Let $z \in \overline{\bf Q}$ be an algebraic number. Define the "denominator" of z to be the least natural number $n$ such that $nz$ is an algebraic integer.

By a rather ad hoc argument (playing with the minimal polynomial of $z$ to compute high powers of $z$ in terms of low powers of $z$, and measuring the coefficients obtained p-adically), I can show the following:

If $z$ is an algebraic number that is not an algebraic integer, then the denominator of $z^m$ grows exponentially in $m$ in the limit $m \to \infty$ (thus there is $c>1$ such that the denominator is at least $c^m$ for all sufficiently large $m$).

This is obvious in the case when $z$ is rational, from the fundamental theorem of arithmetic, but I couldn't find a similarly quick proof in the general case; presumably, unique factorisation into prime ideals for a suitable number field is the key, but (somewhat to my embarrassment) my algebraic number theory is too rusty to figure out how to exploit this here. So I am posing the question here to see if anyone can find an easy proof of this fact.

  • $\begingroup$ The point should be that integrality is a local condition ($z$ is an algebraic integer iff $\nu(z) \ge 0$ for every valuation $\nu$ on $\mathbb{Q}(z)$) and I feel like there should be a clean proof of this but I don't know it. $\endgroup$ – Qiaochu Yuan Nov 22 '13 at 1:20
  • 5
    $\begingroup$ @QiaochuYuan: Take the ideal $I$ of algebraic integers $a\ in \mathbb Q(z)$ such that $az$ is an integer. $I$ is a proper nontrivial ideal, so it has some proper nontrivial prime factor. This defines a valuation. The valuation of $z$ by that valuation cannot be nonnegative, or else $z$ could be written as $b/a$ for $b$ an algebraic integer and $a$ a unit mod the prime ideal, which is a contradiction because $a$ is not in $I$. $\endgroup$ – Will Sawin Nov 22 '13 at 3:02

There is some number field $K$ containing $z$, and there we have a prime factorization $$ (z)=\frac{\mathfrak{p}_1\ldots\mathfrak{p}_a}{\mathfrak{q}_1\ldots\mathfrak{q_b}} $$ of fractional ideals, where no $\mathfrak{p}_i$ is equal to a $\mathfrak{q}_j$. If $nz^m$ is an algebraic integer, then $(\mathfrak{q}_1\ldots\mathfrak{q_b})^m$ divides $(n)$ (as ideals of $\mathcal{O}_K$). This means $$ N(\mathfrak{q}_1\ldots\mathfrak{q_b})^m|N(n)=n^{[K:\mathbb{Q}]}, $$ so that $$ n\geq \left(N(\mathfrak{q}_1\ldots\mathfrak{q_b})^{1/[K:\mathbb{Q}]}\right)^m. $$ We can take $$ c=N(\mathfrak{q}_1\ldots\mathfrak{q_b})^{1/[K:\mathbb{Q}]} $$

| cite | improve this answer | |
  • 2
    $\begingroup$ Ah, that was indeed simple - somehow I had not figured out that one should be taking norms :-) $\endgroup$ – Terry Tao Nov 22 '13 at 1:55

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.