Statistics for p-ranks of Artin-Schreier covers

The first part of the statement follows from the proof of [sankar, Proposition 2.3]. The remaining assertions are left as an exercise (see [pries2012p, Remark 3.9]).

Let g be such that \mathcal{C}=\mathcal{C}_g. Let \delta be such that for f(x)=g( x)+\delta^p-\delta condition (?) is satisfied. In greater detail, all monomials involving x^j where p|j in f_\alpha(x) can be removed by adding (\delta^p-\delta). It is easy to see that all other conditions are satisfied.

Let \sigma\in \operatorname{Gal}(\bar{\mathbb{F}}_p/\mathbb{F}_q) be such that \alpha'=\sigma(\alpha). Note that since f(x)\in \mathbb{F}_q(x) we have that (\sigma f)(x)=f(x). Here, (\sigma f)(x) is the rational function obtained by applying \sigma to the coefficients of f(x). This implies that

On the other hand,

Setting y=x_{\alpha'}, we find that f_{\alpha'}(y)=\sigma(f_\alpha)(y), and hence, d_{\alpha'}=d_\alpha.

The result is well known, see [fulman2016generating, Theorem 2.2].

Note that the degree of g(x) is (d+1-r). For i=1,\dots, t, we set

We observe that \lambda_1=\sum_{i=1}^t \beta_i and \lambda_2=\sum_{i=1}^t \xi_i. By Lemma 3.3, the number of squarefree monic polynomials over \mathbb{F}_q of degree \kappa_1 is \left(1-\frac{1}{q}\right)^{\xi_1} q^{\kappa_1}, and this is the number of choices for g_1(x). For a given choice of g_1(x), the number of choices for g_2(x) that are coprime to g_1(x) is \leq \left(1-\frac{1}{q}\right)^{\xi_2} q^{\kappa_2}. Assume that \kappa_2>0. Since there are at most (d+1-r) factors that divide g_1(x), the total number of choices for g_2(x) is

If \kappa_2=0, we find that the number of choices of g_2(x) is q^{\kappa_2}=1. Combining the above statements, we find that the total number of choices for g_2(x) is \geq \left(1-\frac{d+2-r}{q}\right)^{\beta_2}q^{\kappa_2}. Suppose we have made choices for g_1(x),\dots, g_j(x), and would like to choose g_{j+1}(x) to be coprime to g_1(x),\dots, g_j(x). The same argument tells us that the number of choices for g_{j+1}(x) lies between \left(1-\frac{d+2-r}{q}\right)^{\beta_{j+1}}q^{\kappa_{j+1}} and \left(1-\frac{1}{q}\right)^{\xi_{j+1}} q^{\kappa_{j+1}}. Therefore, we find that the number of choices for g(x) lies in between

and \left(1-\frac{1}{q}\right)^{\lambda_2} q^{r+1}. Note that \operatorname{deg}g(x)=d+1-r and h(x) has degree \leq \operatorname{deg} g(x). We find that for a given choice of g(x), the number of choices for h(x) lies in between

and q^{d+2-r}. Therefore, we obtain the bounds

Recall that we have expressed f(x)=h(x)/g(x), where g(x)=\prod_j g_j(x)^j and \kappa_j=\operatorname{deg} g_j(x), where g_j(x) is a squarefree polynomial with coefficients in \mathbb{F}_q. Assume without loss of generality that g_j(x) is a monic polynomial for all j. Note that the degree of g(x) is

Since f_{\infty}(x) is a constant, we find that \operatorname{deg} h(x)\leq \operatorname{deg} g(x). Furthermore, note that by construction, the polynomials g_i(x) and g_j(x) are coprime for i\neq j.

First, we consider the case when \kappa_j=0 for all j\geq p. In this case, for every pole \alpha of f(x), we have that d_{\alpha}<p. Thus, the condition (?), which requires that the coefficient of x^j be zero in f_\alpha(x) whenever p|j, is automatically satisfied. Furthermore, h(x) is coprime to g(x). Therefore, the result in this case follows from Lemma 3.4. In greater detail, the lower bound proceeds from the following observation

Next, assume that for some j>p, we have that \kappa_j\neq 0. In this case, the same calculation goes through with the one difference being that there is an additional constraint requiring that the coefficient of x^{j} in f_\alpha(x) is 0 for all j>0 such that p|j. This condition was automatically satisfied in the previous case. In this case, this gives an additional condition on h(x), which forces the number of choices to be

Let us explain this in greater detail.

Since \kappa_j\neq 0 for some j>p, there exists a pole \alpha of f(x) such that d_{\alpha}=j>p. The coefficient of x^p in f_\alpha(x) is required to be 0. Set \eta(x)\in \mathbb{F}_q(x) to denote the function given by

where \alpha' ranges over the \mathbb{F}_q-conjugates of \alpha. Note that \eta(x)^p=\sum_{\alpha'} \frac{1}{(x-\alpha')^p}. This function \eta(x)^p has coefficients in \mathbb{F}_q and is expressed as \eta(x)^p=\frac{\eta_1(x)}{\eta_2(x)}, with

Moreover, note that \eta_2(x) divides \prod_{\alpha'} (x-\alpha')^p, and hence divides \prod_{\alpha'} (x-\alpha')^j. Since this latter product divides g(x), it follows that \eta_2(x) divides g(x). For any choice of 0\neq c\in \mathbb{F}_q, the function f_c(x):=f(x)+c \eta(x)^p does not satisfy condition (?) since now the coefficient of x^p in f_{c,\alpha}(x) is nonzero, and p<j=d_\alpha. We may express f_c(x)=h_c(x)/g(x), where

The number of functions f(x)=h(x)/g(x) satisfying the conditions of Lemma 3.4 is \leq \left(1-\frac{1}{q}\right)^{\lambda_2} q^{d+3}. On the other hand, from an admissible function f(x)=\frac{h(x)}{g(x)}, we have constructed q new functions f_c(x), as c ranges over \mathbb{F}_q. Only one of which is admissible (i.e., when c=0). It is easy to see that if f(x) and F(x) are distinct admissible functions, then f_c(x)\neq F_{c'}(x) for all c,c'. Hence, the total number of admissible functions for the partition vector \kappa is

The proof of the result is rather straightforward and not particularly interesting, hence it is omitted.

Since it is assumed that d is even for p=2, it follows from Lemma 3.8 that M_p(d+2) is not equal to 0. Note that

where we recall that g=d\left(\frac{p-1}{2}\right) and \tau= r(p-1). It follows from Lemma 3.6 that \# \mathcal{S}_{g,\tau}(\mathbb{F}_q)\sim M_p(r+1, d+2)q^{d+3}. Likewise, we have that \# \mathcal{S}_{g}(\mathbb{F}_q)\sim M_p( d+2)q^{d+3}, and it thus follows that

On the other hand, if f and g are both admissible, then, by Proposition 2.3,

Hence, each isomorphism class of covers (unramified at \infty) consists of exactly (p-1) admissible functions. As a result,

and consequently, \mathfrak{d}(p,g, \tau)=\mathfrak{d}'(p, g, \tau).

Note that when (r+1)p<d+2, it is not possible to partition d+2 into r+1 numbers, all of which are \leq p. In this case, M_p(r+1, d+2)=0, and the result follows from Theorem 3.10.

The result follows from the first part of Lemma 3.6.

The condition requiring \kappa_j=0 for all j\geq p is satisfied when p is large, and the result thus follows from Lemma 3.14.

The proof is similar to that of Theorem 3.10. Assume that p is large enough so that \kappa_j=0 for all j\geq p. Recall that by abuse of notation, g:=g(d,p), \tau=\tau(r,p). Since d and r are fixed, g and \tau increase linearly with p (but this is suppressed in our notation). We have that

It follows from Lemma 3.14 and Corollary 3.15 that as p\rightarrow \infty,

Putting everything together, we obtain that

That \mathfrak{p}(d,r)=\mathfrak{p}'(d,r) follows from the equalities:

see the proof of Theorem 3.10.