Let $K$ be a local field (therefore complete, discrete non-archimedian valuation field) with perfect residual field $\kappa_K:= \mathcal{O}_K/\pi_K$.

Assume that $L/K$ is a field extension of $K$ of degree $n= [L:K]$ and $L$ contains all $n$-th roots and is tamely ramified. How to show that then $L/K$ is a cyclic extension (Therefore that $Gal(L/K)$ is a cyclic group)?

It looks (taking into account the condition that it contains the $n$-th roots) that possibly I could use the Kummer theory to get the claim but I don’t see how to apply it here. Especially I’m not really familar with local fields. Could anybody help?

Btw: I know that the main result of Kummer theory provides a bijective correspondence between abelian extensions of $K$ and subgroups $W$ of multiplicative group: $(K^*)^n \subset W \subset K^*$. Can group this correspondence be derived a correspondence to cyclic extensions?- as in my case


I keep your notation $L/K$ of degree $n$, and write as usual $n=ef$, where $e$ (resp. $f$) is the ramification (resp. inertia) index. The maximal unramified subextension $F/K$ (= inertia subfield) of $L/K$ is classically known to be cyclic of degree $f$.

1) Let us first study $L/F$, which, by hypothesis, is tamely ramified (the residual characteristic $p$ of $K$ does not divide $e$) and contains the $n$-th (hence the $e$-th) roots of unity. Denote by $\pi$ (resp. $\Pi$) a uniformizer of $F$ (resp. $L$), and by $U_i , i\ge 0$, the subgroup of units $U_i=1+ (\pi^i)$ of $F$. It is known that $U_0/U_1$ is canonically isomorphic to the residual multiplicative group ${\kappa}^*$ of $F$, and for $i \ge 1, U_i/ U_{i+1}$ is non canonically isomorphic to $(\kappa,+)$, see e.g. Cassels-Fröhlich, chap. 1. By definition $\Pi^e /\pi =u\in U_0$, and $u=u_1w$, with $u_1 \in U_1$ and $w$ representing a class of $U_0/U_1$. But, because $p \nmid e$, multiplication by $e$ in $U_1/U_2$ is an isomorphism, hence we can repeat the approximation process to get $u_1=u_2{w_1}^e$, with $u_2 \in U_2$ and $w_1$ representing a class of $U_1/U_2$, etc. On taking the projective limit, we obtain (with a change of notations) $\Pi^e=\pi$. This shows that $L=F(\sqrt [e]\pi)$ is an Eisenstein extension, thus tamely totally ramified with ramification index $e$. Since $F$ contains a primitive $e$-th root of unity, Kummer theory tells us that $L/F$ is a cyclic Galois extension of degree $e$.

2) We can't say much more about $L/K$ since we don't even know if it is Galois. However, even in the Galois situation, I think that your proposition (the cyclicity of $L/K$) does'nt hold in general. For example, fix an uniformizer $\pi$ of $K$ and take $L=E.F$, with $E=K(\sqrt [e]\pi)$. For ramification reasons, $E, F$ are linearly disjoint, hence $L/K$ is an abelian extension, with Galois group $G \cong Z/eZ \times Z/fZ$. If $e, f$ are coprime, $G$ is cyclic. But if for instance $e=f=$ a prime $q$, $G$ is not cyclic.

  • $\begingroup$ Thank you for your answer. One question: Why is the maximal unramified subextension $F/K$ of L/K cyclic? $\endgroup$ – KarlPeter Dec 19 '18 at 0:09
  • $\begingroup$ This is a classical result : an unramified extension is cyclic, obtained by adding a primitive root of 1 of order not divisible by the residual characteristic. See e.g. Cassels-Fröhlich, chap. 1. $\endgroup$ – nguyen quang do Dec 19 '18 at 7:33

If $\kappa_K$ is contained in $\overline{\mathbb{F}_p}$, or if $L/K$ is totally ramified (see comments below)

The valuation gives an absolute value $|x| = q^{-v(x)} $.

  • If $x \in L, |x| < 1$ then $(1+x)^{1/n} = \sum_{m=0}^\infty {1/n \choose k} x^k $ converges and is $\in L$.

    Proof : (In characteristic $p \nmid n$ then ${1/n \choose k} \in \mathbb{Z}_p$ then reduce it modulo $p$)

    So $|{1/n \choose k}| \le 1$ and the series converges, and since $L$ is complete the limit is in $L$.

  • Since $L/K$ is totally ramified of degree $n$ then $\pi_L^n = u^{-1} \pi_K$ where $|u| = 1$. So there is a root of unity $\zeta \equiv u \bmod (\pi_L)$ such that $u \zeta^{-1} = 1+x, |x| < 1$, so that $\varpi_L = \zeta^{1/n} (1+x)^{1/n} \pi_L \in L$ satisfies $ \varpi_L^n = \pi_K$.

Whence $L = K(\pi_K^{1/n})$ and $Gal(L/K) = \langle \sigma \rangle$ with $\sigma(\pi_K^{1/n}) = \zeta_n \pi_K^{1/n}$.

  • $\begingroup$ Hi. How do you deduce that $L:K$ is totally ramified from the given condition that it is tamely ramified? $\endgroup$ – KarlPeter Dec 16 '18 at 21:23
  • $\begingroup$ I’m using the definitions for totally/tamely ramification from math.uga.edu/~pete/8410Chapter4.pdf It says especially that if we consider a finite field extension $L/K$. Denote by $k := \kappa_K$ (resp l:= \kappa_L) is the residual fields of $K$ (resp L) then: $L/K$ totally ramified iff e(L/K) =[L:K] (or as you used $\pi_K \mathcal{O}_L = \pi_L ^{e(L/K)} \mathcal{O}_L$. This is equivalently to $l=k$ and $L/K$ is tamely ramified iff $e(L/K)$ is prime to the characteristic of $K$ $char(K)=p$. $\endgroup$ – KarlPeter Dec 17 '18 at 1:53
  • $\begingroup$ We know that $k$ is perfect. I don't see how do you conclude that $L/K$ totally ramified. What can we say about the resudual field $l$ of $L$. We need an argument that $l=k$... $\endgroup$ – KarlPeter Dec 17 '18 at 1:53
  • $\begingroup$ @KarlPeter Let $\kappa_K = \mathbb{Q}(\zeta_{n^\infty})$ then it has non-cyclic Galois extensions $\kappa_L/\kappa_K$ of degree $n$ so $K = \kappa_K((t)),L = \kappa_L((t))$ is a non-cyclic unramified extension of degree $n$. That's why your statement meant that $L/K$ is totally ramified (and doesn't have wild ramification, since all the $\zeta_{n^r}$ are then in $\kappa_K$). $\endgroup$ – reuns Dec 17 '18 at 18:35
  • $\begingroup$ Two points aren't clear to me: Firstly, why we can assume wlog that the residual field $\kappa_K$ is already the perfect closure $\mathbb{Q}(\zeta_{n^\infty})$ Naively - since by assumption it is perfect - it is just a subfield of $\mathbb{Q}(\zeta_{n\infty})$ in case of characteristic zero. Or did you implicitely used an argument that allows to reduce this to the case $\kappa_K = \mathbb{Q}(\zeta_{n^\infty})$? $\endgroup$ – KarlPeter Dec 17 '18 at 22:54

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.