6
$\begingroup$

Bhargava 2021 proves van der Waerden's conjecture about Galois groups of random integer polynomials: over all $x^n + a_{n-1} x^{n-1} + \cdots + a_0 = 0$ with $a_k \in \{-H, \ldots, H\}$, the number of polynomials $E_n(H)$ with Galois group not equal to the full symmetric group $S_n$ is $O(H^{n-1})$.

  • First (and this is a very basic question), am I correct that the $O$ here is in terms of $H$ only, so that in detail we have $E_n(H) < c_n H^{n-1}$ for some constants $c_n$ that depend on $n$?

  • Second, is there a conjectured growth rate for $c_n$?

From Bhargava 2021 it seems like in the worst case $c_n$ might have a factor related to the number of nonisomorphic groups of order $n$ (possibly with other factors), but presumably most of the count concentrates on fewer of the groups.

$\endgroup$

1 Answer 1

10
$\begingroup$

There are two different questions you could ask here.

One is the dependence on $n$ in the conjectured asymptotic best constant, i.e. the $c_n$ such that we have $E_n (H) < c_n H^{n-1} + o(H^{n-1})$ with the little $o$ depending on $n$. This one has a precise answer, because we do not need to worry about all the groups that are proven or expected to occur with frequency $o(H^{n-1})$. So we just need to deal with the reducible polynomials, where it was computed by Chela that we can take

$$ c_n = 2^n (\zeta(n-1) - 1/2) + 2 k_n $$ where $k_n$ is the volume of the region in $\mathbb R^{n-1}$ consisting of tuples $x_1,\dots, x_{n-1}$ with $$-1 \leq x_1,\dots, x_{n-1}, \sum_{i=1}^{n-1} x_i \leq 1 .$$

It is straightforward to check that this has growth rate proportional to $2^n$ in $n$.

The other is the dependence on $n$ of the conjectured best constant valid for all $H$. This is a somewhat annoying question because it will depend heavily on small values of $H$. For example, we must take $c_n \geq 3^{n-1}$ to deal with the case where $H$ is $1$ (and the constant coefficient vanishes). It's possible that we can take $c_n \leq 3^n$ - I don't see a reason against it.

$\endgroup$
8
  • 1
    $\begingroup$ Thank you, that makes sense. So very roughly we might have something like $E_n(H) = O(2^n H^{n-1}) + O(1)^n o(H^{n-1})$ where all constants are absolute. Also, to clarify: do you mean "occur with frequency $o(H^{n-1})$"? $\endgroup$ Apr 22, 2022 at 21:52
  • 2
    $\begingroup$ @GeoffreyIrving Yes, that sounds exactly right, and that's what I meant to say. $\endgroup$
    – Will Sawin
    Apr 22, 2022 at 22:56
  • $\begingroup$ Sorry, do you mean $k_n$ as written? Isn't it just an (n-1)-simplex of volume $2^{n-1}/(n-1)!$? $\endgroup$
    – liuyao
    Apr 24, 2022 at 2:37
  • $\begingroup$ @liuyao I think it's not quite - if you take the standard $n-1$-simplex, double all coordinates, and then subtract $1$ so they range from $-1$ to $1$, then the sum of the coordinates would range from $-(n-1)$ to $-(n-3)$, not from $-1$ to $1$. $\endgroup$
    – Will Sawin
    Apr 24, 2022 at 11:08
  • $\begingroup$ @WillSawin I see, it is an (n-1)-simplex still, and has fairly simple expression for the volume. The base edge isn't 2, but n-1. $\endgroup$
    – liuyao
    Apr 25, 2022 at 4:01

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.