Moduli of bundles and semiorthogonal decomposition

Kai Xu · 2025

Read Moduli of bundles and semiorthogonal decomposition by Kai Xu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-19948.txt.

Introduction

Kai Xu decomposes spaces of bundles on a curve into simpler building blocks. The objects are moduli of principal bundles, the stack of all bundles for a group G over a curve and the coarse moduli space of semistable ones. Semiorthogonal decomposition is a way of splitting a triangulated category into ordered pieces, and Xu constructs such splittings whose pieces are the curve's symmetric powers.

The proof assembles heavy machinery: Borel-Weil-Bott theory for loop groups, the highest-weight structure of current-group representations, derived theta-stratification, and local-global compatibility of Kac-Moody localization. Both the stack and the semistable space receive decompositions, unifying their derived categories with concrete geometric ingredients. Results of this kind feed into broader programs relating moduli spaces, representation theory, and the geometric Langlands circle of ideas.

Catalog record

CREATORKai Xu
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE72 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-19948

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Moduli of bundles and semiorthogonal decomposition Kai Xu Abstract . In this paper we construct semiorthogonal decompositions of moduli of principal bundles on a curve into its symmetric powers, for both the moduli stack of all G G -bundles and the coarse moduli space of semistable G G -bundles. The essential ingredients in the proof include Borel-Weil-Bott theory for loop groups, highest weight structure of current group representation, derived Θ \Theta -stratification and local-global compatibility of Kac-Moody localization. 1. Introduction Fano varieties occupy a central place in birational geometry as part of the building blocks of general algebraic varieties. Semiorthogonal decompositions of the derived category of coherent sheaves on a Fano variety are deeply related to its birational geometry and motive [ 19 ] . In contrast to the Calabi-Yau and general type case, the derived category of a Fano variety is always decomposable, and its semiorthogonal decomposition is closely related to its motivic decomposition and rationality problem. Moduli theory provides a rich source of higher-dimensional Fano varieties: let X be a smooth projective curve of genus g>1 , L be a line bundle on X , then the coarse moduli space 1 1 1 We will always use Roman type for coarse moduli space and Italic type for moduli stack. Bun_{n}^{L,ss} of semistable vector bundles on X with rank n and determinant L , are Fano varieties of dimension (g-1)(n^{2}-1) (smooth if \delta=(n,\deg L)=1 and with Gorenstein canonical singularity in general [ 2 ] ). Analogous construction exists for any semisimple algebraic groups G . In low rank examples, they are known to admit a motivic decomposition into products of symmetric powers \mathrm{Sym}^{\lambda_{1}}X\times\cdots\times\mathrm{Sym}^{\lambda_{n-1}}X , and they are believed to lift to a semiorthogonal decomposition. [ 22 , 21 ] This is one of our main results: 1 . Let X be a smooth projective curve of genus g>1 , L be a line bundle on X so that \delta=(n,\deg L)=1 . We have a semiorthogonal decomposition for {Bun}_{n}^{L,ss} : \mathrm{D}^{b}({Bun}_{n}^{L,ss})=\langle\theta^{\otimes k}\otimes\mathrm{D}^{b}(\mathrm{Sym}^{\lambda_{1}}X\times\cdots\times\mathrm{Sym}^{\lambda_{n-1}}X),\cdots\rangle where \theta is the theta line bundle (positive generator of the Picard group), integers (k,\lambda_{1},\cdots,\lambda_{n-1}) satisfy 0\leq k\leq 1,0\leq\lambda_{i} , \sum_{i\leq l}(\lambda_{i}-(g-1))i(n-l)+\sum_{i>l}(\lambda_{i}-(g-1))l(n-i)\leq 0 for all 0<l<n , and when k=1 all other upper bounds are strict. \cdots is the semiorthogonal complement of the previous subcategories. The derived categories and semiorthogonal decompositions are well-behaved (i.e. saturated) in the smooth case, and in the singular case, we can replace \mathrm{D}^{b}(Bun_{n}^{L,ss}) by a categorical/non-commutative resolution of singularity \mathrm{D}^{b}(Bun_{n}^{L,ss})^{\sim} . There are natural constructions from the rigidified moduli stack \mathrm{D}^{b}(\mathrm{Bun}_{n}^{L,ss}/\mathrm{B}\mu_{n}) (and similarly for \mathrm{D}^{b}(Bun_{n}^{d,ss}) ). 2 . Let X be a smooth projective curve of genus g>1 , L be a line bundle on X , we have a non-commutative resolution of {Bun}_{n}^{L,ss} and a semiorthogonal decomposition: \mathrm{D}^{b}({Bun}_{n}^{L,ss})^{\sim}=\langle\theta^{\otimes k}\otimes\mathrm{D}^{b}(\mathrm{Sym}^{\lambda_{1}}X\times\cdots\times\mathrm{Sym}^{\lambda_{n-1}}X),\cdots\rangle where \theta is the theta line bundle (positive generator of the Picard group), integers (k,\lambda_{1},\cdots,\lambda_{n-1}) satisfy 0\leq k\leq\delta , \delta divides \sum i\lambda_{i} , \sum_{i\leq l}(\lambda_{i}-(g-1))i(n-l)+\sum_{i>l}(\lambda_{i}-(g-1))l(n-i)\leq 0 for all 0<l<n , and when k=\delta all other upper bounds are strict. \cdots is the semiorthogonal complement of the previous subcategories. More general semiorthogonal decomposition exist for moduli of principal bundles: we can define a moduli space {Bun}_{G}^{\xi,ss} of twisted G -bundles depending on a Z(G) -gerbe \xi\in H^{2}(X,Z(G)) which recovers {Bun}_{n}^{L,ss} for G=\mathrm{SL}_{n} and \xi=\xi(L) the n -th root gerbe of L . 3 . Let X be a smooth projective curve of genus g>1 , G be a semisimple algebraic group of rank r , Z(G) be the center of G and \xi\in H^{2}(X,Z(G)) , we have a semiorthogonal decomposition for {Bun}_{G}^{\xi,ss} : \mathrm{D}^{b}({Bun}_{G}^{\xi,ss})^{\sim}=\langle\theta^{\otimes k}\otimes\mathrm{D}^{b}(\mathrm{Sym}^{\lambda_{1}}X\times\cdots\times\mathrm{Sym}^{\lambda_{r}}X),\cdots\rangle where \theta is the theta line bundle, integers (k,\lambda_{1},\cdots,\lambda_{r}) satisfy 0\leq k\leq\delta , (g-1)\rho-\lambda lies in the cone generated by positive roots, and lies in the interior if k=\delta . Here \rho is half the sum of positive roots, \lambda=\sum_{i}\lambda_{i}\omega_{i} is the dominant weight with coefficients \lambda_{i} , \delta=\frac{h^{\vee}}{\mathrm{ord}(\xi)} , h^{\vee} is the dual Coxeter number, and o is the order of the central character of the determinant line bundle. \cdots is the semiorthogonal complement of the previous subcategories, and \mathrm{D}^{b}({Bun}_{r}^{L,ss})^{\sim} is a categorical resolution of singularity of {Bun}_{G}^{\xi,ss} (which is \mathrm{D}^{b}({Bun}_{r}^{L,ss}) for smooth {Bun}_{G}^{\xi,ss} ) We now record two simple examples explicitly: for G=\mathrm{SL}_{2} the moduli is smooth only when \deg L=2k+1 , when G=\mathrm{SL}_{3} the moduli is smooth when \deg L=3k\pm 1 but they are isomorphic by conjugation. Corollary 1 . We have a semiorthogonal decomposition for {Bun}_{2}^{L,ss} when \mathrm{deg}\,L is odd where the last \cdots is a phantom category and conjecturally 0 : \displaystyle\mathrm{D}^{b}({Bun}_{2}^{L,ss})=\langle \displaystyle\mathrm{D}^{b}(pt),\mathrm{D}^{b}(X),\cdots,\mathrm{D}^{b}(\mathrm{Sym}^{g-1}X), \displaystyle\theta\otimes\mathrm{D}^{b}(pt),\theta\otimes\mathrm{D}^{b}(X),\cdots,\theta\otimes\mathrm{D}^{b}(\mathrm{Sym}^{g-2}X),\cdots\rangle This celebrated conjecture, raised independently by Belmans-Galkin-Mukhopadhyay and Narasimhan, was proved 2 2 2 None of the two papers prove \cdots=0 half of the conjecture, it was proved later by Tevelev in [ 33 ] . independently by Tevelev-Torres [ 32 ] and Xu-Yau [ 36 ] . See also [ 20 ] for following works. Our previous method does not seem to yield a complete decomposition for r>3 , although it provides partial decompositions as discussed above. Corollary 2 . We have a semiorthogonal decomposition for {Bun}_{3}^{L,ss} when \mathrm{deg}\,L is not divisible by 3 where k_{1}+2k_{2},2k_{1}+k_{2}\leq 3(g-1) and both inequality should be strict for components with the factor \theta : \displaystyle\mathrm{D}^{b}({Bun}_{3}^{L,ss})=\langle\mathrm{D}^{b}(pt),\theta\otimes\mathrm{D}^{b}(pt),\cdots,\mathrm{D}^{b}(\mathrm{Sym}^{k_{1}}X\times\mathrm{Sym}^{k_{2}}X), \displaystyle\theta\otimes\mathrm{D}^{b}(\mathrm{Sym}^{k_{1}}X\times\mathrm{Sym}^{k_{2}}X),\cdots,\mathrm{D}^{b}(\mathrm{Sym}^{g-1}X\times\mathrm{Sym}^{g-1}X),\cdots\rangle Although we started with the Fano variety Bun_{n}^{L,ss} , it is more natural to view it as the coarse moduli space of the underlying stack. We have natural morphisms Bun_{n}^{L,ss}\longleftarrow\mathrm{Bun}_{n}^{L,ss}/\mathrm{B}\mu_{n}\longleftarrow\mathrm{Bun}_{n}^{L,ss}\longrightarrow\mathrm{Bun}_{n}^{L} , where Roman type is used for moduli stacks and the map \mathrm{Bun}_{n}^{L,ss}/\mathrm{B}\mu_{n}\longrightarrow Bun_{n}^{L,ss} gives a categorical resolution of singularity. So the study of \mathrm{D}^{b}(\mathrm{Bun}_{n}^{L,ss})^{\sim}\simeq\mathrm{D}^{b}(\mathrm{Bun}_{n}^{L,ss})^{\mathrm{B}\mu_{n}} is reduced to the study of \mathrm{D}^{b}(\mathrm{Bun}_{n}^{L,ss}) and Theorem 2 is similarly reduced to the following: 4 . We have a semiorthogonal decomposition for {\mathrm{Bun}}_{G}^{\xi,ss} : \mathrm{D}^{b}({\mathrm{Bun}}_{n}^{L,ss})=\langle\mathcal{L}^{\otimes k}\otimes\mathrm{D}^{b}(\mathrm{Sym}^{\lambda_{1}}X\times\cdots\times\mathrm{Sym}^{\lambda_{n-1}}X)\rangle where \mathcal{L} is the fundamental determinant line bundle, k\in\mathbb{Z} and \lambda\in\mathbb{N}^{r} . The open substack \mathrm{Bun}_{G}^{\xi,ss}\longrightarrow\mathrm{Bun}_{G}^{\xi} is the semistable locus determined by the Harder-Narasimhan filtration, and the machinery of derived \Theta -stratification developed by Halpern-Leistner allows us to relate their derived categories. \mathrm{D}^{b}(\mathrm{Bun}_{G}^{\xi,ss})\longrightarrow\mathrm{D}^{b}(\mathrm{Bun}_{G}^{\xi}) is a full subcategory which satisfies certain restrictions of weights on the polystable locus. (This is how the finite subsets arise.) So Theorem 3 is reduced to the following: 5 . For a smooth projective curve X we have the following semiorthogonal decomposition \mathrm{D}^{b}(\mathrm{Bun}_{G}^{\xi})=\langle\mathcal{L}^{\otimes k}\otimes\mathrm{D}^{b}(\mathrm{Sym}^{\lambda}X)\rangle where \mathcal{L} is the fundamental determinant line bundle, k\in\mathbb{Z} , \lambda\in\Lambda^{+}\simeq\mathbb{N}^{r} are dominant weights, and \mathrm{Sym}^{\lambda}X=\prod\mathrm{Sym}^{\lambda_{i}}X are the corresponding variety of colored divisors. This also has a non-projective analogue: 6 . For a smooth non-projective curve X=\mathrm{Spec\,}A we have the following semiorthogonal decomposition \mathrm{D}^{b}(\mathrm{Bun}_{G})=\mathrm{Rep}\,G(A)=\langle\mathrm{D}^{b}(\mathrm{Sym}^{\lambda}X)\rangle where \lambda\in\Lambda^{+}\simeq\mathbb{N}^{r} are dominant weights, and \mathrm{Sym}^{\lambda}X=\prod\mathrm{Sym}^{\lambda_{i}}X=\bigotimes(A^{\otimes\lambda_{i}})^{S_{i}} are the corresponding variety of colored divisors. The full moduli stack {\mathrm{Bun}}_{G}^{\xi} is much easier to study because of the Weil uniformization, which realizes {\mathrm{Bun}}_{G}^{\xi} as a double coset. This is most well-studied for trivial \xi , where we have {\mathrm{Bun}}_{G}^{\mathcal{O}}=\mathrm{Bun}_{G}=\prod_{x\in X}G(\mathcal{O}_{x})\backslash\prod_{x\in X}G(\mathcal{K}_{x})/G(K_{X}) Here we are taking the restricted product, the reader is referred to Section 2 for details. We prove that for nontrivial \xi we need to twist the right-hand side by the outer-automorphism of G(K_{X}) corresponding to \xi : {\mathrm{Bun}}_{r}^{L}=\mathrm{Bun}_{G}^{\xi}=\prod_{x\in X}G(\mathcal{O}_{x})\backslash\prod_{x\in X}G(\mathcal{K}_{x})/G(K_{X})^{\xi} This double coset representation allows us to reduce the study of tautological bundles and derived category to representation theoretical problems about loop group G(\mathcal{K}_{x}) and current group G(\mathcal{O}_{x}) . G(K_{X})

Read the complete text