Topology of Stratified Spaces MSRI Publications Volume 58, 2011
Intersection homology Wang sequence
FILIPP LEVIKOV
ABSTRACT. We prove the existence of a Wang-like sequence for intersection homology. A result is given on vanishing of the middle dimensional inter- section homology group of “generalized Thom spaces”, which naturally occur in the decomposition formula of S. Cappell and J. Shaneson. Based upon this result, consequences for the signature are drawn. For non-Witt spaces X , signature and L-classes are defined via the hyper- i cohomology groups H .X I ICL/, introduced in [Ban02]. A hypercohomology i Wang sequence is deduced, connecting H . I ICL/ of the total space with that of the fibre. Also here, a consequence for the signature under collapsing sphere-singularities is drawn.
1. Introduction
The goal of this article is to add to the intersection homology toolkit another useful long exact sequence. In [Wan49], H. C. Wang, calculating the homol- ogy of the total space of a fibre bundle over a sphere, actually proved an exact sequence, which is named after him today. It is a useful tool for dealing with fibre bundles over spheres and it is natural to ask: Is there a Wang sequence for intersection homology? Given an appropriate notion of a stratified fibration, the natural framework for dealing with a question of the kind above would be an intersection homology analogue of a Leray–Serre spectral sequence. Greg Friedman has investigated this and established an appropriate framework in [Fri07]. For a simplified setting of a stratified bundle, however, i.e., a locally trivial bundle over a manifold with a stratified fibre, it seems more natural to explore the hypercohomology spectral sequence directly. In the following we are going to demonstrate this approach.
Mathematics Subject Classification: 55N33. Keywords: intersection homology, Wang sequence, signature.
251 252 FILIPP LEVIKOV
Section 3 is a kind of foretaste of what is to come. We prove the monodromy case by hand using only elementary intersection homology and apply it to calcu- late the intersection homology groups of neighbourhoods of circle singularities with toric links in a 4-dimensional pseudomanifold. We recall the construction of induced maps in Section 4.1. Because of their central role in the application, the Cappell–Shaneson decomposition formula is explained in Section 4.2. Section 5 contains a proof of the Wang sequence for fibre bundles over simply connected spheres. It is shown that under a certain assumption the middle-dimensional middle perversity intersection homology of generalized Thom spaces of bundles over spheres vanish. The formula of Cappell and Shaneson then implies, that in this situation the signature does not change under the collapsing of the spherical singularities. In Section 6, we demonstrate a second, concise proof — this is merely the sheaf-theoretic combination of the relative long exact sequence and the suspen- sion isomorphism. However, this proof is mimicked in Section 7 to derive a Wang-like sequence for hypercohomology groups H.X I S/ with values in a self-dual perverse sheaf complex S 2 SD.X /. In Section 8, finally, together with Novikov additivity, this enables us to identify situations when collapsing spherical singularities in non-Witt spaces does not change the signature.
2. Basic notions
We will work in the framework of [GM83]. In the following X D Xn Xn 2 X0 X 1 D ? will denote an oriented n-dimensional strati- fied topological pseudomanifold. The intersection homology groups of X with pN respect to perversity pN are denoted by IH i .X /, and the analogous compact- cpN support homology groups by IH i .X /. The indexing convention is also that of [GM83]. Most of the fibre bundles to be considered in the following are going to be stratified bundles in the following sense (see also [Fri07, Definition 5.6]):
DEFINITION 2.1. A projection E ! B to a manifold is called a stratified bundle if for each point b 2 B there exist a neighbourhood U B and a stratum- preserving trivialization p 1.U / Š U F, where F is a topological stratified pseudomanifold. We will also restrict the automorphism group of F to stratum preserving auto- morphisms and work with the corresponding fibre bundles in the usual sense. Since we will basically need the local triviality, Definition 2.1 is mostly suf- ficient. When we pass to applications for Whitney stratified pseudomanifolds, however, the considered bundles will actually be fibre bundles — this follows from the theory of Whitney stratifications. A stratification of the fibre induces an obvious stratification of the total space with the same l-codimensional links, INTERSECTIONHOMOLOGYWANGSEQUENCE 253 namely by EkCn l — the total spaces of bundles with fibre Fk l and n the dimension of B.
3. Mapping torus
PROPOSITION 3.1 (INTERSECTION HOMOLOGY WANG SEQUENCE FOR S 1). Let F D Fn Fn 2 F0 be a topological stratified pseudomanifold, W F ! F a stratum and codimension preserving automorphism, i.e., a stratum preserving homeomorphism with stratum preserving inverse such that both maps respect the codimension. Let M be the mapping torus of , i.e., the quotient space F I=.y; 1/ s ..y/; 0/. Denote by i W F D F 0 Œ M the inclusion. Then the sequence
cpN id cpN i cpN @ cpN IH k .F/ ! IH k .F/ IH k .M/ IH k 1.F/ is exact. PROOF. The proof is analogous to the one for ordinary homology. Start with the quotient map q W .F I; F/ ! .M; F/ and look at the corresponding diagram of long exact sequences of pairs. The boundary of F I is a codimension 1 stratum and hence not a pseudomanifold. We have either to introduce the notion of a pseudomanifold with boundary here or work with intersection homology for cs-sets [Kin85; HS91]. However, we can also manage with a work-around: Define
I" WD . "; 1 C "/; @I" WD . ";"/ [ .1 "; 1 C "/; F" WD F . "; "/:
We extend the identification .y; 1/ s ..y/; 0/ to F I" by introducing the quotient map q W F I" ! M, . 1.y/; 1 C t/ if t2 . "; 0 q.y; t/ D 8.y; t/ if t2 .0; 1/ <..y/; t 1/ if t2 Œ1; 1 C "/:
Evidently, M D q.F I"/.: Now F @I" D .F @I"/nC1 .F @I"/ .F @I"/0 is an open subpseudomanifold of F I" and F D Fn Fn 2 F0 sits normally nonsingular in M. Hence the inclusions induce morphisms on intersection homology and we get a morphism of the corresponding exact sequences of pairs:
0 ... cpN @ ... cpN j ... cpN 0 ... ::: ...... ::: IH k .F I"; F @I"/ IH k 1.F @I"/ IH k 1.F I"/ ...... q . q . q ...... cpN @ ... cpN i ... cpN ... ::: ...... ::: IH k .M ; F"/ IH k 1.F"/ IH k 1.M / 254 FILIPP LEVIKOV
The “boundary” of F I" is the disjoint union of two components of the form F R, so j is surjective and the outer arrows are zero maps. The connecting morphism @ is injective and therefore an isomorphism onto its image, i.e., onto ker j cpN cpN D .˛;ˇ/ j ˛ 2 IH k .F. "; C"//; ˇ 2 IH k .F.1 "; 1C"//; Œ˛Cˇ D 0 ˚ cpN R cpN « Df.˛; ˛/gŠ IH k .F / Š IH k .F/: cpN cpN The middle q maps .˛; ˛/ to .˛ .˛// 2 IH k .F"/ Š IH k .F/. Since q commutes with @, one has
@ ı q ı @j 1 D q j cpN D id : ker jŠIH k .F / Hence, we have the diagram q j ... cpN ker j ... cpN i... cpN ... cpN : : : ...... IH .M / ...... IH .M ; F / IH k .F/ IH k .F/ k k " ...... ko . . . . . ker j . . @ ...... 1 ...... o..... @j . cpN q... cpN ...... IH kC1.F I"; F @I"/ IH kC1.M; F"/ where the top sequence is exact and the bottom square is commutative. As in ord- cpN cpN inary homology one can show that q W IH .F I"; F @I"/ ! IH .M; F"/ k cpNk is an isomorphism. Finally, observe that on the right hand side IH k .M; F"/Š cpN 1 ˜ IH k 1.F/ via @ ı q .
Let X D X4 X1 X0 be a compact stratified pseudomanifold with X0 D ?. Then, the stratum of codimension 3 is just a disjoint union of circles X1 D 1 1 S tt S . If we assume X to be PL, the link L at a point p 2 X1 is independent of p within a connective component of X1. Furthermore, in X4, there is a neighbourhood U of the circle containing p, which is a fibre bundle 1 over S and hence homeomorphic to the mapping torus M with Š W c˚.L/ ! c˚.L/: Putting this data together and using the Wang sequence we can compute cpN cpN IH k .U /, with U a neighbourhood of X1 X4. The group IH k .X / can then be computed via the Mayer–Vietoris sequence. In this section we restrict ourselves to the case of L being a torus T 2. While the orientation preserving mapping class group of the torus is known to be SL.2I Z/, we have to make the following restriction on its cone: In the following, we look only at those automorphisms W c˚.T 2/ ! c˚.T 2/ which are induced INTERSECTIONHOMOLOGYWANGSEQUENCE 255
Š by an automorphism of the underlying torus W T 2 ! T 2.1 It is given by a matrix ˛ 2 SL.2I Z/ and by abuse of notation we will again write ˛ for this torus automorphism. ˚ cpN 2 id .c.˛// cpN 2 Defining k to be the map IH k .c˚.T // !IH k .c˚.T //, we obtain the sequence
cpN 2 k IH k .c˚.T // cpN 2 i cpN @ cpN 2 IH k .c˚.T // IH k .M˛/ IH k 1.c˚.T // For the open cone we have
cpN 2 IH 0 .T / for k D 0; cpN 2 cpN 2 N IH k .c˚.T // D 8IH 1 .T / for pN D 0 and k D 1; <0 else: cpN : Clearly, IH k .M˛/ D 0 for k 3. Now examine the nontrivial part of the sequence
cpN @ cpN 2 1 cpN 2 i 0 IH 2 .M˛/ IH 1 .c˚.T // IH 1 .c˚.T // cpN @ cpN 2 0 cpN 2 i cpN IH 1 .M˛/ IH 0 .c˚.T // IH 0 .c˚.T // IH 0 .M˛/ 0:
Since c˚.˛/0 maps a point to a point, clearly c˚.˛/0 D id, hence 0 D id id D 0. cpN Z It follows that IH 0 .M˛/ D . Note that the only possible perversities in this example are 0N and tN. So far ctN 2 we have not distinguished between them. Due to IH 1 .c˚.T // D 0 there is ctN ctN Z IH 2 .M˛/ D 0 and IH 1 .M˛/ D . For the zero perversity, we have
c0N @ c0N 2 1 c0N 2 c0N 0 IH 2 .M˛/ ! IH 1 .c˚.T // ! IH 1 .c˚.T // IH 1 .M˛/ ko Z ˚ Z c0N 2 IH 0 .c˚.T // 0 c0N 2 2 The group IH 1 .c˚.T // is isomorphic to H1.T / and is therefore generated by the corresponding homology classes of the torus. Hence, .c˚.˛// is just c0N Z Z the matrix ˛. If ˛ D id; 1 D 0 and IH 2 .M˛/ Š ˚ . In the general c0N case, IH 2 .M˛/ is isomorphic to im @ D ker 1. We examine the determinant
1I believe that in the PL context this does not constitute a real restriction. With [Hud69, Theorem 3.6C] we can find an admissible triangulation of c˚.T 2/, such that becomes simplicial. Furthermore, it is not difficult to see, that the simplicial link of the cone point L.c/ is preserved under . Since the geometric realization of L.c/ is a torus, we get a candidate for . By linearity, every slice between L.c/ and the cone point c is mapped by . If we could extend the argument to the rest of c˚.T 2/ the goal would be achieved. 256 FILIPP LEVIKOV of the matrix: det 1 D det.id ˛/ D p˛.1/, where p˛.t/ is the characteristic polynomial of ˛, which is
a11 a12 p˛.t/ D det t id a a D .t a11/.t a22/ a21a12 21 22 2 D t .a11 C a22/t C .a11a22 a21a12/ D t 2 tr ˛ t C det ˛ D t 2 tr ˛ t C 1:
Here, tr ˛ is the trace of ˛ 2 SL.2I Z/. Thus we have det 1 D 2 tr ˛ and get Z ˚ Z if ˛ D id; c0N Z IH 2 .M˛/ Š 8 if tr ˛ D 2 and ˛ ¤ id; <0 if tr ˛ ¤ 2: c0N 2 : Since IH 0 .c˚.T // is free, the sequence above reduces to a split short exact sequence c0N c0N 2 0 coker 1 IH 1 .M/ IH 0 .c˚.T // 0: c0N Z Hence IH 1 .M/ Š ˚ coker 1. In this final case our interest reduces to a cokernel calculation of the 2 2-matrix 1 D id ˛. The image im 1 Z ˚ Z is of the form nZ ˚ mZ; n; m 2 Z and so every group Z ˚ Z=nZ ˚ Z=mZ can c0N be realized as IH 1 .M/. In particular a torsion intersection homology group may appear. Using det 1 D det.id ˛/ D 2 tr ˛ as above, we immediately see that Z ˚ Z if ˛ D id; coker 1 Š 0 if tr ˛ D 1; 3: Summarizing all these results we get: 2 PROPOSITION 3.2. Let M˛ be the mapping torus over the open cone c˚.T / of Š a torus glued via ˛ W T 2 ! T 2. Then its intersection homology groups are
ctN Z if k D 0; 1; IH .M˛/Š k 0 if k 2;
Z if k D 0; 8Z ˚ Z ˚ Z if k D 1 and ˛ D id ˆZ ˆ if k D 1 and tr ˛ D 1; 3; ˆZ c0N ˆ ˚ coker.id ˛/ if k D 1 (in general); IH k .M˛/ŠˆZ <ˆ if k D 2; tr ˛ D 2 and ˛ ¤ id; Z ˚ Z if k D 2 and ˛ D id; ˆ ˆ0 if k D 2 and tr ˛ ¤ 2; ˆ ˆ0 if k 3: ˆ :ˆ INTERSECTIONHOMOLOGYWANGSEQUENCE 257
EXAMPLE 3.3. Let X be the fibre bundle over S 1 with fibre ˙T 2 and mon- odromy ˛ 2 SL2.Z/. We assume ˙˛ to be orientation preserving, i.e., the sus- 1 1 pension points are fixed under it. The space X has a filtration X4 X1 DS tS 2 1 and our situation applies. Let ˛ be 1 0 . First, using the ordinary Wang sequence, we compute the homology of the total space E of the fibre bundle T 2 ! E ! S 1 with the same monodromy:
@ 2 id ˛ 2 @ 2 0 ! H3.E/ ! H2.T / ! H2.T / ! H2.E/ ! H1.T / id ˛ 2 @ 2 id ˛ 2 ! H1.T / ! H1.E/ ! H0.T / ! H0.T / ! H0.E/ ! 0:
In degrees 2 and 0, the map ˛ is the identity, so we substitute zeros for id ˛ to see that H3.E/ Š Z Š H0.E/. In degree 1, the map ˛ is just the matrix ˛. Using im @2 D ker.id ˛/ Š Z, we get the sequence
2 @ 0 ! H2.T / ! H2.E/ ! Z ! 0; which yields H2.E/ Š Z ˚ Z and
0 ! coker.id ˛/ ! H1.E/ ! Z ! 0:
It follows by coker.id ˛/ Š Z that H1.E/ Š Z ˚ Z. Let us now compute the intersection homology groups of X via the Mayer–Vietoris sequence. The neighbourhoods of the two S 1 are of the desired form, i.e., mapping tori over c˚.T 2/ and their intersection is a fibre bundle over S 1 with fibre T 2 R, so that the intersection homology groups are just H.E/ from above. Looking at the exact sequence cpN cpN cpN cpN 0 IH 4 .X / ! IH 3 .E/ !IH 3 .M˛/˚IH 3 .M˛/! ; jj jj 0 0 cpN Z we see that IH 4 .X / Š H3.E/ Š . In degree 0 the inclusion of E induces an injection on homology, i.e.,
cpN cpN cpN 0 H0.E/ ! IH 0 .M˛/ ˚ IH 0 .M˛/ ! IH 0 .X / ! 0; cpN Z and IH 0 .X / Š as it should be. Turning to the interesting degrees, we look at ctN ctN pN D .0; 1;:::/ first. Due to IH 2 .M˛/ D 0, there is IH 2 .X / Š ker .i ˚ i/1. ctN Finally IH 1 .X / is isomorphic to the cokernel of the inclusion .i ˚ i/1 W ctN ctN Z Z H1.E/ ! IH 1 .M˛/˚IH 1 .M˛/ Š ˚ , which is the diagonal map; hence ctN Z ctN Z IH 1 .X / Š and IH 2 .X / Š ker .i ˚i/1 Š . Similarly, for pN D .0; 0;:::/ c0N Z c0N Z Z we have IH 3 .X / Š ker .i ˚ i/2 Š ; with IH 1 .M˛/ Š ˚ it follows c0N Z Z c0N Z IH 1 .X / Š ˚ . And IH 2 .X / Š coker .i ˚ i/2 Š . Because all the 258 FILIPP LEVIKOV groups are free, the duality is already seen working with integral coefficients, especially c0N ctN Z Z IH 3 .X / Š IH 1 .X / Š ˚ ; c0N ctN Z IH 1 .X / Š IH 3 .X / Š :
4. Some more advanced tools 4A. Normally nonsingular maps. Intersection homology is not a functor on the full subcategory of Top consisting of pseudomanifolds, since induced maps do not exist in general. However, on the category of topological pseudomanifolds and normally nonsingular maps, intersection homology is a bivariant theory in the sense of [FM81]. This fact is often suppressed. Since most of the maps which we will encounter are normally nonsingular, we recall in this section how induced maps are constructed. See particularly [GM83, 5.4].
DEFINITION 4.1. A map f W Y ! X between two pseudomanifolds is called normally nonsingular (nns) of relative dimension c D c1 c2 if it is a com- position of a nns inclusion of dimension c1 — meanining that Y is sitting in a c1-dimensional tubular neighbourhood in the target — and a nns projection, i.e., a bundle projection with c2-dimensional manifold fibre.
EXAMPLE 4.2. An open inclusion U Œ X is normally nonsingular. The inclu- sion of the fibre F D b F Œ E, where E is fibred over a manifold is normally nonsingular. The projection Rn X ! X is normally nonsingular.
PROPOSITION 4.3 [GM83, 5.4.1, 5.4.2]. Let f W Y ! X be normally non- singular of codimension c. Then there are isomorphisms
! f ICpN .X / Š ICpN .Y /Œc and f ICpN .X / Š ICpN .Y /: DEFINITION 4.4. If f W Y ! X is a proper normally nonsingular map of codimension c, we have induced homomorphisms pN pN pN pN f W IH k .Y / ! IH k .X / and f W IH k .X / ! IH k c.Y /: They are constructed by considering the adjunction morphisms of the adjoint ! pairs .Rf!;f / and .f ; Rf/,
! Rf!f ICpN .X / ! ICpN .X / and ICpN .X / ! Rff ICpN .X /; by combining them with the proposition above and by finally applying hyper- cohomology. We will also need induced maps on intersection homology with compact sup- ports, which are not discussed in [GM83]. The above construction of Goresky cpN and MacPherson works equally well for IH . If f is not proper, the map INTERSECTIONHOMOLOGYWANGSEQUENCE 259
cpN cpN f still exists. For the case of compact supports, f! W IH k .Y / ! IH k .X / can be constructed in the same manner. These different maps are listed in the following table — note that only f and f for proper f are explicitly men- tioned in [GM83, 5.4].
f proper f not proper pN pN f W IH k .Y / ! IH k .X / cpN cpN cpN cpN f! W IH k .Y / ! IH k .X / f! W IH k .Y / ! IH k .X / pN pN pN pN f W IH k .X / ! IH k c.Y / f W IH k .X / ! IH k c.Y / ! cpN cpN f W IH k .X / ! IH k c.Y / 4B. Behaviour under stratified maps. Computing intersection homology in- variants of one space out of the invariants of the other often relies on the de- composition formula of S. Cappell and J. Shaneson [CS91]. Since we will need it in the application below, we briefly recall it in this section. Let f W X n ! Y m be a stratified map between closed, oriented Whitney stratified sets of even relative dimension 2t D n m, Y having only even- b codimensional strata. Let S 2 Dc .X / be a self-dual complex. Denote by V the 2 set of components of pure strata of Y . For each y 2 Vy 2 V, define