🔢 mathematics

Sections of Jacobian fibrations over lines

This paper investigates the cohomology classes of sections of the universal Jacobian over lines in a linear system on a smooth surface, with a specific focus on K3 surfaces where the findings highlight a discrepancy with certain steps in a related work by BKV25.

Original authors: János Kollár, Giulia Saccà

Published 2026-08-04

📖 7 min read🧠 Deep dive

CC BY 4.0

Original authors: János Kollár, Giulia Saccà

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are exploring a vast, multi-dimensional landscape made of pure mathematics. In this world, there are special shapes called "surfaces," and on these surfaces, you can draw families of curves, like a bundle of spaghetti where each noodle is a slightly different shape. Mathematicians call this a "linear system." Now, for every single curve in this bundle, there is a hidden, complex machine attached to it called a "Jacobian." Think of the Jacobian as a magical control panel that records every possible way you can wrap a string around that specific curve.

When you look at all these control panels together, they form a giant, twisting tower called a "Jacobian fibration." The big question in this corner of geometry is: Can you walk through this tower and pick a single, smooth path (a "section") that touches every floor exactly once? Sometimes, you can find a path that works for a small slice of the tower, like a straight line drawn through the bundle of spaghetti. But can you stretch that path out to cover the entire tower? This matters because these towers are often "hyperkähler manifolds," which are incredibly rare and symmetrical shapes that physicists and mathematicians love to study because they might hold secrets about the fundamental structure of space and time.

The paper you are about to read, written by János Kollár and Giulia Sacca, tackles a specific puzzle about these paths. They investigate whether a path that exists on a small slice of the tower can be extended to cover the whole thing. Their answer is a firm "no" in many cases due to topological obstructions. However, they note a potential tension with a recent, related study (by BKV25). While the main theorem of that paper might still stand, the authors state that they "do not see how to reconcile" their results with some of the intermediate steps in [BKV25]. The authors show that while you can find many paths on a small slice, the only one that can be extended to the whole tower without breaking is the "zero section"—essentially, the path that just sits at the very bottom, doing nothing.

The Story of the Impossible Stretch

Let's dive into the adventure. Imagine you have a giant, flexible sheet (our surface $S$ ) covered in a grid of lines. You pick a specific set of lines, called a linear system $|H|$ , and you start drawing curves on them. For every curve you draw, there is a corresponding "Jacobian" space. If you stack all these Jacobian spaces on top of their curves, you get a massive, multi-story building. The floor plan of this building is your set of lines, and the height of the building at any point is the Jacobian of the curve at that spot.

Now, imagine you draw a straight line $L$ across the floor plan. This line cuts through a specific slice of the building, creating a smaller tower $J_L$ . The authors ask: If you can find a path (a section) that walks smoothly through this small slice $J_L$ , can you stretch that path out to walk through the entire giant building?

The paper proves that, in many cases, you cannot. It's like trying to stretch a rubber band from a small circle to a giant sphere; the geometry of the sphere simply doesn't allow the band to stay smooth and connected everywhere. The authors show that the "Mordell-Weil group"—which is just a fancy name for the collection of all possible paths you can take on your small slice—is huge. It looks like a grid of $r-1$ dimensions, where $r$ is the number of base points (the spots where all your lines cross).

However, here is the twist: Even though there are thousands of paths on the small slice, almost none of them can be extended to the whole building. The paper demonstrates that if a path on the slice comes from a "global" path (one that exists for the whole building), it must be the zero section. Think of the zero section as the "ground floor" path. It's the only one that doesn't wiggle or twist in a way that would cause a collision when you try to stretch it out. Any other path you find on the slice is a "local trick" that breaks the moment you try to expand it to the whole system.

The Topological Obstacle Course

How do they know this? They use a clever trick involving "monodromy." Imagine you take your small slice (the line $L$ ) and spin it around the floor plan. As you spin it, the base points (the crossing spots) swap places, like dancers in a choreographed routine. The paths on your slice get shuffled around too.

The authors argue that if a path could be extended to the whole building, it would have to look the same no matter how you spin the slice. It would have to be "invariant" under this dancing. But they show that the only way a path stays the same while the dancers swap places is if the path is the zero section. Any other path gets scrambled by the dance. Therefore, the only path that survives the stretch to the whole building is the one that does nothing.

This leads to a surprising observation about a recent paper by BKV25. That paper suggested that you might be able to extend these paths by changing the "complex structure" of the building (essentially, reshaping the material of the tower). Kollár and Sacca point out a discrepancy: they "do not see how to reconcile" their findings with some of the intermediate claims in [BKV25], though they clarify that this does not necessarily contradict the main theorem of that paper. They argue that the obstruction isn't just about the shape of the material; it's about the fundamental topology—the way the space is knotted. Even if you change the material, the knot remains. So, the steps in the BKV25 paper that claim you can extend these paths might be inconsistent with these topological rules.

The Takeaway

In short, this paper is a "stop sign" for a specific idea in high-level geometry. It says: "You can find many interesting paths on a small slice of these Jacobian towers, but don't try to stretch them to cover the whole thing. The universe of these shapes has a hidden rule that says only the 'zero' path can go global."

The authors are very sure of this result; they provide a rigorous proof using cohomology (a way of counting holes and shapes in higher dimensions) and the Shioda-Tate isomorphism (a bridge between geometry and algebra). They don't just suggest it; they prove that for a wide class of surfaces, the answer is definitively negative.

They also note that while this works for these specific towers, it's still an open question whether this rule applies to all Lagrangian fibrations (a broader class of these geometric structures). But for the specific case of Jacobians over lines, the door is closed: you can't extend the non-zero paths. It's a beautiful example of how a local discovery (a path on a line) can hit a hard, global wall (the topology of the whole space), teaching us that what looks possible in a small neighborhood isn't always possible in the grand scheme of things.

Technical Summary: Sections of Jacobian Fibrations over Lines

Problem Statement
The paper investigates the existence and extension of sections in universal Jacobian fibrations, specifically focusing on the cohomology classes of sections over lines within a linear system. The primary motivation arises from a tension between the authors' findings and intermediate claims in a recent work, [BKV25], regarding Lagrangian fibrations of smooth hyperkähler varieties.

Specifically, [BKV25] studies whether a section $\sigma_L$ defined over a general line $L \subset \mathbb{P}^n$ in a Lagrangian fibration $p: X \to \mathbb{P}^n$ can be extended to a meromorphic section $\sigma: \mathbb{P}^n \dashrightarrow X$ , potentially after altering the complex structure of $X$ . The authors posit that the correct geometric setting for this problem is the universal Jacobian for linear systems of curves on a surface $S$ . They aim to demonstrate that topological obstructions often prevent such extensions, even when the complex structure is varied.

Methodology
The authors work over the field of complex numbers $\mathbb{C}$ . Their approach combines the geometry of linear systems on surfaces with the theory of Mordell-Weil groups and the Shioda-Tate isomorphism.

Geometric Setup: Let $S$ be a smooth projective surface with $\text{Pic}(S) = \mathbb{Z}[H]$ , where $|H|$ is a basepoint-free linear system. The authors consider the universal compactified Jacobian $p: J(S, H) \to |H|$ . They restrict this fibration to a general line $L \subset |H|$ , denoting the restricted fibration as $J_L \to L$ .
Section Construction: Using the blow-up of the base points of the pencil corresponding to $L$ , the authors construct sections of $J_L \to L$ derived from exceptional curves. These sections correspond to elements in the Mordell-Weil group $\text{MW}(J_L/L)$ .
Cohomological Analysis: The core methodological step involves analyzing the cohomology classes of these sections. The authors utilize the Shioda-Tate isomorphism, which relates the Mordell-Weil group to the orthogonal complement of the fiber class in the Néron-Severi group.
Monodromy Arguments: The authors employ monodromy actions on the cohomology of the total space. They argue that while specific sections constructed from exceptional curves have cohomology classes that vary under monodromy (as the base points permute), classes arising from the restriction of global cohomology $H^{2g}(J(S, H), \mathbb{Z})$ must be monodromy invariant.
Intersection Theory: To establish the injectivity of the map from sections to cohomology classes, the authors prove a special case of a conjecture regarding the injectivity of the map from the Mordell-Weil group to numerical equivalence classes. This relies on the Hodge index theorem and the Theorem of the Square for line bundles on abelian varieties.

Key Contributions and Results

Proposition 1: The authors establish two main results for a smooth projective surface $S$ $S$ with $\text{Pic}(S) = \mathbb{Z}[H]$ $Pic (S) = Z [H]$ :
1. The Mordell-Weil group of the restricted fibration $J_L \to L$ is isomorphic to $\mathbb{Z}^{r-1}$ , where $r = (H^2)$ .
2. If a section $\sigma_L: L \to J_L$ has a cohomology class that lies in the image of the restriction map from the global cohomology of the universal Jacobian, then $\sigma_L$ must be the zero section.
Corollary 2: This result addresses the extension problem directly. If a holomorphic section $\sigma_L$ over a line $L$ extends to a meromorphic section $\sigma$ over the entire linear system $|H|$ (possibly after changing the complex structure of the total space $X = J(S, H)$ ), then $\sigma_L$ must be the zero section.
Proposition 9: A variant of Proposition 1 is provided for surfaces where $h^1(S, \mathcal{O}_S) = 0$ , showing that the Mordell-Weil group of the universal Jacobian over the open subset of irreducible nodal curves is isomorphic to the orthogonal complement of $[H]$ in $\text{NS}(S)$ .
Conjecture 7: The paper proposes a general conjecture stating that for a smooth group scheme over a curve with irreducible fibers, the map sending a section to its numerical equivalence class is injective modulo torsion, provided the generic fiber has no Abelian subvarieties defined over the base field. The authors prove a special case of this for surfaces mapping to $\mathbb{P}^1$ .

Significance and Claims
The paper claims to identify topological obstructions that prevent the extension of sections in Lagrangian fibrations, challenging the feasibility of certain steps in [BKV25].

Conflict with [BKV25]: The authors note that their results appear inconsistent with intermediate claims in [BKV25] regarding the extension of sections, although they state these results do not contradict the main theorem of that work.
Topological Nature of Obstructions: The authors emphasize that their obstructions are topological. Consequently, they apply to all deformations of the fibration $J \to \mathbb{P}^d$ , not just those arising from deformations of the underlying pair $(S, H)$ . This is illustrated by the example of O'Grady 10-folds deforming to LSV 10-folds.
Regularity of Sections: Remark 10 highlights that every rational section of the map $p: J(S, H) \to |H|$ is regular. The authors explicitly state they do not know if this holds for all Lagrangian fibrations of smooth hyperkähler varieties.

The work relies on standard tools in algebraic geometry (Shioda-Tate isomorphism, Hodge index theorem, monodromy) to derive constraints on the existence of global sections, suggesting that the "right setting" for these problems is the universal Jacobian of linear systems rather than general Lagrangian fibrations.

The Story of the Impossible Stretch

The Topological Obstacle Course

The Takeaway

More like this