Exponents of Jacobians and relative class groups

Technical Summary: Exponents of Jacobians and Relative Class Groups

Problem Statement
The paper addresses the structural properties of the class group $Cl(K)$ of a global function field $K = \mathbb{F}_q(X)$, which is isomorphic to the group of rational points on the Jacobian $Pic^0 X$ of a smooth, proper, integral curve $X$ over a finite field $\mathbb{F}_q$. While the size of the class group is well-understood via the Weil bounds and Frobenius eigenvalues, the group structure—specifically the exponent $e$ (the smallest integer such that $eD \sim 0$ for all $D \in Pic^0 X$)—remains a delicate question.

Previous work, notably by Madan, Madden, and Stichtenoth, established that the exponent grows with the genus $g$. Stichtenoth's bound, $e > g^{1/3} / (4 \log g)$, was the state-of-the-art for the absolute case ($X_2 = \mathbb{P}^1$). However, no bounds existed for the "relative" case: the exponent of the relative class group $Pic^0 X_1 / \phi^* Pic^0 X_2$ for a covering $\phi: X_1 \to X_2$ where $X_2$ is a curve of genus $g_2 > 0$.

Methodology
The authors employ a direct geometric approach, moving away from the field-theoretic language used in previous proofs. The core strategy involves:

1. Existence of Low-Degree Points: Utilizing the Weil bounds to demonstrate the existence of distinct Galois orbits of points (divisors) on the curve $X$ with degrees $k \approx \log_q g$.
2. Linear Equivalence Contradiction: Assuming the exponent $e$ is small, the authors construct a linear equivalence $e\tilde{P} \equiv e\tilde{Q}$ between distinct effective divisors of low degree.
3. Factorization via Castelnuovo–Severi: If such a linear equivalence exists, it implies the existence of a rational function $f: X \to \mathbb{P}^1$ of degree $e \cdot k$. If this degree is sufficiently low relative to the gonality of the curve, the Castelnuovo–Severi inequality forces $f$ and a fixed gonal map $\gamma$ to factor through a common intermediate curve $Z$.
4. Ramification Analysis: The authors analyze the ramification data of these maps. By carefully selecting points that are "non-fibral" (avoiding ramification loci and specific fibers), they show that the required factorization leads to a contradiction regarding the structure of the divisors or the genus of the intermediate curve $Z$.

In the relative setting, the argument is adapted to handle divisors $D$ on $X_1$ such that $eD$ is not the pullback of a divisor from $X_2$. This involves analyzing the equivalence $e(\phi^*\phi_* D) \equiv (e \deg \phi) D$ and constructing rational functions with controlled ramification patterns that cannot exist under the assumption of a small exponent.

Key Contributions and Results

• Improved Bound for Absolute Class Groups (Theorem 1.2):
  The authors improve upon Stichtenoth's bound by incorporating the gonality of the curve, $\text{gon } X$. For a curve $X$ of genus $g$ and gonality $\text{gon } X > 1$, the exponent $e$ satisfies:
  $$e \ge \max \left( \frac{\text{gon } X}{2 \lceil \log_q(2g+1) \rceil}, \frac{g}{4(\text{gon } X - 1) \lceil \log_q(7g+1) \rceil} \right)$$
  This yields an asymptotic improvement of $e \gtrsim g^{1/2} / \log g$, surpassing the previous $g^{1/3}$ bound.

• First Bounds for Relative Class Groups (Theorem 1.3 / Theorem 5.1):
  The paper establishes the first lower bounds for the exponent $e$ of the relative class group $Pic^0 X_1 / \phi^* Pic^0 X_2$ for a separable covering $\phi: X_1 \to X_2$ with genera $g_1, g_2$. The exponent satisfies:

  1. $e \ge \lfloor c \sqrt{g_1}/g_2 \rfloor$, where $c = \min(1/16, \frac{g_2}{4 \log_q(14g_1+1)})$.
  2. $e \ge \frac{g_1^{1/2}}{8 \lceil \log_q(19g_1) \rceil (\deg \phi - 1)}$.

• Quantitative Estimates:
  Theorem 4.3 provides not only a bound on the exponent but also a lower bound on the number of elements in $Pic^0 X$ of a given order $m$ (and their pairwise differences), which is crucial for the relative case proofs.

Significance and Limitations
The paper claims that the geometric approach offers increased flexibility, allowing for the derivation of results in the genuinely relative situation where no prior bounds existed. The results improve the asymptotic shape of the bounds for the absolute case and are entirely new for the relative case.

The authors are modest regarding the limitations of their method. They note that for covers where $g_1 \approx g_2^2$ and $\deg(\phi) \approx g_2$ (almost unramified covers), the derived bounds become trivial. They explicitly raise Question 5.3: whether the set of separable coverings with a fixed relative class group exponent $N$ is finite. While their bounds restrict the possible triples $(g_1, g_2, \deg \phi)$, they do not resolve this finiteness question, particularly for small $N$ (e.g., $N=2$).

The paper does not claim these results extend to general abelian varieties, noting that counterexamples exist (e.g., powers of elliptic curves) where the exponent does not grow with the dimension. The work is strictly confined to the geometry of curves over finite fields.