Professional Documents
Culture Documents
Remark 1.6
Remark 1.6
where the induced action on the coarse moduli space is trivial. We also estab-
lish a general local structure theorem for morphisms of algebraic stacks.
1. Introduction
The main goal of this paper is to provide a description of Deligne–Mumford
stacks X with a torus action where the induced action on the coarse moduli space
Xcms is trivial. Such actions arise naturally in moduli theory as the fixed locus
of a torus action on a Deligne–Mumford stack (see Remark 1.6). The motivation
for this paper is that an explicit understanding of such actions could be useful for
applications of localization theorems.
A prototypical example is the non-trivial Gm -action on Bµr arising from the
r
Kummer sequence 0 → µr → T = Gm − → Gm → 0: the induced map Bµr → BT
is a Gm -torsor as it’s the base change of Spec k → BGm . The main theorem below
states essentially that every Gm -action on a Deligne–Mumford stack X, which is
trivial on Xcms , is built from this nontrivial action on Bµr for some r.
1.1. The main theorem. We work over an algebraically closed field k.
Theorem 1.1. Let X = [X/Gnm ] → BGnm be a separated, finite type, and relatively
tame Deligne–Mumford morphism of algebraic stacks (i.e., X is a separated, finite
type, and tame Deligne–Mumford stack). Assume that X is reduced and connected,
and that the Gnm -action on the coarse moduli space Xcms of X is trivial.
(1) There is a diagonalizable group µr := µr1 × · · · × µrn for a unique tuple
r = (r1 , . . . , rn ) of integers with r1 |r2 | · · · |rn and a central closed subgroup
µr,X ⊂ IX /BGnm of the relative inertia stack restricting to a subgroup µr,X ⊂ IX
such that the Gnm -action on X descends to a trivial Gnm -action on the Deligne–
Mumford stack Y := X( µr,X and there is a factorization
X PP // X( µr,X ∼ = Y × BGnm
PPP
PPP
PPP
PPP
''
Xcms × BGnm
which satisfies the following universal property: if X → Y ′ × BGnm → Xcms ×
BGnm is another factorization where Y ′ is a Deligne–Mumford stack, then there
is a morphism g : Y → Y ′ unique up to unique isomorphism fitting in the
diagram
X ■ // Y × BGn // Xcms × BGn
■■ ✤ m 77 m
■■
■■ ✤ g×id ♥♥ ♥♥♥
♥
■■ ✤ ♥♥♥
■$$ ♥♥♥
Y ′ × BGnm .
Date: February 19, 2024.
1
2 J. ALPER AND F. JANDA
X ❖❖ // X( µr,X ∼
= Y × BGnm
❖❖❖
❖❖❖
❖❖❖ p1
❖❖❖
''
X( T ∼ =Y
and X( T is identified with the rigidification (X( µr,X )( Gnm by Gnm = T /µr .
Furthermore, X → Y is the T -gerbe associated to the µr -gerbe X → Y with
respect to the canonical inclusion of group schemes µr ⊂ T .
1.2. Special cases and our proof strategy. Given the technical nature of the
statement and the subtleties arising from torus actions on Deligne–Mumford stack,
we will try to unpack the conclusion in a series of remarks and examples.
Remark 1.2 (Torus equivariant interpretation). Part (1) asserts that Y is the
largest Deligne–Mumford stack with trivial Gnm -action factoring Gnm -equivariantly
as X → Y = X( µr,X → Xcms . Note that there is an identification Xcms ∼ = Ycms
of coarse moduli spaces. Note also that the assumption that the Gnm -action on
Xcms is trivial translates to the condition that the relative coarse moduli space of
X → BGnm is Xcms × BGnm .
For part (2), let Xtriv be the µr -gerbe over Y which is isomorphic to X as
Deligne–Mumford stacks over Y but is given the trivial Gnm -action. Then the Gnm -
equivariant µr -gerbe X → Y is obtained from the possibly non-trivial µr -gerbe
Xtriv → Y with trivial Gnm -action by twisting by the trivial µr -gerbe Bµr,Y → Y
with non-trivial Gnm -action, i.e. there is a Gnm -equivariant isomorphism of µr -gerbes
X∼
= Xtriv ∧µr Bµr,Y .
While the Gnm -action on X may be non-trivial, part (2) also asserts that there is a
smallest integer r such that the action becomes trivial after the reparameterization
T = Gnm → Gnm defined by (t1 , . . . , tn ) 7→ (tr11 , . . . , trnn ).
Example 1.3 (Case of a classifying stack). The crucial example—both to un-
derstand the conclusion of the main theorem and its proof—is a Gnm -action on a
classifying stack X = BG of a finite group G. Such an action corresponds to an
extension
(1.1) 1 → G → Γ → Gnm → 1
of algebraic groups (see Example 3.2) where X = [X/Gnm] = BΓ. We let T =
Γ0 ⊂ Γ be the identity component. As T is isomorphic to Gnm (see §4.1), the
intersection T ∩ G is isomorphic to µr for a unique tuple of integers r = (r1 , . . . , rn )
with r1 | · · · |rn . By Lemmas 4.1 and 4.2, T and µr are central in Γ, and there
is an identification Γ ∼ = G × Gnm where G := G/µr and Γ := Γ/µr . Part (1) of
Theorem 1.1 is the identifications
X( µr ∼
= BΓ ∼
= BG × BGnm
| {z }
Y ×BGn
m
STRUCTURE RESULTS FOR TORUS FIXED LOCI 3
|{z} BG .
BΓ → |{z}
BΓ → |{z}
X X( µr X( T
On the other hand, the conclusion of Theorem 1.1 may fail if X is nonreduced.
Consider the affine scheme U = Spec(k[ǫ]/ǫ2 ), and consider the action of Γ =
Gm on U defined by ǫ 7→ tǫ. The quotient stack X = [U/µ2 ] by the subgroup
µ2 ⊂ Γ is a nonreduced, connected Deligne–Mumford stack with coarse moduli
U/µ2 = Spec k. There is an induced action of Gm = Γ/µ2 on X, and the quotient
X = [X/Gm ] = [U/Γ] is a nonreduced and connected algebraic stack over BGm via
the structure morphism X → BGm induced by the squaring map Γ → Gm . Note
that X → Spec k and X → BGm are not µ2 -gerbes.
On the other hand, the theorem does extend to the nonreduced case if we modify
the hypotheses:
Remark 1.10. In fact, as we will see in Remark 3.3 below, when X is separated
and reduced, a torus action trivial on coarse moduli is automatically also trivial
after a reparameterization, and so Theorem 1.9 generalizes Theorem 1.1.
Remark 1.11 (Restriction to stabilizers). In the setting of Theorem 1.1, for every
point x ∈ X(k), there is a short exact sequence 1 → StabX (x) → StabX (x) →
Gnm → 1. The central subgroup µr,X ⊂ IX /Gnm restricts to the subgroup µr =
StabX (x)0 ∩ StabX (x) ⊂ StabX (x).
Remark 1.12. This paper was inspired and strengthens the second author’s work
[CJR22] (joint with Q. Chen and Y. Ruan) on punctured R-maps in the context
of gauged linear sigma models (GLSM). In fact, as explained in Section 2.2 below
Gm -actions play a central role in the geometry of GLSM. In particular, the notation
for the groups in the sequence (1.1) follows the one used in the work of Polishchuk–
Vaintrob [PV16, § 3.2], and Fan–Jarvis–Ruan [FJR18, § 2].
As µ-gerbes play a central role in the structure of torus fixed loci of Deligne–
Mumford stacks, it is interesting to study the behavior of invariants such as Picard
and Chow groups under taking µ-gerbes. We hope to explore this in future work.
It would also be interesting to study the following generalizations. Are there
analogous results for (tame) Artin stacks with a torus action or Deligne–Mumford
stacks with a non-abelian action? Is there an analogous description for the attractor
locus of a Gm -action?
1.3. A local structure theorem for morphisms. Our proof of Theorem 1.1
uses a special case of the following theorem, which we believe is independently
interesting. It is established using techniques from [AHR20] and [AHR21].
g f
BGy oo [Spec B/Gy ] // Y
2.1.2. Stable maps to projective space. Consider the Gm -action on PN given by dis-
tinct weights λ0 , . . . , λN , and the induced Gm -action on Mg,n (PN , d). We illustrate
Theorem 1.1 for each connected component of the fixed locus Mg,n (PN , d)Gm . Fol-
lowing [GP99, §4], each connected component corresponds to a marked graph Γ as
follows. If f : C → PN is a stable map fixed by Gm , then f (C) is a Gm -invariant
curve, and the image of every marked point, node, contracted component, and
ramification point is a Gm -fixed point of PN . Since the weights are distinct, the
Gm -fixed points of PN are the points p0 , . . . , pN corresponding to the standard ba-
sis elements and the Gm -invariant curves are the lines connecting pi to pj . Each
non-contracted component maps to one of these Gm -invariant curves and is rami-
fied over two points, which implies that each such component is rational and that
the map f restricted to this component is determined by its degree. The graph
Γ corresponding to f has vertices corresponding to the connected components of
f −1 ({p0 , . . . , pN }) and has edges corresponding to the non-contracted component.
6 J. ALPER AND F. JANDA
We now determine the unique integer r arising from Theorem 1.1 for the con-
nected component containing f . For each edge e connecting vertices v1 and v2 , let
λe,1 = λi(v1 ) and λe,2 = λi(v2 ) . We claim that
de
(2.1) r = lcm e edge of Γ .
gcd(de , λe,1 − λe,2 )
To see this, we consider the case of a single edge as the general case is similar
and only notationally more complicated. Let f : C → PN be a stable map with
a single component Ce that is not contracted and that maps d to 1 to the line
connecting pi to pj . We may assume that the other pointed components of C have
no automorphisms. The stabilizer of f in [Mg,n (PN , d)/Gm ] is
Stf = {(s, t) | sd = tλe,1 −λe,2 } ⊂ Aut(C) × Gm ,
where s and t are the coordinate of the Gm ’s corresponding to the automor-
phism group of the two-pointed Ce and the two-pointed line in PN . Letting
h = gcd(d, λe,1 − λe,2 ), the connected component St0f of Stf is given by the van-
ishing of sd/h − t(λe,1 −λe,2 )/h . The map St0f → Gm is therefore d/h to 1, and thus
r = d/h, which agrees with (2.1).
log
map1 to P◦k twisted by the logarithmic dualizing line bundle ωC according to the
Gm -action. One way to make this precise is using the notion of R-maps of [CJR21]:
An R-map is a morphism f : C → P◦ := [P◦k /Gm ] together with a 2-commutative
triangle
◦
P
f ②②
②<<
②②
②②②log
② ωC
C // BGm ,
BT × BGm // BGm
over T ,
G|T
■■
ψx ✈✈✈ ■■ ψy
■■
✈
✈✈✈ ■■
zz✈✈ ■$$
Innα
AutT (x) // Aut (y).
T
3.2. Group actions on stacks. We follow [Rom05] for our conventions for group
actions on stacks. A group action of an algebraic group G on an algebraic stack X
is a morphism µ : G × X → X of stacks such that the usual diagrams
idG ×µ (e,idX )
G ×S G ×S X // G ×S X X❍ // G ×S X
❍❍
❍❍
m×idX µ ❍❍ µ
idX ❍❍❍
µ
// G ##
G ×S X X
strictly commute.
Example 3.2 (Actions on BG). If G is a finite group, then a Gnm -action on BG is
equivalent to a short exact sequence
1 → G → Γ → Gnm → 1.
The action is trivial if and only if the sequence splits if and only if the sequence
splits trivially.
STRUCTURE RESULTS FOR TORUS FIXED LOCI 9
To see this, observe that such a short exact sequence induces a cartesian diagram
Gn m
// BG // Spec k
Spec k // BΓ // BGnm .
Proposition 3.4. Under the hypotheses of Theorem 1.1, assume that there is a
central diagonalizable closed subgroup scheme µr,X ⊂ IX /BGnm with r = (r1 , . . . , rn )
and a factorization
(3.1) X → X( µr,X ∼ = Y × BGn → Xcms × BGn ,
m m
// T i // Γ // K // 1
1 ❃❃
❃❃ j
p q ❃❃ k
❃❃
1 // Gnm // Γ // G // 1
with exact rows. Since p is surjective, and ker(p) → ker(q) is an isomorphism, the
snake lemma implies that k is an isomorphism, yielding the exact sequence (4.5).
Since T is normal in Γ, conjugation induces a homomorphism Γ → Aut(T ) =
GLn (Z). Similarly, since Gnm is normal in Γ, there is a homomorphism Γ →
12 J. ALPER AND F. JANDA
∼
Γ // Aut(Gn ).
m
Since Γ ∼
= G × Gnm , we conclude that Γ → Aut(Gnm ) is trivial and thus so is
Γ → Aut(T ).
Proposition 4.3. Theorem 1.1 holds for X = BΓ.
Proof. Lemma 4.2 gives central subgroups µr ⊂ Γ and T ⊂ Γ with isomorphisms
BΓ( µr = ∼ BΓ and BΓ( T ∼ = BG
where as above Γ = Γ/µr and G = G/µr . Lemma 4.1 gives an identification of
Γ∼= G × Gnm . Defining Y := BG, the factorization BΓ → BΓ( µr ∼ = Y × BGnm →
n
BGm satisfies the universal property by Proposition 3.4, yielding Part (1). Part (2)
follows from the exact sequence (4.4), while Part (3) follows from the identification
BΓ( T ∼ = BG = Y and (4.4).
4.2. Quotient case. The case of a quotient stack follows easily from the case of a
classifying stack. We will use the group-theoretic notation from Section 4.1.
Proposition 4.4. Theorem 1.9 holds for X = [U/Γ] where U is a connected and
quasi-separated algebraic space U of finite type over k.
Proof. By the hypotheses of Theorem 1.9, X := [U/G] → BG is a Gnm -equivariant
morphism such that the Gnm -action on X becomes trivial after a reparameterization
T ′ of Gnm . We may assume without loss generality that T ′ → Gnm factors through
T . Since the action of T ′ on X is trivial, we have [U/(Γ ×Gnm T ′ )] ∼ = [U/G] × BT ′
′ ′
over BT . We claim that since p : U → X is finite étale, the T action on U is
trivial. Indeed, as the T ′ action on X is trivial, A := p∗ OU can be written as the
direct sum A0 ⊕ A′ , where A0 is the weight 0 subspace and A′ is the direct sum
of the non-trivial weight subspaces. If I ⊂ OX denotes the nilradical, then IOU is
the nilradical of U (since U → X is étale) and IA = p∗ (IOU ) is the nilradical of
A. As T ′ acts trivially on Ured , A′ ⊂ IA = IA0 ⊕ IA′ , and it follows from weight
considerations that A′ ⊂ IA′ . As A′ is a coherent OX -module (since U → X is
finite), Nakayama’s Lemma implies that A′ = 0.
This implies that the action of T on U is also trivial. Therefore, the µr -action on
U is trivial and we have an isomorphism X( µr ∼ = [U/Γ]. Using the isomorphism
Γ∼ = G × Gnm from Lemma 4.1, we have further identifications
X( µr ∼= [U/Γ] ∼ = [U/(G × Gn )] ∼
m = [U/G] × BGn ,m
where in the final equivalence we have used that the Gnm -action on U must be trivial
(as the T -action is trivial).
For each geometric point x ∈ X(L), restricting the Gnm -torsor X → X to residual
gerbes at x induces a Gnm -torsor BGx → BΓx , where Gx = StabX (x) ⊂ G and
Γx = StabX (x). As Γ0x = Γ0 , we have the identification µr = Gx ∩ Γ0x . By
Proposition 3.4, the factorization
[U/Γ] → [U/Γ] = [U/G] × BGnm → U/G × BGnm
satisfies the universal property, yielding part (1). As in the case of the classifying
stack, parts (2) and (3) follow from the exact sequence (4.4) and the identification
BΓ( T ∼ = BG.
STRUCTURE RESULTS FOR TORUS FIXED LOCI 13
Corollary 4.5. Theorem 1.1 holds for X = [U/Γ] where U is a connected, reduced,
and separated algebraic space U of finite type over k.
Proof. By Remark 3.3, there is a reparameterization T ′ → T of tori such that the
T ′ -action on X = [U/G] is trivial. Therefore, the result follows from Proposition 4.4.
Alternatively, we can argue similarly to Proposition 4.4. By the hypotheses of
Theorem 1.1, X = [U/G] → BG is a Gnm -equivariant morphism such that the in-
duced Gnm -action on the quotient algebraic space U/G = [U/G]cms is trivial. The
map U → U/G is quasi-finite and equivariant under T → Gnm . Since U is re-
duced, the T -action on U must also be trivial. From here, the proof proceeds as in
Proposition 4.4.
[Spec B/Gy ] // BGy
is commutative.
Let p, q : [Spec A/Gx ] → BGy denote the two composition morphisms, and let P
and Q denote the corresponding Gy -torsors over [Spec A/Gx ]. Since (5.1) and (5.2)
∼
induce isomorphism of stabilizers at y ′ and x′ , there is a 2-isomorphism p|BGx′ →
q|BGx′ of the restrictions of p and q along the inclusion BGx′ → [Spec A/Gx ] of
the residual gerbe of x′ , and therefore an isomorphism α0 : P|BGx′ → Q|BGx′ of
Gy -torsors. The stack I := Isom[Spec A/Gx ] (P, Q) parameterizing isomorphisms of
Gy -torsors is smooth and affine over [Spec A/Gx ]. The isomorphism α0 defines
a section of I → [Spec A/Gx ] over BGx . By [AHR21, Prop. 7.18], there exists
an étale neighborhood Spec R → Spec AGx such that α0 extends to a section of
I → [Spec A/Gx ] over Spec R ×Spec AGx [Spec A/Gx ]. Therefore, after replacing A
with R ⊗AGx A, there is a 2-isomorphism of p and q.
The addendums follow from [AHR21, Prop. 5.7].
There is a stronger version when X → Y is a smooth morphism of smooth
noetherian algebraic stacks. For a point x ∈ X (k), we let Nx denote the normal
space to x, viewed as a Gx -representation. Explicitly, as x ∈ |X | is locally closed,
14 J. ALPER AND F. JANDA
g f
[Ny /Gy ] oo [Spec B/Gy ] // Y
Remark 5.4. The following are special cases of the above corollary:
(1) If the Gm -action on X is trivial, i.e. X = X × BGm , then this follows from the
Luna étale slice theorem for algebraic stacks [AHR20, Thm. 1.1].
(2) If X is an algebraic space, then this is Sumihiro’s theorem for torus actions
generalized to algebraic spaces [AHR20, Thm. 4.1].
5.3. Proof of the general case of Theorem 1.1.
Proof of Theorem 1.1. For every x ∈ X , we use Corollary 5.3 to obtain an étale
morphism
(5.3) U = [Spec A/Γx ] = [U/Gnm ] → [X/Gnm] = X
over BGnm and a preimage u ∈ U(k) of x, where U = [Spec A/Gx ]. By Corollary 4.5,
the theorem holds for U. This gives a unique tuple r = (r1 , . . . , rn ) and a central
closed subgroup µr,U ⊂ IU /BGnm . Moreover, the construction of the subgroup µr,U
is canonical: writing Γx as an extension 1 → Gx → Γx → Gnm → 1, then µr =
Gx ∩ Γ0x ⊂ Γx acts trivially on Spec A and this defines µr,U . As X is connected, the
tuple r = (r1 , . . . , rn ) is the same for every étale neighborhood. Moreover, covering
X with étale neighborhoods as in (5.3), we have canonical isomorphisms of the
subgroups of the inertia over intersections, and this provides descent data for the
construction of a central closed subgroup µr,X ⊂ IX /BGnm .
The rigidification X ( µr,X is the quotient stack of X( µr,X by Gnm . An étale
neighborhood (5.3) induces étale neighborhoods U( µr,U → X( µr,X and U( µr,U →
X( µr,X . Since Gnm acts trivially on U( µr,U , it acts trivially on Y := X( µr,X .
This gives a factorization
X → X( µr,X ∼ = Y × BGn → Xcms × BGn ,
m m
which satisfies the universal property by Proposition 3.4, yielding part (1). The
factorizations in parts (2) and (3) also follow from étale descent.
References
[ACGS20] Dan Abramovich, Qile Chen, Mark Gross, and Bernd Siebert, Punctured logarithmic
maps, 2020.
[AHR20] Jarod Alper, Jack Hall, and David Rydh, A Luna étale slice theorem for algebraic
stacks, Ann. of Math. (2) 191 (2020), no. 3, 675–738. MR 4088350
[AHR21] Jarod Alper, Jack Hall, and David Rydh, The étale local structure of algebraic stacks,
2021.
[AKL+ ] Dhyan Aranha, Adeel Khan, Alexei Latyntsev, Hyeonjun Park, and Charanya Ravi,
Localization theorems for algebraic stacks.
[Bor91] Armand Borel, Linear algebraic groups, second ed., Graduate Texts in Mathematics,
vol. 126, Springer-Verlag, New York, 1991. MR 1102012
[CGL21] Huai-Liang Chang, Shuai Guo, and Jun Li, Polynomial structure of Gromov-Witten
potential of quintic 3-folds, Ann. of Math. (2) 194 (2021), no. 3, 585–645. MR 4334973
[CJR21] Qile Chen, Felix Janda, and Yongbin Ruan, The logarithmic gauged linear sigma model,
Invent. Math. 225 (2021), no. 3, 1077–1154. MR 4296354
[CJR22] , Punctured logarithmic R-maps, 2022.
[dCHN23] Mark Andrea de Cataldo, Andres Fernandez Herrero, and Andrés Ibáñez Núñez, Rel-
ative étale slices and cohomology of moduli spaces, 2023.
[FJR18] Huijun Fan, Tyler J. Jarvis, and Yongbin Ruan, A mathematical theory of the gauged
linear sigma model, Geom. Topol. 22 (2018), no. 1, 235–303. MR 3720344
[Gir71] Jean Giraud, Cohomologie non abélienne, Springer-Verlag, Berlin-New York, 1971, Die
Grundlehren der mathematischen Wissenschaften, Band 179. MR 0344253
[GJR18] Shuai Guo, Felix Janda, and Yongbin Ruan, Structure of higher genus Gromov-Witten
invariants of quintic 3-folds, 2018.
[GP99] Tom Graber and Rahul Pandharipande, Localization of virtual classes, Invent. Math.
135 (1999), no. 2, 487–518. MR 1666787 (2000h:14005)
[Kre99] Andrew Kresch, Cycle groups for Artin stacks, Invent. Math. 138 (1999), no. 3, 495–
536. MR 1719823
16 J. ALPER AND F. JANDA
[PV16] Alexander Polishchuk and Arkady Vaintrob, Matrix factorizations and cohomological
field theories, J. Reine Angew. Math. 714 (2016), 1–122. MR 3491884
[Rom05] Matthieu Romagny, Group actions on stacks and applications, Michigan Math. J. 53
(2005), no. 1, 209–236. MR 2125542
[Wit93] Edward Witten, Phases of N = 2 theories in two dimensions, Nuclear Phys. B 403
(1993), no. 1-2, 159–222. MR 1232617 (95a:81261)