my question can be seen as a extension to this question.

Let $\overline{\mathbb{Q}}$ denote an algebraic closure of $\mathbb{Q}$. Given a polynomial with algebraic coefficients $f \in \overline{\mathbb{Q}}[X_1,...,X_n]$, say $$f=\sum_{(\alpha_1,...,\alpha_n)}c_{(\alpha_1,...,\alpha_n)}X^{\alpha_1}\cdot...\cdot X^{\alpha_n}.$$ Let $(b_1,...,b_n)$ be a root of this polynomial, where $b_1,...b_n \in \mathbb{C}$ (so the $b_{i}$ are not necessarily algebraic numbers individually). By root i mean that we have $$f(b_1,...,b_n) = \sum_{(\alpha_1,...,\alpha_n)}c_{(\alpha_1,...,\alpha_n)}b^{\alpha_1}\cdot...\cdot b^{\alpha_n} = 0.$$ Is it possible to construct a polynomial with rational coefficients $g \in {\mathbb{Q}}[X_1,...,X_n]$, for which $(b_1,...,b_n)$ is a root?

I would already be happy with hints to relevant literature. Thank you!


Here is a try. Let $L/Q$ be a finite Galois extension that contains all the coefficients of $f$. Denote by $G$ the corresponding Galois group. Take $$ g={\rm Norm}_{L/Q}(f) = \prod_{\sigma \in G} \sigma(f), $$ where by $\sigma(f)$, I mean the polynomial one gets by applying $\sigma$ to the coefficients of $f$. Then the coefficients of $g$ lie in $Q$, since they are invariant to the Galois action and clearly $$ g(b_1,\ldots,b_n) = f(b_1,\ldots, b_n) \prod_{1\neq \sigma\in G} \sigma(f)(b_1,\ldots, b_n) = 0 $$ as needed.

Edit: More details. Write $f = \sum c_{\alpha_1, \ldots, \alpha_n} X^{\alpha_1}\cdots X^{\alpha_n}$. Then $\sigma(f) = \sum \sigma( c_{\alpha_1, \ldots, \alpha_n}) X^{\alpha_1}\cdots X^{\alpha_n}$. Now, write $$ g = \prod_{\sigma \in G} \sigma(f) = \sum d_{\alpha_1,\ldots, \alpha_n} X^{\alpha_1}\cdots X^{\alpha_n}. $$ Note that, on the one hand $\tau (g) = \prod_{\sigma\in G} \tau\sigma(f) = \prod_{\rho\in G} \rho(f) = g$ (I've substituted $\rho = \tau\sigma$ and noted that as $\sigma$ runs over the elements of $G$, so does $\rho$). On the other hand, $\tau(g) = \sum \tau(d_{\alpha_1,\ldots, \alpha_n}) X^{\alpha_1}\cdots X^{\alpha_n}$. Therefore, by comparing coefficients, we get that $\tau(d_{\alpha_1,\ldots, \alpha_n}) = d_{\alpha_1,\ldots, \alpha_n}$ for all $\tau\in G$ and $\alpha_1,\ldots, \alpha_n$, so $d_{\alpha_1,\ldots, \alpha_n}\in Q$.

  • $\begingroup$ Oops; I've just noticed that essentially the same solution was given by Lubin in the linked question. $\endgroup$ – Lior B-S Mar 26 '17 at 13:03
  • $\begingroup$ Thank you for your answer! You are right, this is essentially the same solution as by Lubin. When i read his answer, i got stuck at the same point as in your answer, which makes it even more helpful: can you provide a reference/proof for the fact that the coefficients of $Norm_{L/Q}(f)$ lie in $Q$? My literature defines $Norm_{L/Q}$ only for single elements of $L$, so you could write $Norm_{L/Q}(a)$ for some $a \in L$. Wikipedia uses the same definition (en.wikipedia.org/wiki/Field_norm). $\endgroup$ – cdwe Mar 26 '17 at 17:40
  • 1
    $\begingroup$ I added details... $\endgroup$ – Lior B-S Mar 27 '17 at 10:47
  • 1
    $\begingroup$ Thanks alot, i get it now! $\endgroup$ – cdwe Mar 27 '17 at 11:03

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.