Let $\bar{\mathbf{Q}}$ be an algebraic closure of the rationals, and $\alpha$ denote an algebraic number in $\bar{\mathbf{Q}}$. We define the height of $\alpha$, denoted by $H(\alpha)$, to be $$H(\alpha) = \left( \prod_v \max(1,\Vert \alpha\Vert_v) \right)^{1/[K:\mathbf{Q}]}.$$ Here $K$ is a number field containing $\alpha$ and the product runs over the set of normalized valuations $v$ of $K$. For a non-empty finite subset $B\subset \bar{\mathbf{Q}}$, we define $$H(B) := \max \{H(\alpha) \ | \ \alpha\in B\}.$$

Now, let $K$ be a number field and let $\alpha$ be an algebraic number contained in $K$. Let $B$ be the set of conjugates of $\alpha$.

Question. Can we bound $H(B)$ from above in terms of data depending only on $\alpha$ and $K$?

Example. The number of elements of $B$ is less or equal to $[K:\mathbf{Q}]$.

I'm looking for a bound of the form $H(B)\leq H(\alpha)^{[K:\mathbf{Q}]}$ if possible.


One always has $H(B)=H(\alpha)$. Indeed, since presumably you normalise the absolute values in such a way that the definition does not depend on the choice of $K$ for a given $\alpha$ (as long as $\alpha\in K$), let's replace $K$ by its Galois closure $K'$ over $\mathbb{Q}$. Let $\sigma$ be any element in the Galois group. Since $\sigma$ acts by permutation on the set of absolute values of $K'$, $H(\sigma(\alpha))$ is just the same product as $H(\alpha)$ with some of the terms permuted.

  • $\begingroup$ Nice. I didn't realize the elements of the Galois group permute the set of absolute values of the Galois closure of $K$. Thanks alot. $\endgroup$ – Dali Aug 1 '11 at 10:44

Alex has answered your question, but if you're working with heights and Galois conjugates of numbers, you're likely at some point to need a relation between the height of $\alpha$ and the heights of the elementary symmetric functions $s_1,\ldots,s_n$ of its conjugates $\alpha_1,\ldots,\alpha_n$. The advantage of working with $s_1,\ldots,s_n$, of course, is that they are in $\mathbb{Q}$. A useful relation (the constants can be improved) is $$\frac{1}{2}H(\alpha) \le H([1,s_1,\ldots,s_n])^{1/n} \le 2H(\alpha).\qquad(*)$$ This is proven in many standard texts, but for concreteness I can point to my book The Arithmetic of Elliptic Curves, Theorem VIII.5.9. (To get $(*)$ from that theorem, you also need to use the fact, as noted in Alex's answer or see [AEC, Theorem VIII.5.10], that Galois conjugates have the same height.)


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.