• Aucun résultat trouvé

Shifted cotangent stacks are shifted symplectic

N/A
N/A
Protected

Academic year: 2022

Partager "Shifted cotangent stacks are shifted symplectic"

Copied!
25
0
0

Texte intégral

(1)

ANNALES

DE LA FACULTÉ DES SCIENCES

Mathématiques

DAMIENCALAQUE

Shifted cotangent stacks are shifted symplectic

Tome XXVIII, no1 (2019), p. 67-90.

<http://afst.cedram.org/item?id=AFST_2019_6_28_1_67_0>

© Université Paul Sabatier, Toulouse, 2019, tous droits réservés.

L’accès aux articles de la revue « Annales de la faculté des sci- ences de Toulouse Mathématiques » (http://afst.cedram.org/), implique l’accord avec les conditions générales d’utilisation (http://afst.cedram.

org/legal/). Toute reproduction en tout ou partie de cet article sous quelque forme que ce soit pour tout usage autre que l’utilisation à fin strictement personnelle du copiste est constitutive d’une infraction pénale. Toute copie ou impression de ce fichier doit contenir la présente mention de copyright.

cedram

Article mis en ligne dans le cadre du

Centre de diffusion des revues académiques de mathématiques

(2)

pp. 67-90

Shifted cotangent stacks are shifted symplectic

(∗)

Damien Calaque(1)

ABSTRACT. — We prove that shifted cotangent stacks carry a canonical shifted symplectic structure. We also prove that shifted conormal stacks carry a canonical Lagrangian structure. These results were believed to be true, but no written proof was available in the Artin case.

RÉSUMÉ. — On démontre que les champs cotangents décalés sont canonique- ment munis d’une structure symplectique décalée. On démontre également que les champs conormaux décalés sont munis d’une structure Lagrangienne canonique. Ces résultats étaient attendus mais aucune démonstration n’était disponible dans le cas des champs d’Artin.

Introduction

Shifted symplectic and Lagrangian structures have been introduced in the inspiring paper [7], where many examples of shifted symplectic stacks are given. They appeared to be very powerful tools, allowing for instance:

• to prove existence of symplectic structures on various moduli spaces ([7]).

• to prove existence and functoriality of perfect symmetric obstruction theories ([7]).

• to construct “classical” topologial field theories ([2, 5]).

Many results of classical symplectic geometry do extend to the shifted/

derived context, sometimes even in a better way thanks to the flexibility of derived geometry. For instance, Lagrangian correspondences compose well, without any transversality assumption. But some obvious results happen to be much more difficult in the derived context, precisely because of the

(*)Reçu le 11 janvier 2017, accepté le 20 février 2017.

(1) IMAG, Univ Montpellier, CNRS, Institut Universitaire de France, Montpellier, France — damien.calaque@umontpellier.fr

Article proposé par Bertrand Toën.

(3)

flexibility of derived geometry. The main example of this phenomenon is the classical (and easy) correspondence between symplectic and non-degenerate Poisson structures; in the shifted/derived setting this correspondence is much harder to prove, even in the affine case (see [3, 8]).

Another classical and rather easy fact in symplectic geometry is that the cotangentTY to a manifoldY admits a canonical symplectic structure and that the conormal NXTY of a submanifold XY is Lagrangian.

An analog of this was expected to be true for shifted cotangent (and conor- mal) stacks of derived Artin stacks, but no written proof was available (the Deligne-Mumford case is treated in [7]).

We provide a proof in this paper. The spirit of the proof is essentially the same as the one in usual symplectic geometry. The challenge is to formulate it in a completely coordinate independent fashion.

We also conclude the paper with several conjectures about the relation between Lagrangian morphisms, (deformed) shifted cotangent stacks, and shifted Poisson structures in the sense of [3, 8].

Basic notation and conventions

kis a field of characteristic zero (or just a NoetherianQ-algebra).

• all algebro-geometric structures are overk. For instance, a derived stack is a derivedk-stack.

• acdgameans a commutative differential gradedk-algebra sitting in non-positive degree. For commutative differential graded algebras without any degree condition we will sayunbounded cdga.

• an Artin stack means a derived geometric k-stack locally of finite presentation in the sense of [11].

• as we work with∞-categories, unless otherwise specified all usual categorical terms must be understood∞-categorically (for instance,

“limit” means “∞-limit” or “homotopy limit”).

Acknowledgments

I thank Pavel Safronov for several stimulating discussions about shifted cotangent stacks. I actually first came up with a different (and quite indirect) strategy for the proof, that uses shifted symplectic groupoids (it will appear in a forthcoming work), and it was after a discussion with Pavel during the conference “Homotopical Methods in Quantum Field Theory” at IBS-CGP in Pohang that I started to look for a simpler and more direct proof.

(4)

I acknowledge the support of the Institut Universitaire de France and ANR SAT.

1. Basics of derived (pre)symplectic geometry

1.1. Some∞-categorical linear algebra

LetC be a stable symmetric monoidal ∞-category. Let V be a perfect (meaning dualizable) object inC.

Definition 1.1. — a) A degree n pairing ω :V⊗21[n], where 1is a unit, isskew-symmetricif it factors through V⊗2→ ∧2V1[n], where2V := S2(V[−1])[2]. From now on, unless otherwise specified, all pairings will be skew-symmetric.

b) A degree npairing ω:V⊗21[n]is non-degenerateif the adjoint morphismω[:VV[n] is an equivalence.

Remark 1.2. —Recall that there are a priori two ways to define the adjoint morphism ofω:

• asω[:= (idVcoevV)◦(ω⊗idV), wherecoevV:1VV is the coevaluation map.

• or as[ω:= (coevVidV)◦(idVω).

Sinceωis skew-symmetric then these two definitions coincide. Now observe that the (shifted) dual ofω[is (coevV[n]idV)◦(idV[n]ω[[−n]⊗idV)◦ (idV[n]evV), where evV : VV1 is the evaluation map. Using that (coevVidV)◦(idVevV) is homotopic to idV in M apC(V,V) we get that the (shifted) dual of ω[ is homotopic to[ω (and thus toω[ itself) inM apC(V,V[n]).

Letf :UVbe a morphism between perfect objects in C.

Definition 1.3. — Let ω be a degree npairing onV.

a) An isotropic structureonf (w.r.t.ω) is a homotopy between0and fω in the spaceM apC2U,1[n]

of degreen pairings onU.

b) An isotropic structure on f is non-degenerate if the induced null- homotopic sequenceUVU[n] is a fiber sequence.

The following result is very easy, though quite useful.

(5)

Lemma 1.4. — Assume that we are given a morphism f : UV between perfect objects together with an n-shifted pairing ω on V and an isotropic structureγ onf. If γ is non-degenerate thenω is non-degenerate as well.

Proof. — By definition, the null-homotopic sequence UVU[n]

is a fiber sequence. Its (shifted) dual UV[n] → U[n] is thus a fiber sequence as well. The main point is that, thanks to Remark 1.2, these two sequences are mapped to each other in the following way:

U //V

ω[

//U[n]

U //V[n] //U[n]

Hence the morphism ω[ : VV[n] is an equivalence and ω is non-

degenerate.

1.2. Shifted (pre)symplectic, isotropic and Lagrangian structures

Recall from [7] that to any derived stack X one can associate a graded mixed (k-)moduleDR(X). We will often denoteddR the mixed differential (of weight 1 and cohomological degree 1)(1), anddintordXthe differential of the underlying graded moduleDR(X)] (of weight 0 and cohomological degree 1), also called theinternal differential.

IfX = Spec(A), for a cofibrant cdgaA, thenDR(X) =DR(A/k) is the graded module

SymA1A[−1](−1) . equipped with the following mixed differential:

=ddR:a0(da1). . .(dan)7→(da0)(da1). . .(dan).

Note that (dai) is a priori not “d applied to ai and is in particular not dintai =dAai. But observe that (dai) =ddRai; hence we will try to use the notationddRai in order to avoid confusion.

Notation 1.5. —We write−dggrk , resp.dggrk , for the category of graded mixed modulesk-modules, resp. gradedk-modules. We refer to [3, Section 1]

for details about the homotopy theory of graded mixed modules.

Definition 1.6. — a) The space of p-forms of degreenonX is Ap(X, n) :=M apdggr

k k,DR(X)][n+p](p) .

(1)We use the weight/grading convention from [3] rather than the one of [7].

(6)

b) The space of closedp-forms of degreenonX is Ap,cl(X, n) :=M ap−dggr

k k,DR(X)[n+p](p) .

c) For a closed p-form α of degree n, we denote byα0 its underlying p-form of degree n, that is to say its image through the forgetful functor(−)].

The functorsAp(n) :=Ap(−, n) and Ap,cl(n) :=Ap,cl(−, n) are actually derived stacks (see [7, Proposition 1.11]).

Remark 1.7. —Note that for a given graded mixed moduleM, M ap−dggr

k (k, M) is the space of a semi-infinite sequences α= (α0, . . . , αi,· · ·)

such thatd(α0) = 0, and for everyi>0,i) =d(αi+1). I.e. it is the space of 0-cocycles in Q

k>0M(p) equipped with the total differential d+. The image of such a sequenceαthrough the forgetful functor (−)]is precisely its leading termα0, which is a 0-cocyle inM(0). We will thus sometimes write, in the context of the above definition, “the leading term” in place of “the underlyingp-form of degreen” of a closedp-form of degreen.

Definition 1.8. — a) An n-shifted presymplectic structure on a stackX is a closed 2-form of degree non X.

b) Let(X, ω)be a stack equipped with ann-shifted presymplectic struc- ture. An isotropic structure on a morphism f : LX is a path from 0 to fω in the space A2,cl(L, n) of n-shifted presymplectic structures onL.

c) We say that ann-shifted presymplectic structureω on anX issym- plecticifX has a perfect cotangent complexLX and if the degreen pairing onTX :=LX given by its leading termω0 is non-degenerate in the sense of Definition 1.1.

d) We say that an isotropic structureγonf :LX isnon-degenerate if both X and L have perfect cotangent complexes and if the lead- ing term γ0 is indeed a non-degenerate isotropic structure on the morphism TLfTX in the sense of Definition 1.3. It is only when the presymplectic structureωitself is symplectic that we name Lagrangian structuresthe non-degenerate isotropic ones.

The main result of Section 2 below is that then-shifted cotangent stack T[n]X of an Artin stack X carries an n-shifted symplectic structure and that the zero section morphismXT[n]Xcarries a Lagrangian structure.

We refer to [7, 2, 9] for many other examples ofn-shifted symplectic and Lagrangian structures.

(7)

Remark 1.9. — a) A stack equipped ann-shifted presymplectic struc- ture (an n-shifted presymplectic stack, for short) is nothing but a stack overA2,cl(n).

b) Let (X, ω) be a stack equipped with an n-shifted presymplectic structure. An isotropic structure on a morphism f : LX is nothing but the data of a commuting square

L f //

X

ω

0 //A2,cl(n)

The above alternative approach to isotropic structures on morphisms (due to Rune Haugseng [5, Section 9]) leads to the following:

Definition 1.10. — Ann-isotropic correspondence is a correspondence in the∞-categorydSt/A2,cl(n) of derived stacks overA2,cl(n): it is the data of a commuting square

L g //

f

Y

ωY

X ω

X

//A2,cl(n)

In other words, ann-isotropic correspondence is the data of twon-shifted presymplectic stacks (X, ωX) and (Y, ωY) together with an isotropic mor- phismf×g:LX×Y, whereX×Y is then-shifted presymplectic stack (X×Y, π2ωYπ1ωX). AssumingωX andωY are symplectic, thenf×g is Lagrangian if and only if the commuting square

TL //

gTY 'gLY[n]

gTX'fLX[n] //LL[n]

is (co)Cartesian. In this case we talk about ann-Lagrangian correspondence.

1.3. Isotropic and Lagrangian fibrations

Recall from [3, Section 1] that to any morphismf : XY of derived stacks one can associate a graded mixed module DR(f) := DR(X/Y). If

(8)

f :X = Spec(A)→Spec(B) =Y comes from a cofibrationBAbetween cofibrant cdgas, thenDR(X/Y) :=DR(A/B) is the graded module

SymA1A/B[−1](−1) . equipped with the mixed differential:

:a0(da1). . .(dan)7→(da0)(da1). . .(dan).

Definition 1.11. — a) The space of relativep-forms of degreen onf (a-k-a p-forms of degreenonX relative to Y) is

Ap(f, n) =Ap(X/Y, n) :=M apdggr

k k,DR(X/Y)][n+p](p) . b) The space of relative closed p-forms of degreenonf is Ap,cl(f, n) =Ap,cl(X/Y, n) :=M ap−dggr

k k,DR(X/Y)[n+p](p) . c) For a relative closedp-formαof degree n, we denote by α0 itsun-

derlying relativep-form of degreen, that is to say its image through the forgetful functor(−)].

Note that we have a null-homotopic sequence DR(Y) f

−→DR(X)−→/Y DR(X/Y).

Remark 1.12. — Recall from [3, Remark 2.4.8] that DR(X/Y) can be obtained as global sections of a stackDR(f) =DR(X/Y) of graded mixed modules over the∞-site of derived affine schemes overY:

DR(X/Y) Spec(A)−→y Y

:=DR yX/Spec(A) .

Observe that DR(X/Y) is actually a graded mixed OY-module as every DR yX/Spec(A)

is a graded mixedA-module.

Let nowX be equipped with ann-shifted presymplectic structureω.

Definition 1.13. — a) An isotropic fibration structure onf is a path from0 toω/Y in the space A2,cl(X/Y, n) of relative closed 2- forms of degreen.

b) An isotropic fibration structure γ on f is non-degenerate (or is a Lagrangian fibration structure) if bothX andY have perfect cotan- gent complexes and if the leading termγ0is indeed a non-degenerate isotropic structure on the morphism TX/Y → TX in the sense of Definition 1.3.

Remark 1.14. — It follows from Lemma 1.4 that ifγ is non-degenerate then then-shifted presymplectic structureω is indeed n-shifted symplectic.

Therefore we don’t need here to make a distinction between non-degenerate isotropic fibration structures and Lagrangian fibration structures.

(9)

We will see in Section 2 that the projection T[n]X → X of a shifted cotangent stack (of an Artin stack) comes equipped with a Lagrangian fi- bration structure.

2. The n-shifted symplectic structure of the n-shifted cotangent stack

LetX be an Artin stack. It has a perfect global cotangent complex LX

(see [11]), its dual TX :=LX being called the tangent complex. Our main object of interest is itsn-shifted cotangent stack

T[n]X :=RSpecX Sym(TX[−n]) ,

that is nothing but the stack of n-shifted 1-forms on X. Note that, given a quasi-coherent module E over X, the stack of sections E := AX(E) is defined, as a stack overX, by

E Spec(A)→x X

:=M apA−mod(A, xE). WheneverE is perfect, we get that this is equivalent to

M apA−mod(xE, A) ∼= M apA−alg SymA(xE), A

=: RSpecX Sym(E)

Spec(A)→x X , and one can show thatRSpecX Sym(E)

is an Artin stack as well (it is an affine stack of finite presentation overX; see [10] for what we mean by an affine stack).

In this Section we prove that the n-shifted cotangent stack T[n]X is n-shifted symplectic, together with variations on this result. All this was mainly already known for Deligne-Mumford stacks.

2.1. Ann-shifted presymplectic structure on T[n]X

Recall that the space of morphisms YT[n]X is nothing but the space of morphisms f : YX equipped with a section of fLX[n]. In particular the identity mapT[n]X → T[n]X corresponds to the data of the projection mapπX :T[n]X →X together with a section ofπXLX[n].

Therefore, using the map πXLX[n] → LT[n]X[n] we obtain that T[n]X carries a tautologicaln-shifted 1-formλX. The morphismddR:A1(−, n)→ A2,cl(−, n) sends it to ann-shifted closed 2-formωX=ddRλX onT[n]X.

(10)

The zero section morphismιX:XT[n]X corresponds to the data of the identity mapXX together with the zero section ofLX[n]. Tautologi- cally, this means that the pull-back ofλXalong the zero section is homotopic to zero. Hence we obtain an isotropic structureγXonιX(w.r.t. then-shifted presymplectic structure previously described).

Finally observe thatλX comes from a section ofπXLX[n] and recall that the sequence

πXLX →LT[n]X →LπX

is fibered. Therefore the image of λX into relative n-shifted 1-forms along πX is homotopic to zero. Hence we obtain that the projection morphismπX

is equipped with an isotropic fibration structureηX.

Remark 2.1. —Note thatLπX is nothing butπXTX[−n].

2.1.1. Compatibility with the Gm-action

Observe that, for a perfect moduleE overX, its stack of sections E is acted on by Gm because Sym(E) is a graded OX-algebra (by symmetric powers). We will refer to this new grading as thefiber grading(not to confuse it with the weight of forms and closed forms). Both the zero sectionXE and the projectionEX areGm-equivariant for the trivialGm-action on X.

Notation 2.2. — a) IfY is an Artin stack equipped with an action ofGm, then the relative cotangent complexL[Y /Gm]/BGm is a quasi- coherent sheaf on [Y /Gm]. Its pull-back along the quotient mapY → [Y /Gm] identifies withLY. We will therefore allow ourselves to abuse notation and to keep writingLY for itsGm-equivariant enhancement L[Y /Gm]/BGm. Similarly, if f : Y1Y2 is a Gm-equivariant map, then we will keep writing Lf for its Gm-equivariant enhancement L[f], where [f] : [Y1/Gm]→[Y2/Gm] is the induced map on quotient stacks.

b) Sheaves on BGm = [∗/Gm] are identified with graded complexes.

For a graded complexM = (Mi)i∈Zwe denote byM{k}itsk-th shift for this grading (meaning that M{k}i = Mi+k). Without further precision, a genuine complex is always assumed to be equipped with the trivialGm-action. This notation extends as well to sheaves on [X/Gm] =X×BGm, that are nothing but graded sheaves onX.

Remark 2.3. — With this notation the identification from Remark 2.1 becomesLπX ∼=πXTX[−n]{−1}.

(11)

Now, ifY is a derived stack equipped with an action ofGm, observe that aGm-equivariant mapYEis determined by the data of aGm-equivariant mapf :YX together with aGm-equivariant section offE{1}. Apply- ing this to the identity map T[n]X → T[n]X we get a Gm-equivariant section of πXLX{1}. As a consequence, the 1-form λXA1 T[n]X, n∼= Γ T[n]X,LT[n]X[n]

of (cohomological) degreenis homogeneous of weight 1 for the fiber grading.

Recall from Remark 1.12 that for a stackY equipped with aGm-action we have a sheafDR [Y /Gm]/BGm

of graded mixedO-modules on BGm. Its pull-back along the canonical point inBGmisDR(Y). This means that DR(Y) carries yet another auxiliary grading that enhances it to a graded mixed gradedk-module. We will follow the (abusing) Notation 2.2 and keep writingDR(Y) for its graded enhancement DR [Y /Gm]/BGm

. Similarly, if f : Y1Y2 is a Gm-equivariant map, then DR(f) also admits Gm- equivariant enhancement that we still denote the same.

Going back toT[n]X we get thatλX belongs to the space A1,{1} T[n]X, n

:=M ap(dggr

k)gr k,DR(T[n]X)][n+ 1](1){1}

and thatωX =ddRλX belongs to the space A2,cl,{1} T[n]X, n

:=M ap(−dggr

k)gr k,DR(T[n]X)[n+ 2](2){1}

of closed 2-forms of (cohomological) degreenon T[n]X that are homoge- neous of weight 1 for the fiber grading. Similarly,ηX is a path in the space

A2,cl,{1} πX, n

:=M ap(−dggr

k )gr k,DR(πX)[n+ 2](2){1}

of relative closed 2-forms of (cohomological) degree n along πX that are homogeneous of weight 1 for the fiber grading.

2.2. Non-degeneracy of then-shifted presymplectic structure

We will now prove thatωX actually defines ann-shifted symplectic struc- ture onT[n]X, that is to say the morphism

TT[n]X −→LT[n]X[n]

induced by its underlyingn-shifted 2-form is an equivalence.

We also prove that the isotropic morphismXT[n]Xand the isotropic fibrationT[n]X →X are Lagrangian.

(12)

Theorem 2.4. — Then-shifted cotangent stackT[n]Xisn-shifted sym- plectic, its zero section is Lagrangian, and the projection down to X is a Lagrangian fibration. More precisely:

(1) Then-shifted presymplectic structureωXonT[n]Xis non-degenerate.

(2) The n-shifted isotropic structure γX on the zero section ιX :XT[n]X is non-degenerate.

(3) Then-shifted isotropic fibration structureηX onπX:T[n]→X is non-degenerate.

Before we start it is important to have in mind the two following fiber sequences, that are (n-shifted) dual to each other:

πXLX[n]∼=TπX → TT[n]XπXTX

πXLX[n] → LT[n]X[n] → LπX[n]∼=πXTX.

Moreover, when pulled-back along the zero section these sequences split and we get that

ιXTT[n]X∼=LX[n]⊕TX and ιXLT[n]X[n]∼=LX[n]⊕TX. Remark 2.5. — The above fiber sequences actually are Gm-equivariant.

More precisely, according to Remark 2.3 we thus get fiber sequences πXLX[n]{1} → TT[n]XπXTX

πXLX[n] → LT[n]X[n] → LπX[n]∼=πXTX{−1}

ofGm-equivariant sheaves onT[n]X that split when pulled-back along the zero section:

ιXTT[n]X ∼=LX[n]{1} ⊕TX and ιXLT[n]X[n]∼=LX[n]⊕TX{−1}.

2.2.1. The isotropic structure on the zero sectionιX is non-degenerate

We prove here the second part of Theorem 2.4.

We have an isotropic structure γX on the zero section morphism ιX : XT[n]X, inducing in particular a commuting diagram

TX

//TX

//0

ιXTT[n]X //ιXLT[n]X[n] //LX[n]

in which the right-hand square is Cartesian (this square is the equivalence ιXLT[n]X[n]∼=LX[n]⊕TX). The arrowTX →TX in the left-most square

(13)

is determined by the leading term ofγX (which is a homotopy between the composed mapTX →LX(n] and the zero map).

We will now show that

Lemma 2.6. — The morphism φX : TX → TX appearing in the above diagram is natural in X, in the sense that for every morphism f :XY we get a commuting square

TX

φX //TX

fTY

fφY //fTY

Corollary 2.7. — The isotropic structure on the zero sectionιY :YT[n]Y is non-degenerate, that is to say the mapφY is an equivalence.

Proof of the Corollary. — We know that it is true for derived affine schemes locally of finite presentation. Let us then assume by induction that we know the result holds for m-Artin stacks (i.e. Artin stacks that are m- geometric in the sense of [11]). Let us then prove that it is true as well for (m+ 1)-Artin stacks. IfY is an (m+ 1)-Artin stack thenY =|G|:=

colimn(Gn), where G = (Gn)n>0 is a smooth Segal groupoid inm-Artin stacks. Letq: X =G0Y be the natural map to the colimit. Note that qTY can be computed as the colimit of the simplicial diagram (enTGn)n>0, whereen:G0Gn is the unit map(2).

Thanks to the above Lemma we have a simplicial diagram (enφGn)n>0of morphisms of perfect modules onX, the colimit of which is the morphism qφY. Thanks to the induction hypothesis every enφGn is an equivalence, and thus qφY is an equivalence as well. One finally gets that φY is an equivalence becauseqis conservative (qis a groupoid quotient morphism).

We get the result by induction on the degree of geometricity.

Remark 2.8. —One can even prove that the equivalenceφXis homotopic to the identity (this is true for derived schemes, and then the inductive proof is exactly the same).

(2)Recall thatQCoh(|G|)=limn QCoh(Gn)

and thatq=limn(en). Under the first identification we have thatL|G|is the diagram (LGn)n>0and thus, using the second identification,qL|G| is equivalent tolimn(enLGn). Dualizing, we get thatqT|G| is equivalent tocolimn(enTGn).

(14)

Proof of the Lemma. — Letf :XY be a map of Artin stacks and consider the diagram

X

xx

f //

Y

ιY

T[n]X oo fT[n]Y //T[n]Y

wherefT[n]Y :=T[n]Y ×Y X is the stack of sections offLY[n].

The morphismφX, resp. φY, is determined by the leading term of the isotropic structureγX, resp. γY. Hence the morphism fφY is determined by the leading term offγY, considered as an isotropic structure on

fιY :X −→fT[n]Y

and where we equipfT[n]Y with then-shifted presymplectic structure ˜ω given by the pull-back of ωY along fT[n]Y → T[n]Y. Now the result follows from the following fact:

• the pull-back of the pair (ωX, γX) along fT[n]Y → T[n]X is naturally homotopic to the pair (˜ω, fγY).

Actually, the pull-back of λX itself along fT[n]Y → T[n]X is natu- rally homotopic to the pull-back of λY along fT[n]Y → T[n]Y(3). All diagrams we have written down live under X, therefore all previously con- sidered homotopies between 0 and pull-backs of λ’s on X are themselves

equivalent. The Lemma is proved.

2.2.2. The isotropic fibration structure on the projection πX is non-degenerate

We now give the proof of the third part of the theorem. Its structure is very similar to the one of the second part (apart from that we will have to useGm-equivariant enhancements at some point).

We have an isotropic fibration structureηXon the projectionπX:T[n]X→ X, inducing in particular a commuting diagram

πXLX[n]

//πXLX[n]

//0

TT[n]X //LT[n]X[n] //πXTX

(3)This is a slight variation on the obvious fact that for any morphismu:AB, idBuanduidAare equivalent.

(15)

in which the right hand square is Cartesian.

Let us write ψX : πXLXπXLX for the (−n)-shift of the morphism πXLX[n]→πXLX[n] appearing in the above diagram. Our aim is to prove thatψXis an equivalence. It follows from the next Lemma that it is sufficient to prove thatιXψX :LX→LX is an equivalence.

Lemma 2.9. — ψX =πXιXψX.

Proof. — Recall that, up to a shift, the morphismψX is the composition of the mapTπXπXLX[n] given by the (leading term of the) isotropic fi- bration structureηX with the canonical identificationπLX[n]∼=TπX (dual to the one from Remark 2.1). Taking into accountGm-equivariant enhance- ments, it becomes the composition of:

• the map TπXπXLX[n]{1} given by (the leading term of the) pathηthat is homogeneous of weight 1 for the fiber grading,

• with the identification πLX[n]{1} ∼= TπX (dual to the one from Remark 2.3).

HenceψX is aGm-equivariant mapπXLX{1} →πXLX{1}. SinceLX{1}is a graded sheaf of pure weight onX, thenψX is the pull-back alongπX of a

morphismLX →LX (that has to be ιXψX).

Let nowf :XY be a map of Artin stacks and consider the diagram T[n]X

πX

&&

fT[n]Y

Tf

oo

fπY

πYf //T[n]Y

πY

X f //Y We will show that

Lemma 2.10. — The morphism ψX :πXLXπXLX is natural in X, in the sense that for every morphismf :XY we get a commuting square

(fπY)LX

(Tf)ψX//(fπY)LX

(fπY)fLY

OO

Yf)ψY

//(fπY)fLY

OO

In particular we have a commuting square LX

ιXψX

//LX

fLY

OO

fιYψY

//fLY

OO

(16)

Corollary 2.11. — The isotropic fibration structure on the projection πX : T[n]X → X is non-degenerate, that is to say the map ψX is an equivalence.

Proof of the Corollary. — It follows from Lemma 2.9 that it is sufficient to show thatιXψX is an equivalence, the proof of which is almost identical to the one of Corollary 2.7. Let us repeat it for the sake of clarity.

We know that it is true for derived affine schemes locally of finite pre- sentation. Let us then assume by induction that we know the result holds form-Artin stacks and prove that it does for (m+ 1)-Artin stacks as well.

IfY is an (m+ 1)-Artin stack thenY =|G|, whereG is a smooth Segal groupoid in m-Artin stacks. Let q : X = G0Y be the natural map to the colimit. Note thatqLY can be computed as the limit of the cosimplicial diagram (enLGn)n>0, where en:G0Gn is the unit map.

Thanks to the above Lemma we have a cosimplicial diagram

(enιGnψGn)n>0of morphisms of perfect modules onX, the limit of which is the morphismqιYψY. Thanks to the induction hypothesis everyenιGnψGn

is an equivalence, and thus qιYψY is an equivalence as well. One finally gets thatιYψY is an equivalence becauseq is conservative (qis a groupoid quotient morphism).

We get the result by induction on the degree of geometricity.

Remark 2.12. — Just as in Remark 2.8 one can even prove that the equiv- alenceψX is homotopic to the identity.

Proof of the Lemma. — The morphismψX, resp.ψY, is determined by the leading term of the isotropic fibration structure ηX, resp. ηY. Hence the morphism (πYf)ψY is determined by the leading term of (πYf)ηY, considered as an isotropic fibration structure on

fπY :fT[n]Y −→X

and where we equip fT[n]Y with the n-shifted presymplectic structureYf)ωY. Now the result follows from the following fact:

• the pull-back (Tf)ηX is naturally homotopic to (πYf)ηY. First of all this claim makes sense as we have already seen that the analogous statement for ω’s is true (see the proof of Lemma 2.6). It is even true for λ’s. Since the part of the above diagram we are interested in lives over X, all images of previously considered homotopies between 0 and images ofλ’s in the space of 1-forms relative toX are themselves equivalent. The Lemma

is proved.

(17)

End of the proof of Theorem 2.4. — The first part of the Theorem follows from the third one. Namely, Lemma 1.4 tells us that if we have a non- degenerate isotropic fibration on ann-shifted presymplectic Artin stack then

it isn-shifted symplectic.

2.3. A Lagrangian structure on shifted conormal stacks

Let f : XY be a morphism of Artin stacks. We have already seen that there is a tautologicaln-shifted 1-form onT[n]Y, mainly determined by the identity morphism ofT[n]Y (and the mapπYLY[n]→LT[n]Y). We now want to consider the morphism of Artin stacks`f :TX[n]Y →T[n]Y, whereTX[n]Y :=T[n−1]f =A(Lf[n−1]) is then-shifted conormal stack off. First of all observe that this morphism factors through fT[n]Y, so that`fωY is homotopic to the pull-back ofωX along the sequence

TX[n]Y →fT[n]Y →T[n]X .

Hence `fωY is equivalent to 0 as the composed morphism of the above se- quence factors through the zero sectionXT[n]X. Let us denoteωf the isotropic structure we have just constructed on`f.

Our aim is to prove the following:

Theorem 2.13. — The n-shifted conormal stack is Lagrangian, that is to say the isotropic structureωf is non-degenerate.

Let us provide two rather extreme examples:

• consider f : Xpt, where pt is the terminal stack. In this case TX[n+ 1]pt ∼= T[n]X gets a Lagrangian morphism to the point equipped with its canonical (n+ 1)-shifted symplectic structure. We get back in a rather involved way then-shifted symplectic structure on ann-shifted cotangent stack.

• consider the identity morphismXX. In this case TX[n]X ∼=X and we get back that the zero section is Lagrangian.

2.3.1. Isotropic squares

Our aim is to apply a strategy similar to what we did in the previous Section. To do so we need to exhibit a structure on the projectionTX[n]Y → X that resembles to the one of a Lagrangian fibration. This is what we do in this Subsection.

(18)

Definition 2.14. — An n-shifted pre-isotropic square is a commuting square of Artin stacks

E g //

F

X f //Y together with the following additional structures:

ann-shifted presymplectic structureω on F:ωA2,cl(F, n).

an isotropic structureγ onEF:γP athgω,0 A2,cl(E, n) .

an isotropic fibration structureη onFY: ηP athω/Y,0 A2,cl(F/Y, n)

.

Note that composingγ/X with (gη)−1 one gets an element ω0∈Ω0A2,cl(E/X, n)∼=A2,cl(E/X, n−1)

Example 2.15. — There is an n-shifted pre-isotropic square structure withF=T[n]Y,E=TX[n]Y,g=`f,ω=ωY,γ=ωf andη=ηY.

Definition 2.16. — Borrowing the above notation, ann-shifted isotropic square is an n-shifted pre-isotropic square equipped with a homotopy ξ in A2,cl(E/X, n)between the two isotropic fibration structuresγ/X andgη on EX.

We let the reader prove the following:

Proposition 2.17. — The n-shifted pre-isotropic square from Exam- ple 2.15 can be upgraded to ann-shifted isotropic square.

2.3.2. Non-degeneracy of the isotropic structure on the n-shifted conormal stack

Assume we are given an isotropic square. Borrowing the previous nota- tion, we set U :=TE, V :=gTF/Y, W:= gTF and Z:=TE/X. Recall that we have a lot of structure:

• a degreenpairing onW(given byω).

• a path from 0 to its pull-back alongUW(given byγ).

• a path from 0 to its pull-back alongVW(given byη).

• these two paths compose to a self-homotopy of 0 in the space of degreenpairings onU×WV, defining a degreen−1 pairingβ on it.

(19)

• a path from 0 to the pull-back of this last pairing alongZWV (given byξ).

Lemma 2.18. — If ξandη are non-degenerate, then so areω,β andγ.

Proof. — We already know from Lemma 1.4 that if ξ, resp. η, is non- degenerate then so is β, resp. ω. Now the data of ω, γ and η provides us with the following commuting diagram

U×WV

//V

//0

U

//W

//V[n]

0 //U[n] //(U×WV)[n]

We observe that the following squares are Cartesian: top-left (by definition), top-right (by non-degeneracy ofη), bottom-right (by non-degeneracy ofω), and the big one (by non-degeneracy ofβ). Therefore the bottom-left square is Cartesian as well, which means thatγis non-degenerate.

End of the proof of Theorem 2.13. — Note that, ifE=A(E),F=A(F) andg is linear in the fiber (i.e.g comes from a morphismEfF) then we have that the cofiber ofZU×WVis

cof ib πXE→TE×gTFgTF/Y

∼=πXTX/Y =πXTf.

This is typically what happens in the situation of Example 2.15. In this case the null-homotopic sequenceZU×WVZ[n−1] then takes the following form:

πXLf[n−1]−→U×WV−→πXTf. Hence we get a mapπXTf ∼=hocof ib ZU×WV

πXTf, which is an equivalence if and only ifξis non-degenerate. As before, one can show that this map must be an equivalence by reasoning along the following steps:

(1) usingGm-equivariant enhancements, one proves (as in the proof of Lemma 2.9) that the mapπXTfπXTf is the pull-back alongπX of a mapψf :Tf →Tf.

(20)

(2) one then proves (along the lines of Lemmas 2.6 and 2.10) thatψf is natural in the following sense: for any morphism commuting square

X0

x

f0 //Y0

y

X f //Y one has a commuting square

Tf0

ψf0

//Tf0

xTf

xψf //xTf

(3) the previous step allows one to reduce to the case when bothX and Y are affine, by induction on the degree of geometricity off (as in Corollaries 2.7 and 2.11).

Remark 2.19. — We say thatf ism-geometricif it can be real- ized as the colimitf =|f|of a simplicial diagram f:XY of morphisms, with bothXandYbeing Segal groupoids in (m−1)- geometric stacks. Every morphism between geometric stacks is m- geometric for some m. Namely, Y being geometric there exists a Segal groupoid Y such that Y = |Y| and every Yn is geomet- ric. Then there exists a Segal groupoidX presenting X such that Xn∼=X×YYnand a simplicial morphismfpresentingf such that fn is equivalent to the second projection (see [11, §1.3.5]). Observe that every Xn is geometric, as geometricity is preserved by fiber products.

(4) one finally proves that the result is true when both X and Y are affine.

We conclude by using Lemma 2.18: since ξ is non-degenerate then so is

γ=ωf.

Remark 2.20. —We won’t do it but one can also prove, using similar methods, that given two morphisms X1Y and X2Y there is an equivalence

TX1[n]Y ×T[n]Y TX2[n]Y ∼=T[n−1](X1×Y X2) of (n−1)-shifted symplectic stacks.

Example 2.21. —The example we provide here is just an illustration of our results, as in this case everything can be directly proven in an ad

Références

Documents relatifs

In Sec- tion 3 we give some results concerning bilateral versions of the change of base in Bailey pairs from [9], yielding m-versions of other multisum Rogers-Ramanujan

We extend this example in the general case of a monoid in an additive symmetric monoidal category without shifting (cf. Theorem 5.2.4) Let A be a Σ-double Poisson algebra in an

The second option uses affine invariants and requires extracting and tracking two points, but ensure higher performances since it allows convergence for huge motion (exceptional for

The theory of spherical functions, representations and orthogonal polynomials was first generalized to the case of free groups and more generally to groups acting isometrically

can write down simple, explicit (strict) k-shifted Poisson structures, and for our “Lagrangian Darboux form” local models for Lagrangians in k-shifted symplectic derived schemes

This prompts us to understand how the shifted cotangent bundle behaves under homo- topy equivalence. Our other main result, in the form of Theorem 3.1 and Corollary 3.6, states that

We introduce the notion of a shifted Poisson structure on a general derived Artin stack, study its relation to the shifted symplectic structures from [PTVV], and construct a

Gemignani , Efficient solution of parameter depen- dent quasiseparable systems and computation of meromorphic matrix functions, arXiv 1611.09107 [math.NA], 2016...