4
$\begingroup$

If $f$ is any monic polynomial/$\mathbb{Z}$ with non-zero constant coefficient. I wish to study the quantities

$$t_n=\sum_{i}\theta_i^n\in\mathbb{Z}$$

where $(\theta_i)_{i=1}^{d}$ are the roots of $f$ counted with multiplicity.

The main question I am interested is finding all the primes $p$ such that $p\mid t_n$ for all large enough $n$. I have proved that this can only happen in the case that either $p\mid c_n$ $\forall$ $n$ where $c_n$ are the non-leading coefficients of $f$, or $p\mid t_n$ $\forall$ $n\geq0$.

The condition that $p\mid c_n$ is easy to check, so my interest has turned to some algebraic interpretation of the condition that $p\mid t_n$ for all $n$. My current idea is to let $K=\mathbb{Q}(\theta_i)_{i=1}^{d}$ be the field attached to $f$, and analyse the ideal $I$ of $\mathcal{O}_K$ generated by $(\theta_i)_{i=1}^{d}$. My idea is that it seems as if $p\mid t_n$ $\forall$ $n$ implies that any element of $I \cap \mathbb{Z}$ will be a multiple of $p$. The converse is obvious.

Is the above observation correct? If so, is there some known property of such primes (i.e. $I\cap \mathbb{Z} \subset p\mathbb{Z}$ iff $p$ ramifies in $\mathcal{O}_K$)?

$\endgroup$
9
  • 3
    $\begingroup$ The observation you propose, that $p \mid t_{n}$ for all $n$ implies that every element of $I \cap \mathbb{Z}$ is a multiple of $p$, isn't true. For example, if $\theta_{1} = 1$, $\theta_{2} = 4$ and $\theta_{3} = 7$, then $3 \mid t_{n}$ for all $n$, but $\langle \theta_{1}, \theta_{2}, \theta_{3} \rangle = \langle 1 \rangle \subseteq \mathbb{Z}$ contains many elements that are not multiples of $3$. $\endgroup$ Commented Oct 29, 2021 at 20:52
  • $\begingroup$ @JeremyRouse Thank you so much for the example! $\endgroup$
    – Milo Moses
    Commented Oct 29, 2021 at 21:02
  • $\begingroup$ It seems that you are counting the roots without multiplicity. Is it obvious why this should be the right approach? \\ Also, TeX note: please use $p \mid t$ p \mid t, not $p | t$ p | t. I have edited accordingly. $\endgroup$
    – LSpice
    Commented Oct 29, 2021 at 21:41
  • $\begingroup$ @LSpice Sorry I meant that the roots should be counted with multiplicity; I totally see how that wasn't clear. I will use $p\mid t$ in the future, thanks for the note! $\endgroup$
    – Milo Moses
    Commented Oct 29, 2021 at 21:58
  • 1
    $\begingroup$ It seems you are assuming implicitly that, say, $f$ has integer coefficients (otherwise, why would $t_n$ be in $\mathbb{Z}$?) At least for $p>\deg f$ the condition $p \mid t_n$ for all $n$ is the same as $p \mid c_n$ for $0 \le n < \deg f$. Indeed, the coefficients of $f$ are polynomials (with rational coefficients having denominator dividing $\deg f!$) in $t_1,\ldots,t_{\deg f}$; see Newton-Girard identities for explicit formulas for these. If $p>\deg f$ the denominators shouldn't worry us. $\endgroup$ Commented Oct 29, 2021 at 22:55

1 Answer 1

7
$\begingroup$

You have $t_k=0$ for all $k$ if and only if $f(x) \bmod p$ is a $p$-th power.

Let $g(x)$ be the image of $f(x)$ in $\mathbb{F}_p[x]$; let $\alpha_1$, $\alpha_2$, ..., $\alpha_n$ be the roots of $g$ (with multiplicity) in $\overline{\mathbb{F}_p}$, let $e_k$ be the $k$-th elementary symmetric function in the roots, and let $p_k = \sum \alpha_i^k$. You want a criterion for when all the $p_k$ are $0$.

If each $\alpha_i$ occurs with multiplicity divisible by $p$ then, clearly, $p_k=0$.

Conversely, suppose all the $p_k$ are $0$. Then, by Newton's identities, $k e_k = 0$ for all $k$. So, whenever $p$ does not divide $k$, we have $e_k=0$. But this means that the coefficient of $x^{(\deg g(x))-k}$ in $g(x)$ vanishes whenever $p$ does not divide $x$, so $g(x)$ is of the form $h(x^p) x^m$, where $m$ is equivalent mod $p$ to the degree of $g$. Since you imposed that $p_0=0$ as well, the degree of $g(x)$ is $0 \bmod p$.

$\endgroup$
3
  • 1
    $\begingroup$ Thank you so much! This is exactly the kind of criterion I was looking for, splendidly simple proof. $\endgroup$
    – Milo Moses
    Commented Oct 31, 2021 at 4:22
  • 1
    $\begingroup$ It's not the greatest idea to be using characteristic $p$ and also writing power sums as $p_k$. Maybe $P_k$? $\endgroup$
    – KConrad
    Commented Oct 31, 2021 at 4:26
  • $\begingroup$ Reading closer, you seem to be using without proof that $t_k \mod{p}=p_k$. This is trivial if we assume that the reduction-mod-$p$ map extends to a full ring homomorphism from the ring of algebraic integers to to $\overline{\mathbb{F}_p}$, but I don't see how it is so clear without that assumption. I know that this extension be done (see here, for example: math.stackexchange.com/questions/897660/…), but I think that you should make it more clear what is happening inside of your answer. $\endgroup$
    – Milo Moses
    Commented Nov 1, 2021 at 22:05

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .