6
$\begingroup$

Let $k$ be a global function field of positive characteristic $p$ (e.g. $k = \mathbb{F}_p[t]$). Let $x \in k$ be non-zero and assume that $x$ is not a $p$th power.

For each place $v$ of $k$, we can consider the $v$-adic expansion of $x$ in the completion $k_v$ of $k$ at $v$. For all but finitely many $v$ this will look like $$x = a_{v,0} + a_{v,1} \pi_v + a_{v,2} \pi_v^2 + \dots$$ where $\pi_v$ denotes a choice of uniformiser at $v$ and each $a_{v,i}$ lives the residue field of $v$.

One of my PhD students has shown that for all but finitely many $v$ we have $a_{v,1} \neq 0$ (note that this property is independent of the choice of uniformiser). I found this a bit surprising, and naturally requires that $x$ is not a $p$th power. Even the case where $k = \mathbb{F}_p[t]$ it is not obvious to me how to prove this in an elementary way. His proof is very indirect and passes through the Brauer group of $k$, which seems overkill for such an elementary looking statement.

Is there any elementary proof of the fact that $a_{v,1} \neq 0$ for all but finitely many $v$?

More generally, I would be interested to know if this result is already known, which it must surely be. Moreover if there is any wider setting in which such a result holds and naturally lives, since it looks like there should be some deep general theory underlying it.

$\endgroup$
8
  • 1
    $\begingroup$ This looks like this might be equivalent to the statement that the field extension K[t]/K[x] is only ramified at finitely many places. $\endgroup$ Commented Jul 14 at 16:29
  • 1
    $\begingroup$ The statement follows immediately for example from Lemma 6.4 in my paper arxiv.org/abs/2401.11930 with Philip Dittmann, where we also give an elementary proof. The wider setting in which we work there is trying to understand the first-order sentences that hold in almost all completions of the global function field. $\endgroup$ Commented Jul 16 at 6:23
  • 2
    $\begingroup$ In a sense, what we show there (Theorem 6.5) is that among those sentences that are boolean combinations of existential sentences and true in almost all completions , the statement your student found is essentially the only maybe surprising one (assuming some consequence of resolution of singularities in positive characteristic). $\endgroup$ Commented Jul 16 at 6:59
  • 3
    $\begingroup$ I hope you’ll pardon me my irremediaby old-fashioned language: it seems to me you’re talking about the exact differential $dx$ and its divisor $(dx)$. This has only finitely many zeros, of course. $\endgroup$ Commented Jul 16 at 12:42
  • 1
    $\begingroup$ @Lubin: I'd be very happy with another answer to the question if you have one $\endgroup$ Commented Jul 17 at 8:39

2 Answers 2

9
$\begingroup$

$\DeclareMathOperator\Der{Der}$Let $f\in \mathbb{F}_p[x]$. The assumption that $f$ is not a $p$-th power implies that the derivative $f'= df/dx$ is not identically zero. For all but finitely many places $v$, $f'$ will have valuation zero at $v$ as you write.

If $v$ corresponds to an irreducible polynomial $g\in \mathbb{F}_p[x]$ then $f'$ has valuation zero at $v$ if and only if $f'(\lambda)\neq 0$, where $\lambda$ is any zero of $g$ in $\overline{\mathbb{F}}_p$.

I think this implies the assertion, for if $\lambda$ is in the algebraic closure of $\mathbb{F}_p$ we have an equality in $\overline{\mathbb{F}}_p[[x-\lambda]]$:

$$f= f(\lambda) + f'(\lambda)(x-\lambda) + O((x-\lambda)^2).$$

To deduce from this that $a_{v,1}\neq 0$ with $\pi_v= g$ one has to use that $\mathbb{F}_p$ is perfect, so that every zero of $g$ occurs with multiplicity $1$ and hence $(dg/dx)(\lambda)\neq 0$, so that $a_{v,1}= f'(\lambda)/g'(\lambda)\neq 0$.

Added: I think the above proof works in general, provided we can find a substitute for the derivative $d/dx$. I think this can be done as follows.

Let $X$ be a smooth curve of finite type over a perfect field $k$ of characteristic $p$. The sheaf of Kähler differentials $\Omega_{X/k}$ is locally free of rank $1$. Let $U$ be an affine open of $X$, where $\Omega$ is free of rank $1$. Working on $U$ results in disregarding finitely many closed points on $X$ (=finitely many places of the function field).

Let $A$ be the ring of global sections of $U$. Then by construction $\Der_{k}(A, A)=\mathbb{Hom}_A(\Omega_X(U), A)$ is a free $A$-module of rank $1$. Let us choose a generator $d$ (=the analog of $d/dx$ in the previous proof).

If $\mathfrak p$ is a maximal ideal of $A$ then the image of $d$ generates $\Der_{k}(A_{\mathfrak p}, A_{\mathfrak p})$ as an $A_{\mathfrak p}$-module, and hence $d(\pi_v)\neq 0$, where $\pi_v$ is a uniformizer of $A_{\mathfrak p}$. In fact, we get that $d(\pi_v)$ is a unit in $A_{\mathfrak p}$.

From $f=\sum_{i\ge 0} a_{v,i} \pi_v^i$ we get

$$df =(\sum_{i\ge 1} a_{v, i} i \pi_v^{i-1}) d\pi_v.$$

Hence, $a_{v,1} = 0$ if and only if $df$ lies in the maximal ideal of $A_{\mathfrak p}$.

If $df \neq 0$ then by shrinking $U$ (which amounts to disregarding finitely many places) we may assume that $df$ has valuation $0$ at all $v\in U$. This allows us to conlude that if $df\neq 0$ then $a_{v,1}\neq 0$ for all $v\in U$.

It remains to make the condition $df=0$ explicit. It is equivalent to $f$ being a $p$-th power in every localization $A_{\mathfrak p}$. Is this equivalent to $f$ being a $p$-th power in $A$? I am not sure.

$\endgroup$
1
  • 5
    $\begingroup$ Regarding your final question, the answer is "yes" -- there is always a map $\mathscr{O}_{X^{(p)}}\to \ker(d)$, and one may check \'etale locally that this map is an isomorphism for $X$ smooth, whence it reduces to an easy computation for affine space. This is part of "Cartier theory." $\endgroup$ Commented Jul 14 at 21:45
3
$\begingroup$

Expanding on my comment. (Edited to add: though, having seen Lubin's comment now, I agree that philosophically the right way to think about this is in terms of the differential form $df$ vanishing at only finitely many points on the algebraic curve.)

Consider the subfield $k' = \mathbb{F}(x)$ of $k$ (note that $x$ is a transcendental element of $k'$ because otherwise it would be a $p$th power).

I claim that the extension $k/k'$ is ramified at $v$ if and only if $a_{v,1} = 0$. There will be finitely many such places exactly when $k/k'$ is separable, which is equivalent to your condition that $x$ is not a $p$th power in $k$. (A proof of this fact can be found in Proposition 3.10.2 of Stichtenoth, Henning, Algebraic function fields and codes, Graduate Texts in Mathematics 254. Berlin: Springer (ISBN 978-3-540-76877-7/hbk). xiii, 355 p. (2009). ZBL1155.14022 where $x$ is called a "separating element" for $K$. Thanks to Arno Fehm's comment and his linked paper for helping me track down this reference!)

Without loss of generality we can assume that our base field $\mathbb{F}$ contains all coefficients $a_{v, i}$, since adjoining scalars to a field doesn't affect the ramification.

Now let $v$ be a place of $k$, and let $v'$ be its restriction to $k'$.

Rearranging your equation as

$$x - a_{v,0} = a_{v,1} \pi_v + a_{v,2} \pi_v^2 + \dotsb,$$

we see that $v(x - a_{v,0}) > 0$, and $v(x - a_{v,0}) > 1$ if and only if $a_{v,1} = 0$. The first statement implies that $v'$ is the place of $k' = \mathbb{F}_p(x)$ corresponding to the ideal $(x-a_{v,0}) \subset \mathbb{F}_p[x]$: in particular that means that $x - a_{v,0}$ is a uniformizer for the place $v'$.

We conclude that $v/v'$ is ramified if and only if $v(x - a_{v,0}) > 1$ if and only if $a_{v,1} = 0$.

$\endgroup$
2
  • 1
    $\begingroup$ In general, $a_{v,0}$ need not lie in $\mathbb{F}$, but in a finite extension, so that $(x-a_{v,0})$ is not an ideal of $\mathbb{F}[x]$. Maybe, you could first assume $\mathbb F$ to be algebraically closed and deduce the general case from this. $\endgroup$ Commented Jul 15 at 15:02
  • $\begingroup$ Good point! I'll correct this. $\endgroup$ Commented Jul 15 at 16:44

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.