I came across an curious exercise in Neukirch's algebraic number theory book. Exercise 2 page 176 (Chapter II section 9) asks the following

Prove that the maximal tamely ramified abelian extension V of $\mathbb{Q}_p$ is finite over the maximal unramified extension $\mathbb{Q}_p^{nr}$ of $\mathbb{Q}_p$.

I suspect that this is wrong because one can always consider the extensions

$$ L_n = \mathbb{Q}_p^{nr} (p^{\frac{1}{p^n - 1}}) $$

$L_n$ is tamely ramified and has degree $p^n - 1$ since $X^{p^n - 1} - p$ is irreducible in $\mathbb{Q}_p$ (by Eisenstein criterion).

Am I missing something here?

Here is why I think $L_n$ is an abelian extension.

Call $K_n = Q_p^{nr}(p^{1/n})$ for $n$ prime to $p$. The extension $K_n/Q_p$ (I think) is abelian for all $n$ prime to $p$ and here is why:

It suffices to show that $K_n/Q_p^{nr}$ is abelian (since $Q_p^{nr}/Q_p$ is already abelian).

Any $Q_p^{nr}$-linear automorphism $\sigma$ of $K_n$ is determined by the image $\sigma(p^{1/n})$.

The Galois group elements of $Gal(K_n/Q_p^{nr})$ are then the elements $\sigma_i$ with

$$\sigma_i(p^{1/n}) = p^{1/n} \zeta_{n}^{i}$$

for $i = 0, 1, \dots, n-1$ and these commute since $\zeta_n$ ( a primitive $n$-th root of the unit) is in $Q_p^{nr}$.

$$ \sigma_i \sigma_j(p^{1/n}) = \zeta_{n}^{i} \zeta_{n}^{j} p^{1/n}$$

  • $\begingroup$ $Gal(Q^{nr}_p(p^{1/(p^\infty-1)}/Q^{nr}_p) $ is abelian not $Gal(Q^{nr}_p(p^{1/(p^\infty-1)}/Q_p) $ $\endgroup$
    – reuns
    Jun 19, 2020 at 21:47
  • $\begingroup$ Isn't $Gal(Q_p^{nr}/Q_p)$ abelian? and a tower of abelian extensions is abelian? $\endgroup$
    – yassine
    Jun 19, 2020 at 21:55
  • 1
    $\begingroup$ Of course no towers of abelian extensions aren't $Q(2^{1/4})/Q(2^{1/2})/Q$. And $Gal(Q(2^{1/4},i)/Q)$ is the dihedral group $D_8$ (same as group of affine transformations $x\to ax+b,Z/(4)\to Z/(4)$) $\endgroup$
    – reuns
    Jun 19, 2020 at 21:59

2 Answers 2


$V$ is a tamely ramified extension of $Q_p$ and it contains $Q_p^{nr}=\bigcup_{p\ \nmid\ m} Q_p(\zeta_m)$ so $$V=\bigcup_j Q_p^{nr}(p^{1/n_j})$$ for some $p\nmid n_j$ and $n_j | n_{j+1}$.

Since any subextension of an abelian extension is abelian

It suffices to find for which $n$ we have $Q_p^{nr}(p^{1/n})/Q_p$ is abelian and tame.

ie. when $Q_p(p^{1/n})/Q_p$ is abelian and tame.

This happens iff $n | p-1$.

Whence $V=Q_p^{nr}(p^{1/(p-1)})$

  • $\begingroup$ I don't see how $Q_p^{nr}(p^{1/n})$ can fail to be abelian when $n$ is prime to $p$. I tried to explain my argument in the edited version of the question. I should have made that clearer, my bad. $\endgroup$
    – yassine
    Jun 19, 2020 at 18:16
  • $\begingroup$ $Q_3(i,3^{1/4})/Q_3$ is Galois but not abelian. And $Q_3(3^{1/4})/Q_3$ is not Galois. $\endgroup$
    – reuns
    Jun 19, 2020 at 19:18

[If $K\subset L\subset M$ are Galois extensions and $M/L, L/K$ are abelian, $M/L$ need not be abelian. In fact, if one has an exact sequence of groups $$1\to H\to G\to K\to 1$$ where $H$ is normal in $G$ and $H, K$ abelian, $G$ can well be non-abelian. ($G = S_3$, for example)]

As for the problem, let $K$ be a field that is complete with respect to a non-trivial discrete valuation $v$ and suppose the residue field is $\mathbb{F}_q$. Let $x$ be a uniformizer in $K$, $d=q-1$, $\mathcal{O}$ the valuation ring in $K$, and $U, \mathfrak{m}$ the group of units and the maximal ideal in $\mathcal{O}$. Denote by $K_d$ the extension $K_{nr}(x^{\frac{1}{d}})$. Since $K$ contains all the $d$-th roots of unity (by Hensel's lemma), $K(x^{\frac{1}{d}})$ is Galois, and is therefore abelian. Thus $K_d$ is abelian.

Let $L/K$ be an abelian tamely ramified extension. The local reciprocity map $\omega: K^*\to \mathrm{Gal}(L/K)$ maps $U$ to the inertia subgroup $I$ and $1+\mathfrak{m}$ to $1$. Then, $\omega$ maps the residue field surjectively to $I$. So $I$ is finite of order dividing $d=q-1$. One now has $L\subset K_d$, and $K_d$ is the maximal abelian unramified extension of $K$.


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.