry,
or Calculus would terminate ... Well, enough of this nightmare.
It is often hard to define basic mathematical notions. The rigor of such definition
depends on the time they are made, and of the depth of the subject where they
are used. For example, a better definition of a function became important with
the development of Calculus, and of abstract algebra. Often the development of
mathematical techniques and accumulation of mathematical facts far preceded the
thorough discussion of the objects being studied. As examples, one can mention
integers, functions, limits, rational and complex numbers. It took many centuries
between their appearance and use in mathematics, and the time when they were
defined at the level meeting now days standards. Of course, mathematics is not
special in this regard. For thousands of years comedies were played in theaters,
but definitions of humor, or of the notion of ‘funny’ appeared very recently. One
may wonder whether those were needed at all, but very few mathematicians will
doubt that in order to discuss rational and irrational numbers it is important to
first define them.
These notes are motivated by the desire to clarify for myself what I wish to say
about the rationals and reals in the courses I teach. There are many accounts of
these topics in the literature, but I have difficulties of using them in my courses.
Either the exposition is too long, or too short, or at the kindergarten level. Or it
requires greater mathematical maturity from the students, or it does not mention
things I find necessary to be mentioned, or I disagree with the emphases, or with

Date: June 30, 2010.

1
2 FELIX LAZEBNIK

the methodology .... These are, probably, typical reasons for one to begin writing
on a classical topic. We often believe that our choices and tastes are better, and
if not, then at least our writings will help students in our courses. Teachers are
like actors, and there are as many Pythagorean theorems as there are Hamlets. So,
here we are.

1. Rational numbers.

Suppose we are familiar with natural numbers

N = {1, 2, . . . , n, . . .},
and with integers
Z = {0, 1, −1, 2, −2, 3, −3, . . . , n, −n, . . .}.

We accept the properties of operations and of order on Z as known.

All similar properties of rational numbers, which we introduce below, will follow
easily from their definition and the corresponding properties of integers.
Construction of rational numbers begins with a set of symbols, called fractions,
and denoted by
m
, or m/n,
n
where m, n ∈ Z and n 6= 0. Two fractions m/n and m0 /n0 are defined to be equal,
if they are precisely the same, i.e., if and only if m = m0 and n = n0 . On the other
hand, every one knows that fractions describe certain objects, like parts of an apple,
and that 1/2, 2/4 or 3/6 of the same apple are indistinguishable. That is why we
write 1/2 = 2/4 = 3/6 as obvious, which contradicts our definition of equalities of
fractions. Why do symbols which look very different are set to be equal? This is
the main reason for the discussion below.

1.1. Let’s complicate things. The standards of rigor in modern mathematics

often require reductions of notions to the set-theoretic language. Usual definition
of a set does not allow equal (same as identical) elements. Notions like an ordered
pair of objects, a relation on a set, a function, etc., are parts of this language. In
what follows we describe how rational numbers can be built from integers in this
way.

This and the next two subsections are for the readers who
are familiar with the notion of an equivalence relation on
a set, and with basic facts related to it. If you are not one
of these readers, just browse through the subsections.
 NOTES ON RATIONAL AND REAL NUMBERS 3

We say that a fraction a/b is equivalent to a fraction c/d, and write it as

a/b ∼ c/d if and only if ad = bc and b, d 6= 0.1 Thus the equivalence of new objects
(fractions) is defined in terms of equality of familiar objects, namely integers.

Examples.
(i) 10/15 ∼ 22/33, since 10 · 33 = 330 = 15 · 22.
(ii) at/bt ∼ a/b for all t ∈ Z, t 6= 0
(iii) Every fraction a/b is equivalent to a fraction a0 /b0 , where gcd(a0 , b0 ) = 1.
In order to prove this, write a = a0 r and b = b0 r, where r = gcd(a, b). E.g.,
10/15 ∼ 22/33 ∼ 2/3.

It is easy to check that the just defined relation ∼ satisfies the following proper-
ties: for all fractions
• a/b ∼ a/b (reflexive property)
• if a/b ∼ c/d, then c/d ∼ a/b (symmetric property)
• if a/b ∼ c/d and c/d ∼ e/f , then a/b ∼ e/f (transitive property).
Let us prove the last property. As a/b ∼ c/d and c/d ∼ e/f , we have ad = bc and
cf = de. Multiplying both sides of the first equality by f , and of the second by b,
we obtain adf = bcf and bcf = bde. This implies adf = bde, and, as d 6= 0, we have
af = be. As b, f 6= 0, we get a/b ∼ e/f .

The readers who are qualified to read this subsection, recognize that the three
properties above mean that ∼ is an equivalence relation on the set of fractions. It
is a standard fact that whenever we have an equivalence relation on a set, the set
becomes partitioned into the equivalence classes. In our case, the equivalence class
containing a/b, which we denote by [a/b], is the following set
[a/b] = {c/d : c/d ∼ a/b},
and classes [p/q] and [x/y] are equal if and only if p/q ∼ x/y. For example,
[1/2] = [−3/ − 6] = [1000/2000] =

{1/2, −1/ − 2, 2/4, −2/ − 4, . . . , 1000/2000, . . .}.

Clearly, [at/bt] = [a/b] for every t ∈ Z, t 6= 0.

Now we define addition and multiplication of the equivalence classes in the fol-
lowing way:
h a i h c i h ad + bc i h a i h c i h ac i
+ = and · = .
b d bd b d bd
It is not obvious that our definition of addition and multiplication of classes is
well-defined in the following sense: since we have different notations for the same
class, like [2/4] = [3/6] and [−2/7] = [−10/35], how can we be sure that
[2/4] + [−2/7] = [3/6] + [−10/35]?

1If one does not think that symbols like a/b belong to the language of set theory, one is welcome
to use (a, b) instead.
4 FELIX LAZEBNIK

In other words, how can we be sure that the result of this newly defined addition
doesn’t depend on the choices of fractions representing the summands? Nevertheless
the last equality is true. Indeed,
h 14 + (−8) i h 6 i h 3 i
[2/4] + [−2/7] = = = ,
28 28 14
h 105 + (−60) i h 45 i h 3 i
[3/6] + [−10/35] = = = .
210 210 14

Let us prove now that the addition is well-defined. Let [a/b] = [a0 /b0 ] and [c/d] =
[c /d0 ]. Then [a/b] + [c/d] = [(ad + bc)/bd] and [a0 /b0 ] + [c0 /d0 ] = [(a0 d0 + b0 c0 )/b0 d0 ].
0

We have to show that [(ad + bc)/bd] = [(a0 d0 + b0 c0 )/b0 d0 ]. Indeed, remembering that
b, d, d, d0 6= 0, we have:
[(ad + bc)/bd] = [(a0 d0 + b0 c0 )/b0 d0 ] ⇔
(ad + bc)/bd ∼ a0 d0 + b0 c0 )/b0 d0 ⇔
adb0 d0 + bcb0 d0 = bda0 d0 + bdb0 c0 ⇔
bcb0 d0 = bdb0 c0 ⇔
cd0 = dc0 ⇔
c/d ∼ c0 /d0 .
The last statement is equivalent to [c/d] = [c0 /d0 ], and the proof is finished.

We call an equivalence class of ∼ a rational number, or just a rational and

denote the set of all rational numbers by Q.
The operations of addition and multiplication of rationals defined above satisfy
the following properties. We will write xy instead of x · y.

Proposition 1.1. For all rational numbers x, y, z, the following hold.

(1) Commutativity: x + y = y + x, and xy = yx.
(2) Associativity: (x + y) + z = x + (y + z), and (xy)z = x(yz).
(3) Identity elements: [0/1] + x = x + [0/1] = x and [1/1]x = x[1/1] = x, and
[0/1] and [1/1] are the only rationals with these properties.
(4) Inverses: for a given rational x, there is a unique rational u such that
x + u = u + x = [0/1];
for a given rational x 6= [0/1], there exists a unique rational v such that
xv = vx = [1/1].
(5) Distributivity: x(y + z) = xy + xz and (x + y)z = xz + yz

The rational u in part (4) is denoted by −x and is called the additive inverse
of x, or the opposite of x. The rational v in part (4) is denoted by x−1 and is
called the multiplicative inverse of x, or the reciprocal of x.

Proof. We prove only two of these statements, and leave others as exercises.
(1) Let x = [a/b] and y = [c/d]. Then x + y = [(ad + bc)/(bd)]. Using the
commutative properties of addition and multiplication of integers, we rewrite ad+bc
as cb + da, and bd as db. Hence,
x + y = [(ad + bc)/(bd) = [(cb + da))/(db)] = y + x.
 NOTES ON RATIONAL AND REAL NUMBERS 5

Similarly for multiplication:

xy = [(ac)/(bd)] = [(ca)/(db)] = yx.

(4) We prove it for the multiplicative inverse only. For the additive inverse the
argument is similar.
Let x = [p/q] 6= [0/1]. Then p 6= 0, and we can consider v = [q/p]. So we have:

xv = [(pq)/(qp)] = [(pq)/(pq)] = [1/1].

As xv = vx by (1), we also have vx = [1/1]. This shows that v is a multiplicative

inverse of x. Why v is unique? Let v 0 x = xv 0 = [1/1]. Then, using the associativity
of the multiplication of rationals (which is left for the reader to check),

v 0 = v 0 · [1/1] = v 0 (xv) = (v 0 x)v = [1/1] · v = v,

hence, v 0 = v. ¤

As an immediate corollary of Proposition 1.1 we obtain several more useful prop-

erties of operations on Q. They all can be proved as logical consequences of state-
ments (1) − (5) of Proposition 1.1, i.e., without any reference to the nature of
elements of Q.

Proposition 1.2. For all rational numbers x, y, z, the following hold.

(1) x + y = x + z if and only if y = z.
(2) If xz = yz and z 6= [0/1], then x = y.
(3) [0/1] · x = x · [0/1] = [0/1].
(4) xy = [0/1] if and only if x = [0, 1] or y = [0/1].
(5) (−x)y = −(xy) = x(−y), −x = (−1)x, (−x)(−y) = xy.
(6) If x, y 6= 0, then (xy)−1 = x−1 y −1 , and (x−1 )−1 = x.

Proof. Let us prove (2), (3), and the first statement of (5). Proofs of other state-
ments are left as exercises.
(2) As z 6= 0, there exists z −1 . Then, as xz = yz, (xz)z −1 = (yz)z −1 . Using
parts (2) and (4) of Proposition 1.1, we obtain:
(xz)z −1 = x(zz −1 ) = x · [1/1] = x, and (yz)z −1 = y(zz −1 ) = y · [1/1] = y. This
implies x = y.
(3) We have: [0/1] · x = ([0/1] + [0/1]) · x = [0/1] · x + [0/1] · x. Adding −([0/1] · x)
to both sides and using (1), we get [0/1] = [0/1] · x. As [0/1] · x = x · [0/1], the
statement is proven.
(5) Using the distributive property (5) of Proposition 1.1, and part (3), we obtain

xy + (−x)y = (x + (−x))y = [0/1] · y = [0/1].

So xy+(−x)y = [0/1], which means that (−x)y is the additive inverse of xy, namely
−(xy), or simply −xy. ¤
6 FELIX LAZEBNIK

1.2. Are integers rational numbers? Strictly speaking: no. What is true, that
a function
f : Z → Q,
defined by a 7→ [a/1], is an injection satisfying the following properties:
f (a + a0 ) = [(a + a0 )/1] = [a/1] + [a0 /1] = f (a) + f (a0 ),
and
f (aa0 ) = [(aa0 )/1] = [a/1] · [a0 /1] = f (a) · f (a0 ).
The properties show that the symbols f (a) = [a/1], when it comes to operations,
just mimic the ones in Z, and differ from them only in appearance. This is often
phrased by saying that the substructure {[a/1] : a ∈ Z} of Q is isomorphic to Z.

1.3. Let’s make things simpler. Now, when we know what rational numbers
are, we will begin to denote them by fractions, as we used to. So, when we write
1/2 ∈ Q, we really mean [1/2] ∈ Q. We also remember that equivalent fractions
denote the same rational number, and from now on, writing 1/2 = 4/8 is O.K.
And, yes, writing a instead of [a/1], or writing Z ⊂ Q, is fine too. Hence, [0/1] can
be written as just 0, and [1/1] as just 1. What a relief!
Well, our time was not exactly wasted: now we know what this all means!!! Don’t
you feel like Monsieur Jourdain, who suddenly discovered that “These forty years
now, I’ve been speaking in prose without knowing it! How grateful am I to you for
teaching me that!”2?

1.4. Order on Q. We remind the reader that we assume known the properties of
the (usual) order relation ≤ on Z. Similar properties of the order on Q, which we
are going to introduce, will follow from them easily.
We say that x = a/b ∈ Q is positive (negative), and write 0 < x (x < 0), if
and only if ab is a positive (negative) integer.3
The sum x + (−y) is called the difference of x and y, and it is usually denoted
by x − y.
For x, y ∈ Q, we say that x is less than y, and write it as x < y, if y − x is
positive. The inequality x < y is often expressed in another way by saying that y is
greater than x, and by using the symbol > to denote it. Hence y > x is equivalent
to x < y.4
The abbreviation for “(x < y) or (x = y)” is x ≤ y. If x ≤ y, we say that x is at
most y, or, equivalently, y is at least x. Similarly, for x ≥ y. Hence, both 3 ≤ 3
and 3 ≤ 5 are true. If 0 ≤ x or, equivalently, x ≥ 0, we say that x is non-negative,
and if x ≤ 0 or, equivalently, 0 ≥ x, we say that x is non-positive.

2From J.B.P. Moliére: Le Bourgeois Gentilhomme, 1670.

3If one wants to be more pedantic, and check that the notion is well-defined for rationals, one
can observe that if a/b ∼ c/d, then ab and cd are both positive or both negative simultaneously.
Indeed, a/b ∼ c/d implies ad = bc, which is equivalent to abd2 = b2 cd (as b, d 6= 0). As b2 and d2
are positive, ab and cd are both positive, or both negative.
4This is an example of a wasteful expansion of mathematical vocabulary which is hard to
justify. Maybe, saying “x is less than y” makes us to concentrate on x more than on y, while
saying “y is greater than x” does otherwise? Same also applies to the next paragraph.
 NOTES ON RATIONAL AND REAL NUMBERS 7

Proposition 1.3. Suppose x, y, z represent rational numbers. Then the following

holds.
(1) If x > 0 and y > 0, then x + y > 0 and xy > 0
(2) If x < 0 and y < 0, then x + y < 0 and xy > 0.
(3) If one of x, y is positive and another is negative, then xy < 0.
(4) 1 > 0.
(5) x > 0 (x < 0) if and only if −x < 0 (−x > 0).
(6) x > 0 (x < 0) if and only if x−1 > 0 (x < 0).
(7) Only one of the following three statements is true for a number x:
x = 0, x > 0, x < 0.
(8) x ≤ x (x ≥ x)
(9) If x ≤ y and y ≤ x, then x = y.
(10) If x ≤ y (x ≥ y) and y ≤ z (y ≥ z), then x ≤ z (x ≥ z).
Proof. We prove only three of these statements, and leave others as exercises.
(1) Let x = a/b and y = c/d. Then both ab and cd are positive integers. As
x+y = (ad+bc)/bd, we consider (ad+bc)(bd) = (ad)(bd)+(bc)(bd) = (ab)d2 +b2 (cd).
We know that integers ab, b2 , cd and d2 are all positive, and sums and products of
positive integers are positive. So we conclude that (ad + bc)(bd) is positive, which
is equivalent to x + y > 0. Similarly, xy = (ac)/(bd), and (ac)(bd) = (ab)(cd) > 0
as the product of two positive integers. Hence, xy > 0.
(4) Since 1 = 1/1, and 1 · 1 = 1, and 1 is a positive integer, then 1 (now viewed
as rational) is a positive rational.
(10) Let us prove the version involving ≥. Since x ≥ y and y ≥ z, both x − y
and y − z are non-negative rationals by definition. Due to (1), which we proved
above, the sum of positive rational is positive. This implies that the sum of two
non-negative numbers is non-negative. So (x − y) + (y − z) = x − z ≥ 0, hence,
x ≥ z. ¤

The order relation and the operations on Q satisfy the following additional prop-
erties. The expression 0 < x < y represents, of course, a shorthand of the statement:
“0 < x and x < y”.
Proposition 1.4. For any rational numbers x, y, z, a, b, the following holds.
(1) x < y if and only if x + z < y + z
(2) For z > 0, x < y if and only if xz < yz
(3) For z < 0, x < y if and only if xz > yz
(4) If a < b and x < y, then a + x < b + y, and a similar statement holds for
any n ≥ 2 inequalities.
(5) If 0 < a < b and 0 < x < y, then ax < by, and a similar statement holds
for any n ≥ 2 inequalities
(6) The inequalities 0 < a < b imply 0 < an < bn , for any n ∈ N. Also, if
b > 0, and n ∈ N, then an < bn implies a < b.
Proof. Here we present proofs of several of the properties, leaving proofs of the
remaining ones to the reader. We wish to note that these proofs will not be based
on our understanding of what rationals are: they are merely logical consequences
of the properties stated in Propositions 1.1, 1.2, 1.3.
8 FELIX LAZEBNIK

(2) The distributive law of Proposition 1.1 (5) holds also when addition is replace
by subtraction. Indeed, applying Proposition 1.2 (5), we obtain:
yz − xz = yz + (−xz) = yz + (−x)z = (y + (−x))z = (y − x)z.
If x < y, then y − x > 0 is positive by definition. Since z > 0 (given), and the
product of positive rationals is positive by Proposition 1.3 (1), (y − x)z is positive.
Hence yz − xz > 0 , and so xz < yz.
Conversely: let xz < yz. Then yz − xz = (y − x)z > 0 by definition. As z > 0,
this implies y − x > 0, otherwise (y − x)z < 0 by Proposition 1.3 (3), or y = x.
Hence, x < y.
(5) Proof 1. By transitivity of inequalities, b > 0. By part (2) of this proposition
proven above, a < b implies ax < bx, and x < y implies xb < yb. By the commuta-
tivity of multiplication of rationals, the latter is equivalent to bx < by. Transitivity
of inequalities gives ax < by.
Proof 2. by − ax = by − bx + bx − ax = b(y − x) + (b − a)x. As all b, y − x, b − a,
and x are positive, by − ax is positive. Hence, ax < by.
Let us generalize this property:
For all n ≥ 2, the inequalities 0 < ai < bi , i = 1, . . . , n, imply
a1 a2 · · · an < b1 b2 · · · bn .
We prove this by the method of mathematical induction (on n). The base case,
n = 2, has been established in (5): a1 = a, b1 = b, a2 = x, b2 = y. Suppose the
statement holds for any n = k ≥ 2 inequalities. We have to show that it holds for
any n = k + 1 inequalities.
Let 0 < xi < yi for all i = 1, . . . , k + 1. We have to show that
x1 x2 · · · xk xk+1 < y1 y2 · · · yk yk+1 .
Note that x1 · · · xk < y1 · · · yk by the induction hypothesis, and we are given that
xk+1 < yk+1 . As all parts of the inequalities are positive rationals, these two
inequalities can be multiplied (base case n = 2). By doing this, we obtain
(x1 · · · xk )xk+1 < (y1 · · · yk )yk+1 .
When several rational numbers are multiplied, the parentheses can be placed in ar-
bitrary way due to the associative property of multiplication. Hence, the statement
holds for any n = k + 1 inequalities, and the proof is finished. ¤
All properties of the order on Q we have seen so far are exactly the same as the
one’s on Z. But there are some fundamental distinctions. For example, unlike for
integers, the order relation on Q satisfies the following ‘density’ property.
Proposition 1.5. (Density property of Q) Between any two distinct rationals there
are infinitely many other rationals.
Proof. Suppose x, y be distinct rationals, and x < y. Let z1 = (x + y)/2. Then
z1 ∈ Q,
x+y y−x
z1 − x = −x= > 0, and
2 2
x+y x−y
z1 − y = −y = < 0.
2 2
 NOTES ON RATIONAL AND REAL NUMBERS 9

Hence, x < z1 < y. Repeating the argument for x and z1 , we set z2 = (x + z1 )/2.
Then z2 is a rational number, and x < z2 < z1 . Continuing this way we obtain an
infinite decreasing sequence of rationals zn+1 = (x + zn )/2, such that x < zn < y
for all n ≥ 1. ¤

On the other hand, an analog of the Well-ordering of integers does not hold
in Q. Indeed, for a ∈ Q, let Q>a = (a, b) := {x ∈ Q : a < x < b}. Then not
every non-empty subset S of (a, b) contains the smallest element. For example,
take S = (a, c), with a < c < b. If m ∈ S were the least element of S, then, by the
density of Q (Proposition 1.5), the interval (a, m) would contain infinitely many
rationals. Each of these rationals is in S and less than m, a contradiction.

1.5. Representing rationals as decimal fractions. It is well known that every

rational number can be written as a decimal fraction, and that the “long division”
algorithm can be used to do it. Have you ever seen the details? If not, here they
are.
First we accept for granted that integers can be written “in base 10”. E.g., the
decimal 2304 stands for the number
2 · 103 + 3 · 102 + 0 · 101 + 4 · 100 .
This implies that fractions with denominators 10, 100, 1000, etc., can also be written
“in base 10”, e.g.,
519006 519006 5 · 105 + 1 · 104 + 9 · 103 + 0 · 102 + 0 · 101 + 6 · 100
= 4
= =
10000 10 104
1 1 1 1
5 · 101 + 1 · 100 + 9 · 1
+ 0 · 2 + 0 · 3 + 6 · 4.
10 10 10 10
This, as we know, can also be shortened to a decimal 51.9006. Let us now review
how we write fractions with denominators different than powers of 10 as decimals
(finite or infinite). What is presented below is the “long division ” algorithm, which
most of us were taught in the middle school, by using a shorthand, like
14|123.
Let p/q ∈ Q, q ≥ 1. Dividing p by q with remainder we obtain p = aq + r1 ,
where 0 ≤ r1 < q. This gives
p r1
=a+ .
q q
If r1 = 0, we write p/q = a, or p/q = a.000 . . .. If r1 ≥ 1, we divide 10r1 by q with
remainder. We get 10r1 = b1 q + r2 , where 0 ≤ r2 < q. Since 0 ≤ 10r1 ≤ 9q, then
0 ≤ b1 ≤ 9. This gives
p r1 10r1 1 b1 q + r2 1
=a+ =a+ · =a+ · =
q q q 10 q 10
³ r2 ´ 1 ³ 1 ´ r2 1
a + b1 + · = a + b1 · + · =
q 10 10 q 10
³ 1 ´ 10r2 1
a + b1 · + · 2.
10 q 10
10 FELIX LAZEBNIK

If r2 = 0, we write p/q = a.b1 , or p/q = a.b1 000 . . .. If r2 ≥ 1, we divide 10r2 by

q with remainder. We get 10r2 = b2 q + r3 , where 0 ≤ r3 < q and 0 ≤ b2 ≤ 9. If
r3 = 0, we wrote p/q = a.b1 b2 , or p/q = a.b1 b2 000 . . .. If r3 ≥ 1, we have
p ³ 1 1 ´ r3 1
= a + b1 · + b2 · 2 + · .
q 10 10 q 102
Continuing this way we obtain
³ 1 1 1 ´ rn+1 1
p/q = a + b1 · + b2 · 2 + · · · + bn · n + · n =
10 10 10 q 10
rn+1 1
a.b1 b2 . . . bn + · n,
q 10
and so forth. This leads to the following representation of p/q:
p/q = a.b1 b2 . . . bn bn+1 . . . , (1)
with 0 ≤ bi ≤ 9 for all i. If rk = 0 for all k > n, we abbreviate a.b1 b2 . . . bn 000 . . .
as a.b1 b2 . . . bn , and write p/q = a.b1 b2 . . . bn . In this case we often say that p/q
is represented as a finite decimal. If rk 6= 0 for infinitely many k, we obtain an
infinite decimal.
The equality (1) may prompt a question: we understand completely its left side,
but do we really have complete information about its right side? It will take too
long to finish the writing of an infinite decimal... It turns out that the sequence
{bn }n≥1 is actually rather simple, and it allows a short description. Let us explain
it.
Writing rational numbers as decimals by using the algorithms above, often re-
ferred to as the long division algorithm, it is easy to notice that we always obtain
a decimal with a repeating group of consecutive digits. For example,
• 23/6 = 3.8333 . . .: digit 3 repeats beginning from the second decimal posi-
tion;
• 1/11 = 0.09090909090909090909 . . ., sequence 09 repeats beginning from
the first decimal position;
• 3991/990 = 4.031313131 . . ., sequence 31 repeats beginning from the second
decimal positions.
Often one expresses this repetition phenomena by writing 23/6 = 3.8(3), 1/11 =
0.(09), 3991/990 = 4.0(31). Note that the same repeating pattern can be expressed
in other forms: 23/6 = 3.8(33), 1/11 = 0.(090909), 3991/990 = 4.031(3131313131),
but usually we use the shortest repeating sequence of decimals at the the position
it begins. Let us show that the repetition always takes place, and that the shortest
repeating sequence for a fraction p/q, q ≥ 2, has length at most q − 1.
Theorem 1.6. Let p/q = a.b1 b2 . . . bn bn+1 . . ., q ≥ 2. Then there exist positive
integers i and j, i < j, such that
p/q = a.b1 b2 . . . bi (bi+1 . . . bj ),
with j − i ≤ q − 1.
Proof. Look again at the long division algorithm. If ri = 0 for some i, then the
statement is proven, as i + 1 = j and bi+1 = bj = 0. If ri 6= 0 for all i, then, as
there are at most q − 1 distinct nonzero remainders when an integer is divided by q,
two of the remainders must repeat. Hence, there exist indices i, j, i < j, such that
 NOTES ON RATIONAL AND REAL NUMBERS 11

ri = rj . Then the ratios and the remainders of the divisions of 10ri and 10rj by q
will be equal, i.e., bi = bj and ri+1 = rj+1 . This, in turn, leads to bi+1 = bj+1 and
ri+2 = rj+2 , and so on. Hence, the repetition of a finite sequence of consecutive
digits in the decimal representation of p/qis inevitable. Let j be the smallest index
greater than i such that ri = rj . Then the remainders ri , ri+1 , . . . , rj−1 are all
distinct, and there are (j − 1) − i + 1 = j − i of them. The remainders are elements
of {1, 2, . . . , q − 1}. Hence, j − i ≤ q − 1. ¤
Theorem 1.6 helps to speed computation of bn for large n. Suppose we wish to
know b100005 in
554867/26000 = 21.3410(384615).
Then b100005 is one of the six digits in the repeating sequence 3, 8, 4, 6, 1, 5. As
the repeating begins with b5 = 3, we divide 100005 − 4 by 6 with remainder,
obtaining 100001 = 16666 · 6 + 5. Hence b100005 is the fifth member of the sequence
3, 8, 4, 6, 1, 5, i.e., b100005 = 1.

1.6. There is no rational number whose square is 2. This was proven by the
Pythagoreans (or maybe even by Pythagoras himself) in the 6th century BC, and
is often considered as one of those “great moments” in mathematics which strongly
affected its development.
Due to the Pythagorean Theorem for a right triangle, the length d of the diagonal
of the unit square satisfies d2 = 12 +12 = 2. Pythagorean believed that integers rule
the world, and everything could be expressed as certain relations among natural
numbers: the motion of planets, physics, geometry, music ... In particular, it is
believed that they believed (see Grattan-Guinness [8]) that the length of every
segment can be expressed as a positive integer or as a ratio of integers. So one can
imagine how shaken were they after proving that this is not the case! We wish to
present and compare several different proofs of this result. For more proofs, see
Bogomolny [4]. See also [22] and Henderson [10] for related discussions.
Proposition 1.7. There is no rational number whose square is 2.

Proof 1. This proof is, probably, the the most popular one. It assumes two facts:
• no odd integer is equal to an even integer, and
• every fraction can be written in its “lowest terms”, i.e., with having its
numerator and denominator relatively prime. (The latter follows from the
Well-ordering of integers).
Suppose the contrary, and let (m/n)2 = 2 with integers m, n being relatively prime.
Then m2 = 2n2 .
We claim that m is even. Indeed, if m were odd, then it could be written as m =
2s+1, for some integer s, and this would lead to m2 = 4s2 +4s+1 = 2(2s2 +2s)+1–
an odd number. The obtained contradiction implies that m is even. Hence, m = 2s
for some integer s. Then, (2s)2 = 2n2 , or 2s2 = n2 . Arguing as before, we obtain
that n is even. Having both m and n even, contradicts the assumption that they
are relatively prime. This ends the proof. ¤
Proof 2. This proof assumes that
• 2 is a prime number,
• every non-prime integer can be written as a product of prime numbers, and
the number of primes in this product is uniquely defined.
12 FELIX LAZEBNIK

(So it uses a little less than the uniqueness of prime factorization).

Suppose the contrary, and let (m/n)2 = 2. Then m2 = 2n2 . A square of any
integer (≥ 2) has an even number of prime factors in its prime factorization, namely
twice as many as the number itself. Therefore, as 2 is prime, the equal numbers
m2 and 2n2 have, respectively, even and odd number of primes in their prime
factorizations. A contradiction. ¤
Proof 3. This proof uses the Well-ordering of integers and simple inequalities.
Let (m/n)2 = 2. We may assume that both m and n are positive integers, otherwise
we replace them with their absolute values. Since the set of m’s in such rationals
m/n is not empty, there must be one with the smallest m. Since 1 < m/n < 2, we
have 0 < 2n − m < m and 0 < m − n < n. As
m 2n − m ³ 2n − m ´2
= , we obtain = 2,
n m−n m−n
a contradiction, since 0 < 2n − m < m. ¤
Proof 4. The following argument is a geometric version of Proof 3, though it may
take a few minutes to see it. It deals with a geometric analog of a common divisor
of two integers. It is very much in the spirit of Euclid’s Elements, and, possibly, it
is close to one of the oldest proofs of this result. See Rademacher and Töplitz [18],
and Courant and Robbins [5] for related discussions.
Two line segments AB and CD are called commensurable if and only if there
exists some third segment EF that can be laid end-to-end a whole number, say m,
of times to produce a segment congruent to AB, and also, with a different whole
number, say n, a segment congruent to CD. In this case EF is called a common
measure of AB and CD. In terms of the lengths of the segments, it means that
AB = m · EF and CD = n · EF .
It is easy to argue that two segments are commensurable if and only if the ratio
of their lengths is a rational number. Indeed, let AB and CD be commensurable.
Then AB = m · EF , CD = n · EF , and AB/CD is a rational number m/n.
Conversely, let the ratio of lengths of line segments AB and CD be rational, say
p/q. Divide CD into q congruent segments, and call one of them EF . Then EF ,
laid end-to-end q times gives CD, and p times gives AB. Hence, segments AB and
CD are commensurable.
If a leg of a right isosceles triangle is commensurable with its hypothenuse, then
the equation x2 = 2 has rational solutions. Indeed, let a be the length of the leg.
Then the length of the hypothenuse is m n a, for m/n ∈ Q, and
³ m ´2 ³ m ´2
a = a2 + a2 ⇔ = 2.
n n
In what follows we show that the leg and the hypothenuse are not commensu-
rable. This will prove that the equation x2 = 2 has no rational solutions.
Let CA be a leg of an isosceles right triangle ABC with the hypothenuse AB,
see Figure 1. Then CA < AB < CA + CB = 2 · CA. Let C 0 be a point on AB
such that BC 0 = BC. Let BB 0 bisect ∠B, where B 0 is on CA.
 NOTES ON RATIONAL AND REAL NUMBERS 13

A
C 00 B 00 0
C

B0

C B
FIGURE 1.

As 4B 0 CB is congruent to 4B 0 C 0 B by Side-Angle-Side axiom, ∠B 0 C 0 B is a

right angle and CB 0 = C 0 B 0 . Hence, ∠AC 0 B 0 is a right angle. As the measure of
∠A is 45◦ , 4AC 0 B 0 is isosceles, so AC 0 = B 0 C 0 = CB 0 .
Suppose that CA and AB are commensurable, and EF is a common measure.
It is clear that if two segments are commensurable, then the segment whose length
is the difference of their lengths is commensurable with them. Hence, EF is also
a common measure of AC 0 and AB 0 . Thus we obtained a smaller right isosceles
triangle AC 0 B 0 such that EF is a common measure of its leg and its hypothenuse.
Repeating the argument for this triangle, we obtain a smaller right isosceles tri-
angle AC 00 B 00 with EF being a common measure of its legs and its hypothenuse.
Continuing this way, we obtain a decreasing (prove it) sequence of lengths
CA > C 0 A > C 00 A > . . . ,
where each number is an integer multiple of EF . Writing CA = m · EF , C 0 A =
m0 · EF , C 00 A = m00 · EF , etc., we get
m > m0 > m00 > . . . ,
which is an infinite decreasing sequence of positive integers. Its existence, clearly
contradicts to the Well-ordering of integers. The obtained contradiction proves that
CA and AB are not commensurable. ¤.

1.7. An unsolved problem. We finish this section by mentioning an open prob-

lem. It is related to representing rationals as decimals. Suppose p is a positive odd
prime number. Considering representations of several fractions 1/p as decimals we
have:
1/3 = 0.(3),
1/7 = 0.(142857),
1/11 = 0.(09),
1/13 = 0.(076923), and
1/17 = 0.(0588235294117647).
From these examples one can see that, when p is prime, the length of the shortest
repeating sequence of decimal digits in the decimal expansion of 1/p can be p − 1
(p = 7, 17), or less (p = 3, 11, 13). It is never greater than p − 1 by Theorem 1. It is
an not known whether there are infinitely many primes p for which this sequence
contains exactly p − 1 digits. For more on this, see [18].
14 FELIX LAZEBNIK

Exercises for Section 1.

All problems below can be solved without any reference to real numbers. That
is why we avoided geometric terminology which would make some statements more
appealing. Solving problems below, one may assume, if needed, that a rational
can be represented by a fraction p/q, where gcd(p, q) = 1. By ∗ we mark harder
problems (in our opinion).

1.1 Prove the reflexive and symmetric properties of the relation ∼ on the set
of fractions as defined in this section.
1.2 Show that the multiplication of rationals introduced in this section is well-
defined.
1.3 Prove all properties of operations on rationals from Proposition 1.1.
1.4 Prove all properties of operations on rationals from Proposition 1.2.
1.5 Prove all properties of inequalities on rationals from Proposition 1.3.
1.6 Suppose a positive integer n is not a square of another integer. Prove that
it is also not a square of a rational number.
1.7 Show that there is no rational number r such that
(i) r5 = 96 (ii) 3r = 7
1.8 A rational solution of the equation x2 + y 2 = 1 is an ordered pairs (a, b)
with both a, b ∈ Q and a2 + b2 = 1. Find ten distinct rational solutions of
x2 + y 2 = 1.
1.9 Check that for every rational number r, the rational numbers
x = (r2 − 1)/(r2 + 1) and y = (2r)/(r2 + 1)
satisfy the equation x2 + y 2 = 1.
This proves that the equation x2 + y 2 = 1 has infinitely many rational
solutions.
∗
1.10 Prove that the equation x2 + y 2 = 3 has no rational solutions, i.e., there
are no rational numbers x and y which satisfy it. This shows that not every
rational number is a sum of squares of two rational numbers.
1.11 Check that
³ ´3 ³ 3(m2 + m) ´3 ³ −m3 + 3m + 1 ´3
9m = m − 1 + +
m2 + m + 1 m2 + m + 1
Substituting m = a/9 gives a representation of any rational a as a sum
of cubes of three rational numbers. Compare this result with the one of
Problem 1.10.
Now, of course, one wants to know how such a representation can be
found, and are there other representations. For this, see Shklarsky, Chent-
zov, Yaglom, Sussman,[20] (an elementary exposition), or Cohen [6] (for
good college students), or Manin [14] (an advanced treatment).
1.12 Can the long division algorithm produce decimal
(a) 0.111 · · · = 0.(1)?
(b) 0.131313 · · · = 0.(13)?
 NOTES ON RATIONAL AND REAL NUMBERS 15

∗
(c) 0.999 . . . = 0.(9)?
Comment. One can show that if a decimal not of the form 0.b1 . . . bk (9), it
can always be obtained by applying the long division to certain integers.
1.13 Let α = p/q = a.b1 b2 . . . bn . . . .... Find b1000 if
(i) α = 0.22(345) (ii) α = 3991/990.

1.14 (Rational root theorem) Let p/q, gcd(p, q) = 1, be a rational solution of

the equation
an xn + an−1 xn−1 + · · · + a1 x + a0 = 0, with all ai ∈ Z, an 6= 0.
Then p divides a0 and q divides an .
Comment. This fact is useful for finding all rational solutions of polyno-
mial equations with integer coefficients (or showing that no such exist). It
reduces the problem to examining only finitely many fractions p/q. The
statement also implies that if an = 1, then every rational solution of the
equation is actually an integer.
1.15 (i) Find all rational solutions of the equation
2x5 − 9x4 + 12x3 − 12x2 + 10x − 3 = 0.
(ii) Show that there is no rational solution of
x5 − 7 = 0.
∗
1.16 Prove that for each rational number r such that r2 < 2, there exists a
rational number t, such that r < t and t2 < 2.
∗
1.17 Prove that Q is a countable set, i.e., there exists a bijection from N to Q.
1.18 Is there a set {a1 , a2 , . . . , ak } of rational numbers such that every rational
number r can be written in the form r = n1 a1 + n2 a2 + . . . + nk ak , for some
integers ni ?
1.19 (a) Let f : Q → Q be a function such that f (a + b) = f (a) + f (b) for
all a, b ∈ Q. Prove that f is always of the form f (x) = cx for for
some c ∈ Q and all x ∈ Q. A function with this property is called an
additive function on Q.
(b) Let f : Q → Q be a function such that f (ab) = f (a)f (b) for all
a, b ∈ Q, and f (1) = 1. A function with this property is called a
multiplicative function on Q. Describe all multiplicative functions
on Q.
(c) Let f : Q → Q be a function which is both additive and multiplicative
on Q. Prove that f is the identity function, i.e., f (x) = x for all x ∈ Q.
16 FELIX LAZEBNIK

2. Real numbers.

2.1. Why are they needed? It is clear that for all everyday life purposes one does
not need numbers other than rationals, especially when one uses decimal fractions.
Every measurement is approximate, and finite decimals can express the result with
needed precision. They are sufficient for building chairs, houses, and even rockets.
But they are not sufficient for measuring the lengths of all segments! Well, segments
are not objects encountered in everyday life... Still it is so restrictive not be able
to say what is the length of the diagonal of a square with a unit side! An easily
perceived correspondence between real numbers and points of a straight line made
real numbers even more ‘real’, and lead first to the Coordinate method, and then to
Calculus. These two theories proved to be extremely useful, both in mathematics
and applications. That is why one needs reals.

2.2. Definition of real numbers. What is a real number? Most of the time, when
one thinks about a real number, one imagines symbols like 0.234, or −21.998167,
or 4532.12121212... which are called decimals or decimal fractions. Though
decimals were used informally in our discussion of rationals, here we need to do
it again, and at a slower pace. Decimals are infinite sequences made out of ten
digits 0, 1, 2, . . . , 9 preceded with a symbol + or −, and one symbol represented
by a period “.” . (The last period is the grammatical symbol, not mathematical!)
Usually, when terms of a sequence are arranged in a row, they are separated by
commas. We do not do it in case of decimals.
A general form of a decimal is
α = ±am am−1 . . . a1 a0 .b1 b2 . . . bn . . . , where all ai , bj ∈ {0, 1, 2, . . . , 9}.
The integers ai and bj are called the decimal digits of α. In such a repre-
sentation, if m ≥ 1, am is assumed to be non-zero. Any decimal of the form
0.00 . . . 0bN bN +1 bN +2 . . ., N ≥ 1, is called a decimal tail of α, or the N th decimal
tail of α.
To simplify appearance of decimals, several conventions are agreed upon. Usually
the + which precedes the decimal is not used: 0.2341984 . . . is just a shorthand for
+0.2341984 . . .. Usually a tail with each term equal to 0 is dropped, and we write
0.234000 . . . as 0.234. The latter is an example of a finite decimal. So every finite
decimal is a decimal with tail of all 0’s: 43.45 = 43.45000 . . .. Usually the period is
not used if it follows by a tail of all 0’s: 34.000 . . . is the same as 34, and −2.0 is
the same as −2 . Another convention, which we will motivate later (see Subsection
2.4), is that any decimal with a tail of all 9’s is assumed to be equal to a finite
decimal with the digit preceding the tail increased by 1:
0.233999 . . . = 0.234,
1.999 . . . = 2,
−1.999 . . . = −2,
−3.24999 . . . = −3.25 .
We call decimals described above real numbers and denote the set of all of them
by R. Often we refer to real numbers as to “reals”.
 NOTES ON RATIONAL AND REAL NUMBERS 17

Having described the symbols we use to identify real numbers provides very little
information about them. The most important property of objects referred to as
numbers is that one can compute with them, i.e., to perform some operations, like
addition or multiplication. Also, often one wants to be able to order them, and be
able to say that one number is smaller or greater than another. For reals represented
by decimals, the latter is much easier than the former. Indeed, everyone knows how
to add or multiply two finite decimals, but how does one add two infinite decimals?
2.3. Order on decimals. Ordering reals is easy, and we all know how to do it:
23.9891 < 111.343434 . . .
23.9891 < 23.9991
23.9891 < 23.989100001
−5 < 0.000001
−5.00001 < −5, etc.
The sign < is read ‘less’. A general rule of ordering reals can be stated this way.
Decimals which begin with + (often omitted) are said to be positive, and those
which begin with −, negative. We denote the set of all positive decimals by R+ ,
and the set of all negative decimals by R− . We assume that the zero decimal
0 = 0.000 . . . is the only decimal which is neither positive nor negative. For any
negative decimal α, and any positive decimal α0 , we declare α < 0, 0 < α0 , and
α < α0 . For any two distinct positive decimals
α = am am−1 . . . a1 a0 .b1 b2 . . . bn . . .
and
β = cs cs−1 . . . c1 c0 .d1 d2 . . . dn . . . ,
we declare
α<β
if one of the following conditions is satisfied:
• m < s, or
• m = s and, proceeding from left to right in both decimals and comparing
the corresponding decimal digits one at a time, we obtain that for the first
time they are distinct, the digit of α is less than the corresponding digit of
β.
The second case, m = s, can be presented more formally in the following way.
(1) m = s, and there exists an integer k, 0 ≤ k ≤ m, such that am = cm ,
am−1 = cm−1 , ... , ak+1 = ck+1 , but ak < ck , or
(2) m = s, and ai = ci for i = 0, 1, . . . , m, and there exist an integer k ≥ 1,
such that b1 = d1 , b2 = d2 , ... , bk−1 = dk−1 , but bk < dk .
Being able to compare any two positive decimals α and β, we compare −α and −β
in the reverse way: −α < −β if and only if β < α. (This was a definition, by the
way.)

The definitions of ‘greater’(>), ‘at most’ (≤), ‘at least’ (≥), ‘non-positive’, and
‘non-negative’ are exactly the same as for the rationals. One can also easily check
that all the properties of the order on rationals listed in parts (7) - (10) of Propo-
sition 1.3 are satisfied by the just introduced order on reals.
We also have an analog of the density property of rationals:
18 FELIX LAZEBNIK

Proposition 2.1. (Density property of R) Between any two distinct reals, there
are infinitely many other reals.
Note that at this place of our discussion we cannot prove this statement in the
way we did it for rationals in Proposition 1.5, as we do not know yet how to add
reals. Nevertheless, a proof based on ordering of reals, can be easily given. The
following numerical example illustrates the idea of such a proof. A general argument
is similar, and we omit it.
Let α = 23.25983 . . . and β = 23.25984 . . .. Then α < β, as no tail of α consists
of 9’s only. The latter also implies that there are infinitely many decimal digits in
the 6-tail of α which are less than 9. Increasing any one of them by 1, and not
changing other decimal digits, we obtain infinitely many decimals which are greater
than α and less than β.
2.4. Forbidding tails of all 9’s. Forbidding tails of all 9’s, allows us to say that
if reals “look” distinct, they are distinct. Otherwise, if we considered 1.999 . . . and
2.000 . . . as distinct reals, we would not be able to find any other real between them,
and the density property will fail. When reals are matched with points of a line,
this would introduce a ‘gap’ on the latter which our intuition resists to accept. A
few other reasons will be explained later.

2.5. Periodic decimals. A decimal α = ±am am−1 . . . a1 a0 .b1 b2 . . . bn . . . , is called

periodic, if there exist a positive integer N and a positive integer p such that for
all n ≥ N , bn+p = bn :
∃N ∈ N ∃p ∈ N ∀n ≥ N (bn+p = bn ).

If a decimal is periodic, numbers N and p are not defined uniquely. For example,
12.333 . . . (all decimal digits are 3) is periodic, with N = 1, p = 1, or with N = 3
and p = 5. Decimal 12.1341341341 . . . (sequence of digits 341 repeats) is periodic,
with N = 2, p = 3, or with N = 4 and p = 6.
If α is periodic, p is called a period of α. It is clear that if p is a period of α,
then kp is also a period for all integers k ≥ 0. Indeed, for n ≥ N ,
bn+2p = b(n+p)+p = bn+p = bn ,
so 2p is a period. Similarly, for n ≥ N ,
bn+3p = b(n+2p)+p = bn+2p = bn ,
so 3p is a period, etc..
It can be shown that if a decimal is periodic, then there exists the smallest
period, and it is called the principal period. For example, if
α = 4.031313131 . . . ,
then periods have length 2, 4, 6, . . ., the value of the principal period is 2, and α is
often written as 4.0(31), or 4.0(31).
Here are some not periodic decimals:
β = 0.101001000100001000001 . . .
(runs of 0’s increase in length and are separated by 1’s; )

γ = 0.12345678910111213 . . .
 NOTES ON RATIONAL AND REAL NUMBERS 19

(consequent natural numbers written next to each other;)

Let us explain why β is not periodic. If it is, then there exist positive integers N
and p such that bn+p = bn for all n ≥ N . As the lengths of sequences of consecutive
0’s grow, there exists a positive integer M ≥ N such that
bM = bM +1 = . . . = bM +p−1 = 0.
Since p is a period, so are 2p, 3p, 4p, etc., then
bM = bM +1 = . . . = bM +p−1 =

bM +p = bM +1+p = . . . = bM +2p−1 =
bM +2p = bM +1+2p = . . . = bM +3p−1 = · · · = 0,
i.e. β is periodic with a tail of all 0’s. This certainly contradicts the definition of
β. Hence, it is not periodic.
The non-periodicity of γ can be shown in a similar way.

2.6. Continuity of reals. Soon we will see that reals posses all good properties of
rationals. Here we discuss a fundamental property of real numbers which rationals
do not possess, the property of continuity, which makes reals superior to rationals
when it comes to Calculus.
Let S ⊆ R. A number A ∈ R is called an upper bound of S if every element
of S is at most A:
A is an upper bound of S if ∀x ∈ S (x ≤ A).
A subset of reals which has an upper bound is called bounded from above.
For example, (−∞, 1) is bounded from above by any number greater or equal to 1.
An upper bound α of S ⊂ R is called a least upper bound of S if it is not
greater than any upper bound of S. If S ⊆ R has a least upper bound, it is, clearly,
unique, so we can speak about the least upper bound.
The least upper bound of S is often called the supremum of S, and it is denoted
sup(S). For example,
sup((−∞, 1)) = sup((−∞, 1]) = sup((0, 1)) = 1.
Similar definitions can be made of a lower bound of S, of a set bounded from
below, and of the greatest lower bound of S. The greatest lower bound of S is
often called the infimum of S, and it is denoted by inf(S).
Theorem 2.2. (Continuity of Real Numbers.) Every nonempty bounded from
above (below) subset of real numbers has the unique least upper (greatest lower)
bound in R.
Before proving the theorem, we wish to explain that it does not hold for rational
numbers, more precisely, that not every bounded from above a nonempty subset of
Q has a supremum in Q.
Proposition 2.3. Let S = {r ∈ Q : r2 < 2}. Then S is a non-empty bounded from
above in Q and has no supremum in Q.
20 FELIX LAZEBNIK

Proof. As 1 ∈ S, S is nonempty. Obviously, 2 is an upper bound of S, so S is

bounded from above. Suppose α = sup(S), where α ∈ Q. We show that this
assumption leads to a contradiction. Note that (if exists) α > 1.
As α2 = 2 is not possible by Proposition 1.7, we have α2 < 2 or α2 > 2.
Case 1: α2 < 2. Then α ∈ S. Consider any rational t such that
0 < t < min{1, (2 − α2 )/(2α + 1)}.
Then
(α + t)2 = α2 + 2αt + t2 = α2 + (2α + t)t < α2 + (2α + 1)t <
2 − α2
α2 + (2α + 1) = α2 + (2 − α2 ) = 2.
2α + 1
Hence, (α + t)2 < 2, and so α + t ∈ S. As α < α + t, α is not an upper bound of
S, a contradiction.5
Case 2: α2 > 2. Then α 6∈ S and α is an upper bound of S. Consider any
rational t such that
0 < t < min{1, (2 − α2 )/(−2α + 1)}.
Then
(α − t)2 = α2 − 2αt + t2 = α2 + (−2α + t)t > α2 + (−2α + 1)t >
2 − α2
α2 + (−2α + 1) = α2 + (2 − α2 ) = 2.
−2α + 1
Hence, (α − t)2 > 2, and so α − t is a positive upper bound of S smaller than α, a
contradiction.
Therefore there is no supremum of S in Q. ¤

We now proceed with a proof of Theorem 2.2. Let bxc denote the integer part of
x, defined as the greatest integer less or equal x. For example, b6.23c = b6c = 6,
and b−6.23c = b−7c = −7.
Proof. Let ∅ 6= S ⊆ R.
Case 1. S contains nonnegative numbers.
Let S0 = {bxc : x ∈ S}. Then S0 is a non-empty subset of integers. As S
is bounded from above, so is S0 . By the Well-Ordering Axiom of integers, S0
contains the greatest element.6 Denote it by a. Then a ≥ 0. Consider a set S1
obtained from S by removing from it all numbers with integer part less than a:
S1 = {x ∈ S : bxc = a}. As a ∈ S1 , S1 is not empty. Each element x ∈ S1 can be
represented as a decimal x = a.x1 x2 x3 . . ., where a denote the representation of a in
base 10. Let b1 = max{x1 : x = a.x1 x2 x3 . . . ∈ S1 }. Removing from S1 all numbers
x with x1 < b1 , we obtain a nonempty subset of S2 = {x ∈ S1 : x = a.b1 x2 x3 . . .}.
Let b2 = max{x2 : x = a.b1 x2 x3 . . . ∈ S2 }. Let S3 = {x ∈ S2 : x = a.b1 b2 x3 . . .}.
Continuing this way we construct a decimal α = a.b1 b2 b3 . . . bn . . ., which is not
5All inequalities in this argument are applied to rational numbers only, and hold due to Propo-
sitions 1.3 and 1.4.
6Often the Well-Ordering Axiom is presented in the following version: every nonempty subset
X of integers bounded from below contains the least element. This is equivalent to the the
statement that every nonempty subset X of integers bounded from above contains the greatest
element. To see the equivalence, consider the set −X = {−x : x ∈ X}.
 NOTES ON RATIONAL AND REAL NUMBERS 21

necessarily in S. Clearly, α is an upper bound of S. We claim that α = sup(S).

Let γ = c.d1 d2 d3 . . . dn . . . (with no tail of all 9’s) is an upper bound of S. We have
to show that α ≤ γ.
Suppose α has no tail of all 9’s. If α = γ, we are done. Suppose α > γ. Then
either a > c, or a = c ≥ 0 and there exists an index n ≥ 1 such that bn > dn
and bi = di for indices smaller than n (if any). If a > c, then c is not an upper
bound of S, a contradiction. In the latter case, S contains a number of the form
x = a.b1 b2 b3 . . . bn . . . ≥ a.b1 b2 b3 . . . bn > γ, a contradiction again.
It may happen that α has a tail of all 9’s. As no element of S has such a tail,
this will happen only if α = a.(9), or for some k ≥ 2, S contains all numbers
of the form a.b1 . . . bk−1 9, a.b1 . . . bk−1 99, a.b1 . . . bk−1 999, . . ., where bk−1 < 9. In
the first case, we rewrite α as a + 1, and in the second – as a.b1 . . . b0k−1 , with
b0k−1 = bk−1 + 1. As in the previous case, S again contains a number of the form
x = a.b1 b2 b3 . . . bn . . . ≥ a.b1 b2 b3 . . . bn > γ, a contradiction.
Hence, α = sup(S).
Case 2. S contains only negative numbers.
Consider a number δ > 0 such that the set S + δ := {x + δ : x ∈ S} contains
nonnegative reals. Clearly such a δ exists. Then sup(S +δ) exists, as we just proved
in Case 1. Let’s call it α. Then, obviously, α − δ = sup(S).
To prove the theorem with respect to infimum, replace a nonempty bounded
from below set S of reals by −S = {−x : x ∈ S}. As −S is bounded from above,
by the argument above, it has the least upper bound in R, say α. Then, obviously,
−α = inf(S). ¤

It seems that a logical way to finish this subsection is to show that there exists a
real number whose square is 2. This would partially justify our efforts of introducing
real numbers. Rereading our argument which preceded the proof of Theorem 2.2,
makes it clear that the supremum of S in R must have this property. The only
problem with this “clear” is that at this time we do not know yet that the laws of
operations and inequalities for reals are similar to the ones for rationals. Moreover,
we have not introduced any operations on R yet! Therefore, we have to address
this issue first.

2.7. Computations with decimals. In this section we use the fact that finite
decimal fraction can be identified with (some) rational numbers, and that usual
operations on them have the properties listed in Proposition 1.1, except the exis-
tence of multiplicative inverses for each nonzero number. (though x = 3 = 3.(0) is
a finite decimal, x−1 = 0.(3) is not.)
Let α = ±am am−1 . . . a1 a0 .b1 b2 . . . bn . . .. For each n ≥ 1, we define a finite
decimal
αn = ±am am−1 . . . a1 a0 .b1 b2 . . . bn 000000 . . .
= ±am am−1 . . . a1 a0 .b1 b2 . . . bn
Then αn is called the nth approximation of α. Let β be another decimal, and let
βn be its nth approximation. Counting with finite decimal fractions in the usual
way, we define two new sequences of finite decimals: for each integer n, n ≥ 1, let
σn := αn + βn , and ρn := αn · βn .
22 FELIX LAZEBNIK

Then, one can prove (but we do not do it here) that there exists a unique
decimal σ, and a unique decimal ρ such that, for every n ≥ 1, σn and ρn are the
nth approximations of σ and ρ, respectively. The decimals σ and ρ are called the
sum and the product of α and β, respectively, and we write σ = α + β and ρ = αβ
(or α · β). Here is an example. Let

α = 2.398745012 . . . and β = 5.137096541 . . .

Then
α1 + β1 = 2.3 + 5.1 = 7.4
α2 + β2 = 2.39 + 5.13 = 7.52
α3 + β3 = 2.398 + 5.137 = 7.535
α4 + β4 = 2.3987 + 5.1370 = 7.5357
α5 + β5 = 2.39874 + 5.13709 = 7.53583
α6 + β6 = ... + ... = 7.535841
α7 + β7 = ... + ... = 7.5358415
α8 + β8 = ... + ... = 7.53584155
α9 + β9 = ... + ... = 7.535841553

and

α1 β1 = 2.3 · 5.1 = 11.73

α2 β2 = 2.39 · 5.13 = 12.2607
α3 β3 = 2.398 · 5.137 = 12.318526
α4 β4 = 2.3987 · 5.1370 = 12.32212190
α5 β5 = 2.39874 · 5.13709 = 12.3225432666
α6 β6 = ... · ... = 12.322583344520
α7 β7 = ... · ... = 12.32258454389250
α8 β8 = ... · ... = 12.3225846912132654
α9 β9 = ... · ... = 12.322584703886203492

Hence σ = 7.53584155 . . . and ρ = 12.322584 . . .. Let us explain why we do know

the first six decimal digits of ρ. To get an upper bound of ρ, we can just multiply
two numbers slightly greater than the nth approximations of α and β for any fixed
n. Taking n = 8, we obtain ,

12.3225846912132654 < ρ < (α8 + 10−8 )(β8 + 10−8 ) =

2.39874502 · 5.1370966 = 12.322584886508932.

Hence, 12.3225846 < ρ < 12.3225848. Therefore ρ = 12.322584 . . ..

One can show that the opposite of α is the same decimal as α but with the
altered sign. It is denoted by −α. One can also show that the difference α − β,
defined as α + (−β), is positive if and only if α > β.
 NOTES ON RATIONAL AND REAL NUMBERS 23

It turns out, but we omit the proofs, that the order, and the operations on reals,
in addition to all properties of Propositions 1.3, 2.1 and Theorem 2.2, also satisfy
all properties of Propositions 1.1, 1.2 and 1.4.

Now it is a good time accept the fact that

There exists a positive real number x whose square is 2.
√
It is denoted by 2. The argument preceding the proof of Theorem 2.2 from
subsection 2.6,√with the set S considered now as a subset of reals, provides a proof.
It shows that 2 = sup(S), which exists in R.√
By a similar argument, one can prove that α exists for every real α ≥ 0.

2.8. Where are we? In the last section we used phrases like ‘one can prove’, or
‘one can show’, or ‘it turns out ... , but we omit the proofs’. In case the reader
is getting tired of these statements, we can assure the reader, that if we decided
to present all these proofs, she or he would be even more tired ... Our goal here
is not to present a self-contained mathematically rigorous exposition of a theory
of real numbers. This is done in undergraduate courses very rarely, and rarely all
the properties are verified even in graduate courses. The main reason is that every
existing thorough exposition is long.7

Our goal in Section 2 was to review basic properties of reals, to mention some
less known ones, and to point at the difficulties with the development of the theory.
We also wanted to provide a model of real numbers. Instead of spending time on
verifying all the details related to the model, we prefer to concentrate on other
questions related to reals which we find more interesting. Let us mention them.
If some of the notions or the properties in the list below are not familiar to the
reader, keep reading and attempt solving problems. These notions and properties
will play little role in the remaining part of these notes.

7A relatively fast and rigorous way of introducing reals is defining them axiomatically, e.g., as
an ordered field with an additional Completeness axiom. Corresponding definitions and exposition
can be found, e.g., in Birkhoff and MacLane [3] or D’Angelo and West [1]. In this approach many
propositions/theorems become axioms, and other properties are derived from them. One can
easily proceed with building new theories which use reals: limits of functions, Calculus, Topology,
Geometry, etc. . The main weakness of this approach is that it does not provide any explicit
description of reals in terms of other mathematical objects known to the reader. Such a description
is often called a model. A model is necessary to ensure that the axiomatic system is consistent
in the sense that no two contradictory statements can be proven in it. The model of decimals
that we provided is, probably, the simplest. It was suggested by K. Weierstrass (1815-1897) in
the second half of the 19th century. Weierstrass never published his theory of real numbers, so
it is hard to date it. But notes of his lectures, taken at different years by several of his students,
contain it.
Among other models of reals, we wish to mention the ones constructed by G. Cantor (1871)
and by R. Dedekind (1856, 1872). For other constructions, see the article “Construction of the real
numbers” on Wikipedia.org. For each construction one has to carefully check that all axioms of
reals are satisfied. After this is done, we say that the construction is a model of reals. It is possible
to show (see, e.g., [3]) that all model of reals are essentially the same (isomorphic), which is not a
property most axiomatic system possess. The readers who knows basics of abstract algebra, can
easily see that none of the axiomatic system for groups, or for rings, or for fields has the property
that all corresponding models are isomorphic.
24 FELIX LAZEBNIK

We take for granted the following additional notions and facts about reals:

• Limits and their properties

• Series and their properties (like convergence,. . .). In particular, that
X∞
bi
0.b1 b2 . . . bn . . . = i
.
i=1
10

• Continuity of elementary
√ functions and related properties. These imply,
e.g., that numbers 10 1.2, sin 3, arctan 2, . . . exist.
• The set of all periodic decimals can be identified with Q, and operations
are consistent with the ones in R. In other words, we assume that the set
of all periodic decimals is isomorphic to Q, and we will write Q ⊂ R.
• Q is a countable set. R is an uncountable set.

We have seen that the long division algorithm applied to a fraction always leads
to a periodic decimal (Theorem 1.6). On the other hand, it is possible to argue
that for every periodic decimal α (no tail of all 9’s), there exist a rational number
p/q such that the long division of p by q leads to α. We do not prove this fact, but
illustrate two practical ways of doing it.

Example 1. Let α = 2.(7). Then (believe us!) 10α = 27.(7), and (believe us
again!) 10α − α = 9α = 25. Hence, α = 25/9. If one doubts the method, one can
just apply long division to 25 and 9. The answer will be 2.(7).
If the reader is familiar with infinite geometric series, then the same result can
be obtained in the following way. First we take for granted that
X∞
7 7 7 7
2.(7) = 2 + + 2 + 3 + ··· = 2 + n
.
10 10 10 n=1
10

Then, using the formula for the sum of the infinite geometric progression, we obtain:
X∞
7 7/10
2.(7) = 2 + n
=2+ = 2 + 7/9 = 25/9.
n=1
10 1 − 1/10

Example 2. Let α = 5.3(71). Then 10α = 53.(71) and 1000α = 5371.(71).

Subtracting, we get 990α = 5318. Hence α = 5318/990 = 2659/495. Again, to
check the result, one can apply long division to 2659/495. The answer will be
5.3(71).
Using infinite geometric progression, the same result can be obtained this way.

3 71 71 71 53 71/103 53 73 2659
5.3(71) = 5 + ++ 3 + 5 + 7 +··· = + = + = .
10 10 10 10 10 1 − 1/102 10 990 495
 NOTES ON RATIONAL AND REAL NUMBERS 25

2.9. Irrationals. Thus we assume that

N ⊂ Z ⊂ Q ⊂ R,
and that rational numbers correspond to precisely the periodic decimals. We have
also shown that some decimals are not periodic, so they do not correspond to
rational numbers. Reals which are not rational are called irrational. They form
the set R \ Q. They correspond to (infinite) non-periodic decimals. Since Q is
a countable set, and R is uncountable, and the union of two countable sets is
countable, the set R \ Q of irrational real numbers is uncountable. Hence, the are
“much more” irrational numbers than there are rationals.

When we deal with irrational numbers, in most cases we do not have them
represented
√ as non-periodic decimals. We describe them in some other way, like
2, or log2 3, cos 1. Descriptions like these are useful if we wish to know which
decimal digit occupies, say the 1000th decimal place, since, in these cases, there are
relatively fast algorithms of computing the nth decimal digit. At the same time,
no one knows what patterns are present in their representations as decimals. For
example, no one knows whether
√ there exists a sequence of 100 consecutive 5’s in
the decimal representing 2.
The questions whether given real numbers are rational or irrational have always
occupied mathematicians. Some of them are relatively easy, others are very hard.
It is hard not to like them. Often these are asked about numbers with simple
definitions, and which often appear in mathematical reasonings: like
e, π, log2 3, cos 1,
arccos(1/3)
,
π
∞
X 1
ζ(3) := ,
i=0
n3
X∞
1
ζ(5) := 5
,
i=0
n
the Euler-Mascheroni constant
³X
n
1 ´
γ := lim − ln n .
n→∞
i=0
i
For example, no one knows whether ζ(5) or γ is irrational. More on this subject
can be found in the following books: Niven [15, 16], Schoenberg [19], Hardy and
Wright [9], Khinchin [11], and Baker [2].
In the next section we consider several relatively simple problems of this kind.

2.10. Rational or irrational? As we know, the set Q of rationals is closed with

respect to arithmetic operations, like the sum, difference, product, division (if the
divisor is not zero). This means that the result of these operations is again a rational
number. This is not true for irrationals. For example, the sum or the product of
two irrationals can be a rational number.
On the other hand, a sum of a rational number and an irrational number is always
irrational. Indeed, if the sum were rational, then the irrational number would be
the difference between two rationals, a contradiction. See more in Problem 5.
26 FELIX LAZEBNIK

√
Example 3. Prove that if n ∈ N and n is not a square of an integer, then n is
irrational.
√
Proof.
√ As n exists, it is either rational or irrational. Suppose it is rational. Then
n = a/b, for some a, b ∈ N. Then a2 = nb2 . As n > 1 and is not a square of
an integer, the exponent of some prime divisor p of n in the prime decomposition
of n is odd. As the exponents of p in the prime factorizations of a2 and of b2 are
even, the exponent is even in √ a2 and odd in nb2 . This contradicts the uniqueness
of prime factorization. Hence n is irrational. ¤

√ √
Example 4. Prove that 2 + 3 5 is irrational.

Proof. We would like to discuss three proofs of this statement. They illustrate
different ideas.
√ √ 2
√
√ Proof 21. Suppose r = 2 + 3 5 is rational. Then r = 47 + 6 6 is rational, and
6√ = (r − 47)/6 is rational. As 6 is not a perfect square of an integer (Example
3), 6 is irrational, a contradiction. Hence, r is irrational.
√ √ √ √ √ √
√Proof √2. Let a1 = 2 + 3 5, a2 = 2 − 3 5, a3 = − 2 + 3 5, and a4 =
− 2 − 3 5. Consider the polynomial

f (x) = (x − a1 )(x − a2 )(x − a3 )(x − a4 ) =

(x2 − (a1 + a2 )x + a1 a2 )(x2 − (a3 + a4 )x + a3 a4 ) =

√ √
(x2 − 2 2x − 43)(x2 + 2 2x − 43) =
√
(x2 − 43)2 − (2 2x)2 = x4 − 94x2 + 432 .
By the Rational root theorem (see Problem 14), the only possible rational roots
of f can be 1, −1, 43, −43, 432 and −432 . One can easily eliminate each of them
as a root (by a direct substitution, for example). Hence, all four real roots of f ,
namely, all ai are irrational. The argument demonstrates the irrationality of all
four roots simultaneously. It also demonstrates that introducing “more symmetry”
into the problem can be useful. This last observation was extremely important in
the development of algebra in the last two centuries.
It is possible to argue, that f is a polynomial of the smallest degree with integer
coefficients which has a1 as its root. We do not do it here.

Proof 3. The polynomial

√ f√from the previous
√ 2 proof√could also be√found in the
2
following way. Let a = 2+3 5. Then √ 2 (a− 2) = (3 5) , or a2 −2 2a+2 = 45.
2 2 4 2 2
From here we have, (a − 43) = (2 2a) , or a − 94a + 43 = 0. Hence, a is a
root of f . Now we apply the Rational root theorem, as in Proof 2. ¤
 NOTES ON RATIONAL AND REAL NUMBERS 27

2.11. Approximating some irrational numbers by rationals. Theorems

of Kronecker and Dirichlet. The fact that for any real number α there exist
rational numbers arbitrarily close to it is obvious: just take the nth approximation
αn of α. The denominator of αn is 10n , and the approximation we get is within
10−(n+1) . Hence, in this case the distance to a number is no greater than a tenth
of the denominator of the approximating fraction. An interesting question is how
close one can get to α by using rational numbers with denominators at most b. We
begin our discussion with the following result, which, I think, is very surprising at
the first glance.

Example 5. Surround each rational number ab ∈ (1, 2), gcd(a, b) = 1, with an

1
open interval Ia/b which is centered at a/b and of radius 3.5b 2 , i.e.,
³a 1 a 1 ´
I ab = − , + .
b 3.5b2 b 3.5b2
Does the union of all such intervals cover the interval (1, 2)? In other words, is it
true that [
(1, 2) ⊆ I ab ?
a
b ∈(1,2)

Solution. We claim that the answer is No. As every rational number is covered,
we
√ should try to find an irrational number which is not covered. Let us prove that
2 is one of such numbers.
√
Proof. Number 2 is not in the union of all Ia/b if and only if it does not belong
to any of them. In terms of inequalities it means that for every ab ∈ (1, 2),
¯√ a ¯¯ 1
¯
¯ 2− ¯≥ ,
b 3.5b2
or, equivalently, that the inequality
1 √ a 1
− 2
< 2− <
3.5b b 3.5b2
does not hold for any rational a/b.
In order to prove this inequality, we first observe that |2b2 −a2 | ≥ 1 for all integer
a, b ≥ 1. We can assume a, b ≥ 1 as a/b > 0. This implies
√ √
|b 2 − a||b 2 + a| ≥ 1 ⇔
√ 1 a
|b 2 − a| ≥ √ ⇒ (as 1 < < 2)
|b 2 + a| b
¯√ a¯¯ 1
¯
¯ 2− ¯> 2 √ ⇒
b b ( 2 + 2)
¯√ a ¯¯ 1
¯
¯ 2− ¯> .
b 3.5b2
¤
Having proved the statement, let us now try to understand it better. Two
decimals are close to each other if their decimal expansions begin with many equal
corresponding digits. We know that any rational number a/b can be presented as
a decimal with a period no longer than |b| − 1. At the same time, an irrational
number can be thought as a decimal with an infinite period. Therefore for the
28 FELIX LAZEBNIK

numbers to be close, |b| must be large. When |b| is large, the radius of Ia/b , namely
1
3.5b2 , is small. The example √
above demonstrates that it is small enough for Ia/b
never ‘capture’ the irrational 2.

The result in Example 5 should be compared with √ the fact that there exist
infinitely many rationals a/b which still come close to 2 in the sense that
1 √ a 1
− 2 < 2 − < 2.
b b b
This is a particular case of the following beautiful theorem of Dirichlet (1842):
Theorem 2.4. (Dirichlet’s Theorem.) For any irrational number α there exist
infinitely many rational numbers a/b such that
1 a 1
− 2 < α − < 2.
b b b
In light of Example 5 and this theorem, a question one may ask is of determining
the maximum positive constant c such that the inequality
1 √ a 1
− 2 < 2− < 2
cb b cb
holds for infinitely many rationals a/b. Theorem 2.4 implies that this maximum
value
√ is at least 1, and, from Example 5 it is at most 3.5. It turns out that it is
2 2 (for details, see Waldschmidt [21]).
Now we discuss another striking result related to distribution of certain numbers
among reals. It has many surprising applications. In particular, the Dirichlet’s
Theorem 2.4 will follow from it.

We say that a set S of real numbers is dense in R if every open interval (a, b) ⊂ R
contains a number from S. It is clear that if S is dense, then every interval actually
contains infinitely many numbers from S.

Theorem 2.5. (Kronecker’s Theorem.) Prove that for any irrational number α,
the set K = {nα + m, m, n ∈ Z} is dense in R.
Proof. Note that both set Z and K is closed with respect to addition (and sub-
straction) of its elements.
We remind the reader that for any real number x, bxc denotes the integer part of
x, defined as the greatest integer which does not exceed x. The fractional part of x
is defined as x − bxc, and is denoted by {x}. For example, if x = 2.12, 5, −3, −3.1,
then bxc = 2, 5, −3, −4, and {x} = 0.12, 0, 0, 0.9, respectively. Clearly, {x} ∈ [0, 1)
for all x ∈ R.
Let us first show that K is dense in (0, 1). The proof of the theorem will easily
follow from this. For every n ∈ N, let xn = {nα}.

Key observation 1. All xn are distinct. Indeed, let m 6= n and xn = xm .

This implies that
{nα} = {mα} ⇔
nα − bnαc = mα − bmαc ⇔
(n − m)α = bnαc − bmαc.
 NOTES ON RATIONAL AND REAL NUMBERS 29

As all bnαc, bmαc, m − n are integers, and m 6= n, the last equality implies that
α is rational, a contradiction. Hence all xn are distinct.
Consider an interval (a, b) ⊂ [0, 1), and let b−a = ². We choose a positive integer
N such that 1/N < ², and divide [0, 1) into N subintervals of equal length:
h 1´ h1 2´ hN − 1 ´
0, , , , ..., ,1 .
N N N N

Then at least two of x1 , x2 , . . . , xN +1 must be in the same subinterval. Suppose

these are xk and xk+h , h ≥ 1. Let δ = xk+h − xk > 0 (if δ < 0, the argument will
continue in a similar way). Then 0 < δ < 1/N < ².

Key observation 2. Consider a subsequence (xk+th ), t ≥ 0, of the sequence (xn ). If

we bend [0, 1) into a circle, then points representing consecutive members of (xk+th )
will be spaced on the circle by the arcs of length δ.
Therefore, as all xk+th are distinct, and the arc between the points representing
a and b is of length ² > δ, at least one of xk+th must be on the arc representing
(a, b).8 This completes the proof of the statement that K is dense in [0, 1).
Let us show that K is dense in R. Let (a, b) ⊂ R, a < b. We want to show that at
least one point of K is in (a, b). Consider interval (a−bac, b−bac). As 0 ≤ a−bac =
{a} < 1, we have (a − bac, b − bac) ∩ (0, 1) = ({a}, c) where c = min{b − bac, 1}. As
K is dense in [0, 1), there exist s, t ∈ Z such that sα + t ∈ ({a}, c). Then
a < (sα + t) + bac = sα + (t + bac) < b.
As t + bac ∈ Z, (sα + t) + bac ∈ K ∩ (a, b), and the proof is finished. ¤

Let us explain that the results obtained in the proof above imply the Dirichlet’s
theorem. We will use the notations and ideas from our proof of the Kronecker’s
theorem. There we have explained that for any N > 1, there are integers k and
k + h, such that 1 ≤ k < k + h ≤ N + 1, and |xk+h − xk | < 1/N , which is equivalent
to |hα − (bxk+h c − bxk c)| < 1/N . Setting b = h and a = |bxk+h c − bxk c|, we can
rewrite the last inequality as |bα − a| < 1/N , or |α − a/b| < 1/(bN ). As 1 ≤ b ≤ N ,
bN ≥ b2 , and so |α − a/b| < 1/b2 for some 1 ≤ b ≤ N . Thus we proved that for
any positive integer N > 1 there exist at least one rational number a/b such that
a 1 1
|α − | < ≤ 2 and 1 ≤ b ≤ N.
b bN b
Note that the numbers a and b in this inequality can be assumed to be coprime.
Indeed, if d = gcd(a, b) > 1, and a = pd, b = qd, then
p a 1 1 1
|α − | = |α − | < ≤ 2 < 2 and 1 ≤ p < b ≤ N.
q b bN b p
Now we explain that this implies that there are infinitely many rationals a/b such
that |α − a/b| < 1/b2 .
If the found a/b corresponded to N > 1, consider an integer N 0 such that
1/N 0 < |bα − a| < 1/N . Such N 0 exists since |bα − a| > 0 due to the irrationality
8One can think of a point jumping around the circle with circumference length 1 in the same
direction, each jump having length equal δ < ². Then for any arc of this circle of length ², at least
one jump will land in it.
30 FELIX LAZEBNIK

of α. As N 0 > N , N 0 > 1, and we know that there exist coprime integers a0 , b0 such
that |b0 α − a0 | < N10 and 1 ≤ b0 ≤ N 0 . This implies that |b0 α − a0 | < |bα − a|.
If a/b = a0 /b0 , then a0 = ±a and b0 = ±b, as two fractions are in the lowest
terms. As both b and b0 are positive, a0 = a and b0 = b, a contradiction with
|b0 α − a0 | < |bα − a|. Hence a0 /b0 6= a/b, and we have shown the existence of a new
fraction a0 /b0 satisfying |α − a0 /b0 | < 1/(b0 )2 . Applying similar argument to a0 /b0 ,
we obtain a fraction a00 /b00 distinct from a0 /b0 and satisfying the inequality. It is
also distinct from a/b as it is in the lower terms and
|b00 α − a00 | < |b0 α − a0 | < |bα − a|.
Now it is clear that continuing this way we can show that there are infinitely many
distinct rationals a/b such that |α − a/b| < 1/b2 . This ends the proof of Theorem
2.4. ¤

2.12.
√ Algebraic
√ √and√transcendental
√ numbers. Some irrational numbers, like
5
3, 1 − 3, or 3 + 5 3/2 − 6 2 + 51, are solutions of polynomial equations with
integer coefficients, i.e., of the the form
an xn + an−1 xn−1 + · · · + a1 x + a0 = 0, all ai ∈ Z, an 6= 0.
For the first number a polynomial can be x5 −3 = 0, for the second — x2 −2x−8 = 0,
and for the third — it can be worked out. Numbers of this type are called real
algebraic numbers. Clearly, every rational is algebraic: m n is a root of nx + m = 0.
A real number which is not algebraic is called transcendental. The existence of
transcendental numbers is not a trivial matter, but it follows easily from the facts
that R is uncountable and the set of all algebraic numbers is countable (Cantor’s
argument, 1874). Therefore there are “much more” transcendental numbers than
algebraic ones.
Liouville was the first who proved a theorem (in 1844) which enables us to
produce examples
P∞ of transcendental numbers. By using it, one can easily show that
the number n=1 10−n! is transcendental.
It is much harder to prove that some particular well-known numbers are tran-
scendental. It is known, for example, that e, π, eπ , log2 3, arcsin(1/3) are transcen-
dental. And it is not known whether π e , 2e , e + π, ln(ln 2) are transcendental. See
[15], [2] , and [21] for much more on this subject.
 NOTES ON RATIONAL AND REAL NUMBERS 31

Exercises for Section 2.

Assume that you know what is the field of real numbers R, and its model via
decimal fractions. If you need it, you can use as known that periodic decimals
correspond precisely to the elements of Q and non-periodic decimals – to irrational
numbers. We assume that N ⊂ Z ⊂ Q ⊂ R. In most problems below, one does not
have to use the fact that reals can be represented by decimals. It is often useful
to represent a non-zero rational by a fraction m n , where m, n ∈ Z, n 6= 0, and
gcd(m, n) = 1. We call a point x = (x1 , x2 , . . . , xn ) ∈ Rn a rational point if all xi
are rational. Harder (in my opinion) problems are marked by ∗ , and even harder –
by ∗∗ .

2.1. Thinking about reals as infinite decimal fractions, prove that between any
two distinct real numbers there are infinitely many rational numbers and
infinitely many irrational numbers.
2.2. (i) Prove that the decimal .12345678910111213 . . . is not periodic.
P∞ 2
(ii) Prove that the decimal 0.1001000010 . . . = i=1 10−n is not periodic.
∗
2.3. Prove that for a periodic decimal, the principal period divides any other
period.
2.4. Suppose n and k are positive integers,
√ k ≥ 2, and let n be not the kth power
of another integer. Assuming that k n is real, prove that it is irrational.
2.5. Let a, b ∈ R and a, b > 0. What can be said (rational or irrational) about
a + b , a · b , and ab if
(i) both a and b are rationals?
(ii) a is rational and b is irrational?
(iii) a is irrational and b is rational?
(iv) both a and b are irrationals?
2.6. There is no a standard notation for the set of all irrational reals. We have
not introduce one either. Why, in your opinion, this is the case?
2.7. Is there a line in R2 passing through the origin and distinct from both x-
and y-axes such that
(i) it contains no point distinct from the origin with both integer coordi-
nates?
(ii) it contains no rational point distinct from the origin?
(iii) every point of the line distinct from the origin has exactly one rational
coordinate?
(iv) it contains no point with both coordinates irrational?
2.8. In R2 , consider the integer lattice Z2 = {(x, y) : x, y ∈ Z}. Is there an
equilateral triangle having three of the lattice as its vertices? Prove your
answer.
2.9. In R2 , consider the integer lattice Z2 = {(x, y) : x, y ∈ Z}. Is it true that
there exists a point in the plane which has different distances from all points
of Z2 ?
32 FELIX LAZEBNIK

2.10. In R2 , consider the integer lattice Z2 = {(x, y) : x, y ∈ Z}. Prove that for
every n ∈ N, there exists a circle in the plane having in its interior exactly
n lattice points.
2.11. Consider Problems 9 and 10 for n ≥ 3, i.e., for Rn and Zn .
√ √
2.12. 3 2 cannot be represented as a + b 2 with a, b ∈ Q.
2.13. Determine
√ whether the following real numbers are rational or irrational:
√
(a) √2 + √3 ; √
(b) 2 + 3 + 5 ;
(c) log10 2 ;
(d) ∗ cos 1◦ , sin 1◦ ;
(e) tan 1◦ ;
(f) ∗ π1 arccos 13 ;
(g) 1 + 21/3 + 22/3 ;
(h) 1 + 31/5 + 32/5 + 33/5 + 34/5 .
2.14. Determiner
whether the following real numbers are rational or irrational:
q p √
(a) an = 2 + 2 + 2 + . . . + 2 (n radicals), n ≥ 1.
q p q p
(b) α = 3 6 + 847/27 + 3 6 − 847/27
∗
2.15. Let 1 < p1 < p2 < . . . < pn be a sequence of primes. Is it true that
√ √ √
p1 + p2 + · · · + pn is irrational? Prove your answer.
2.16. Prove that the first√999 decimal digits after the period in the decimal rep-
resentation of (6 + 37)999 are all zeros.
2.17. Prove that if a line in R2 contains two rational points, then it contains
infinitely many rational points.
2.18. Given a circle in R2 . What is the greatest number of rational points can
be on it if its center is not a rational point?
∗
2.19. Is there a circle in R2 with exactly 51 rational points on it? What about
with exactly 56 rational points on it?
2.20. Consider the unit circle S 1 : x2 + y 2 = 1, and a point A : (−1, 0) on it.
(i) Let B be a rational point of S 1 distinct from A, and let M : (0, r) be
the intersection of line AB with the y-axis. Prove that r ∈ Q.
(ii) Let M : (0, r) where r ∈ Q, and let B be the intersection of line AM
with S 1 . Then B is a rational point of S 1 .
(iii) Parts (i) and (ii) can be combined in the following statement: there
exists a bijective correspondence between Q and the set of all rational
points of S 1 \ {A}. Prove the set of all rational points on S 1 coincides
with the following set:
n³ 1 − r2 2r ´ o
{(−1, 0)} ∪ , : r ∈ Q .
1 + r2 1 + r2
We say that the set of all rational points of S 1 \ {A} is parametrized
by Q, or that we have found a general form of rational points on S 1 .
2.21. Using the ideas described in Problem 20, parametrize the set of all rational
points of the ellipse x2 /4 + y 2 /9 = 1.
 NOTES ON RATIONAL AND REAL NUMBERS 33

2.22. Describe all rational numbers x for which 3x2 − 5x + 9 is a square of a

rational number.
∗
2.23. Prove that there are no rational points on the circle x2 + y 2 = 3.
2.24. Show that for any n ∈ N, one can find n points in a plane, not all colinear,
such that the distance between any two of them is an integer.
∗∗
2.25. Suppose we have an infinite set of points in a plane with the integer
distances between any two of the points. Prove that the points are colinear.

∗∗
2.26. Can one find an infinite subset of rational points of the unit circle
x2 + y 2 = 1 such that all distances between them are also rational?
2.27. A man walks along the x-axis towards +∞. The length of his step is an
irrational number α. Each integer point n is surrounded with a “narrow
ditch” of radius ² > 0: (n − ², n + ²). Prove that sooner or later the man
will step into the ditch.
2.28. Prove that for some n ∈ N, the decimal representation of 2n begins with
9999999999 (ten nines).
2.29. Assuming that π is irrational, prove that for some integer n 6= 0,
| sin n| < 0.000000001.
2.30. In R , consider the integer lattice Z2 = {(x, y) : x, y ∈ Z} and a line
2

y = mx, where m is irrational. For a given ² > 0, surround every point of

the lattice with a disc centered at the point and of radius ². Prove that no
matter how small ² is, the line intersects infinitely many of such discs.
2.31. Prove that Theorem 2.4 does not hold for rational α.
∗
2.32. Let α and β be two positive irrational numbers such that 1/α + 1/β = 1.
Let A = {bnαc : n ∈ N} and B = {bnβc : n ∈ N}. Prove that the sets A
and B partition N, which means that A ∪ B = N and A ∩ B = ∅.
2.33. (i)∗∗ Prove that for n > 10 there is no rational number m/n such that
1 √ m 1
− 3 < 3− < 3.
n n n
(ii) Find all rational numbers m/n which satisfy the inequality in part (i).
34 FELIX LAZEBNIK

Hints and Answers to Some Exercises from Section 1

1.1 Use definitions and an example from the section.

1.2 Use definitions and an example from the section.
1.3 Use definitions and an example from the section.
1.4 Use the corresponding properties of integers.
1.5 Use definitions and an example from the section.
1.6 Hint: Use ideas of proofs in subsection 1.6. Or use Problem 1.14.
1.7 Hint: (i) Use ideas of proofs in subsection 1.6. Or use Problem 1.14. (ii)
Hint: If r = m/n, then 3m = 7n .
1.8 First find ten solutions of x2 + y 2 = z 2 in integers. Can use computer, if
you wish.
1.9 Just check.
1.10 Hint: first show that if x ∈ Z, then x2 ≡ 0 or 1 mod 3. Then show that if
3|(x2 + y 2 ) for x, y ∈ Z, then both x and y are divisible by 3. Reduce the
problem about rational solutions to the problem about integer solutions.
1.11 Just check.
1.12 (a) Yes (b) Yes (c) No.
1.13 (i) 4 (ii) 3 4.031313131 . . .
1.14 Hint: Let p/q be a solution of the equation. We may assume gcd(p, q) = 1.
Substitute p/q into the equation, and get rid of the denominators. Proof
of this fact can be found in many algebra books.
1.15 Answers: (i) 1, 1/2, 3. (ii) No rational solutions.
1.16 Hint 1: Let r = a/b ∈ Q. Check that t = (2a + 2b)/(a + 2b) satisfies the
required property. (I learned this idea from Elizabeth Sieminski (F.L.).)
Hint 2: Let t = r + x. Take a rational number x such that 0 < x <
min{1, (2 − r2 )/(2r + 1)}. Prove that t2 < 2. A well known argument; easy
to locate.
1.17 Find a bijection between N and N × N. If you cannot do it, try to read
about it in a book or on the web. Then find a bijection f : N → Q+ , where
Q+ denote the set of positive rational numbers. Then find a bijection
g : Q+ → Q. The composition of f and g is an example of desired the
desired bijection.
1.18 No.
1.19 (a) Hint: f (1) = f (1/n + 1/n + . . . + 1/n) = nf (1/n).
(b) Hint: What is f (0)? Is there any relation between f (a) and f (a−1 , if
a 6= 0? Think about the prime factorization of integers.
(c) Hint: Use parts (a) and (b).
 NOTES ON RATIONAL AND REAL NUMBERS 35

Hints and Answers to Some Exercises from Section 2.

2.1. Try to use an idea from Section2. Or begin by showing that there are
arbitrarily small positive rational and arbitrarily small positive irrational
numbers.
2.2. Both (i) and (ii) are similar to a problem from Section2.
2.3. Suppose p is the principal period and p0 is a period. Let p0 = qp + r, where
0 ≤ r < p. Show that if 0 < r, then r is a period.
2.4. Hint: Find a similar example in the text.
2.5. Answers:
(i) Rational; rational; either.
(ii) Irrational; irrational; either.
(iii) Irrational; irrational; either.
(iv) Either; either; either.
2.6. The set of irrational numbers is not closed under arithmetic operations,
like addition or multiplication. This makes it a much less interesting object
from the algebraic point of you when they are compared to N, Z, Q or R.
2.7. (i) Yes.
(ii) Yes.
(iii) No.
(iv) No. √
◦
2.8. No. Use that√ tan 60 = 3.
2.9. Yes, and ( 2, 1/3) is such a point. If the reader is familiar with the car-
dinalities of infinite sets, the existence also follows from the countability of
perpendicular bisectors defined by all pairs of lattice points.
2.10. Follows from Problem 9.
2.11. For n = 3, one can consider a point (log 2/ log 7, log 3/ log /7, log 5/ log 7).
Generalize. Another approach for those who are familiar with existence
of transcendental numbers: consider a point (α, α2 , . . . , αn ), where α is
transcendental.
2.12. Assume the contrary.
2.13. (a) Irrational. Can use one of several ideas for a similar problem discussed
in Section 2.
(b) Irrational. Can use one of several ideas for a similar problem discussed
in Section 2.
(c) Irrational. Use uniqueness of prime factorization.
(d) Both irrational. Show that cos nα, n ∈ N, can be represented as a
polynomial of cos α with integer coefficients. Show that sin nα, n ∈ N
and odd, can be represented as a polynomial of sin α with integer
coefficients.
(e) Irrational. The ideas in the hint to the previous problem can be used,
but there are also other ways.
(f) Let x = cos−1 (1/3). Prove that for every positive integer n,
an
cos nx = n ,
3
where an is an integer and gcd(an , 3) = 1.
(g) Irrational.
(h) Irrational.
36 FELIX LAZEBNIK

√
2.14. (a) Irrational. Hint: for n ≥ 2, an = 2 + an−1 .
(b) Rational. Hint: (a + b)3 = a3 + b3 + 3ab(a + b).
2.15. Yes. Generalize and √ use induction on√n.
2.16. First show√that (6 + 37)999 + (−6 + 37)999 is an integer. Then show that
0 < −6 + 37 < 1/10.
2.17. Consider an equation of the line.
2.18. At most two. Show that it is possible to have 0, or exactly 1, or exactly
two rational points.
2.19. For 51 points, the answer is No. If there are more than 2 rational points,
then the center is a rational point. Consider symmetries with respect to
the center.
The answer to the second question is No, and the reason is deeper. Prove
that if there are more than 2 rational points on the circle, then it contains
infinitely many of them.
2.20. (i) Consider an equation of the line.
(ii) Consider an equation of the line.
(iii) There is nothing to add. Maybe this: does the result remind you a
very useful formula from trigonometry? Expressing sin x and cos x in
terms of tan(x/2)? Can you explain the connection?
2.21. Either repeat the argument from Problem 20, or reduce the problem to
Problem 20 by a simple change of variables.
2.22. Using the ideas described in Problem 20, parametrize the set of all rational
points of the hyperbola 3x2 − 5x + 9 = y 2 . Point (0, 3) is on the curve.
2.23. First, explain that a remainder of the division of square of an integer by
3 can be only 0 or 1. Then explain that if the sum of squares of two
integers is divisible by 3 then each of them is divisible by 3. Then assume
that x2 + y 2 = 3 has a rational solution, represent it as a pair of reduced
fractions, and obtain a contradiction with the previous statement.
2.24. One can use Problem 20, for example.
2.25. Assume the contrary. Use the fact that in any triangle ABC, |AB − AC| <
BC. Then show that the set can be represented as the union of points of
intersection of finitely many hyperbolas, and that two distinct hyperbolas
can intersect in at most 4 points.
2.26. Prove that there exists α > 0 such that α/π is irrational, but both cos(α/2)
and sin(α/2) are rational. Then consider A = {(cos nα, sin nα) : n ∈ N}.
2.27. Use Theorem 2.5.
2.28. The decimal representation of 2n begins with ten nines if and only if there
exists a positive integer k such that
9999999999 · 10k < 2n ≤ 10k+10 ,
which is equivalent to
log10 (0.9999999999) < n log10 2 − (k + 10) < 0.
Now use Theorem 2.5.
2.29. First show that | sin x − sin y| ≤ |x − y| for any real x, y. Then explain that
there exist nonzero integers m, n such that |n − πm| < 10−9 .
 NOTES ON RATIONAL AND REAL NUMBERS 37

2.30. We have to show that the distance from some lattice point (a, b) to the line
will be less than ². Using the formula for the distance from a point to a
line, this is equivalent to proving the existence of integers a, b such that
|ma − b| p
√ < ² ⇔ |ma − b| < ²0 , where ²0 = ² m2 + 1.
m2 + 1
2.31. Easy.
2.32. Hint: the statement is equivalent to the following: for each positive integer
N exactly one member of A ∪ B is in the interval (N, N + 1). A great
proof for the latter statement is obtained by counting how many members
of A ∪ B are less than N .
2.33. See Niven [15].
38 FELIX LAZEBNIK

References
[1] John D’Angelo and D.B. West, Mathematical Thinking: Problem-Solving and Proofs, Second
edition: Prentice Hall 2000.
[2] A. Baker, Transcendental Number Theory, Cambridge University Press, 1990.
[3] G. Birkhoff and S. MacLane, A Survey of Modern Algebra, The Macmillan Comp., New York,
1941.
[4] A. Bogomolny: http://www.cut-the-knot.org/proofs/sq− root.shtml
[5] R. Courant and H. Robins, What is Mathematics, Oxsford University Press, 1978.
[6] H. Cohen, Number Theory: Volume I: Tools and Diophantine Equations (Graduate Texts in
Mathematics) Springer 2007.
[7] T. Gowers, Mathematics: A Very Short Introduction, Oxford Univ. Press, 2002.
[8] I. Grattan-Guinness, The Rainbow of Mathematics: A History of the Mathematical Sciences,
W.W. Norton & Company, 2000.
[9] G.H. Hardy and E.M. Wright,An Introduction to the Theory of Numbers, Fifth Edition,
Clarendon Press, 1990.
[10] D. W. Henderson, http://www.math.cornell.edu/∼ dwh/papers/sulba/sulba.html
[11] A. Ya. Khinchin, Continued Fractions, Dover Publ., 1997.
[12] A.A. Leman, Collection of Moscow mathematical competitions problems, Moscow,
Prosveshchenie, 1965. In Russian.
[13] A.A. Lyapunov, Real Numbers, M athematicheskoe Prosveshchenie, v. 2, Moscow, 1957, 149–
156. In Russian.
[14] Y. I. Manin, Cubic Forms: Algebra, Geometry, Arithmetic, North-Holland, 1974; second
enlarged edition 1986.
[15] I. Niven, Numbers: Rational and Irrational, Random House, 1961.
[16] I. Niven, Irrational Numbers, MAA, 3rd printing, 1967.
[17] J J O’Connor and E F Robertson,
(1) The real numbers, Pythagoras to Stevin,
http://www-groups.dcs.st-and.ac.uk/∼ history/HistTopics/Real− numbers− 1.html
(2) Stevin to Hilbert, http://www-groups.dcs.st-and.ac.uk/∼ history/HistTopics/Real− numbers− 2.html
(3) The real numbers: Attempts to understand, http://www-groups.dcs.st-
and.ac.uk/∼ history/HistTopics/Real− numbers− 3.html
[18] H. Rademacher and O. Toeplitz, The Enjoyment of Mathematics, Princeton University Press,
1957.
[19] I.J. Schonberg, Mathematical Time Exposures, MAA, 1982.
[20] D. O. Shklarsky, N. N. Chentzov, I. M. Yaglom, Irving Sussman, The USSR Olympiad
Problem Book: Selected Problems and Theorems of Elementary Mathematics Courier Dover
Publications, 1993.
[21] M. Waldschmitdt, http://people.math.jussieu.fr/∼ miw/articles/pdf/IntroductionDiophantineMethods.pdf
[22] http://en.wikipedia.org/wiki/Square− root− of− 2
[23] Real Number, http://en.wikipedia.org/wiki/Real− number