Holomorphic Disks and Knot Invariants: Article in Press

ARTICLE IN PRESS
Advances in Mathematics 186 (2004) 58–116

http://www.elsevier.com/locate/aim
Holomorphic disks and knot invariants

Peter Ozsvátha,1 and Zoltán Szabób,,2
a
Department of Mathematics, Columbia University, New York, NY 10027, USA
b
Department of Mathematics, Princeton University, Washington Road, Princeton, NJ 08544, USA
Received 6 January 2003; accepted 20 May 2003
Communicated by Tomasz Mrowka
Abstract
We define a Floer-homology invariant for knots in an oriented three-manifold, closely

related to the Heegaard Floer homologies for three-manifolds defined in an earlier paper. We
set up basic properties of these invariants, including an Euler characteristic calculation, and a
description of the behavior under connected sums. Then, we establish a relationship with HF þ
for surgeries along the knot. Applications include calculation of HF þ of three-manifolds
obtained by surgeries on some special knots in S3 ; and also calculation of HF þ for certain
simple three-manifolds which fiber over the circle.
r 2003 Elsevier Inc. All rights reserved.
1. Introduction
The purpose of this paper is to define and study an invariant for null-homologous
knots K in an oriented three-manifold Y : These naturally give rise to invariants for
oriented, null-homologous links in Y ; since oriented, null-homologous links in Y
correspond to certain oriented, null-homologous knots in Y #n1 ðS2 S1 Þ (where n
denotes the number of components of the original link). In the interest of exposition,
we describe first some special cases of the construction in the case where the ambient
three-manifold is S3 —the case of ‘‘classical links.’’

Corresponding author.
E-mail addresses: petero@math.columbia.edu (P. Ozsváth), szabo@math.princeton.edu (Z. Szabó).
1
Supported by NSF Grant DMS 9971950 and a Sloan Research Fellowship.
2
Supported by NSF Grant DMS 0107792 and a Packard Fellowship.
0001-8708/$ - see front matter r 2003 Elsevier Inc. All rights reserved.
doi:10.1016/j.aim.2003.05.001
ARTICLE IN PRESS
! Z. Szabo! / Advances in Mathematics 186 (2004) 58–116
P. Ozsvath, 59
1.1. Classical links
Suppose Y DS3 ; and suppose L is an oriented link. In its simplest form, our
construction gives a sequence of graded Abelian groups HFK d
HFKðL; iÞ; where here iAZ:
If L has an odd number of components, the grading is a Z-grading, while if it has an
even number of components, the grading function takes values in 12 þ Z:
These homology groups satisfy a number of basic properties, which we outline
presently. Sometimes, it is simplest to state these properties for HFK d
HFKðL; i; QÞ; the
homology with rational coefficients: HFK d
HFKðL; i; QÞDHFK d
HFKðL; iÞ#Z Q:
First, the Euler characteristic is related to the Alexander–Conway polynomial of
L; DL ðTÞ by the following formula:
X
d
wðHFK
HFKðL; i; QÞÞ T i ¼ ðT 1=2 T 1=2 Þn1 DL ðTÞ; ð1Þ
where n denotes the number of components of L (it is interesting to compare this

with [1,6,15]). The sign conventions on the Euler characteristic here are given by
X n 1
d
wðHFK
HFKðL; i; QÞÞ ¼ d d ðL; i; QÞÞ:
ð1Þdþ22 rkðHFK
n1
dAð 2 ÞþZ
Our grading conventions are justified by the following property. If L% denotes the
mirror of L (i.e. switch over- and under-crossings in a projection for L), then
d d ðL; i; QÞDHFK
HFK % i; QÞ:
d d ðL; ð2Þ
Another symmetry these invariants enjoy is the following conjugation symmetry:
HFK d d2i ðL; i; QÞ;

d d ðL; i; QÞDHFK ð3Þ
refining the symmetry of the Alexander polynomial. Finally, the invariants remain
unchanged after an overall orientation reversal of the link,
HFK d d ðL; i; QÞ:

d d ðL; i; QÞDHFK ð4Þ
These groups also satisfy a Künneth principle for connected sums. Specifically, let
L1 and L2 be a pair of disjoint, oriented links which can be separated from one
another by a two sphere. Choose a component of L1 and L2 ; and let L1 #L2 denote
the connected sum performed at these components. Then,
M
d 1 #L2 ; i; QÞD
HFK
HFKðL d 1 ; i1 ; QÞ#Q HFK
HFK
HFKðL d 2 ; i2 ; QÞ
HFKðL ð5Þ
i1 þi2 ¼i
(see Corollary 7.2 for a more general statement). Of course, this can be seen as a
refinement of the fact that the Alexander polynomial of links is multiplicative under
connected sums.
ARTICLE IN PRESS
60 ! Z. Szabo! / Advances in Mathematics 186 (2004) 58–116
P. Ozsvath,
− 0 +
Fig. 1. Skein moves at a double-point.
‘ L1 and L2 be two links as above. Then, we can also form their disjoint union
Let
L1 L2 : We have that
a
d L1
HFK d 1 #L2 ; i; QÞ#W ;
L2 ; i; Q DHFK
HFKðL ð6Þ
where W is a two-dimensional graded vector space splitting into two one-

dimensional pieces
W DW1=2 "W1=2 ;
where W71=2 has grading 712: The above is a manifestation of the advantage of HFK d
over the Alexander polynomial: the Alexander polynomial of any split link vanishes.
These invariants also satisfy a ‘‘skein exact sequence’’ (cf. [4,7,12,22]). Suppose
that L is a link, and suppose that p is a positive crossing of some projection of L:
Following the usual conventions from skein theory, there are two other associated
links, L0 and L ; where here L agrees with Lþ ; except that the crossing at p is
changed, while L0 agrees with Lþ ; except that here the crossing p is resolved. These
three cases are illustrated in Fig. 1. There are two cases of the skein exact sequence,
according to whether or not the two strands of Lþ which project to p belong to the
same component of Lþ :
Suppose first that the two strands which project to p belong to the same
component of Lþ : In this case, the skein exact sequence reads:
?-HFK
HFKðL d 0 Þ-HFK
d Þ-HFK
HFKðL d þ Þ-?;
HFKðL ð7Þ
where all the maps above respect the splitting of HFK d

HFKðLÞ into summands (e.g.
d d
HFKðL ; iÞ is mapped to HFK
HFK d 0Þ
HFKðL0 ; iÞ). Furthermore, the maps to and from HFK
HFKðL
drop degree by 12: The remaining map from HFK d þ Þ to HFK
HFKðL d Þ does not
HFKðL
necessarily respect the absolute grading; however, it can be expressed as a sum of
homogeneous maps,3 none of which increases absolute grading. When the two
3
A graded group is one which splits as A ¼ "kAQ Ak : A homogeneous element is one which is
supported in Ak for some k: A map between graded groups f : A-B is said to be homogeneous of degree d
if f maps all the Ak into Bkþd : The map f is said to be homogeneous if it is homogeneous of degree d; for
some d:
ARTICLE IN PRESS
P. Ozsvath, 61
strands belong to different components, we obtain the following:

d Þ-HFK
?-HFK
HFKðL d 0 Þ#V -HFK
HFKðL d þ Þ-?;
HFKðL ð8Þ
where V denotes the four-dimensional vector space
V ¼ V1 "V0 "V1 ;
where here V71 are one-dimensional pieces supported in degree 71; while V0 is a
two-dimensional piece supported in degree 0: Moreover, the maps respect the
decomposition into summands, where the ith summand of the middle piece
d 0 Þ#V is given by
HFKðL
HFK
d 0 ; i 1Þ#V1 Þ"ðHFK
ðHFK
HFKðL d 0 ; iÞ#V0 Þ"ðHFK
HFKðL d 0 ; i þ 1Þ#V1 Þ:
HFKðL
The shifts in the absolute gradings work just as they did in the previous case.
In addition to establishing the above properties, we calculate this invariant for
some examples. For a more detailed discussion of the link invariants in the classical
case, we refer the reader to the sequel [23]. In particular, in that article, we determine
the invariants for alternating links. In a different direction, if L is a fibred link and d
denotes the degree of DL ðTÞ; then we have that HFK d d þ ðn1
HFKðL; 2 ÞÞDZ: This
statement is a generalization of the fact that the Alexander–Conway polynomial of a
fibered link is monic. We do not prove this fact in the present paper, but rather give
the proof in [21].
Suppose now that our link consists of a single component, i.e. it is a knot K: In
this case, the homology groups also give bounds on the genus g of the knot K: if
d
HFKðK;
HFK sÞa0; then
jsjpg:
Since the knot homology groups can be viewed as a refinement of the Alexander
polynomial, it is natural to ask whether they contain more information (as they did
in the case of links). In fact, it is easy to see that they do—as graded Abelian groups,
d distinguishes the right-handed from the left-handed trefoil. More interestingly,
HFK
building on the material in [23], we show in [24] that HFK d distinguishes all non-
trivial pretzel knots of the form Pð2a þ 1; 2b þ 1; 2c þ 1Þ from the unknot (note that
these include knots with trivial Alexander polynomial). Indeed, in that paper, it is
also shown that HFKd can distinguish knots which differ by a Conway mutation.
Moreover, we conjecture that the genus bounds stated earlier are sharp:
Conjecture 1.1. If KCS 3 is a knot with genus g; then

d
HFK
HFKðK; gÞa0:
Our evidence for this conjecture is based on three ingredients. The first is a
relationship between the knot Floer homology groups and Heegaard Floer
ARTICLE IN PRESS
P. Ozsvath,
homology, which is established in Section 4. Next, we appeal to the conjectured

relationship between Heegaard Floer homology and Seiberg–Witten Floer homology
(cf. [20]). And finally, non-triviality of the corresponding Seiberg–Witten objects is
established by work of Kronheimer–Mrowka [13], which in turn rests on work of
Gabai [8] and Eliashberg–Thurston [5].
1.2. General case
In the above discussion, we restricted attention to a rather special case of our

constructions, in the interest of exposition. As we mentioned, the constructions we
give here actually generalize to the case where the ambient three-manifold Y is an
arbitrary closed, oriented three-manifold and K is a null-homologous knot; and
indeed, the algebra also gives more information than simply the knot homology
groups. In their more general form, the constructions give a Z"Z-filtration on the
chain complex CF N ðY ; sÞ used for calculating the Heegaard homology HF N ðY ; sÞ
defined in [20] (where here s is any Spinc structure over Y ). This filtration also induces
a Z filtration on CFc ðY ; sÞ; and the knot homology groups we described above are the
homology groups of the associated graded complex (in the case where Y DS3 ).
The paper is organized as follows. In Section 2, we set up the relevant topological
preliminaries: the passage from links to knots, Heegaard diagrams and knots, and
some of the algebra of filtered chain complexes. In Section 3 we give the definition of
the knot filtration, and prove its topological invariance. In that section, we also
establish some of the symmetries of the invariants. In Section 4, we explain how the
filtration can be used to calculate the Floer homology groups of three-manifolds
obtained by ‘‘large surgeries’’ along the knot K: In Section 5, we derive an
d and the genus of the knot. In Section 6,
‘‘adjunction inequality’’ which relates HFK
we give some sample calculations for the knot homology groups of some special
classes of classical knots. With this done, we return to some general properties: the
Künneth principle for connected sums, and the surgery long exact sequences in
Sections 7 and 8, respectively. As an application of these general results, we conclude
with some calculations of HF þ ðY ; sÞ for certain simple three-manifolds which fiber
over the circle (in the case where the first Chern class of the Spinc structure s
evaluates non-trivially on the fibers). The three-manifolds considered here include
Y DS 1 Sg for arbitrary g; and also mapping tori of a single, non-separating Dehn
twist. Finally in Section 10 we point out how the results from the paper specialize to
the properties and formulae stated in Section 1.1.
We will return to other applications of these constructions in future papers. For
example, the constructions play a central role in [21], where they are used to define
invariants of contact structures for three-manifolds.
1.3. Further remarks
The fact that a knot in Y induces a filtration on CF N ðY Þ has been discovered

independently by Rasmussen [25], where he uses the induced filtration to compute
ARTICLE IN PRESS
P. Ozsvath, 63
the Floer homologies of three-manifolds obtained as surgeries on two-bridge knots.

In fact, we have learned that many of the constructions presented here have been
independently discovered by Rasmussen in his thesis [26].
In another direction, it worth pointing out the striking similarity between the knot
homology groups described here and Khovanov’s homology groups [12] (see also
[2]). Although our constructions here are quite different in spirit from Khovanov’s,
the final result is analogous: we have here a homology theory whose Euler
characteristic is the Alexander polynomial, while Khovanov constructs a homology
theory whose Euler characteristic is the Jones polynomial. It is natural to ask
whether there is a simultaneous generalization of these two homology theories.
2. Topological preliminaries
2.1. From knots to links
Although we discussed links in the introduction, we will focus mainly on the case
of knots in the paper. Our justification for the apparent loss of generality is the
observation that oriented n-component links in Y correspond to oriented knots in
Y #n1 ðS 2 S 1 Þ: For related constructions, see [10]; compare also [4,6].
More precisely, suppose that L is an n-component oriented three-manifold Y :
Then, we can construct an oriented knot in Y #n1 ðS2 S 1 Þ as follows. Fix 2n 2
points fpi ; qi gn1
i¼1 in L; which are then pairwise grouped together (i.e., n 1
embedded zero-spheres in L), in such a manner that if we formally identify each pi
with qi in L; we obtain a connected graph. We view pi qi as the feet of a one-handle to
attach to Y : Let kðY ; fpi ; qi gÞ denote the new three-manifold. Of course,
kðY ; fpi ; qi gÞDY #n1 ðS 2 S1 Þ: Now, inside each one-handle, we can find a band
along which to perform a connected sum of the component of L containing pi with
the component containing qi : We choose the band so that the induced orientation of
its boundary is compatible with the orientation of the link. Our hypotheses on the
number and distribution of the distinguished points ensures that the newly
constructed link (after performing all n 1 of the connected sums) is a single-
component knot, which we denote kðL; fpi ; qi gÞ; inside kðY ; fpi ; qi gÞ:
Proposition 2.1. The above construction induces a well-defined map from the set of
isotopy classes of oriented, n-component links in Y to the set of isotopy classes of
oriented knots in Y #n1 ðS 2 S1 Þ: More precisely, if L and L0 are isotopic links, an
fpi ; qi gn1 0 0 n1
i¼1 and fpi ; qi gi¼1 are auxiliary choices of points as above, then there is an
orientation-preserving diffeomorphism
kðY ; fpi ; qi gÞ-kðY ; fp0i ; q0i gÞ
which carries the oriented knot kðL; fpi ; qi gÞ to a knot which is isotopic to
kðL; fp0i ; q0i gÞ:
ARTICLE IN PRESS
P. Ozsvath,
Proof. First, we argue that for a fixed set of points fpi ; qi gn1
i¼1 in L as above, the
isotopy class of the induced knot in Y #n1 ðS2 S1 Þ is independent of the choices of
bands connecting the pi to qi (inside the one-handles). This can be seen using an
isotopy supported inside the one-handle.
Next, we argue that the isotopy class of kðLÞ is independent of the choice of points
fpi ; qi gn1
i¼1 :
The key observation is that if we have three components K1 ; K2 ; K3 ; and
distinguished points p1 AK1 with q1 AK2 ; and p2 AK1 and q2 AK3 ; then we can
handleslide the one-handle specified by fp2 ; q2 g over fp1 ; q1 g to obtain a new one-
handle fp3 ; q3 g; where p3 AK2 ; q3 AK3 (cf. Fig. 2).
We say that the points are arranged in a linear chain if for some ordering on the
components of L; fKi gni¼1 ; if for each i ¼ 1; y; n 1; we have that pi AKi and
qi AKiþ1 : First, we claim that given any initial collection of points fp0i ; q0i gni¼1 ; we can
pass to a linear chain. To see this, we let K1 denote a circle with the minimal number
of distinguished points on it. Clearly K1 has only one distinguished point, which we
denote p1 : By induction, we can find a sequence K1 ; y; Ki so that K1 contains only
one distinguished point, K2 ; y; Ki1 all contain only two distinguished points;
indeed, pi AKi and qi AKiþ1 for i ¼ 1; y; i 1: If Ki contains no other distinguished
pi ; then it follows from the connectivity hypothesis that i ¼ n; and we have arranged
all our points in a linear chain. Otherwise, we fix some pi AKi : Its corresponding point
qi lies on some other component of L which, according to the hypothesis, is not
among K1 ; y; Ki : We denote that circle by Kiþ1 : Again, by a sequence of handleslides
over this one-handle, we can push all other marked points from on Ki to Kiþ1 :
Finally, any two linear chains can be connected by handleslides as above.
Specifically, choose any linear ordering of the components of L; and consider the
induced linear chain. In the case where K1 ; K2 ; K3 are the first three components in
this linear ordering, the handleslide described illustrated in Fig. 2 induces the
permutation of K1 and K2 : More generally, it is easy to realize a permutation of two
consecutive components as a composition of two such handleslides. &
The above construction is more than a curiosity: it plays a central role in the
‘‘skein exact sequence’’ described in Section 8 below.
,
P1 P1 P2
K1 K1
K2 Q1 P2 Q1
K2
,
K3 Q2 K3 Q2
Fig. 2. Handleslides. An illustration of the handleslide used in Proposition 2.1. We slide the handle
specified by fp2 ; q2 g over that specified by fp1 ; q1 g to obtain the new one-handle specified by fp01 ; q01 g:
ARTICLE IN PRESS
P. Ozsvath, 65
2.2. Heegaard diagrams and knots
Definition 2.2. For us, a knot will consist of a pair ðY ; KÞ; where Y is an oriented
three-manifold, and KCY is an embedded, oriented, null-homologous circle.
A knot ðY ; KÞ has a Heegaard diagram
ðS; a; b0 ; mÞ;
where here a is an unordered g-tuple of pairwise disjoint attaching circles a ¼

fa1 ; y; ag g; b0 is a ðg 1Þ-tuple of pairwise disjoint attaching circles fb2 ; y; bg g; m
is an embedded, oriented circle in S which is disjoint from the b0 ; and, of course g is
the genus of S: This data is chosen so that ðS; a; b0 Þ specifies the knot-complement
Y ndðKÞ; i.e. if we attach disks to S along the a and b0 ; and then add a three-ball,
we obtain the knot-complement. Moreover, m represents the ‘‘meridian’’ for the knot
in Y ; thus, ðS; a; fmg,b0 Þ is a Heegaard diagram for Y :
Definition 2.3. A marked Heegaard diagram for a knot ðY ; KÞ is a quintuple

ðS; a; b0 ; m; mÞ; where here mAm-ðS a1 ? ag Þ:
The distinction between pointed Heegaard diagrams (in the sense of [20]) and
marked Heegaard diagrams is that in the former, the distinguished point lies in the
complement of all the g-tuples, while in the latter, the distinguished point lies on one
of the curves.
A marked Heegaard diagram for a knot K can also be used to construct a doubly
pointed Heegaard diagram for Y :
Definition 2.4. A doubly pointed Heegaard diagram for a three-manifold Y is a tuple

ðS; a; b; w; zÞ; where ðS; a; bÞ is a Heegaard diagram for Y ; and w and z are a pair of
distinct basepoints in S which do not lie on any of the a or b:
We extract a doubly pointed Heegaard diagram for Y from a Heegaard diagram

for ðY ; KÞ as follows. Let ðS; a; b0 ; m; mÞ be a marked Heegaard diagram for a knot
complement ðY ; KÞ: We write b ¼ b0 ,m: Fix an arc d which meets m transversely in
a single intersection point, which is the basepoint m; and which is disjoint from all
the a and b0 : Then, let z be the initial point of d; and w be its final point.
Note that the ordering of the two points w and z is specified by the orientation of
K: Specifically, choose a longitude l for K; which we think of this as a curve in the
Heegaard surface. Clearly, the orientation on K gives an orientation for l: Now
choose w so that if d is oriented as a path from z to w; then we have an equality of
algebraic intersection numbers: #ðd-mÞ ¼ #ðl-mÞ (for either orientation of m).
Let ðS; a; b; w; zÞ be a doubly pointed Heegaard diagram of genus g: As in [20], we
consider the g-fold symmetric product Symg ðSÞ; equipped with the pair of tori
T a ¼ a1 ? ag and Tb ¼ b1 ? bg :
ARTICLE IN PRESS
P. Ozsvath,
Following notation from [20], if x; yATa -Tb ; we consider spaces of homotopy

classes of Whitney disks p2 ðx; yÞ: Letting vAS a1 ? ag b1 ? bg ; and
choosing fAp2 ðx; yÞ; we let nv ðfÞ denote the oriented intersection number
nv ðfÞ ¼ #f1 ðfvg Symg1 ðSÞÞ:
A complex structure on S induces a complex structure on Symg ðSÞ with the property
that any Whitney disks f which admits a holomorphic representative has nv ðfÞX0:
For a suitable small perturbation of the holomorphic curve condition (which can be
encoded as a one-parameter family J of almost-complex structures over Symg ðSÞ),
we obtain a notion of pseudo-holomorphic disks, which form moduli spaces
satisfying suitably transverse Fredholm theory (along with the above non-negativity
hypothesis). We refer the reader to [20] for a detailed discussion.
2.3. Intersection points and Spinc structures
Recall that in [20], for a Heegaard diagram for Y ; ðS; a; b; wÞ; we give a map
sw : Ta -Tb -Spinc ðY Þ:
When Y is equipped with an oriented, null-homologous knot K; and corresponding

marked Heegaard diagram, this map has a refinement as follows. Note first that
there is a canonical zero-surgery Y0 ðKÞ: We define a map
sm : Ta -Tb -Spinc ðY0 ðKÞÞ;

%
where here in the definition of Tb we use b ¼ m,b0 : Sometimes, we denote the set
Spinc ðY0 ðKÞÞ by Spinc ðY ; KÞ; and call it a set of relative Spinc structures for ðY ; KÞ:
Note that
Spinc ðY ; KÞDSpinc ðY Þ Z:
The map from Spinc ðY ; KÞ-Spinc ðY Þ is obtained by first restricting t to Y K; and

%
then uniquely extending it to Y (note that the composition of these two maps is the
map sw described earlier). Projection to the second factor comes from evaluation
2/c1 ðtÞ; ½F̂ S; where here F̂ denotes a surface in Y0 ðKÞ obtained by capping off some
1
%
fixed Seifert surface F for K: If tASpinc ðY ; KÞ projects to sASpinc ðY Þ; we say that t
% %
extends s:
To define sm ; we replace the meridian with a longitude l for the knot, chosen to
wind once along % the meridian, never crossing the marked point m; so that each
intersection point x has a pair of closest points x0 and x00 : Let ðS; a; c; wÞ denote the
corresponding Heegaard diagram for Y0 ðKÞ (i.e. c ¼ l,b0 ). We then define sm ðxÞ to
be the Spinc structure over Y0 ðKÞ given by s0w ðx0 Þ ¼ s0w ðx00 Þ; where % here
s0w : Ta -Tg -Spin ðY0 ðKÞÞ denotes the usual map from intersection points to
c
Spinc structures (which we distinguish here from the analogous map sw for Y ). Of
ARTICLE IN PRESS
P. Ozsvath, 67
µ
w z
x ,, x x, α1
Fig. 3. Winding once along the meridian. We have pictured the winding procedure to illustrate the relative
Spinc structure corresponding to an intersection point. Note that the pictured region (all of which takes
place in a cylinder—i.e. the two dark curves are to be identified) is only a portion of the Heegaard diagram:
the tuples x and x0 in principle contain additional intersection points not pictured here. The orientation of
l induced from K is illustrated by the arrow.
course, the points w and z lie in the same component of S a1 ? ag g1

? gg ; so s0w ¼ s0z : See Fig. 3 for an illustration.
Recall (cf. [20]) that we can express the evaluation of /c1 ðsm ðxÞÞ; ½F̂ S in terms of
%
data on the Heegaard diagram. More precisely, each two-dimensional homology
class for Y0 has a corresponding ‘‘periodic domain’’ P in S; i.e. a chain in S whose
local multiplicity at w is zero, and which bounds curves among the a and c: Let
yATa -Tg be an intersection point, let wðPÞ denote the ‘‘Euler measure’’ of P (which
is simply the Euler characteristic of P; if all its local multiplicities are zero or one),
and let n% y ðPÞ denote the sum of the local multiplicities of P at the points yi
comprising the g-tuple y (taken with a suitable fraction if it lies on the boundary of
P; as defined in Section 7 of [18]; compare also Eq. (15) below). Then, it is shown in
Proposition 7.5 of [18] that
/c1 ðs0w ðyÞÞ; ½F̂ S ¼ wðPÞ þ 2n% y ðPÞ; ð9Þ
provided that P is the periodic domain representing F̂:
Lemma 2.5. Let ðS; a; b0 ; m; mÞ be a marked Heegaard diagram for an oriented knot.
Given x; yATa -Tb ; for any fAp2 ðx; yÞ; we have that
sm ðxÞ sm ðyÞ ¼ ðnz ðfÞ nw ðfÞÞ PD½m;

% %
where here ½mAH1 ðY0 ðKÞ; ZÞ is the homology class obtained by thinking of the
meridian of K as a loop in Y ndðKÞCY0 ðKÞ; and oriented so that #ðm-F Þ ¼ 1:
Proof. We begin with some preliminary observations. Consider the natural

Heegaard triple for the the cobordism from Y to Y0 ðS; a; b; c; wÞ; where c
is obtained as before by replacing the meridian m in the b by the longitude l: Note
also that there is a second basepoint zAS; which lies in the same component of
S a1 ? ag g1 ? gg as w:
ARTICLE IN PRESS
P. Ozsvath,
We claim that more generally if cAp2 ðx; Y; yÞ is a Whitney triangle (where,

as usual, Y is an intersection point to Tb -Tg representing the canonical
top-dimensional homology generator for HF p0 ð#g1 ðS 2 #S 1 ÞÞ), then we
have that
s0w ðyÞ ¼ sm ðxÞ þ ðnw ðcÞ nz ðcÞÞ ½PDðmÞ: ð10Þ

%
We see this as follows. First, fix xATa -Tb and an integer kAZ; and consider the
set Sðx; kÞ of all yATa -Tg for which there is a Whitney triangle cAp2 ðx; Y; yÞ with
nw ðcÞ nz ðcÞ ¼ k: Clearly, this set is a single Spinc -equivalence class of intersection
points on Y0 ðKÞ: Now it is an easy consequence of the definition of s that Eq. (10)
holds in the case where k ¼ 0: %
In view of the above remarks, to verify Eq. (10), it suffices to verify it in the
case where the triangle is supported in the winding region. More precisely, fix
an an integer k; and consider the ‘‘small’’ triangle ck supported in the winding
region (after winding l sufficiently many times in a neighborhood of m). For
example, when kp0; after winding sufficiently many times, we can find a triangle ck
which lies on one side of m; so that nw ðcÞ ¼ 0 and nz ðcÞ ¼ k: Now ck Ap2 ðx; Y; x0k Þ;
where the x0i are the various points for Ta -Tg closest to x representing various Spinc
structures ½Sðx; kÞ: It is an easy consequence now of Eq. (9) and our orientation
conventions that
/c1 ðs0w ðxi ÞÞ; ½F̂ S /c1 ðs0w ðxj ÞÞ; ½F̂ S ¼ 2ði jÞ:
This completes the verification of Eq. (10).

The equation stated in the lemma now follows from Eq. (10) at once:
let cAp2 ðy; Y; y0 Þ be a triangle with nz ¼ nw ¼ 0; and then apply the formula
for the juxtaposed triangle c0 ¼ f cAp2 ðx; Y; y0 Þ with nw ðc0 Þ nz ðc0 Þ ¼
nw ðfÞ nz ðfÞ: &
2.4. Filtered complexes
Fix a partially ordered set S: An S-filtered group is a free Abelian group C

generated freely by a distinguished set of generators S which admit a map
F : S-S:
We write elements of C as sums

X
as s;
sAS
where as AZ: If
X X
a¼ as s and b¼ bs s
sAS sAS0
ARTICLE IN PRESS
P. Ozsvath, 69
are elements of S-filtered groups ðC; F; SÞ and ðC 0 ; F0 ; S0 Þ; respectively, then we

write apb if
max FðsÞp min F0 ðsÞ:

fsASjas a0g fsAS0 jbs a0g
A morphism of S-filtered groups
f : ðC; F; SÞ-ðC 0 ; F0 ; S0 Þ
is a group homomorphism with the property that
fðaÞpa
for all aAC:

An S-filtered chain complex is an S-filtered group equipped with a differential
which is an S-filtered morphism; a morphism of S-filtered chain complexes is a chain
map which is also an S-filtered morphism. Let TCS be a subset of S with the
property that if bAT; then all elements aAS with apb also are contained in T: If
TCS is such a subset, and if ðC ; @; FÞ is an S-filtered complex, then T gives rise to a
subcomplex of C ; which transforms naturally under morphisms.
For example, recall that CF N ðY ; tÞ defined in [20] is generated by pairs
½x; iAðTa -Tb Þ Z
(where the tori Ta and Tb lie in the g-fold symmetric product of a genus g Heegaard
surface for Y ). Indeed, the map
F½x; i ¼ i
induces a natural filtration on this complex, and hence endows it with a canonical
subcomplex corresponding to the negative integers, CF ðY ; tÞ; and a quotient
complex CF þ ðY ; tÞ:
In the present paper, we will typically encounter complexes filtered by Z Z; with
the partial ordering ði; jÞpði0 ; j 0 Þ if ipi0 and jpj 0 : These complexes have a number of
naturally associated subcomplexes. For instance, there is a subcomplex correspond-
ing to the quadrant fði; jÞjip0 and jp0g; and there is also a subcomplex
corresponding to the union of three quadrants fði; jÞjip0 or jp0g:
3. The knot filtration: definitions and basic properties
In this section, we begin by giving the definition of the knot filtration. In Section
3.2 we prove its topological invariance. In Section 3.3, we describe an absolute
grading used on the knot filtration. In Section 3.5 we describe several symmetries of
the filtration: first under orientation reversal of the ambient space, then under
ARTICLE IN PRESS
P. Ozsvath,
orientation reversal of the knot, and finally under conjugation invariance of the
underlying relative Spinc structures.
3.1. Definition of the knot filtration
Let ðS; a; b; w; zÞ be a doubly pointed Heegaard diagram, and J be an allowed one-

parameter family of almost-complex structures over Symg ðSÞ: We can associate to
this data a Z2 -filtered chain complex, following the constructions of [20]. Specifically,
we let CF N ðS; a; b; w; zÞ be the free Abelian group generated by triples ½x; i; j with
xATa -Tb ; i; jAZ: We endow this with the differential:
X X
@ N ½x; i; j ¼ #
#ðMðfÞÞ½y; i nw ðfÞ; j nz ðfÞ;
yATa -Tb ffAp2 ðx;yÞjmðfÞ¼1g
#
where here MðfÞ denotes the quotient of the moduli space of J-holomorphic disks
representing the homotopy type of f; MðfÞ; divided out by the natural action of R
on this moduli space, and mðfÞ denotes the formal dimension of MðfÞ: The signed
count here uses a coherent choice of orientations, as described in [20]. Moreover, we
can endow the chain complex with the structure of a Z½U-module, by defining
U ½x; i; j ¼ ½x; i 1; j 1:
Of course, the filtration on the chain complex
F : ðTa -Tb Þ Z2 -Z2
is given by
F½x; i; j ¼ ði; jÞ;
where Z Z is given the partial ordering ði; jÞpði0 ; j 0 Þ when ipi0 and jpj 0 :
The complex CFK N naturally splits into a sum of complexes. Specifically,
generators ½x; i; j and ½y; c; m lie in the same summand precisely when there is a
homotopy class fAp2 ðx; yÞ with
nw ðfÞ ¼ i c and nz ðfÞ ¼ j m:
In the case studied here—where the doubly pointed Heegaard diagram comes
from from the marked Heegaard diagram of an oriented, null-homologous knot
K—this splitting can be interpreted in terms of Spinc structures over Y0 ðKÞ:
Specifically, fix a Spinc structure s over Y and let tASpinc ðY ; KÞ ¼ Spinc ðY0 ðKÞÞ
%
be a Spinc structure which extends it (cf. Section 2.3). consider the subset
CFK ðY ; K; tÞCCF ðS; a; b0 ,m; w; zÞ generated by triples ½x; i; j with sw ðxÞ ¼ s
N N
and %
sm ðxÞ þ ði jÞPD½m ¼ t; ð11Þ

% %
ARTICLE IN PRESS
P. Ozsvath, 71
where ½mAH1 ðY0 ðKÞÞ is the homology class obtained by thinking of the meridian m
as a closed curve in Y0 ðKÞ: According to Lemma 2.5, this is a subcomplex. Note that
if t1 and t2 represent the same Spinc structure over Y ; then the complexes
CFK% N ðY ; t% Þ and CFK N ðY ; t Þ are isomorphic as chain complexes, and indeed, the
1 2
%
only difference % Z"Z-filtration.
is a shift in the
Fix t0 ASpinc ðY ; KÞ and let sASpinc ðY Þ be the compatible element. We can view
CFK N ðY ; K; t0 Þ as an extra Z filtration on CF N ðY ; sÞ; by using the isomorphism
%
P1 : CFK N ðY ; K; t0 Þ-CF N ðY ; sÞ
%
given by
P1 ½x; i; j ¼ ½x; i;
and declaring the extra Z filtration to be induced from the projection of ½x; i; j to j:
c ðY ; sÞ: Specifically, let
There is also a corresponding filtration structure on CF
CFK ; ðY ; K; t0 ÞCCFK N ðY ; K; t0 Þ

% %
denote the subcomplex corresponding to ði; jÞ with io0; which in turn has a quotient
complex CFK þ; ðY ; K; t0 Þ: Of course, these complexes can be thought of as induced
% and CF þ ðY ; sÞ; respectively. Consider the subcomplex
filtrations on CF ðY ; sÞ
CFK ðY ; K; t0 ÞCCFK þ; ðY ; K; t0 Þ generated by elements in the kernel of the
0;
%
induced U-action, %
i.e. these generators all have i ¼ 0: This can be thought of as a
c
filtration on CF ðY ; sÞ: The associated graded complex for CFK 0; ðY ; K; t0 Þ admits
an extra Z-grading which, of course, depended on our initial choice of t0 : %However,
we can think of this object more invariantly as graded by Spinc ðY ; KÞ: %Specifically,
for each tASpinc ðY0 ðKÞÞ; we have a complex CFK d ; K; tÞ which is generated by
CFKðY
%
intersection points x with sm ðxÞ ¼ t; and its boundary% operator counts only
homotopy classes fAp2 ðx; yÞ% with nw ðfÞ% ¼ n ðfÞ ¼ 0: Then the associated graded
z
0;
graded complex for CFK ðY ; K; t0 Þ is identified with
%
M
d ; K; tÞ;
CFK
CFKðY
c
ftASpin ðY ;KÞjt extends sg %
% %
where the summand of the associated graded object for CFK 0; ðY ; K; t0 Þ belonging
to the integer j corresponds to the summand of the above complex %belonging to
t ¼ t0 þ j PD½m:
% When
% considering the complex CFK N ðY ; K; tÞ we always work with a Heegaard
%
diagram for the knot which is strongly s-admissible as a Heegaard diagram for Y
c
equipped with the Spin structure s which is extended by t; while when we consider
% d it
its quotient complexes which are bounded below (such as CFK þ; and CFK CFK),
suffices to consider only weakly s-admissible Heegaard diagrams.
ARTICLE IN PRESS
P. Ozsvath,
Theorem 3.1. Let ðY ; KÞ be an oriented knot, and fix a Spinc structure tASpinc ðY ; KÞ:
%
Then the filtered chain homotopy type of the chain complex CFK N ðY ; K; tÞ is a
% i.e.
topological invariant of the oriented knot K and the Spin structure tASpinc ðY ; KÞ;
c
%
it is independent of the choice of admissible, marked Heegaard diagram ðS; a; b0 ; m; mÞ
used in its definition.
Indeed, we shall see that when the first Chern class of the corresponding Spinc
structure s over Y is torsion, then the above chain complex inherits in a natural way
an absolute grading from the absolute grading on CF N ðY ; sÞ (cf. Lemma 3.6 below).
We postpone the proof of the above theorem till the next subsection, first stating
some of its immediate consequences.
d
Corollary 3.2. The ‘‘knot homology groups’’ HFK
HFKðY d ; K; tÞÞ are
; K; tÞ ¼ H ðCFK
CFKðY
%
topological invariants of the knot KCY and tASpinc ðY ; KÞ: %
%
Theorem 3.1)Corollary 3.2. It is clear from the above construction that the chain
d ; K; tÞ is uniquely determined by the filtered chain
homotopy type of CFK
CFKðY
%
homotopy type of CFK N ðY ; K; tÞ:
%
In view of Proposition 2.1, passage from knot invariants to link invariants is
straightforward.
Definition 3.3. Let LCY be an n-component oriented link in Y ; and let kðLÞ denote
the induced knot inside kðY ÞDY #n1 ðS2 S 1 Þ as in Proposition 2.1. Then, given
tASpinc ðkðY Þ; kðLÞÞ we define
%
CFK N ðY ; L; tÞ ¼ CFK N ðkðY Þ; kðLÞ; tÞ:
% %
Corollary 3.4. Let ðY ; LÞ be an oriented link, and fix tASpinc ðkðY Þ; kðLÞÞ: Then,
CFK N ðY ; L; tÞ is an invariant of the link LCY : %
%
Theorem 3.1)Corollary 3.4. This implication is immediate, in view of Proposition 2.1.
As we see from above, the case of links is no more general than the case of knots.
Thus, we focus on the case of knots for most of the present paper, returning to the
disconnected case in Section 10.
3.2. Proof of topological invariance of the knot homologies
The proof of Theorem 3.1 is very closely modeled on the proof of the topological
invariance of Floer homology proved in [20]. Thus, we highlight here the difference
between the two results, leaving the reader to consult [20] for more details.
Verification of the topological invariance rests on the following description of how
the invariant depends on the underlying Heegaard diagram:
ARTICLE IN PRESS
P. Ozsvath, 73
Proposition 3.5. If ðS; a; b0 ; mÞ and ðS0 ; a0 ; b00 ; m0 Þ represent the same knot complement,
then we can pass from one to the other by a sequence of the following types of moves
(and their inverses):
* handleslides and isotopies amongst the a or the b0 ;
* isotopies of m;
* handleslides of m across some of the b0 ;
* stabilizations (introducing canceling pairs agþ1 and bgþ1 and increasing the genus
of S by one).
Proof. This follows from standard Morse theory (cf. Lemma 4.5 of [19]). &
For our proof, we adapt some of the Floer-homology constructions for (singly)
pointed Heegaard diagrams described in [20] to the doubly pointed case.
Let ðS; a; b; w; zÞ; ðS; b; c; w; zÞ be a pair of doubly pointed Heegaard diagrams.
Then, there is an induced (filtered) map:
F : CF N ðS; a; b; w; zÞ#CF N ðS; b; c; w; zÞ-CF N ðS; a; c; w; zÞ
defined as follows:
X
F ð½x; i; j#½y; c; mÞ ¼ ð#MðcÞÞ ½w; i þ c nw ðcÞ; j þ m nz ðcÞ;
fcAp2 ðx;y;wÞjmðcÞ¼0g
where #MðcÞ denotes a signed count of pseudo-holomorphic triangles.
Proof of Theorem 3.1. First, we observe that the filtered chain homotopy type
CFK N is independent of the perturbation of complex structure J: This is a
straightforward modification of the corresponding discussion in [20]. Similarly,
independence of the complex under isotopies of the b0 is a straightforward
modification of the corresponding invariance of HF N : for an isotopy consisting of a
single pair creations, one considers the natural chain homotopy equivalence induced
by an exact Hamiltonian isotopy realizing the isotopy.
For independence of the groups under handleslides amongst the b0 ; follow the
arguments of handleslide invariance of HF þ as well. Now, we observe that if b00
is obtained from b0 by a handleslide, then the doubly pointed Heegaard diagram
ðS; m,b0 ; m0 ,b00 ; w; zÞ (where m0 is a small exact Hamiltonian translate of m) has the
property that w and z lie in the same connected component of S m b2 ?
bg m0 b02 ? b0g : We arrange for this Heegaard diagram to be strongly s0 -
admissible as a pointed Heegaard diagram for #g ðS 2 S 1 Þ (using either basepoint),
where here s0 is the Spinc structure with trivial first Chern class. We let CFdN denote
the ‘‘diagonal’’ summand of CF N ðS; m,b0 ; m0 ,b00 Þ generated by ½x; i; i (where ½x; i
represents s0 ). It is easy to see then that (cf. Lemma 9.1 of [20]),
HFdp0 ðm,b0 ; m0 ,b00 ; w; zÞDZ½U#Z L H1 ðT g Þ:

ARTICLE IN PRESS
P. Ozsvath,
Letting Y represent the top-dimensional non-zero generator, we use
x/F ðx#YÞ
to define the map associated to handleslides. The proof that this is a filtered chain
homotopy equivalence now follows from a direct application of the methods from
Section 9 of [20]. The same remarks apply when one slides m over one of the circles in
b0 : (Note that if one were to try to slide b2 over m; the arguments would breakdown
in view of the fact that now w and z lie in different components of the complements
of the curves. This move, however, is not required for proving the topological
invariance of the knot homology, according to Proposition 3.5 above.)
The above argument shows that the chain homotopy type of the doubly filtered
chain complex is invariant under handleslides. It is not difficult to see that the map
also preserves the splitting according to tASpinc ðY ; KÞ: (This is shown in a
%
somewhat more general context in Proposition 8.1 below.)
Constructing such an isomorphism for isotopies and handleslides along the a
proceeds in an analogous manner: one shows that the filtered chain homotopy type
remains invariant under isotopies and handleslides which do not cross the arc d
connecting w and z: Observe also that an isotopy of the a which does cross this arc
can be realized by a sequence of isotopies and handleslides amongst the a which do
not cross the arc (cf. Proposition 7.1 of [20]).
Stabilization invariance, too, follows directly as in [20]. &
3.3. Absolute gradings
Fix a Spinc structure sASpinc ðY Þ; whose first Chern class is torsion. Recall (cf.
[19]) that in this case, we defined an absolute Q-grading on CF N ðY ; sÞ which enjoys
the following formula: if W is a cobordism from Y1 to Y2 ; rASpinc ðW Þ is a Spinc
structure whose restrictions s1 and s2 to Y1 and Y2 ; respectively, are both torsion,
then the induced map FW ;r shifts degree by
c1 ðrÞ2 2wðW Þ 3sðW Þ

; ð12Þ
4
where wðW Þ and sðW Þ denote the Euler characteristic and signature of W ; respectively.
The torsion hypothesis also gives us a canonical t0 ASpinc ðY ; KÞ extending s which
%
satisfies
/c1 ðt0 Þ; ½F̂ S ¼ 0 ð13Þ
%
for any choice of Seifert surface F for K (indeed, the condition that t0 is independent
% is torsion). We
of the choice of Seifert surface is equivalent to the condition that c1 ðsÞ
can endow CFK ðY ; K; t0 Þ with the absolute grading induced from the map
N
%
P1 : CFK N ðY ; K; t0 Þ-CF N ðY ; sÞ
%
given by P1 ½x; i; j ¼ ½x; i:
ARTICLE IN PRESS
P. Ozsvath, 75
Since t0 ASpinc ðY ; KÞ is uniquely determined by sASpinc ðY Þ; we typically write

%
CFK N ðY ; K; sÞ for CFK N ðY ; K; t0 Þ:
%
Lemma 3.6. Let Y be an oriented three-manifold KCY be a knot, and fix a Spinc
structure sASpinc ðY Þ: Then, there is a convergent spectral sequence of relatively
graded groups whose E 1 term is
M
d
HFK
HFKðY ; K; tÞ;
c
ftASpin ðY ;KÞjt extends sg %
% %
and whose E N -term is HF c ðY ; sÞ: Moreover, when c1 ðsÞ is torsion, the spectral
sequence respects absolute gradings.
Proof. This is simply the Leray spectral sequence associated to the filtration of
c ðY ; sÞ given by CFK 0; ðY ; t0 Þ (where tASpinc ðY ; KÞ is any choice of Spinc
CF
% %
structure which extends sASpinc ðY Þ). The statement in the absolutely graded case
follows from our definition of the absolute grading on HFK d
HFK: &
Of course, we have analogous spectral sequences converging to HF þ ðY ; sÞ;
HF ðY ; sÞ; and HF N ðY ; sÞ whose E1 terms are derived from CFK N ðY ; KÞ in a
straightforward way.
3.4. Notational shorthand
We make several further simplifications when considering knots K in S 3 : We write

simply CFK N ðS3 ; KÞ for the absolutely Z-graded complex CFK N ðS 3 ; K; t0 Þ; where
%
here t0 ASpinc ðS3 ; KÞ is uniquely characterized by the property that c1 ðt0 Þ ¼ 0:
%
Moreover, given an integer nAZ; write HFK d 3 ; K; nÞ for HFK
HFKðS d 3 ; K; tn Þ; where
HFKðS
%
tn ASpinc ðS 3 ; KÞ is characterized by the property that /c1 ðtn Þ; ½F̂ S ¼ 2n; where F is
% %
a compatible Seifert surface for the oriented knot K:
In this setting, we also write HF þ ðS03 ðKÞ; nÞ to denote HF þ ðS03 ðKÞ; sn Þ (where we
%
think of sn as a Spinc structure over Y0 ).
%
3.5. Symmetries
Proposition 3.7. The dependence on the orientation of Y is given by
d ðY ; K; sÞDHFK
HFK d ðY ; K; sÞ
% %
(where here the left-hand side denotes homology, and the right-hand side denotes
cohomology). Moreover, if s extends a torsion Spinc structure over Y ; then
%
d d ðY ; K; sÞDHFK
HFK d d ðY ; K; sÞ:
% %
ARTICLE IN PRESS
P. Ozsvath,
Proof. This follows from the behavior of HF c ðY Þ under orientation reversal.

Specifically, if ðS; a; b; w; zÞ describes the knot K in Y ; then ðS; a; b; w; zÞ describes
the knot K in Y : The identification of the complexes now follows as in [18]. &
Proposition 3.8. Fix a torsion Spinc structure s over Y : For each extension
sASpinc ðY ; KÞ of s; we have that
HFK d d2m ðY ; K; sÞ;
% %
where
m ¼ 12/c1 ðsÞ; ½F̂ S
%
is calculated using the surface F̂ obtained by capping off a Seifert surface F for the
oriented knot K:
Proof. Observe that for the chain complex underlying CFK N ðY ; K; t0 Þ; the roles of i
and j can be interchanged; i.e. the same complex can be viewed at the % same time as
CFK ðY ; K; t0 Þ: (Recall that t0 is chosen to satisfy Eq. (13).) Correspondingly, we
N
% map
have a projection %
P2 : CFK N ðY ; K; t0 ÞDCFK N ðY ; K; t0 Þ-CF N ðY ; sÞ

% %
given by P2 ½x; i; j ¼ ½x; j: (By all rights, the complex CF N ðY ; sÞ ought to include
the base point in its notation: in this case we use z rather than w). This projection,
too induces an absolute grading on CFK N ðY ; K; t0 Þ; but in fact it coincides with the
absolute grading induced from P1 ; since there is % a grading-preserving homotopy
equivalence from CF ðY ; sÞ; as defined using the base point w; to the corresponding
N
complex as defined using z:

Now, the stated isomorphism is induced by the map
d ; K; sÞD CFK 0;m ðY ; K; t0 Þ-CFK m;0 ðY ; K; t0 Þ

U m : CFK
CFKðY
% % %
d ; K; sÞ:
D CFK
CFKðY &
%
There is a conjugation symmetry
J : Spinc ðY ; KÞ-Spinc ðY ; KÞ
induced from the usual conjugation symmetry on Spinc ðY0 ðKÞÞ:
Proposition 3.9. For each sASpinc ðY ; KÞ; there is an identification

%
CFK N ðY ; K; sÞDCFK N ðY ; K; J sÞ:
% %
ARTICLE IN PRESS
P. Ozsvath, 77
Proof. First notice that in our definition the Heegaard diagram for a knot, there is
an asymmetry in the role of the a and the b; specifically, our choice of meridian was
chosen to be among the b: This was, of course, unnecessary. We could just as well
have chosen our meridian to be an a-circle, repeated the above definition, and
obtained another knot filtration which is also a knot invariant. In fact, these two
filtrations coincide, provided that we switch the roles of w and z: Indeed, we can
always choose a Heegaard diagram for the knot K for which there are
representatives for K in both handlebodies, with both a1 and b1 representing
meridians for the two representatives, reversing the roles of w and z: Such a diagram
can be obtained by stabilizing a doubly pointed Heegaard diagram for which the
attaching circle b1 is a meridian for K: Specifically, we stabilize our original
Heegaard surface by attaching a handle with feet near w and z; respectively, the circle
a1 is supported inside the newly attached handle, and we introduce a canceling circle
b2 which runs along a longitude for K except near b1 ; which it avoids by running
through the one-handle. This is illustrated in Fig. 4.
With these remarks in place, we realize the conjugation symmetry by the natural
identification of the complexes ðS; a; b; w; zÞ and ðS; b; a; z; wÞ; following [20]. It is
easy to see that this identification induces an isomorphism of CFK N ðY ; K; sÞ with
CFK N ðY ; K; J sÞ: & %
%
Fix a torsion Spinc structure s over Y ; and let CFK N ðY ; K; sÞ0 denote the Z"Z-
filtered chain complex CFK N ðY ; K; sÞ; endowed with the filtration
F½x; i; j ¼ ð j; iÞ:
α1
w z
β2
β1
Fig. 4. Moving the meridian. This illustrates the symmetry between the roles of a1 and b1 as meridians for
a given knot K: This knot K is isotopic to either of the two (unlabeled) dotted circles in S; and hence can
be pushed into either handlebody so that either a1 or b1 is a meridian. Note that there may be additional a-
and b-circles entering the picture, but it is important that they do not cross the dotted circles, and do not
separate w and z from these dotted circles.
ARTICLE IN PRESS
P. Ozsvath,
The above proposition gives a degree-preserving Z"Z-filtered chain homotopy

equivalence
F : CFK N ðY ; K; sÞ-CFK N ðY ; K; sÞ0 :
This gives the following corollary, which we state for the knot Floer homology:
Proposition 3.10. Let Y be a three-manifold equipped with a torsion Spinc structure,

then
HFK d d2m ðY ; K; J sÞ;
% %
where 2m ¼ /c1 ðsÞ; ½F̂ S:
%
Proof. Combine the results of Propositions 3.8 and 3.9. &
4. Relationship with three-manifold invariants
We consider the following two Z"Z-filtered, Z½U-subcomplexes of

CFK N ðY ; KÞ:
* CFK ðY ; KÞ; which denotes the subcomplex of CFKðY ; KÞ generated by ½x; i; j
with maxði; jÞo0;
* b
CFK ðY ; KÞ (‘‘big’’ CFK ), which denotes the subcomplex generated by ½x; i; j
with minði; jÞo0:
These have quotient complexes denoted
b
CFK þ ðY ; KÞ ¼ CFK N ðY ; KÞ=CFK ðY ; KÞ;
CFK þ ðY ; KÞ ¼ CFK N ðY ; KÞ=b CFK ðY ; KÞ:
Our aim is to identify the homologies of CFK þ ðY ; KÞ (resp. b CFK þ ðY ; KÞ) with
HF þ ðYp ðKÞÞ (resp. HF þ ðYp ðKÞÞ) for sufficiently large integer surgery coefficients
p; to be made precise shortly.
Fix a Seifert surface F for KCY ; compatible with the orientation of K: We can
cap this off to obtain a surface F̂CY0 ðKÞ: For each mAZ; let
M
CFK N ðY ; K; F ; mÞ ¼ CFK N ðY ; K; tÞ:
ftASpinc ðY ;KÞj/c1 ðtÞ;½F̂ S¼2mg
%
% %
For fixed integer p; let Wp ðKÞ denote the four-manifold obtained by attaching a
two-handle to ½0; 1 Y along K with framing p: Let S denote the surface in
Wp ðKÞ obtained by closing off F ; to obtain a surface of square p: Now, given
½mAZ=pZ; let TðF ; p; ½mÞCSpinc ðYp ðKÞÞ denote the set of Spinc structures which
ARTICLE IN PRESS
P. Ozsvath, 79
can be extended over Spinc ðWp ðKÞÞ to give a Spinc structure r with
/c1 ðrÞ; ½SS þ p 2m ðmod 2pÞ:
Correspondingly, we let
M
CF N ðYp ðKÞ; F ; ½mÞ ¼ CF N ðYp ðKÞ; F ; tÞ:
tATðF ;p;mÞ
When F is understood from the context, then we drop it from the notation.
To relate these two complexes, we will use the following map. Fix a positive
integer p; and let ðS; a; b; c; mÞ be a marked Heegaard triple for the cobordism from
Y to Yp : Consider the chain map
F : CFK N ðY ; KÞ-CF N ðYp Þ
defined by
X X
F½x; i; j ¼ ð#MðcÞÞ ½y; i nw ðcÞ:
yATa -Tg fcAp2 ðx;Y;yÞjnw ðcÞnz ðcÞ¼ij;mðcÞ¼0g
Theorem 4.1. Let ðY ; KÞ be an oriented knot, and fix a compatible Seifert surface F
for K: Then, for each integer mAZ; we have an integer N ¼ NðmÞ so that for all
integers pXN; there is a Heegaard diagram for which the map F above induces
isomorphisms of chain complexes which fit into the following commutative diagram:
Similarly, if we let CFK 0 ðY ; K; mÞCCFK þ ðY ; K; mÞ be the subset generated by ½x; i; j

with minði; jÞ ¼ 0; then F also induces isomorphisms in the following diagram:
Proof. First, note that F maps b CFK ðY ; KÞ into CF ðYp Þ: if c has a holomorphic
representative, then nw ðcÞX0 and nz ðcÞX0; so if ½x; i; jAb CFK ðY ; KÞ; then either
io0; so clearly, i nw ðcÞo0; or iX0 and jo0; so
i nw ðcÞ ¼ j nz ðcÞo0:
Fix a marked Heegaard diagram ðS; a; b; mÞ for ðY ; KÞ; where mAb1 ; so that b1 is
the meridian. Fix an annular neighborhood of b1 and a longitude l for K: We will
ARTICLE IN PRESS
P. Ozsvath,
consider Heegaard triples ðS; a; b; c; wÞ so that the gi for iX2 are all exact
Hamiltonian translates of the corresponding bi ; and g1 is a small (embedded)
perturbation of the juxtaposition pb1 þ l: We will situate g1 so that b1 intersects it in
one point towards the middle of this winding region (cf. Fig. 5).
We say that an intersection x0 ATa -Tg is supported in the winding region if the
coordinate on x0 in g1 is supported there. In this case, there is a uniquely determined
intersection point xATa -Tb which is closest to x0 : Moreover, there is a canonical
‘‘small triangle’’ c0 Ap2 ðx; Y; x0 Þ supported in the winding region. Conversely, given
any xATa -Tb ; there are p distinct intersection points for Ta -Tg ; representing the p
Spinc structures over Yp ðKÞ which are cobordant to the Spinc structure represented
by x:
We claim that
/c1 ðsðxÞÞ; ½F̂ S þ 2ðnw ðcÞ nz ðcÞÞ ¼ /c1 ðsw ðcÞÞ; ½SS p: ð14Þ
%
Fix a Spinc structure sASpinc ðY Þ; and let x1 be some representative for s: For
c ¼ c0 ðx1 ; Y; x01 ÞAp2 ðx; Y; x0 Þ with nw ðcÞ ¼ nz ðcÞ ¼ 0; this is a straightforward
application of the first Chern class formula from [19]. Now, any c which induces the
same Spinc structure has the form c ¼ f c0 f0 with fAp2 ðx2 ; x1 Þ (for Ta -Tb ;
f0 Ap2 ðx01 ; x02 Þ). It follows from Lemma 2.5 Eq. (14) is unaffected by the juxtaposition
of f; while it is clear that it is unaffected by the juxtaposition of f0 : Finally, it is
straightforward to see that the equation is preserved by adding a triply periodic
domain belonging to S: This completes the verification of Eq. (14) for all Whitney
triangles c:
It follows from Eq. (14), that the restriction of F to CFK N ðY ; K; mÞ maps to
CF þ ðYp ðKÞ; ½mÞ:
Fix an integer mAZ: We claim that if p is sufficiently large, then that all the
intersection points between Ta and Tg which represent any Spinc structure
sATðF ; p; ½mÞ are supported in the winding region. This can be seen, for example,
from Eq. (14). In this case, consider the map
F0 : CFK N ðY ; K; mÞ-CF N ðYp ðKÞ; ½mÞ
which carries ½x; i; j to ½x0 ; i nw ðc0 Þ: By the Riemann mapping theorem, we have
that #Mðc0 Þ ¼ 1; so we can think of F0 as the summand of F corresponding to the
w z
β
γ
1
x2 x’
1
α
Fig. 5. llustration of Theorem 4.1. The integers denote (non-zero) local multiplicities of the ‘‘small
triangle’’ F0 connecting x and x0 : This picture is taking place in a cylindrical region in S: the top dark line
is identified with the bottom.
ARTICLE IN PRESS
P. Ozsvath, 81
homotopy class c0 : Moreover, it is easy to see that F0 induces isomorphisms of

groups
CFK N ðY ; K; mÞD CF N ðYp ðKÞ; ½mÞ;

b
CFK ðY ; K; mÞD CF ðYp ðKÞ; ½mÞ;
CFK þ ðY ; K; mÞD CF þ ðYp ðKÞ; ½mÞ;
c ðYp ðKÞ; ½mÞ:

CFK 0 ðY ; K; mÞD CF
Now, if the winding region has sufficiently small area relative to the areas of the
regions in S a b (divided by n), then it is clear that
F ¼ F0 þ lower order;
with respect to the energy filtrations on CFKðY ; K; sÞ and CF ðYn ðKÞ; ½sÞ
(induced by areas of domains). Thus, by elementary algebra, the map F; which
we know is a chain map, is also an isomorphism (carrying b CFK ðY ; K; sÞ
to CF ðY ; ½sÞ). &
For convenience, we state a version of the above result in the case where we
restrict to Spinc structures over Y whose first Chern class is torsion. In this case,
recall that we defined CFK N ðY ; K; sÞ to be CFK N ðY ; K; t0 Þ; where t0 ASpinc ðY ; KÞ
% %
extends and also satisfies /c1 ðt0 Þ; ½F̂ S ¼ 0; a notion which now is independent of
the choice of Seifert surface for% K: We let
CFK N ðY ; K; sÞfiX0 or jX mgCCFK N ðY ; K; sÞ
denote the induced (quotient) complex generated by ½x; i; jACFK N ðY ; K; sÞ; subject
to the stated constraint iX0 or jX m: Moreover, let ½s; mASpinc ðYp Þ denote
the Spinc structure over Yp which is cobordant to s; via a Spinc structure
rASpinc ðWp Þ with
/c1 ðrÞ; ½SS ¼ 2m þ p:
Corollary 4.2. Let KCY be a knot, and fix sASpinc ðY Þ whose first Chern class
is torsion. Then, there is an integer N with the property that for all pXN;
we have that

Hk ðCFK N ðY ; K; sÞfiX0 and jX mgÞ if jmjpg;
HFcþ ðYp ðKÞ; ½s; mÞD
HFkþ ðY Þ otherwise;
ARTICLE IN PRESS
P. Ozsvath,
where
!
p ð2m þ pÞ2
c¼kþ :
4p
Proof. If jmj4g; this follows readily from the adjunction inequality for HF þ ðY0 Þ;
together with long exact sequence for integral surgeries (cf. [18]) (provided that
p4g).
The cases where jmjog; this follows from the above theorem. Specifically, observe
that if tASpinc ðY ; KÞ extends s; then we have an isomorphism of chain complexes
%
CFK N ðY ; K; tÞDCFK N ðY ; K; t0 Þ
% %
defined by
½x; i; j/½x; i; j þ 12/c1 ðtÞ; ½F̂ S:

%
This preserves the absolute Z grading as defined in Section 3.3, shifting the filtration
in the obvious way. Thus, the isomorphism in its relatively graded form is a direct
consequence of Theorem 4.1. For the graded statement, we recall that the map F
inducing the isomorphism is a filtered version of the map from CF N ðY Þ to
CF N ðYp Þ induced by counting holomorphic triangles (cf. [19]), associated to the
Spinc structure over Wp ðKÞ with
/c1 ðrÞ; ½SS ¼ 2m þ p;
according to Eq. (14). Thus, the shift in dimension follows from Eq. (12). &
Remark 4.3. It follows from the adjunction inequality for HF þ ðY0 Þ; together with
properties of the integer surgeries long exact sequence, that we can take N ¼ 2g 1;
where g denotes the genus of the knot.
In the same vein, for each mAZ; we define a map
C : CF N ðYp ; ½mÞ-CFK N ðY ; K; mÞ
by
X X
C½x; i ¼ #MðcÞ½y; i nw ðcÞ; i nz ðcÞ:
yATa -Tg fcAp2 ðx;Y;yÞjmðcÞ¼0;nw ðcÞnz ðcÞ¼mg
Note that the analogue of Eq. (14) in this case guarantees that the right-hand side is
contained in CFK N ðY ; K; mÞCCFK N ðY ; KÞ: We now have the following analogue
of Theorem 4.1:
Theorem 4.4. Let ðY ; KÞ be an oriented knot, and fix a compatible Seifert surface F :
For each integer m; we have an integer N so that for all integers pXNðmÞ; there is a
ARTICLE IN PRESS
P. Ozsvath, 83
Heegaard diagram for which the map C above induces isomorphisms of chain
complexes which fit into the following commutative diagram:
Similarly, if we let b CFK 0 ðY ; KÞCCFK þ ðY ; KÞ be the subset generated by ½x; i; j with

maxði; jÞ ¼ 0; then C also induces isomorphisms in:
Proof. This is a straightforward modification of the proof of Theorem 4.1. &
It is worth pointing out the following result:
Corollary 4.5. Suppose that K is a knot in S3 ; and let d be the largest integer for which
d
HFKðY
HFK ; K; dÞa0; and suppose that d41: Then,
d 3 ; K; dÞ
HF þ ðS03 ðKÞ; sÞDHFK
HFKðS
as Z=2Z graded Z-modules, where here sASpinc ðS03 ðKÞÞ is the Spinc structure with
/c1 ðsÞ; ½F̂ S ¼ 2d 2:
Proof. On the one hand, we have the short exact sequence

C
0-Cfio0 and jXd 1g-CfiX0 or jXd 1g ! CfiX0g-0:
Now, it is easy to see by taking filtrations that
d 3 ; K; dÞ:
H ðCfio0 and jXd 1gÞDHFK
HFKðS
Of course, according to Theorem 4.4, for sufficiently large integers n;
H ðCfiX0 or jXdgÞDHF þ ðSn3 ðKÞ; ½d 1Þ;
and indeed the map C is modeled on the map induced by the cobordism from Sn3 ðKÞ
to S 3 ; endowed with the Spinc structure r with
/c1 ðrÞ; ½F̂ S ¼ 2d 2 þ n:

ARTICLE IN PRESS
P. Ozsvath,
Note that the map c induced by C on homology is surjective, since it is a Z½U

module map and
H ðCfiX0gÞDHF þ ðS 3 Þ;
and c induces an isomorphism in all sufficiently large degrees.

We compare this with the integral surgeries long exact sequence, according to
which we have
F
?-HF þ ðS03 ðKÞ; sÞ-HF þ ðSp3 ðKÞ; ½d 1Þ ! HF þ ðS 3 Þ-? :
Now, if we endow HF þ ðSp3 ðKÞ; ½d 1Þ and HF þ ðS 3 Þ with the filtrations given by
the absolute grading, then F has the form C þ L; where L is a sum of homogeneous
maps which map to lower order than C (by at least 2d 2; cf. Eq. (12)). It then
follows that the induced map on homology c þ c is surjective, and also that this
kernel is identified with the kernel of c: &
5. Adjunction inequalities
d
In this section, we prove the following adjunction inequality for HFK
HFK:
Theorem 5.1. Let KCY be an oriented knot, and suppose that HFK d
HFKðY ; K; sÞa0:
Then, for each Seifert surface K for F of genus g40; we have that %
j/c1 ðsÞ; ½F̂ Sjp2gðF Þ:

%
Proof. Following Section 7 of [18] (see especially Lemma 7.3), we can construct a
special Heegaard diagram ðS; a; b0 ,m; w; zÞ for the knot K with the following
properties. In the corresponding diagram for Y0 ðKÞ (where the meridian m is
exchanged for the longitude l; but we have not yet wound the longitude along m; as
pictured in Fig. 3), there is a periodic domain P which bounds l,a1 ; all its
multiplicities are one and zero and its Euler characteristic is 2g: Moreover, the
meridian m meets the support of P in an arc which meets none of the other attaching
circles.
Let x; now, be any intersection point generating CFK d ; KÞ; and let x0 be the
CFKðY
nearby intersection for the Y0 Heegaard diagram, obtained now after winding.
Observe that for this Heegaard diagram, there is a variant of P; denoted P0 ; where
there are multiplicities 1; as well, now along some neighborhood of the meridian m:
Indeed, if x1 Ax denotes the intersection point on m; then x01 is supported in this
region. However, none of the other xi Ax0 are supported in the winding region. Note
that there must be another x2 Ax0 which is supported on a1 (i.e. it lies on the
boundary of P0 ; where the multiplicity on one side is þ1). See Fig. 6 for an
illustration.
ARTICLE IN PRESS
P. Ozsvath, 85
α1
,
x1
− 0
x1 +
,
0 λ
w λ
Fig. 6. Adjunction inequality. An illustration for the proof of Theorem 5.1. Here, the dashed circle is the
original longitude, while l0 is the one obtained after winding. We have indicated some of the local
multiplicities of the periodic domain P0 obtained after winding, by writing þ for the regions where its local
multiplicity is þ1; where it is 1; and 0 where it vanishes. (These can be obtained by realizing that the
boundary of P is a1 ,l0 ; with the orientations indicated by the arrows.)
Now, according to Eq. (9), we see that
/c1 ðs0 ðx0 ÞÞ; ½F̂ S ¼ 2g þ #ðxi in the interior of PÞX 2g:
The result now follows by conjugation invariance (Proposition 3.10). &
6. Examples
Before turning to some general calculational devices for CFK N ; we give here a few
Heegaard genus two examples where CFK N ðY ; KÞ can be explicitly determined
through more direct means.
The key trick which facilitates these calculations is the following:
Proposition 6.1. Let ðS; a; b; w; zÞ be a doubly pointed Heegaard diagram of genus g:

Suppose that the curves a1 and b1 intersect in a single point, and suppose that b1 meets
none of the other ai for i41: Consider the doubly pointed Heegaard diagram
ðS0 ; a0 ; b0 ; w0 ; z0 Þ of genus g 1; where S0 is obtained from S by surgering out b1 ; and
a0 ; b0 ; w0 ; and z0 are obtained by viewing fa2 ; y; ag g; fb2 ; y; bg g; w; and z as supported
in S0 : Then,
CF N ðS; a; b; w; zÞDCF N ðS0 ; a0 ; b0 ; w0 ; z0 Þ:
Proof. In the case where a1 meets none of the curves in fb2 ; y; bg g; this is a
statement of the stabilization invariance of the doubly pointed complex. It is easy to
reduce to this case by performing a series of handleslides of the bi (for i41) which
ARTICLE IN PRESS
P. Ozsvath,
meet a1 ; over b1 (along a subarc in a1 which connects bi with b1 ). Note that these
handleslides do not cross either basepoint. &
6.1. The trefoil
Start with a genus two Heegaard diagram for the left-handed trefoil. This can be
destabilized once (Proposition 6.1) to obtain the following picture shown in Fig. 7,
which takes place inside a torus. Note that we have destabilized out the meridian of
the knot.
Proposition 6.2. The filtered chain complex CFK N ðS 3 ; KÞ is freely generated as a

Z-module by generators ½x1 ; i; i þ 1; ½x1 ; i þ 1; i; and ½x0 ; i; i; where i is an arbitrary
integer. The differential given by
@½x0 ; i; i ¼ 0;
@½x1 ; i; i þ 1 ¼ ½x0 ; i; i;
@½x1 ; i þ 1; i ¼ ½x0 ; i; i:
Moreover, the absolute grading of the generator ½x0 ; 0; 0 is 1:
Proof. That the chain complex has the stated form follows readily from the fact
there are two orientation-preserving embedded Whitney disks in the torus: f1 ;
which connects x1 to x0 ; and f1 ; which connects x1 to x0 : So, by the Riemann
mapping theorem, both admit unique holomorphic representatives (up to transla-
tion). Moreover, f1 contains the basepoint w (and not z), while f1 contains z (and
not w).
Strictly speaking, this information suffices to identify the chain complex CF N for
the doubly pointed Heegaard diagram (generated by ½x; i; j for intersection points
xAfx1 ; x0 ; x1 g; and i; jAZ), which splits as a direct sum of chain complexes
isomorphic to CFK N ðS3 ; KÞ: It is immediate from the symmetry property of HFK d
3
(cf. Proposition 3.7) that the summand CFK ðS ; KÞ is generated by ½x0 ; i; i;
N
½x1 ; i; i þ 1 and ½x1 ; i þ 1; i:

To calculate the absolute grading, we use Lemma 3.6, and the observation that the
complex generated by ½x1 ; 0; 1; ½x0 ; 0; 0; and ½x1 ; 0; 1 (thought of as a quotient of
a subcomplex of CFK N ðY ; KÞ) calculates HF c ðS 3 Þ: In this complex, the generator
½x1 ; 0; 1 is the only one which persists in homology, so it must have degree
zero. &
As an easy application of the results from Section 4, one could give yet another
calculation of HF þ for three-manifolds which are surgeries along the trefoil.
Moreover, the technique employed above can be readily generalized to give a
combinatorial description of CFK for an arbitrary two-bridge knot. However, since
HF þ of three-manifolds obtained as integral surgeries along such knots has already
ARTICLE IN PRESS
P. Ozsvath, 87
x1 x_1
β x0
α
z
Fig. 7. Destabilized doubly pointed Heegaard diagram for the trefoil. This picture is meant to take place
in a torus, where the two hollow circles are identified.
been calculated by Rasmussen in [25], we do not pursue this any further here, turning
instead to a related calculation.
6.2. Examples from two-bridge links
There is a class of knots admitting genus two Heegaard diagrams, but which are
not two-bridge knots in the usual sense. Our aim here is to show how to reduce the
calculation of the complex CFK for this class of knots to a purely combinatorial
problem (cf. Propositions 6.3 and 6.4 below).
Recall that a two-bridge link in S3 is a link which is contained in a Euclidean
neighborhood B on which there is function f : B-R whose restriction to the link has
exactly two local maxima. We consider links of this type which have two
components. Such links can be described by certain (four-stranded) braids s:
Specifically, think of a braid as a mapping class s of the plane with four marked
points fa1 ; b1 ; a2 ; b2 g which permutes the points. Let I1 and I2 be a pair of disjoint
arcs which connect a1 to b1 and a2 to b2 ; respectively. We construct (the projection
of ) a link by joining I1 and I2 to sðI1 Þ and sðI2 Þ; respectively, along their boundaries,
so that I1 and I2 pass ‘‘over’’ their images under s: When the braid s fixes the subsets
fa1 ; b1 g and fa2 ; b2 g; the link has two components. In fact, by introducing twists
which leave the isotopy class of the link unchanged, we can assume the braid fixes the
four points (and indeed that it fixes a small tubular neighborhood of these points).
Let K1 ,K2 be a two-component two-bridge link. It is easy to see that the two
components of the link are each individually unknotted. Thus, if we perform 71
surgery on K1 (or, more generally, 1=n where n is an arbitrary integer), then the
ambient three-manifold is still diffeomorphic to S 3 ; and we let K denote the image of
K2 under this diffeomorphism. Thus, K is obtained from K2 by introducing a
‘‘Rolfsen twist’’ along the linking circle K1 (see for example [9] for a detailed
description of this operation; see Fig. 9 for a picture of the result of performing the
operation on the link shown in Fig. 8).
Let s denote the braid associated to a two-component link K1 ,K2 : We construct
a doubly pointed genus one Heegaard diagram for K from s as follows. Delete a pair
of disks A1 and B1 centered at a1 and b1 : The Heegaard surface is formed by
attaching a cylinder C1 to S2 A1 B1 along its two boundary circles. We let a2 and
ARTICLE IN PRESS
P. Ozsvath,
K1
K2
Fig. 8. A two-bridge link.
Fig. 9. A knot with Heegaard genus two. This knot is obtained by ‘‘blowing down’’ (with sign þ1) the
component K1 of the link from Fig. 8.
b2 serve as our two basepoints w and z; respectively. We consider the curve b

obtained from closing up the arc I1 -ðS2 A1 B1 Þ by an arc inside the attached
cylinder C1 : Similarly, we consider the curve a obtained by closing up sðI1 Þ by
an arc inside C1 : Note that there is some freedom in how we close off a and b
(inside C1 ). In particular, by introducing Dehn twists supported in the cylinder
on the a; we can arrange for the algebraic intersection number #a-b to be an
arbitrary integer.
ARTICLE IN PRESS
P. Ozsvath, 89
Proposition 6.3. Consider the doubly-pointed, genus one Heegaard diagram obtained
as above, where we have arranged for the curves a and b to have #a-b ¼ 71: This
Heegaard diagram is a destabilization of the Heegaard diagram for the knot K
obtained from the two-bridge link K1 ,K2 by ‘‘blowing down’’ K1 (given framing 71).
Proof. In the standard way (see, for example, [9]), a two-bridge link gives a genus
two Heegaard diagram of S 3 : This is related to the automorphism s as follows. We
consider the sphere with four distinguished disks A1 ; B1 ; A2 ; and B2 (neighborhoods
of the a1 ; b1 ; a2 ; and b2 ; respectively) each fixed by s: The Heegaard surface S2 is
obtained by attaching cylinders C1 and C2 to S 2 A1 B1 A2 B2 which connect
@A1 to @B1 and @A2 to @B2 ; respectively. Meridians m1 and m2 for the two
components K1 and K2 are appropriately oriented, embedded, homologically non-
trivial circles supported in C1 and C2 ; respectively. The remaining circles, g1 and g2 ;
are obtained by closing off the arcs sðI1 Þ and sðI2 Þ inside C1 and C2 ; respectively.
Longitudes l1 and l2 for K1 and K2 are obtained by closing off I1 and I2 in C1 and
C2 ; respectively. Thus, ðS2 ; fg1 ; g2 g; fm1 ; m2 gÞ describes S3 : The Heegaard diagram
for a three-manifold obtained by integral surgery along K1 is obtained from the
above by replacing m1 by the curve d obtaining by juxtaposing l1 and some number
of copies of m1 : Thus, when d is chosen so that #ðg1 -dÞ ¼ 71; then
ðS2 ; fg1 ; g2 g; fdg; m2 Þ is a Heegaard diagram for the knot induced from K2 inside
3
S71 ðK1 Þ: The corresponding doubly pointed Heegaard diagram ðS2 ; fg1 ; g2 g;
fd; m2 g; w; zÞ clearly has the property that m2 meets g2 in a single point (inside C2 ).
Thus, we can destablizing as in Proposition 6.1, to obtain the doubly pointed
Heegaard diagram described in the statement of the present proposition. &
Thus, CFK for knots obtained from two-bridge links is reduced to a calculation of
the doubly-pointed chain complex for the torus. Such a calculation is purely
combinatorial, i.e. independent of the choice of complex structure on the torus, as we
shall see in the following results (see especially Proposition 6.4 below).
When analyzing Floer homology taking place in the torus T; we find it convenient
to pass to the universal covering space p : C-T: To this end, we will assume that a
and b are a pair of homotopically non-trivial, embedded curves in the torus T: In this
case, p1 ðaÞ (and also p1 ðbÞ) is a properly embedded submanifold of C
homeomorphic to R Z: Let a* be one connected component of p1 ðaÞ and b* be a
connected component of p1 ðbÞ: Given reference points w and z in the torus, p1 ðwÞ
and p1 ðzÞ lift to lattices W and Z; respectively. Moreover, CFK N ðS3 ; KÞ is
generated by triples ½x̃; i; j subject to the relation in Eq. (11) where here the x̃ are
intersection points of a* and b: * Indeed, by path lifting, given x; yAa-b; the space of
homotopy classes of disks p2 ðx; yÞ is identified with the space of homotopy classes of
Whitney disks in C interpolating connecting the corresponding lifts x̃; ỹA*a-b*
(a space denoted fAp* *
2 ðx̃; ỹÞ). Clearly, if fAp2 ðx; yÞ and fAp2 ðx̃; ỹÞ is its lift, then
*
nw ðfÞ is obtained as a sum over all w̃AW of nw̃ ðfÞ: It is also not difficult to see an
identification of moduli spaces MðfÞDMðfÞ: *
ARTICLE IN PRESS
P. Ozsvath,
Proposition 6.4. Suppose that a and b are two homotopically non-trivial, embedded
curves in the torus T; and let a* and b* be corresponding curves in the universal cover C:
Let x̃; ỹA*a-b* be a pair of intersection points, and let fAp
* 2 ðx̃; ỹÞ be the homotopy
* ¼ 1; we have that #Mð
class of Whitney disk connecting x̃ and ỹ: Then if mðfÞ * ¼1
# fÞ
*
if and only if DðfÞX0:
Before proving this proposition, it is useful to have the following formula for the
Maslov index. To state it, we define the local multiplicity of f * at x̃ as follows. Given
*fAp2 ðx̃; ỹÞ; there are four regions in C a* b* which contain x̃ as a corner point.
Choosing interior points z1 ; y; z4 in these four regions, define
* * *
* ¼ nz1 ðfÞ þ nz2 ðfÞ þ nz3 ðfÞ þ nz4 ðfÞ:
n% x̃ ðfÞ ð15Þ
4
* similarly, only now choosing the four points in the region containing
We define n% ỹ ðfÞ
ỹ as a boundary.
Lemma 6.5. Suppose that a and b are two homotopically non-trivial, embedded curves
in the torus T; and let a* and b* be corresponding curves in the universal cover C: Given
*
x̃; ỹAp* 2 ð*a; bÞ;
* ¼ 2ðn% x̃ ðfÞ
mðfÞ * þ n% ỹ ðfÞÞ:
*
Proof. As we go from x̃ to ỹ in a* ; we encounter finitely many intersection points

a* -b:* Let fx̃1 ; y; x̃n g denote this sequence of points, arranged in the order they are
encountered, so that x̃1 ¼ x̃ and x̃n ¼ ỹ: There is a canonical Whitney disks
f* i Ap* 2 ðx̃i ; x̃iþ1 Þ: the domain is bounded by the Jordan curve obtained by juxtaposing
the two subarcs of a* and b* which connect x̃i and x̃iþ1 (oriented so that we go from x̃i
to x̃iþ1 in the a* -arc). Thus, the support of Dðfi Þ is an embedded disk whose local
multiplicity is either þ1 or 1: Clearly, the, homotopy class f* can be decomposed as
juxtapositions of these canonical Whitney disks, i.e. f * ¼ f* 1 ? f * n1 :
We prove the lemma by induction on the length n: Clearly, when n ¼ 1; both
quantities are zero.
For the inductive step decompose f * ¼f *1 f
* 0 ; where f* 0 Ap* 2 ðx̃2 ; x̃n Þ has length
n 1: Let e denote the local multiplicity inside Dðf1 Þ: A case-by-case analysis shows
that
* þ nx̃ ðfÞÞ
2ðnx̃1 ðfÞ * ¼ 2ðnx̃ ðf* 0 Þ þ nx̃ ðf
* 0 ÞÞ þ e; ð16Þ
2 2 n
* 1 Þ ¼ e: Thus, the result follows from Eq. (16), using the

while it is easy to see that mðf
inductive hypothesis, and the additivity of the Maslov index under juxtaposition.
ARTICLE IN PRESS
P. Ozsvath, 91
The cases required for establishing Eq. (16) are divided according to the order in
* Observe that for this ordering,
which x̃1 ; x̃2 ; and x̃n are encountered as we traverse b:
x̃n ox̃1 :
For example, suppose that x̃n ox̃1 ox̃2 : In this case, the local multiplicities of the
four regions containing x̃1 have the form m þ e; occurring three times, and m;
occurring once, for some integer m: Now, traversing the arc in a* from x̃1 to x̃2 ;
we see that the four neighboring multiplicities are: m (with multiplicity two) and
m þ e (also with multiplicity two). It follows readily that
nx̃2 ðf* 0 Þ ¼ nx̃1 ðfÞ

* 1 e;
2 nx̃n ðf* 0 Þ ¼ nx̃n ðfÞ;
*
verifying Eq. (16). The case where x̃2 ox̃n ox̃1 works the same way.
Finally, when x̃n ox̃2 ox̃1 ; we have instead that
nx̃2 ðf* 0 Þ ¼ nx̃1 ðfÞ;

* * 0 Þ ¼ nx̃ ðfÞ
nx̃n ðf 2
* 1 e;
2
also giving Eq. (16). &
We now turn to a proof of Proposition 6.4.
Proof of Proposition 6.4. Clearly, if f* has a holomorphic representative, then

*
DðfÞX0:
In the more interesting direction, suppose that fAp * 2 ðx̃; ỹÞ is a homotopy class
with mðfÞ ¼ 1 and DðfÞX0: Clearly, f is determined by the restriction of f* to its
* * *
boundary. Moreover, its boundary curve C is formed from an arc a in a* from x̃ to ỹ
and an arc b in b* from ỹ to x̃: Of course, C is null-homotopic. Our goal is to
construct a covering space of the plane with two disks removed, in which the lift C 0
of C is a null-homotopic, embedded circle.
To this end, Lemma 6.5, together with the hypothesis that DðfÞX0; * shows that
three out of the four local multiplicities around each of x̃ and ỹ vanish, and the
remaining ones are þ1: We choose a pair of disks, D1 and D2 ; centered at points on b*
with D1 just after x̃; and the other just before ỹ in b* (with respect to the orientation
induced on b* from @DðfÞ), * choosing the disks small enough to be disjoint from the
support of DðfÞ; * and so that they are disjoint from a* : Clearly, ðD1 ,D2 Þ-b*
*
separates b into three components, one of which is compact (containing both x̃ and
ỹ) and two of which are non-compact. Let g denote the compact subarc. Intersection
number with g induces a homomorphism of p1 ðC D1 D2 Þ to Z; and hence a
covering space X of C D1 D2 : This covering space is easily seen to be an infinite
strip with a discrete, countable set of points L removed (corresponding to the point
at infinity in C D1 D2 ). In X ; the pre-image of g is an infinite collection of
separating arcs (connecting the two boundaries), of which we choose one, denoted g0 :
Similarly, the pre-image of a* is an infinite collection of arcs (which begin and end in
the puncture points), of which we choose one, a* 0 : Let x̃0 and ỹ0 be the lifts in a* 0 -g0 of
x̃ and ỹ respectively. (cf. Fig. 10 for an illustration).
ARTICLE IN PRESS
P. Ozsvath,
D2
0 ∼
0 ,
y ∼
0 1 ∼
α
y
∼ 0
1β 2
... ∼γ ,
0∼x 1 1
...
00
∼x, ∼,
α
D1
Fig. 10. Lifting a domain. On the left-hand-side, we have illustrated a domain connecting x̃ and ỹ;
corresponding to a map f* with mðfÞ
* ¼ 1: The ambient space is C; with the disks D1 and D2 deleted. On the
right, we have illustrated an induced domain in X : Here, the picture is an infinite strip with an infinite
sequence of puncture points, indicated here by the grey circles. The top boundary corresponding to the
universal cover of @D2 ; and the bottom corresponding to the universal cover of @D1 :
The lift C 0 of C to X ; of course, is obtained by adjoining the subarc a0 in a* 0 from x̃0

to ỹ0 to the subarc b0 in b* 0 from ỹ0 to x̃0 : We claim that a0 and b0 meet only along their
boundary, and hence that C 0 is an embedded curve. To this end, observe that g0
separates X into two halves. Call the ‘‘right’’ half the one which the curve a0 points
towards, at x̃0 : Let q̃0 be the first crossing (with respect to the a0 orientation) after x̃
encountered as we traverse a0 : With respect to the induced orientation on g; x̃0 and ỹ0
are extremal. It is an easy consequence now that one of the four local multiplicities
near q̃0 is 1: Indeed, by following parallel to a0 ; we can find a curve d supported on
the ‘‘right’’, connecting one of the regions near x̃0 with zero multiplicity to one of the
four neighboring regions near q̃0 ; so that d is supported near a* 0 ; its projection to C is
disjoint from a* ; and, since its endpoints are both near g0 but to the right,
the intersection number of the projection of d with b* is zero. This shows that in
R2 D1 D2 there is a region containing q̃ (the projection to R2 D1 D2 of q̃0 ) to
the right of g; which has multiplicity zero. Since q̃0 falls between x̃0 and ỹ0 ; it follows
that the facing region on the other side of g has multiplicity 1: (See the illustration
in Fig. 11.) This contradicts the hypothesis that DðfÞX0: *
0
We have established that C is an embedded curve in X : Thus, by the Jordan curve
theorem, C 0 separates the strip into two regions, a disk D; and another region which
contains the boundary of the strip. Since DðfÞ * is compact it follows readily that
none of the points in L is contained in the disk; i.e. C 0 bounds a disk in X : (More
precisely, the disk has two corner points at the two points in C 0 where a0 and b0 meet.)
It follows now by the Riemann mapping theorem that there is a unique holomorphic
Whitney disk representing this disk; and hence by path-lifting, it follows that
#MðfÞ * ¼ 1; as well. &
As an illustration of these techniques, consider the link pictured in Fig. 8. Label

the generators of the braid group on four elements by t1 ; t2 ; t3 (so that ti transposes
the ith and ði þ 1Þth strands). The braid element s corresponding to the link, then
ARTICLE IN PRESS
P. Ozsvath, 93
~,
0 y 0
0 1
~ ,
γ
−1 0
... ,
q~ 1 ...
~, δ
α
~,
0 x 1
0 0
Fig. 11. Proof of Proposition 6.4. In the case where the arc in a0 meets the arc b0 in than two points, as we
have illustrated, DðfÞ * is forced to be non-positive, provided mðfÞ* ¼ 1: The light curve d illustrated here
has projection to R D1 D2 which is disjoint from a* ; and intersection number zero with b:
2 * Thus, the
*
multiplicities of DðfÞ at the two endpoints must coincide (indeed, they must vanish). It follows that just on
the other side of b* 0 from the endpoint near q̃0 ; the local multiplicity of the domain DðfÞ
* is 1:
can be written as a product t22 t23 t4

2 : The Heegaard diagram, then, is obtained by
corresponding Dehn twists in the sphere with four marked points. Blowing down the
other component (with sign þ1), we obtain the diagram illustrated in Fig. 12. Note
that this gives us the knot pictured in Fig. 9 (which appears in knot tables as the knot
942 ). Moreover, lifting this doubly pointed Heegaard diagram to the universal
covering space C for the torus, we obtain the picture shown in Fig. 13.
In this case, a* -b* consists of n ¼ 9 points. Thus, C a* b* has eight compact,
connected components (and, more generally, n 1 compact connected components),
which we label D1 ; y; D8 ; under the numbering convention as follows. Let Ci denote
the Jordan curve obtained by juxtaposing the segment of a* connecting x̃i to x̃iþ1 with
the corresponding segment in b: * Then, Di is the component of C a* b* which is
inside Ci ; and whose boundary contains x̃i : For example, as we see from the figure,
C1 ; whose interior represents a flow from x̃1 to x̃2 ; can be decomposed as a sum
D1 þ D3 ; C2 ; which gives a flow from x̃3 to x̃2 ; bounds simply D2 ; while C3 bounds
D3 ; giving a flow from x̃3 to x̃4 : It follows that the domain of the Whitney disk from
x̃4 to x̃1 is given by D1 þ D2 : This disk has m ¼ 1; and it clearly supports a single
flowline (indeed, it is represented by an embedded disk in the plane). Moreover, the
domain contains no points point from the W lattice and one from the Z lattice, so we
have determined that the x̃4 coefficient of @½x̃1 ; i; j is 7½x̃4 ; i; j 1:
Proceeding in the like manner, one verifies that CFK N ðS 3 ; KÞ (with the notational
shorthand from Section 3.4) is generated as a Z-module by generators indexed by
iAZ of the form:
½x̃1 ; i þ 1; i; ½x̃2 ; i; i; ½x̃3 ; i; i 1;

½x̃4 ; i þ 1; i 1; ½x̃5 ; i þ 1; i þ 1; ½x̃6 ; i 1; i þ 1; ð17Þ
½x̃7 ; i 1; i; ½x̃8 ; i; i; ½x̃9 ; i; i þ 1;
ARTICLE IN PRESS
P. Ozsvath,
β z
w
Fig. 12. A doubly pointed Heegaard diagram. This is the Heegaard diagram for the knot from Fig. 9, as
constructed in Proposition 6.3. The two empty circles are glued along a cylinder, so that no new
intersection points are introduced between the curve a (the darker curve) and b (the lighter, horizontal
curve). Note that there are nine intersection points between a and b; which, when counted with sign, give
an intersection number of þ1:
~
α
~
β
1 4 3 2 5 8 7 6 9
* are pictured here as dark

Fig. 13. Lift of Fig. 12 to the universal cover. Two of the lifts of a and b; a* and b;
curves, while other lifts are shown with dashed lines. The lattice p1 ðfwgÞ is represented by the lightly filled
circles, while the lattice of p1 ðfzgÞ is represented by the darkly-filled circles, while. The nine intersection
points of a* and b* are labeled in the order in which they are encountered along a* :
ARTICLE IN PRESS
P. Ozsvath, 95
and the boundary operator on CFK N ðS 3 ; KÞ (up to some signs) is given by
@½x̃1 ; i þ 1; j ¼ ½x̃2 ; i; i þ ½x̃4 ; i þ 1; i 1;
@½x̃2 ; i; i ¼ ½x̃3 ; i; i 1;
@½x̃3 ; i; i 1 ¼ 0;
@½x̃4 ; i þ 1; i 1 ¼ ½x̃3 ; i; i 1;
@½x̃5 ; i þ 1; i þ 1 ¼ ½x̃2 ; i; i þ ½x̃4 ; i þ 1; i 1 þ ½x̃6 ; i 1; i þ 1 þ ½x̃8 ; i; i;
@½x̃6 ; i 1; i þ 1 ¼ ½x̃7 ; i 1; i;
@½x̃7 ; i 1; i ¼ 0;
@½x̃8 ; i; i ¼ ½x̃7 ; i 1; i;
@½x̃9 ; i; i þ 1 ¼ ½x̃6 ; i 1; i þ 1 þ ½x̃8 ; i; i:
In calculations, it is often useful to depict this schematically, as pictured in Fig. 14.

Consider the complex CFK 0; ðY ; KÞ; which is generated by the nine generators as
enumerated in Eq. (17), but with i ¼ 0: We obtain a complex whose homology is Z;
supported in a single dimension. In fact, this homology is represented by the cycle
½x̃1 ; 0; 17½x̃5 ; 0; 0: Indeed (as in to Lemma 3.6) this generator represents the Floer
homology class of HF c ðS3 ÞDZ; so its absolute grading is zero.
6 9
5
7
8
2
1
3 4
Fig. 14. The chain complex. This depicts the differential for CFK N ðS3 ; KÞ: The integers here correspond
to intersection points with the labeling conventions from the text: i.e. the integer k corresponds to x̃k : An
arrow denotes a non-trivial differential, and (rough) coordinates in the plane correspond to the indices i; j:
Of course, the complex CFK N ðS3 ; KÞ consists of infinitely many copies of this complex; more precisely it
is the tensor product of this complex with Z½U; U 1 :
ARTICLE IN PRESS
P. Ozsvath,
In particular, we have the following:
Proposition 6.6. Let K denote the knot 942 (illustrated in Fig. 9). Various invariants for
the knot K are given by
8
> Zð3Þ if i ¼ 2;
>
> 2
>
> Zð2Þ if i ¼ 1;
>
>
>
>
< Z2 "Z if i ¼ 0;
d 3 ; K; iÞD ð1Þ ð0Þ
HFKðS
HFK
>
> 2
Zð0Þ if i ¼ 1;
>
>
>
>
> Zð1Þ
> if i ¼ 2;
>
:
0 otherwise;
8
> T
< 1=2 "T 1=2 if i ¼ 0;
þ 3
HF ðS0 ðKÞ; iÞD Z if i ¼ 71;
>
:
0 otherwise:
In particular, d71=2 ðS03 ðKÞÞ ¼ 71=2:
Proof. The calculation of HFKðS3 ; KÞ follows from inspection of the chain complex.
For instance, observe that CFK d 3 ; K; 2Þ is generated by a single element
CFKðS
c ðS 3 ; K; 0Þ is generated by three elements ½x̃2 ; 0; 0; ½x̃5 ; 0; 0; and
½x̃6 ; 0; 2; ; while CF
½x̃8 ; 0; 0:
To calculate HF þ ðS03 ðKÞÞ; we first use Theorem 4.1 to identify HF þ ðSp 3
ðKÞ; ½sÞ for
b þ
large p with H ð CFK ðY ; K; sÞÞ (with the grading shift). The graded long exact
sequence for integral surgeries then shows that HF þ ðS03 ðKÞÞ has the claimed form. &
7. Connected sums of knots
Recall that there are multiplication maps
c ðY1 Þ#Z CF
CF c ðY1 #Y2 Þ;
c ðY2 Þ-CF
CF ðY1 Þ#Z½U CF ðY2 Þ-CF ðY1 #Y2 Þ;
CF N ðY1 Þ#Z½U;U 1 CF N ðY2 Þ-CF N ðY1 #Y2 Þ;
which induce homotopy equivalences.

To state the generalization for knot invariants, we use the following convention:
the tensor product of two Z-filtered (resp. Z"Z-filtered) filtered complexes ðC1 ; F1 Þ
and ðC2 ; F2 Þ is naturally a Z-filtered (resp. Z"Z-filtered) complex ðC1 #C2 ; F# Þ;
ARTICLE IN PRESS
P. Ozsvath, 97
where the tensor product filtration is defined by
F# ðx1 #x2 Þ ¼ F1 ðx1 Þ þ F2 ðx2 Þ
for arbitrary homogeneous generators x1 AC1 and x2 AC2 :

Given t1 ASpinc ðY1 ; K1 Þ and t2 ASpinc ðY2 ; K2 Þ; let t3 ¼ t1 #t2 ASpinc
ðY1 #Y2 ; K1 #K2 Þ be characterized by the properties that
* if si ASpinc ðYi Þ denote the Spinc structures underlying ti for i ¼ 1; y; 3; where
Y3 ¼ Y1 #Y2 ; then s3 is cobordant to s1 ,s2 under the natural cobordism from Y3
to Y1 ,Y2
* if F1 and F2 are Seifert surfaces for K1 and K2 ; and F3 is the corresponding Seifert
surface for K1 #K2 obtained by boundary connected sum, then
/c1 ðt3 Þ; ½F3 S ¼ /c1 ðt1 Þ; ½F1 S þ /c1 ðt2 Þ; ½F2 S:
Theorem 7.1. There are filtered chain homotopy equivalences
CFK 0; ðY1 ; K1 Þ#Z CFK 0; ðY1 ; K1 Þ-CFK 0; ðY1 #Y2 ; K1 #K2 Þ;
CFK o; ðY1 ; K1 Þ#Z½U CFK o; ðY1 ; K1 Þ-CFK o; ðY1 #Y2 ; K1 #K2 Þ;
CFK N ðY1 ; K1 Þ#Z½U;U 1 CFK N ðY1 ; K1 Þ-CFK N ðY1 #Y2 ; K1 #K2 Þ;
which respect the splittings of the complexes according to their summands; in

particular,
CFK N ðY1 #Y2 ; t3 Þ

M
D CFK N ðY1 ; t1 Þ#CFK N ðY2 ; t2 Þ:
ft1 ;t2 ASpin ðY1 ;K1 ÞSpinc ðY2 ;K2 Þjt1 #t2 ¼t3 g
c
Proof. We sketch here the definition of the usual connected sum map (see Section 6
of [18]). Let ðS1 ; a1 ; b1 ; w1 Þ and ðS2 ; a2 ; b2 ; w2 Þ be pointed Heegaard diagrams
representing Y1 and Y2 : There are maps
CF N ðY1 Þ-CF N ðY1 #g2 ðS 2 S 1 ÞÞ;
c ð#g1 ðS 2 S1 Þ#Y2 Þ
CF N ðY2 Þ-CF
sending x1 and x2 to x1 Y1 and Y2 x2 ; respectively, where Y1 and Y2 are

intersection points representing the canonical (top-dimensional) HF p0 class for the
connected sum of S 2 S 1 : We then compose this with a count f of holomorphic
triangles in the Heegaard triple ðS1 #S2 ; a1 a2 ; b1 a2 ; b1 b2 Þ: In the above
expression, any time some g-tuple of circles is repeated, it is to be understood that
one takes a small exact Hamiltonian translate to make the intersections transverse.
ARTICLE IN PRESS
P. Ozsvath,
The refinement
f N : CFK N ðY1 ; K1 Þ#Z½U;U 1 CFK N ðY2 ; K2 Þ-CFK N ðY1 #Y2 ; K1 #K2 Þ
is defined as follows. Consider the map
f ð½x1 ; i1 ; j1 #½x2 ; i2 ; j2 Þ
X X
¼ #MðcÞ ½y; i1 þ i2 nw1 ðcÞ; j1 þ j2 nz2 ðcÞ;
yATa1 a2 -Tb1 b2 fcAp2 ðx1 Y2 ;Y1 x2 ;yÞg
where we continue with the notation from above, with the understanding that the
meridians m1 and m2 appear in the g-tuples b1 and b2 ; respectively. Of course, the
above map is a perturbation of the map sending ½x1 ; i1 ; j1 #½x2 ; i2 ; j2 to ½x1
x2 ; i1 þ i2 ; j1 þ j2 : The map clearly respects filtrations, since nz2 ðcÞX0 whenever c
has a holomorphic representative. We go from this Heegaard diagram to a Heegaard
diagram to the following Heegaard diagram for the knot ðY1 #Y2 ; K1 #K2 Þ;
ðS; a1 ,a2 ; b3 ,ðb1 m1 Þ0 ,ðb2 m2 Þ0 ; m3 Þ;
where here the primes denote small exact Hamiltonian translates, m3 is a curve
obtained from m1 by a small exact Hamiltonian translate, and b3 is obtained as a
handleslide of m2 over m1 —in particular, the new Heegaard diagram for Y1 #Y2
differs from the ðS1 #S2 ; a1 a2 ; b1 b2 Þ by a single handleslide. Moreover, it is
easy to see that the composite map here respects relative filtrations.
We verify that this map respects the splitting according to Spinc ðYi ; Ki Þ: This is
clear on the level of Spinc ðYi Þ; thus, it remains to show that if F1 and F2 are
compatible Seifert surfaces for K1 and K2 ; then
# 1 x2 ÞÞ; ½F1d
/c1 ðtðx1 ÞÞ; ½F̂1 S þ /c1 ðtðx2 ÞÞ; ½F̂2 S ¼ /c1 ðtðx #b F2 S; ð18Þ
where F1 #b F2 is the Seifert surface for K1 #K2 inside Y1 #Y2 obtained by boundary
connected sum. To this end, observe that if l1 and l2 are curves in the Heegaard
diagrams for Y1 and Y2 representing longitudes for the knots K1 and K2 ;
respectively, then l1 #l2 represents a longitude for K1 #K2 : Correspondingly, if P1
and P2 are the periodic domains (corresponding to F̂1 and F̂2 ) in ðY1 Þ0 ðK1 Þ and
ðY2 Þ0 ðK2 Þ; respectively, then the periodic domain for F1 #b F2 in ðY1 #Y2 Þ0 ðK1 #K2 Þ
is obtained from P1 and P2 by a boundary connected sum (see Fig. 15). Under this
operation, the Euler measure satisfies
wðP1 #b P2 Þ ¼ wðP1 Þ þ wðP2 Þ 1;
and it is easy to see that if x3 denotes the intersection point corresponding to x1 x2 ;

then
n% x3 ðP1 #b P2 Þ ¼ n% x1 ðP1 Þ þ n% x2 ðP2 Þ þ 12:

ARTICLE IN PRESS
P. Ozsvath, 99
µ1 µ2
w1 z1 w2 z2
λ1 λ2
x1 x,1 α1 x2 x,2 α2
µ3 β3 β3
w3 z3
λ3 λ3 λ3
x3 x,3 α1 x,,4 α2
Fig. 15. Connected sum of knots: before and after. The top picture represents the (disconnected) union of
two original manifolds ðY1 ; K1 Þ and ðY2 ; K2 Þ; while the second represents their connected sum (in
particular, in the second picture, the two open circles are identified). The curve m3 (corresponding to m1 in
the previous diagram) represents a meridian for K1 #K2 while the curve b2 which is obtained by
handlesliding m2 over m1 (after forming the connected sum) is a new attaching circle for Y3 ¼ Y1 #Y2 : The
circle l3 is the longitude for K1 #K2 : Note that b3 and l3 are the only curves which go through the
connected sum neck. The curve b3 is easily seen not to appear in the boundary of the periodic domain for
K1 #K2 ; while the curve l3 appears in it with multiplicity one. The pair x03 x004 corresponds under the
tensor product map to x1 x2 :
(Again, the reader is asked to consult Fig. 15.) Eq. (18) now follows immediately
from these observations and Eq. (9).
Having defined f N ; it is now easy to adapt the arguments for three-manifolds to
verify that f N is a chain homotopy equivalence of filtered complexes. (Moreover, the
corresponding constructions for CFK o; and CFK 0; proceed in the same way.) &
Note that the above theorem has the following immediate corollary (which can be
thought of as a refinement of the multiplicativity of the Alexander polynomial under
connected sums):
Corollary 7.2. Let t1 ASpinc ðY1 ; K1 Þ and t2 ASpinc ðY2 ; K2 Þ be Spinc structures, and let
t1 #t2 ASpinc ðY1 #Y2 ; K1 #K2 Þ denote the Spinc structure whose underlying Spinc
structure is cobordant to s1 ,s2 over Y1 ,Y2 (under the natural cobordism to Y1 #Y2 ),
and
/c1 ðt1 #t2 Þ; ½F1d c1 S þ /c1 ðt2 Þ; ½F
#b F2 S ¼ /c1 ðt1 Þ; ½F c2 S:
ARTICLE IN PRESS
P. Ozsvath,
Then, for each t3 ASpinc ðY1 #Y2 ; K1 #K2 Þ; we have

M
d 1 #Y2 ; K1 #K2 ; t3 ÞD
HFK
HFKðY d 1 ; K1 ; t Þ#CFK
H ðCFK
CFKðY d 2 ; K2 ; t ÞÞ:
CFKðY
1 2
t1 #t2 ¼t3
Proof. This follows immediately from the theorem. &
8. Long exact sequences
One can readily adapt the surgery long exact sequences from [18] to the context of
knot invariants, in several ways. We give here a discussion whose level of generality
is suitable for the applications to follow.
Recall that if g is a framed knot in a three-manifold Y ; and s is a Spinc structure
on the four manifold W ¼ Wg ðY Þ; then there are maps induced by counting
holomorphic triangles
c ðY Þ-CF
fˆW ;s : CF c ðYg Þ;

fW ;s : CF ðY Þ-CF ðYg Þ;
N N N
fW ;s : CF ðY Þ-CF ðYg Þ:
In the presence of a Seifert surface F for a knot, for each sASpinc ðY Þ; there is a
t0 ASpinc ðY ; KÞ extending s; which is uniquely characterized by the property that
/c1 ðt0 Þ; ½F̂ S ¼ 0:
Thus, we can identify,
CF N ðY ; sÞDCFK N ðY ; t0 Þ;
as Z-filtered complexes; and this isomorphism endows CF N ðY ; sÞ with an extra Z

grading (i.e. our representatives for CFK N are triples ½x; i; j; and the Z filtration is
given by the value of j). We sometimes suppress t0 from the notation, writing
CFK N ðY ; K; sÞ when F is understood from the context. (Note that this shorthand
coincides with the corresponding shorthand we used in the case where c1 ðsÞ is
torsion.)
Suppose now that K is an oriented knot in Y which is disjoint from g; and whose
linking number with g is zero. In this case, we can choose a Seifert surface F for Y
which is disjoint from g; thus, this can be closed up in Y or Yg to obtain surfaces F̂
and F̂ 0 :
Proposition 8.1. Let Y be a three-manifold with a null-homologous knot K; and let g be

a framed knot which is disjoint from K; and whose linking number with K is zero, and
ARTICLE IN PRESS
P. Ozsvath, 101
let Y 0 be the three-manifold induced from Y by surgery along g and K 0 be the induced
knot. Then, for each Spinc structure s; the map fW ;s induced by counting holomorphic
triangles respects the above induced Z-filtrations.
Proof. We describe here the case of CF N ðY Þ; the other two cases follow similarly.
Let ðS; a; b; c; w; zÞ be a Heegaard triple, where for i ¼ 2; y; g; bi is a small exact
Hamiltonian translate of gi ; with m ¼ b2 ; while the b1 and g1 are different (the g1 here
corresponds to the longitude of g; and b1 to its meridian). Following [19], the map
N N N 0 0
fW ;s : CFK ðY ; KÞ-CFK ðY ; K Þ
induced by the cobordism is defined by

X X
N
fW ;s ð½x; i; jÞ ¼ #MðcÞ ½y; i nw ðcÞ; j nz ðcÞ:
fyATa -Tg g fcAp2 ðx;Yb;g ;yÞjmðcÞ¼0g
We must verify that the range of this map is CFK N ðY 0 ; K 0 Þ as claimed, i.e. that if
/c1 ðsm ðxÞ; ½F ÞS þ ði j Þ ¼ 0; ð19Þ
then if cAp2 ðx; Y; yÞ; we also have that
/c1 ðsm ðyÞÞ; ½F# 0 S þ ði nw ðcÞ j þ nz ðcÞÞ ¼ 0;
as well.
As a preliminary remark, we claim that if P is a triply periodic domain for
the Heegard triple describing the cobordism from Y to Y 0 ; then nw ðPÞ ¼ nz ðPÞ:
The reason for this is that P can be viewed as a null-homology of g in Y (as the only
curve which appears in @P which is not isotopic to an attaching circle for Y is a
longitude for g). Thus, the multiplicity with which b2 appears in @P can be
interpreted as the linking number of g with m; which we assumed to vanish. But this
multiplicity is given by nw ðPÞ nz ðPÞ:
In view of these remarks, it suffices to verify that
/c1 ðsm ðxÞÞ; ½F S ¼ /c1 ðsm ðyÞÞ; ½F 0 S; ð20Þ
for two particular intersection points x; y; which can be connected by a triangle

cAp2 ðx; Y; yÞ with nw ðcÞ ¼ nz ðcÞ ¼ 0 (by appealing to Lemma 2.5). Indeed, it
suffices to consider the case where the point in x lying on b2 is a small perturbation of
the point on y lying on g2 (recall here that g2 is a small perturbation of b2 ). It is easy
to see that c induces a triangle c0 Ap2 ðx0 ; Y; y0 Þ; and hence a Spinc structure on the
cobordism from Y0 ðKÞ to Y00 ðK 0 Þ whose restrictions to Y0 ðKÞ and Y0 ðK 0 Þ are sw ðx0 Þ
and sw ðy0 Þ; respectively. Since ½F̂ and ½F# 0 are homologous in this cobordism (again,
we are using the fact that the linking number of g with K vanishes), Eq. (20) follows
at once, and hence also Eq. (19).
As defined, the map clearly respects filtrations. &
ARTICLE IN PRESS
P. Ozsvath,
We have now the following easy consequence of these observations, as combined

with the long exact sequences of [18]. Note that in the following statement, we let
M
d
HFK
HFKðY ; K; mÞ ¼ d
HFK
HFKðY ; K; tÞ:
ftASpinc ðY ;KÞj/c1 ðtÞ;½F̂ S¼2mg
Theorem 8.2. Let ðY ; KÞ be a knot, and g be a framed knot in Y K: Then, by

choosing all Seifert surfaces compatibly, the counts of holomorphic triangles give a long
exact sequence for each integer m:
f1 f2 f3
d 1 ðgÞ; K; mÞ ! HFK
?-HFK
HFKðY d 0 ðgÞ; K; mÞ ! HFK
HFKðY d
HFKðY ; K; mÞ ! ?
In cases where the groups are graded (e.g. when Y is a homology sphere), the maps f1
and f2 are homogeneous maps which lower the absolute gradings by 12; while f3 is a sum
of homogeneous maps, which are non-increasing in the absolute grading.
Proof. This follows easily from inspecting the proof of the surgery long exact sequence
d (see Theorem 9.12 of [18]), and observing that the homotopy equivalences also
for CFK
respect the filtrations, following arguments from Proposition 8.1 above.
Recall that the maps f1 ; f2 ; and f3 are sums of maps induced by Spinc structures
over the corresponding cobordisms. The statements about gradings now follow
immediately from the dimension formula of [19], stated in Eq. (12) above. &
It is not difficult to see that the above theorem generalizes the skein exact sequence
stated in the introduction. We return to this point in Section 10.
9. Some fibered examples
As an illustration of some of the above techniques, we give some calculations of

knot homology groups for some simple (genus one) fibered knots. These
calculations, together with basic properties of the knot invariants, allow us to
calculate HF þ ðS1 Sg ; tÞ (and also some other three-manifolds which fiber over the
circle), where t is some non-torsion Spinc structure. It is interesting to compare these
with Seiberg–Witten results, cf. [16,17], with the quantum cohomology calculations
of [3], and also with the ‘‘periodic Floer homology’’ of Hutchings, see [11].
Consider the three-component Borromean rings in S 3 ; and fix m; nAZ,fNg
(indeed, we will typically consider m; nAf0; 1; Ng). Perform m-surgery on one
component, n-surgery on another, and let Bðm; nÞ denote the remaining component,
thought of as knot in the surgered three-manifold. Thus, Bðm; nÞ is a null-
homologous knot inside Lðm; 1Þ#Lðn; 1Þ (with the understanding that Lð0; 1ÞDS 2
S 1 and Lð1; 1ÞDS3 ). By Kirby calculus, one can see that Bð1; 1ÞCS 3 is diffeomorphic
to the right-handed trefoil knot, and Bðm; NÞCLðm; 1Þ is an unknot (see Fig. 16).
ARTICLE IN PRESS
P. Ozsvath, 103
Fig. 16. A Kirby calculus picture of the ‘‘Borromean knot’’ BC#2 ðS2 S1 Þ:
In general, Lðm; 1Þ#Lðn; 1Þ Bðm; nÞ is diffeomorphic to the mapping torus of an

automorphism of the torus with a disk removed T 2 D: In particular, the complement
ðS2 S1 Þ Bð0; 0Þ is diffeomorphic to the product of a circle with T 2 D; and ðS 2
S 1 Þ Bð1; 0Þ is the mapping torus of a non-separating Dehn twist acting on T 2 D:
Remark 9.1. In the following three-manifolds Y ; there is a unique tASpinc ðY Þ with

c ðY ; sÞ ¼ 0 for all other Spinc
trivial first Chern class (and indeed, it is clear that HF
structures). Thus, there is no ambiguity in writing HFK d
HFKðY ; K; j Þ to denote
d c
HFKðY ; K; tj Þ; where tj ASpin ðY ; KÞ is characterized by the fact that it extends t;
HFK
and satisfies /c1 ðtj Þ; ½F̂ S ¼ 2j:
d of Borromean knots are given by:

Proposition 9.2. The knot homology groups HFK
8
> Zð0Þ if j ¼ 1;
>
>
<Z if j ¼ 0;
d 3 ; Bð1; 1Þ; j ÞD ð1Þ
HFKðS
HFK
> Zð2Þ if j ¼ 1;
>
>
:
0 otherwise;
8
> Z1 if j ¼ 1;
>
> ð2Þ
>
>
< 2 if j ¼ 0;
d 2 S 1 ; Bð0; 1Þ; j ÞD Zð12Þ
HFKðS
HFK
>
>
>
> Zð3=2Þ if j ¼ 1;
>
:
0 otherwise;
8
> Zð1Þ if j ¼ 1;
>
> 2
< Zð0Þ if j ¼ 0;
d
HFKð#
HFK 2
ðS2 S1 Þ; Bð0; 0Þ; j ÞD
>
> Z if j ¼ 1;
>
:
ð1Þ
0 otherwise:
ARTICLE IN PRESS
P. Ozsvath,
Proof. The first isomorphism follows immediately from the trefoil calculation from
Section 6.1.
d for Bð0; 1Þ could be made in the same spirit—by reducing
The calculation of HFK
to a genus one Heegaard diagram for S 2 S1 : Alternatively, one can use the long
exact sequence of Theorem 8.2:
f1 f2 f3
d 3 ; Bð1; NÞ; j Þ !
HFK
HFKðS d 2 S 1 ; Bð1; 0Þ; j Þ !
HFKðS
HFK d 3 ; Bð1; 1Þ; j Þ !
HFKðS
HFK ? :
Since Bð1; NÞCS3 is an unknot, HFK d 3 ; Bð1; NÞ; j Þ ¼ 0 when ja0: Moreover
HFKðS
d 2
the map f2 lowers degree by 12 (cf. Eq. (12)). Thus, the calculation of HFK
HFKðS
1
S ; Bð1; 0Þ; j Þ for ja0 is complete.
In the case where j ¼ 0; of course,
d 3 ; Bð1; NÞ; 0ÞDZð0Þ

HFK
HFKðS d 3 ; Bð1; 1Þ; 0ÞDZð1Þ ;
and HFKðS
HFK
while the map f3 is non-increasing in grading. Thus, it must vanish identically, and,
d 3 ; Bð1; 0Þ; j Þ for all j is
since f1 also lowers degree by 12; the computation of HFK
HFKðS
complete.
For Bð0; 0Þ; we now have
f1
d 2 S 1 ; Bð0; NÞ; j Þ !
HFK
HFKðS d
HFKð#
HFK 2
ðS2 S1 Þ; Bð0; 0Þ; j Þ
f2 f3
d 2 S 1 ; Bð0; 1Þ; j Þ ! ?
! HFK
HFKðS ð21Þ
Thus, the case where ja0 follows exactly as before.

Before continuing the calculation when j ¼ 0; it is useful to make some
observations. Specifically, recall that
M
d
HFK
HFKð# 2
ðS 2 S 1 Þ; Bð0; 0Þ; j Þ
j
is the homology of the graded object associated to a filtration of HFc ð#2 ðS 2 S1 ÞÞ:
d
Thus, since HFK
HFKð# 2 2 1
ðS S Þ; Bð0; 0Þ; j Þ ¼ 0 for jo 2; we obtain a natural map
d
Zð1Þ DHFK
HFKð# 2 c ð#2 ðS 2 S 1 ÞÞDZð1Þ "Z2 "Zð1Þ :
ðS 2 S1 Þ; Bð0; 0Þ; 1Þ-HF ð0Þ
We claim that this map actually induces an isomorphism onto the Zð1Þ factor on the
right-hand side. This follows immediately from the following commutative diagram:
D
d 3 ; Bð1; 1Þ; 1Þ
HFK
HFKðS ! c ðS3 ÞDZð0Þ
HF
fk gk
i
d
HFK
HFKð# 2
ðS2 S1 Þ; Bð0; 0Þ; 1Þ c ð#2 ðS2 S1 ÞÞ
! HF
ARTICLE IN PRESS
P. Ozsvath, 105
where here the two horizontal maps are induced by inclusions, the fact that the first
is an isomorphism follows readily from the calculation of Bð1; 1Þ; while the vertical
maps f and g are induced by the obvious cobordisms (here g is induced by the two-
handle additions, and f is the induced map on the knot complex). In particular, it is a
straightforward calculation that (cf. [19]) that g induces an isomorphism to
c ð1Þ ð#2 ðS2 S 1 ÞÞDZ: It follows that i is an isomorphism for degree 1; as
HF
claimed.
We return now to the calculation of HFK d
HFKð# 2
ðS 2 S 1 Þ; j Þ when j ¼ 0:
Recall that HFKd
HFKðS 2
S ; Bð0; NÞ; 0ÞDZð1=2Þ "Zð1=2Þ ; and of course HFK
1 d 2
HFKðS
S 1 ; Bð0; 1Þ; j ÞDZ2ð1=2Þ : Verifying the calculation thus reduces to showing that f3
( from long exact sequence (21)) surjects onto the Zð1=2Þ summand of HFK d 2
HFKðS
S 1 ; Bð0; NÞ; 0Þ: If it did not, the map would have to have cokernel, and hence the
rank (over possibly some finite field) of
M
G¼ d 1 ð#2 ðS 2 S 1 Þ; Bð0; 0Þ; j Þ
HFK
j
would be two. But this is impossible: we have already shown that HFK d
HFKð#ðS 2

1 c 2 2 1
S Þ; Bð0; 0Þ; 1Þ represents a generator for HF1 ð# ðS S ÞÞ: It also follows from
d
the previous calculation that HFK
HFKð# 2
ðS 2 S 1 ÞÞ is supported in dimension 1; so the
remaining generator must come from HFK d 1 ð#2 ðS 2 S 1 Þ; 0Þ: Indeed, by dimension
reasons, this generator cannot be killed by any higher differential (in the spectral
sequence connecting HFK d
HFKð# 2
ðS2 S1 Þ; Bð0; 0ÞÞ to HF c ð#2 ðS2 S 1 ÞÞ), and hence
contradicting the fact that HFc 1 ð#2 ðS 2 S 1 ÞÞDZ: &
We would like to use the above calculations to calculate HF þ of a number of

three-manifolds. Indeed, the manifolds we will consider have positive first Betti
number, and in this case, it is interesting to obtain information of Floer homology as
a module over the ring Z½U#Z L H1 ðY ; ZÞ=Tors (cf. Section 4.2.5 of [20]). We
pause now to recall the construction of this action.
Given a curve g in a Heegaard surface for Y which is disjoint from all the ai -bj ;
the action of the corresponding class ½gAH1 ðY ; ZÞ is induced by the chain map
Ag : CF N ðY Þ-CF N ðY Þ
which decreases absolute degree by one, and is defined by

X X
Ag ½x; i ¼ #
#MðfÞ /g; fS ½y; i nw ðfÞ;
y ffAp2 ðx;yÞjmðfÞ¼1g
where here /g; fS is the intersection number of the restriction of
f : ½0; 1 R-Symg ðSÞ

ARTICLE IN PRESS
P. Ozsvath,
to f1g R (where it maps into Ta ) with the subset of ðg Symg1 ðSÞÞ-Ta

(cf. Section 4.2.5 of [20]). We can obviously extend this to CFK N ðY ; KÞ by defining
X X
Ag ½x; i; j ¼ #
#MðfÞ /g; fS ½y; j nw ðfÞ; j nz ðfÞ;
y ffAp2 ðx;yÞjmðfÞ¼1g
so that it respects the filtration (and indeed, it is not hard to see that the filtered chain
homotopy type of the induced map Ag is an invariant of the knot K; as well). In
particular, this induces also an action, which we denote by A0g on HFK d
HFKðY ; KÞ: In
2 1 2 2 1
both examples ðY ; KÞ ¼ ðS S ; Bð0; 1ÞÞ and ð# ðS S Þ; Bð0; 0ÞÞ calculated
above, for each fixed integer j; HFKd
HFKðY ; K; j Þ is concentrated in a single degree, so it
follows at once that the action A0g is trivial. In these cases, there are higher maps, for
example
d
A1g : HFK
HFKðK; d
j Þ-HFK
HFKðK; j 1Þ;
which count f with nw ðfÞ ¼ 0 and nz ðfÞ ¼ 1 (weighted, once again, by /g; fS).
We can use Proposition 9.2 to obtain information about HF þ for some simple
three-manifolds which fiber over the circle. We begin with the case where the
monodromy is trivial. The key point for these calculations is that if we perform
zero-surgery on the knot #g Bð0; 0ÞC#2g ðS 2 S 1 Þ; then we get the three-manifold
Sg S1 :
We now explain the notation required to state the following result. Fix an integer
k; and let HF þ ðSg S 1 ; kÞ denote the summand of HF þ ðS S 1 Þ corresponding to
the Spinc structure s with
/c1 ðsÞ; ½Sg S ¼ 2k; and /c1 ðsÞ; g S 1 S ¼ 0
for all curves gCSg (of course, HF þ is trivial on Spinc structures which do not satisfy
this latter condition, by the adjunction inequality).
Next, for integers g and d; consider the group
d
M
X ðg; dÞ ¼ L2gi H 1 ðSg Þ#Z ðZ½U=U diþ1 Þ: ð22Þ
i¼0
This can be viewed as a module over the ring Z½U#Z L H1 ðSg Þ; where the action of
gAH1 ðSg Þ is given by
Dg ðo#U j Þ ¼ ðig oÞ#U j þ PDðgÞ4o#U jþ1 ; ð23Þ
where here
ig : Li H 1 ðSg Þ-Li1 H 1 ðSg Þ
denotes the contraction homomorphism (and U acts in the obvious way). In fact, we
can endow X ðg; dÞ with a relative grading so that multiplication by U lowers degree
ARTICLE IN PRESS
P. Ozsvath, 107
by two, and Dg lowers degree by one. It is interesting to note that if Symd ðSÞ denotes
the d-fold symmetric product of S; then
X ðg; dÞDH ðSymd ðSg ÞÞ ð24Þ
as relatively graded groups, according to MacDonald, cf. [14]. Indeed,

H ðSymd ðSg ÞÞ is naturally a module over Z½U#Z L H1 ðSg Þ; where U acts as cap
product with the Poincaré dual of fxg Symd1 ðSÞ; and also H1 ðSg ÞD
H 1 ðSymd ðSg ÞÞ acts by cap product with one-dimensional cohomology. With this
action, the isomorphism of Eq. (24) is given as Z½U#Z L H1 ðSg Þ-modules, again
following [14].
Of course, HF þ ðS S1 Þ is also a module over Z½U#Z L H1 ðS S1 Þ and hence,
under the natural inclusion H1 ðSÞ in H1 ðS S 1 Þ; it can be viewed as a module over
Z½U#L H1 ðSg Þ (in fact, the following proof also shows that the circle factor in
S S 1 annihilates HF þ ðS S1 ; sÞ; provided c1 ðsÞa0).
In the following result, it is very suggestive to compare the results on the module
structure with the quantum cohomology ring calculations of the d-fold symmetric
product of S obtained by Bertram and Thaddeus [3].
Theorem 9.3. Fix an integer ka0: Then, there is an identification of Z-modules
HF þ ðS S 1 ; kÞDX ðg; dÞ; ð25Þ
where
d ¼ g 1 jkj:
(Of course, the group is trivial if jkj4g 1:) Indeed, there is a filtration on HF þ ðS
S 1 ; kÞ as a Z½U#Z L H1 ðSÞ-module with the property that the identification of
Eq. (25) is an isomorphism between the associated graded modules. Indeed, when
3do2g 1; there is an isomorphism as in Eq. (25) on the level of Z½U#Z L H1 ðSÞ-
modules.
Proof. In view of the conjugation symmetry, it suffices to consider the case where
k40:
First, note that it follows from the calculation of HFK d for Bð0; 0Þ given in
d
and Corollary 7.2 that HFK
Proposition 9.2 HFKð# 2g
ðS 2 S 1 Þ; #g Bð0; 0Þ; j Þ is a free Z-
2g
module of rank gþj supported entirely in grading j: More invariantly, we can write
d
HFK
HFKð# 2g
ðS2 S1 Þ; #g Bð0; 0Þ; j ÞDLgþj H 1 ðSg ; ZÞð j Þ ð26Þ
(where as usual the subscript ð j Þ denotes the absolute grading of the corresponding
group).
ARTICLE IN PRESS
P. Ozsvath,
Now we can view CFK N ðY Þ as a Z-filtered subcomplex, by defining the filtration

level of ½x; i; j ¼ i þ j: And hence, we obtain a spectral sequence whose E1 term is
Z½U; U 1 #Z HFK d
HFKðY ; KÞ; converging to HFK N ðY Þ: Moreover, as the above
calculations show, the filtration (on the E1 term) is proportional to the absolute
degree of each group, so the spectral sequence collapses after the E2 stage. The same
remarks apply to the complexes CFK 0; and CFK ;0 ; as well. Indeed, we argue that
the d2 differentials vanish, as well. This follows immediately from the observations
that CFK 0; and CFK ;0 must both calculate HF c ð#2g ðS 2 S 1 ÞÞ; so for dimension
reasons, all higher differentials must vanish. It also follows readily that
H ðCfiX0 or jXkgÞDHfiX0 or jXkg;
where we have adopted the convention that HfiX0 or jXkg denotes the quotient
module of H ðCÞ by the submodule of H ðCfip0 and jpkgÞ: (We use this notation
later in this proof, as well.)
We claim that with respect to the induced identification
HF N ð#2g ðS 2 S 1 Þ; #g Bð0; 0ÞÞDL H 1 ðSg ; ZÞ#Z Z½U; U 1 ;
the action of
Z½U#Z L H1 ð#2g ðS 2 S1 ÞÞDZ½U#Z L H1 ðSÞ
is the action Dg defined above. (Note that the induced action A0g is trivial, A1g is given
by the map Dg above, and all higher actions vanish for dimension reasons.)
To see why this is true, we proceed as follows. We restrict to a ‘‘vertical slice’’
d
HFK
HFKð# 2g c ð#2g ðS 2 S 1 ÞÞ;
ðS 2 S 1 Þ; #g Bð0; 0ÞÞDL H 1 ðSg ; ZÞDHF
where here as usual we view CFK 0; ð#2g ðS 2 S 1 Þ; #g Bð0; 0ÞÞ as a filtration for
c ð#2g ðS2 S 1 ÞÞ; by forgetting about z: In the model for HF
CF c ð#2g ðS 2
1 1
S ÞÞDL H ðSg ; ZÞ; the action by g is given by ig (cf. Section 3 of [18]). This verifies
that if
o#U j AL2gi H 1 ðSg Þ#Z ðZ½U=U diþ1 Þ;
then the component of Ag ðo#U j Þ in L2gi1 H 1 ðSg Þ#Z ðZ½U=U diþ1 Þ—the
‘‘vertical component’’—is given by Dg ðo#U j Þ: For the component in L2giþ1 #Z
d ;0 ð#2g ðS 2 S 1 Þ; #g Bð0; 0ÞÞ as a filtration of
ðZ½U=U diþ1 Þ; we consider now CFK
c ð# ðS S ÞÞ; by forgetting about w: In this model,
CF 2g 2 1
M
c ð#2g ðS 2 S1 ÞÞD
HF Lgþj H 1 ðSg ; ZÞ#U j :
j
ARTICLE IN PRESS
P. Ozsvath, 109
Now, the action of g is forced to be given by the map
ðo#U j Þ/PDðgÞ4o#U jþ1 ;
verifying that the Dg describes the horizontal component of Ag : Since Ag decreases

dimension by one, there are no other components to be verified.
Let Yn denote the three-manifold obtained by performing þn surgery along B:
This manifold is, in fact, the total space of a circle bundle over S with Euler number
n: It follows from the above remarks, together with Theorem 4.4 that for large
þ
enough n; HFred ðYn Þ ¼ 0: Indeed, if we consider the cobordism W obtained by the
natural two-handle addition, and let f1 : HF þ ðYn ; sk Þ-HF þ ð#2g ðS 2 S 1 ÞÞ denote
the map associated to the Spinc structure r with
/c1 ðrÞ; ½SS ¼ 2k þ n
(where S is obtained by completing the Seifert surface for B inside W ), then we see
that the kernel of f1 is given by
Hfio0; jXkgDX ðg; dÞ;
where d ¼ g 1 þ jkj: We claim in fact that the homology of this kernel of f1 is

identified with HF þ ðS 1 S; kÞ:
To this end, recall the integer surgeries long exact sequence, according to which we
have
F G
? ! HF þ ð#2g ðS 1 S2 ÞÞ-HF þ ðS 1 S; kÞ ! HF þ ðYn ; ½kÞ-? ;
where here the map F is the sum of the maps associated to all the Spinc structures
þ
over the cobordism W : Note also that since HFred ð#2g ðS1 S2 ÞÞ ¼ 0; and
þ
HFred ðS S; kÞDHF ðS S; kÞ (since ka0), it follows that HF þ ðS 1
1 þ 1
Sg ÞDKer F : Now, we can write
F ¼ f1 þ f2 ;
where f1 is the homogeneous map described earlier, and f2 has lower order, in the
sense that if x is a homogeneous element of dimension d; then f2 ðxÞ is a sum of
elements of dimension less than f1 ðxÞ: More concretely, according to Eq. (12), if x
supported in a single dimension, then so is f1 ðxÞ; and f2 ðxÞ can be written as a sum of
terms, all of which are supported in dimension at least 2k smaller than the dimension
of f1 ðxÞ:
In this setting, there is an identification A ¼ Ker f1 with B ¼ Ker F ; as Z-modules.
To see this, let
R1 : HfiX0g-HfiX0 or jXkg
ARTICLE IN PRESS
P. Ozsvath,
denote the natural inclusion. Since f2 has lower order than f1 ; it is easy to see that
there is an injective Z-linear map
P : Kerð f1 þ f2 Þ-Kerð f1 Þ;
given by taking the an element xAA to its homogeneous term in the highest degree,
and hence, rkðBÞXrkðAÞ: In the other direction, consider the map
X
N
R¼ ðR1 3f2 Þi : ð27Þ
i¼0
Since R is non-increasing in the filtration induced by the absolute grading, while f2 is

strictly decreasing in this filtration (and, of course, this filtration is bounded below),
it follows that RðxÞ is well-defined for any x: Indeed, it is easily seen that R defines an
injection from A to B; and hence, their ranks agree.
To analyze the Z½U#Z L H1 ðSÞ-module structure, we proceed as follows. Of
course, the filtration mentioned in the statement of the theorem is induced by
identifying HF þ ðS 1 SÞ ¼ Ker F CHF þ ðYn Þ; and noting that the latter group has a
filtration induced by absolute grading. The first assertion about the module structure
is clear, now. For the second, observe that in general neither the maps P nor R are
module homomorphisms. However, the restriction of R1 to X ¼ HfiX0 and jokg
does respect the module structure (as does f2 ; since it is induced by a cobordism). In
fact, since in our case, the absolute degree of Hfi; j g is given by i þ j; it follows that
any element of HF þ ð#2g ðS 2 S 1 ÞÞ with degree og k is contained in X : Since the
maximal dimension of any element of Hfio0 and jXkg is g 2; and f2 decreases
absolute degree by at least 2k; the inequality 2g 143d implies that f2 maps Hfio0
and jXkg into X ; and hence R is a module homomorphism and hence also an
isomorphism. &
More generally, let f be an automorphism of Sg ; and let Mf denote the mapping

torus of f;
½0; 1 Sg
Mf D :
ð0; xÞBð1; fðxÞÞ
When f can be written as a product of disjoint non-separating Dehn twists, one can
use the above techniques to calculate HF þ ðMf ; kÞ in many cases. We content
ourselves here with a calculation in the case where f is a single negative Dehn twist
along a non-separating simple, closed curve gCS:
To this end, we define another differential on X ðg; dÞ: Write S ¼ T#S0 ; where T is
a torus and gCT is a non-separating simple, closed curve in T: Let
Lþ H 1 ðSÞ ¼ ðL0 H 1 ðTÞ#Z L ðS0 ÞÞ"ðL2 H 1 ðTÞ#Z L ðS0 ÞÞ;
L H 1 ðSÞ ¼ L1 H 1 ðTÞ#Z L ðS0 Þ;

ARTICLE IN PRESS
P. Ozsvath, 111
so that L H 1 ðSÞ ¼ Lþ H 1 ðSÞ"L H 1 ðSÞ: Let X ðg; dÞ ¼ Xþ ðg; dÞ"X ðg; dÞ be the
corresponding splitting. We define D0g so that
D0g jXþ ðg; dÞ 0 and D0g jX ðg; dÞ Dg jX ðg; dÞ: ð28Þ
Theorem 9.4. Let f denote a negative Dehn twist along a non-separating simple, closed
curve gCS; and let Mf denote its mapping torus. For each integer k; let HF þ ðMf ; kÞ
denote HF þ of Mf evaluated on the Spinc structure s characterized by the properties
that /c1 ðsÞ; ½F S ¼ 2k and c1 ðsÞ vanishes on all homology classes represented by tori.
Suppose that 0ak and also suppose that
3do2g 1; ð29Þ
where d ¼ g 1 jkj: Then, there is an isomorphism of Z½U#Z L H1 ðSÞ-modules
HF þ ðMf ; kÞDH ðX ðg; dÞ; D0g Þ; ð30Þ
where D0g is the differential defined by Eq. (28). When, Eq. (29) does not hold, we still
obtain an isomorphism of the form stated in Eq. (30), as Z-modules.
Proof. If f is a negative Dehn twist along g; then we can realize Mf as zero-surgery

along the connected sum Bð0; 1Þ#ð#g1 Bð0; 0ÞÞ: We adopt the notation from the
proof of Theorem 9.3 before. Note that HFK d of Bð0; 0Þ and Bð0; 1Þ differ by only a
shift in dimension, so it is still the case that the knot filtration is proportional to the
absolute degree, so the spectral sequence collapses after the E2 stage. There is one
crucial difference now: it is no longer the case that the d2 differential is trivial.
d 2 S 1 ; Bð0; 1ÞÞ is the homology of the graded object associated
Indeed, since HFK
HFKðS
c ðS 2 S 1 Þ; and
to a filtration of CF
c ðS2 S1 ÞDZ
HF 1 "Z 1 ;
ð2Þ ð 2Þ
it follows that the d2 differential is non-trivial; indeed, the map
d2 : Z 21 DHFK
d
HFKðS 2 d 2 S1 ; Bð0; 1Þ; 1ÞDZð3=2Þ
S 1 ; Bð0; 1Þ; 0Þ-HFK
HFKðS
ð2Þ
surjects. In the notation from the theorem, we can write this as the map
H 1 ðT 2 Þ-H 0 ðT 2 Þ given by evaluation against ½g (i.e. this is the ‘‘horizontal
component’’ of D0g :
We now proceed as before, now letting H denote the homology of C with respect
to the d1 differential. Thus, the group Hfio0; jXkgDX ðg; dÞ; which is the kernel of
f1 ; is now endowed with the d2 differential induced by D0g : (The fact that this
differential is given by D0g follows the proof of Theorem 9.3, in the calculation of the
module Z½U#Z L H1 ðSg Þ-module structure of HF þ ðS S 1 ; kÞ: In particular, it
ARTICLE IN PRESS
P. Ozsvath,
follows from dimension reasons, followed by a consideration of the filtration

c ðS2 S1 Þ:)
CFK ;0 ðS 2 S 1 ; Bð0; 1ÞÞ of CF
We once again consider the integral surgeries long exact sequence, and let Yn
denote the three-manifold obtained by large n surgery on the link Bð1; 0Þ#g1 Bð0; 0Þ;
and identify HF þ ðMf ; kÞDH ðKerð f1 þ f2 Þ; d2 Þ: Proceeding as before, we construct
the identification of Kerð f1 þ f2 Þ with Kerð f1 Þ; by considering the map R defined as
in Eq. (27). The inequality 3do2g 1 now ensures that R is a chain map, using the
d2 differential (and hence also an isomorphism of chain complexes).
For the final remark, when Eq. (29) is violated, we still observe that the map
induced by f1 is surjective on homology. We then appeal to the same argument as in
Corollary 4.5. &
It follows from the above result that when g42; HF þ ðMf ; g 2Þ is isomorphic to
the relative singular cohomology H ðS; gÞ; while HF þ ðMf1 ; g 2ÞDH ðS gÞ:
This should be compared with a result of Seidel [27], see also [11].
10. Links
Throughout most of the paper, we assumed that our oriented knot K is connected.
In view of Proposition 2.1, passing to the case of oriented links is quite
straightforward. Our aim here is to highlight some issues in this generalization,
with the aim of showing how the properties of the link invariant stated in the
introduction follow from the other results proved above.
As a first point, recall that in some applications, especially to exact sequences
(cf. Section 8), we found it convenient to fix a Seifert surface F for the knot K:
Correspondingly, when considering a link L; we also fix a Seifert surface F for L:
Lemma 10.1. Fix a Seifert surface F for L in Y, and let F 0 be any extension of F to the
zero-surgery of kðY Þ along kðLÞ: Then, the group
M
d
HFK
HFKðY ; L; mÞ ¼ d
HFK
HFKðY ; L; tÞ
ftASpinc ðkðY Þ;kðLÞÞj/c1 ðtÞ;½F 0 S¼2mg
depends on F 0 only through the (relative homology class of the) induced Seifert surface
F in Y (i.e. it is independent of the extension F 0 ).
Proof. The indeterminacy of F 0 comes from the attached one-handles—more

precisely writing kðY Þ ¼ Y ##n1 ðS 2 S1 Þ; the indeterminacy amounts to adding
homology classes represented by spheres coming from S 2 S 1 ; which in turn can be
represented by embedded tori in the complement of kðKÞ inside kðY Þ: However, it is
straightforward to see that if the first Chern class of tASpinc ðkðY ÞÞ evaluates non-
trivially on such a homology class, then we can find an admissible Heegaard diagram
for the knot kðLÞ in kðY Þ with the property that no intersection point represents t:
ARTICLE IN PRESS
P. Ozsvath, 113
This follows from a straightforward adaptation of the proof of the adjunction

inequality for HF þ (Theorem 7 of [18], see also Theorem 5.1 above) that
d
HFKðkðY
HFK Þ; kðLÞ; tÞ ¼ 0: &
Now, consider the case where Y is an oriented three-manifold, equipped with an

oriented link Lþ ; and let gCY be an unknot which spans a disk D which meets L in
two algebraically cancelling transverse points. Let L denote the new link induced
from L after introducing a full twist, as pictured in Fig. 1. In comparing with this
picture, the disk D can be thought of as a horizontal disk. Let L0 denote the new link
obtained from L by resolving the link, so that it no longer meets D:
Recall that in Section 8, we chose a Seifert surface F for L; and use it to define the
integral splitting of HFKðY ; LÞ as in Lemma 10.1.
Theorem 10.2. Let Lþ ; L0 ; and L be the links related by a skein move as above. Then
we have a long exact sequence of the form:
d
?-HFK
HFKðY d
; L Þ-HFK
HFKðY d
; L0 Þ-HFK
HFKðY ; Lþ Þ-? ; ð31Þ
while if they belong to different components of L; we have a long exact sequence of the
form
d
?-HFK
HFKðY d
; L Þ-HFK
HFKðY 0 d
; L00 Þ-HFK
HFKðY ; Lþ Þ-? ; ð32Þ
where here ðY 0 ; L00 Þ denotes the link obtained by connected sum of ðY ; L0 Þ with the
‘‘Borromean knot’’ ð#2 ðS 2 S1 Þ; Bð0; 0ÞÞ from Fig. 16. Moreover, the maps in the
exact sequence respect the Z-splittings of the homology groups induced from a Seifert
surface F for Lþ in Y (and corresponding Seifert surface for L and L0 ).
Proof. This result is a special case of Theorem 8.2, after some remarks.
First, observe that under the diffeomorphism Y1 ðgÞDY ; the image of the link
obtained from viewing Lþ as a link in Y gCY1 ðgÞ is L :
Suppose next that both strands of L meeting D belong to the same component of
L: In this case, we can trade the zero-framed two-handle attached along g for a one-
handle, without changing the underlying three-manifold Y0 ðgÞDY #ðS2 S 1 Þ: In
the case where L had two components, in fact, we see that the knot induced from K
then coincides with the knot kðL0 Þ: It is easy to see that in this case, Theorem 8.2
translates into long exact sequence (31).
Suppose next that two strands of L meeting in D belong to different components
of L: In fact, for simplicity, we assume that L has two components L1 and L2
(indeed, the case where L consists of more than two components can be reduced to
this case, after attaching sufficiently many one-handles.) In this case, by definition,
the link invariant of L inside Y0 ðgÞ agrees with a knot invariant of the knot induced
from kðLÞ inside kðY Þ0 ðgÞ: Recall that kðY Þ is the three-manifold obtained from Y
by attaching a one-handle. We can replace the one-handle in kðY Þ by a zero-framed
ARTICLE IN PRESS
P. Ozsvath,
,
γ ,
γ
L1 L2
γ
γ γ
Fig. 17. Trading link crossings. The identification of ðkðY Þ0 ðgÞ; kðLÞÞ with ðY ; L0 Þ#ðS2 S1 ; Bð0; 0ÞÞ:
two-handle g0 ; to obtain Y 0 ¼ Y #2 ðS 2 S1 Þ with a new knot K 0 : It is easy to see
ðY 0 ; K 0 Þ ¼ ðY ; KÞ#ð#2 ðS2 S 1 Þ; Bð0; 0ÞÞ
as claimed, with the knots g and g0 playing the role of the two zero-framed circles in
the Borromean knot, as illustrated in Fig. 17. Strictly speaking, to see that the
operation is local on the knot K; it is useful to trade the two-handle specified by g for
a three-handle (with the curve g0 running through it), and slide the one-handle freely
along K:
With these remarks in hand, we see long exact sequence (32) as a special case of
Theorem 8.2. &
In view of the calculation of HFK d

HFKð# 2
ðS 2 S1 Þ; Bð0; 0ÞÞ in Proposition 9.2,
together with the Künneth principle for connected sums from Corollary 7.2, we see
that the versions of the skein exact sequence given in the introduction coincide with
those stated in Theorem 10.2.
d stated in Eq. (1) is an easy application
The Euler characteristic calculation of HFK
of the skein exact sequence (though an alternate proof could also be given in the
spirit of Theorem 5.2 of [18]). More precisely, note that HFK d 3 ; LÞ inherits an
HFKðS
absolute Z=2Z grading from HF c ð# ðS S ÞÞ (where here n is the number of
n1 2 1
components of L). As in [22], we see that this grading is preserved by both the map
from HFK d 3 ; L Þ to HFK
HFKðS d 3 ; L0 Þ and the map from HFK
HFKðS d 3 ; Lþ Þ to
HFKðS
d 3 ; L Þ; while it is reversed by the remaining map. It follows from the two
HFKðS
HFK
skein exact sequences then that if we write
X
d 3 ; LÞÞ ¼
wðHFK
HFKðS wðHFKd 3 ; L; iÞÞ T i ;
HFKðS
iAZ
then
d 3 ; L ÞÞ wðHFK
wðHFK
HFKðS d 3 ; L0 ÞÞ wðHFK
HFKðS d 3 ; Lþ ÞÞ ¼ 0
HFKðS
if L0 has more components than Lþ ; otherwise,

d 3 ; L ÞÞ ðT 1=2 T 1=2 Þ2 wðHFK
wðHFK
HFKðS d 3 ; L0 ÞÞ wðHFK
HFKðS d 3 ; Lþ ÞÞ ¼ 0:
HFKðS
ARTICLE IN PRESS
P. Ozsvath, 115
Since wðHFKd 3 ; uÞÞ ¼ 1 (when u is the unknot), Eq. (1) now follows from the usual
HFKðS
skein relation characterization of the Alexander–Conway polynomial. The fact that
this absolute Z=2Z grading is related to the absolute Q-grading in the classical case
discussed in the introduction follows easily from [22].
Eq. (2) follows immediately from Proposition 3.7, Eq. (3) follows from Proposi-
tion 3.9, Eq. (3) follows from Proposition 3.10 (noting that when we pass from K to
K; we also must reverse the orientation of the Seifert surface, so
/c1 ðsÞ; ½F̂ S ¼ /c1 ðJsÞ; ½F̂ S), and Eq. (5) follows from Corollary 7.2.
For the behavior of HFK d under disjoint union, note that by definition of the link
invariant,
d 3 ; L1 ,L2 Þ ¼ HFK
HFK
HFKðS d 2 S 1 ; L1 #L2 Þ
HFKðS
(where here we think of L1 #L2 as supported in a ball in S2 S1 ). Now, Eq. (6) from
a straightforward adaptation of the proof that HF c ðS3 #ðS 2 S 1 ÞÞD
c ðY Þ#H ðS Þ given in [18].
HF 1
Acknowledgments
The authors warmly thank Paolo Lisca, Tomasz Mrowka, Jacob Rasmussen, and
András Stipsicz for some interesting discussions. We would also like to thank the
referee for some helpful comments.
References
[1] S. Akbulut, J. McCarthy, Casson’s invariant for oriented homology 3-spheres—an exposition, in:
Annals of Mathematics Studies, Vol. 36, Princeton University Press, Princeton, NJ, 1990.
[2] D. Bar-Natan, On Khovanov’s categorification of the Jones polynomial, Algebraic Geom. Topol. 2
(2002) 337–370.
[3] A. Bertram, M. Thaddeus, On the quantum cohomology of a symmetric product of an algebraic
curve, Duke Math. J. 108 (2) (2001) 329–362.
[4] P. Braam, S.K. Donaldson, Floer’s work on instanton homology, knots, and surgery, in: H. Hofer,
C.H. Taubes, A. Weinstein, E. Zehnder (Eds.), The Floer Memorial, Progress in Mathematics,
Vol. 133, Birkhäuser, Basel, 1995, pp. 195–256.
[5] Y.M. Eliashberg, W.P. Thurston, Confoliations, in: University Lecture Series, Vol. 13, American
Mathematical Society, Providence RI, 1998.
[6] R. Fintushel, R.J. Stern, Knots, links, and 4-manifolds, Invent. Math. 134 (2) (1998) 363–400.
[7] A. Floer, Instanton homology, surgery, and knots, in: Geometry of Low-Dimensional Manifolds, 1
(Durham, 1989), London Mathematical Society, Lecture Note Series, Vol. 150, 1990, Cambridge
University Press, Cambridge, pp. 97–114.
[8] D. Gabai, Foliations and the topology of 3-manifolds, J. Differential Geom. 18 (3) (1983) 445–503.
[9] R.E. Gompf, A.I. Stipsicz, 4-manifolds and Kirby calculus, in: Graduate Studies in Mathematics,
Vol. 20, American Mathematical Society, Privdence, RI, 1999.
[10] J. Hoste, Sewn-up r-link exteriors, Pacific J. Math. 112 (2) (1984) 347–382.
ARTICLE IN PRESS
P. Ozsvath,
[11] M. Hutchings, M. Sullivan, The periodic Floer homology of a Dehn twist, 2002. http://
math.berkeley.edu/hutching/pub/index.htm.
[12] M. Khovanov, A categorification of the Jones polynomial, Duke Math. J. 101 (2000) 359–426.
[13] P.B. Kronheimer, T.S. Mrowka, Scalar curvature and the Thurston norm, Math. Res. Lett. 4 (6)
(1997) 931–937.
[14] I.G. MacDonald, Symmetric products of an algebraic curve, Topology 1 (1962) 319–343.
[15] G. Meng, C.H. Taubes, SW=Milnor torsion, Math. Res. Lett. 3 (1996) 661–674.
[16] J.W. Morgan, Z. Szabó, C.H. Taubes, A product formula for Seiberg–Witten invariants and the
generalized Thom conjecture, J. Differential Geom. 44 (1996) 706–788.
[17] V. Muñoz, B.-L. Wang, Seiberg–Witten–Floer homology of a surface times a circle, math.DG/
9905050.
[18] P.S. Ozsváth, Z. Szabó, Holomorphic disks and three-manifold invariants: properties and
applications, math.SG/0105202, Ann. of Math., to appear.
[19] P.S. Ozsváth, Z. Szabó, Holomorphic triangles and invariants for smooth four-manifolds, math.SG/
0110169.
[20] P.S. Ozsváth, Z. Szabó, Holomorphic disks and topological invariants for closed three-manifolds,
math.SG/0101206, Ann. of Math., 2001, to appear.
[21] P.S. Ozsváth, Z. Szabó, Heegaard Floer homologies and contact structures, math.SG/0210127, 2002.
[22] P.S. Ozsváth, Z. Szabó, Absolutely graded Floer homologies and intersection forms for four-
manifolds with boundary, Adv. Math. 173 (2) (2003) 179–261.
[23] P.S. Ozsváth, Z. Szabó, Heegaard Floer homology and alternating knots, Geom. Topol. 7 (2003)
225–254.
[24] P.S. Ozsváth, Z. Szabó, Knot Floer homology, genus bounds, and mutation, math.GT/0303225,
2003.
[25] J.A. Rasmussen, Floer homologies of surgeries on two-bridge knots, Algebr. Geom. Topol. 2 (2002)
757–789.
[26] J. Rasmussen, Floer homology and knot complements, Ph.D. Thesis, Harvard University, 2003.
[27] P. Seidel, The symplectic Floer homology of a Dehn twist, Math. Res. Lett. 3 (6) (1996) 829–834.

Holomorphic Disks and Knot Invariants: Article in Press

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Holomorphic Disks and Knot Invariants: Article in Press

Uploaded by

Copyright:

Available Formats

ARTICLE IN PRESS

Advances in Mathematics 186 (2004) 58–116

Holomorphic disks and knot invariants

Received 6 January 2003; accepted 20 May 2003

Communicated by Tomasz Mrowka

We deﬁne a Floer-homology invariant for knots in an oriented three-manifold, closely

1.1. Classical links

where n denotes the number of components of L (it is interesting to compare this

Another symmetry these invariants enjoy is the following conjugation symmetry:

HFK d d2i ðL; i; QÞ;

HFK d d ðL; i; QÞ:

Fig. 1. Skein moves at a double-point.

where W is a two-dimensional graded vector space splitting into two one-

where all the maps above respect the splitting of HFK d

strands belong to different components, we obtain the following:

where V denotes the four-dimensional vector space

V ¼ V1 "V0 "V1 ;

Conjecture 1.1. If KCS 3 is a knot with genus g; then

homology, which is established in Section 4. Next, we appeal to the conjectured

1.2. General case

In the above discussion, we restricted attention to a rather special case of our

1.3. Further remarks

The fact that a knot in Y induces a ﬁltration on CF N ðY Þ has been discovered

the Floer homologies of three-manifolds obtained as surgeries on two-bridge knots.

2.1. From knots to links

kðY ; fpi ; qi gÞ-kðY ; fp0i ; q0i gÞ

2.2. Heegaard diagrams and knots

A knot ðY ; KÞ has a Heegaard diagram

where here a is an unordered g-tuple of pairwise disjoint attaching circles a ¼

Deﬁnition 2.3. A marked Heegaard diagram for a knot ðY ; KÞ is a quintuple

Deﬁnition 2.4. A doubly pointed Heegaard diagram for a three-manifold Y is a tuple

We extract a doubly pointed Heegaard diagram for Y from a Heegaard diagram

Following notation from [20], if x; yATa -Tb ; we consider spaces of homotopy

nv ðfÞ ¼ #f1 ðfvg  Symg1 ðSÞÞ:

2.3. Intersection points and Spinc structures

When Y is equipped with an oriented, null-homologous knot K; and corresponding

sm : Ta -Tb -Spinc ðY0 ðKÞÞ;

The map from Spinc ðY ; KÞ-Spinc ðY Þ is obtained by ﬁrst restricting t to Y K; and

course, the points w and z lie in the same component of S a1 ? ag g1

/c1 ðs0w ðyÞÞ; ½F̂ S ¼ wðPÞ þ 2n% y ðPÞ; ð9Þ

provided that P is the periodic domain representing F̂:

sm ðxÞ sm ðyÞ ¼ ðnz ðfÞ nw ðfÞÞ  PD½m;

Proof. We begin with some preliminary observations. Consider the natural

We claim that more generally if cAp2 ðx; Y; yÞ is a Whitney triangle (where,

s0w ðyÞ ¼ sm ðxÞ þ ðnw ðcÞ nz ðcÞÞ  ½PDðmÞ: ð10Þ

This completes the veriﬁcation of Eq. (10).

2.4. Filtered complexes

Fix a partially ordered set S: An S-ﬁltered group is a free Abelian group C

We write elements of C as sums

are elements of S-ﬁltered groups ðC; F; SÞ and ðC 0 ; F0 ; S0 Þ; respectively, then we

max FðsÞp min F0 ðsÞ:

A morphism of S-ﬁltered groups

is a group homomorphism with the property that

for all aAC:

½x; iAðTa -Tb Þ  Z

3. The knot ﬁltration: deﬁnitions and basic properties

3.1. Definition of the knot filtration

Let ðS; a; b; w; zÞ be a doubly pointed Heegaard diagram, and J be an allowed one-

U  ½x; i; j ¼ ½x; i 1; j 1:

Of course, the ﬁltration on the chain complex

F : ðTa -Tb Þ  Z2 -Z2

F½x; i; j ¼ ði; jÞ;

nw ðfÞ ¼ i c and nz ðfÞ ¼ j m:

nv ðfÞ ¼ #f1 ðfvg Symg1 ðSÞÞ:

/c1 ðs0w ðyÞÞ; ½F̂ S ¼ wðPÞ þ 2n% y ðPÞ; ð9Þ

sm ðxÞ sm ðyÞ ¼ ðnz ðfÞ nw ðfÞÞ PD½m;

s0w ðyÞ ¼ sm ðxÞ þ ðnw ðcÞ nz ðcÞÞ ½PDðmÞ: ð10Þ

½x; iAðTa -Tb Þ Z

U ½x; i; j ¼ ½x; i 1; j 1:

F : ðTa -Tb Þ Z2 -Z2

F½x; i; j ¼ ði; jÞ;

sm ðxÞ þ ði jÞPD½m ¼ t; ð11Þ

P1 ½x; i; j ¼ ½x; i;

HFdp0 ðm,b0 ; m0 ,b00 ; w; zÞDZ½U#Z L H1 ðT g Þ:

We consider the following two Z"Z-ﬁltered, Z½U-subcomplexes of

/c1 ðrÞ; ½SS þ p 2m ðmod 2pÞ:

Similarly, if we let CFK 0 ðY ; K; mÞCCFK þ ðY ; K; mÞ be the subset generated by ½x; i; j

F0 : CFK N ðY ; K; mÞ-CF N ðYp ðKÞ; ½mÞ

CFK N ðY ; K; mÞD CF N ðYp ðKÞ; ½mÞ;

CFK þ ðY ; K; mÞD CF þ ðYp ðKÞ; ½mÞ;

c ðYp ðKÞ; ½mÞ:

/c1 ðrÞ; ½SS ¼ 2m þ p:

½x; i; j/½x; i; j þ 12/c1 ðtÞ; ½F̂ S:

/c1 ðrÞ; ½SS ¼ 2m þ p;

Similarly, if we let b CFK 0 ðY ; KÞCCFK þ ðY ; KÞ be the subset generated by ½x; i; j with

H ðCfiX0 or jXdgÞDHF þ ðSn3 ðKÞ; ½d 1Þ;

/c1 ðrÞ; ½F̂ S ¼ 2d 2 þ n:

Note that the map c induced by C on homology is surjective, since it is a Z½U

j/c1 ðsÞ; ½F̂ Sjp2gðF Þ:

@½x1 ; i; i þ 1 ¼ ½x0 ; i; i;

@½x1 ; i þ 1; i ¼ ½x0 ; i; i:

Moreover, the absolute grading of the generator ½x0 ; 0; 0 is 1:

½x1 ; i; i þ 1 and ½x1 ; i þ 1; i: