62
$\begingroup$

Is an automorphism of the field of real numbers $\mathbb{R}$ the identity map? If yes, how can we prove it?

Remark An automorphism of $\mathbb{R}$ may not be continuous.

$\endgroup$
5
  • 8
    $\begingroup$ Hint: it's fairly easy to see that any order-preserving automorphism is the identity. $\endgroup$ Jul 22, 2013 at 12:05
  • 2
    $\begingroup$ I was sure this is a duplicate given that I have a recollection of spelling this out here at least twice. But the best match I could find right away is this. I'm not sure we can call it a duplicate. Sure, the answers of that question also answer this, but... $\endgroup$ Jul 22, 2013 at 13:27
  • $\begingroup$ Also, Jacobson's Basic Algebra I :-) $\endgroup$ Jul 22, 2013 at 13:28
  • $\begingroup$ @JyrkiLahtonen It is an interesting fact that there are infinitely many automorphisms of $\mathbb{C}$, even though $[\mathbb{C}:\mathbb{R}]=2$. Why is this fact not a contradiction to this problem? $\endgroup$
    – user425181
    Sep 4, 2017 at 20:50
  • 4
    $\begingroup$ @user425181 It means that $\Bbb{C}$ has infinitely many subfields isomorphic to $\Bbb{R}$. Mind you, this result depends heavily on the axiom of choice. IIRC (ask a set theorist) if we drop out the axiom of choice, it may (or may not?) happen that $\Bbb{C}$ has only finitely many automorphisms. $\endgroup$ Sep 4, 2017 at 21:04

7 Answers 7

90
$\begingroup$

Here's a detailed proof based on the hint given by lhf.

Let $\phi$ be an automorphism of the field of real numbers. Let $x \gt 0$ be a positive real number. Then there exists $y$ such that $x = y^2$. Hence $\phi(x) = \phi(y)^2 \gt 0$.

If $a \lt b$, then $b - a \gt 0$. Hence $\phi(b) - \phi(a) = \phi(b - a) \gt 0$ by the above. Hence $\phi(a) \lt \phi(b)$. This means that $\phi$ is strictly increasing.

If $n$ is a natural number, it can be written in the form $1 + \ldots + 1$, so $\phi(n) = n$. Now, any rational number is of the form $r = (a - b)c^{-1}$, for $a, b, c$ natural numbers, so it follows that $\phi(r) = r$ for any rational number.

Let $x$ be a real number. Let $r, s$ be rational numbers such that $r \lt x \lt s$. Then $r \lt \phi(x) \lt s$. Since $s - r$ can be arbitrarily small, $\phi(x) = x$. This completes the proof.

$\endgroup$
2
  • 1
    $\begingroup$ @MakotoKato It is an interesting fact that there are infinitely many automorphisms of $\mathbb{C}$, even though $[\mathbb{C}:\mathbb{R}]=2$. Why is this fact not a contradiction to this problem? $\endgroup$
    – user425181
    Sep 4, 2017 at 20:49
  • $\begingroup$ @user425181 Because in one case you're looking at automorphisms of $\mathbb{C}$, and in the other you're only looking at those which restrict to the identity on $\mathbb{R}$. The restriction in the second case is strong enough to only leave two options (and this is one of the key realizations in Galois theory). $\endgroup$
    – Ducky
    Mar 8, 2021 at 2:52
63
$\begingroup$

Hint: Let $\phi$ be a field automorphism of $\mathbb R$. Then prove:

  • $\phi$ sends positive numbers to positive numbers

  • $\phi$ is increasing

  • $\phi$ is continuous

  • $\phi$ is the identity on $\mathbb Q$

  • $\phi$ is the identity on $\mathbb R$.

$\endgroup$
4
  • 2
    $\begingroup$ I just feel tempted to write how one's intuition might develop in such cases. $\phi$ is an automorphism $\implies$ it's a function. To know about this function, we must know in every possible way how this function will behave in the real domain. And that's why we want to investigate all properties of $\phi$. A real domain follows the laws of trichotomy, it has positive and real numbers. It's worthy to note how the function will react to such kinds of inputs. Again, a beautiful question :-) $\endgroup$
    – MathMan
    Jul 17, 2014 at 19:55
  • $\begingroup$ How do you prove the continuity? $\endgroup$
    – user422112
    May 5, 2018 at 12:50
  • 2
    $\begingroup$ @ThatIs, actually, as Makoto's answer shows, continuity is not needed. But my point was that an increasing function can only have jump discontinuities and then it cannot be surjective. See math.stackexchange.com/questions/1207135/…. $\endgroup$
    – lhf
    May 5, 2018 at 13:28
  • $\begingroup$ My intuition for this proof is that once we know $\phi$ is the identity on the rational numbers, we want to extend $\phi$ by continuity. One way to do that is to show $\phi$ is increasing. But an automorphism is something that only "knows" about algebraic properties of the field, involving the field operations. Why would it "know" about the order relation on $\mathbb R$? We have to find a way to encode the order relation in algebraic terms, which we do by noting an algebraic difference between positive and negative numbers: having square roots. $\endgroup$
    – Jack M
    Jul 22, 2018 at 11:41
46
$\begingroup$

I have never liked the proofs of this that use analysis more than necessary. Once you get that $\phi$ is order-preserving and the identity map on the rationals, take an arbitrary real number $a$. If $\phi(a) \neq a$, then there is a rational $q \in \mathbb{Q}$ between $a$ and $\phi(a)$. If $a \leq q \leq \phi(a)$, then $\phi(a) \leq \phi(q) = q$, so $\phi(a) \leq q$ and $q \leq \phi(a)$, so $\phi(a) = q = \phi(q)$, so $a = q$, contradicting the fact that $\phi(a) \neq a$. Similarly if $\phi(a) \leq q \leq a$.

$\endgroup$
16
$\begingroup$

For related but slightly stronger results, see $\S$ 16.7 of these field theory notes.

Highlights:

(i) Every Archimedean ordered field $K$ admits a unique homomorphism of ordered fields $K \hookrightarrow \mathbb{R}$.
(ii) Let $(F,<)$ be an ordered field in which every positive element is a square (e.g. any real-closed field, e.g. $\mathbb{R}$). Then the ordering $<$ is unique, so every homomorphism of fields between two such fields is necessarily a homomorphism of ordered fields. Thus:

The identity map on $\mathbb{R}$ is the unique field homomorphism from $\mathbb{R}$ to $\mathbb{R}$: "$\mathbb{R}$ is strongly rigid".

(In the Lemma that occurs just before the "Main Theorem on Archimedean Ordered Fields" -- currently numbered Lemma 192 and on p. 106, but both of these are subject to change -- where it says "topological rings", I think it should say "Hausdorff topological rings".)

$\endgroup$
13
$\begingroup$

Here is a quick formulation of the proof which bypasses explicitly showing the ordering is preserved. Let $f:\mathbb{R}\to\mathbb{R}$ be an automorphism. We know $f$ fixes all rational numbers. Suppose $f$ is not the identity; say $f(x)\neq x$ for some $x$. We may assume $f(x)>x$ (if $f(x)<x$, then $f(-x)>-x$, so we can replace $x$ by $-x$). Now let $q$ be a rational number such that $x<q<f(x)$ and let $y=\sqrt{q-x}$. Note that $$f(q)=f(x+y^2)=f(x)+f(y)^2\geq f(x).$$ But $f(q)=q$ since $q$ is rational, so this contradicts the fact that $q<f(x)$. Thus $f$ must be the identity.

$\endgroup$
2
  • 1
    $\begingroup$ How do we know, that f fixes all rational numbers ? $\endgroup$
    – martina
    Mar 13, 2018 at 8:17
  • 2
    $\begingroup$ @martina: $f$ fixes $1$ and then additive and multiplicative properties forces it to fix all rationals. $\endgroup$
    – Bumblebee
    May 20, 2018 at 0:08
1
$\begingroup$

I think the correct way to use density of rationals in this problem is this:

Let $a$ be a real number and suppose $\phi(a) \neq a$. Then, there exists a rational number $q$ such that $a<q<\phi(a)$. Hence, $q < \phi(a)$ and $q>\phi(a)$. This is a contradiction since $\phi$ is a bijection.

$\endgroup$
4
  • $\begingroup$ Of course you first must show that $\phi$ preserves the ordering. $\endgroup$
    – GEdgar
    Jun 15, 2016 at 14:57
  • 1
    $\begingroup$ Yes. This is done above. $\endgroup$ Jun 15, 2016 at 15:43
  • $\begingroup$ @GEdgar It is an interesting fact that there are infinitely many automorphisms of $\mathbb{C}$, even though $[\mathbb{C}:\mathbb{R}]=2$. Why is this fact not a contradiction to this problem? $\endgroup$
    – user425181
    Sep 4, 2017 at 20:51
  • 6
    $\begingroup$ @user425181 ... only one non-identity automorphism of $\mathbb C$ maps $\mathbb R$ into itself. The others don't. Thus, no contradiction. $\endgroup$
    – GEdgar
    Sep 4, 2017 at 22:24
0
$\begingroup$

Since $f$ preserves addition, $f(0+0)=f(0)+f(0)$, and so $f(0)=0$. Since $f$ preserves mutliplication, for all $x\in\mathbb R$ we have $f(x)=f(1\cdot x)=f(1)\cdot f(x)$. Therefore, either $f(1)=1$ or $f$ is identically $0$; the second possibility is ruled out by the fact that $f$ maps $\mathbb R$ onto $\mathbb R$. Let $n\in\mathbb N$. It can be easily proven by induction that if $a_1,\dots,a_n$ is any list of numbers, then $$ f\left(\sum_{i=1}^{n}a_i\right)=\sum_{i=1}^{n}f(a_i) \, , $$ and so in particular $$ f(n)=f\left(\sum_{i=1}^{n}1\right)=\sum_{i=1}^{n}f(1)=\sum_{i=1}^{n}1=n \, . $$ Since $f(n)+f(-n)=f(0)=0$, we have $f(-n)=-f(n)=-n$. This shows that $f$ is the identity map on $\mathbb Z$. Let $a,b\in\mathbb Z$, with $b\neq0$. We have $f(1/b)\cdot f(b)=f(1)$, so $f(1/b)=f(1)/f(b)=1/b$. Hence $f(a/b)=f(a)\cdot f(1/b)=a/b$. This shows that $f$ equals the identity on $\mathbb Q$.

For the final stage of the proof, we utilise three facts that rest upon $\mathbb R$ being a complete ordered field:

  • Every positive real has one positive square root.
  • The square of a nonzero real number is positive.
  • $\mathbb Q$ is dense in $\mathbb R$: between any two distinct real numbers there is a rational number in between them.

Let $x>0$. Then $f(x)=f(\sqrt x)f(\sqrt x)>0$. This means that if $x>y$, then as $x-y>0$, $$ f(x-y)=f(x)-f(y)>0\implies f(x)>f(y) \, . $$ Therefore, $f$ preserves order. Now, let $z$ be irrational. If it were the case that $f(z)<z$, then there would an $r\in\mathbb Q$ such that $f(z)<r<z$. But as $f(r)=r$, this contradicts the fact that $f$ is order-preserving. The case $f(z)>z$ similarly leads to a contradiction. Hence $f(z)=z$ and so $f$ is the identity on $\mathbb R\setminus\mathbb Q$, completing the proof.

$\endgroup$
2
  • $\begingroup$ I failed to see what this answer brings new. $\endgroup$
    – user26857
    Mar 15 at 16:49
  • $\begingroup$ @user26857: It doesn't present a different method to the other answers, but it does flesh things out a bit more, which might be helpful for readers who find it difficult to fill in the details of the other answers. $\endgroup$
    – Joe
    Mar 15 at 16:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.