Skip to content

Chapter 1 — Fundamental Concepts

1.1 — The Real Numbers

Every exact science begins by naming what it is going to speak about.


Every mathematical study needs two things: objects to work with and precise rules for operating on them. In this book, which runs from elementary algebra to the threshold of Calculus, the fundamental object is the real number, and the set of all real numbers is denoted R\mathbb{R}.

This section establishes what R\mathbb{R} is made of and which laws govern its operations. We will proceed in layers. Each number set arises from an operation that the previous one could not carry out, and in this way we will reach the real line with its addition, its multiplication, its order and its notion of distance. What is established here will be used throughout the rest of the book without being justified again.

The first numbers we use are the ones that serve for counting. They form the set of natural numbers, denoted N\mathbb{N}. We can describe it by listing its elements between braces, separated by commas:

N={1,2,3,4,5,… }\mathbb{N} = \{1, 2, 3, 4, 5, \dots\}

The ellipsis is not decoration. It indicates that the list continues without end, following the same pattern, so that N\mathbb{N} is an infinite set. Nor does the sign == express an equality between numbers here; it expresses an equality between sets: it says that the set we call N\mathbb{N} is exactly the one on the right. What “following the same pattern” means rigorously is a harder question than it seems. We answer it in §10.1, where the natural numbers are defined without an ellipsis.

Note. We follow the usual convention of algebra and analysis: zero is not a natural number. In other areas, such as logic or computer science, it is convenient to include it, and one then writes N0=N∪{0}\mathbb{N}_0 = \mathbb{N} \cup \{0\} to distinguish the two sets. In this book N\mathbb{N} always starts at 11.

With the natural numbers we can add and multiply without leaving the set, but we cannot always subtract: 3−53-5 is not a natural number. For subtraction to be always possible, we must add zero and the negative numbers. This extension came late in history and met resistance. Negative numbers were first used to keep track of debts, and centuries passed before they were accepted as numbers in their own right.

Adding to the natural numbers zero and the opposite of each natural number, we obtain the set of integers, denoted Z\mathbb{Z}, from the German Zahlen, “numbers”. Listing its elements,

Z={…,−3,−2,−1,0,1,2,3,… }\mathbb{Z} = \{\dots, -3, -2, -1, 0, 1, 2, 3, \dots\}

and, if we want to highlight its structure,

Z=Z−∪{0}∪Z+\mathbb{Z} = \mathbb{Z}^- \cup \{0\} \cup \mathbb{Z}^+

where Z+\mathbb{Z}^+ are the positive integers, which coincide exactly with N\mathbb{N}, and Z−\mathbb{Z}^- the negative integers. The symbol ∪\cup is read “union”: an element belongs to A∪BA \cup B if it belongs to AA, to BB, or to both. The expression {0}\{0\} denotes the set whose only element is zero. It should not be confused with the number 00: we can think of {0}\{0\} as a box that contains zero. The box has one element; zero is that element.

In Z\mathbb{Z} subtraction no longer poses a problem, but division does: 1÷31 \div 3 is not an integer. Sharing a quantity into equal parts, or measuring a length with a unit that does not fit a whole number of times, forces us to introduce fractions.

A rational number is a number that can be written as a quotient of two integers. The set of rational numbers is denoted Q\mathbb{Q}, for quotient, and we define it in set-builder notation:

Q={mn  |  m,n∈Z, n≠0}\mathbb{Q} = \left\{ \frac{m}{n} \;\middle|\; m,n \in \mathbb{Z},\ n \neq 0 \right\}

Set-builder notation does not list the elements; it describes them by the property they satisfy. The vertical bar ∣\mid is read “such that” and introduces that property. The symbol ∈\in indicates membership: "m∈Zm \in \mathbb{Z}" is read ”mm belongs to Z\mathbb{Z}”. The symbol ≠\neq negates equality. The whole expression is thus read “the set of fractions m/nm/n such that mm and nn are integers and nn is different from zero”.

The condition n≠0n \neq 0 is indispensable: it excludes division by zero, which is not defined. Note also that different fractions can represent the same number, such as 12\frac{1}{2} and 24\frac{2}{4}. The criterion for deciding when two fractions are equal appears in part 6 of this section.

Between any two rational numbers there is always another, their average for instance, so it might seem that fractions already fill the line. They do not. Greek mathematicians discovered that the diagonal of a square of side 11 cannot be measured by any fraction of that side. The discovery was so disturbing that tradition surrounded it with legends, and it forced the admission that there are lengths that are not quotients of integers.

Before giving the definition, a remark on the order in which we present things. So far we have used R\mathbb{R} only intuitively, as the complete number line. Its definition as a set comes in the next part, and that definition relies in turn on the irrational numbers. There is no vicious circle here, only a choice of order. First we characterize the irrational numbers by a property of their own, which does not require R\mathbb{R} to have been defined. Then, once R\mathbb{R} is established, we also describe them as a part of R\mathbb{R}.

An irrational number is a real number that cannot be written as a quotient of integers. Equivalently, it is a number whose decimal expansion is neither finite nor periodic. That the two descriptions agree, that is, that the rational numbers are exactly the numbers with a finite or periodic decimal expansion, is proved in §10.4. The set of irrational numbers is denoted I\mathbb{I}, or also Qc\mathbb{Q}^c, the complement of Q\mathbb{Q}. Once R\mathbb{R} is available, we can write it as follows:

I={x∈R∣x∉Q}=R∖Q\mathbb{I} = \{x \in \mathbb{R} \mid x \notin \mathbb{Q}\} = \mathbb{R} \setminus \mathbb{Q}

The symbol ∉\notin negates membership. The symbol ∖\setminus denotes set difference: A∖BA \setminus B consists of the elements of AA that do not belong to BB. The formula is read “the irrational numbers are the real numbers that are not rational”.

Saying that irrational numbers exist is not enough; we must exhibit one. The proof that follows is one of the oldest in mathematics, and it is also the first in this book.

Proposition. There is no rational number whose square is 22.

Proof. Suppose, for the sake of contradiction, that there is a rational number qq with q2=2q^2 = 2. Since (−q)2=q2(-q)^2 = q^2, we may assume that qq is positive and write it as q=m/nq = m/n, with mm and nn positive integers. Every fraction can be simplified, by dividing numerator and denominator by their common factors, until none remain. We therefore choose mm and nn with no common factors. From q2=2q^2 = 2 we obtain

m2n2=2⟹m2=2n2\frac{m^2}{n^2} = 2 \quad\Longrightarrow\quad m^2 = 2n^2

The last equality says that m2m^2 is even. This forces mm to be even. Indeed, if mm were odd, we would have m=2j+1m = 2j+1 for some integer jj, and then m2=4j2+4j+1=2(2j2+2j)+1m^2 = 4j^2 + 4j + 1 = 2(2j^2+2j) + 1 would be odd. We therefore write m=2km = 2k with kk an integer, and substitute:

(2k)2=2n2⟹4k2=2n2⟹n2=2k2(2k)^2 = 2n^2 \quad\Longrightarrow\quad 4k^2 = 2n^2 \quad\Longrightarrow\quad n^2 = 2k^2

Now it is n2n^2 that turns out to be even, and by the same argument nn is even as well. But we had chosen mm and nn with no common factors, and we have just seen that both are divisible by 22. The contradiction comes solely from having assumed that qq existed. Therefore no rational number has a square equal to 22. ■\blacksquare

The proposition says that if there is a positive real number whose square is 22, the number we write 2\sqrt{2} and which measures the diagonal of the square of side 11, then that number is irrational: 2∉Q\sqrt{2} \notin \mathbb{Q}. That such a number exists in R\mathbb{R} is not obvious, and part 9 of this section shows where it comes from. The method of the proof, assuming the opposite of what we want to prove until we reach an absurdity, is called reductio ad absurdum, or proof by contradiction, and we will use it many times.

We now bring the two families together. The set of real numbers is the union of the rational and the irrational numbers:

R=Q∪I\mathbb{R} = \mathbb{Q} \cup \mathbb{I}

The union is disjoint: no number is both rational and irrational, which is written Q∩I=∅\mathbb{Q} \cap \mathbb{I} = \emptyset. The symbol ∩\cap is read “intersection” and designates the elements common to two sets. The symbol ∅\emptyset designates the empty set, the only set with no elements. Two sets whose intersection is empty are called disjoint.

The sets we have built are thus linked in a chain, each contained in the next:

N⊂Z⊂Q⊂R\mathbb{N} \subset \mathbb{Z} \subset \mathbb{Q} \subset \mathbb{R}

The symbol ⊂\subset is read “is contained in”. The irrational numbers occupy exactly what Q\mathbb{Q} leaves free inside R\mathbb{R}.

The most important question remains open: why are the irrational numbers needed? What exactly does Q\mathbb{Q} lack? The answer, the property that distinguishes R\mathbb{R} from Q\mathbb{Q}, is developed in part 9 of this section.


§6. The algebraic properties and the order of ℝ

Section titled “§6. The algebraic properties and the order of ℝ”

On R\mathbb{R} there are two operations, addition and multiplication. We will not justify their properties from something more elementary: we take them as a starting point. They are axioms, and the algebra of the whole book is deduced from them.

Both operations are commutative: the order of the terms or of the factors does not change the result.

a+b=b+aab=baa + b = b + a \qquad\qquad ab = ba

Both are also associative, which allows us to write sums and products of three or more terms without indicating how they are grouped:

(a+b)+c=a+(b+c)(ab)c=a(bc)(a+b)+c = a+(b+c) \qquad\qquad (ab)c = a(bc)

Distributivity is the property that links the two operations. Multiplying a sum by a number is the same as multiplying each term and adding the products:

a(b+c)=ab+aca(b+c) = ab + ac

Each operation has an identity element, which leaves unchanged whatever it operates on, and each number has an inverse, which undoes the operation. Zero is the identity for addition, a+0=aa + 0 = a, and every real number aa has an opposite −a-a with a+(−a)=0a + (-a) = 0. With it we define subtraction, a−b:=a+(−b)a - b := a + (-b). One is the identity for multiplication, a⋅1=aa \cdot 1 = a, and every real number a≠0a \neq 0 has a multiplicative inverse 1/a1/a with a⋅(1/a)=1a \cdot (1/a) = 1. With it we define division, a÷b:=a⋅(1/b)a \div b := a \cdot (1/b), provided b≠0b \neq 0. It is also assumed that 0≠10 \neq 1. The symbol :=:= is read “is defined as”.

The opposite of each number is unique. If a+b=0a + b = 0 and a+c=0a + c = 0, then

b=b+(a+c)=(b+a)+c=0+c=cb = b + (a + c) = (b + a) + c = 0 + c = c

using the identity, associativity and commutativity. In the same way one proves that the multiplicative inverse is unique.

Signs. The rules of signs are not conventions to be memorized: they are deduced from the axioms. First, a⋅0=0a \cdot 0 = 0 for every aa. Indeed, by distributivity, a⋅0=a(0+0)=a⋅0+a⋅0a \cdot 0 = a(0+0) = a \cdot 0 + a \cdot 0, and adding the opposite of a⋅0a \cdot 0 to both sides leaves 0=a⋅00 = a \cdot 0. Second, (−1)a=−a(-1)a = -a, because

a+(−1)a=1⋅a+(−1)a=(1+(−1))a=0⋅a=0a + (-1)a = 1 \cdot a + (-1)a = \big(1 + (-1)\big)a = 0 \cdot a = 0

and the opposite of aa is unique. Third, (−1)(−1)=1(-1)(-1) = 1. By the previous rule with a=−1a = -1, the product (−1)(−1)(-1)(-1) is the opposite of −1-1, and that opposite is 11, because (−1)+1=0(-1) + 1 = 0. From these three rules, together with commutativity and associativity, the complete table follows:

(−1)a=−a−(−a)=a(−a)b=a(−b)=−(ab)(-1)a = -a \qquad -(-a) = a \qquad (-a)b = a(-b) = -(ab) (−a)(−b)=ab−(a+b)=−a−b−(a−b)=b−a(-a)(-b) = ab \qquad -(a+b) = -a-b \qquad -(a-b) = b-a

For example, (−a)(−b)=(−1)a(−1)b=(−1)(−1)ab=ab(-a)(-b) = (-1)a(-1)b = (-1)(-1)ab = ab. The others are proved in the same way, and it is worth doing so as an exercise. “Minus times minus is plus” is thus a consequence of distributivity and of the uniqueness of the opposite.

Fractions. For b≠0b \neq 0 we write ab:=a⋅1b\dfrac{a}{b} := a \cdot \dfrac{1}{b}. The rules for fractions are also deduced from the axioms. Assuming all denominators are nonzero, and also cc in the division,

ab⋅cd=acbdab÷cd=ab⋅dc\frac{a}{b}\cdot\frac{c}{d} = \frac{ac}{bd} \qquad\qquad \frac{a}{b}\div\frac{c}{d} = \frac{a}{b}\cdot\frac{d}{c}

When two fractions have the same denominator, their sum is distributivity applied to the factor 1c\frac{1}{c}:

ac+bc=a⋅1c+b⋅1c=(a+b)⋅1c=a+bc\frac{a}{c}+\frac{b}{c} = a\cdot\frac{1}{c} + b\cdot\frac{1}{c} = (a+b)\cdot\frac{1}{c} = \frac{a+b}{c}

When the denominators are different, we multiply each fraction by a fraction equal to 11, dd\frac{d}{d} or bb\frac{b}{b}, to bring them to a common denominator, and apply the previous rule:

ab+cd=adbd+bcbd=ad+bcbd\frac{a}{b}+\frac{c}{d} = \frac{ad}{bd}+\frac{bc}{bd} = \frac{ad+bc}{bd}

In calculations it is convenient to use the least common multiple of the denominators as the common denominator. It is obtained by factoring each denominator into primes and taking each factor with its largest exponent, and this keeps the numbers smaller. Finally, if c≠0c \neq 0 we may cancel a common factor, acbc=ab\dfrac{ac}{bc} = \dfrac{a}{b}, and two fractions are equal exactly when their cross products agree: ab=cd  ⟺  ad=bc\dfrac{a}{b} = \dfrac{c}{d} \iff ad = bc. The symbol   ⟺  \iff is read “if and only if”. This shows, for example, that 12=24\frac{1}{2} = \frac{2}{4}, because 1⋅4=2⋅21 \cdot 4 = 2 \cdot 2.

Order. Real numbers are not only added and multiplied: they are also compared. Order is governed by a second group of axioms. There is a subset of R\mathbb{R}, that of the positive numbers, with two properties. The first is trichotomy: for each real number aa, exactly one of the statements ”aa is positive”, "a=0a = 0" and ”−a-a is positive” holds. The second is that the sum and the product of two positive numbers are positive. We write a>0a > 0 when aa is positive, and we define a<ba < b, read ”aa is less than bb”, as b−a>0b - a > 0. The expression a≤ba \le b means a<ba < b or a=ba = b.

From these axioms come the rules for working with inequalities. If a<ba < b and b<cb < c, then a<ca < c, because c−a=(c−b)+(b−a)c - a = (c-b) + (b-a) is a sum of positive numbers. If a<ba < b, then a+c<b+ca + c < b + c for any cc, because the difference is still b−ab - a. If a<ba < b and c>0c > 0, then ac<bcac < bc, because bc−ac=(b−a)cbc - ac = (b-a)c is a product of positive numbers. On the other hand, if c<0c < 0, the inequality is reversed, ac>bcac > bc, because ac−bc=(b−a)(−c)ac - bc = (b-a)(-c) is positive. Finally, the square of every nonzero real number is positive: if a>0a > 0 it is a product of positive numbers, and if −a>0-a > 0, then a2=(−a)(−a)a^2 = (-a)(-a). In particular 1=12>01 = 1^2 > 0. Section 1.7 develops these rules.


We have already used ∪\cup, ∩\cap, ∅\emptyset and ∖\setminus in building R\mathbb{R}. It is now worth fixing their general definitions. Given two sets SS and TT,

S∪T={x∣x∈S or x∈T}S∩T={x∣x∈S and x∈T}S∖T={x∣x∈S and x∉T}S \cup T = \{x \mid x \in S \text{ or } x \in T\} \qquad S \cap T = \{x \mid x \in S \text{ and } x \in T\} \qquad S \setminus T = \{x \mid x \in S \text{ and } x \notin T\}

The union gathers what is in at least one of the two; the intersection keeps only what they share; the difference removes from SS what is in TT. When S∩T=∅S \cap T = \emptyset, the sets are disjoint.

One kind of subset of R\mathbb{R} appears constantly: the interval, made up of all the real numbers between two endpoints a<ba < b. The notation indicates with parentheses and brackets whether each endpoint is included or not.

NotationSetEndpoints
(a,b)(a,b){x∣a<x<b}\{x \mid a < x < b\}both excluded (open interval)
[a,b][a,b]{x∣a≤x≤b}\{x \mid a \le x \le b\}both included (closed interval)
[a,b)[a,b){x∣a≤x<b}\{x \mid a \le x < b\}includes aa, excludes bb
(a,b](a,b]{x∣a<x≤b}\{x \mid a < x \le b\}excludes aa, includes bb

When the interval has no end on one side, the symbol ∞\infty, read “infinity”, is used, and always with a parenthesis. Infinity is not a real number, so there is nothing to include:

(a,∞)[a,∞)(−∞,b)(−∞,b](a, \infty) \qquad [a,\infty) \qquad (-\infty, b) \qquad (-\infty,b]

For example, [a,∞)[a, \infty) is the set of xx with a≤xa \le x. The extreme case, (−∞,∞)(-\infty,\infty), is all of R\mathbb{R}.


On the line, 33 and −3-3 are at the same distance from zero, one on each side. The absolute value measures that distance while forgetting the side. The absolute value of a real number aa is denoted ∣a∣|a| and is defined by cases:

∣a∣={aif a≥0−aif a<0|a| = \begin{cases} a & \text{if } a \ge 0 \\ -a & \text{if } a < 0 \end{cases}

If aa is nonnegative, its absolute value is aa itself. If it is negative, it is its opposite, which is positive by trichotomy. In both cases ∣a∣≥0|a| \ge 0, as befits a distance.

Four properties follow from the definition:

∣a∣=∣−a∣∣ab∣=∣a∣∣b∣∣ab∣=∣a∣∣b∣ (b≠0)∣an∣=∣a∣n|a| = |-a| \qquad |ab| = |a||b| \qquad \left|\frac{a}{b}\right| = \frac{|a|}{|b|} \ (b\neq 0) \qquad |a^n| = |a|^n

The first three are checked by separating cases according to the signs of aa and bb. For example, if a≥0a \ge 0 and b<0b < 0, then ab≤0ab \le 0 and ∣ab∣=−ab=a(−b)=∣a∣∣b∣|ab| = -ab = a(-b) = |a||b|; the other cases are analogous. The fourth, for a natural exponent nn, results from applying the second repeatedly. The first says that a number and its opposite are equidistant from the origin; the others, that the absolute value respects products, quotients and powers.

A fifth property deserves a place of its own, because the rest of the book relies on it again and again: the triangle inequality,

∣a+b∣≤∣a∣+∣b∣|a+b| \le |a|+|b|

To prove it we use a translation that will appear everywhere. If c≥0c \ge 0, then

∣x∣≤c  ⟺  −c≤x≤c|x| \le c \iff -c \le x \le c

Indeed, if x≥0x \ge 0, the condition ∣x∣≤c|x| \le c is x≤cx \le c, and −c≤x-c \le x holds anyway. If x<0x < 0, the condition is −x≤c-x \le c, that is, x≥−cx \ge -c, and x≤cx \le c holds anyway. The same argument works with strict inequalities: ∣x∣<c  ⟺  −c<x<c|x| < c \iff -c < x < c.

Proof of the triangle inequality. By the previous equivalence with c=∣a∣c = |a| and with c=∣b∣c = |b|, we have −∣a∣≤a≤∣a∣-|a| \le a \le |a| and −∣b∣≤b≤∣b∣-|b| \le b \le |b|. Adding term by term,

−(∣a∣+∣b∣)≤a+b≤∣a∣+∣b∣-\big(|a|+|b|\big) \le a+b \le |a|+|b|

and the equivalence, now read from right to left with c=∣a∣+∣b∣c = |a| + |b|, gives ∣a+b∣≤∣a∣+∣b∣|a+b| \le |a|+|b|. ■\blacksquare

The absolute value of a sum never exceeds the sum of the absolute values. Equality holds when aa and bb have the same sign, or one of them is zero, and strict inequality when they have opposite signs, because then they partly cancel. This inequality, modest in appearance, is the central tool for bounding sums and differences, and it reappears in the study of limits and approximations.

With the absolute value we define the distance between any two points of the line. Given a,b∈Ra, b \in \mathbb{R},

d(a,b)=∣b−a∣d(a,b) = |b-a|

Since b−ab - a and a−ba - b are opposites, the first property gives d(a,b)=d(b,a)d(a,b) = d(b,a): the distance does not depend on the order in which we name the points.

Finally, a translation between inequalities and geometry that we will use constantly. Let ε\varepsilon be a positive number; the Greek letter is read “epsilon” and usually designates a small quantity. The inequality ∣x−a∣<ε|x-a| < \varepsilon says that xx is less than ε\varepsilon away from aa. By the equivalence with strict inequalities, this means −ε<x−a<ε-\varepsilon < x - a < \varepsilon, that is, a−ε<x<a+εa - \varepsilon < x < a + \varepsilon. The inequality therefore describes the open interval centered at aa:

Bε(a)=(a−ε,a+ε)={x∈R∣∣x−a∣<ε}B_\varepsilon(a) = (a-\varepsilon, a+\varepsilon) = \{x \in \mathbb{R} \mid |x-a| < \varepsilon\}

This interval is called the neighborhood of aa of radius ε\varepsilon, and the notation Bε(a)B_\varepsilon(a) recalls the word “ball”. It looks like a minor detail, but it is the language in which we will later say precisely what it means for one point to come as close as we like to another.

The four types of bounded interval and the neighborhood of radius epsilon centered at a

Figure 1.1 — The four types of interval and the neighborhood Bε(a)=(a−ε,a+ε)B_\varepsilon(a)=(a-\varepsilon,a+\varepsilon) just defined.


§9. Completeness of ℝ: bounds, supremum and the axiom that distinguishes it from ℚ

Section titled “§9. Completeness of ℝ: bounds, supremum and the axiom that distinguishes it from ℚ”

Let us return to the question left open in part 5 of this section: what does Q\mathbb{Q} lack in order to be R\mathbb{R}? The usual picture says that Q\mathbb{Q} is full of holes and that the irrational numbers fill them. The picture is correct, but it misleads if taken literally, because the rational numbers are not separated from one another. Between two rational numbers p<qp < q there is always another, their average p+q2\frac{p+q}{2}, and repeating the argument gives infinitely many. This property is called density. The hole in Q\mathbb{Q} does not lie in the distance between its points, but in something subtler: there are sets of rational numbers that do not extend without limit and yet have no exact rational “ceiling”. To say this precisely we need three definitions.

Bounds. Let S⊆RS \subseteq \mathbb{R} be nonempty; the symbol ⊆\subseteq is read “is a subset of”. A number MM is an upper bound of SS if no element of SS exceeds it, that is, if x≤Mx \le M for every x∈Sx \in S. If some upper bound exists, we say that SS is bounded above. Symmetrically, mm is a lower bound of SS if m≤xm \le x for every x∈Sx \in S, and SS is bounded below if it has one. An upper bound need not belong to SS, and if SS has one, it has infinitely many, because every number greater than an upper bound is also one.

Supremum. Among all the upper bounds of SS we are interested in the tightest. The supremum of SS, written sup⁡S\sup S, is the upper bound of SS that is less than or equal to every other upper bound of SS. There is an equivalent way of saying this, which is the one most used in proofs: s=sup⁡Ss = \sup S if and only if ss is an upper bound of SS and, for every ε>0\varepsilon > 0, there is some x∈Sx \in S with x>s−εx > s - \varepsilon.

Let us see why the two formulations say the same thing. If ss is the least upper bound and ε>0\varepsilon > 0, the number s−εs - \varepsilon is less than ss and is therefore not an upper bound, so some x∈Sx \in S exceeds it. Conversely, suppose that ss is an upper bound with that property and let t<st < s. With ε=s−t\varepsilon = s - t there is x∈Sx \in S with x>s−ε=tx > s - \varepsilon = t, so tt is not an upper bound. No upper bound is less than ss, and ss is the least.

When sup⁡S\sup S belongs to SS, it is called the maximum of SS. But a set can have a supremum without having a maximum: the interval (0,1)(0,1) has supremum 11, which does not belong to it. The number 11 is an upper bound, and given ε>0\varepsilon > 0, the larger of the numbers 12\frac{1}{2} and 1−ε21 - \frac{\varepsilon}{2} belongs to the interval and exceeds 1−ε1 - \varepsilon. The symmetric notion, the greatest of the lower bounds, is called the infimum and is written inf⁡S\inf S.

With this vocabulary we state the property that separates R\mathbb{R} from Q\mathbb{Q}.

Supremum axiom (completeness of R\mathbb{R}). Every nonempty subset of R\mathbb{R} that is bounded above has a supremum in R\mathbb{R}.

This statement is not deduced from the algebraic and order axioms of part 6. It is added to them as one more axiom, as fundamental as commutativity. What we can prove is that Q\mathbb{Q} does not have this property: there are nonempty subsets of Q\mathbb{Q}, bounded above, with no rational supremum. That is exactly the hole.

Proposition. The set S={x∈Q∣x>0, x2<2}S = \{x \in \mathbb{Q} \mid x > 0,\ x^2 < 2\} is nonempty and bounded above, but has no supremum in Q\mathbb{Q}.

Proof. SS is nonempty, because 1∈S1 \in S. It is bounded above by 22: if we had x≥2x \ge 2, then x2≥2x≥4>2x^2 \ge 2x \ge 4 > 2, and xx would not be in SS.

We will use a fact about positive numbers: if 0<y≤x0 < y \le x, then y2≤x2y^2 \le x^2, because multiplying y≤xy \le x by yy and by xx gives y2≤xy≤x2y^2 \le xy \le x^2. Consequently, if x,y>0x, y > 0 and x2<y2x^2 < y^2, then x<yx < y.

Now suppose, for the sake of contradiction, that SS has a rational supremum rr. Since 1∈S1 \in S, we have r≥1r \ge 1. By the proposition of part 4, r2=2r^2 = 2 is impossible, so two cases remain, and in each one we reach a contradiction.

Case r2<2r^2 < 2. We will find an element of SS greater than rr, which prevents rr from being an upper bound. Let

h=2−r22r+1h = \frac{2-r^2}{2r+1}

which is rational and positive, because 2−r2>02 - r^2 > 0. Since r≥1r \ge 1, the numerator satisfies 2−r2≤12 - r^2 \le 1 and the denominator satisfies 2r+1≥32r + 1 \ge 3, so that 0<h<10 < h < 1. Multiplying h<1h < 1 by h>0h > 0 we get h2<hh^2 < h, and then

(r+h)2=r2+2rh+h2<r2+2rh+h=r2+h(2r+1)=r2+(2−r2)=2(r+h)^2 = r^2 + 2rh + h^2 < r^2 + 2rh + h = r^2 + h(2r+1) = r^2 + (2-r^2) = 2

The number r+hr + h is rational, positive, and its square is less than 22: it belongs to SS and is greater than rr. Hence rr is not an upper bound, a contradiction.

Case r2>2r^2 > 2. We will find an upper bound of SS less than rr, which prevents rr from being the least. Let

r′=r−r2−22r=r2+22rr' = r - \frac{r^2-2}{2r} = \frac{r^2+2}{2r}

which is rational and positive, and less than rr because it is obtained by subtracting a positive quantity from rr. Its square exceeds 22:

(r′)2−2=r4+4r2+44r2−8r24r2=r4−4r2+44r2=(r2−2)24r2>0(r')^2 - 2 = \frac{r^4 + 4r^2 + 4}{4r^2} - \frac{8r^2}{4r^2} = \frac{r^4 - 4r^2 + 4}{4r^2} = \frac{(r^2-2)^2}{4r^2} > 0

Now every x∈Sx \in S satisfies x2<2<(r′)2x^2 < 2 < (r')^2, and since xx and r′r' are positive, the previous fact gives x<r′x < r'. Thus r′r' is an upper bound of SS less than rr, another contradiction.

Both cases lead to an absurdity, so SS has no supremum in Q\mathbb{Q}. ■\blacksquare

In R\mathbb{R}, on the other hand, the supremum axiom guarantees that SS has a supremum. And the same two-case reasoning, applied now to the set of positive real numbers whose square is less than 22, proves something more important: that there is a positive real number whose square is exactly 22. In §1.2 we carry out that proof for roots of any index. That number is 2\sqrt{2}, and it exists in R\mathbb{R} precisely because R\mathbb{R} satisfies the supremum axiom.

That is the underlying reason why the irrational numbers are needed. They are not there to complete a list of numbers that were “missing”, but so that every bounded set has, without exception, an exact ceiling. Without them, the line would have gaps in places as concrete as the diagonal of a square.