1.3 The completeness axiom

In the system of rational numbers, the least upper bound for S:={a:a22}S:=\{a\in\mathbb{Q}:a^{2}\leq 2\} is missing. In order to find it, we need to pass to a larger system of numbers: this is where the completeness axiom and the real numbers \mathbb{R} come in.

Axiom 1.15 (Completeness).

Every nonempty subset AA\subseteq\mathbb{R} which is bounded above has a least upper bound in \mathbb{R}.

Since this is an axiom, it does not need to be proved. We will simply define the real numbers \mathbb{R} to be a system of numbers which satisfy this property. In Section 1.6, we will derive consequences from this axiom to establish facts about the real numbers.

Note that we cannot replace \mathbb{R} with \mathbb{Q} in 1.15; that is, the number system \mathbb{Q} does not satisfy the completeness axiom. We can see this from Example 1.14, which shows that the set S:={a:a22}S:=\{a\in\mathbb{Q}:a^{2}\leq 2\}\subset\mathbb{Q} is bounded above and nonempty but does not have a least upper bound in the rational number system \mathbb{Q}.

Remark 1.16.

Note that the completeness axiom applies only to AA which are nonempty. This is because every number xx\in\mathbb{R} is an upper bound for ∅︀\emptyset in \mathbb{R}, and so ∅︀\emptyset cannot possibly have a least upper bound. Indeed, given any xx\in\mathbb{R}, it vacuously holds that axa\leq x for all a∅︀a\in\emptyset, since there are no elements in the empty set to check the condition on!

We shall invest some time becoming acquainted with the completeness axiom, studying sets where the supremum (i.e., least upper bound) can be explicitly computed.

Intervals and their suprema

In Example 1.2 and Example 1.12, we considered the set A:={a:0a1}A:=\{a\in\mathbb{R}:0\leq a\leq 1\} and showed supA=1\sup A=1. This set is an example of an interval, with the endpoints 0 and 1, and we could represent it as [0,1][0,1], using the interval notation introduced in IMU. In this section, we will look at other examples of intervals, and consider their suprema.

Given a,ba,b\in\mathbb{R} with aba\leq b, we define the sets

(1.1) (1.1) [a,b]:={t:atb},(,b]:={t:tb},[a,):={t:at}.\displaystyle\begin{split}[a,b]&:=\{t\in\mathbb{R}:a\leq t\leq b\},\\ (-\infty,b]&:=\{t\in\mathbb{R}:t\leq b\},\\ [a,\infty)&:=\{t\in\mathbb{R}:a\leq t\}.\end{split}

We say II\subseteq\mathbb{R} is a closed interval if it is of the form of one of the three sets in (1.1), I=I=\mathbb{R} or I=∅︀I=\emptyset.

Warning 1.17.

Here we just think of ‘±\pm\infty’ as symbols used in the notation rather than any actual object. In particular, \infty and -\infty do not represent numbers in \mathbb{R}.

The argument used in Example 1.2 can be applied to solve the following exercise.

Exercise 1.18.

For a,ba,b\in\mathbb{R} with aba\leq b, show that

sup[a,b]=bandsup(,b]=b.\sup\,[a,b]=b\quad\text{and}\quad\sup\,(-\infty,b]=b.

What about the remaining closed intervals [a,)[a,\infty), \mathbb{R} and ∅︀\emptyset: do they have suprema?

It is useful to visualise closed intervals as line segments on the real line: see Figure 1.1. It is customary to draw vertical straight lines at the endpoints to signify that the endpoints are included in the interval.

Figure 1.1: A closed interval.

Given a,ba,b\in\mathbb{R} with a<ba<b, we define the sets

(1.2) (1.2) (a,b):={t:a<t<b},(,b):={t:t<b},(a,):={t:a<t}.\displaystyle\begin{split}(a,b)&:=\{t\in\mathbb{R}:a<t<b\},\\ (-\infty,b)&:=\{t\in\mathbb{R}:t<b\},\\ (a,\infty)&:=\{t\in\mathbb{R}:a<t\}.\end{split}

We say II\subseteq\mathbb{R} is an open interval if it is of the form of one of the sets in (1.2), I=I=\mathbb{R} or I=∅︀I=\emptyset.22 2 Note that, by definition, \mathbb{R} and ∅︀\emptyset are both open intervals and closed intervals. It is again useful to visualise open intervals as line segments on the real line: see Figure 1.2. Here it is customary to draw round brackets at the endpoints to signify that the endpoints are not included in the interval.

Figure 1.2: An open interval.

In addition, we define the half-open intervals,

(1.3) (1.3) [a,b):={t:at<b},(a,b]:={t:a<tb}.\displaystyle\begin{split}[a,b)&:=\{t\in\mathbb{R}:a\leq t<b\},\\ (a,b]&:=\{t\in\mathbb{R}:a<t\leq b\}.\end{split}

We say II\subseteq\mathbb{R} is an interval if it is either an open interval, a closed interval, or of the form of one of the sets in (1.3).

Example 1.19.

Consider the open interval (0,1):={a:0<a<1}(0,1):=\{a\in\mathbb{R}:0<a<1\}, consisting of all real numbers lying strictly between 0 and 11. We claim that sup(0,1)=1\sup\,(0,1)=1. This looks similar to the set [0,1][0,1] from Example 1.12, but is in fact harder since now 1(0,1)1\notin(0,1). We need to check both parts of Definition 1.10:

  1. 1.

    Since a<1a<1 for all a(0,1)a\in(0,1), we see that 11 is an upper bound for (0,1)(0,1).

  2. 2.

    Suppose xx\in\mathbb{R} and x<1x<1.

    • If x0x\leq 0, then a=1/2(0,1)a=1/2\in(0,1) and a>xa>x, so that xx is not an upper bound for (0,1)(0,1).

    • If 0<x<10<x<1, then the midpoint t:=x+12t:=\frac{x+1}{2} between xx and 11 satisfies x<t<1x<t<1. Moreover, t(0,1)t\in(0,1) and t>xt>x, so that xx is not an upper bound for (0,1)(0,1).

    This shows that if xx\in\mathbb{R} is an upper bound for (0,1)(0,1), then x1x\geq 1.

Thus, by definition, sup(0,1)=1\sup\,(0,1)=1.

Exercise 1.20.

Draw a sketch to illustrate the proof used in Example 1.19, including the interval (0,1)(0,1), the point xx and the midpoint t:=1+x2t:=\frac{1+x}{2}.

Exercise 1.21.

For a,ba,b\in\mathbb{R} with a<ba<b, show that

sup(a,b)=sup(a,b]=sup[a,b)=bandsup(,b)=b.\sup\,(a,b)=\sup\,(a,b]=\sup\,[a,b)=b\quad\text{and}\quad\sup\,(-\infty,b)=b.

What about the remaining open intervals (a,)(a,\infty), \mathbb{R} and ∅︀\emptyset: do they have suprema?

Maxima

The concept of a maximum of a set is closely related to the supremum, although – as we will see – the two are not the same!

Definition 1.22.

Let AA\subseteq\mathbb{R}. We say MM\in\mathbb{R} is a maximum of AA if MAM\in A and xMx\leq M for all xAx\in A. In this case we write M=maxAM=\max A.

In other words, MM is maximum of AA if MAM\in A and MM is an upper bound of AA.

Exercise 1.23.

Let AA\subseteq\mathbb{R}. Show that if a maximum for AA exists, then it must be unique. In other words, if M1M_{1}, M2AM_{2}\in A are maxima for AA, then M1=M2M_{1}=M_{2}.

In light of the previous exercise, if AA\subseteq\mathbb{R} has a maximum, then we can refer to it as the (definite article) maximum of AA.

Example 1.24.

We claim that [0,1][0,1] has a maximum and, moreover, max[0,1]=1\max\,[0,1]=1. Indeed, this follows directly from the definition, since 1[0,1]1\in[0,1] and a1a\leq 1 for all a[0,1]a\in[0,1]. Recall from Example 1.12 that sup[0,1]=1\sup\,[0,1]=1, so in this case we have sup[0,1]=max[0,1]=1\sup\,[0,1]=\max\,[0,1]=1.

Exercise 1.25.

For each of the following subsets of \mathbb{R}, show that both the maximum and the supremum exists and compute their values.

  1. (i)

    {1,2,3}\{1,2,3\};

  2. (ii)

    {2/n:n}\{2/n:n\in\mathbb{N}\}.

The above observations suggest that suprema and maxima are closely related. However, the following important example shows that the two concepts are different.

Example 1.26.

We claim that the set (0,1)(0,1) does not have a maximum. This follows by the same argument we used to compute the supremum in Example 1.19. Indeed, for any M(0,1)M\in(0,1), the midpoint t:=M+12t:=\frac{M+1}{2} satisfies M<t<1M<t<1. Moreover, t(0,1)t\in(0,1) and MtM\not\geq t and so MM cannot be a maximum for (0,1)(0,1).

On the other hand, recall from Example 1.19 that sup(0,1)=1\sup\,(0,1)=1. Thus, here the supremum exists but no maximum exists for the set.

Exercise 1.27.

Using the fact that sup(0,1)=1\sup\,(0,1)=1, give another proof that (0,1)(0,1) does not have a maximum as follows.

  1. (i)

    Suppose (0,1)(0,1) does have a maximum MM. Using sup(0,1)=1\sup\,(0,1)=1, show that M1M\geq 1.

  2. (ii)

    Use part (i) to obtain a contradiction.

Suprema: further examples

We round off this section with some more involved examples of least upper bounds / suprema, to allow us to further master the concept.

Example 1.28.

Consider the set A:={21/n:n}A:=\{2-1/n:n\in\mathbb{N}\}. We claim that supA=2\sup A=2.

  1. 1.

    Since 21/n22-1/n\leq 2 for all nn\in\mathbb{N}, it follows that 22 is an upper bound for AA;

  2. 2.

    Suppose xx\in\mathbb{R} with x<2x<2. Our aim is to show that xx cannot be an upper bound for AA, so we need to find an element of AA that is larger than xx.

    Let r:=2xr:=2-x. Since \mathbb{N} is not bounded above (see Example 1.5), there exists some NN\in\mathbb{N} such that N>1/rN>1/r and therefore 1/N<r1/N<r. We have 21/NA2-1/N\in A and 21/N>2r=x2-1/N>2-r=x, so that xx is not an upper bound for AA.

    This shows that if xx\in\mathbb{R} is an upper bound for AA, then x2x\geq 2.

Thus, by definition, supA=2\sup A=2.

Exercise 1.29.

Does the set A:={21/n:n}A:=\{2-1/n:n\in\mathbb{N}\} from Example 1.28 have a maximum? Either prove a maximum exists and compute its value, or prove that no maximum exists.

Exercise 1.30.

For each of the following subsets of \mathbb{R}, show that the supremum exists and compute its value. Determine whether a maximum exists.

  1. (i)

    {3n21n2:n}\displaystyle\Big{\{}\frac{3n^{2}-1}{n^{2}}:n\in\mathbb{N}\Big{\}};

  2. (ii)

    (0,1)\mathbb{Q}\cap(0,1).