. 9
( 9)

Let us begin by considering the effect of changing the orientation on This switches
the orientation on 77 and so our bundle map remains the same over BE - Nand
_ ˜ ). This has the effect of putting
over N it is multiplied by the matrix ( -˜
full twists into the framing around any circle that intersects F n V geometrically
t times where s == t (mod 2). Hence the class in HI (F; Z/2Z) that measures the
change in Pin- structure is just the class dual to F n V. If F n V has several
components and we switch the orientation of only one of them then the class
in HI (F; Z/2Z) that measures the change in Pin- structure is just the class dual
to that component of F n V.
eand on M - V. This time the
Now suppose that we switch the orientation on
aE by multiplication by the matrix ( - ˜ _ ˜) .
two bundle maps differ over all of
The effect of this is to change the Pin- structure on F via w}(F). This follows
from Lemma 1.6.
From the two results above the reader can work out the effect of the other
possible changes of orientations. Finally, the diligent reader should work through
the relative version.
This ends our description of the Pin- Structure Correspondence.•
As an application of the Pin- Structure Correspondence and Reduction 6.10
we present
Kirby & Taylor: Pin structures on low-dimensional manifolds 227

Theorem 6.11. There exists a homomorphism R: n˜ ˜ n˜˜2- (BO (2»). Given an
object, x E Q˜, let F denote the submanifold dual to W2 + wf. This manifold has a
map F ˜ BO(2) classifying the normal bundle. Use the above construction to put
a Pin- structure on F: R(x) is the bordism class of this Pin structure on F.
Variants of this map enter into the discussions below.
Corollary 6.12. If MFKr denotes the r-th bordism group of Freedman-Kirby,
then there exists a long exact sequence
nSpin i MF}r? R nSpin(B ) a nSpin
80(2) ---+˜'r-l --+ •••
• • • --+ ˜£r ·\.r---+˜'r-2

where R takes the Spin bordism class of the classifying map for the normal bundle
to F in M, and a takes the Spin structure we put on the total space of the associated
circle bundle. The V we always take is the empty set.
Remark 6.13. There are definitely non-trivial extensions in this sequence.
Remark 6.14. The Freedman-Kirby bordism theory is equivalent to the bordism
theory Spine, the theory of oriented manifolds with a specific reduction of W2 to
an integral cohomology class. This bordism theory has been computed, e.g. Stong
[Stong], and is determine˜d by Stiefel-Whitney numbers, Pontrjagin numbers, and
rational numbers formed from products of Pontrjagin numbers and powers of the
chosen integralization of W2.
Remark 6.15. There are versions of this sequence for the bordism theory studied
by Guillou-Marin and for our bordism theory. In both of these cases we replace
n Spin by the Pin- bordism groups nPin -. We also replace n;?:.;n(BSO (2») by the
bordism groups of O(2)-bundles over Pin- manifolds with some extra structure.
The bordism groups of O(2)-bundles over Pin-manifolds can be identified with
the homotopy groups of the Thom spectrum formed from BPin- X BO(2) using the
universal bundle over B pin - and the trivial bundle over BO(2). The associated
bordism groups are denoted n˜˜2- (B O (2»). In the Guillon-Marin case we define
BGM as the fibre of the map BPin- X B O (2) --+ K(Z/2Z, 1) where the map is the
sum of WI of the universal bundle over B pin - and WI of the universal bundle over
BO(2). In our case we let BE be the fibre of the map B pin - X B O (2) ˜ K(Z/2Z, 2)
where the map is the product of two I-dimensional cohomology classes: namely 0

WI of the universal bundle over B pin - and WI of the universal bundle over B O (2).
Over either BGM or BE we can pull back the universal bundle over BPin- plus the
trivial bundle over BO(2) and form the associated Thom spectrum. The homotopy
groups of these spectra fit into the analogous exact sequences for the bordism theory
studied by Guillou-Marin and by us.
Remark 6.16. All the bordism groups defined in Theorem 6.11, Corollary 6.12
and its two other versions are naturally modules over the Spin bordism ring, and
all the maps defined above are maps of n˜pin-modules.
228 Kirby & Taylor: Pin structures on low-dimensional manifolds

§7. Geometric calculations of characteristic bordism.
In this section we will calculate the characteristic bordism introduced in the
last section up through dimension 4.
The first remark is that any manifold M of dimension less than or equal to 4
has a characteristic structure. Hence !-bordism is onto unoriented bordism through
dimension 4. We show next that
Theorem 7.1. The forgetful map

ni ˜ 0.
f2b ˜ n˜ ˜ Z/2Z and
is an isomorphism for r = 0,1, and 2. Hence
Proof: Since the forgetful map is onto, it is merely necessary to show that the
!-bordism groups are abstractly isomorphic to Z/2Z or o. We begin in dimension
0. The only connected manifold is the point and it has a unique characteristic
structure: F and V are empty. Hence n˜ is a quotient of Z. It is easy to find a -:(
characteristic structure on [0, 1] which has 2 times the oriented point as its boundary:
F is empty and V = {1/2}. Hence ˜ Z/2Z given by the number of points mod 2.
In dimensions at least 1, it is easy to add I-handles to show any object is
bordant to a connected one. Hence in dimension 1, the only objects we need to
consider are characteristic structures on 51. Here F is still empty, and V is an even
number of points. The circle bounds B 2 , the 2-disk, and it is easy to extend V
to a collection of arcs in B 2 and to extend the orientation on 8 1 - V. The Pin-
structure on the circle either bounds a 2-disk, in which case extend it over B2, or
it does not, in which case take F to be a point in B 2 which misses the arcs and A

ni˜ 0.
extend the Pin- structure over B 2 - pt. Hence It
In dimension 2 we can assume that M is connected and that it bounds as
an unoriented manifold. The goal is to prove that it bounds as a characteristic
structure. Note V is a disjoint union of circles, and F is a finite set of points with
F nV being empty. Since every surface has a Pin- structure, F is an even number .:˜
of points. Let W be a collection of embedded arcs in M x [0,1] which miss M x 1
wi, ˜˜
and have boundary F. Since W is a dual to W2 + there is a Pin- structure
on M X [0,1]- W which extends across no component of W. This induces such a
structure on M x O. Since HI (M; Z/2Z) acts on such structures, it is easy to adjust :4
to get a Pin- structure on M x [0,1] - W which extends across no component of W

and which is our original Pin- structure on M x O. Given V C M x 0 we can extend

to an embedding V x [0, 1) in M x [0,1]. The orientation on M - V extends to one
on M x [0,1] - V x [0,1]. Clearly this orientation extends across no component of .:1

V x [0,1], so this submanifold is dual to Wt. Hence we may assume our surface has
empty F with no loss of generality: i.e. M has a fixed Pin- structure.
Let Ek denote the total space of the non-trivial 2-disk bundle over the circle.
The boundary of Ek is 1(2, the Klein bottle and HI (K; Z/2Z) is spanned by a fibre
Kirby & Taylor: Pin structures on low-dimensional manifolds 229

circle,C/, and a choice of circle which maps non-trivially to the base, Ct. Consider
the Pin-structure on K2 whose quadratic enhancement satisfies q(C/) = 2 and
q(Ct ) = 1. This structure does not extend across EK so let F be the core circle in
EK. Let V be a fibre 2-disk. Orient the normal bundle to V n F in F any way one
likes. It is easy to check that this gives a characteristic structure on Ek extending
,. the one on K2 which does not bound as a Pin- manifold. By adding copies of this
structure on K2 to M, we can assume that M is a Pin- boundary, so let W 3 be a
Pin- boundary for M.
Inside W we find a dual to WI, say X2, which extends V in M. There is some
orientation on W - X which extends across no component of X and this structure
restricts to such a structure on M - V. Since M is connected, there are only two
such structures and both can be obtained from such a structure on W - X. Hence
our original characteristic structure is a characteristic boundary assuming nothing
more than that it was an unoriented boundary.•
The results in dimensions 3 and 4 are more complicated. We begin with the
3-dimensional result.
Theorem 7.2. The homomorphism R of Theorem 6.11, followed by forgetting the
map to BO(2) yields an isomorphism

Proof: We first show that Ris onto and then that it is injective.
Let Ek denote the disk bundle with boundary the Klein bottle as in the-last
proof. The Pin- structure received by F in this structure is seen to be the Lie
group Pin- structure. There is a similar story for the torus, T 2 • There is a 2-disk
bundle over a circle, E}, and a Pin- structure on the torus which does not extend
across the disk bundle so that the core circle receives the Lie group Pin - structure.
Indeed, E} is just a double cover of Ek. IT we take two copies of K2 with its Pin-
structure and one copy of T 2 with its Pin- structure, the resulting disjoint union
bounds in nf in -. Let W 3 denote such a bordism. Let M 3 = JlE˜. II E˜ lL W 3
with the boundaries identified. Let F be the disjoint union of the three core circles,
and note F is a dual to W2 + wf since the complement has a Pin-structure which
does not extend across any of the cores. Let V be a dual to WI and arrange it to
meet F transversely. Indeed, with a little care one can arrange it so that V n F
consists of 2 points, one on each core circle in a Ef(. This is our characteristic
in - .
structure on M. Our homomorphism applied to M is onto the generator of
It remains to show monicity. Let M be a characterized 3-manifold. By adding
I-handles, we may assume that M is connected. First we want to fix it so that
V n F is empty. In general, V n F is dual to W2WI +wr and, for a 3-manifold, this
Kirby & Taylor: Pin structures on low-dimensional manifolds

vanishes. Hence V n F consists of an even number of points. We explain how to
remove a pair of such points.
Pick two points, po and PI, in V n F. Each point in F has an oriented normal
bundle. The normal bundle to each point in V is also trivial and V is oriented,
so the normal bundle to each point in V is oriented. Attach a I-handle, H =
(BI X [0, 1]) X B2 so as to preserve the orientations at po and Pl. Let W 4 denote the
resulting bordism. Inside W 4 , we have embedded bordisms, Vl and Fl beginning
at V and F in M. Notice that at the "top" of the bordism, the "top" of VI and
the "top" of F 1 intersect in 2 fewer points. Moreover, the orientation of the normal
bundle of V n F in F clearly extends to an orientation of the normal bundle of
Vi n F I in Fl.
Since F 1 is a codimension 2 submanifold of W, it is dual to some 2-dimensional
cohomology class. Since H* (W, M; Z/2Z) is 0 except when * = 1 (in which case
it is Z/2Z), this class is determined by its restriction to H2 (M; Z/2Z). Hence
F I is dual to W2 + so choose a Pin - structure on W - F I which extends
across no component of Fl. This restricts to a similar structure on M, and since
HI (W; Z/2Z) --. HI (M; Z/2Z) is onto, we can adjust the Pin- structure until it
extends the given one on M - F.
The above argument does not quite work for Vi, but it is easy in this case to see
that W - Vi has an orientation extending the one on M - V. Any such orientation
can not extend over any components of VI. Hence we have a characteristic bordism
as required.
We may now assume that V n F is empty. Since F is a union of circles and
V n F = 0, F has a trivial normal bundle in M. If our homomorphism vanishes
on our element, F is a Pin- boundary, which, in this dimension, means that it is
a Spin boundary: i.e. F bounds Q2, an orientable Pin-manifold. Glue Q2 x B 2
to M x [0, 1] along F X B 2 C M x 1 to get a bordism X 4 • Since Q is orientable,
V x [0, 1] is still dual to WI, and it is not hard to extend the Pin-structure on
M - F to one on X - Q which extends across no component of Q. Since Q and
V x [0, 1] remain disjoint, the "top" of X is a new characteristic pair for which the
dual to W2 + wi is empty: i.e. the "top", say N3, has a Pin- structure. Since
nfin- = 0, N3 bounds a Pin- manifold, y4. Since M was connected, so is Nand
there is no obstruction to extending the dual to WI in N, say VI, to a dual to WI
in Y, say U, and extending the orientation on N - Vi to an orientation on Y - U
which extends across no component of U. The union of X 4 and y4 along N3 is a
characteristic bordism from M 3 to o.•
The last goal of the section is to compute n˜. Since the group is non-zero,
we begin by describing the invariants which detect it. Given an element in n˜,
we get an associated surface F2 with a Pin - structure, and hence a quadratic
enhancement, q. We may also consider 1], the normal bundle to F in our original
4-manifold. We describe three homomorphisms. The first is {3: n˜ -+ Z/8Z which
Kirby & Taylor: Pin structures on low-dimensional manifolds 231

just takes the Brown invariant of the enhancement q. The second homomorphism
01 ˜ Z/4Z given by the element Q(Wl("'» E Z/4Z. The third homomorphism
is W:
w2:n1 ˜ Z/2Z given by (W2(1J), [F]) E Z/2Z. We leave it to the reader to check
that these three maps really are homomorphisms out of the bordisID group, n˜.
Theorem 7.3. The sum of the homomorphisms

f3 ED WEf) W2: n˜ ˜ Z/8Z EB Z/4Z EB Z/2Z
is an isomorphism.

Proof: First we prove the map is onto and then we prove it is 1-1. Recall from
Lemma 6.7 that a surface, M, with a Pin- structure and a 2-plane bundle, 7], can
be completed to a characteristic bordism element iff (WI (M) + WI (7]» U WI (7]) = O.
Notice that this equation is always satisfied since cupping with WI (M) and squaring
are the same. Hence we will only describe the surface with its Pin- structure and
the 2-plane bundle.
First note that Rp2 with the trivial 2-plane bundle generates the Z/8Z and
maps trivially to the Z/4Z and the Z/2Z.
The Hopf bundle over the 2-sphere maps trivially into the Z/8Z and the Z/4Z
since 52 is a Pin- boundary and \II vanishes whenever the 2-plane bundle has
trivial WI. However, 8 2 and the Hopf bundle maps non-trivially to the Z/2Z.
Let [(2 denote the Klein bottle, and fix a Pin- structure for which K2 is a
Pin- boundary. Let 1] be the 2-plane bundle coming from the line bundle with
WI being the class in HI (K2; Z/2Z) with non-zero square. Since K2 is a Pin-
boundary, (3(K2) = O. Since TJ comes from a line bundle, W2(TJ) = O. However,
Q(Wl(7]» is an element in Z/4Z of odd order and is hence a generator.
This shows that our map is onto. Before showing that our map is 1-1, we need
a lemma.
Lemma 7.4. There exists a 2-disk bundle B 2n over the punctured 8 1 x 52, 8 1 X
S2 - int B 3 , whose restriction to the boundary S2 has Euler class 2n, n E Z.

Proof: Start with the 2-disk bundle En over 8 2 with Euler number n and pull it
back over the product 8 2 xl. Now add a I-handle to 8 2 xl, forming 8 1 xS 2 -int B 3 ,
and extend the bundle B", over the I-handle so as to create a non-orientable bundle
B 2n • Then X(B 2n l s 2) = 2n.•
Suppose M 4, V 3, F2, TJ2 is a representative of an element of n˜ and that
(3(F 2) = 0, 'I1(W] (1]» = 0, and W2(7J) = 0. We need to construct a !-bordism to 0.
Since we may assume that F, M and V are connected, there is a connected
I-manifold, an 8 1 , which is Poincare "dual to Wl(7]); then the normal vector to
8 1 in F makes an even number of full twists jp. the Pin- structure on F as 8 1
is traversed. It follows that we can form a !-bordism by adding to F a B 2 x B l
Kirby & Taylor: Pin structures on low-dimensional manifolds

where 51 x B l is attached to the dual 8 1 to wl(71) and its normal Bl bundle.
Clearly the Pin- structure on F extends across the bordism. Since the dual to 8 1
has self-intersection zero in F, 71 restricted to 51 is orientable, so 1] extends over
B2 x Bl •
Since W2(71) = 0, it follows that X(7])[F] = 2n for some n E Z. By Lemma 7.4
there is a bundle B_ 2n over a punctured SI x 52 with x(B- 2n l s 2) = -2n. We form
a 5-dimensional bordism to the boundary connected sum, i.e. in M 4 X 1 C M 4 xl,
choose a 4-ball of the form B 2 x B 2 where B2 x 0 C p2 - (V n F) and p x B 2
is a normal plane of 71 over p, and identify B2 X B2 with B-2nls˜ where 5˜ is a
hemisphere of 52.
The new boundary to our !-bordism, which we shall denote (M, V, F, 1]) now
has a trivial normal bundle 71.
Since I3(F2) = 0, F Pin- bounds a 3-manifold N 3 , so we add N 3 x B 2 to
M x 1 along the normal bundle TJ to F, F X B 2 , where it does not matter how
we trivialize 71. The Pin- structure on M - F extends over the complement of N
(using the Pin- Correspondence Theorem, 6.9, and the Pin- structure on N), so
the new boundary to our !-bordism consists of a Pin- manifold M with empty F2.
Since'4-dimensional Pin- bordism, nf in -, is zero, we can complete our !-bordism
by gluing on to M x 1 a 5-dimensional Pin - manifold. •

Remark. It is worth comparing this argument with the argument in [F-I(J showing
that if (M 4 , P2) is a characteristic pair with M 4 and F 2 orientable and with
sign(M 4 ) = 0 and F.F = 0, then (M, F) is characteristically bordant to zero.
The arguments would have been formally identical if we had also assumed that the
Spin structure on F, obtained from the Pin- Correspondence Theorem, bounded
in 2-dimensional Spin bordism, pin = Z/2Z (corresponding to f3(F) = 0 above).
However, it is possible to show that n˜har = Z EB Z without the extra assumption
on F, and this ZJ2Z improvement leads to Rochlin's Theorem (see (F-K], [[{i}, ...).

Further Remark. The image of the Guillou-Marin bordism in this theory can be
determined as follows. The group is Z E9 Z generated by (54, RP2) and (C p2 , S2 ).
Both f3 and '¥ vanish on (CP2, S2), but W2 is non-zero. On (S4, RP2), W2 evaluates
o(the nonnal bundle comes from a line bundle): f3 is either 1 or -1 depending on
which embedding one chooses. Moreover, '11 is either 1 or -1 (the same sign as (3).

§8. New knot invariants.

The goal here is to describe some generalizations of the usual Ad invariant of
a knot (or some links) due to Robertello, [R].
We fix the following data. We have a 3-manifold M 3 with a fixed Spin structure
and a link L: JJ-51 -+ M 3 • Since M is Spin, w2(M) = 0 and we require that [L] E

HI (M; Z/2Z) is also 0, hence dual to w2(M). We next require a characterization of
˜. ·˜.7·'''' - -.,. ......., ....... ,˜.,.
Kirby & Taylor: Pin structures on low-dimensional manifolds 233

the pair, (M, L): i.e. a Spin structure on M - L which extends across no component
of L. We call such a characterization even iff the Pin- structure induced on each
component of L by Lemma 6.2 is the structure which bounds. We say the link is
even iff it has an even characterization.
One way to check if a link is even is the following. Each component of L has a
normal bundle, and the even framing of this normal bundle picks out a mod 2 lon-
gitude on the peripheral torus. The link is even iff the sum of these even longitudes
is 0 in H 1 (M - L; Z/2Z)
a are even: the Hopf link in S3 is an
Remark. Not all links which represent
example where any structure which extends across no component of L induces the
Lie group Spin structure on the two circles. We shall see later that a necessary and
sufficient condition for a link in S3 to be even is that each component of the link
should link the other components evenly. This is Robertello's condition, [RJ.

Definition. A link, L, in M 3 with a fixed Spin structure on M and a fixed Spin
structure on M - L which extends across no component of L and induces the
bounding Pin- structure on each component of L is called a characterized link.

Given a characterized link, (M, L), we define a class i E HI (M - L; Z/2Z): 1
is the class which acts on the fixed Spin structure on M - L to get the one which is
the restriction of the one on M. The class i is defined by the characterization and
conversely a characterization is defined by a choice of class i E HI (M - L; Z/2Z) so
that, under the coboundary map, the image of i in H2 (M, M - L; Z/2Z) hits each
generator. (Recall that by the Thorn isomorphism theorem, H2 (M, M - L; Z/2Z)
is a sum of Z/2Z's, one for each component of L.)
Let E be the total space of an open disk bundle for the normal bundle of L, and
let S be the total space of the corresponding sphere bundle. Note S is a disjoint
union of a peripheral torus for each component of L. The class '1 is dual to an
embedded surface F C M - E and 8F n S is a longitude in the peripheral torus
of each component of L. Let l denote this set of longitudes. We will call.e a set of
even longitudes. We will call F a spanning surface for the characterized link.
The set of even longitudes is not well-defined from just the characterized link. It
is clear that two surfaces dual to the same / must induce the same mod 2 longitudes.
But if we act on one component of L by an even integer, we can find a new sutface
dual to 1 which has the same longitudes on the other components and the new
longitude on our given component differs from the old one via action by this even
integer. Hence the characteristic structure only picks out the mod 2 longitudes and
any set of integral classes which are longitudes and which reduce correctly mod 2
can be a set of even longitudes. Moreover, any set of even longitudes is induced by
an embedded surface.
Since M is oriented, the normal bundle to any embedded surface, F, is isomor-
phic to the determinant bundle associated to the tangent bundle of F. The total
. ..."..
Kirby & Taylor: Pin structures on low-dimensional manifolds

space of the determinant bundle to the tangent bundle is naturally oriented. The
total space to the normal bundle to F is M is oriented by the orientation on M.
Choose the isomorphism between the normal bundle to F in M and the determi-
nant bundle to the tangent bundle of F so that, under the induced diffeomorphism
between the total spaces, the two orientations agree. Under these identifications,
Corollary 1.15 picks out a Pin- structure on F from the Spin structure on M. We
apply this to an F which is a spanning surface for our link. Of course we could apply
the same result but use the Spin structure on M - L. It is not hard to check that
the two structures on F differ under the action of wl(F) since this is the restriction
of / to F. Hence it is not too crucial which structure we use but to fix things we
use the structure on M.
We can restrict this structure on F to a component of L. If we put the Spin
structure on F coming from that on M - L it is easy to see that we get the bound-
ing Pin- structure on each component of L. Hence this also holds for the Pin-
structure on F coming from the one on M. Hence, a spanning surface for a char-
acterized link has an induced Pin- structure which extends to the corresponding
closed surface uniquely.
Our link invariant is a mod 8 integer which depends on the characterized link
and the set of even longitudes.
Definition 8.1. Given a characterized link, (M, L), and a set of even longitudes,
f, pick a spanning surface F for L which induces the given set of longitudes. Then
f3( L, f, M) = (3(F)
where F is F with a disk added to each component of L; the Pin- structure is
extended over each disk; and f3 is the usual Brown invariant applied to a closed
surface with a Pin-structure.
i) Notice that unlike Robertello's invariant, our invariant does not require that
the link be oriented.
ii) It follows from the proof of Theorem 4.3 that a knot is even iff it is mod 2
iii) IT each component of L represents 0 in HI (M; Z/2Z) then the mod 2 linking
number of a component of L with the rest of the link is defined. If F is
an embedded surface in M with boundary L, the longitude picked out for a
component of L is even iff the mod 2 linking number of that component of L
with the rest of the link is O.
iv) IT M is an oriented Z/2Z homology 3 sphere, then it has a unique Spin structure
and there is a unique ˜ay to characterize an even link L.
v) Let M be an integral homology 3 sphere containing a link L. Orient each
component of the link. Let fi be the linking number of the ith component of L
Kirby & Taylor: Pin structures on low-dimensional manifolds 235

with the rest of the link. Each component of L has a preferred longitude, the
one with self-linking 0, so li also denotes a longitude. The link L is even iff
each ti is even. Robertello's Arf invariant is equal to f3( L, -l, M), where the
Spin structure and characterization are unique and t is the set of longitudes
obtained by using -ti on each component. Notice that ti depends on how the
link is oriented.

It is not yet clear that our invariant really only depends on the characterizations
and the even longitudes.

Theorem 8.2. Let L be a link in a 3-manifold M. Suppose M has a Spin structure
and that L is characterized: Let t be a collection ofeven longitudes. Then (1(L, l, M)
is well-defined. Let W 4 be an oriented bordism between M 1 and M 2 • Let Li C Mi,
i = 1,2 be characterized links. Let FeW be a properly embedded surface with
F n Mi = Li. Suppose W - F bas a Spin structure which extends across no
component of F and which gives a Spin bordism between the two structures on
Mi - L j , i = 1,2, given by the characterizations.
The normal bundle to F in W has a section over every non-closed component
of F so pick one. This choice selects a longitude for each component of each link.
Suppose the longitudes picked out for each Li, say ti, are even. The surface F
receives a Pin- structure by Lemma 6.2. With this structure, each component of
8F bounds and hence F has a f3 invariant. H we orient W so that M 1 receives the
reverse Spin structure then tbe following. formula bolds.

Proof: We begin by discussing some constructions and results involving 3-
manifold N and a spanning surface, V for a characterized link, L. To begin, given
e: V 2 C N 3 , define V C N x [0,1] as the image of e x I, where I: V --+ [0,1/2] is any
map with 1- 1 (0) = aVe If N has a Spin structure, N x [0, 1] receives one. The class
represented by [V, L] in H 2 (N x [0, 1], N x 0 JL N x 1; Z/2Z) ˜ HI (N x 0; Z/2Z)
is the same as that represented by [L] in HI (N x 0; Z/2Z). Hence it represents
o. Since w2(N X [0,1]) is also trivial, there is a Spin structure on N x [0, 1] - V
which does not extend across any component of V. Such structures are acted on
simply transitively by HI (N; Z/2Z), so it is easy t˜ construct a unique such Spin
structure which restricts to the initial one on N X 1.
We proceed to identify the Spin structure induced on N X 0 - L. Let X =
V x [0, 1] and embed two copies of V in the boundary so that 8X = V U V where
the union is along 8V thought of as 8V x 1/2. First observe that we can em-
bed X in N x [0, 1] so that ax is V c N x 0 union V x 1 = V. Since X
has codimension 1, the Poincare dual to W is a I-dimensional cohomology class
Kirby & Taylor: Pin structures on low-dimensional manifolds

x E HI (N X [0,1] - V; Z/2Z). Suppose we take the Spin structure on N x [0,1]
and restrict it to N x [0,1] - V and then act on it by x. This is a Spin structure on
N x [0,1]- V which extends across no component of V and which is the original
one on N x 1. On N x 0- L it can be described as the one obtained by taking the
given Spin structure on N x 0, restricting it, and then acting on it by the restriction
of x. But the restriction of x is just the Poincare dual of FeN x so it is
the Spin structure which characterizes the link. By Lemma 6.2, there is a preferred
Pin- structure on V, which is easily checked to be the same as the one we put on
it in §4. The above Spin structure on N x [0,1] - V will be called the standard
characterization of the pair (N x [0, 1], V).
With this general discussion behind us, let us turn to the situation described
in the second part of the theorem. Recall W 4 is an oriented bordism between M I
and M 2 ; L 1 C M I and L 2 C M 2 are characterized links; F2 C W be a properly
embedded surface with FnMi = Li; and W -F has a Spin structure which extends
across no component of F and which gives a Spin bordism between the structures
on Mi - Li. Define sets of even longitudes f,j as in the statement of the theorem.
Let Fi C Mi be a spanning surface for L i • Inside W = M 1 X [-1, 0] U W U M 2 X
[0,1] embed F = F U F U F , where F is defined with function I: F 1 ˜ [-1/2,0]
I 2
and still 1- (0) = 8Ft • There is a Spin structure on W - F which extends across

no component of F. It is just the union of the standard characterization of M 1
[-1,0], F , the given Spin structure on W - F and the standard characterization
of M 2 x [0,1],F2 •
By Lemma 6.2 again, there is a preferred Pin- structure on F, which agrees
with the usual ones on F I and F 2 • In particular, F also receives a Pin- structure
which only depends on W, not on the choice of F I or F2 • However, from F 1 and
F2 , we see that the Pin- structure induced on each component of each link is the
bounding one. Moreover, f3(p) = f3(F) + f3(F2 ) - f3(F1 ).
By construction, F."F is 0, so 6.4 says that

where the J.l invariants arise because 6.4 only applies to closed manifolds.
Apply this to the case W = M x [0, 1], F = L x [0, 1] embedded as a product.
Since we may use different spanning surfaces at the top and bottom, this shows f3 is
well-defined. The formula in the theorem now follows from the formula immediately
The next thing we wish to discuss is how our invariant depends on the longi-
tudes. Given two different sets of even longitudes, i and i', for a characterized link
L C M 3 , there is a set of integers, one for each component of L defined as follows.
The integer for the ith component acts on the longitude for f, t6 give the longitude
for i'. Since both these longitudes are even, so is this integer.
Kirby & Taylor: Pin structures on low-dimensional manifolds 237

Theorem 8.3. Let L C M 3 be a characterized link with two sets of even longitudes
R. and it. Let 2r be the sum of the integers which act on the longitudes i to give
the longitudes i'. Then
+r (mod 8) .
f3(L,i',M) = f3(L,i,M)

Proof: Given F1 , a spanning surface for the longitude i, we can construct a spanning
surface for i' as follows. Take a neighborhood of the peripheral torus, which will
have the form W = T2 X [0,1]. Inside W embed a surface V which intersects
T2 X 0 in the longitude R., which intersects T2 x 1 in the longitude i', which has no
boundary in the interior of W; and which induces the zero map H 2 (V, av; Z/2Z) --+
H2 (W, aw; Z/2Z). The Spin structure on M restricts to one on W which is easily
described: it is the stabilization of one on T2 and this can be described as the
°on °on
one which has enhancement the longitude and the meridian. Since the
Pin- structure induced from Corollary 1.15 is local, we see that F 2 = V U F1 has
invariant the invariant for F1 plus the invariant for V. We further see that the
invariant for V only depends on the surface and the Spin structure in W. But
these are independent of the link and so we can calculate the difference of the f3's
using the unknot.
Furthermore, we see that the effect of successive changes is additive, so we only
need to see how to go from the 0 longitude to the 2 longitude, and the 2 longitude
is given by the Mobius band, which inherits a Pin- structure. This Pin- structure
extends uniquely to one on Rp2 and this Rp2 has (3 invariant +1.
Remark. Even in the case of links in S3, the longitudes used enter into the answer.
It is just in this case that there is a unique set of longitudes given by using an
orientable spanning surface.
Unfortunately, in general there is no natural choice of longitudes so it seems
simplest to incorporate them into the definition. The drawback comes in discussing
notions like link concordance. In order to assert that our invariant is a link concor-
dance invariant, we need to describe to what extent a link concordance allows us
to transport our structure for one link to another. Recall that a link concordance
between L o C M and L l C M is an embedding of (Jl.. SI) x [0,1] C M x [0,1] with
is (1L SI) X i being Li for i = 0,1. Suppose L o is an even link with f o a set of even
longitudes. There is a unique way to extend this framing of the normal bundle to
L o in M to a framing of the normal bundle of (Jl.. SI) x [0,1] in M x [0,1]. Hence
the concordance picks out a set of longitudes for L 1 which we will denote by R.I.
There is a unique way to extend a characterization of L o to a Spin structure on
M x [0,1] - (1L SI) x [0,1] and hence to M - L 1 •
Corollary 8.4. Let L o and L l be concordant links in M. Suppose L o is charac-
terized and that i o is a set of even framings. Then the transport of framings and
238 Kirby & Taylor: Pin structures on low-dimensional manifolds

Spin structures described above gives a cbaracterization of L 1 and £1 is a set of
even framings. Furtbermore P( L o,f o, M) = f3( L 1 , £1 , M).
Proof: The proof follows immediately from Theorem 8.2 and the fact that (ll
8 1 ) X [0, 1], when capped off with disks, is a union of 8 2 's and so has f3 invariant
We do know one scheme to remove the longitudes which works in many cases.
Suppose that each component of the link represents a torsion class in HI (M; Z).
Each component has a self-linking and by Lemma 4.1 the framings, hence longitudes
are in one to one correspondence with rational numbers whose equivalence class in
Q/Z is the self-linking number. There is a unique such number, say qi for the ith
component, so that qi represents an even framing and 0 ˜ qi < 2. We say that this
is the minimal even longitude. To calculate linking numbers it is necessary to orient
the two elements one wants to link, but the answer for self-linking is independent
of orientation.

Definition 8.5. Let L be a link in M so that each component of L represents a
torsion class in HI (M; Z). Suppose L is characterized. Define

= f3(L,i,M)

where /, is the set of even longitudes such that each one is minimal.

/3 is a concordance invariant.
Remark. It is not hard to check that

As we remarked above, (3 and /3 (if it is defined) do not depend on the orien-
tation of the link. H we reverse the orientation of M, and also reverse the Spin
structure on M and on M - L, it is not hard to check that the new Pin- structure
on F is the old one acted on by WI (F) so the new invariant is minus the old one.
The .remaining point to ponder is the dependence on the two Spin structures.
To do this properly would require a relative version of the (3 function 4.8. It does
not seem worth the trouble.

Remark. We leave it to the reader to work out the details of starting 'with a
characteristic structure on M 3 with the link as a dual to W2 + w˜ (i.e. represents 0
in HI (M; Z/2Z)).

§9. Topological versions.
There is a topological version of this entire theory. Just as Spin(n) is the
double cover of SO(n) and Pin±(n) are the double covers of O(n), we can consider
the double covers of STop(n) and Top(n). We get a group TopSpin(n) and two
groups TopPin±(n). A Top(n) bundle with a TopPin±(n) structure and an O(n)
structure is equivalent to a Pin±(n) bundle.
Kirby & Taylor: Pin structures on low-dimensional manifolds 239

Any manifold of dimension $ 3 has a unique smooth structure, so there is no
difference between the smooth and the toplogical theory in dimensions 3 and less.
The 3-dimensional bordism groups might be different because the bounding objects
are 4-dimensional, but we shall see that even in bordism there is no difference.
We turn to dimension 4. First recall that the triangulation obstruction (strictly
speaking, the stable triangulation obstruction) is a 4-dimensional cohomology class
so evaluation gives a homomorphism, which we will denote ˜, from any topological
bordism group to Z/2Z. Since every 3-manifold has a unique smooth structure,
the triangulation obstruction is also defined for 4-manifolds with boundary. Every
connected 4-manifold M 4 has a smooth structure on M - pt, and any two such
structures extend to a smoothing of M x [0,1] - pt x [0,1].
Some of our constructions require us to study submanifolds of M. In particular,
the definition of characteristic requires a submanifold dual to WI and a submanifold
dual to W2 + w˜. We require that these submanifolds be locally-fiat and hence,
by [Q], these submanifolds have normal vector bundles. Of course we continue to
require that they intersect transversely. Hence we can smooth a neighborhood of
these submanifolds. The complement of these smooth neighborhoods, say U, is a
manifold with boundary, which may not be smooth. If we remove a point from the
interior of each component of U, we can smooth the result. With this trick, it is not
difficult to construct topological versions of all our "descent of structure" theorems.
In particular, the [nw˜], [nWl] and R maps we defined into low-dimensional Pin±
bordism all factor through the corresponding topological bordism theories.
Theorem 9.1. Let Smooth-bordism. denoten;pin, n;in:l:, n˜, or the Freedman-
Kirby or Guillou-Marin bordism theories. Let Top - bordism* denote the topolog-
ical version. The natural map

Smooth - bordism3 Top - bordism3

is an isomorphism.
Smooth - bordism4 Top - bordism4-+Z/2Z 0
-+ -+

is exact.

Proof: The E s manifold, [F], is a Spin manifold with non-trivial triangulation
obstruction. Suppose M 3 is a 3-manifold with one of our structures which is a
topological boundary. Let W 4 be a boundary with the necessary structure. Smooth
neighborhoods of any submanifolds that are part of the structure. This gives a new
4-manifold with boundary U 4 • If the triangulation obstruction for a component of
U is non-zero, we may form the connected sum with the E s manifold. Hence we
may assume that U has vanishing triangulation obstruction. By [L-S] we can add
some 8 2 x 8 2 's to U and actually smooth it. The manifold W can now be smoothed
Kirby & Taylor: Pin structures on low-dimensional manifolds

so that all submanifolds that are part of the structure are smooth. Hence M 3 is
already a smooth boundary.
The E s manifold has any of our structures, so the map Top-bordism4 ˜ Z/2Z
given by the triangulation obstruction is onto.
Suppose that it vanishes. We can smooth neighborhoods of any submanifolds,
so let U be the complement. Each component of U has a triangulation obstruction
and the sum of all of them is O. We can add Es's and - E s 's so that each component
has vanishing triangulation obstruction and the new manifold is bordant to the old.
Now we can add some S2 x 5 2 's to each component of U to get a smooth manifold
with smooth submanifolds bordant to our original one.•

Theorem 9.2. The topological bordism groups have the following values. OTopSpin
˜ z; nropPin- ˜ Z/2Z; OrOpPin+ ˜ Z/8Z Ea Z/2Z; and n[op-! ˜ Z/8Z $ Z/4Z EB
Z/2Z e1 Z/2Z. The triangulation obstruction map is split in all cases except the
Spin case: the smooth to topological forgetful map is monic in all cases except
the TopPin+ case where it has kernel Z/2Z. The triangulation obstruction map
is split onto for the topological versions of the Freedman-Kirby and Guil1ou-Marin
theories and the smooth versions inject.
Proof: The TopPin- case is easy from the exact sequence above. The TopSpin
case is well-known but also easy. The Es manifold has non-trivial triangulation
obstruction and twice it has index 16 and hence generates nfpin.
There is a [nw˜] homomorphism from nropPin+ to nfin- ˜ Z/8Z which is
onto. Consider the manifold M = E S #S2 X RP2. The oriented double cover
of M is Spin and has index 16, hence is bordant to a generator of the smooth
Spin bordism group. It is not hard to see that the total space of the non-trivial
line bundle over M has a Pin+ structure, so the !{ummer surface is a TopPin+
boundary. Hence there is a Z/2Z in the kernel of the forgetful map and the [nw˜]
map shows that this is all of the kernel. Furthermore, E s represents an element of
order 2 with non-trivial triangulation obstruction.
The homomorphisms used to compute n˜ factor through nJop-!, so -! ˜

n˜ E9 Z/2Z.
Likewise, the homomorphisms we use to compute smooth Freedman-Kirby or
Guillou-Marin bordism factor through the topological versions.•

Corollary 9.3. Let M 4 be an oriented topological 4-manifold, and suppose we
have a characteristic structure on the pair (M, F). The following formula holds:

2· (J(F) = F.F - sign(M) +8· K(M) (mod 16)

where the Pin- structure on F is the one induced by the characteristic structure
on (M, F) via the topological version of the Pin-Structure Correspondence, 6.2.
Kirby & Taylor: Pin structures on low-dimensional manifolds 241

Proof: Generators for the topological Guillou-Marin group consist of the smooth
generators, for which the formula holds, and the E g manifold, for which the formula
is easily checked.•

Remark. The above formula shows that the generator of H 2 ( ; Z) of Freedman's
Chern manifold, [F, p. 378], is not the image of a locally-fiat embedded 8 2 •

[ABP1] D. W. Anderson, E. H. Brown, Jr. and F. P. Peterson, The structure of the
Spin cobordism ring, Ann. of Math., 86 (1967), 271-298.
, Pin cobordism and
related topics, Comment. Math. Helv., 44 (1969), 462-468.
[ABS] M. F. Atiyah, R. Bott and A. Shapiro, Clifford modules, Topology, 3 (Suppl.
1) (1964), 3-38. .
E. H. Brown, The Kervaire invariant of a manifold, in "Proc. of Symposia
in Pure Math.", Amer. Math. Soc., Providence, Rhode Island, XXII 1971,
[F) M. H. Freedman, The topology of four-dimensional manifolds, J. Differen-
tial Geom., 17 (1982), 357-453.
and R. C. Kirby, A geometric proof of Rochlin's theo-
rem, in "Proc. of Symposia in Pure Math.", Amer. Math. Soc., Providence,
Rhode Island, XXXII, Part 2 1978, 85-97.
[G-M] L. GuiDou and A. Marin, Une extension d'un theoreme de Rohlin sur 1a
signature, in "A la Recherche de la Topologie Perdue", edited by Guillou
and Marin, Birkhauser, Boston - Basel - Stuttgart, 1986, 97-118.
N. Habegger, Une variete de dimension 4 avec forme d'intersection paire et
signature -8, Comment. Math. Helv., 57 (1982), 22-24.
S. J. Kaplan, Constructing framed 4-manifolds with given almost framed
boundaries, Trans. Amer. Math. Soc., 254 (1979), 237-263.
[Ki] R. C. Kirby, "The Topology of 4-Manifolds", Lecture Notes in Math. No.
1374, Springer-Verlag, New York, 1989.
and L. R. Taylor, A calculation of Pin+ bordism groups, Com-
ment. Math. Helv., to appear.
[L-8] R. Lashof and J. Shaneson, Smoothing 4-manifolds, Invent. Math., 14
(1971), 197-210.
[Mat] Y. Matsumoto, An elementary proof of Rochlin's signature theorem and its
extension by Guillou and Marin, in "A la Recherche de la Topologie Per-
due", edited by Guillou and Marin, Birkhauser, Boston - Basel - Stuttgart,
1986, 119-139.

Kirby & Taylor: Pin structures on low-dimensional manifolds

J. W. Milnor and J. D. Stasheff, "Characteristic Classes", Annals of Math.
Studies # 49, Princeton University Press, Princeton, NJ, 1974.
[Q] F. Quinn, Ends of maps, III: dimensions 4 and 5, J. Diff. Geom., 17 (1982),
R. A. Robertello, An invariant of knot cobordism, Comm. Pure and App.
Math., XVIII (1965), 543-555.
V. A. Rochlin, Proof of a conjecture of Gudkov, Funkt. Analiaz. ego Pri!.,
6.2 (1972), 62-24: translation; Funct. Anal. Appl., 6 (1972), 136-138.
[Stolz] S. Stolz, Exotic structures on 4-manifolds detected by spectral invariants,
Invent. Math., 94 (1988), 147-162.
[Stong] R. E. Stong, "Notes'on Cobordism Theory", Princeton Math. Notes, Prin-
ceton Univ. Press, Princeton, New Jersey, 1958.
L. R. Taylor, Relative Rochlin invariants, Gen. Top. Appl., 18 (1984),
V. G. Turaev, Spin structures on three-dimensional manifolds, Math. USSR

Sbornik, 48 (1984),65-79.

Department of Mathematics
Department of Mathematics
University of California, Berkeley University of Notre Dame
Berkeley, California 94720 Notre Dame, Indiana 46556

UW 8102787 X



. 9
( 9)