WITT SPACES: A GEOMETRIC CYCLE THEORY FOR KO-HOMOLOGY AT ODD PRIMES by Paul Howard Siegel B.S., Massachusetts Ind'itute of Teclnology (1975) SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY at the INSTITUTE OF TECHNOLOGY MASSACHUSETTS May 4, 1979 Signature of Author........ Department of_ Mathematics, May 4, 1979 Certified by......... Thesis Supervisor .. . Accepted by.. Chairman, Departmental Committee on Graduate Students ARCHIVES OF %&Y:'LcOY JUN .i 1979 LIBRARIES 2 WITT SPACES: A GEOMETRIC CYCLE THEORY FOR KO-HOMOLOGY AT ODD PRIMES by Paul Howard Siegel Submitted to the Department of Mathematics on May 4, 1979 in partial fulfillment of the requirements for the Degree of Doctor of Philosophy. ABSTRACT We solve a problem posed by Dennis Sullivan: to construct a class of piecewise-linear cycles with signature which represent the connective version of KO homology at odd primes, ko 0 2[-]. The construction is one of the first applications of the recent intersection homology theory of Goresky and MacPherson. It demonstrates the power of the theory in the study of the geometry of spaces with singularities. We also present what appears to be the first direct adaptation of classical surgery (spherical modification) to the setting of singular spaces. Specifically, we consider a class of stratified p.l. pseudomanifolds which we call Witt spaces, whose rational intersection homology groups satisfy a Poincare duality theorem. To each Witt space X, we associate a p.l. invariant w(X) taking values in W(Q), the Witt ring of the rationals. This invariant generalizes the signature of manifolds, and it possesses many of the same beautiful properties: cobordism invariance, additivity, and a product formula. Sullivan has shown how to use such an invariant to define a canonical orgientation in ko* ® 2[ ], The orientations induce a natural transformation p of homology theories, from Witt space bordism to ko, 0 Z[\]. To complete the construction, we use rational surgery to prove that the cobordism class of a Witt space X is determined by w(X), and we obtain an explicit description of the cobordism groups of Witt spaces. The only nontrivial groups occur in dimensions 4k, and for k > 0, they are precisely W(Q). The structure of W(Q) is known, and we conclude that y is an equivalence at odd primes. Thesis Supervisor: Title: Edward Y. Miller Associate Professor of Mathematics 3 TABLE OF CONTENTS Abstract............. .. ....... . 2 .. .. Introduction......... Chapter I: Chapter II: Chapter III: 4 Geometric Preliminaries............. Algebraic Preliminaries............ 12 33 Witt Spaces and Generalized Poincar6 Duality.................. 43 Chapter V: Chapter VI: Appendix... References. Properties of the Witt Class w(X).. Rational Surgery on Witt Spaces..... 65 85 Witt Space Bordism Theory.......... 112 .. . . ...0................ ......... 121 ... . ......................... . - Chapter IV: 131 Figures........... 134 Acknowledgements.. .137 Biographical Note. 139 4 WITT SPACES: A GEOMETRIC CYCLE THEORY FOR KO-HOMOLOGY AT ODD PRIMES Introduction In this thesis, we construct a geometric cycle theory with signature, primes, representing connected KO-homology at odd . ko,* 0[] The construction is one of the first applications of the intersection homology theory of Goresky and MacPherson. We also present what appears to be the first direct adaptation of classical surqery modification) (spherical to the setting of "manifolds with singularity." To put the result in perspective, we give some of the historical background to the problem before describing the construction. Dennis Sullivan Floyd type theorem the signature, Theorem [9] relating [22] discovtred the Conner- smooth oriented bordism, and connected KO-theory at odd primes: (Sullivan): For compact p.l. pairs, there is a canonical isomorphism (equivalence) of homology theories: (XA) ® S Q, Z[2] ko, (X,A) 02Z [] , So Q, (pt) where the theories are regarded as 2Z/42Z-graded. 5 He found (1) two interesting applications: A geometric description of connected KO horhology obtained by geometrically singularities at as a bordism theory with join- ko* 0 Z [3], odd primes, like (at least) [See generators of smooth bordism. "killing" [8], [6], [23]]. There is a commutative diagram: so SSO i J * (X, A) p 0s Q * (X,.0A) Q* in which J, k*(X,A) 0 2Z[ 1] Z (PO) is the bordism theory with singularities, is the natural inclusion (gotten by regarding a smooth manifold as a cycle in the bordism theory), natural projection, and i p the is gives an equiv1,ence of theories p [8]. (2) A construction of a canonical KO* o 2Z [%] block bundle over a finite for a p.l. a canonical particular, KO* manifolds via Alexander duality 0 [%] [22]. complex, orientation orientation and, in for p.l. The techniques used for this construction generalize to give a more general result. If the collection of cycles of a geometric homology 6 theory is closed under the operations of taking cartesian product and transversal intersections with p.l. manifolds, and if it possesses a "nice" signature invariant extending then the cycles have canonical the classical signature, KO* 0 2[k] orientations. Note that among the properties of a "nice" signature are cobordism invariance, additivity, and a product formula with respect to closed p.l. manifolds. For further details see Chapter VI and the Appendix. A natural synthesis of the themes of these two applications would be achieved by solving the following Problem: Construct a geometric cycle theory for built from a class of p.l. cycles invariant, induced by the orientations - ko, * ] with "nice" signature (at odd primes) such that the equivalence : is 6' M[ ko* 0 2Z described in (2) above. Goresky and MacPherson were also interested in finding a class of spaces with cobordism invariant signature for the purpose of extending the Hirzebruch L-class to the setting of "manifolds with singularity." Their investigation of cycle intersection phenomena in stratified spaces resulted in their beautiful intersection homology theory proved that for a stratified p.l. pseudomanifold [11]. X They of 7 with only even codimension 4k dimension intersection homology group H k (XA) strata, the and is self-dual, they proceeded to define a cobordism invariant signature and L-class for the collection of spaces with only even strata. codimension This paper begins where [11] left off, and makes use of intersection homology theory to solve the problem stated above. Techniques for computing the cobordism groups of stratified spaces with even codimension strata are not currently available, so we enlarge the class of spaces to permit certain odd codimension strata. Specifically, impose the following local homological If condition: L 0. There are three observations which confirm definition is link is the link of an odd codimension H m (L;V) = stratum, then intersection we that this reasonable. Atiyah in Observation 1: [4] studied examples of smooth fibrations M 29 X N2k with sign (X) / signature of X. 0, where sign (X) denotes the classical Using a theorem of Putz, we may regard 8 fibre bundle as being p.l., and the mapping cylinder the of the fibration provides a stratified p.l. cobordism of X The cobordism has odd codimension stratum to zero. with link If the extended signature is to be cobordism M. invariant, we need a general condition on sign that H N (X) (M;Q) = 0. = 0. Atiyah's For a manifold M which ensures analysis provides M, coincides with ordinary homology, one: intersection homology so we get the same link condition as above. Observation 2: additive. We want the extended signature to be An appealing geometric proof of additivity follows immediately from cobordism invariance if the "pinch cobordism" 1.). is a permissible cobordism is the only odd codimension v The pinch point Its link singularity introduced. carried out in Chapter IV, H Observation 3: (L;) = is a suspension, L A geoiLctric calculation, with suspension points identified. a fortiori, (see Figure shows that H (L;2Z) = 0; 0. An intersection homology spectral sequence argument proves that H (X2k;Q) condition is satisfied in X. is self-dual if the link This result is proved in Chapter III. Using Observation 3, we associate to each permissible 9 stratified p.l. pseudomanifold satisfying X4k space Hm the (X;V). X4k link condition), (that is, to each the inner product The inner product space is determined up to permissible cobordism by its equivalence class in the Witt group of the rationals. class dim the Witt class of (X) / 0 incorporates (mod 4), X, we set = 0. a "nice" signature in the Applying Sullivan's result, in We call this equivalence and denote it w(X) ko* 0 Z[1] W(O), by w(X) . This p.l. If invariant sense used above. we obtain canonical orientations for permissible cycles. It remains to compute the cobordism groups. applying a technique of rational surgery, By adapted from classical spherical modification to the setting of stratified spaces and intersection homology, we show that the Witt class w(X) determines the cobordism class of X. We also construct geometrical representatives in positive dimensions q E 0 Therefore, (mod 4) for each class Q Witt . and denote the bordism theory These results can be rephrased as: Theorem: W(). we give the collection of permissible spaces the name Witt spaces, based on them by in There are isomorphisms 10 QWitt(pt) =Z W(), Q Witt(pt) q QWitt(Pt) q q > 0 q / = 0 0 0 (mod 4) q- and (mod 4) determined by the Witt class invariant. ko, 0 2Z [%] of homology theories: p transformation induce a natural orientations Witt Witt which reduces to the signature groups when sign q 0 : @ ko > 7 2Z[%] . The homomorphism on coefficient (mod 4): Witt q (pt) q The group structure of c -Z W(V) is Z[%] classical ko (pt) 0 2[%] [13]. There q is a canonical isomorphism W(2) where 4. T @ T, is a direct sum of cyclic 2-groups of order 2 or The composition cyclic summand gotten by projection onto the infinite 11 Witt Q q (pt) W(Q) -1 Z, coincides with the signature homomorphism above. Tensoring with Theorem: 2[%2], we conclude: The natural transformation Witt Witt 0 2Z [12-- ko* 0 2 [%; is an equivalence of homology theories, signature homomorphism coefficient (tensored with reducing to the Z[6]) on groups. Finally, we note here that John Morgan has also extended the signature to a class of stratified whose cobordism groups he has computed. spaces These spaces give a cycle theory for the homology theory ass.ciated to G/PL. He uses different techniques from those in this thesis, and thoroughly studies the 2-torsion in his theory. We gratefully acknowledge some helpful remarks regarding 2-torsion which he provided early on during the preparation of this thesis. 12 CHAPTER I: homology theory and the theory of bordism from intersection All with singularities. Stratified p.l. E space 6 are taken for the most [11]. Definition 1.1.: manifold are piecewise pseudomanifolds. The contents of sections 1 part from spaces topological stated. unless otherwise Section 1: and results definitions chapter we present In this linear Preliminaries Geometric X A (closed) n-dimensional (p.l.) pseudo- is a compact space containing a closed subsatisfying: (1) dim (E (2) X - E ) < n - 2 xis an n-dimensional oriented (p.l.) manifold (3) C(X -E) Definition 1.2.: = X. A stratification of a pseudomanifold is a filtration by closed subspaces X=xX n such that xXn-1 for each point n-2 p E X x n-3 -X i , ....2X 1 D X 0 Xi, there is a X 13 filtered compact p.l. space: V = Vn D Vn- and a mapping V. J x B i in X.. to V V (p.1.) x which, B i is the p.1. (interior point of X If -*X for each j, takes homeomorphically to a neighborhood of Here, I B x = a point i D'..D is not empty, i-ball and p p corresponds B ). it is a manifold of dimension i, called the i-dimensional stratum of the stratification. We also consider relative objects: Definition 1.3.: An n-dimensional boundary is a compact pair (1) A is singular set (2) dim E an (n-1)-dimensional A neighborhood [0,1) = N. (p.l.) pseudomanifold with There is a closed subspace X - such that n-manifold which is dense ing satisfying the properties: ZA* < n - 2 (3) (X,A) pseudomrnifold with in of A homeomorphism S(EA in U A) with is an oriented X. is collared in N GE E X; X i.e., there is and an orientation 0 : A x x [0,1] -N a closed preservsuch that 14 Here is the relative definition of stratification. A stratification of an n-dimensional Definition 1.4.: (X,A) pseudomanifold with boundary is a filtration by closed subspaces: n stratifies A, and the obvious sense), (A A [0,1]) _x Section 2: = 0 1 n-3 A given by stratify = X - A X ( A (in the respect the collaring filtrations By the last X. in A - A. 1 X the x 2 of the filtration such that of n- n-l we mean condition, X. r N. Piecewise-linear chains and perversity Assume throughout that X is an n-dimensional pseudomanifold with fixed stratification. chains is complexes of X the complex of C* (X), Definition 2.1.: the direct T C,(X), T C e C (X), triangulations ranging over all compatible with the p.1. For structure. the support of C, denoted the union of the closures of the i-simplices coefficient in c is geometric under refinement of the simplicial limit with .. not zero. As a p.l. a JC, is whose subspace of X, 15 D E A perversity Definition 2.2.: = p (P2 '''''n) + 1. If Y C X such that is i (p,i) ) < i n-c 2) IDCI C (X) is (p,i) is We will the condition in 2.2. < i and a subspace dim (Y) < i and c > 2. is the subgroup of C C (X) such that allowable (p,i-l) allowable. Remark 2.4.: (Y) or on occasion. a perversity, p for all consisting of those chains IC p, pc+l =c c- - Definition 2.3.: 1) and allowable if c + p - of integers a sequence for an integer and is called is p2 = 0 p(c) We also write dim (Y 0 X dim |DI. has a well defined support, C*(X) a chain Therefore, is invariant under refinement. JCJ frequently use an ec-uivalent Y Namely, dim (Y r) Xn c) is < i (p,i) - form of allowable if c + p for all c > 2. Proof of equivalence: dim (Y so the second condition Since X n-c C Xn-c, we have nx )n-c < dim (Y () X n-c) implies the first. On the other 16 hand, (Y r dim Xn max = j=O,...,n-c j, For each Pc+j c + {dim (Y nX .n-c-j j, so the second condition dim (Y ) Xnc) < i + p c+ (c+j) - < i + c + p . implies, Therefore, dim The H(X) , (Y C i th intersection Xn-c < i - c + pc homology group of perversity is the i th homology group of the chain pP complex CP (X). There are homomorphisms for each perversity a- : H*(X) p -+ HP W- : HP (X) p -+. H,(X) n-* p: (X) 0 [X] : H*(X) -+Hn-* . which factor the classical cap product homomorphisms: These are roughly defined as follows. The map w- p is 17 T is p by sending the elementary cochain a where , an is a chain supported on the T, to the p.1. n-i i-simplex in assume p on the cochain level a- Define a triangulation. a-, For the map induced by chain level inclusion. i a dual to in class in fundamental Section X, corresponding to the appropriate Transversality and Intersection products. 3: Definition 3.1.: Suppose dimensionally transverse is [11], Lemma 3.2.: C A @D, D E and j - n = 2. are said to be CW(X) J C fi D) (written i + ICI () IDI if allowable. (r2,k) In p + q = r and a chain C E C (X) A chain H n-i(D(a),DD(a)). the following lemma is Let be as in 3.1. C, D then the intersection chain proved: If C Ax D, C f) D E Cr DC A D, is defined and a(C 0 D) = 3C t) D + (-1)n-iC 0 D. is The transversality lemma, which we use frequently, proved in [16]. Lemma 3.3.: Suppose p + q = r, A e C?(X) 1 and B E C9(X). J 18 Suppose K : Then there is a p.2 . isotopy DA A B. X x [0,1] -- +X such that (1) K(x,0) = x, (2) K(x,t) = x for all (3) K(x,t) e X. - (4) K(IAIl) = K for all X (JAI) dimensionally x X E xE all t E whenever x E X - supports K1 j|Aj, transverse to B. denotes the induced map on Corol lary 3. 4.: p + q = r. and 3C' Suppose Suppose C e DC A D. such that E e C i+l (X) C Here K C, (X). (X), D E C 1 = C' X (A), which is J (X), and Then there exists DE [0,1] - C, C' e Cw(X) C' A D, and C. = Now we come to the homology theory its Theorem 3.5.: intersection (1) result which gives intersection name. Suppose p + q < r. There pairing: S:HP (X) i x H0 (X) 3 -- + H i+j-n defined by: [C] f) [D] = [C r) D] (X) is a unique 19 for every dimensionally transverse pair of cycles C e CP(X) I (2) D E and If C (X). I f E H9(X), and 3 (Ol,2,...,n-2), then E EH p + c + s < t = C 0 T1 r) (3) The (T, ) p + q = r, That p p = A n W-(C) p = <A,w- (C)>= p and c : H Section 4: w fn is, if A, = A 0 [X] a-(A U B) p <,) T) B E H* (X), C H E is -- + 2Z x-(A) 0 p (cap nroduct) a-(B) q O(a-(A) n C) q r E(w-(a -(A) r p () C)). the Kronecker pairing H* H* x --- 2Z is the usual augmentation. The basic sets QP and the intersection pairing theorem. For any compatible triangulation perversity (X) then w-a-(A) Here (c r = T (n-j) (n-i) (1) and if intersection pairings are compatible with the cup and cap product. and r = HS(X), k WE T of X and p, Goresky and MacPherson define the basic 20 subsets lQ}. and Q 1 C. 1+1 is a full i-dimensional Q. T, the first barycentric subdivision of T', subcomplex of 0 < i < n, For for all (modulo i n). The relation of these sets to intersection homology is given by the lemmas below. Lemma 4.1.: is an isomorphism There S:Im(H 1 (Q1 ) -- >H HP (X) (Q?+) 111 induced by inclusion of cycles. Lemma 4.2.: If > 1 i and p + q = t = (0,1,2, ...,n-2), then there are canonical simplex preserving deformation retractions: X i+1 where Tn-2 Lemma 4.3.: is the If n-2 i + H(X) i - QP n-i+l - n-2 skeleton of j=n, ~Im(Hj(Q and ) T. p + q=t, H3 (Q)), 5 21 N canonically. Lemma 4.1 implies that the groups H (X) are finitely X, since generated and independent of the stratification of any two stratifications have a common subordinate triangulation. Lemmas 4.2 and 4.3, algebra, yield the Theorem 4.4.: Intersection E Let augmentation, where p + q = t, and combined with some elementary : H t 0 (X) -+ Pairing Theorem. Z be the usual (0,l,2, ...,n-2). = j i + If = n then the augmented intersection pairing: HP(X) 1 x H9(X) 3 +Ht 0 'E X) is nondegenerate if these groups are tensored with the Section 5: 0. sets The basic Whereas the basic sets proving the intersection different basic applications. if p < q, R Q are perfectly pairing theorem, sets which are better suited to there are adapted The main drawback of the sets then it is The basis sets on the other hand; generally not true that R W(. and complexes to other Q is that QC QC . rationals behave well under inclusion relations, they play an important role in Chapter III. 22 They are defined as follows. T be a triangulation of centric subdivision X R Define Tn I D = T, IT n-2 I = IT n -3 simplexes which are (p,i) ' .Z.jT T' to be the subcomp]ex of 1 with first bary- Consider the stratification of T'. skeleta of X given by the X 0 . Let consisting of all allowable with respect to this stratification. W Let those be the subgroup of simplicial properties Property 5.1.: The inclusion i of the complexes : W Im (H.1 (R?) 1 If -- p < H.(RP))~ 1 1+1 ~, W R are: C 1 (X) i an isomorphism on homology, Property 5.2.: consisting of i-chains with boundary supported on Two fundamental induces C T(R ) so H? (X). 1 there is an inclusion of chain complexes: WPC0 W *. 23 Intersection homology with coefficients and Section 6: the G Let Q/Z intersection pairing. be an abelian group. the subgroup of chains C e C (p,i) allowable and (X) 0 G JaCj is This complex may differ from CE(X;G) Define 1 such that (p,i-1) to be jCj allowable. is (Note: 0 G). CP(X) One can check that: 1 ~Im(H 1 (Q?; G) -+Hi +?; G) 1+ ) HP(X;G) and HP(X;G) 1 Suppose there is and p+q=r D E C J a pairing i+j in C ( D holds. [11] E C A D, H.i (RP 1+1 ;G)) -- G 0 G' If -n9=. (X,G'), with arguments chain ~Im(H.i (R?;G) 1 DC A D, -+G", and suppose CY(X;G) CE and C can be modified to yield A aD, an intersection we get intersection products with coefficients. If p + q < r, there is an intersection pairing: 0 : then the Cr (X;G"), and the Lefschetz boundary formula By applying the transversality lemma, Theorem 6.1.: and H1 (X; G) 1 x H9 (X;G') 3 -- H + 1 j-n (X;G") 24 such that [C] 0 [D] = [C 0 transverse pair of cycles D] for every dimensionally C E CP(X;G) D e C and 1 let In particular, G' = 2Z, G = Q/2Z, J (X;G'). G" = Q/2Z. and Consider the pairing induced by the abelian group structure q/zZ: of '2/2Z ® 2Z -+/2Z. The group Q/2Z is divisible, so the universal coefficient theorem for cohomology implies, finite for A a complex, H i (A; Q/2Z) Hom (H (A) , /2Z), with the isomorphism induced by chain level evaluation. Some elementary algebra, and an argument very much like that used to prove Theorem 4.4 yields the Q/2Z intersection pairing theorem: Theorem 6.2.: is Let i + j = n a nonsingular augmented H ?(X;J/2Z) x and let intersection H.(X;2Z) -- + p + q = t. There pairing: H (X;Q/2Z) :,.Q/2Z. I 25 Section 7: In Bordism with singularities. Sullivan introduced the concept of a bordism [23], theory based upon a class of cycles and homologies with Let or "bordism with singularities." singularities, specified rn Akin pairs. be a sequence of classes of compact p.l. n e 2Z, , gave a list [2] of axioms which ensure that is rich enough to support a homology theory {yn} defined for compact p.l. (T,T0 ), built upon cobordisms of continuous maps where (X,X0 ) necessities, E tnL, some and produces n. and f : (X,X0 ) -- We present slightly criteria for {Xn} (T,T O) The list represents the bare Z/2Z-bordism theories unoriented). oriented). (T,T ), pairs (that is, stronger axioms providing to support a Z-bordism theory (that is, In chapter VI, we give an interesting example of such a theory. First, recall the following standard definitions: Definition 7.1.: X C X X C N that (X,X ) is collared if C X, and a p.I. h(X 0 x 0) special be a compact p.l. pair. there is a closed neighborhood N, homeomorphism such = X0 C N. h (Note definition : X x 1.3 I -* N (3) is a case). Definition 7.2.: totally Let A locally compact polyhedron n-dimensional if X = C9(In X) X _ In-1(X)). is Here 26 Remark 7.3.: sional iff Akin some [2] shows that X is (not necessarily proper) Now, let of X Every simplex of X n Assume . K. n > 0. pairs, , E, is all n. satisfies the following axioms. {yn} Axiom 1: If (XX 0) If (XX0 and YY) 0( (X,X0 ) ,n, then E En Axiom 2: and K face of an n-simplex of (X,#) for Suppose (Y,Y0 ) n-dimen- be a sequence of compact p.1. X E yn Write k), then {fn} [1] totally triangulation (and hence any) satisfies the following condition: a as in . is X, k-skeleton of the intrinsic Ik (X) X0 C X is E.n, then is totally n-dimensional X collared. n Axiom 3: If (X,X0 ) E J' and ' (X,X 0 ) J ~ Hn(XX;) i=j~ (X,X0 ), denoted Axiom 4: If (X,X0) Ey n an orientation for (X,X0 ), is an orientation for by [X,X 0 ]. Here for is a sum of generators of the [X,X 0 ] cyclic summands of infinite An orientation J. some finite Z, Hn (XX0' then X0 then If E .n-l. 3[X,X 0 ] E [X,X 0 ] Hn-(X ;') X 0 , called the orientation induced D* : Hn(XIX0;) b. Hn- 1 (X 0 ;2Z) is the is 27 connecting homomorphism Axiom 5: Axiom 6 (X,X 0 ) E Y If (X x I,X x I UX0 xIn+ then (V,V ) A If Y = CS(X-V) n, (V,Bd V U V0 ) in (X ,X 0 ) . Axiom 7 Vr1Y ) A C. X ,E (X,X0 subpolyhedron, (Bd VBd be a closed a regular neighborhood pair for = CZ(X0 Y and n 0 ), then and n-1 Let (Pasting). X : be p.1. triples withZ X and 0Y disjoint. (X , 0 ) Suppose Let [Xtx 0 i Z [Y y 0 ], X0 -+ n 3 (YY0 [Z,Z 0 ] and iy : and , 6n Z (ZZ0 ) and n-1 Let be orientations. be p.1. -- embeddings which are orientation preserving and reversing, respectively, with respect to the restrictions of the induced orientations on Y Furthermore, suppose X0 Y = orientation XC W and (X U [W,W0 ] Y - W ' 1 (Z) U Y (X,i,,(Z )), (W,W0 ) = Then, Y (Z) UY i X and X , and , X0 ' Z , Let (Cutting): (Y,Bd VUY 0 ) C X0 X1, Y D Y in ordinary homology. (Yl Z 1i = ZO 1) (Z) n E n-1 and there is an such that the natural embeddings are orientation preserving. Now we define the notions of oriented cobordism and 28 oriented map. Definition 7.4.: [X,X 0 ] orientations (X,X0 ) Let [Y,Y 0 ], and respectively. (Y,Y ) and (X,X0 ) An oriented cobordism between an oriented pair yn, with be in (Y,Y ) and n+ , with orientation (W,W0 ) [W,W 0] satisfying 1) i are p.1. There W : Y -+ i embeddings W : X - and preserving and which are orientation reversing respectively with respect to the induced orientation on * W0 2) i 3) There is a compact subspace r) i a) and (Y) = i (X) W i (Y) n W WO - i = i(X) U (W ,i (X 0) U b) An oriented map of such that W x) (Y W1 U i C iy (Y) Y )) EIn (X,X0 ), with its orientation [X,X ], is a continuous map (T,T0 ) is denote this i W ' (X) f : (XX a compact p.l. pair with 0) -- T0 C T {f,[X,X 0 ]1, and refer to ' 0 )T, where closed. {f,[X,X 0 ]} singular 3-cycle. Given oriented maps (TT0),an oriented cobordism g between : (Y,Y f and 0) ' f : (X,X0 (T,T 0 ) and We as a is 29 (W,W ) is an oriented cobordism (Y,Y ), together with a continuous map F o such that and f = x F o i= F as is easily : (W,W) (T,T ) (T,T0 ). checked. Under disjoint union, group structure: and *(T,TOr is an equival- Denote by the set of equivalence classes of oriented maps into and g. y Cobordism of oriented maps into ence relation, (X,X0 ) between Q n T,T 0 ) g {f,[X,X 0]} this has an abelian $ the zero element is (4,$) ---. (T,T0 ), : {f, [X,X 0 ]} + {f,-[X,X 0 ]} = 0, using the product ~I, xIUx0 X]}. {f oa 7 cobordism Define the connecting homomorphism: n Q 0 (T, T0 n-l(T 0 by 9n fi[XiX 0 where define Finally, F* if : Qn (S'S0 F Q* * (T,T ) : (S,S0 ) Qn(T,T ) F*({f,[XX0]}) Let I ffIX0 is the induced orientation D*[X,X 0 ] before. = denote the functor = from on X 0 , as is continuous, by {F o f,[X,X 0]' compact p.1. pairs to graded abelian groups, along with the connecting morphisms 30 n Theorem 7.5.: Q, is a homology theory. The proof proceeds as in Proof: simply keep track of [2]; orientations. 3 called an oriented bordism Such a homology theory is theory with singularities. Definition 7.6.: of compact p.l. n > for spaces, be a sequence of classes {&n} a&= Let We say 0. d& is a class of singularities if: (1) 4 (2) If L (3) If L E C" n, all E E n ,n , then L is totall r n-dimensional J some J > 1. L 74 $, then and An orientation for n (L;2Z) is denoted L = 7Z, [L], as before. (4) So For n > 0, is the 0-sphere. L E iff .6n S * L E n+ , where This is closure unier suspension and desuspensi n. L n = x,X0) ' (X0) is a totally Let n-dime]=sional compact pair, x C orientable, X - X0 ' with x0 k (x; X) E .6} . X Here collared, kk (x; X) such that for denotes the link 31 x of in Then X. is readily verified satisfies axioms 1 - =P.& [2]. 7, as The associated homology theory called the oriented bordism theory based on the is * J, class of singularities. o7 Bordism theory with 8: Section 2/kZ coefficients. QT. Suppose we have an oriented bordism theory in an abelian Coefficients * ;G). The [8], yield- as described in geometrically into the theory, ing the theory can be introduced G group G = 2Z/k2Z case admits a particularly simple and convenient description first ;Z/kZ) ;,O [18] in the setting of . by Morgan and Sullivan in realized In general, a Z/kE-cycle of dimension n is obtained , equipped with an oriented (X,X0 ) En from an oriented pair k embedding Y0 X 0 of k copies of Y onto XO' 1= Y where n-1, ori:entation by identifying the copies of preserving manner. is defined similarly, into A ,n-. by embedding k copies of (Y0 Y 1 ). The boundary consists of k - -{interior 1=1 along the images of of image of Y1 . X0 in an 2Z/kE-cycle with boundary X0 , and identifying the copies of (YOY1 ) 1 Y Y01' with identifications (Y0 Y 1) Here 32 ( The homology theory built from maps of ;2Z/k2Z) Z/k2Z-cycles, is the bordism theory modulo -naps of 2Z/kZ- cycles with boundary. >f bordism with We need the following properties 2Z/k2Z coefficients, play a role in the discussion Property 8 .1.: is both easily verified. in the Appendix. The homomorphiEm multiplication by k, These facts i : 2Z/k2Z induces a natural -2Z/k-.ZZ, which transformation of theories: Q*( This is Property reduction ;2Z/k2Z) ;2Z/k2Z4) ,( -- Sullivan's "replication" homomorphism. 8.2.: The projection mod k, Q*( ZZ/k'-9I -. 2Z/k2Z, which is induces a natural transfurmation ;2Z/k-9,2Z) -- +Q*,( ; 2Z/k2Z) I 33 Chapter Algebraic Preliminaries. II: This chapter contains a description of the Witt theory of inner product spaces. of classification Section 1: Witt rings. We employ the terminology of is found in Let finitely R be a commutative ring with generated projective 1 : X Another treatment [3]. Definition 1.1.: form [13]. x module over An inner product X -- R unit. 3 on Let X to HomR(X,R) be a R. X which is non-degenerate, two homomorphisms from X is a bilinear i.e., the given by: x i (x00,-) y 0 I-'P(',y 0 ) and are bijective. We call (X,1) an inner product space; there is the obvious notion of isomorphism of inner product spaces over R. The inner product is according as symmetric or skew-symmetric 34 r(x,y) = f(y,x) or In this R = 2Z/p2Z We can associate to (X,?) p el,...,e B = = where 1.2.: has matrix a) B b) If B' (X,r') prime, and R = 2Z. if p. ( . .) 3(e.,e.). :ji J invertible (X, ) has a matrix and only if (e1 , . . = ABAt B B and ,e )), (XI') then , foi some A. 3] We sometimes write I (XS) = Given inner product spaces with the form: <B>. (X , ) , ... , (X 1 . .EDXn') . define the orthogonal sum X 1 1....Xn X. the mratric: (with respect to matrix [13, is be a basis for . invertible Proof: for is necessarily a free R-module (a vector space or Let (X,3)~ -1(y,x). we will be concerned primarily with free abelian group). Lemma = . X thesis, R = C, the cases Then (x,y) to be n,' we 35 f(x e.. .x Clearly (X1 6... 'ye. . .ey n) is 6Xn) (iFyi) an inner product space. may also construct the tensor product X 1. . .QXn is with the unique form X... Xn') (X1 We which satisfying ' n '(x 1 Again, (X1 Q...Xn,') [13, 10]. p. N C X its R. Let X such that orthogonal Oxn' is Definition 1.3.: space over .- N split be a symmetric inner product if there exists is a direct complement. a submodule summand of X, N = NL, and (NL= {x e XJr (x,n) = 0 for n e N}). all Remark 1.4.: A split "metabolic" in inner product space space (X,) is called to us, we have the following convenient description of split Lemma 1.5.: X [3]. In the cases of interest it T i (xi'01 yi i=1 n an inner product space (X,r) is ' 1 Let over R spaces. be a ring such that R is free. Then every inner product (X) is split iff has a basis with respect to which the matrix of 6 has 36 0 I n the form: ,where is I the identity matrix of I nAn rank n. If 2 is a unit, then with respect to some 0 the matrix of basis, a (In Proof: [13, p. 1 is 0) U 13]. In the case of skew-symmetric X (X, ) is symplectic if 6 (x,x) = 0, all E X. Parallel Lemma 1.7.: to Lemma 1.5 If R a field), inner product space over symplectic basis; 0 is: is either a Dedekind domain or a local ring (e.g. S spaces, we concept of being split. have an analogue of the Definition 1.6.: inner product that R is, (e.g., Z) then every symplectic is free, and possesses a a basis such that the matrix of 1 n is (n Proof: [13, p. 7]. Finally, we distill the collection of symmetric inner product spaces over R into the Witt ring W(R) by means 37 of an equivalence relation. Two symmetric inner product spaces Definition 1.8.: X' X belong to the same Witt class, denoted there exist split spaces isomorphic to X' , equivalent to the form a useful S' such that if X @ S is we omit explicit s'). By abuse of terminology, This is and (For convenience, S'. mention of the forms S ~ X', X, we say that a split space is <0). relation, it satisfies the following properties. Lemma 1.9.: X - X' If 1) X 2) X E Y 3) (X , ) E (X,- ) 4) (1> - X Proof: ) Y ~X' 9 Y' ~ E Y' 0 X Y - Y', then: ~ <0) 3 [13, p. 14]. Consequently, W(R) and the collection has attractive structure, of Witt classes, denoted suitable for study in terms of standard algebraic concepts. Theorem 1.10.: W(R) is a commutative ring with unit, using orthogonal sum as addition operation, and tensor 38 product as multiplication operation. U Immediate from 1.9. Proof: Structure of Section 2: We derive the additive In the process, W(). a prime, and W(V). structure of the Witt ring we determine W ZZ/p7Z), for p W(2Z). We begin with the statement of a lemma concerning the "diagonalization" of the matrix of an inner product space. Lemma 2.1.: If R is then every symmetric a local ring in which inner product space possesses an orthogonal basis. basis e 1 ... ,e so that X = (u 1 )G.. .$(un>, Proof: [13, p. That is, (e.,e.) = 0 where the u Now, p (X,) given in (Xr) X a unit, over R possesses a for i are units in j. - Thus R. a nonsingular rational on Lemma 2.1. (X,3) in W(Z/pZ), for following result, is 6]. Anyone who has diagonalized matrix has relied 2 W(V) , we associate elements p prime, by means of the due to Springer and Knebusch. First, 39 if we fix a prime rational number n E 2Z of rationals E 2Z, and u is any non-zero n p = *u, a unit in 2Z () the subring which, when written in reduced denominators prime to Lemma 2.2.: additive p. Fix a prime (That is, p E 2Z. 2Z form, have (2Z-(p)) -I ) There is a unique homomorphism -+ ap : W( W(2Z/P2Z) which maps each generator <pnu> as 0 (mod 2). n E 1 (mod 2) or residue class of Proof: then can be written uniquely as: q q where p u n in 2Z to (u > /(P) .Z () or (Here 0, u according is the = 2Z/p&). U [13, p. 85]. Remark 2.3.: This homomorphism is the second residue class form homomorphism. We can collect p . note that the homomorphisms P for all primes p and thereby obtain an additive homomorphism 40 S:W(M (@)+W (2Z/pZ) p prime Note that we also have i a homomorphism : W (2Z) inner product space as an integral gotten by regarding -W(W a rational one. The next, remarkable theorem is proved in 90] . recall First Definition space is d+ the integer - d- d+ where inner product d~) (resp. diagonal of a diagonalized matrix for the form on 0 1) The sequence .W(2Z) -*W(') -()W(Z/p2Z) -p0 is split exact. W(2Z) ~Z canonically, and the isomorphism is : W (2Z ) -+. Z , given by the signature homomorphism, sign z is the space. p 2) 88- entries on the the number of positive (resp. negative) Theorem 2.5.: p. a standard definition. The signature of a rational 2.4.: [13, 41 the signature of 3) sign 1 to its (XI) e W(2Z) i(X,3) e W(Q)) The exact sequence o Sign is split (that is, by the homomorphism , where --+W(Z) W(Q) signature . which takes sign : W(0) 2Z - is the signature homomorphism. Therefore there is a canonical e D W(2Z/p2z) W(2Z) W() isomorphism: p 2Z ED W (2Z/p2Z) p It remains to compute Lemma 2.6.: The additive structure 1) W(2Z/22Z) Z/22Z, 2) W(2Z/p2Z) 2Z/42Z, p = 3 3) Proof: for all primes of W(2Z/p2Z) p. is: generated by (1>. generated by <) for (mod 4) W(2Z/p2Z) <d> for p = 1 2Z/22Z E and [19, p. Remark 2.7.: W(2Z/pZZ) where 2Z/27Z, with generators d is a non-square in <1> 2Z/p2Z, (mod 4). 3 47]. [19, p. 47] structure is described by: actually shows that the ring 42 1) <d> 0 (d> 2) O>) 0 ) =<' 3) (1> 0 <1) = (d) 0 <d) =<(l> (1> = for p = 2 for p E 3 (mod 4) for p =1 (mod 4). 1> We isolate the following obvious corollary to Theorem 2.5 and Lemma 2.6, as it has important ramifications in Chapter VI. Corollary 2 .8.: W(Q) ~ Z T (canonically), where T is an infinite direct sum of cyclic 2-groups of order 2 or 4. 2 It is also useful to state the following fact about . W(O) Proposition 2.9.: If (Xr3) is a symmetric rational product space such that its equivalence class in the zero element, Proof: Witt's then (X,f) [13, p. W(V) is is split. This follows from the definition of theorem inner W(Q) and 8]. M 43 Chapter III: Witt spaces and generalized Poincar6 duality. Introduction. Section 1: In section 2, we define the class of Witt spaces. Roughly, a Witt space is a stratified p.l. pseudomanifold link homological the following intersection satisfying condition: is the link of an odd cocimension stratum L2' at a point x X, then in H where M = = 0, (0,0,11,2,2,...); that In section pseudomanifold Wn(X;r) (L;Q) 3, for an arbitrary is mn(c) = triangulated c-2 [-2 * If p.l. X, we define an increasing filtration on corresponding to the increasing sequence of perversities: M dim X-2 (0,0,ll,2,2, ... , [ 2 dim X-1 (0,0,1,2,2,3,...,[dim Xn = (0,1,1,2,2,3, dim X-1 2 44 Recall = n(c) [ Notice that c-1 -I. the perversity index of each odd co- dimension stratum gets incremented by 1, in order of decreasing codimension. Associated to the filtration converging to collapses Wn (X;e). is an E1 -spectral sequence, We prove that for a Witt space. That is, the E spectral = 0 sequence for s > 0. The generalized Poincare duality theorem for closed Witt spaces X, stating m that H*(X;Q) is augmented intersection product, section pairing theorem of a Witt space values in W(T), dim X = 4k, X, a p.l. in W (Q). invariant taking its the Witt group of the rationals. k > 0, w(X) then follows from the inter- we introduce the Witt class w(X) is Otherwise, If is the equivalence class of the rational inner product space dim X = 0, under (1.4.4). Finally in section 4, w(X) self-dual H k (X; () merely the rank of we set w (X) in W(Q). H 0 (X;Q) If times 6]> = 0. For discussion and applications of a special case of the Witt class, Section 2: Let see [3], [10] Witt spaces. X be a q-dimensional p.l. pseudomanifold with 45 Then there is a p.l. homeomorphism: Z(x) [1, = q p. X ~ at x. Sd(x)-l * L(x). is a pseudomanifold of dimension L(x) The space kk(x,X) DX. homeomorphism dimension of be the intrinsic d(x) Let [20]. up to p.1. unique kk (x,X), is x, of The link X E X - Let X. collared boundary (possibly empty) - d(x) homeomorphism unique up to p.l. 1, 420 and 424]. as above. manifold X Let Definition 2.1.: We say be a q-dimensional p.l. pseudo- Z(x)/2 for all X E X Remark 2.2.: - DX is X a Witt space (L(x); such that k(x) ) = if 0, E 0 (mod 2). The definition places intersection homological restrictions on the links of odd codimensi*n intrinsic strata only. section The discussion of the spectral sequence in 3 will provide justification. For applications, we must translate the Witt condition in 2.1 into a statement about arbitrary stratifications of X. To this end, we prove a few lemmas which culminate in Proposition 2.6. with m(c) = c-2 Recall that m refers to the perversity 46 Let dimension 22. manifold K L be a closed p.l. pseudomanifold of L = S0 * K Suppose of dimension 2k-l. for some Then H p.l. (L;E) pseudo- =M . Lemma 2.3.: 0k Proof: T The 0-sphere is the discrete set be a triangulation of ci stratification: S S0 = K, and T T2 Z-1 D ZT be the triangulation of let ef be the c7 L {a,b}. Let the associated T 2Z.-30 23 'T 0 ). Let induced by suspension, stratification on L similarly and induced. X (S Note that S *T 2 29i-l .4 *Z= * T 2k-3... T * T0 D 22-3~ 0 is not the stratification asbociated to z e Cm (L) , where perversity Consider a cycle 4. defined with respect to a triangulation S k0 Then, of L Izi ) Sj = $. subordinate to S z is a sum of k-simplices with coefficients. v of L. If If v Izi in the Int(a) C K, E Int(a*T) define of lies Izi. f(a) = T. By interior define or v E of a unique f(a) [20, p. 17], chain which we denote One checks easily that f f(z) f(z) is Choose such that Each vertex a of = C, the barycenter of a. Int(b*T), Extend the map S. where f simplex Int(T) C. linearly K, then to simplices determines a p.1. geometric in C ,(L), supported in is a cycle and, in fact, IKI. ) That is: 47 If(z)I the same homology class as of Izi IzI v structure on Izi x [0,1]. chain H(z DH(z x [0,1]) = f(z) - z. [0,l])j is |H(z x for instance, See, -* L [0,1] x a p.l. that The triangulation and subordinate to the product cell x [0,1]) = v H (v,0) by [11]. and over simplices of [20, p. Appealing again to [0,1]. x (L). the set and extend linearly f(v), = IzI) E IZI H : Define Izi H vertices having as its {(v,0),(v,1)I H(v,1) in can be used to construct a triangulation of [0,1] x z represents f(z) We claim that (rn,Z)-allowable. is in we obtain 17], CP+1(L), with it is easy to see Moreover, (mi, Z+l) -allowable. This proves the claim. Let IWI = a W * be the obvious p.1. If(z)I chain in DW = f(z). and trivial. Therefore was arbitrary, Corollary 2.4.: dimension 2k. manifold Proof: represents f(z) Suppose L in be a closed p.l. L Let 0 L = Si 0 < j as S0 < 2k. r0 in H C with (L), H m(L) (L). is Since z (L) = 0. H we conclude K, where Rewrite z W E Then implying that the homology class of Ck+1 (L) * K for some p.l. pseudo- Then, ~1 pseudomanifold of (Sj1* K) , H (L) = 0. and apply 2. 3. I 48 Corollary 2.5.: Proof: Let H' (L;Q) = C*(L) C*(L;V) Hm(L) Then Hm(L;Q) = 0. manifold, with X Let .4 = Xi, 0 Xi Xi / and boundary Xn.n-2 Xn (Xn be the link of (X-DX) be a stratified p.l. pseudo- (possibly empty) stratification: L(Xi,x) induced by the definition: 0 Q, 0 (. Proposition 2.6.: E be as in 2.4. The result follows from 2.4 and the isomorphism of groups: x L X.1t = $. - I DX Then, 0). '' at _1 X and Let x, xherwhere is a Witt space if and only if H (L(Xix);Q) = for i = q - Proof: k > 1. (2k+l), whenever Suppose There are p.l. homeomorphisms: is Witt. X Sd(x) k(x,X) 9k(x,X) Let j = d(x) - i. L(Xi;x) Then z S 0 ~ * L(Xi;x) S j * L(x) > 0, and * L(x) by [1, p. 423]. 49 Case 1: j - If follows result Case 2: 1 = -1, The converse of stratification X from the fact j - 1 > If 0, Pk(xi;x) then a Witt space. is the result follows follows from the fact that is X so the ~L(x), from 2.5. any p.l. to the intrinsic subordinate stratification. I Section 3: homology The intersection spectral sequence and duality. triangulation r T, and associated stratification 2r+1 < q be the largest integer satisfying q > 2) . pseudomanifold with is a q-dimensional p.l. X Assume U7. Let (assume Assume rational coefficients unless otherwise specified. Definition 3.1.: Let be the perversity defined by: >k m(c) = Pkc(c) 1 < k < q for c < k [-2 for f c-1 c > k = n(c) where c-2 [2] and = 2 < c < q. 50 Since odd. = n(c) m(c) Note that for and m P2r+l k = n. p on the chain complex F There is a filtration [see even, we may assume c Wn(X) 1.5]: (0 (X) F ={W(2r+1)2s(X) W(X) for s < 0 for 0 < s < r for s > r The filtration is induced by theinclusions of complexes: 0c . = w* P2r-1 2r+l Following E 1 -spectral c W* [21, C- W p. 469], 3 C there is p1 a convergent sequence with st = H s~t (FSWr* (X)/Fs-1i W Est and E nO isomorphic (X)) to the bigraded module G (H (X)), associated graded to the filtration F (H (X)) = Im(H Theorem 3.2.: (2 r+l) -2s (X) Assume (1) E (2) E St = 0 Proof: 0 t ott 0( Part (1) Im(H X is a Witt space. (X) --. H for H n(X)). Then t (X)) s > 0. is immediate, since F (H (X) ) = 0. the 51 Part (2) follows from the equality 2s3 Im(H, (3) 0 < s < r We now prove (3). Fix W 2s+3 J E W. t _ s + 1 < j s 2s+1 (X). with < q - 0 < s < for 2s+l J 2s+l (X), for r - 1. j < s + 1 j > q - Note that and s, s. be represented by the cycle H 2s+1(X) Let z ~ (X)) = H 1. - so assume 2s+l H (X) - . We construct a chain w E W P2s+1 (X) j+a such that @w = z - y, where y e W. P2s+3 (3) This will prove (X). and therefore prove the theorem. The chain stratum of dim( IzI is constructed locally to reduce the I z I K X q-(2s+3), where dimension of q-(2s+3) w n (f. Xq-(2s+3) Xq-(2s+3) To begin with, - j = j is the we have: (2s+3) + p2s+1 (2s+3) - s - 2. 52 If, in fact, (4) then z the following stronger inequality holds: dim(IzI 0 Xq-(2s+3)) 92s+3 W. E (X). < j- (2s+3) = j- s - + p2s+3 (2s+3) 3, This follows from the fact that P2s+p(c) = P2s+3(c) y = z In this case, we may set / 2s+3. c if and w 0, and we're = done. Suppose (4) does not hold. The open stratum is the disjoint union of interiors of simplexes dim a = q - where (2s+3). subset of these simplices j - s - (where Let 2. I is a e C, a finite Let C denote for which index set) the set j of in the first barycentric subdivision of Izi n Orient each simplices Int(T ) in is: Int(a) The subchain of T. T' whose closures - in \ Int(a')) {T s - 2 a a intersect = i i}iEI simplices satisfying D Int(T). z T, the non-empty dim(jzJ and consider Xq-(2s+3) consisting of Iny(a) jin m 53 Z. T. 1 1 where v. an d e C s+1(k(u,T')), k(c,T') boundary of the classical dual to Lemma 3.3.: Dv. = 0, for all Proof: i I. E Dz. = 3z - Now, -a (- * 1 Z., I The re fore, / * v v. + Int (c) z as contains By definition of 0. ) Int ( does not contain (5), i. z = z. + 1 (z-z.). 1 (-)s-1. * 3v. 1 I|a(Z-z.i)| in dual(a,T'). (z-z.). =T. Dv. T', is a cycle, we have: z (5) if Decompose in ) Since Fix a denotes the z., Int(T.) if and only however, 10 Int (a) It follows, by taking supports v. = 0. that 2 The canonical simplicial isomorphism of s+l k(a,T)' maps the complex v. to a cycle WT'(9,k(a,T)) V. 1 e C tk(a,T') (kk(a,T)). and We have associated to the restriction of 54 kk(a,T). to (k(cr,T)), v. e W s+1 I Lemma 3.4.: (s+l)-simplices in the is a sum of v. Clearly, Proof: first barycentric subdivision of for each Fix (6) Qk(a,T). Iv | it remains to show only that i. for all av. Since 0, = (m,s+)-allowable, is i. i v The cycle I. E can be written uniquely as v = a(<p0''.. ' s+1-i ' T G<y04'''.'ps+1 Here the simplex in d(p) T' (7) ' spanned by k to be the maximum integer d(p) For such that p> a, ^ = >'Os+l> with non-zero coefficient in simplex Set is the rational coefficient of a(y 0 '..'s+l) = -1 if p does not appear. dim(Iz.I (I Int()) On the other hand, since z j E P2s+1 C. (X), we get the s - 2 + d(y) (6). + 1. J following inequality, valid when codimension of is greater than or equal to 2 in a Then = - define (codim(p,X) > 2): p in X 55 n dim(lzl (8) < P2s+l (codim(-p,X)) Now, implying codim(y,X) d(p) (9) y > a Each < (s+1) r) Int(y)) < dim(jzj Int(p)) j - codim(p,X) + p 2s+(codim(p,X)). = m(codim(y,X)), < 2s+l. Combining since (7) and P > a, (8), we get: + m(codim(P,X)). - codim(p,X) has a unique expression as: P = a * V, where V e 9k (a,T). The cycle v (10) v. that P = a * v, v = Vk then d(P) in a simplex 0 is the maximum integer ,. ' (v 0 '' ps+ <' 0 s+1 > d(y) = dim(|v' Int(v)). Observe that (12) codim(vZk(a,T)) k such with non-zero coefficient in (10). Therefore, (11) .vS+1 >. a( a<p 0 <..'Ps+1 If then has the expression: = codim(p,X). 56 Then (9), and (11), dim( V- I (12) yield the inequality: < ) Int(v)) - (s+l) codim(v,Yk(o,T)) + r(codim(v,Zk(a,T))) This codim(v,Pk(a,T)) > inequality implies that I. 2 is . when (ms+l)-allowable. I By hypothesis, m that Hs+1 (9k(a,T)) x. i s+2 E WM (k(a,T)) X is Witt. = 0. Therefore, wit h Dx. corresponding chain in = v.. 1 w implies there is a chain Let be the x. i WJe define: 9,k(cr,T'). w. = and Proposition 2.6 (-l) j-sT * X w = iE I Note that: Dw. = Lemma 3.5.: Proof: It W a (-1) j D. 1 * x. 1 + T. 1 * ax.. 1 EW Wp2s+1M (X). j+1 suffices to prove W. 1 1 2s+ E W. 3+1 (X , 1 f or all i. 57 The argument proceeds as in 3.4. is Iw I (p2s+l, j+1)-allowable. codim(p,X) 1: Case P E Let star(cy,T), with A counting argument analogous p > a. 3.4 proves: dim(|wj | < ) Int(I)) codim(p,X) Since we show that > 2. Suppose to that in First (j+1) < 2s+1 - codim(p,X) + m(codim(p,X)). = p 2s+l(codim(p,X)) m(codim(p,X)) so Int(p)) dim(jwl 1) Case 2: Suppose dim(|w | n P < Int()) j+l - codim(p,X) < + p2s+l(codim(p,X)). Then a. = dim(|T. * xi| = dim(JT. * v < dim(Iz I Int int(1)) fl I ) Int(W)) ()). (p2s+1,j )-allowable, Since Izi is dim(Iwil n Int(p)) < j - codim(p,X) + p2s+l (codim(p,X)) 58 p Suppose star lw. l|n Then (0,T) . Case 3: Int(P) These three cases prove the assertion that 0. is j+l)-allowable. |3w J is * Xii U ITi * v Next, we must verify that able. Note be as above. Case 1: (13) that I|wji = aTi three cases We treat a. p dim( * xj sInt(y)) ( Ti 11 again. = dim(IT = dim(Iwi 1A < Let Then Suppose > (p2s+ 1 , j)-allow- . (p2s+l, Iw | = j - Int(y)) (codim(yp,X) i* xi(A - Int (p) )-1 1 + 62s+l (codim(p,X)). (14) dim(IT S* Int xj I r) (p) ) Also, = dim(I-u <dim(|zi n j Together, dim(IwI (13) and r (14) Int(p)) - * vi I Int(p)) Int(p)) codim(p,X) + P2s+ 1 (codim(y,X)). imply that < j - codim(p,X),X) + 2s+l (codim(y,,X)). 59 n dim(0|Dwi 3: Case Then P < a. Suppose (p) Int Suppose y ) Case 2: ' < dim(IT1 < dim( zi < j Int(y)) ) Int(p)) - codim(p,X) star(cyT). These three cases prove that 0 Then I|w I + 62s+l(codim(y,X)) Dwi is .,) Int(y) = 0. (P2s+ 1 ,j)-allowable, completing the proof of the lemma. a Let Y CY of E y (X). W. J w ar z - = w. Then Moreover, y is a cycle and it is clear from the construction that dim(IyI n Int(cy)) < j - s - 3. Now repeat the argument for each of the simplices remaining Denote the in C. by {w } . j+l-chains obtained at all of the steps If we set w = and w CreE 60 y = z - Dw we have: dim(IyI1 for all G E Tq-(s+3. Int(a)) < j - Therefore, yE s - (4) 3 holds and W 2s+3 (X). W W. J The theorem is proved. 2 Remark 3.6.: The proof of 3.2 condition on links is satisfied shows that if the Witt only for odd codimension strata of codimension less than or equal to 2k+1, then the spectral sequence collapses partially. Esit = 0 We now use for s > (r-k). 3.2 to prove a proposition which, when combined with the intersection pairing theorem, yields Theorem 3.8, generalized Poincare duality for Witt spaces. We include the rational coefficients for emphasis. 61 Proposition 3.7.: X Let q > 0. be a Witt space, The homomorphism of graded groups, H 1,(X;Q) i, - H (X induced by inclusion of chain complexes, Proof: For q For q = 0 or 1, the result claimed is obvious. > 2, Theorem 3.2 (15) Im(H*(X;Q) is an isomorphism. implies: -,H*(X;Q)) = H (X;O) Therefore: (16) (X;Q) > dim H(X;Q) dimH pairing theorem The intersection (17) dimH Combined, (18) (X;C) 0 i (16) and dim H M (X;Q) i = dimZH q j (17) = 0 for .(X;qD) (I.4.4) < j < q. implies: for 0 < j for 0 < j < q. < q. yield: dim H'(X;D) MO 3 The proposition follows from (15) and (18). I 62 Finally, we get generalized Poincar6 duality for Witt spaces. Let Theorem 3.8.: X There is a be a Witt space. nondegenerate rational pairing: H m(X;O) 21~ for i + j = q; i,j > x Hm(X; J ) Q 0. The pairing is given by augmented intersection product. Proof: Theorem 1.4.4 provides a nondegenerate rational pairing O)x Proposition on the H n(X; () for -+ 3. 7 gives an isomorphism i + j = q; ij > 0 . H (X; (ind iced by inclusion chain level): i. J J (X; J (X; for 0 < j < q. Intersection product of cycles is compatible with this homomorphism by [11]. The theorem follows. a 63 Section 4: Let The Witt class X w(X). be a Witt space of dimension Theorem 3.8 gives a nondegenerate on the vector space H (X;V). 4k, k > 0. symmetric bilinear That is, form is H k(X;Q) a symmetric inner product space. Definition 4.1.: q, w(X), 2k X Let be a Witt space of dimension q > 0. q = 4k, If Hm (1) is k > 0, the Witt class of the equivalence (X;Q) in denoted of the inner product space W(), the Witt group of the rationals. 0, w(X) If q = If q / 4k, set (2) If (X,DX) set w(X) = rank(H0 (X; C)).(l> Remark 4.2.: sign = 0 w(X) in in Clearly : W(T) W(D). W((). is a Witt space of dimension = w(R), where Let class X, q 0 (mod 4), X = X U cone(X). w(X) -- Z is a p.l. invariant. be the signature homomorphism of 11.2.5. Definition 4.3.: integer sign The signature of (w(X)). X, sign(X), is the 64 Remark 4.4.: This signature extends that for pseudo- manifolds which can be stratified with strata of only even codimension and, therefore, p.l. manifolds [111. the classical signature for 65 Chapter IV: Section 1: w(X). Properties of the Witt class Introduction. In this chapter, we study properties of the Witt class, For the most part, these are analogues of properties w(X). of the signature of manifolds. (1) Cobordism invariance: X If (2) (3) = 4k+1 k, then = 0. w(X) Additivity: Z 4k-1 = aX4k If then 4k w(X UZ Y) = w(X) -Z 4k-1 = 3Y4k and + w(Y). Multiplicativity with respect to signature of closed manifolds: X If is a Witt space, and M is a closed manifold, (*) w(M x X) = sign(M) - w(X). These properties are crucial in the application of Sullivan's ko, ® ZZ [ bordism theory. ] orientation argument to Witt See Chapter VI and the Appendix. In section 2, we give a proof of Witt cobordism invariance of the Witt class. The proof is modeled on the classical proof of cobordism invariance of the signature for 66 manifolds [12]. In section 3, we give a geometric proof of additivity. "pinch cobordism," we deduce additivity By means of the from cobordism invariance. Incidentally easily this result implies the Novikov additivity theorem for the signature of [cf. manifolds. 5, 588] p. Section 4 contains a Kunneth formula for intersection homology of a product and X M X, x where M is a closed manifold Specifically, we prove that a closed Witt space. chain level cross-product induces an isomorphism: H*(M;Q) 0 H(X; for all perversities formula [10] (*), ) H (MxX; From this, we deduce the product p. which generalizes a similar formula stated in for the case where X is the quotient space of a prime order diffeomorphism on a smooth manifold. Section 2: Cobordism invariance. Theorem 2.1.: Let space with boundary. (Y,DY) Then be a w(DY) Proof: The argument resembles follows [12]. that (4k+l)-dimensional Witt = 0. in [11], which in turn 67 Y = Y U cone(DY) Let Y be the space obtained from We will construct by adjoining the cone on the boundary. a commutative diagram of rational vector spaces (assume rational coefficients): i k+() = (c) i + 2k, n-(4k+1) = m(c) is surjective, and algebra in m + = n(4k+l) 4k, in The vertical map a 3 (over Q). is the isomorphism induced by implies that A standard i(H (Y) ) k(aY). =dimnH Here are the details. in linear arguent is self-annihilating Hmk (3Y), under the pairing of III.3.3, dimQ i(H k+1 (Y)) (4k+l) 1. as in 111.3.7. [12] c < 2 < (0y) the row sequences are exact and, dual to each other inclusion, for = 2k - = m(4k+1) In the diagram, fact, 2 < c < 4k, for = n(c) n-(c) and H H2k (0Y) H H2k+1 (Y Here H k Y -H k(0Y) and It follows that w(DY) The top exact row fits into the long exact sequences on homology corresponding to the inclusion of chain complexes: (1) C'2(Y) Similarly, C C = 0. (Y). the bottom row is a segment of the long 68 exact sequence gotten from: (2) C (Y) -- C (Y) Allowability is defined with respect to the stratifications on and 3Y Let be the set of point singularities in the {v } By reasoning just and C k 2k+t (Y) '2k+l C +2(Y)/Cm (here (1) I a finite is index set). the quotient we find that [11], as in from complex arising I2 links of 4k-dimensional be the corresponding {L } Y, and let stratification of set Y. induced by that of Y has only two non-zero terms: Z . (D ) s Z2 k (L k B k(L ) ( PY) B k iE I The differential on the quotient complex identifies with the inclusion homomorphism. H2k+1( (Y) /C (Y)) The Witt condition on Y H k (Li) Thereforo (L EHk ) H k ensures that 0, all i c I, so m m+" H 2k+l1(C * (Y)/C (Y) m H2 k (3y). 69 This proves the existence of the top row of the diagram. An entirely analogous argument yields the bottom row. The duality of the row sequences follows from the intersection pairing theorem a remains to show that is (M+)+(n-)= t. Note that (1.4.4). For thlis, surjective. refer It to Remark 111.3.6, noting that the Witt condition on links is v, satisfied for all odd codimension strata except the cone I point. Section 3: Additivity. w(X) The cobordism invariance of the Witt class suggests a simple geometric proof of the additivity of w(X). "Additivity" refers to the property described in the following proposition. Let Proposition 3.1.: Z, Witt space, and (1) Y = X (2) X1 (3) (XVZ) be an oriented 4k-dimensional Y X , X2 subspaces of and Z (X2 -Z) = w(X ) + w(X2 is biTollared in Y are oriented 4k-dimensional [Y] = [X 1 1 + [X2 . X2 = Z, and ' w(Y) such that: U X2 Witt spaces with boundary, so Then, Y 70 Proof: Let K1 , K 2 respectively. be triangulations of Denote by K , K2 X1 , X 2 the complexes with isomorphic simplicial collars added on the outside The induced triangulation of Z the collars) L. Then as in [20, will be denoted is p.l. homeomorphic to "pinched space" (on the outside edge of IKI p. IK = 24]. the polyhedron underlying UL K2I The the complex defined by: J = where and Li, v L2 [K U cone(L 1 1 )] v ILI induced map p, to in X1 , X 2 v I|J. in p : Orient Ii Figure 2. Attach a collar x C with collared boundary. To check check that homeomorphic so that the IKI I to P is H k(Zk(v,P);Q) Z x CP, P See forming the space is a pseudomanifold It can be oriented so that We call that to by is triangulable. It is not difficult to see that lJI. JlI IKj IJ By constructing a triangulation explicitly, we see that the mapping cylinder - -- on homology carries the orientation of IJ|. to that of IKI Z is the common cone vertex. collapsing aP = [K 2 Ui cone(L 2)] are the triangulations of Now, define a continuous map: P. [n ' J is Y, [20]. P the pinch cobordism. a Witt space, = 0. But [-1,1] U cone(3(Z it is ik(v;P) x enough to is p.l. [-1,1])). This is 71 Z, with suspension homeomorphic to the suspension of normalization as S Therefore, points identified. has the same Z * Zk(v,P), implying H k(S0 H k(kk(a,P);O) * Z;Q) 0, with the latter isomorphism from 111.2.5. By cobordism invariance of the Witt class, we have w(Y) = w([J ) = w(X1 = Section 4: + w( 2) w(X1 ) + w(X 2 ) 111.4.1) (by definition Multiplicativity with respect to signature of manifolds. Suppose space, M a p.1. is manifold, both without boundary. without boundary. Then X and M x X is is a Witt a Witt space In this section, we prove the product formula (1) where w(M x X) = sign(M)-w(X) sign(M)-w(X) in the abelian group is sign(M) W(Q). times the element w(X) 72 The proof proceeds as follows. M For closed manifold X, we construct an iso- and arbitrary pseudomanifold morphism ip : H * (M; Q) for every perversity Assume X 0 HP (X; Q) i+j=2k X; Q), x This is done in Theorem 3.1. p. is a Witt space, specialize to the case -H+PH(M p = m. dim(M x X)= 4k, and We have an isomorphism H . (M;Q) 0 HJ (X;Q) i 1 H MXX;Q) -2k which can be regarded as an isomorphism of rational inner product spaces. The inner product on the right hand side is that given by Theorem 111.3.8, while that on the left side is uniquely determined by: (a b ,cd>= (-1 )dim b dim c<a, cM -(b , d)X if dim a + dim c = dim M dim b + dim d = dim X otherwise 0 where <,>M H* (M;Q) if and and dim M E H 1 <,>X (X; (mod 2) are the intersection pairings on It follows at once that or dim M = 2 (mod 4) w(MxX) and = 0 1313 73 dim M = 0 if = sign(M)-w(X) w(MxX) (mod 4). First, Now we formulate and prove Theorem 3.1. define a chain homomorphism C*(M) 0 C (X) -b. C (MxX) Given (resp. d.) be the homology class in (resp. H.(Id.I,I3d.i)) corresponding to J Denote by c. H. J1J J d. x c. (M) 0 C*(X) , let c . 0 d. e C 1 J1 as follows. the chain in , |3ci ) i p: d.). (resp. c. x ( 'c.x d1)C C (MxX) C. corresponding to the ordinary exterior cross product 6. x d. J 1 in H. j (Ic.IxdIrIcixld.I U respect to the product stratification on lies in i'j is a chain map. CP (MxX). 1+j Set i (c. 0 d.) 1 Let ip = c. Ic.xIDdi). c. X, M x x d.. With x 1 d. J Clearly be the homomorphism induced on rational homology: i i Theorem 3.1.: Proof: : H*(C*(M;V) 0 C is (X;Q)) -+H (MXX;0). an isomorphism. The proof breaks down into two main lemmas (injectivity and surjectivity) and two technical lemmas. We assume rational coefficients throughout. Lemma 3.2.: ip is injective. 74 there is a canonical By the classical Kunneth theorem, isomorphism Let q (X; = H*(M;Q) ) 0 C H* (C* (M;Q) 0 H (X;0). p + q = t, be the perversity such that the homomorphism: iq H*(M) (X) 0 H -0 H*q (MxX). The dual homomorphism is: (i q)* : (H (MxX))* There are canonical V, 0 H (X)) (H,(M) isomorphisms (2) (H (MxX))* ~ H (3) (H* (M) (X))* 0 H (MXX) (H*(M) )* H,(M) 0 (H (X))* 0 HP (X) by 111.3.8. So, we regard (i )* : H (ii') (MxX) as a homomorphism H(M) + 0 H (X). and consider 75 Injectivity of follows from the following claim. ip 0 i (ia)* The composition Claim: is the identity homomorphism. observe To verify the claim, (iP (~ where <, H*(M) 0 H 0 6) E 6,E iq( intersection numbers (MxX), with above. and such that and 0 m+x 0 < k < m+x. are obvious. Given a representa- z' = Yc. x d.. SJ homologous to z' a, we produce a cycle for z (3) (2) and The cases of dimension tive cycle with is surjective. i* a E H r-Q (M) ® HP(X) (X) , and that, by 111.3.8, Lemma 3.3.: z H is the pairing of determine the isomorphisms Assume that In the course of the construction, we use some technical lemmas whose proofs Let skeleton Let skeleton S be a triangulation of Si, and T T., perversities M, with i-dimensional (m-j)-dimensional coskeleton be a triangulation of and basic sets p (3.4 and 3.5). and with respect to perversity q. p {Qk} X and Sm . are deferred to the end of the section with i-dimensional {Q } for (We will refer to a chain as allowable if its support is so.) 76 (p,Z)-allowable with respect to the is z The cycle (product) intrinsic stratification on stratification refines it. k > m We treat the case since any other MxX Z < m. initially that Assume later. The first step is to apply stratified general position [16] so as to be disjoint from z to shift SS(in- t+1)q [Qq nT X z . Denote the shifted chain by is a strong deformation r : M x X - ] >MZ~) retraction: S(m-Z+l) X[Q intrinsic skeleta of 1zl1 follows from 3.4 that v the map of vertices 1zlJ, simplices of support in M x Q U imply that S x X. 12 is and Moreover, T1 2 and, it also has a linear extension to -. r(v) z 2 = r*(z ) Also, (M x QP) () (S z2 S U S can be triangulated so that yielding a cycle and the fact that Z-1, MxX. x Q -+ M 2 which preserves products of simplices in therefore, there By lemma 3.4, Z-2 with the delfinition of x X) r, has dimension (p,t)-allowable. 111.2.3 proves that Decompose z2 z2 is homologous to z . An argument entirely analogous to that in the proof of into the sum of two chains: XX 77 z2 Jul C MXQ where l|ul = Q Note that M x Qp is a regular cell complex with 0-skeleton w + y1 , where lylp ,1 C Z-1 M x X, lies in the nonsingular part of homologous to u XQP 13CSXQP IU S 1 X S y1C 3w = 3u with |y 0 '_ C QP, and and -Say Q By lemma 3.5, J - . for chains C c E C i Since lvi keeping n 3v Tx 2 fixed. x X, X-2 YJC-2S Int (aP stratified (M), E C (X) d E C z 3 = w + v, with levI d and (X). w E Image(i ). is disjoint from using stratified general position so as to be disjoint from this subspace, v chain in a d .. c = iP( y and is (p,t)-allowable and S (m-Z+2) X [Qq we shift (M), E C So we now have a cycle K .- c 1x dj) ip y0 ) . Q1-l Z 0 9-2 = y0 x X. k-2 X we can find a chain w IVI C XQP _1 l3vI C S Since and and 1v = u + V. 2 xX The shifts can be chosen to keep the since they take place in which is the interior of pseudo manifold of dimension a x + k - 2 and 78 (S X) x ,-2 k, of dimension x q_ 0 x-1 x x-2 ) U a -2 x- Denote the shifted chain x-3. of dimension and v1 . by \ (S (m-t+2) Tx x-2 [Qq_ x-1 |vj subsets involve closed Now, r : S use the deformation retraction, X - x k-2 S(m-Z+2) [Qq_ x 0T x-l -3 S with support in as y 2 + v, Sk-3 where x c,-2 e C Z-X22 (M) fashion, The chain x d2) and shifting r*(v{) v' L C to Now, 2 (X) v where c e C (M) proceed and deforminq y3 E in the same v' to Image(i ), and so on. We will then have Continue until the process stops. y, homologous to y and v r*(V Decompose will have a decomposition r*(v!) = y 3 + v", where constructed a cycle homologous to by 3.5, where E d Q x x X U St - 2 is the subchain with support in - Z-2 x QP. Then y2 ii S)-2 X U y2 - c r,(v ) (p,Z)-allowable chain x-2 . to get a given by 3.4, = d j+k=E. k E i (y c i0 d) i Ck(X). k z, with k r,(v!). 79 k < m. This completes the proof for the case Z > If m, the procedure is very similar, in indices being the primary modifications. first shift z with changes Specifically, so as to be disjoint from (0)- +1 T 1, yielding a chain z. - X Use the deformation retraction: X - : M x r S() x ) Tx [Qx-(-m)+ 2 Mx QP US k-m in-i to get a chain r*(z ) M x QP - S x X, Decompose r*(z ) homologous and which is to lies y-1 Now, -S xXUS xQ . v, in with support in into a sum of two chains and C M x QP e + V, |vi C Sm-l x X. By 3.5, Image (ip ) . 1yi-mn 1y1 repeat the shift, deform, decompose procedure on using the deformation retraction: r : Sr M-1 xX-S X (p,k) -allowable. r,(z1 ) = y 1 where z, x M x [Q q_- x-(k-M) n T(xm-2 P M-1 x-M+1 80 k < as in the case where And so on, m. 5 The first retractions lemma provides the deformation technical used in the proof of 3.3. We thank Javier Bracho for his assistance on this argument. Given integers Lemma 3.4.: 0 < j < m, M Xx there M is S x i, j 1 < i < x with and an embedding: [Q 1 n T x-2 ] *[S m-j-l X X U M X Q]x-i+1 which is the identity on S [Q9 r) T ] 1 x-2 x and S and which takes M X x x-i+1 to a union of join lines. It suffices to define compatible embeddings on products of simplices in APE S' and T': bxAq Ap xAd (A A(AA S', a _P 5 Ab (j) i ) r) S = A AP - .- - A(x where x Q ( Proof: . x XUM M-3-1 81 AC = A T ', and A Let {v 1, {v 1 }, ' d- A q f-QiP 1 T {v 1 }, in these decompositions: x e A Any points be vertices of the simplices {v.} 1 and A = Aa = AC * A y 6 Aq Ab have unique expressions in terms of barycentric coordinates: and y = a Zc.v Define and Both are p.1. b V a + + , x = with Ya. + b d Vi, with Yc. + d. 1 1 : A -+ I by t(x) = b s : A - I by s(y) = Yd. t = 1, a.,b. > 0 = 1, c.,d. > 0. 1 1- maps. X = {(x,y)Jt(x) < s(y)} Y = {(x,y)|t(x) > s(y)} Z = {(x,y)|t(x) = s(y)} Ap x A : . Consider the closed convex subspaces of See Figure 3. We define the embedding separately on these subspaces. It is easy tb check compatibility on intersections. Define by iZ : Z-Aa x Ac *Ab xAdc aa x Ac (Ab q p d 82 a 1t, Ea.v. (J c.vc 11) Ea. Ec. 1 (x, y) b + (1-t) Ed.v. Eb. Ed. ( ) 1 1 H-0 (x,y) Define di Eb.v. Ac P a x hX if t = t(x) if t = t(x) 0,1. = Ap d q Aa xAC *(Ab d 0,1 by Ea.v ( -t + ( 1-t S if (x,y) Define (-s) + (I3- t t(x) = (-),3 S t # 1 S )s s = s(y) and 0 0 - (x, y) 1-S Ed.vd Zb.v. Za v Ec.v hy : ~a if c s(y) = 0 qe b t(x) = 1. or a x p~x ad ql cb similarly: Zava ( l-t i (x,y) -+ (x,y) EC.V I)(1-) 1-s + ( zc.v Eb.vb t if t if t(x) = 0 = t(x) # 0 or and Ed.vd + (IS 1t tS s s s(y) = # 1 s(y) = 1. Continuity is checked using barycentric coordinates, and the definitions agree on overlaps. seen to be a union of join lines. Since The image is easily Ap x Aq is compact Hausdorff and the map is injective, we have an embedding as 83 claimed. I Thus, the strong deformation retractions along join lines in the join induce the strong deformation retractions in Ap x Aq which we used in 3.3. Specifically, we obtain the deformation retraction r : - Ap x Aq which preserves the convex sets A them to the convex sets b x Ac a X, Y, Ad x U A x A and Z, Ab x Aq, and x Ad and carries Ab x Ad respectively. The next lemma was used to conclude that the chains in a product space Lemma 3.5.: with Let dim X = x, S (X,X0 ) c E Zx+y(X chains (X,X0), with a] E Z J and (Y,Y ) dim X0 = x-l Given a chain are in the image of Q x X YX bj e Z and dim Y = y, x U Y x Y) (Y,Y0), for j X dim Y= there are = l,...,J, finite, such that J c = . aj x bi. j=1 Proof: ip. be tinite complexes . y. By the Kunneth formula [21, p. 235] dimensional chains we have the isomorphism: for top- y-. 84 H x (X,X ) ® H 0 y (YY ) 0 = Z x (x,X ) 0 Z 0 y z +y(X x Y,X (YY x 0 YO U X 0 xY) H +y(X x Y,XxY C E Z +y(X x YX x Y 0 U element of the tensor product under the isomorphism. X0 x Y), we can find an a Therefore, 0 b3 c = which maps to a x b c . Since x Y). This completes the proof of Theorem 3.1. 2 85 Chapter V: Rational surgery on Witt spaces. Section 1: Description of program. In this chapter, we develop a technique for performing We use it to prove that rational surgery on Witt spaces. the Witt class In particular, 4k, and determines the cobordism class of w(X) if X. is a closed Witt space of dimension X w(X) = 0, we find a . n'n }I "surgery basis" with respect to which H k (X;Q) for 01 the inner product has matrix 1 diag[( 1 0'' 0 0 1 By collapsing a regular neighborhood of an irreducible representative for (see Section 2), we obtain X with (1) w(X1 ) (2) rank H' k(X1,) space and al 0 = = rank H (X;Q) - that X X and X . We say by an elementary surgery, X is obtained from are killed. (with a collar the mapping cylinder of the collapse gives a Witt cobordism between added) 2. {ctL31 } Roughly speaking, the homology classes Moreover, a Witt and call the cobordism the trace of the surgery. Iterating this procedure space between X with n X and Hm 2k Xn ;Q) X n . = n 0 times, we construct a Witt and a Witt cobordism Coning off the copy of produces a Witt cobordism of X to zero. X n in Y DY ;Z) = 2Z A similar procedure works when the dimension of 4k+2. In that case, we find a { (ca ),.,(Ct is symplectic basis with respect to which the skew- n)} symmetric form on H k diag[(diag[( 0 0)]. 0 2k+1 1 X has matrix (X;Q) Then proceed as above. In section 2 we The chapter is organized as follows. define and prove existence of irreducible representative cycles. In section 3, we study the regidlar neighborhood of an irreducible cycle and show that the "mapping cylinder" of a collapse as described above is a Witt cobordism. in section 4, Finally, we construct a surgery exact diagram very much in the classical vein [17] to compute the effect of surgery on intersection homology, verifying and (2) above. Irreducible p.l. Section 2: Definition 2.1.: G = 2Z (1) A representative cycle H . (2) the generator |z of H.(I z;Z) J on every j-simplex of of Remark 2.2.: for a e Hk(X;G), J Z (1) J z is irreducible if Q, or geometric cycl-s. Izi, has coefficient 1 in some triangulation |z1. Condition (2) implies that the generator of 87 H.(Izl;E) has coefficient 1 on every j-simplex of J Izi. in any triangulation of be a p.l. pseudomanifold of dimension X2k Let we say X 2k By abuse of [X]. with irreducible fundamental class language, Izi Thissection is devoted is irreducible. to a proof of the following lemma. Let Lemma 2.3.: a E H Given z which is Proof: J X2k be an irreducible p.l. pseudomanifold. (X;q), there exists a representative cycle irreducible, if 0 _ j < 2k-1. by a general position argument, lvi such that p. 67] E v Step 2 uses the technique of piping to produce a cycle satisfies the condition j-cycle, a representative cycle supports a simplicial cycle with coefficient 1 on every j-simplex. [20, Step 1 constructs, The proof proceeds in 2 steps. z Hi.(Izl, Z) which, additionally, = If Z. J is a denotes the complement of the intrinsic open j-stratum. Step 1: In some triangulation T representative simplicial cycle for j-simplices if necessary, choose a X, of a. By reorienting we may assume all non-zero Clear denominators coefficients of j-simplices are positive. so as to obtain an integral cycle y = Y GET n(a)a G such that 88 all distinct non-zero coefficients share no common factor. If = 1 n(a) whenever / n1(a) 0, v = y then set and Step 1 will be completed. If not, let be the connected components of E 1 ,...,E the intrinsic open j-stratum of Y lyl. = we can associate a unique coefficient of simplices in that component. union of n. copies of the obvious p.l. map. Let obtained by identifying f(x) = f(x'). Then .W having coefficient +1 f*(w) = y, where with respect to X - g : W -+ shifts HO = to x' ZX H1 if x,x' [1, W f S(wx [0,1]) g,(w) - f*(w) p. 418], - W; = g. homotopy Then (Z ) and y j-cycle w [20, p. We move 51]. H X. Let The linear trace of the : V x [0,1] -+ X with is a p.l. geometric cycle on each j-simplex. (p,j)-allowable. [0,1] = wx 1- wx 0. g,(w) sequence that is, only if g*(w) x = W into general position the nonsingular part of Its support is also W -l E f Y W -+ and such that which, when triangulated, has coefficient +1 Give : f Perform a finite f, bringing a p.1. and and be the quotient space of be the reulting map. X yields f = 1,...,n, on all j-simplices, ZW Zx, be the disjoint supports a simplicial Int(T) C only if W the coefficient denotes the induced map on chains. f* of shifts of the map f(Int(T)) C x f(W-E W) C: X - Note that f(T) W n., Let i E., To each component the orientation such that - y. Then Since H*(w x H(x,t) [0,1]) = H(x,0) = = f(x) 89 when x e x e W - is ZW, EW for all t, and for all t, y such that in H [v] = g,(w) (X;V). Let an orientation respecting pipe E6 H.(X;Q). lvI. P jvj of (X-EXv l r) (X-E X)) complete. F to Fi+1 [20, p. 67). is a subcomplex of some triangulation is 3 cally unknotted at all points in the interiors of j-dimensional lvI. v We now construct from Note that since of be the Then J 1 < i < Z-1, with distinct pipes disjoint. X, the pair v be the ccnnected components {F . } ofthe intrinsic open j-stratum of for [0,1])| represents Let is the desired representative, and Step 1 ir Step 2: when IH,(w x it follows that the same homology class as g*(w) c X - EX We conclude that (p,j+l)-allowable. multiple of H(x,t) simplices Therefore the piping construction can always be carried out. Let Z be the p.l. the piping is done. Then subspace of Z X we obtain after supports a simplicial cycle which has coefficient 1 on every simplex in Z - E Z is connected, this cycle generates Z. H Since (Z;2Z). (p,j)-allowable and, by "filling Clearly, its support is in the pipes," we see that it represents the same homology class as g,(w) in this cycle such that H (X;Q). [z] satisfies the conditions Let [v] (1) and E z be the multiple of Hk (X;Q). J Then z (2) of definition 2.1. I 90 Remark 2.4.: If X is an irreducible p.l. manifold of dimension 2, class a e H 1 (X;Q) pseudo- it is easy to see that any is represented by a cycle whose support is a p.1. embedded S 1 91 Regular neighborhoods. Section 3: Let X2k U e H Iz| in be an irreducible representative for a (closed) regular neighborhood of X U = U U X. The stratification of - Since Izj is and Let 437]. adjoined. DU U: induces a stratification on U U in X corresponding to a stratum X U cone(X r) DU) p. [1, U Then aU, with the cone on U be cone(DU) v0 {v 0 }, where (r,k)-allowable, U (X;), with is a pseudomanifold with collared boundary stratification induced from that of z Let be an irreducible pseudomanifold. is the stratum is the cone point. the cone point v0 constitutes the entire 0-stratum of the stratification of U. The main result of this section is: Proposition 3.1.: If <a,a> = 0, is a Witt space and X then: H ;(U.) = 0. To prove this, we prove a series of preliminary lemmas. Let i : C (U;2Z) complexes, where for 2 < c < 2k-1 -+Cm(U; mand ) be the inclusion of chain is the perversity: m - (2k) = m(2k) - m (c) - 1 = k = M(c) 2. - An argument identical to that given in the proof of IV.2.1 proves that the quotient complex only two non-trivial terms, C (O;2)/C ~(0; ) has and there are canonical iso- 92 morphisms of these terms: k+2 /C mk+2 B m( U) Cm Clm /C k_ k+1 k+l k = Z (U). identified The boundary homomorphism on the left is We therehy obtain: with inclusion on the right. Lemma 3.2.: There is an exact sequence of integral inter- section homology groups H + 1 (U) - (integer coefficients assumed): H k(DU) . H () -. F(U) k0 -k a of We made no use of the irreducibility proof of 3.2, Lemma 3.3.: Then H but it is u Let in the nexL be a generator = 2Z (U;) crucial of z in the lemma. Hk(lzl;2) and the homology class of u = Zk(lzI). is a generator. Remark 3.4.: (1) Since Jul is (m,k)-allowable, it is also (m-,k)-allowable. (2) [u] Since e Hk(U;eZ) [z] E Hm(X;C) is non-trivia 1. is non-trivial, If u = w, for some 93 W E Cm k+l |wl projection, we may assume also, this would imply Therefore Proof of 3.3.: Let C [u] E cone(9U) * = v0 E Hk~ (U; 2Z) , assume |yI * $ y jl A then Int(a), Izi f(v) = with * -- f,(y) in jz Jul () Wu so that each simplex lies in Izi ITI = U. 3 If a vertex lul let and we c, 0 v T E IyI denote Iz|. Extend this map on vertices of T. linearly over simplices, :yj (U; 2Z) * A = A, by the usual convention. = the simplex in f y E C satisfies: a e U Then every a unique simplex with interior in Define with By pseudoradial projection, we can Triangulate the cycle lies in U. the structure of a derived neighborhood 33]. A a Suppose cycle Jl= Jlt () lI where (U;2) U0 0 U Give p. C . k+l in DU cone(DU) UO = Cl(U-N). Let with representative lyI A V 0 =v Then [20, U. C is non-trivial. of N U There is a p.l. homeomorphism of E (X;2Z) H (U;2Z) H denote a collar N w Since U. is trivial in [u] contradiction. so by pseudoradial ' Iwi r) v 0 then (U; 2Z) obtaining a p.l. |y| map U C . Then f determines a geometric cycle Ck (U; Z) . Since the support of f*(y) lies in 94 Iz i, E Ck (Y) (U;2Z) to define Proceed as in 111.2.3. . kA a p.l. lyt H map DH*(yxI) satisfying implies that represents and H*(yxI) (. Hk (Izl;2Z) j for some E and geometric chain x I -U = f -(y) y. Therefore e Ck+l (U; 2Z). f*(y) = 2Z by hypothesis. and so (yxI) f The definition of But, 7Z, H f*(y) Izi, is a k-cycle supported in ( = Therefcre f*(y) = j-[u]. By the intersection pairing theorem (1.4.4), there is a nondegenerate pairing induced by intersection product: Hk That n + is, (c) = n(c) S E n+ Hk (U;Q) 2 < c for n+(2k) Choose OwQ. (U;D) is the perversity satisfying n+ Here, (U;Q) x H k (i-) < 2k-1 = n(2k)+l = k a generator <UaI> = Lemma 3.5.: j-u such that 1. The following diagram commutes: + and (n+) = t. 95 p H (U;0) k n+ H k(U;) All The map Proof: C n+ k 4- Hk Ii (U;Q) H - n Hk(U;q~ j 0 Hk is given by: g (U;Q), rn~ Finally, the map m-- = Ck (U;Q) j Since implies that of g immediate from the 3.1. Refer to the diagram of 3.5. hypothesis of the proposition, a surje-tive. 3. We may now prove Proposition Proof of 3.1.: is h is Ck+6(U;) Similarly, is injective. the description definition of that of chain complexes. maps are induced by inclusion Commutativity is clear. C k(U;) 0 -+ generates m-^ Hk (U;E)). commutativity imply that ioh g Under the (Reca l is the zero map. Injectivity of j and is the zero map also. inherits a Witt space structure from X, so 111.3.7 implies that i map. The proposition now follows from surjectivity of is an isomorphism, so h U must in fact be the zero h. I 96 One corollary will be useful when we analyze the surgery exact diagram in Section 4. Refer to 3.2 for notation. Corollary 3.6.: Assume rational coefficients. Under the condition of 3.1, H 3,(Y) (3U) ~{y} Image (q*) , where E H (3U) with I = a. Section 4: Let The surgery exact diagram. X be an irreducible Witt space of dimension 2k. Assume there is a decomposition of H where y VO = (X;) = 1 = (Yl''O.ryd' H V E (a ,6 _(X;,): ) and with respect this basis the intersection pairing is given by: ,3 0a ,Yi) = 01 lY Oilal )= = < ,1> 1 . (Xlal> = = 0, 0 i = 1,...,F to 97 Let U be a regular neighborhood of an irreducible z representative cycle is in bicollared X (X - X= The passage from Note X [1, X for p. 437]. Int(U)) U X to U We may assume that TU Form the space cone( U). is called an elementary surgery. is homeomorphic to by collapsing a1. X, X the space obtained from to a point. The following sequence of propositions, whose proofs will be deferred briefly, partially describe the effect of the elementary surgery. is an irreducible Witt space. Proposition 4.1.: X Proposition 4.2.: Let collapse X -+ Y be the mapping cylinder of the X, with a collar X x I attached. can be given the structure of a Witt cobordism between and on X . (We call Y There is an isomorphism Hk (X which preserves the X the trace of the elementary surgery X). Proposition 4.3.: Y Then ;) ~V0 intersection form. 98 Repeated application of these propositions yields the result alluded to in the introduction. Theorem 4.4.: Let dimension 2k, k > 1. X be an irreducible Witt space of If = 0, w(X) then X Then w(X) is Witt cobordant to zero. Proof: dim X 2 0 Suppose (mod 4). that the inner product space Choose a basis {a, 1 ,a 2 H '3 2 ,.. (X;O) which the matrix of the form is Let 4.1 H VO ( 4.3. 2 4'2 '..,'c'n'n) We obtain (X,) X isomorphic to rational inner product space. V; (al,Y ), and apply Xl, Witt cobordant to -. setting 01)( )]. - V = (a ,,... Now decompose ,a.',') and H cobordant to Hk(X;Q) 0. as (X ,q), and V= Continue until we obtain = with V repeat the procedure. X, with implies with respect to diag[( and 0 is split, by 11.2.9. 4an' } ., = Xn, The cobordism to X is obtained by pasting the traces of the elementary surgeries along their boundaries. Now attach cone(X ) to the cobordism. If H m(X;V) {a 1 dim X is skew-symmetric. 1 1 ... form is = 2(mod 4), then the intersection form on ,n' diag- 6 n} Choose a symplectic basis with respect to which the matrix of the 0 1 1 0 1 0'***''l0)]. above shows that a sequence of The same reasoning as n surgeries will yield a 99 cobordism of X to zero. a We now supply the proofs of 4.1 - Proof of 4.1.: The fact that is bicollared in Witt space. X X k = 1, Therefore Six is a Int(U) is path connected. There- is irreducible. If to X aU general position proves that the X - nonsingular part of fore is a Witt space and immediately implies that k > 1, If X 4.3. z then X has at most point singularities. is an embedded [-1,1]. S1 and The assumption that X - trivial implies Int(U) U is homeomorphic E Hm(X;Q) [z] is connected, so X is nonis irreducible. U Proof of 4.2.: The trace of the elementary surgery is homeomorphic to the following polyhedron, defined analogously to the trace of a spherical modification (X - Int(U)) cone (OU) x U [0,1], attach along DU x 1. the resulting space as attached as above. aU x 0 along [17]. and There is an embedding of aU x I Now attach with U cone(U) and To U in cone(DU) = v*U along U via this embedding, yielding the trace of the surgery Y. Y and is clearly a pseudomanifold with collared boundary, it can be oriented so that aY = X - X. is a Witt space, it suffices to check that To prove that kk(a,Y) Y satisfies 100 H But Zk(v,Y) U, Proof of 4.3.: Steps 1 - 4 (Pk(v,Y);q) = 0. so Proposition 3.1 completes the proof. The proof is broken into a number of steps. provide the ingredients for the surgery exact diagram, which constitutes step 5. Step 6 completes the proof using the diagram. Step 1: Transverse chains. Let U denote cycle of X restricted to the regular neighborhood aU (by abuse of notation) its boundary chain. C*(X) Let consisting of chains TC*(X) c dimensionally transverse to c e TC and, 1 c and by the Lefschetz boundary formula: = DU n U + (-1)2 c (icU. This defines a homomorphism t by t(c) m : TC(X) = c r\ U. 1 - m C(U) 1 Dc are To each element (X), we can associate the intersection chain (C 0 U) U, and be the subcomplex of for which U. the fundamental for each i c r) U, 101 U C U). (We use the inclusion m C Define (U,3U) Give the graded group to be the image of C t in C (U). the structure of a chain (U,9U) complex, defining 6 by 6(t(c)) Note that 6 C (U, 3U) = 3c C -+ 3- (U, 3U) CM 1-1 U = t(cc). if is well-defined: 3(t(c)) = 3(c r) U) U + c = t(c) = 0, (-1) 2 then k-ic r 3U = 0. Dimensional transversality implies that dim(1 ac n U so | c f )U|) cP) UI A < i-2 0 = and c n In particular, 6 (t (c) ) = aU = Dc n It is easy to check that t is a chain homomorphism. 0. U = 0. 6o6 Let = K 0. We conclude that denote the kernel of t, with the induced chain complex structure. short exact sequence of complexes: There is a 102 0 - K M~ -- m TC* (XW m C* (UF - U) -~-V 0, -aHk(K ( M(UDU)) Hk In steps 2 - ) H (CM (Ua U)) H+ -- Hk(TCm(X)) 4 we analyze the homology groups occurring in this sequence. Step 2: = H H,(TC M(X)) The inclusion (X). of chain complexes i : TC(X) InCo(X) induces an isomorphism on homology. In other words, Hm (X) . H,*(TC *(X)) Proof of surjectivity: Apply the stratified general position lemma(I.3.3) (or its corollary 1.3.4) z E C to a cycle (X). Proof of injectivity: m a chain w E Step 3: The complex Let m+ C +(X), Apply the relative part of 1.3.4 to where Dw = z and C +(U). denote the sequence with z m E TC (X). + and an associated long exact sequence in homology: 103 m+(c) m+(2k) Although m+ m(c) = = 2 < c < 2k-i for m(2k)+1 = k is not a perversity, we may define (m+,j)- allowability of the support of a geometric chain with respect to a stratification. consisting of chains is c Consider the complex of dimension IOcl (m+,j)-allowable and is respect to the stratification on j C, such that (U) Icl (m+,j-l)-allowable with U. Recall that the cone point is the 0-stratum. The inclusion C mm+ (U) sequence of complexes. Cp '-. (U) induces a short exact We can analyze the homology of the quotient as in Lemma 3.1 and obtain an isomorphism: m ATfl+ H k+l(U) A Hk+l(U) and an exact sequence 0-*Hk (U PHk () Hk- 1 O) Hk- 1( We get another long exact sequence from the short sequence induced by the inclusion C>(U) n+ is the perversity: --+ C, exact (U), where 104 n+(2k) = 2 for n (c) n+ (c)= n(2k + 2k- 1 < c < k. )= There is a commutative ladder: 0 -- 0 H -- + (k- lH (U) -~+Hk _ ($ -- --0 HHk ()) 0 k- 1 k-l1 U) Hn kl(DU) ~H -+0+H Hn kl(U) k k -+ H' Hk (^) (U) All vertical maps are induced by chain complex inclusions. is Witt, X Since and U DU Proposi- are also Witt. $ 0 0 tion 111.3.7 and the 5-lemma imply that is an isomorphism. We now identify j > k. Let N m J H . (U, U) m (U) J H. with DU be a simplicial collar of cone( U) U is a p.l. homeomorphism cone lines in the cone on the left. N In cone( U) U will N cone(DU) U N. denote the chain map induced by pseudoradial projection along cone lines from the cone vertex boundary of p,(v) preserving U, we can triangulate be subordinate to the simplicial structure on p* There U. in ~ cone(aU) any chain so that its intersection with Let in dimensions = v, Cl(U-N), and pop, p: = p,. Define homomorphisms: C, (U) -+C*(U). v to the Note that 105 (U), --. C : C .(U,DU) J J J j > k + 1 and -- w (U,3U) : Z (cycles) Z(U) by j+lv1 (c (C U)) is 'P in Step 1 shows that t The discussion of * $k (t(c)) = p,(t(c) well- defined. For j > k + 1, 1P _ (6t(c)) = $ _ (t(0c)) = P*(t(0c) + = p,(0(t(c) + = Dp,(t(c) (-l) n\ DU)) (x V (-1)vj+V*(c n\ DU))) j+1V*(C i (t(c)). = $. J induces a homomorphism ( H. (UaU) -aH J (U) j >k - on homology for In fact, ), * Therefore na eU)) ($' J .* Proof of surjectivity: is an isomorphism for Let m+ (). J z E Z. j > k. The cycle p*(z) 106 M+ H . (U), by an in z as the same class represents J argument just like that used in previous pseudoradial C m(U, U) = 3w, C M+fUU) where z E Let Z w E C +1 (U). (U,U) C. J (z) = P, (z + $ and (z) p,(w) (\ U Then = *p(w) t'\ U = p*(3w) Cn U = P* z + (-l) (-l) j+lv* (z n DU) p~op* since U, j+1 v* (z A (U), via a chain we denote by U)), are homooqous y. Therefore H Cm j+l (U, yAUe . U and The complex D (X - Int Let X-U = be the subgroup of such that: 6(yAU) ($P) in By the relative = y j DU. ($n(z)AU) j represents the same class as We conclude that (U, U). Step 4: (z) n U) =p. the chains stratified general position lemma, we can assume Then lies in \ U (Z) i (z) / i and suppose and 6 (p* (w) t'\ U) Since lies in U = p*(z). $p(p,(z) n U) and satisfies: Proof of injectivity: i (z) p*(z) / The chain projection arguments. z is injective. (X-U). (U) ) U cone ( U) . C. (X-U) J Let DT (X-U) consisting of those chains z. in 107 dim(Ic.I dim(Iac. J where v v) ( J j < r) v) < - 2k + (j-1) - is the cone vertex of X-U. (k-2) 2k + (k-2), Allowability of chain supports is with respect to the stratification on induced by that on X-Int(U). Thus the cone point is contained in, but may not be all of, K. (by means of the inclusion $ (c) = c (X D - X - i (X-Int(U)), where the 0-stratum. j: Define homomorphisms for all S X-U Int(U) C X-U) by: by abuse of notation, X-Int (U) represents the restriction of the fundamental cycle of X (1) to X-Int(U). Then c = c C\ (X-Int(U)) $. J + c,1 U is for injective, because c e TCm(X) J and (2) = c (\ U = t(c) Clearly the c. J 0 for c E K.. J are chain maps and we regard them as an inclusion of chain complexes. We will show that the induced maps on homology are isomorphisms Proof of surjectivity: for j Let N < k. be a collar of 3U in 108 X - p* Int(U), and be the chain map on induced C*(X^U) by pseudoradial projection along cone lines to the inner boundary of Izi v A $. = m(X-U) J If N. If z E D. H . (DM (X-U)). Proof of injectivity: a, then p,(z) = aw, where Iw v = l( $, and a p,(z) p,(w) E then represents p,(z) the e K., X-U. Let w e Dj+ < k, But JJ since its support lies in j p*(z) is a cycle, z same homology class in represents for e H.(K*). represents 1 (X-U). Kj+ ap*(w) = If the cycle J 1. a j If also. < k, z Assume then Moreover, P*(Dw) = p*p,(z) = p*(z) This proves the claim about ($4),, j < k. Finally, analysis of the quotient complex Am-A m C,(X-U)/D* H m(aU) Step 5: - (X-U) yields an exact sequence: Hk (Dm ~(X-U)) - H m(X-U) -+0. The surgery exact diagram. Recall that X is a Witt space of dimension 2k. There is a commutative diagram of rational homology groups, with exact row and column (assume rational coefficients): 109 H (J) k q 1H 3, Hk (DM*~ (X-*U) ) -- H(X) k+1 (U) +1~l(U) (U) i III Hk (U) H k (XU) 0 appears in Lemma 3.2. q* The homomorphism The isomorphism on the left was proved in Step 3. The horizontal sequence is a segment of the exact sequence in Step 1. We have used the isomoiphisms proved in Steps 2, 3, and 4 to relabel some of the groups. The vertical exact sequence is a segment of the long exact homology sequence obtained from the inclusion of complexes D'*(X-U) mA -- + C*(X-U), as discussed in Step 4. The commutativity of the upper left follows triangle from the chain-level definitions of the connecting homomorphism of the horizontal and vertical sequences and the map q,. The homomorphism triangle commute. Step 3, and t of t, Step 1. Here is defined to make the lower right g $ is the rational isomorphism of is the homomorphism induced by the chain map In fact, g is given by: g(y) = (,ay)6, 110 S generates inner product on Step 6: n+ Hk (U) Conclusion. {a , Recall the basis and the decomposition of H (X) ,,yl,...,Y,} (X) is the H (X). Assume rational coefficients. H (,> as in 3.5, and 1. ~V0 By remarks in Step 5, t*(y ) = 0, i = 1,. .. , where and t(a)= 0 Similarly, $(t,( 1 )) - 3. By exactness, @ (a!) @ Image Hk(D*(X2U)) = as a rational vector space, and to V 00 product. (a ) V'0 0 (a') 1 (3*) is isomorphic by an isomorphism compatible with intersection Now, corollary 3.6 and commutativity of the upper left triangle imply: Image (j) = (a!) @ Image(3*) 111 Finally, since the vertical sequence is exact, we conclude that: =V H X(X-U) H; V0k V The isomorphism is compatible with intersectiun product. This completes the proof of 4.3. 0 112 Witt space bordism theory. Chapter VI: Section 1: Introduction. In this a chapter, show that we Witt spaces provide KO geometric cycle theory for connected odd primes. Specifically, in section 2, Witt based on the class of Witt spaces, QW homology at the bordism theory is , introduced and using results from Chapter V and a simple construction given in section 3 of this chapter, it follows that the coefficient groups are: WiA-4- " Q z (pt) Witt Qq (pt) q ~0 Witt Qq (pt) W(0) q The isomorphisms are Finally, [24] ko* if q if > 0 and q > 0 q and $ 0 q (mod 4) = 0 (mod 4) induced by the Witt class. in section 4, the technique of D. Sullivan is used to construct canonical orientations in 0 2[%Z for Witt spaces, transformation of P Witt : and the induced natural theories: Witt -, A ko, 0 Z [%], is shown to be an equivalence at odd primes. 113 Section 2: Let The bordism theory Xd be the class QWitt of oriented pseudomanifolds L satisfying: It (1) L (2) H is is a Witt space (L;Q) = 0 not difficult if to check (oriented)singularities bordism sequence spaces. Let dim L {, Witt Q. n} Witt Q Wit(pt) The Z is a class of The associated is the class of all oriented Witt eo. The coefficient groups Wi tt (pt) are: = 2 Witt (pt) = 0, q Witt (pt) q that denote the bordism theory based on QW Qq 22. (use 111.2.3). the class of singularities Proposition 2.1. = = if W(Q), q if isomorphisms in dimension t 0 (mod 4) q > 0 q > 0 and qg= 0 (mod 4). are induced by the Witt class. Proof: The case q = 0 is evident. When q E 1 (mod 2), observe that any odd dimensional Witt space bounds the cone on itself. 114 The case invariance V.4.4, is q : 2 (IV.2.1) (mod 4) follows from cobordism and surgery In applying we can use the fact that any Witt space not irreducible space X'. is Witt cobordant to Simply take connected sum 2k-stratum of X, say 2k-stratum of X' is Finally, show that in E 1 , ... the case 0 Witt of components of the so that the (mod 4), IV.2.1 and V.4.4 the homomorphism Witt 4k (t is . . +W0 injective. construct explicit generators Section which an irreducible E ,MIin order, q induced by the Witt class w X2k # E 2 #...#E E w that (V.4.4). In section 2, Witt k (pt) 4k for we and prove is also surjective. 3: ____ Generators for __ Q Witt * (pt). We use the method of plumbing disk bundles over spheres, due to Milnor [7 ] to construct representative Witt spaces for each bordism class in and q = 0 (mod 4). QWitt (pt), when q In particular, > 0 The representatives can be stratified in such a way that the singularity subset point. q they have stratifications even dimensional strata. E is a single with only This yields information on ev 115 [11]. Theorem 3.1. entries, [7]: Let symmetric, Then, (M,3M) 1, M is H 2 k(M) (2) be an nxn and with even there is of dimension (1) and k > for B 4k matrix with integer entries on the diagonal. an manifold with boundary such that: (2k-l)-connected, DM is (2k-2)-connected, is free abelian; The matrix of intersections is given by H 2 k (M) ® H 2 k(M) -ZZ B. I Given an element a basis the {v } for inner product, with even entries E W(Q), it is possible to find V with respect to which the matrix of B, is an integral symmetric matrix on the diagonal. manifold with boundary represents [V,3] (M,3M) the Witt class Theorem 2.1 produces a such that [V,3] in W(Q). exact sequence of rational homology groups H 2 k(M;) In the long (rational coefficients assumed): 2 - k(3M) the transformation bases the {v.} intersection H 2k(M) i, - H 2 k (MDM) -- + H 2 k-l(aM) has matrix and {v*}, where pairing of {v} H 2 k(M) B - i* -H with respect to the is dual to with {v.} H2k(MaM): under 116 H 2 k(M) Therefore H 2 k(MrM) the Novikov group of m H2k (M;Q), where isomorphic group represent x the class of (V,3). w(M) This proves the result Proposition 2.2. (M, M = V. M) and the M U cone(DM), That is: [V,f] = + E W(Q). used in section 2: The homomorphisms w Witt 4k (t k > +W0, which are induced by the Witt class, 0, are surjective. 2 There is Corollary 3.3. and where q = 0 also the corollary: Every cobordism class in (mod 4), (M,TM) is has a representative Q M Witt , for q = M U cone q > 0 (TM), a q-dimensional manifold with boundary, obtained by plumbing disk bundles over spheres. I In particular, since M has a natural with only even dimensional strata, concrete result about ev stratification we obtain the following 117 The homomorphism of graded cobordism groups Corollary 3.4. Qev (t QWitt (t is surjective. Remark 3.5.: Simple examples show that the homomorphism in 3.4 is not injective. Section 4: ko* Suppose that a, of p.1. (1) a geometric cycle theory for Witt spaces: 2Z[% ]. Q* is a bordism theory based on a class spaces satisfying the following properties: Or is closed under the operations of taking cartesian product with a p.l. manifold, and intersecting transversely with a closed p.l. manifold in Euclidean space. (2) Y has a signature invariant defined on the cycle level which is cobordism invariant: There is a homomorphism sign : Q* (pt) 2Z which extends the identification Q*(pt) -+ 2Z. 118 relative cycles The signature can be extended to (3) (X,3X), so that it is additive: If X,Y sign (X UZ then z Y) = sign (X, Z) The signature (4) Z = are relative cycles such that is 3X,-Z = Y, + sign (Y, Z). multiplicative with respect to closed manifolds: If X is a cycle in t1, and M is a closed manifold, then: sign (M sign Then, x X) (M) coefficient denotes a natural theories: is [2Z[]. Z/4 2Z -periodic. The groups are: ko q(pt) 0 2Z[%] ~ Z [4] 0 Sullivan shows that q = 0 signature the classical -- + ko, @ ko* 0 2Z [2] where has defined [24] of homology p: that sign (M) -sign (X) , Dennis Sullivan transformation Recall = (mod 4), for if q : 0 (mod 4) otherwise. [X] e Q q (pt), where of M. 119 E 2Z[] In chapters III and IV, (1) through (4) hold koq (pt) 7Z[ ] . Ppt([X]) = sign(X) we demonstrated that properties for the class of Witt spaces, so we Witt Witt ko,* -* In Chapter V and Chapter VI, computed QWitt (pt). Recall 2 Z [%]12 . receive a natural transformation: sections 1 and 3, there are canonical we iso- morphisms: Witt pt)~ Witt Qq (pt) q Witt Qqit(pt) q and Recall from (11.2.8) W(0) if 0 otherwise. T The projection is a direct Witt , (mod 4) and 4) T, summand: ZZ ~ q > isomorphism sum of 2-groups of order 2 onto the q - 0 the canonical W(0) where q and 4. 0 120 when q > Therefore, 0, (mod 4), is homomorphism. the signature the natural transformation Witt (5) q - 0 ® 2[ ] : Witt 1t O 2Z[ ] -~+ko, 0 2Z[ ] induces an isomorphism on coefficient groups. We have proved: Theorem 4.1: The natural transformation (5) is an equivalence of homology theories. I For details of Sullivan's construction, Appendix. see the 121 Appendix: Sullivan's construction of ko* 0 Z[ ] orientations. In chapter VI, p we defined a natural transformation Witt : Witt -, ako, 0 2Z[%] (X by assigning to every Witt space orientation class in koq(XX) , X) 0 Z[%]. a canonical What is needed to carry out this assignment is the alchemy of Dennis Sullivan: Given a geometric homology theory cycles Y, properties with signature invariant, in section listed data into signature * for 1-cycles. canonical Specifically, VI.4, based on a class of, and satisfying the the Sullivan translates 2Z[4-] ko* orientations he constructs a natural transformation of homology theories: 11Y: Q -- ko* [% such that the induced homomorphism on coefficient groups is the signature homomorphism. The construction is found in may be unavailable to the reader, [22] and and [24] [24]. Since [22] will almost certainly be so, we sketch the arguments here. The first ingredient is the following algebraic result, which Sullivan calls Pontrjagin-duality for KO-theory. Theorem: [Sullivan] group (X) KO 0 2[z] Let X be a finite complex. The is naturally isomorphic to the group 122 of commutative diagrams: a KO . (X; 0) 1I KO.(X;Q/ where ]) - b C/ [ of coefficient groups, i be specified. diagrams [ is the map on homology induced by the homo- i* morphism and a and T are to (The group structure on the collection of is the obvious one). Sullivan constructs an exact sequence: Sketch of proof: (2) ji , (1) 0 -+KO 2Z[4] 3--KO^i (X) ) KO (X;Q) -+KO^ (X) 0 -+0 (X) from the arithmetic square: 2Z Here 2Z - = lim Ikt [%2]^ zz 2[%]^, 2Z [%] /n2Z[], 0 . 2[ ] -' Q where the inverse limit n odd taken over the direct system of odd positive integers, is 123 partially ordered by the relation: if m/n . m<n The corresponding homomorphism: 2Z[ k]/n2Z [ .- ] Similarly, KO 2Z [ ]I/m2Z[ ]i m. KO i(X; 2Z [ (X) ]/nZ [ ] ) . is reduction modulo -!+ n odd The inclusion: 2Z [%]antu/nZ [%map] ohmo o/Z v+ ) KO . (X; 2Z [%]/n2Z [%- KO . (X; (Q/2Z [ %]l By composition with the homomorphism T ) induces a natural map on homology: in the diagram, we get a collection of homomorphisms: T n : KO L (X; 2Z [-1-/n2Z [%2-]) -+p 2Z[ ]121/n2Z [ -2] C /2Z[%] 124 Now, there is a canonical isomorphism of cohomology theories induced by evaluation KG (X;Z[%]/nZ[%]) Tn n odd: P-Hom(KO. (X;2Z[%I/n2Z[3]),2Z[ This isomorphism is discussed in regard each for (cap product) [22]. as an element of KO T = 1iM T 4 thereby obtain an element ni ]/n2Z[%]) We can therefore (X;ZZ [/nZZ%]) of and KO(X). Using the isomorphism obtained from the universal coefficient theorem for KO-theory, namely: KO we identify KO ^ (X;Q). a (X;Q) -Hom(KO.(X; Q),Q), in the diagram with a unique element of The commutativity of the diagram implies that 0 a e KO (X) @ KO (X;0) in the short exact sequence associate a unique element is sent to (2). v E KO 0 in KO (X) 0 Therefor-, we can (X) 0 Z[Z] with the diagram (1). The.next result, also Sullivan's, enables us to translate the above isomorphism into more geometric terms. It is a Conner-Floyd type theorem (cf. cobordism to KO-theory at odd primes. [9]) relating smooth 125 [Sullivan]: Theorem periodic theories: so Q* where (XA) Sketch of proof: KO f : L MSO(4q) A Q 2Z[ ],;21 and where pair has 2Z[k] MSO(4r) of A4k with the property = the universal is KO* (XIA) Sullivan constructs an element -1 ch(A 4 k 0 C) where -+ structure induced by the signature. [] Q (MSO(4k)) 2Z [12] a compact p.1. is (X,A) (p t 'FQ Q*(pt)-module the is a canonical isomorphism of ) Z/42Z There L- 1 e H*(BSO(4k);q)), Moreover, if L-class. is -*MSO(4(q+r)) the natural map, then f*A 4 (q+r) ">x" A4q x A4 r, where denotes the external product in KO-cohomology. Proceeding as in [9], one uses A to define a natural transformation of cohomology theories Q*O(XA) KO* (XA) G) 2Z[%1-1, 126 or, a natural by Alexander duality, transformation of homology theories: so (XA) KO* (X, A) A E 2Z [%] such that A so KO 4k (Pt) Q4k (pt) the is 0 2Z [ ]2 ~ 2Z[] homomorphism: signature [M] F- sign(M). (That is, Q4k(pt) is so =~ -4k A* -4k SO -1, (pt) (pt) KO ph Q 2Z [-] 2Z[% homomorphism). the signature By the multiplicativity of the elements deduces that, A 4 k' one in fact, there is a natural transformation of 2Z/42Z-periodic theories: so Q* C (XA) Q*,(pt) Z[ -- + KO*(XA) 0 Z[ Since this transformation induces an isomorphism on 127 coefficient groups, we conclude that it is an equivalence of theories. I From the two theorems above, we see that elements of KO (X) are commutative diagrams: 0 2Z[] y Q i+4*()0 ji* where are homomorphisms satisfying the cy,T "periodicity relations": f T((V -- X) x (M -+pt)) = sign(M).T(V G((V f -. X) x (M -+pt)) f = sign (M) -cy(V -. X) and The elements of Now, suppose homology theory be a p.l. of (XA) Q 2Z [-] (that is, in the relative are analogously described. case) X KO X). X that Q* a cycle in the geometric mentioned earlier. Let h : X v- ]RN embedding into a large Euclidean space such that has codimension 4k, X. is and let A canonical orientation be a regular neighborhood U y X in ko n (X) G 2Z [%] corresponds by Alexander duality to a canonical element of 128 ko 0 2Z (U, aU) That is, [4] it corresponds 0 4*(U,9U) 0 4 *(U,3 U;Q/ 2Z[ 2]) to a commutative "diagram": (3) [2] V/2Z which satisfies the periodicity relation. Sullivan described how to obtain such a diagram from the signature invariant (in Given and transversality. cl) [(M,DM),f] transversality theorem assume f (X) C M in (U ,U), Q ([8], [15]) the relative block implies that we can is a cycle with the same local as X, with dimension 4k, in Y, and has say. a signature That is, sign(f 1 (X)) signature, this procedure associates to a ([(M, M),f]). well-defined integer . a For T, it n Z r Define bordism of recall that 2Z/n2Z-manifolds. assign to a singular By relative T : Q(U, 3u;2Z/n2Z) Q*(-;2Z/n2Z) [M,g] in (3) to -+Z/nZ 2Z[%]. is, Using transversality 2Z/nZ-cycle a a a compatible odd, which we can then tensor with To do this, is a cycle [(MM),f] to define suffices collection of homomorphisms: for (X) and by cobordism invariance of the transversality again, be f- structure geometrically, as above, a 2Z/n2Z-cycle 129 g (X) Z/n2Z-coefficients in the bordism theory with associated to If Q 1.8). (see obtained by identifications on the boundary, sign(g as signature invariant of we define: of the Additivity guarantees that the correspondence: Y -+ Q*(u , DU; 2Z/n2Z) [R,g] I- Z/n2Z sign(g is a well-defined homomorphism. Let (X)) T = lim 4- T n odd of the signature By multiplicAtivity manifolds, it is clear that relations. T commutes. Finally, is sign(Y), Z/nZ-manifolds. in the case of Tn : - (X)) (x) g is the relative cycle from which Y a, T the diagram The definition (3) n 0 a;[']. with respect to satisfy the periodicity corresponding to of orientations a, for relative cycles proceeds similarly. c*(pX) where c : X -*Ppt Specializing = sign(X) satisfies: p Note that the orientation E kon (pt) ® Z[ ] is the collapse map. now to the bordism theory associated to 130 the class of Witt spaces, Theorem: Every Witt space a canonical X= the theorem: (X, aX) orientation class P of dimension E ko q > (X, X) 0 2Z1.%]. 0 has If then the homomorphism C* : ko (X) carries we have px to 0 Z[%] -- ko (pt) 0 Z [%] sign (X). a 131 References. 1. E. Akin, Manifold phenomena in the theory of polyhedra, (1969), p. 413-473. Trans. Amer. Math. Soc. 143 2. E. Akin, Stiefel-Whitney homology classes and bordism, Tr'ans. Amer. Math. Soc. 3. J. Alexander, G. Hamrick, (1975), p. 341-359. 205 and J. Vick, and maps of odd prime order, Soc. 4. 221 M.F. Atiyah, Analysis (1976), 1969, M.F. Atiyah and I.M. p. 73-84. Singer, The index of elliptic 87 (1968), p. 546-604. N.A. Baas, On bordism theory of manifolds with W. Browder, Math. Surgery Springer-Verlag, 8. Global (eds. Spencer and Iyanaga), University of singularity, 7. Math. 169-185. operators III, Ann. of Math. 6. Trans. Amer. The signature of fibre bundles, Tokyo Press, 5. p. forms Linking S. Scan. 33 (1973), p. 279-302. on Simply Connected Manifolds, New York and Berlin, Buoncristiano, C.P. Rourke, and B.J. 1972. geometric approach to homology theory, Soc. Lect. Note Series 18, A Sanderson, London Math. Cambridge Univ. Press, 1976. 9. P.E. Conner and E.E. Floyd, The relation between co- bordism and K-theory, No. 28, 10. P.E. Springer-Verlag Lecture Notes 1966. Conner and F. Raymond, A quadratic form on the 132 quotient of a periodic map, Semigroup Formum 7 (1974), p. 310-333. 11. R.M. Goresky and R. MacPherson, theory, Intersection homology to appear. (Announcement of results appears in: La dualitd de Poincar4 pour les espaces singuliers, C.R. Sci. Paris, 12. F. s4rie A, 284 New York and Berlin, D. Husemoller and J. Milnor, Springer-Verlag, 14. K. Janich, McCrory, Trans. 16. C. Mccrory, Symmetric Bilinear Forms, der Signatur von Mannig- (1968) , 6 p. Eigenschaft, 35-40. Cone complexes and p.1. Amer. Math. Soc. 207 Stratified general Symposium on Algebraic transversality, (1975), position, J. Milnor, 664, p. 141-145. A procedure for killing homotopy groups of differentiable manifolds, Proc. Symp. v. 18. 3 Proc. of and Geometric Topology, Springer-Verlag Lecture Notes No. 17. 1973. durcheine Additivitats Inventiones Math. C. 1966. New York and Berlin, Characterisierung faltigkeiten 15. 1549-1551. Hirzebruch, Topological Methods in Algebraic Geometry, Springer-Verlag, 13. (1977), p. (1961), J. Morgan and D. in Pure Math. p. 39-55. Sullivan, The transversality characteristic class and linking cycles theory, 19. Acad. A. Pfister, Ann. of Math. 99 (1974), in surgery p. 463-544. Quadratic Forms, Lecture notes by A.D. 133 McGettrick, 20. C.P. University of Cambridge, Rourke and B.J. Sanderson, Piecewise Linear Topology, York and Berlin, 21. E. Spanier, 1967. Introduction to Springer-Verlag, New 1972. New York, Algebraic Topology, McGraw-Hill, 1966. 22. D. Sullivan, Geometric Topology Notes, Cambridge, 23. D. Sullivan, Singularities MA, Part I, 1970. Singularities in spaces, Symposium II, Proc. of Liverpool Springer-Verlag Notes No. 209, p. 196-206. 24. D. Sullivan, MIT, unpublished (handwritten) notes. Lecture 134 Figure 1. 135 ( z ~ Y- V X, U 1I x.1 U 7u x[-I A z v,v Figure 2. 4-i- 136 12 i / CIL~ / -, -- / /Z A Figure 3. y 137 ACKNOWLEDGEMENTS It gives me great pleasure to thank Mark Goresky for all of the support he provided during the preparation of this thesis. His most obvious mathematical contribution is the remarkable intersection homology theory, my primary which he developed jointly with Bob MacPherson at Brown tool, University. In addition, he first described for me the and our many conversations problem which this thesis solves, Mark later showed were crucial in leading me to a solution. me Atiyah's example, which has become my favorite "motivation" for the definition of Witt spaces. His remarks also clarified the connection between additivity and cobordism invariance of the signature. I could go on. Finally, he commented helpfully on a swarm of preliminary drafts. On the less technical side, Mark generously shared with me his personal and mathematical wisdom, and gave unfailing encouragement at some critical junctures. I am also happy to express my gratitude to Ed Miller for the active interest he took in my work and mathematical. career. ideas, His broad understanding of topology, wealth of and willingness to discuss mathematics have been valuable and inspiring. And I warmly thank Professor James Munkres, teacher of mathematics I have encountered, the finest for introducing 138 me in such beautiful fashion to much of the topology I know. The Department of Mathematics has kindly permitted me to finance my graduate education with a teaching assistant- ship; I thoroughly enjoyed the experience of working with students. Finally, of its forms) I thank my parents for their support and Eugenia Parnassa, (in all without whom I would have finished this thesis six monthsearlier. 139 BIOGRAPHICAL NOTE The author was born September 12, 1953 in Berkeley, California. He grew up in Framingham, Massachusetts during its quaint pre-"Golden Mile" era, and attended Framingham North High School. In September 1971, he began studies at M.I.T., where he was elected to Phi Beta Kappa in 1974. B.S. in Mathematics in June 1975, his graduate work in topology. He graduated with a and stayed on at M.I.T. for He held a four year teaching assistantship in the Department of Mathematics. Lest he overstay his welcome at M.I.T., the author will spend next year as a Visiting Member at the Courant Institute of Mathematical Sciences in New York.