arXiv:1203.4869v4 [math.AG] 1 May 2013
Microlocal Euler classes and Hochschild homology
Masaki Kashiwara
∗and Pierre Schapira May 2, 2013
Abstract
We define the notion of a trace kernel on a manifold M. Roughly speaking, it is a sheaf onM×M for which the formalism of Hochschild homology applies. We associate a microlocal Euler class to such a ker- nel, a cohomology class with values in the relative dualizing complex of the cotangent bundleT∗M over M and we prove that this class is functorial with respect to the composition of kernels.
This generalizes, unifies and simplifies various results of (relative) index theorems for constructible sheaves,D-modules and elliptic pairs.
Contents
1 Introduction 2
2 A short review on sheaves 5
3 Compositions of kernels 7
4 Microlocal homology 9
5 Microlocal Euler classes of trace kernels 16
6 Operations on microlocal Euler classes I 24
∗This work was partially supported by Grant-in-Aid for Scientific Research (B) 22340005, Japan Society for the Promotion of Science.
7 Operations on microlocal Euler classes II 27 8 Applications: D-modules and elliptic pairs 34
9 The Lefschetz fixed point formula 36
1 Introduction
Our constructions mainly concern real manifolds, but in order to introduce the subject we first consider a complex manifold (X,OX). Denote byωXholthe dualizing complex in the category of OX-modules, that is, ωholX = ΩX[dX], wheredX is the complex dimension of X and ΩX is the sheaf of holomorphic forms of degree dX. Denote by O∆
X and ω∆holX the direct images of OX and ωXholrespectively by the diagonal embeddingδ: X ֒→ X×X. It is well-known (see in particular [Ca05,CaW07]) that the Hochschild homology of OX may be defined by using the isomorphism
δ∗HH (OX)≃RHomOX×X O∆
X, ω∆holX (1.1) .
Moreover, if F is a coherent OX-module and DOF := RHomO
X(F, ωXhol) denotes its dual, there are natural morphisms
O∆
X −→F ⊠DOF −→ω∆holX (1.2)
whose composition defines the Hochschild class of F: hhO(F)∈HSupp(0 F)(X;HH(OX)).
These constructions have been extended when replacing OX with a so-called DQ-algebroid stackAX in [KS12] (DQ stands for “deformation-quantization”).
One of the main results of loc. cit. is that Hochschild classes are functorial with respect to the composition of kernels, a kind of (relative) index theorem for coherent DQ-modules.
On the other hand, the notion of Lagrangian cycles of constructible sheaves on real analytic manifolds has been introduced by the first named author (see [Ka85]) in order to prove an index theorem for such sheaves, after they first appeared in the complex case (see [Ka73] and [McP74]). We refer to [KS90, Chap. 9] for a systematic study of Lagrangian cycles and for historical comments. Let us briefly recall the construction.
Consider a real analytic manifold M and let k be a unital commutative ring with finite global dimension. Denote by ωM the (topological) dualizing complex of M, that is, ωM = orM[dimM] where orM is the orientation sheaf of M and dimM is the dimension. Finally, denote by πM: T∗M −→ M the cotangent bundle to M. Let Λ be a conic subanalytic Lagrangian subset of T∗M. The group of Lagrangian cycles supported by Λ is given by HΛ0(T∗M;π−M1ωM). Denote by DbR-c(kM) the bounded derived category ofR- constructible sheaves on M. To an object F of this category, one associates a Lagrangian cycle supported by SS(F), the microsupport ofF. This cycle is called the characteristic cycle, or the Lagrangian cycle or else the microlocal Euler class of F and is denoted here by µeuM(F).
In fact, it is possible to treat the microlocal Euler classes ofR-constructi- ble sheaves on real manifolds similarly as the Hochschild class of coherent sheaves on complex manifolds. Denote as above by k∆M and ω∆M the direct image of kM and ωM by the diagonal embedding δM: M ֒→ M ×M. Then we have an isomorphism
HΛ0(T∗M;πM−1ωM)≃HΛ0 T∗M;µhom(k∆M, ω∆M) (1.3) ,
where µhomis the microlocalization of the functor RHom. ThenµeuM(F) is obtained as follows. Denote by DMF := RHom(F, ωM) the dual of F. There are natural morphisms
k∆M −→F ⊠DMF −→ω∆M, (1.4)
whose composition gives the microlocal Euler class of F.
In this paper, we construct the microlocal Euler class for a wide class of sheaves, including of course the constructible sheaves but also the sheaves of holomorphic solutions of coherent D-modules and, more generally, of ellip- tic pairs in the sense of [ScSn94]. To treat such situations, we are led to introduce the notion of a trace kernel.
On a real manifold M (say of class C∞), a trace kernel is the data of a triplet (K, u, v) where K is an object of the derived category of sheaves Db(kM×M) and u, v are morphisms
u: k∆M −→K, v: K −→ω∆M. (1.5)
One then naturally defines the microlocal Euler class µeuM(K, u, v) of such a kernel, an element of HΛ0(T∗M;µhom(k∆M, ω∆M)) where Λ = SS(K)∩ T∆∗M(M ×M). By (1.4), a constructible sheaf gives rise to a trace kernel.
IfXis a complex manifold andM is a coherentDX-module, we construct natural morphisms (over the base ring k=C)
C∆
X −→ΩX×X
⊗LD
X×X(M⊠DDM)−→ ω∆X, (1.6)
where DDM denotes the dual of M as a D-module. In other words, one naturally associates a trace kernel onX to a coherentDX-module. Moreover, we prove that under suitable microlocal conditions, the tensor product of two trace kernels is again a trace kernel, and it follows that one can associate a trace kernel to an elliptic pair.
We study trace kernels and their microlocal Euler classes, showing that some proofs of [KS12] can be easily adapted to this situation. One of our main results is the functoriality of the microlocal Euler classes: the microlocal Euler class of the composition K1◦K2 of two trace kernels is the composition of the microlocal Euler classes of K1 and K2 (see Theorem 6.3 for a precise statement). Another essential result (which is far from obvious) is that the composition of classes coincides with the composition forπM−1ωM constructed in [KS90] via the isomorphism between µhom(k∆M, ω∆M) and πM−1ωM.
As an application, we recover in a single proof the classical results on the index theorem for constructible sheaves (see [KS90,§ 9.5]) as well as the index theorem for elliptic pairs of [ScSn94], that is, sheaves of generalized holomorphic solutions of coherent D-modules. We also briefly explain how to adapt trace kernels to the formalism of the Lefschetz trace formula.
We call here µhom(k∆M, ω∆M) the microlocal homology of M, and this paper shows that, in some sense, the microlocal homology of real manifolds plays the same role as the Hochschild homology of complex manifolds.
To conclude this introduction, let us make a general remark. The cate- gory DbR-c(kM) of constructible sheaves on a compact real analytic manifold M is “proper” in the sense of Kontsevich (that is, Ext finite) but it does not admit a Serre functor (in the sense of Bondal-Kapranov) and it is not clear whether it is smooth (again in the sense of Kontsevich). However this category naturally appears in Mirror Symmetry (see [FLTZ10]) and it would be a natural question to try to understand its Hochschild homology in the sense of [McC94, Ke99]. We don’t know how to compute it, but the above construction, with the use of µhom(k∆M, ω∆M), provides an alternative ap- proach of the Hochschild homology of this category. This result is not totally surprising if one remembers the formula (see [KS90, Prop. 8.4.14]):
DT∗M(µhom(F, G))≃µhom(G, F)⊗πM−1ωM.
Hence, in some sense, π−M1ωM plays the role of a microlocal Serre functor.
Note that thanks to Nadler and Zaslow [NZ09], the category DbR-c(kM) is equivalent to the Fukaya category of the symplectic manifold T∗M, and this is another argument to treat sheaves from a microlocal point of view.
Acknowledgments
The second named author warmly thanks St´ephane Guillermou for helpful discussions.
2 A short review on sheaves
Throughout this paper, a manifold means a real manifold of class C∞. We shall mainly follow the notations of [KS90] and use some of the main notions introduced there, in particular that of microsupport and the functor µhom.
LetM be a manifold. We denote byπM: T∗M −→M its cotangent bundle.
For a submanifold N of M, we denote by TN∗M the conormal bundle to N. In particular, TM∗ M denotes the zero-section. We set ˙T∗M :=T∗M \TM∗M and we denote by ˙πM the restriction of πM to ˙T∗M. If there is no risk of confusion, we write simply π and ˙π instead of πM and ˙πM. One denotes by a: T∗M −→ T∗M the antipodal map, (x;ξ)7→ (x;−ξ) and for a subset S of T∗M, one denotes by Sa its image by this map. A set A ⊂ T∗M is conic if it is invariant by the action of R+ onT∗M.
Letf: M −→N be a morphism of manifolds. Tof one associates as usual the maps
T∗M
πM
))
❙❙
❙❙
❙❙
❙❙
❙❙
❙❙
❙❙
❙❙
❙ M ×N T∗N
π
fd
oo fπ //T∗N
πN
M f //N.
(2.1)
(Note that in loc. cit. the map fd is denoted by tf′−1.)
Let Λ be a closed conic subset ofT∗N. One says thatfisnon-characteris- ticfor Λ if the mapfdis proper onfπ−1Λ or, equivalently,fπ−1Λ∩fd−1(TM∗ M)⊂ M ×N TN∗N.
Let k be a commutative unital ring with finite global homological di- mension. One denotes by kM the constant sheaf on M with stalk k and by Db(kM) the bounded derived category of sheaves of k-modules onM. When M is a real analytic manifold, one denotes byDbR-c(kM) the full triangulated subcategory of Db(kM) consisting of R-constructible objects.
One denotes byωM the dualizing complex onM and byωM⊗−1its dual, that is, ωM⊗−1 = RHom(ωM,kM). More generally, for a morphism f: M −→ N, one denotes by ωM/N := f!kN ≃ ωM ⊗f−1(ωN⊗−1) the relative dualizing complex. Recall that ωM ≃orM[dimM] where orM is the orientation sheaf and dimM is the dimension of M. Also recall the natural morphism of functors
ωM/N ⊗f−1 −→f!. (2.2)
We have the duality functors
D′MF = RHom(F,kM), DMF = RHom(F, ωM).
For F ∈ Db(kM), one denotes by Supp(F) the support of F and by SS(F) its microsupport, a closedR+-conic co-isotropic subset ofT∗M. For a morphism f: M −→N andG∈Db(kN), one says thatf is non-characteristic for G if f is non-characteristic for SS(G).
We shall use systematically the functorµhom, a variant of Sato’s microlo- calization functor. Recall that for a closed submanifold N of M, there is a functor µN: Db(kM) −→ Db(kTN∗M) constructed by Sato (see [SKK73]) and for F1, F2 ∈Db(kM), one defines in [KS90] the functor
µhom:Db(kM)op×Db(kM)−→Db(kT∗M),
µhom(F1, F2) :=µ∆RHom(q2−1F1, q1!F2)
where q1 and q2 are the first and second projection defined on M ×M and
∆ is the diagonal. This sheaf is supported by T∆∗(M ×M) that we identify with T∗M by the first projectionT∗(M×M) ≃T∗M×T∗M −→T∗M. Note that
Supp(µhom(F1, F2))⊂SS(F1)∩SS(F2) (2.3)
and we have Sato’s distinguished triangle, functorial in F1 and F2: Rπ!µhom(F1, F2)−→Rπ∗µhom(F1, F2)−→R ˙π∗ µhom(F1, F2)|T˙∗M
+1
−→. (2.4)
Moreover, we have the isomorphism
Rπ∗µhom(F1, F2)≃RHom(F1, F2), (2.5)
and, assuming thatM is real analytic andF1 is R-constructible, the isomor- phism
Rπ!µhom(F1, F2)≃D′MF1
⊗L F2. (2.6)
In particular, assuming that F1 is R-constructible and SS(F1)∩SS(F2) ⊂ TM∗ M, we have the natural isomorphism (see [KS90, Cor 6.4.3])
D′MF1
⊗L F2 −→∼ RHom(F1, F2).
(2.7)
As recalled in the Introduction, assuming that M is real analytic, we have the formula (see [KS90, Prop. 8.4.14]):
DT∗M(µhom(F1, F2))≃µhom(F2, F1)⊗πM−1ωM for F1, F2 ∈DbR-c(kM).
(2.8)
3 Compositions of kernels
Notation 3.1. (i) For a manifold M, let δM: M −→ M ×M denote the diagonal embedding, and ∆M the diagonal set of M ×M.
(ii) Let Mi (i= 1,2,3) be manifolds. For short, we write Mij :=Mi×Mj
(1≤ i, j ≤ 3), M123 =M1 ×M2×M3, M1223 = M1×M2×M2 ×M3, etc.
(iii) We will often write for shortki instead of kMi and k∆i instead of k∆Mi and similarly withωMi, etc., and with the indexireplaced with several indicesij, etc.
(iv) We denote by πi, πij, etc. the projection T∗Mi →− Mi, T∗Mij −→ Mij, etc.
(v) We denote by qi the projectionMij −→Mi or the projectionM123 −→Mi
and by qij the projection M123 −→ Mij. Similarly, we denote by pi the projection T∗Mij −→ T∗Mi or the projection T∗M123 −→ T∗Mi and by pij the projection T∗M123 −→T∗Mij.
(vi) We also need to introduce the maps pja or pija, the composition of pj
orpij and the antipodal map onT∗Mj. For example, p12a((x1, x2, x3;ξ1, ξ2, ξ3)) = (x1, x2;ξ1,−ξ2).
(vii) We let δ2: M123 −→M1223 be the natural diagonal embedding.
We consider the operation of composition of kernels:
◦2 : Db(kM12)×Db(kM23)−→Db(kM13) (K1, K2)7→K1◦
2K2 := Rq13!(q12−1K1
⊗L q−231K2)
≃ Rq13!δ−21(K1 L
⊠K2).
(3.1)
We will use a variant of ◦:
∗2 : Db(kM12)×Db(kM23)−→Db(kM13) (K1, K2)7→K1∗
2K2:= Rq13∗ q2−1ω2⊗δ2!(K1 L
⊠K2) (3.2) .
We also have ωM123/M1223 ≃ q2−1ωM⊗−21 and we deduce from (2.2) a morphism δ2−1 −→q2−1ωM2⊗δ2! . Using the morphism Rp13! −→Rp13∗ we obtain a natural morphism for K1 ∈Db(kM12) andK2 ∈Db(kM23):
K1◦K2 −→K1∗K2. (3.3)
It is an isomorphism if p−121aSS(K1)∩p23−1aSS(K2)−→T∗M13 is proper.
We define the composition of kernels on cotangent bundles (see [KS90, Prop. 4.4.11])
◦a
2 : Db(kT∗M12)×Db(kT∗M23) −→ Db(kT∗M13) (K1, K2) 7→ K1
◦a
2K2:= Rp13!(p−121aK1
⊗L p−231K2)
≃Rp13a!(p−121aK1⊗L p−231aK2).
(3.4)
We also define the corresponding operations for subsets of cotangent bundles.
Let A⊂T∗M12 and B ⊂T∗M23. We set A×a
2 B =p−121a(A)∩p−231(B), A◦a
2B =p13(A×a
2 B)
=
(x1, x3;ξ1, ξ3)∈T∗M13; there exists (x2;ξ2)∈T∗M2 such that (x1, x2;ξ1,−ξ2)∈A,(x2, x3;ξ2, ξ3)∈B
. (3.5)
We have the following result which slightly strengthens Proposition 4.4.11 of [KS90] in which the composition ∗ is not used.
Proposition 3.2. For G1, F1 ∈ Db(kM12) and G2, F2 ∈ Db(kM23) there ex- ists a canonical morphism (whose construction is similar to that of [KS90, Prop. 4.4.11] ):
µhom(G1, F1)◦a
2µhom(G2, F2)−→µhom(G1∗
2G2, F1◦
2F2).
Proof. In Proposition 4.4.8 (i) of loc. cit., one may replace F2 ⊠LS G2 with j!(F2⊠L G2)⊗ωX⊗−×1
SY /X×Y. Then the proof goes exactly as that of Proposi-
tion 4.4.11 in loc. cit. Q.E.D.
Let Λij ⊂ T∗Mij (i= 1,2, j =i+ 1) be closed conic subsets and consider the condition:
the projectionp13: Λ12×a
2 Λ23−−→T∗M13 is proper.
(3.6) We set
Λ13= Λ12
◦a 2Λ23. (3.7)
Corollary 3.3. Assume that Λij (i= 1,2, j =i+ 1) satisfy (3.6). We have a composition morphism
RΓΛ12µhom(G1, F1)◦a
2RΓΛ23µhom(G2, F2)−→RΓΛ13µhom(G1∗
2G2, F1◦
2F2).
Convention 3.4. In (3.1), we have introduced the composition ◦
2 of kernels K1 ∈ Db(kM12) and K2 ∈ Db(kM23). However we shall also use the nota- tion M22 = M2×M2 and consider for example kernels L1 ∈ Db(kM122) and L2 ∈ Db(kM223). Then when writing L1◦
2L2 we mean that the composition is taken with respect to the last variable of M22 for L1 and the first variable for L2. In other words, set M4 =M2 and consider L1 and L2 as objects of Db(kM142) and Db(kM243) respectively, in which case the composition L1◦
2L2 is unambiguously defined.
4 Microlocal homology
LetM be a real manifold. Recall thatδM: M ֒→M×M denotes the diagonal embedding. We shall identify M with the diagonal ∆M of M ×M and we
sometimes write ∆ instead of ∆M if there is no risk of confusion. We shall identify T∗M with T∆∗(M ×M) by the map
δTa∗M: T∗M //T∗(M ×M), (x;ξ)7→(x, x;ξ,−ξ).
We denote by k∆M, ω∆M and ω∆⊗−M1 the direct image by δM of kM, ωM and ωM⊗−1:= RHom(ωM,kM), respectively.
The next definition is inspired by that of Hochschild homology on complex manifolds (see Introduction).
Definition 4.1. Let Λ be a closed conic subset of T∗M. We set MHΛ(kM) := RΓΛ(δaT∗M)−1µhom(k∆M, ω∆M),
MHΛ(kM) := RΓ(T∗M;MHΛ(kM)), MHk
Λ(kM) := Hk(MHΛ(kM)) =Hk(T∗M;MHΛ(kM)).
(4.1)
We call MHΛ(kM) themicrolocal homology of M with support in Λ.
We also write MH(kM) instead of MHT∗M(kM).
Remark 4.2. (i) We have µhom(k∆M, ω∆M) ≃ (δTa∗M)∗πM−1ωM. In par- ticular, we have MHΛ(kM) ≃ RΓΛ(T∗M;πM−1ωM) and MH(kM) ≃ RΓ(M;ωM). Assuming that M is real analytic and Λ is a closed conic subanalytic Lagrangian subset of T∗M, we recover the space of La- grangian cycles with support in Λ as defined in [KS90, §9.3].
(ii) The support of µhom(k∆M, ω∆M) is T∆∗M(M ×M). Hence, we have RΓδa
T∗MΛµhom(k∆M, ω∆M)≃(δTa∗M)∗MHΛ(kM).
(iii) IfM is real analytic and Λ is a Lagrangian subanalytic closed conic sub- set, then we haveHk(MHΛ(kM)) = 0 fork <0 (see [KS90, Prop. 9.2.2]).
In the sequel, we denote by ∆i (resp. ∆ij) the diagonal subset ∆Mi ⊂Mii
(resp. ∆Mij ⊂Miijj).
Lemma 4.3. We have natural morphisms:
(i) ω∆12 ◦
22(k∆2
L
⊠ω∆3)−→ω∆13, (ii) k∆13 −→k∆12 ∗
22(ω⊗−∆21⊠L k∆3).
Proof. Denote by δ22 the diagonal embedding M112233 ֒→M11222233. (i) We have the morphisms
ω∆12 ◦
22(k∆2
L
⊠ω∆3) = Rq1133!δ22−1(ω∆12
L
⊠k∆2
L
⊠ω∆3)
≃ Rq1133!ω∆123
−
→ ω∆13. (ii) The isomorphism
δ22! (k∆2 ⊠ω∆2)≃k∆2
gives rise to the isomorphisms k∆12 ∗
22(ω∆⊗−21⊠L k∆3) = Rq1133∗ q−11331 ω22⊗δ22! (k∆12
L
⊠ω⊗−∆21⊠L k∆3)
≃ Rq1133∗δ22! (k∆1
L
⊠ω∆2
L
⊠k∆23)
≃ Rq1133∗k∆123
and the result follows by adjunction from the morphism
q1133−1 k∆13 ≃k∆1 ⊠L k22⊠L k∆3 −→k∆1 ⊠L k∆2 ⊠L k∆3 =k∆123.
Q.E.D.
Proposition 4.4. Let Mi (i = 1,2,3) be manifolds. We have a natural composition morphism (whose constructions will be given in the course of the proof):
µhom(k∆12, ω∆12)◦a
22µhom(k∆23, ω∆23)−→µhom(k∆13, ω∆13).
(4.2)
In particular, let Λij be a closed conic subset of T∗Mij (ij = 12,13,23). If Λ12
◦a
2Λ23⊂ Λ13, then we have a morphism MHΛ12(k12) ◦a
2
MHΛ23(k23)−→MHΛ13(k13).
(4.3)
Proof. Consider the morphism (see Proposition 3.2 and Convention 3.4) µhom(ω∆⊗−21, ω∆⊗−21)◦a
2µhom(k∆23, ω∆23) −→ µhom(ω∆⊗−21∗
2k∆23, ω∆⊗−21◦
2ω∆23)
≃ µhom(ω∆⊗−21⊠L k∆3,k∆2
L
⊠ω∆3).
It induces an isomorphism
µhom(k∆23, ω∆23)≃µhom(ω∆⊗−21⊠L k∆3,k∆2
L
⊠ω∆3).
(4.4)
Note that this isomorphism is also obtained by
µhom(k∆23, ω∆23) ≃ µhom (ω⊗−2 1⊠L k233)⊗L k∆23,(ω2⊗−1⊠L k233)⊗L ω∆23
≃ µhom(ω∆⊗−21⊠L k∆3,k∆2 ⊠L ω∆3).
Applying Proposition 3.2, we get a morphism:
µhom(k∆12, ω∆12)◦a
22µhom(k∆23, ω∆23)
−
→µhom(k∆12 ∗
22(ω∆⊗−21⊠L k∆3), ω∆12 ◦
22(k∆2
L
⊠ω∆3)).
(4.5)
It remains to apply Lemma 4.3. Q.E.D.
Corollary 4.5. Let Λij (i = 1,2, j = i+ 1) satisfying (3.6) and let Λ13 = Λ12
◦a
2Λ23. The composition of kernels in (4.3) induces a morphism
◦a
2 : MHΛ12(k12) ⊗L MHΛ23(k23)−→MHΛ13(k13).
(4.6)
In particular, each λ∈ MH0
Λ12(k12) defines a morphism λ◦a
2 : MHΛ23(k23)−→MHΛ13(k13).
(4.7)
Proof. These morphisms follow from (4.3).The second assertion follows from the isomorphism H0(X)≃HomDb(k)(k, X) in the category Db(k). Q.E.D.
Theorem 4.6. (i) We have the isomorphisms
µhom(k∆M, ω∆M) ≃ (δTa∗M)∗π−M1RHom(kM, ωM)
≃ (δTa∗M)∗π−M1ωM. (ii) We have a commutative diagram
µhom(k∆12, ω∆12)◦a
22µhom(k∆23, ω∆23) //
≀
µhom(k∆13, ω∆13)
≀
(δTa∗M13)∗ π−M112ωM12
◦a
2π−M123ωM23
//(δTa∗M13)∗πM−113ωM13. (4.8)
Here the top horizontal arrow of (4.8) is given in Proposition4.4, and the bottom horizontal arrow is induced by
p−121aπ−M112ωM12
⊗L p−231πM−123aωM23 ≃πM−11ωM1
L
⊠πM−12(ωM2
⊗L ωM2)⊠L πM−13ωM3, π−M12(ωM2
⊗L ωM2)≃ωT∗M2, Rp13! πM−11ωM1
L
⊠ωT∗M2
L
⊠πM−13ωM3
−−→π−M11ωM1
L
⊠π−M13ωM3. Proof. (i) is obvious.
(ii)–(a) By [KS90, Prop. 4.4.8], we have natural morphisms for (i, j) = (1,2) or (i, j) = (2,3):
µhom(k∆i, ω∆i)⊠L µhom(k∆j, ω∆j) −→ µhom(k∆ij, ω∆ij)
and it follows from (i) that these morphisms are isomorphisms. These iso- morphisms give rise to the isomorphism
µhom(k∆12, ω∆12)◦a
22µhom(k∆23, ω∆23) ≃ µhom(k∆1, ω∆1)⊠L µhom(k∆2, ω∆2)◦a
22µhom(k∆2, ω∆2) L
⊠µhom(k∆3, ω∆3) Similarly, we have an isomorphism
π−M112ωM12
◦a
2π−M123ωM23 ≃ πM−11ωM1 ⊠ πM−12ωM2
◦a
2πM−12ωM2
⊠πM−11ωM1.
Hence, we are reduced to the case where M1 = M3 = pt, which we shall assume now.
(ii)–(b) We change our notations and we set:
M :=M2, Y :=M ×M,
δM: M ֒→Y the diagonal embedding,∆M =δM(M), j: Y ֒→Y ×Y the diagonal embedding, ∆Y =δY(Y), δaT∗M: T∗M ֒→T∗Y, (x;ξ)7→(x, x;ξ,−ξ),
δaT∗Y : T∗Y ֒→T∗Y ×T∗Y, p: T∗Y −→pt the projection, aY : Y −→pt the projection.
With these new notations, the composition ◦a
22 will be denoted by ◦a
T∗Y. Consider the diagram4.9 similar to Diagram (4.4.15) of [KS90]
T∗M ×T∗M i //T∗Y ×T∗Y δ ? _T∗Y
a T∗Y
o
p
""
❉❉
❉❉
❉❉
❉❉
❉❉
❉❉
❉❉
❉❉
❉❉
❉❉
❉❉
T∗Y ×Y T∗Y
jπ
OO
jd
T∆∗Y(Y ×Y)
e ? _
o s
πY
p1
∼
OO
T∗Y
s Y
oo a
Y //pt.
(4.9)
Here, iis the canonical embedding induced byδTa∗M,p1 is induced by the first projection T∗Y ×T∗Y −→ T∗Y, s: Y ֒→T∗Y is the zero-section embedding and es is the natural embedding. Note that the square labelled by is Cartesian. We have
Rp! ◦(δaT∗Y)−1 ≃ RaY! ◦RπY! ◦p−11◦(δaT∗Y)−1
≃ RaY! ◦RπY! ◦es−1 ◦jπ−1
≃ RaY! ◦s−1◦Rjd! ◦jπ−1. Therefore,
µhom(k∆M, ω∆M) ◦a
T∗Y µhom(k∆M, ω∆M)
≃Rp!(δTa∗Y)−1 µhom(k∆M, ω∆M)⊠L µhom(k∆M, ω∆M)
≃RaY!s−1Rjd!jπ−1µhom(k∆M
L
⊠k∆M, ω∆M
L
⊠ω∆M).
Hence, by adjunction, to give a morphism µhom(k∆M, ω∆M) ◦a
T∗Y µhom(k∆M, ω∆M)−→k is equivalent to giving a morphism in Db(kY)
s−1Rjd!jπ−1µhom(k∆M
L
⊠k∆M, ω∆M
L
⊠ω∆M)−→aY! kpt. (4.10)
Note that the left hand side of (4.10) is supported on ∆M. Hence in order to give a morphism (4.10), it is necessary and sufficient to give a morphism in Db(kM)
δM
−1s−1Rjd!jπ−1µhom(k∆M
L
⊠k∆M, ω∆M
L
⊠ω∆M)−→δM!aY! kpt. (4.11)
Hence, it is enough to check the commutativity of the upper square in the following diagram in Db(kM)
δM
−1s−1Rjd!jπ−1µhom(k∆M
L
⊠k∆M, ω∆M
L
⊠ω∆M) //
∼
δM! aY! kpt
id
δM−1s−1Rjd!jπ−1i∗ π−M1ωM L
⊠πM−1ωM
//
∼
δM! aY! kpt
∼
ωM id
//ωM.
(4.12)
The top horizontal arrow is constructed by the chain of morphisms (see [KS90,
§ 4.4]):
Rjd!jπ−1µhom(k∆M
L
⊠k∆M, ω∆M
L
⊠ω∆M)
−
→µhom(j!(k∆M ⊠L k∆M)⊗L ωY, j−1(ω∆M ⊠L ω∆M))
≃µhom(ω∆M, ω∆M ⊗ω∆M)≃(δTa∗M)∗πM−1ωM
and
δM−1s−1Rjd!jπ−1µhom(k∆M ⊠L k∆M, ω∆M ⊠L ω∆M)
−
→δM−1s−1(δTa∗M)∗πM−1ωM ≃ωM. (4.13)
Hence, the commutativity of the diagram (4.12) is reduced to the commuta- tivity of the diagram below:
δM−1s−1Rjd!jπ−1µhom(k∆M
L
⊠k∆M, ω∆M ⊠L ω∆M)
λ
++
❱❱
❱❱
❱❱
❱❱
❱❱
❱❱
❱❱
❱❱
❱❱
❱❱
❱❱
❱❱
❱❱
❱❱
❱❱
δM−1s−1Rjd!jπ−1i∗ πM−1ωM ⊠L π−M1ωM ∼
//ωM.
(4.14)
where the morphism λ is given by the morphisms in (4.13). All terms of (4.14) are concentrated at the degree −dimM. Hence the commutativity of (4.14) is a local problem in M and we can assume that M is a Euclidean space. We can checked directly in this case. Q.E.D.
Remark 4.7. Theorem 4.6 may be applied as follows. Let Λij be a closed conic subset of T∗Mij (i = 1.2, j = i+ 1). Assume (3.6), that is, the projection p13: Λ12×a
2 Λ23−−→T∗M13 is proper and set Λ13 = Λ12◦a
2Λ23. Let λij ∈MH0
Λij(kMij)≃HΛ0ij(T∗Mij;π−1ωij). Then λ12
◦a 2λ23=
Z
T∗M2
λ12∪λ23
(4.15)
where the right hand-side is obtained as follows. Set Λ := Λ12×a
2 Λ23 and consider the morphisms
HΛ012(T∗M12;π−1ω12)×HΛ023(T∗M23;π−1ω23)
−
→HΛ0(T∗M123;π−1ω1⊠L ωT∗M2
L
⊠π−1ω3)
−
→HΛ013(T∗M13;π−1ω13).
The first morphism is the cup product and the second one is the integration morphism with respect to T∗M2.
5 Microlocal Euler classes of trace kernels
In this section, we often write ∆ instead of ∆M.
Definition 5.1. Atrace kernel(K, u, v) onM is the data ofK ∈Db(kM×M) together with morphisms
k∆ −−→u K and K −−→v ω∆. (5.1)
In the sequel, as far as there is no risk of confusion, we simply write K instead of (K, u, v).
For a trace kernelK as above, we set
SS∆(K) := SS(K)∩T∆∗(M ×M) = (δTa∗M)−1SS(K).
(5.2)
(Recall that one often identifies T∗M and T∆∗(M ×M) by δTa∗M: T∗M ֒→ T∗M ×T∗M.)
Definition 5.2. Let (K, u, v) be a trace kernel.
(a) The morphism u defines an element ˜u in HSS0 ∆(K)(T∗M;µhom(k∆, K)) and the microlocal Euler class µeuM(K) of K is the image of ˜u by the morphism µhom(k∆, K)−→µhom(k∆, ω∆) associated with the mor- phism v.
(b) Let Λ be a closed conic subset of T∗M containing SS∆(K). One denotes by µeuΛ(K) the image of ˜u in HΛ0 T∗M;µhom(k∆, ω∆)
. Hence,
µeuΛ(K)∈MH0
Λ(kM)≃HΛ0(T∗M;π−1ωM).
(5.3)
Let ˜v be the element ofHSS0 ∆(K)(T∗M;µhom(K, ω∆)) induced byv. Then the microlocal Euler classµeuM(K) ofK coincides with the image of ˜v by the morphism µhom(K, ω∆M) −→ µhom(k∆, ω∆) associated with the morphism u, which can be easily seen by the commutative diagram:
(δTa∗M)−1µhom(K, K) v //
u
(δTa∗M)−1µhom(K, ω∆)
u
(δaT∗M)−1µhom(k∆, K) v //(δaT∗M)−1µhom(k∆, ω∆).
One denotes by eu(K) the restriction of µeu(K) to the zero-section M of T∗M and calls it the Euler class of K. Hence
euM(K)∈HSupp(K)0 ∩∆(M;ωM).
(5.4)
It is nothing but the class induced by the composition k∆M −→K −→ω∆M. We say thatL∈Db(kM) isinvertibleifLis locally isomorphic tokM[d] for some d∈Z. Then, L⊗−1:= RHom(L,kM) is also invertible andL⊗LL⊗−1 ≃ kM.
Proposition 5.3. Let L be an invertible object in Db(kM) and K a trace kernel. Then K⊗L (L⊠L L⊗−1) is a trace kernel and µeu K⊗L (L⊠L L⊗−1)
= µeu(K).
Proof. L⊠L L⊗−1 is canonically isomorphic to kM×M on a neighborhood of
the diagonal set ∆M of M ×M. Q.E.D.