5
$\begingroup$

I asked this question at MSE but I did not receive an answer. So I ask it at MO:

We denote the field of rational numbers by $\mathbb{Q}$. The Galois group of a polynomial $f$ is denoted by $Gal(f)$. The commutator subgroup of a group $G$ is denoted by $G'$.

Is there a $\mathbb{Q}$-linear map $T$ over $\mathbb{Q}[x]$ such that for all polynomials $f\in \mathbb{Q}[x]$ we have $$Gal(T(f))=(Gal(f))'$$

For a related post see the following question:

Endofunctors on the category of groups which are Galois- related to a linear map on $\mathbb{Q}[x]$

$\endgroup$
2
  • 9
    $\begingroup$ Can you say something about the motivation for this question? Why on earth should would one even suspect that such a thing might exist? $\endgroup$ Sep 23, 2015 at 21:22
  • 5
    $\begingroup$ The problem mixes multiplicative and additive structures of $\mathbb{Q}[x]$, which likely makes this very hard. For example, if such a $T$ existed, there would exist two polynomials $q$ and $p$, both of which split over $\mathbb{Q}$, with the property that $p-aq$ splits over $\mathbb{Q}$ if and only if $a$ is a perfect cube ($a\in\mathbb{Q}$), and if $a$ is not a perfect cube, then $p-aq$ is a product of a polynomial that splits and the power of an irreducible cubic polynomial with cyclic Galois group of order $3$ ($p=T(x^3)$, $q=T(1)$); it seems hard to find $p,q$, or show no such exist. $\endgroup$ Sep 23, 2015 at 22:16

2 Answers 2

17
$\begingroup$

The answer is no. First we see that $T$ is injective: Let $0\ne g\in\mathbb Q[X]$ with $T(g)=0$. Then $\textrm{Gal}(f-\alpha g)'=\textrm{Gal}(f)'$ for all $f\in\mathbb Q[X]$ and $\alpha\in\mathbb Q$. Pick $\beta\in\mathbb Q$ with $g(\beta)\ne0$, and $f\in\mathbb Q[X]$ of degree $n$ larger than $\deg g$ and $2$ such that $\textrm{Gal}(f)=S_n$. Then $\textrm{Gal}(f)'=S_n'=A_n$. But the polynomial $f(X)-\frac{f(\beta)}{g(\beta)}g(X)$ has the root $\beta$, so its Galois group is a subgroup of $S_{n-1}$. But no subgroup of $A_{n-1}=S_{n-1}'$ is isomorphic to $A_n$.

Thus we find $f\in\mathbb Q[X]$ of degree $2$ such that $F=T(f)$ has degree at least $2$. Since $\textrm{Gal}(F)=\textrm{Gal}(f)'=1$, $F$ splits into linear factors. Next pick $\gamma\in\mathbb Q$ such that $T(X+\gamma)=T(X)+\gamma T(1)$ has no common root with $F$. (Recall that $T(1)\ne0$.) Then $F$ and $G=T(X+\gamma)$ are relatively prime. As $f+\alpha(X+\gamma)$ has degree at most $2$ for all rational $\alpha$, we see that $F(X)+\alpha G(X)$ splits into linear factors for all rational $\alpha$. However, $F(X)+YG(X)$ is irreducible in $\mathbb Q[Y][X]$, so by Hilbert's irreducibility theorem there are infinitely many $\alpha\in\mathbb Q$ such that $F(X)+\alpha G(X)$ is irreducible. On the other hand, we saw that this polynomial factors into linear factors, so $\deg F\le 1$, contrary to $\deg F\ge2$.

$\endgroup$
0
4
$\begingroup$

I haven't worked out the details, but I think the following shows that the answer must be no. If $T$ were such a map, it should decrease the degree of polynomials. Otherwise, it should be easy to come up with a polynomial $p$ whose image has bigger Galois group than $p$ does. Secondly, every quadratic polynomial has Abelian Galois group, so that its image must factorize over $\mathbb Q$. How hard could it be to show that $T(X^2)\in\text{lin}(1,X)$? Similarly for $T(1)$ and $T(X)$. Now $T$ has a kernel since it maps a 3-dimensional space into a 2-dimensional space, say $q=aX^2+bX+c$. Now for any polynomial $p$ that factors over $\mathbb Q$, so must $p+\lambda q$ for arbitrary $\lambda\in\mathbb Q$.

This cannot be!

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.