map.mpim-bonn.mpg.de

Intersection form - Manifold Atlas

This page has not been refereed. The information given here might be incomplete or provisional.

[edit] 1 Introduction

Let N be an oriented n-manifold. After Poincaré one studies the intersection number of transverse submanifolds (or immersions) in N. This was generalized to a bilinear intersection product

\displaystyle I_N=\cap_N=\cdot_N=\lambda_N \colon H_k(N;\Z)\times H_{n-k}(N;\Z) \to \Z

defined on the homology of N, when N is a PL manifold (recall that any smooth manifold N has a triangulation making N a PL manifold). For n=2k this is the intersection form of N denoted by q_N. For n=4k and N closed the signature of this form is the signature \sigma(N) of N. The intersection product is closely related to the notions of characteristic classes and linking form. These are important invariants used in the classification of manifolds.

The exposition follows [Kirby1989, Chapter II], [Skopenkov2015b, \S6, \S10].

In this page N is a compact PL n-manifold (possibly, with boundary).

[edit] 2 A short direct definition of the intersection product

We use a simple definition of homology groups. In this section

Define the modulo 2 intersection product

\displaystyle  \cap_{N,2}: H_k(N) \times H_{n-k}(N) \to \Z_2\quad\text{by}\quad [x]\cap_{N,2} [y] := |x\cap y|\mod2,

where x and y are modulo 2 k-cycle in T and (n-k)-cycle in T^*.

Lemma 2.1. This product is well-defined.

Definition of the integer intersection product for oriented N. Take oriented dual faces \sigma of T and \sigma^* of T^* intersecting at a point S.

If T is a triangulation of a smooth manifold N, then N is contained in \R^d for some d. In the tangent space of N at S take a base of the tangent subspace corresponding to the orientation of \sigma. Take an analogous base for \sigma^*. If the ordered pair of these bases forms the orientation of N, the orientations on \sigma and on \sigma^* are said to be agreeing.

Assume that T is a triangulation of a PL manifold N. Denote k:=\dim\sigma. Let T' be the barycentric subdivision of T, one of whose vertices is S. Take an ordering (S,A_1,\ldots,A_k) of vertices of a k-face of T' contained in \sigma, corresponding to the orientation of \sigma. Analogously, take an ordering (S,B_1,\ldots,B_{n-k}) of vertices of an (n-k)-faces of T' contained in \sigma^*, corresponding to the orientation of \sigma^*. The vertices of the k-face and of the (n-k)-face form together an n-face of T'. Then (S,A_1,\ldots,A_k,B_1,\ldots,B_{n-k}) is an ordering of vertices of the n-face. If this ordering forms the orientation of N, the orientations on \sigma and on \sigma^* are said to be agreeing.

Analogously one defines agreeing orientations on faces of T and of T^* when T is a cellular decomposition.

Take agreeing orientations on faces of T and of T^*. In this definition we make summations over oriented k-faces \sigma of T. Take an integer k-cycle x = \sum_\sigma x_\sigma\sigma in T. Analogously, take an integer (n-k)-cycle y = \sum_\sigma y_{\sigma^*}\sigma^* in T^*. Define the integer intersection product

\displaystyle \cap_{N;\Z}:H_k(N;\Z)\times H_{n-k}(N;\Z)\to\Z \quad\text{by}\quad [x]\cap_{N;\Z}[y]:= \sum_\sigma x_\sigma y_{\sigma^*}.

Analogously to the modulo 2 case, the product of an integer k-cycle and a boundary of an (n-k+1)-face is zero. This and the PL invariance of homology imply that the integer intersection product is well-defined.

Remark 2.2. Using the notion of cup product, one can give a dual (and so an equivalent) definition:

\displaystyle I_N(x,y) = \langle x^*\smile y^*,[N]\rangle \in \Z,

where x^*\in H^{n-k}(N), y^*\in H^n(N) are the Poincaré duals of x, y, and [N] is the fundamental class of the manifold N. We can also define the cup (cohomology intersection) product

\displaystyle  I_N^*: H^k(N;\Zz) \times H^{n-k}(N;\Zz) \to \Zz \quad\text{by}\quad  I_N^*(p,q) = \langle p \smile q , [N] \rangle .

The definition of a cup product is `dual' (and so is analogous) to the above definition of the intersection product on homology, but is more abstract. However, the definition of a cup product generalizes to complexes (and so to topological manifolds). This is an advantage for mathematicians who are interested in complexes and topological manifolds (not only in PL and smooth manifolds). See [Skopenkov2005, Remark 2.3].

[edit] 3 Bilinearity and supersymmetry

The following properties are easy to check using the simple direct definition; they also follow from simple properties of the cup product.

The intersection product is bilinear. Hence it vanishes on torsion elements (for \Z-coefficients). Thus it descends to a bilinear (integer) intersection pairing

\displaystyle H_k(N;\Z)/\text{Torsion}\times H_{n-k}(N;\Z)/\text{Torsion} \to \Z.

on the free modules.

We have

\displaystyle I_N(x,y) = (-1)^{k(n-k)}I_N(y,x).

Hence for n=2k

[edit] 4 Superadditivity and transversal intersections

In this section the results and arguments work for both \Z_2- and \Z-coefficients, which we omit.

By the rank of a bilinear form on a \Z-module we mean its rank over \Q. We abbreviate

Tex syntax error
to
Tex syntax error
. E.g. if M is the connected sum of s copies of S^k\times S^k, then
Tex syntax error
; also
Tex syntax error
. The rank (of the intersection form) of manifolds is not additive. Indeed, if M_1=M_2=S^1 \times I and M_1\cup M_2=S^1 \times S^1, then
Tex syntax error
.

Lemma 4.1. Let M_1 and M_2 be compact orientable 2k-manifolds (possibly with boundary).

(a) (monotonicity) If M_1\subset M_2, then

Tex syntax error
.

(b) (superadditivity) Let M_1\cup M_2 be the union along some boundary components. Then

Tex syntax error
.

Proof. Part (a) is clear. Let us prove part (b). Let M_1' be the complement in M_1 to a collar of \partial M_1. Then by (a)

Tex syntax error

\square

Let V and W be k- and (n-k)-submanifolds of N. They are (more precisely, the pair V,W is) called transversal if for any x\in V\cap W there exists a closed neighborhood Ox of x in N, and a PL homeomorphism \varphi:Ox\to[-1,1]^n such that

\displaystyle \varphi(V\cap Ox)= [-1,1]^k\times0^{n-k} \quad\text{and}\quad \varphi(W\cap Ox)= 0^k\times[-1,1]^{n-k}.

Lemma 4.2. Let V and W be closed transversal k- and (n-k)-submanifolds of N. Then [V]\cap_N [W] = V\cdot W, where the right-hand part is the sum of signs of the intersection points of V,W for \Z-coefficients, and is the parity of |V\cap W| for \Z_2-coefficients.

Cf. Theorem 2.1 on intersection number of immersions. A simpler proof of Lemma 4.2 is given by

\displaystyle [V]\bigcap\limits_N [W] = [V\cap OW]\bigcap\limits_{OW} [W] = [V\cap OW]\bigcap\limits_{OV\cap OW} [W\cap OV] = V\cdot W.

Here

\displaystyle \cap_{OW}:H_k(OW,\partial OW) \times H_{n-k}(OW) \to R

\displaystyle \cap_{OV\cap OW}:H_k(OV\cap OW,\partial OW) \times H_{n-k}(OV\cap OW,\partial OV) \to R

\displaystyle \cap : H_k(B^k\times B^{n-k},S^{k-1}\times I^{n-k}) \times H_{n-k}(B^k\times B^{n-k},B^k\times S^{n-k-1}) \to R,

where R is either \Z or \Z_2, are defined analogously to the above;

  • the last equality holds by the transversality and because [B^k\times0]\cap[0\times B^{n-k}]=1.

[edit] 5 Poincaré duality

Theorem 5.1.[Poincaré duality] (a) The modulo 2 intersection product is non-degenerate.

(b) The integer intersection pairing is unimodular (in particular non-degenerate).

Proof of (a). (This proof is folklore, but in this short and explicit form is absent from textbooks.) Recall that T is a triangulation of a manifold N, and T^* is the dual cell subdivision. We use orthogonal complements with respect to the modulo 2 intersection product I_{T,2}:C_s(T)\times C_{n-s}(T^*)\to\Zz_2. It suffices to prove that

\displaystyle \phantom{}^\bot Z_{n-s}(T^*)= B_s(T)\quad\text{and}\quad Z_s(T)^\bot=B_{n-s}(T^*).

Let us prove the left-hand equality; the right-hand equality is proved analogously. Since I_{T,2} is non-degenerate, we only need to check that B_s(T)^\bot=Z_{n-s}(T^*). The inclusion B_s(T)^\bot \supset Z_{n-s}(T^*) is obvious. The opposite inclusion follows because if I_{T,2}(\partial c,d)=0 for an (s+1)-cell c of T and a chain d\in С_{n-s}(T^*), then \partial d does not involve the cell c^* dual to c.

[edit] 6 On classification of bilinear forms

Let q and q' be bilinear forms on free \mathbb{Z}-modules (or \Zz_2-vector spaces) V and V' respectively. The forms q and q' are called equivalent or isomorphic if there is an isomorphism f:V \to V' such that q=f^*q'.

[edit] 6.1 Invariants: rank, type, signature

The rank of a bilinear form q is the rank of the underlying \Z-module (or \Z_2-vector space) V.

A bilinear form q is even if q(x,x) is an even number for any element x. Equivalently, if q is written as a square matrix in a basis, it is even if the elements on the diagonal are all even. Otherwise, q is odd.

Let q be a symmetric bilinear form on a free \Z-module. Denote by b^+(q) (b^-(q)) the number of positive (negative) eigenvalues. Note that q is symmetric, so is diagonalisable over the real numbers, so b^+(q) (b^-(q)) is the dimension of a maximal subspace on which the form is positive (negative) definite.

The signature of q is defined to be

Tex syntax error

If n is divisible by 4, the signature

Tex syntax error
is defined to be the signature of the intersection form of N.

[edit] 6.2 Classification over integers modulo 2

Clearly, different sums from the theorem are not isomorphic.

[edit] 6.3 Classification of skew-symmetric forms

Theorem 6.2. Every skew-symmetric unimodular bilinear form q over \Zz is isomorphic to the sum of some number of hyperbolic forms H_-(\Zz) of rank 2 defined by the following matrix

\displaystyle \left( \begin{array}{cc} ~0 & ~1 \\ -1 & ~0 \end{array} \right).

In particular the rank of q is even.

Clearly, different sums from the theorem are not isomorphic.

[edit] 6.4 Examples of symmetric indefinite forms

A form is called definite if it is positive or negative definite, otherwise it is called indefinite. Here we show that

(O) for any pair (r,s) there is an odd unimodular symmetric indefinite form of rank r and signature s;

(E) for any pair (r,8s) there is an even unimodular symmetric indefinite form of rank r and signature 8s.

Cf. Theorem 6.3 and Proposition 6.4.

All values (O) are realised by direct sums of the forms of rank 1,

\displaystyle b^+ (+1) \oplus b^- (-1).

An even positive definite form of rank 8 is given by the E_8 matrix

\displaystyle  E_8 = \left( \begin{array}{c c c c c c c c} \ 2 \ & \ 1\  & \ 0\  & \ 0\ & \ 0\ & \ 0\ & \ 0\ & \ 0\ \\ 1 & 2 & 1 & 0 & 0 & 0 & 0 & 0 \\ 0 & 1 & 2 & 1 & 0 & 0 & 0 & 0 \\ 0 & 0 & 1 & 2 & 1 & 0 & 0 & 0 \\ 0 & 0 & 0 & 1 & 2 & 1 & 0 & 1 \\ 0 & 0 & 0 & 0 & 1 & 2 & 1 & 0 \\ 0 & 0 & 0 & 0 & 0 & 1 & 2 & 0 \\ 0 & 0 & 0 & 0 & 1 & 0 & 0 & 2 \\ \end{array}  \right) .

Likewise, the matrix -E_8 represents a negative definite even form of rank 8.

On the other hand, the matrix H given by

\displaystyle  H = \begin{pmatrix} \ 0 \ & \ 1 \ \\ 1 & 0  \end{pmatrix}

determines an indefinite even form of rank 2 and signature 0. It is easy to see that the direct sums

\displaystyle  k E_8 \oplus l H

with k,l\in\Zz, l>0 realise all forms from (E). Here we use the convention that k q is the k-fold direct sum of q for k>0 and kE_8 is the (-k)-fold direct sum of -E_8 for k<0.

[edit] 6.5 Classification of symmetric indefinite forms

The classification of unimodular definite symmetric bilinear forms is a deep and difficult problem. However the situation becomes much easier when the form is indefinite. Fundamental invariants are rank, signature and being odd or even (aka type). There is a simple classification result of indefinite forms [Serre1970],[Milnor&Husemoller1973]:

Theorem 6.3. (a) Every odd indefinite unimodular symmetric bilinear form over \Zz is isomorphic to the sum of some number of `unity' forms (1) and some number of `minis unity' forms (-1).

(b) Two indefinite unimodular symmetric bilinear forms over \mathbb{Z} are equivalent if and only if they have the same rank, signature and type.

Part (a) follows by part (b).

There is a restriction for values of the above invariants.

Proposition 6.4. The signature of an even (definite or indefinite) form is divisible by 8.

This follows from Proposition 6.5 below.

An element c \in V is called a characteristic vector of the form q if

\displaystyle  q(c,x) \equiv q(x,x) \ (\text{mod} \ 2)

for all elements x \in V. Characteristic vectors always exist. In fact, when reduced modulo 2, the map x \mapsto q(x,x) \in\Z/2 is linear. Hence by unimodularity there exists an element c such that the map q(c,-) equals this linear map.

The form q is even if and only if 0 is a characteristic vector. If c and c' are characteristic vectors for q, then by unimodularity there is an element h with c' = c + 2h. Hence the number q(c,c) is independent of the chosen characteristic vector c modulo 8. One can be more specific:

Proposition 6.5. For a characteristic vector c of the unimodular symmetric bilinear form q one has

\displaystyle  q(c,c) \equiv \text{sign}(q) \ (\text{mod} \ 8)

[edit] 7 References

[edit] 8 External links