🔢 mathematics

Semiorthogonal decompositions for stacks

This paper presents a systematic construction of semiorthogonal decompositions for derived categories of coherent sheaves on quasi-smooth derived algebraic stacks over $\mathbb{C}$ , where the summands are indexed by the component lattice and defined via weight conditions and parabolic induction, yielding new decompositions for moduli stacks of $G$ -bundles, $G$ -Higgs bundles, and $G$ -local systems for reductive groups of arbitrary type.

Original authors: Chenjing Bu, Tudor Pădurariu, Yukinobu Toda

Published 2026-05-26

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Chenjing Bu, Tudor Pădurariu, Yukinobu Toda

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 understand a massive, complex, and slightly messy city. This city is built on strange, warped geometry (mathematical "stacks"), and it's filled with millions of different types of buildings (mathematical "objects"). The goal of this paper is to figure out how to take this chaotic city apart, piece by piece, in a very organized way so that we can study each piece individually without losing the big picture.

Here is the breakdown of what the authors, Chenjing Bu, Tudor Pădurariu, and Yukinobu Toda, have achieved, using simple analogies.

1. The Problem: A Messy, Warped City

In mathematics, there are objects called derived algebraic stacks. Think of these as cities where the ground isn't flat; it's warped, folded, and has "ghost" layers (this is what "derived" means). These cities often represent collections of solutions to complex equations, like all possible ways to wrap a rubber band around a shape or all possible configurations of a physical system.

The authors are interested in the "derived category of coherent sheaves" on these stacks. In our analogy, this is like the complete library of every possible blueprint, map, and instruction manual for every building in this warped city. The problem is that this library is too huge and tangled to read all at once.

2. The Solution: A Systematic "De-Construction"

The paper provides a systematic method to break this giant library into smaller, manageable, and non-overlapping sections. In math, this is called a Semiorthogonal Decomposition.

Think of it like sorting a giant pile of mixed-up LEGO bricks. Instead of trying to build one giant, confusing tower, you sort the bricks into separate bins:

Bin A: All red 2x4 bricks.
Bin B: All blue 1x2 bricks.
Bin C: All special transparent pieces.

The magic of this paper is that they found a rule to sort these "bricks" (mathematical objects) so that:

No overlap: You never find a brick in two bins at once.
No confusion: If you look at a brick in Bin A, you don't need to worry about Bin B to understand it.
Complete: If you put all the bins back together, you get the exact original pile.

3. The Sorting Rule: "Weight" and "Windows"

How do they decide which brick goes in which bin? They use a concept called weights.

Imagine the city has a giant, invisible wind blowing through it. Some buildings are heavy and stay put; others are light and get blown around. The "weight" is a measurement of how a building reacts to this wind.

The authors use a "window" to catch specific types of buildings.

The Window: Imagine a window frame with a specific size and position.
The Rule: Only buildings with a "weight" that fits inside this window are allowed in.
The Sorting: They create a series of windows, each tuned to catch a different range of weights. By moving these windows around, they can sort the entire city into distinct groups.

4. The "Component Lattice": The City's Blueprint

To know exactly how to set up these windows, the authors use a tool called the Component Lattice.

Analogy: Imagine the city has a hidden skeleton or a grid system that dictates how the buildings are connected. This grid is the "Component Lattice."
What it does: It maps out the different "types" of symmetry in the city. Just as a crystal has a repeating pattern, these mathematical cities have repeating patterns of symmetry. The lattice is the map of these patterns.
The Result: The authors use this map to index their bins. Instead of just saying "Bin 1, Bin 2," they say "Bin A (Levi subgroup L, weight w)," which tells you exactly where that bin sits in the grand scheme of the city's geometry.

5. The "Parabolic Induction": The Delivery Truck

Once they have sorted the bricks into the bins, they need to put them back into the main library in a specific order. They use a tool called Parabolic Induction.

Analogy: Think of this as a specialized delivery truck. It doesn't just drop bricks anywhere; it takes a sorted bin from a smaller, simpler neighborhood (a "sub-stack") and delivers it to the main city in a way that fits perfectly with the existing structure.
Why it matters: This ensures that when you look at the whole library again, the pieces fit together seamlessly, preserving the mathematical relationships between them.

6. Real-World Examples (The "Cities" They Studied)

The authors didn't just do this for abstract theory; they applied it to real, famous mathematical "cities":

G-bundles on a curve: Imagine wrapping a ribbon around a loop. There are millions of ways to do this. The authors showed how to sort all these wrapping methods.
Higgs bundles: Think of these as ribbons with a special "magnetic field" attached.
Local systems: Imagine a map where every point has a secret code, and you need to sort all possible code combinations.
Quivers: These are diagrams of dots and arrows. The authors sorted all the possible ways to assign numbers to these dots and arrows.

7. Why This Matters (According to the Paper)

The paper connects this sorting method to Donaldson–Thomas (DT) theory.

The Connection: DT theory is like a census. It tries to count the "buildings" in these cities to understand their shape and properties.
The Breakthrough: By breaking the city into these sorted bins (which the authors call quasi-BPS categories), they are creating a "categorical census." Instead of just counting the total number of buildings, they are counting how many buildings are in each specific "weight" category.
The Benefit: This helps mathematicians understand the deep structure of these spaces, potentially revealing hidden symmetries and relationships that were previously invisible because the data was too jumbled.

Summary

In short, this paper gives mathematicians a universal sorting algorithm for complex, warped geometric spaces. It takes a chaotic collection of mathematical objects, uses a "wind" (weights) and a "map" (component lattice) to sort them into neat, non-overlapping piles, and then shows how to reassemble them perfectly. This allows for a deeper, more precise understanding of the geometry of these spaces, particularly in the context of counting and classifying their structures.

Technical Summary: Semiorthogonal Decompositions for Stacks

Problem Statement
The paper addresses the construction of systematic semiorthogonal decompositions (SODs) for the derived categories of coherent sheaves, $\text{Coh}(X)$ , on quasi-smooth derived algebraic stacks $X$ over $\mathbb{C}$ . While SODs for linear quotient stacks and specific moduli spaces (such as those for quiver representations or vector bundles on curves) have been established in prior works by Ballard–Favero–Katzarkov, Halpern-Leistner, Špenko–Van den Bergh, and Pădurariu–Toda, these constructions often relied on specific linear structures or required restrictive stability conditions. The challenge lies in generalizing these decompositions to a broad class of non-linear, quasi-smooth stacks arising in intrinsic Donaldson–Thomas (DT) theory, including moduli stacks of $G$ -bundles, Higgs bundles, connections, and local systems for general reductive groups $G$ (not limited to type $A$ ).

Methodology
The authors employ the framework of intrinsic Donaldson–Thomas theory (developed in [17–19]) to define a combinatorial structure called the component lattice, denoted $\text{CL}_\mathbb{Q}(X)$ . This lattice encodes the cocharacters and Weyl groups of the automorphism groups of points in $X$ .

The methodology proceeds through the following steps:

Component Lattice and Graded Points: The decomposition is indexed by the rationalized component lattice $|\text{CL}_\mathbb{Q}(X)| = \pi_0 \text{Map}(\ast/\mathbb{G}_m, X)$ . Each element $\lambda$ corresponds to a stack of graded points $X_\lambda$ and a stack of filtered points $X^+_\lambda$ .
Hall Induction: The authors utilize the Hall induction functor $\star_\lambda: \text{Coh}(X_\lambda) \to \text{Coh}(X)$ , defined via the correspondence $X_\lambda \xleftarrow{\text{gr}_\lambda} X^+_\lambda \xrightarrow{\text{ev}_\lambda} X$ . This functor generalizes the parabolic induction functor.
Window Subcategories: For each $\lambda$ , a "window subcategory" $W_{X_\lambda}(\delta_\lambda) \subset \text{Coh}(X_\lambda)$ is defined. These are full subcategories determined by weight conditions relative to the $\mathbb{G}_m$ -action on $X_\lambda$ , centered by a linear function $\delta: \text{CL}_\mathbb{Q}(X) \to \mathbb{Q}$ and a quadratic norm $q$ on the lattice.
Local-to-Global Argument:
- Linear Case: The decomposition is first established for linear quotient stacks $V/G$ where $V$ is a quasi-symmetric representation (Theorem 2.1.6).
- Smooth Case: Using the fact that smooth stacks with good moduli spaces are étale locally modeled on linear quotients, the result is globalized via descent arguments (Theorem 3.2.3).
- Quasi-Smooth Case: For quasi-smooth stacks (which are local models for $(-1)$ -shifted symplectic stacks), the authors utilize Koszul duality to relate $\text{Coh}(X)$ to categories of matrix factorizations $\text{MF}(Y, f)$ on smooth stacks. The decomposition for matrix factorizations (Theorem 3.5.4) is then transferred back to the quasi-smooth setting (Theorem 4.1.3).
Singular Supports: The construction is refined to include subcategories of objects with singular support contained in a saturated closed subset $Z \subset |T^*[-1]X|$ (Theorem 4.3.3).

Key Contributions and Results
The paper provides a unified construction of SODs for a wide range of moduli stacks. The main results are:

General Decomposition (Theorem 4.1.3): For a quasi-symmetric, quasi-smooth, qca derived stack $X$ with a good moduli space, there exists an SOD:
$\text{Coh}(X) = \left\langle \star_\lambda W_{X_\lambda}(\delta_\lambda) \mid \lambda \in |\text{CL}_\mathbb{Q}(X)| \right\rangle$
The summands are indexed by the component lattice, ordered by the $q$ -norm of $\lambda$ . The functors $\star_\lambda$ are fully faithful, and the images are admissible and $\text{Perf}(X)$ -linear.
Singular Support Decomposition (Theorem 4.3.3): A parallel decomposition exists for $\text{Coh}_Z(X)$ , the subcategory of coherent sheaves with singular support in a saturated subset $Z$ . This is particularly relevant for the nilpotent cone.
Matrix Factorizations (Theorem 3.5.4): An SOD is constructed for the category of matrix factorizations $\text{MF}(X, f)$ on a smooth stack, which serves as a local model for critical categories in DT theory.
Concrete Applications: The theory is applied to obtain explicit SODs for:
- Moduli stacks of semistable $G$ -bundles ( $\text{Bun}^{ss}_G$ ), $G$ -Higgs bundles ( $\text{Higgs}^{ss}_G$ ), and $G$ -connections ( $\text{Conn}_G$ ) on curves.
- Moduli stacks of $G$ -local systems ( $\text{Loc}_G$ ) on 2-manifolds.
- Moduli stacks of semistable representations of quivers and preprojective algebras.
- Moduli stacks of semistable coherent sheaves on surfaces.

In these examples, the summands are indexed by Levi subgroups $L \subset G$ and specific weight parameters, generalizing previous results that were often restricted to $GL(n)$ or required complex ordering conditions.

Significance and Claims
The paper claims to provide a systematic and general construction that removes previous restrictions on the type of reductive group (allowing non-type $A$ groups) and the nature of the stack (moving from linear categories to intrinsic DT stacks).

Simplification of Ordering: The authors note that their approach yields a much simpler description of the ordering of summands compared to previous works (e.g., [64; 68; 71]). The order is determined by the $q$ -norm of the lattice elements, whereas earlier methods often required intricate stability refinements or complicated inequalities.
Categorical DT Theory: The decompositions are presented as categorical analogues of cohomological integrality. The summands, referred to as quasi-BPS categories, are intended to categorify the BPS invariants. The paper notes that while additive invariants (like periodic cyclic homology) do not directly recover the cohomological integrality decomposition, the structure of these categories is expected to be fundamental for understanding categorical DT theory and BPS structures.
Geometric Langlands: The decompositions for stacks like $\text{Conn}_G$ and $\text{Loc}_G$ are linked to the geometric Langlands program. The authors formulate a refined version of the Dolbeault geometric Langlands conjecture and a similar conjecture for the de Rham setting (Conjecture 4.4.9), suggesting that the window subcategories correspond to specific microlocal categories of D-modules under the Langlands equivalence.
Limitations: The authors modestly state that they do not prove the equivalence of these categorical decompositions with the BPS cohomology decomposition via additive invariants, nor do they fully characterize the structure of quasi-BPS categories (e.g., whether they are non-commutative resolutions). These are identified as directions for future work.

In summary, the paper establishes a robust combinatorial framework for decomposing derived categories of coherent sheaves on quasi-smooth stacks, unifying various known results and extending them to a broader geometric context central to modern enumerative geometry and representation theory.