Abstract
For a smooth projective curve $C$ with genus $g \ge 2$ and a degree 1 line bundle $L$ on $C$, let $M := SU_{C}(r;L)$ be the moduli space of stable vector bundles of rank $r$ and with the fixed determinant $L$. In this paper, we study the small rational curves on $M$ and estimate the codimension of the locus of the small rational curves. In particular, we determine all small rational curves when $r = 3$.
Introduction
In algebraic geometry, the moduli space $M$ of stable vector bundles over a curve $C$ is a fundamental object of study. These spaces are known to be Fano varieties—geometric spaces characterized by "positive" curvature and a wealth of rational curves. A major open question in the field is whether every smooth Fano variety with a simple Picard group (Picard number 1) contains a "line," defined as a rational curve of the smallest possible degree.
While it is well-established that $M$ is covered by Hecke curves of degree $r$, curves of smaller degree—referred to here as "small rational curves"—are significantly rarer. These curves do not fill the entire space; instead, they are contained within smaller, closed subsets. This research addresses two primary objectives: estimating the codimension of the locus containing these small rational curves and providing a complete classification for the specific case of rank $r=3$.
Methodology
The study utilizes Sun’s degree formula, which provides a critical link between the degree of a rational curve in the moduli space and the properties of a vector bundle defined on the product surface $C \times \mathbb{P}^1$. To analyze these structures, the author employs several key algebraic techniques:
- Harder-Narasimhan Filtrations: These are used to decompose the vector bundles associated with rational curves into simpler components, facilitating a systematic study of their degrees.
- Hecke Transformations: A standard technique for modifying vector bundles at specific points, used here to construct families of stable bundles that parameterize the rational curves.
- Dimension Counting: By calculating the dimensions of various extension groups ($\text{Ext}^1$) and the dimensions of the moduli spaces of the component bundles, the author derives rigorous codimension estimates.
Key Insights
Filtration Correspondence
A central insight of this work is that every small rational curve corresponds to a vector bundle on $C \times \mathbb{P}^1$ that admits a specific type of filtration. This correspondence allows the geometric problem of finding curves to be translated into the algebraic problem of analyzing bundle structures.
The Role of Genus
The "rarity" of small rational curves is directly proportional to the genus $g$ of the underlying curve $C$. As the genus increases, the codimension of the locus of small rational curves grows. This implies that in higher-genus cases, these curves become increasingly "hidden" within the larger moduli space.
The Rank 3 Threshold
For the case of rank $r=3$, the degree of a small rational curve is restricted to 1 or 2. The author demonstrates that these curves are essentially images of lines or conic-like curves residing within specific projective bundles over the moduli space of lower-rank bundles.
Principal Results
Codimension Estimates
The author establishes a general lower bound for the codimension of the locus of small rational curves: $$\min \left\{ \sum_{i=2}^n r_{i-1}(r_i - r_{i-1})(g - 2) + r_{n-1}(r_n - r_{n-1} - 1) \right\}$$ For the specific case of rank $r=3$, this estimate simplifies to a codimension of at least $2g - 3$.
Classification for $r=3$
Theorem 3.7 provides a complete classification of small rational curves for rank 3, identifying four distinct origins:
- Images of degree 2 curves in the fibers of a specific projective bundle.
- Double covers of lines in those same fibers.
- Images of lines that map to lines in a lower-rank moduli space $UC(2, \mathcal{L}')$.
- Images of lines in the fibers of a different projective bundle constructed via Hecke transformations.
Conclusion and Future Work
While the codimension estimates provided are general, the explicit classification is currently limited to rank $r=3$. Extending this classification to $r > 3$ would require managing significantly more complex filtration types and higher-degree curves. Furthermore, the assumption of a fixed determinant of degree $d=1$ is a specific case; while geometrically significant for Fano varieties, other degrees may yield different results.
Ultimately, these findings serve as a foundation for studying the Chow group of the moduli space $M$, providing new tools to understand the algebraic cycles and homology of these complex geometric objects.
Summary updated