Yes, this is yet another "foundational" question in valuation theory.

Here's the background: it is a well known classical fact that the dimension (in the purely algebraic sense) of a real Banach space cannot be countably infinite. The proof is a simple application of the Baire Category Theorem: see e.g.


Suppose now that $(K,| \ |)$ is a complete non-Archimedean (edit: nontrivial) normed field. One has the notion of a $K$-Banach space, and the Baire Category Theorem argument works verbatim to show that such a thing cannot have countably infinite $K$-dimension.

Now let $\overline{K}$ be an algebraic closure of $K$. Then $\overline{K}$, by virtue of being a direct limit of finite-dimensional normed spaces over the complete field $K$, has a canonical topology, and indeed a unique multiplicative norm which extends $|\ |$ on $K$.

My question is: does there exist a complete normed field $(K, | \ |)$ such that:
(i) $[\overline{K}:K] = \infty$ and
(ii) $\overline{K}$ is complete with respect to its norm?

As with a previous question, it is not too hard to see that this does not happen in the most familiar cases. Indeed, by the above considerations this can only happen if $[\overline{K}:K]$ is uncountable. But $[\overline{K}:K]$ will be countable if $K$ has a countable dense subfield $F$ [to be absolutely safe, let me also require that $F$ is perfect]. Indeed, the algebraic closure of any infinite field has the same cardinality of the field, so $\overline{F}$ can be obtained by adjoining roots of a countable collection of separable polynomials $P_i(t) \in F[t]$. It follows from Krasner's Lemma that by adjoining to $K$ the roots of these polynomials one gets $\overline{K}$.

What about the general case?

  • 1
    $\begingroup$ yes: let K be the rationals and |.| the trivial norm (|0|=0, |x|=1 otherwise). $\endgroup$ Feb 11 '10 at 7:46
  • 4
    $\begingroup$ More helpfully: if the norm is (non-trivial and) discrete then probably an easy argument (analogous to one of the proofs that Q_p-bar isn't complete) gives that it can't happen. For the norm on K-bar won't be discrete (take roots of a uniformiser pi of K) so you can build some awkward x in the completion of K-bar with v(x)=1, v(x-pi)=3.5, v((x-pi)^2-pi^7)=10!+1/3 and so on and so on, meaning that the valuation on K(x) induced by K contains numbers with arbitrarily large denominator, so K(x) can't be discrete, so x can't be algebraic over K. $\endgroup$ Feb 11 '10 at 7:48

No, there exists no such field (with a non-trivial norm). A proof can be found in Bosch, Güntzer, Remmert: Non-Archimedean Analysis, Lemma 1, Section 3.4.3.

  • 1
    $\begingroup$ So in fact the proof is the same idea as the one in my comment above. Replace "has valuation n+1/d" by "is close to an element that generates an extension of K of dgree d" and then use Krasner's Lemma. $\endgroup$ Feb 11 '10 at 9:16

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.