1

How could one substitute the (15th) axiom of completeness with Archimedean and Cantor's axiom?

We discussed Cantor's axiom as well as Archimedean in analysis lectures and were told this question might appear on the exam. The first 14 axioms which also work for $\Bbb Q$ are left, and I need to derive, as mentioned in comments, the axiom of completeness of the field $\Bbb R.$ I am looking for proof that the supremum of a non-empty $S\subset(0,+\infty)$ is unique, therefore I have to rewrite the statement that, no matter how small positive real number $\varepsilon$ I subtract from the supremum, there will be some $x\in S$ such that $$\sup S-\varepsilon<x.$$ But, since I need to use the Archimedean axiom, I tried to use it here, but I'm not sure how. Then I wanted to replace the term supremum when describing the properties of the non-empty intersection segments in Cantor's theorem using the statement: $\forall x,y\in \Bbb R^+, \exists n\in\Bbb N,$ such that $ny\ge x$

Am I at least at the right starting trail?

PinkyWay
  • 4,565
  • Presumably you have list of axioms for the real numbers and are asking something about the 15th one. There are many axiomatizations. Talking about "the $\epsilon$ part" something makes no sense without context. I think you have been asked to prove that your 15th axiom can be derived from some other things. If you [edit] the question to show us what you have tried and where you are stuck perhaps we can help. You must provide complete definitions and statements. – Ethan Bolker Nov 04 '19 at 18:04
  • 1
    Is it correct that the axiom of completeness in your lecture is the least upper bound property, "every nonempty subset $S$ of $\mathbb{R}$ that is bounded above has a least upper bound", and Cantor's axiom is the nested interval property, "If $I_1 \supseteq I_2 \supseteq \ldots$ is a nested sequence of nonempty closed intervals, then $\bigcap_{n = 1}^{\infty} I_n \neq \varnothing$"? – Daniel Fischer Nov 04 '19 at 21:21
  • Yes, that is exactly what we learnt. – PinkyWay Nov 04 '19 at 22:11
  • You can look here (and at linked/related questions). I'm not sure if you can piece together a correct proof from the attempts in the questions and the hints in the answers, however. But do give it a try. – Daniel Fischer Nov 04 '19 at 22:39
  • Thank you. I will. – PinkyWay Nov 04 '19 at 23:47
  • Did you manage to assemble a proof? – Daniel Fischer Nov 06 '19 at 19:47
  • Almost, I discussed with my colleague who reformulated the Archimedean axiom from the $nx>y$ into $\frac{x}{y}>\frac{1}{n}$ and I proposed to use that in the Cantor axiom, but now I am pretty sure I could replace $\epsilon$ by $\frac{1}{n}$. So there would be a fixed point $a$ like in the comment section on the page you recommended. That is a positive real number so I thought of using the absolute value for the, at first, symmetric (unempty) segments, and $x_n$ would be a term in a increasing sequence. – PinkyWay Nov 06 '19 at 20:03
  • Cantor's axiom guarantees there will be a point in the intersection after infinte steps of cutting those segments, I can manipulate with numbers from both sides and move the point on the line and cover the whole real number line? – PinkyWay Nov 06 '19 at 20:03
  • Sorry, I missed your reply until now. I wasn't notified. To notify a user in a comment thread they've participated in you need to address them, @userName, as in @Praskovya2.718281828, generally. The exceptions are that the post author [in this case you] is automatically notified of comments under their post, and if only one other user has commented, they are also automatically notified if the post author comments because it is then assumed that the author replies to the other user. Rule of thumb, if you're not totally sure a user will be notified of your comment, @-ping them. – Daniel Fischer Nov 08 '19 at 21:50
  • 2
    Now, I'm not sure what exactly you are referring to, nor what construction you have in mind. Let's look at this comment and the following. One constructs a sequence of nested intervals $I_n = [a_n, b_n]$ such that for all $n$, there is an $s \in S$ with $s \geqslant a_n$, and $b_n$ is an upper bound of $S$ ($a_n$ may happen to be an upper bound too, but generically it isn't), and such that the length of $I_{n+1}$ is half the length of $I_n$. By Cantor's … – Daniel Fischer Nov 08 '19 at 21:51
  • 2
    … axiom/theorem the intersection $\bigcap I_n$ isn't empty. By the Archimedian property and the halving of successive lengths it follows that the intersection contains exactly one point, call it $x$. Since the $b_n$ form a decreasing sequence converging to $x$, it follows that $x$ is an upper bound. Since the $a_n$ form an increasing sequence converging to $x$ it follows that an $y$ with $y < x$ is not an upper bound (for then there is an $n$ and an $s \in S$ with $y < a_n \leqslant s$), thus $x$ is the least upper bound of $S$. – Daniel Fischer Nov 08 '19 at 21:51
  • Thank you very much! I think this argumentation is what I was asked to do. – PinkyWay Nov 08 '19 at 22:18

1 Answers1

0

We can use the Cantor-intersection theorem and the Archimedean axiom to prove the Dedekind's theorem (source:Enrico Gregorio's answer).

Suppose $A,B\subseteq\Bbb R$ are non-empty s. t. $(\forall a\in A)(\forall b\in B)\quad a\le B$ and $A\cup B=\Bbb R$. Then $\exists ! c\in\Bbb R$ s. t. $a\le c\le b, \forall a\in A,\forall b\in B$.

Let's abbreviate the above with $A\le B$ and $A\le c\le B$.

Since $A\ne\emptyset\space\land\space B\ne\emptyset$, let's choose $a_0\in A$ and $b_0\in B$.

Now, we choose $c_0=\frac{a_0+b_0}2$.

If $A\le c_0\le B$, we're done. Otherwise, it is either $$c_0<a,\text{for some }a\in A\space\text {or}\space c_0>b\text{ for some }b\in B.$$

If $c_0<a$, then let $a_1=a, b_1=b_0$.

If $c_0>b$, then let $a_1=a_0, b_1=b$.

In either case, $a_0\le a_1\space\land\space b_0\ge b_1\implies [a_0,b_0]\supseteq[a_1,b_1]$.

Let $d=b_0-a_0$. We have $b_0-c_0=c_0-a_0=\frac{b_0-a_0}2=\frac{d}2$.

If $c_0<a$, then $\frac{d}2=b_0-c_0>b_0-a=b_1-a_1$.

If $c_0>b$, then $\frac{d}2=c_0-a_0>b-a_0=b_1-a_1$.

We can repeat the procedure to build a sequence of nested closed intervals:

$$[a_0,b_0]\supseteq[a_1,b_1]\supseteq\cdots\supseteq[a_n,b_n]$$

If at some point $A\le c_n\le B$, we're done. Otherwise, we get a sequence of closed nested intervals $[a_n,b_n]$ with $a_n\in A,b_n\in B$ and $b_n-a_n\le\frac{d}{2^n}.$

By the Cantor-intersection theorem (now assumed as an axiom), there is $c\in\bigcap_n[a_n,b_n]$.

Now we want to prove $A\le c\le B$. Let's assume the opposite: there is either

$$a\in A\text{ with } c<a\space\text{ or }\space b\in B\text{ with } c>b.$$

Suppose $a>c$.

Let's take $\varepsilon=\frac{a-c}2$. We can take $n\in\Bbb N$ s. t. $\frac{d}{2^n}<\varepsilon$. $(*)$

Then we obtain: $b_n<a+\varepsilon\le c+\varepsilon=c+\frac{a-c}2=\frac{a+c}2\le a$ which contradicts $A\le B$.

Analogously in case of $c>b$.


$(*)$ As $2^n>n, n\in\Bbb N$, we have $\frac1{2^n}<\frac1n\implies\frac{d}{2^n}<\frac{d}n$. In order to make $\frac{d}{2^n}<\varepsilon$, we could take $n$ s. t. $\frac{d}n<\varepsilon$, which is guaranteed to hold true by the Archimedean axiom.


Now, we want to prove the Dedekind's theorem is equivalent to the axiom of completeness (like Henno Brandsma did here):

Every non-empty subset of $\Bbb R$ that is bounded above has a supremum in $\Bbb R$.

Direction $\boxed{\Rightarrow}$

Suppose Dedekind's theorem holds.

Let $S\subset\Bbb R$ with some upper bound $u$ and let's define: $$A=\{x\in\Bbb R\mid\exists s\in S, x<s\}\\\text{and}\\ B=\{x\in\Bbb R\mid\forall s\in S, x\ge s\}$$

$B\ne\emptyset$ because $u\in B$.

Then our $c=\sup S$.

Direction $\boxed{\Leftarrow}$

Assume the axiom of completeness holds.

Then we take $c=\sup A$.

PinkyWay
  • 4,565