15
$\begingroup$

Is it true that in the category of connected smooth manifolds equipped with a compatible field structure (all six operations are smooth) there are only two objects (up to isomorphism) - $\mathbb{R}$ and $\mathbb{C}$?

$\endgroup$
6
  • 2
    $\begingroup$ What about finite fields (considered as discrete 0-dimensional manifolds). Am I missing something? $\endgroup$ Oct 28, 2021 at 20:13
  • 4
    $\begingroup$ @SamGunningham Yes, they are not connected. $\endgroup$ Oct 28, 2021 at 20:14
  • 2
    $\begingroup$ Sorry, missed that asssumption! $\endgroup$ Oct 28, 2021 at 20:31
  • $\begingroup$ Well, they can be equipped with a topology that makes them connected... But in that case they are not manifolds modelled on $\mathbb{R}^n$, is that what you have in mind ? (I must confess I don't know where the theory of smooth manifolds breaks over $\mathbb{Q}$) $\endgroup$ Oct 28, 2021 at 20:39
  • 1
    $\begingroup$ @LoïcTeyssier Yep, a non-discrete finite topological space is not a real manifold. Yes, the question is about that. $\endgroup$ Oct 28, 2021 at 20:49

2 Answers 2

22
$\begingroup$

Here is a series of standard arguments.

Let $(\mathbb{F},+,\star)$ be such a field. Then $(\mathbb{F},+)$ is a finite-dimensional (path-)connected abelian Lie group, hence $(\mathbb{F},+) \cong \mathbb{R}^n \times (\mathbb{S}^1)^m$ as Lie groups. Since $\mathbb{F}$ is path-connected, there is in particular a path $\gamma: [0,1] \to \mathbb{F}$ with $\gamma(0) = 0_{\mathbb{F}}$ and $\gamma(1) = 1_{\mathbb{F}}$. Now consider the homotopy $H: \mathbb{F} \times [0,1] \to \mathbb{F}$, $(x,t) \mapsto \gamma(t) \star x$. This gives a contraction of $\mathbb{F}$ and so we can exclude all the circle factors.

Now, fix $y_0 \in \mathbb{F}$ and consider the map $\widehat{y_0}: \mathbb{R}^n \to \mathbb{R}^n$, $x \mapsto x \star y_0$. Then $\widehat{y_0}$ is an additive map (but at the moment not necessarily linear with respect to the natural vector space structure on $\mathbb{R}^n$). It is not too difficult to see that by additivity we have $\forall q\in \mathbb{Q}: \widehat{y_0}(qx) = q \widehat{y_0}(x)$. Since $\widehat{y_0}$ is continuous (as being smooth), it now follows that it's actually $\mathbb{R}$-linear.

Thus $\mathbb{F}$ is an $\mathbb{R}$-algebra. From this point on one can finish either by the Frobenius theorem on the classification of finite-dimensional associative $\mathbb{R}$-algebras or invoke a theorem of Bott and Milnor from algebraic topology that $\mathbb{R}^n$ can be equipped with a bilinear form $\beta$ turning $(\mathbb{R}^n,\beta)$ into a division $\mathbb{R}$-algebra (not necessarily associative) only in the cases $n=1,2,4,8$.

EDIT: Another finishing topological argument is a theorem of Hopf saying that $\mathbb{R}$ and $\mathbb{C}$ are the only finite-dimensional commutative division $\mathbb{R}$-algebras. This is less of an overkill compared to invoking Frobenius or Bott–Milnor as the proof is a rather short and cute application of homology, see p.173, Thm. 2B.5 in Hatcher's "Algebraic Topology".

$\endgroup$
8
  • 1
    $\begingroup$ It's beautiful! Thanks! $\endgroup$ Oct 29, 2021 at 16:42
  • 3
    $\begingroup$ You don't really need connected, just positive dimensional with smooth multiplication. If you pick some $x\ne 0$ in the path component of $0$, and divide by $x$, you take the $x$ path component to the $1$ path component, but you fix $0$. So $0$ and $1$ are on the same path component. Then your homotopy above shows there is only one path component. $\endgroup$
    – Ben McKay
    Nov 1, 2021 at 18:00
  • 1
    $\begingroup$ @BenMcKay: nice observation that one can weaken the initial hypothesis on $\mathbb{F}$! $\endgroup$
    – M.G.
    Nov 1, 2021 at 18:27
  • 2
    $\begingroup$ One can have even less "overkill", because the question was specifically about fields and the only finite field extensions of $\mathbb{R}$ are $\mathbb{R}$ itself and $\mathbb{C}$ which is an easy consequence of the fact that all real polynomials of odd degree have a real root. $\endgroup$ Nov 1, 2021 at 19:34
  • 2
    $\begingroup$ Is the intermediate value theorem not topological enough? :-) $\endgroup$ Nov 1, 2021 at 19:44
5
$\begingroup$

Finite-dimensional manifolds are locally compact, and the only non-discrete locally compact topological fields are the reals and the complex numbers. So the answer is yes.

$\endgroup$
8
  • 1
    $\begingroup$ Hmm, it seems p-adic numbers are also a locally compact field en.wikipedia.org/wiki/Locally_compact_field $\endgroup$ Oct 28, 2021 at 23:07
  • 2
    $\begingroup$ Here is the complete classification jstage.jst.go.jp/article/jjm1924/19/2/19_2_189/_pdf $\endgroup$ Oct 28, 2021 at 23:11
  • 4
    $\begingroup$ Of course, the $p$-adic numbers are not connected (in fact totally disconnected, though not discrete, so that the statement exactly as written is indeed false). $\endgroup$
    – LSpice
    Oct 28, 2021 at 23:26
  • 1
    $\begingroup$ If formal power series (I guess meaning Laurent polynomials?) over a finite field are given their topology coming from the valuation, then they, too, are totally disconnected. $\endgroup$
    – LSpice
    Oct 28, 2021 at 23:27
  • 3
    $\begingroup$ The formal power series over a field (a finite field or infinite field) form a commutative ring. If you pass to its fraction field you get the Laurent series over that field. It is strange (in English) to refer to that as a division ring when you don't have any actual noncommutative division rings under discussion too: it's better to call $\mathbf R$, $\mathbf Q_p$, $\mathbf F_p((t))$, and their finite extension fields rather than division rings. $\endgroup$
    – KConrad
    Oct 29, 2021 at 0:35

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.