🔢 mathematics

Higher genus reduced Gromov--Witten invariants via desingularizations of sheaves

This paper introduces two desingularization constructions for coherent sheaves on algebraic stacks—the Rossi and Hu-Li methods—to define higher genus reduced Gromov-Witten invariants for a broad class of GIT quotients.

Original authors: Alberto Cobos Rabano, Etienne Mann, Cristina Manolache, Renata Picciotto

Published 2026-04-07

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Alberto Cobos Rabano, Etienne Mann, Cristina Manolache, Renata Picciotto

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 trying to count the number of ways a rubber band can wrap around a complex, bumpy sculpture. In the world of mathematics, this is called counting curves on a shape. This is the heart of Gromov–Witten theory, a field that tries to understand the geometry of shapes by counting the paths (curves) that live on them.

However, there's a problem. When you try to count these paths, your counting machine (a mathematical tool called a "moduli space") gets confused. It starts counting not just the nice, smooth rubber bands, but also:

Broken bands: Rubber bands that have snapped into pieces.
Ghost bands: Paths that look like they exist but are actually just mathematical noise (degenerate maps).

These "bad" paths mess up the count. To get the true answer, mathematicians need to filter them out. They need to build a "cleaner" version of their counting machine that only sees the smooth, perfect paths. This filtered count is called a Reduced Gromov–Witten invariant.

For a long time, mathematicians could only build this "cleaner machine" for simple cases (like counting paths on a sphere or a torus). But for more complex shapes and higher levels of complexity (called "higher genus," which is like counting paths on a pretzel with many holes), the machine was too broken to fix.

This paper is the instruction manual for fixing that machine for any level of complexity.

Here is how the authors did it, using some creative analogies:

1. The Problem: A Tangled Knot of Data

Imagine the shape you are studying is a city, and the rubber bands are delivery routes. The "moduli space" is a giant map of all possible routes.

The Issue: In the "bad" areas of the city (where the roads are broken or the terrain is weird), the delivery trucks (the mathematical data) get stuck. The map becomes a tangled knot. Some parts of the map are smooth, but others are crumpled up, making it impossible to count the routes accurately.
The Goal: You need to smooth out the crumpled parts of the map without changing the actual city. You want a "desingularized" map where every route is clear and distinct.

2. The Solution: The "Sheaf" as a Bundle of Strings

In math, the data that gets tangled is called a sheaf. Think of a sheaf as a bundle of strings attached to every point in your city.

In the "smooth" parts of the city, these strings are neat and organized (mathematically, they are "locally free").
In the "bad" parts, the strings get knotted, tangled, or even disappear (they have "torsion" or become "singular").

The authors' main trick is to blow up the city. In math, "blowing up" doesn't mean an explosion; it means taking a messy point and replacing it with a whole new, clean space that resolves the mess.

3. The Two New Tools: Rossi and Hu-Li

The paper introduces two specific ways to "blow up" the city to untangle the strings. Think of these as two different construction crews with different blueprints:

The Rossi Crew (The Minimalist):
- Goal: They just want to untangle the main bundle of strings so they are neat and free.
- Method: They look at the "generic" (average) state of the strings and build a new map that forces the strings to be neat everywhere.
- Analogy: Imagine you have a knotted necklace. The Rossi method is like carefully pulling the knot apart just enough so the chain is straight again, without adding any extra beads. It's the most efficient way to fix the mess.
The Hu-Li Crew (The Diagonalizer):
- Goal: They want to not only untangle the main strings but also organize the "trash" (the torsion) that was left behind.
- Method: They rearrange the strings so that they line up perfectly in a grid (diagonal). This makes the "trash" strings sit neatly on their own separate shelves.
- Analogy: If the Rossi method just straightens the necklace, the Hu-Li method also organizes the clasp and the clasp's chain into a perfect, labeled box. It's more work, but it gives you a much clearer view of everything that happened, including the broken pieces.

4. The Big Breakthrough: "It Doesn't Matter Which Crew You Hire"

The most surprising and important result of the paper is this: Both crews produce the exact same final count.

Even though the Rossi map and the Hu-Li map look different (one is minimal, the other is more detailed), when you use them to count the rubber bands, you get the same number.

This is like two different architects designing two different bridges to cross a river. One is a sleek, modern suspension bridge; the other is a sturdy, multi-level concrete bridge. They look totally different, but if you drive a car across both, you end up at the exact same destination at the same time.

5. Why This Matters

Before this paper, we could only count these curves for simple shapes (genus 1 and 2). This paper proves that we can now count them for any shape, no matter how complex (genus 3, 4, 100, etc.).

The Impact: This opens the door to solving deep mysteries in physics and geometry. For example, in string theory (physics), these counts help predict how the universe might behave. In pure math, they help us understand the hidden structure of shapes that were previously too messy to study.

Summary

The authors took a messy, tangled mathematical problem (counting curves on complex shapes) and invented two new, powerful tools (the Rossi and Hu-Li blow-ups) to untangle it. They proved that no matter which tool you use, the final answer is the same. This allows mathematicians to finally count these curves for any shape, unlocking new secrets in geometry and physics.

In short: They found a universal way to clean up the mathematical "noise" so we can finally hear the true "music" of the universe's shapes.

1. Problem Statement

The paper addresses the definition and computation of reduced Gromov–Witten (GW) invariants for smooth projective varieties (specifically hypersurfaces and GIT quotients) in arbitrary genus $g$ .

Context: Standard GW invariants count stable maps but include contributions from degenerate maps where the source curve is reducible or singular. "Reduced" invariants are designed to isolate the contribution from the main component of the moduli space (the closure of the locus where the source curve is smooth).
The Obstacle: While reduced invariants were well-defined for genus 1 and 2 (via works by Vakil, Zinger, Hu, Li, etc.), extending this to higher genera is difficult. The moduli space of stable maps $\mathcal{M}_{g,n}(X, d)$ is generally singular and has components of different dimensions. The obstruction sheaf defining the virtual fundamental class is not locally free on the main component, preventing a direct definition of reduced invariants via standard intersection theory.
Goal: Construct a birational modification of the moduli space that resolves these singularities and makes the relevant obstruction sheaves locally free, thereby allowing a rigorous definition of reduced invariants for all $g$ .

2. Methodology

The authors' approach shifts the focus from blowing up the moduli space of maps directly (which is combinatorially complex) to desingularizing the coherent sheaves that govern the geometry of these moduli spaces.

A. Desingularization of Coherent Sheaves

The core technical tool is the construction of birational modifications of a base stack $P$ (typically the Picard stack parametrizing curves and line bundles) to "flatten" a coherent sheaf $F$ . They provide two distinct constructions:

The Rossi Construction (Minimal Desingularization):
- Based on the Nash blow-up and Raynaud–Gruson flattening.
- Goal: Make the torsion-free part of the pullback sheaf, $(p^*F)_{tf}$ , locally free.
- Construction: Defined as the closure of the graph of the map from the generic locus of $P$ to the relative Grassmannian of rank- $r$ quotients of $F$ .
- Property: It satisfies a universal property: it is the minimal birational modification making $(p^*F)_{tf}$ locally free.
The Hu–Li Construction (Diagonalization):
- Based on the work of Hu and Li (and Grivaux).
- Goal: Make the pullback sheaf $p^*F$ diagonal. A sheaf is diagonal if its Fitting ideals are locally principal, which implies a filtration where the graded pieces are locally free on Cartier divisors.
- Construction: An iterated blow-up of the base stack along the Fitting ideals of $F$ .
- Property: This construction is more refined than the Rossi blow-up; it not only flattens the torsion-free part but also organizes the torsion part into a well-behaved structure (a union of vector bundles over closed substacks).

B. Application to Moduli of Stable Maps

The authors apply these sheaf desingularizations to the moduli space of stable maps to projective space, $\mathcal{M}_{g,n}(\mathbb{P}^r, d)$ .

Embedding: $\mathcal{M}_{g,n}(\mathbb{P}^r, d)$ is realized as an open substack of an abelian cone $\text{Spec}(\text{Sym}(F))$ over the Picard stack $\text{Pic}_{g,n,d}$ , where $F = R^1\pi_*(L^\vee \otimes \omega_{\pi})$ .
The Main Component: The "main component" of the moduli space corresponds to the main component of the abelian cone, which is the closure of the locus where the source curve is smooth.
The Resolution: By pulling back the Picard stack via the Rossi or Hu–Li blow-up ( $\tilde{\text{Pic}} \to \text{Pic}$ ), the authors construct a new moduli space $\tilde{\mathcal{M}}_{g,n}(\mathbb{P}^r, d)$ . On this new space, the sheaf defining the obstruction theory becomes locally free (or diagonal), allowing the definition of a perfect obstruction theory relative to the base.

3. Key Contributions

All-Genus Construction of Reduced Invariants:
The paper defines reduced GW invariants for hypersurfaces in projective spaces and GIT quotients for any genus $g$ , provided the degree $d$ satisfies $d > 2g-2$ (ensuring the main component is well-behaved).
Equivalence of Constructions:
The authors prove that the reduced invariants defined via the Rossi blow-up and the Hu–Li blow-up coincide. This is a crucial result because the Rossi construction is minimal (conceptually simpler), while the Hu–Li construction provides the necessary geometric structure (diagonalization) to analyze boundary components and prove conjectures.
Universal Properties on Stacks:
They generalize the Rossi and Hu–Li constructions from schemes to Noetherian integral Artin stacks. They establish that these constructions satisfy universal properties (Theorems 5.2.2 and 5.2.6), ensuring the resulting stacks are independent of the choice of presentation.
Decomposition of Abelian Cones:
A significant technical result (Section 6) describes the irreducible components of the abelian cone of a diagonal sheaf. They show that the cone is topologically a union of vector bundles supported on specific subschemes. This decomposition is essential for understanding the geometry of the moduli space after blow-up.
Maps with Fields:
The paper extends the construction to maps with fields (following Chang–Li). They show that the Hu–Li blow-up of the moduli space of maps with fields yields a space where the irreducible components are smooth over the base. This provides a powerful tool for splitting the virtual fundamental class, which is necessary for proving relations between standard and reduced invariants.

4. Main Results

Theorem 1.0.2: Existence of integral Noetherian Artin stacks $\text{Bl}^F_P$ (Rossi) and $\text{Bl}^{HL}_F P$ (Hu–Li) with representable proper birational morphisms to $P$ , satisfying universal properties for flattening and diagonalizing a coherent sheaf $F$ .
Definition 7.4.4: Definition of reduced GW invariants for hypersurfaces $X \subset \mathbb{P}^r$ as intersection numbers against the virtual fundamental class of the main component of the blown-up moduli space.
Proposition 7.4.3: Proof that these invariants are independent of the specific desingularization chosen (Rossi vs. Hu–Li).
Theorem 7.5.1: The irreducible components of the blown-up moduli space of maps with fields are smooth over their image in the base stack, and the relevant obstruction sheaves are locally free.
Section 9: A detailed comparison in genus 1 showing that the Rossi blow-up yields a moduli space different from the classical Vakil–Zinger blow-up, yet producing the same invariants.

5. Significance and Impact

Unification: The paper unifies the definition of reduced invariants across all genera, removing the previous restriction to low genus ( $g \le 2$ ).
Conjecture Resolution: The construction provides the necessary geometric framework to attack Conjecture 1.0.4 (relating reduced GW invariants to Gopakumar–Vafa invariants for Fano threefolds) in higher genus. The Hu–Li blow-up provides the "maps with fields" space required to decompose the virtual class, a technique previously limited to low genus.
Conceptual Clarity: By moving the desingularization to the level of sheaves on the Picard stack rather than the moduli space of maps itself, the authors avoid the intractable combinatorics of local equations for high-genus moduli spaces. The gluing of local constructions is handled via the universal properties of the blow-ups.
Future Directions: The authors note that while the construction is intrinsic, a modular interpretation of the resulting stack (e.g., in terms of log maps or maps from singular curves) remains an open and desirable problem.

In summary, this work provides a robust, general machinery for defining and computing reduced Gromov–Witten invariants in higher genus by leveraging advanced techniques in the desingularization of coherent sheaves on algebraic stacks.