1.5 Greatest lower bounds

We conclude our discussion of the completeness axiom by introducing the notation of a greatest lower bound (or infimum), which is a close cousin to the notion of a least upper bound.

Definition 1.39.

Let AA\subseteq\mathbb{R}.

  1. 1

    We say xx\in\mathbb{R} is a lower bound if xax\leq a for all aAa\in A.

  2. 2

    We say AA is bounded below if there exists a lower bound xx\in\mathbb{R} for AA.

  3. 3

    We say ii\in\mathbb{R} is a greatest lower bound (or infimum) for AA if both

    1. (i)

      ii is a lower bound for AA, and

    2. (ii)

      if xx\in\mathbb{R} is a lower bound for AA, then xix\leq i.

    In this case, we write i=infAi=\inf A.

  4. 4

    We say AA is bounded if it is both bounded above and bounded below. Otherwise, AA is unbounded.

If an infimum for AA exists, then it must be unique (why?). Thus, we refer to it as the infimum of AA, which is denoted by infA\inf A.

Example 1.40.

Consider the set A:={n2n1:n}A:=\big{\{}\frac{n}{2n-1}:n\in\mathbb{N}\big{\}}. We claim that infA=1/2\inf A=1/2.

Note that

n2n1\displaystyle\frac{n}{2n-1} =12(2n1)+12n1\displaystyle=\frac{1}{2}\cdot\frac{(2n-1)+1}{2n-1}
=12(1+12n1)\displaystyle=\frac{1}{2}\Big{(}1+\frac{1}{2n-1}\Big{)}
(1.4) (1.4) 12for all n.\displaystyle\geq\frac{1}{2}\qquad\text{for all $n\in\mathbb{N}$.}

Hence 1/21/2 is a lower bound for AA.

We now show 1/21/2 satisfies a version of the approximation property for the infimum. For this, we note that

2(2n1)2(2nn)=2n>nfor all n.2(2n-1)\geq 2(2n-n)=2n>n\qquad\text{for all $n\in\mathbb{N}$.}

Let ε>0\varepsilon>0 be given. Let NN\in\mathbb{N} satisfy N>1/εN>1/\varepsilon so that 0<1/N<ε0<1/N<\varepsilon. If we define a:=N2N1a:=\frac{N}{2N-1}, then aAa\in A and

(1.5) (1.5) a=12+12(2N1)<12+1N<12+ε.a=\frac{1}{2}+\frac{1}{2(2N-1)}<\frac{1}{2}+\frac{1}{N}<\frac{1}{2}+\varepsilon.

Combining (1.40) and (1.5), we have 1/2a<1/2+ε1/2\leq a<1/2+\varepsilon. This tells us that 1/2+ε1/2+\varepsilon cannot be a lower bound for AA. As ε>0\varepsilon>0 was chosen arbitrarily, we conclude that infA=1/2\inf A=1/2.

Many properties of infima can be deduced directly from properties of suprema. In particular, the following lemma establishes a version of the completeness axiom in terms of infima.

Lemma 1.41.

Let AA\subseteq\mathbb{R} be nonempty and bounded below. Then infA\inf A exists and, moreover, infA=sup(A)\inf A=-\sup(-A) where A:={x:xA}-A:=\{-x:x\in A\}.

Proof.

Since AA is nonempty, there exists some aAa\in A. Hence, by definition aA-a\in-A so that A-A is nonempty. Since AA is bounded below, there exists some mm\in\mathbb{R} such that xmx\geq m for all xAx\in A. Thus, xm-x\leq-m for all xAx\in A so that ymy\leq-m for all yAy\in-A. Hence, m-m is an upper bound for A-A so that A-A is bounded above.

Since A-A is nonempty and bounded above, sup(A)\sup(-A) exists. We claim that infA\inf A exists and infA=sup(A)\inf A=-\sup(-A).

Since sup(A)\sup(-A) is, by definition, an upper bound for A-A, we have ysup(A)y\leq\sup(-A) for all yAy\in-A. Thus, xsup(A)-x\leq\sup(-A) for all xAx\in A and so xsup(A)x\geq-\sup(-A) for all xAx\in A. Hence sup(A)-\sup(-A) is a lower bound for AA.

It remains to show sup(A)-\sup(-A) is the greatest lower bound for AA. Indeed, suppose mm\in\mathbb{R} is any lower bound for AA. Then by our earlier arguments, we know m-m is an upper bound for A-A. Hence, by the definition of the supremum, sup(A)m\sup(-A)\leq-m. But this implies that msup(A)m\leq-\sup(-A). Hence sup(A)-\sup(-A) is indeed the greatest lower bound for AA and thus, by definition, infA=sup(A)\inf A=-\sup(-A). ∎

Definition 1.42.

Let AA\subseteq\mathbb{R}. We say mm\in\mathbb{R} is a minimum of AA if mAm\in A and xmx\geq m for all xAx\in A.

If a minimum for AA exists, then it must be unique (why?). Thus, we refer to it as the minimum of AA, which is denoted by minA\min A.

Exercise 1.43.

Let AA\subset\mathbb{R}.

  1. (i)

    Show that if a minimum minA\min A exists, then infA\inf A exists and infA=minA\inf A=\min A.

  2. (ii)

    Give an example of a nonempty, bounded set AA\subseteq\mathbb{R} for which no minimum exists.