Você está na página 1de 34

A KONTSEVICH INVARIANT FOR COLOURED KNOTS

DANIEL MOSKOVICH Abstract. Using the non-commutative surgery presentation of a knot in [29], we construct a non-commutative version of the rational Kontsevich invariant for pcoloured knots as a D2p equivariant invariant of their irregular dihedral covering spaces. As part of the construction we prove a non-commutative analogue of the Kirby theorem for untying links of pcoloured knots.

1. Introduction A knot is a smooth embedding K : S 1 S 3 . A coloured knot is a pair K, consisting of a knot and an epimorphism : 1 S 3 K G onto a xed discrete group G called a colouring. We may obtain invariants of coloured knots as G equivariant invariants of the covering space determined by of S 3 branched over K. When G Z, is uniquely determined. A fruitful approach in this case has been to rst replace the knot K with a surgery presentation of itself a unitframed link L made up of unknotted components in ker in S 3 O (the complement of the unknot) such that surgery by L on S 3 , O recovers S 3 , K . We may recover invariants of K from invariants of L by surgery formulae. The link L lifts to a link L in the innite cyclic covering of the complement of the unknot (which is S 3 ), such that surgery on L recovers C K the innite cyclic covering of S 3 K. The is called a surgery presentation of C K . We may recover Zequivariant link L invariants of C K from invariants of L by surgery formulae. Garoufalidis and Kricker used these ideas in a series of papers [14, 15, 16, 19, 20] in which they construct a Zequivariant invariant of the C K which they call a rational Kontsevich invariant of K. The loop expansion property allows us to view the rational Kontsevich invariant as a Zequivariant LMO invariant [36], in other words as a universal LMO invariant for all cyclic coverings of the knot complement. Its 1loop part may be expressed as a rational function depending on the Alexander polynomial of the knot as conjectured by Rozansky [34]. The above approach can be generalized to any class of coloured knots whose surgery presentations are known. In [29] we discovered a surgery presentation for the next simplest class of coloured knots, those with G D2p with p 3, 5 the dihedral groups of order 6 and 10. Using this result, in the present paper we construct a D2p equivariant invariant of a dihedral covering space which gives a non-commutative analogue of the rational Kontsevich invariant. Coloured knots K, with : 1 S 3 K D2p are called pcoloured knots and is called a pcolouring of K. The following are commutative diagrams illustrating the process of translating between invariants of coloured knots and Gequivariant invariants of covering spaces in the case G Z and in the case G D2p with p 3, 5:
1

DANIEL MOSKOVICH

C (1.1)

surg K  L

S3

MK,

surg  L

S3

In the right diagram K, is a pcoloured knot with coloured untying invariant n (Section 2) and OK, is a connect-sum of n copies of p1 , 1 (see Figure 2.1). By abuse of notation we also denote by 1 the pcolouring of the connect-sum. The space MK, is the irregular dihedral cover of S 3 branched over K associated to , and S 3 is the irregular branched dihedral cover of S 3 branched over OK, associated to 1 . The associated degree p covering maps are and 1 correspondingly. We begin by replacing a pcoloured knot K, with a 1framed link L S 3 OK, whose components are unknotted and represent homotopy classes in ker 1 . Such a link is called an untying link for K, since surgery by L on S 3 , OK, , 1 recovers S 3 , K, . In order to obtain invariants of K, as invariants of L, we must know how to relate two untying links for the same pcoloured knots. In Section 2.2 we explain that the non-commutative surgery presentation of a knot is unique up to a certain set of moves analogous to the Kirby moves of the abelian case (this is proved in Section 6.1). The next stage is to lift the surgery presentation to S 3 the irregular branched dihedral cover of S 3 , OK, , 1 . In Section 3 we recover L from L. Taking a diagram-valued Kontsevich integral of L in Section 4 and diagrammatically integrating it in Section 5, we obtain a D2p -equivariant invariant of MK, which may be seen as a pcoloured analogue of the rational Kontsevich invariant (that this is indeed a 3manifold invariant is proved in Section 6.2). In the case that MK, is an integral homology 3sphere the denition of the above invariant simplies to yield a pcoloured diagrammatic counterpart to Gaussian integration (Section 5.2). During our construction we use many spaces of links and diagrams, the main ones of which are listed in Table 1 for easy reference. 1.1. Notations and Conventions. We work over the rational numbers Q. We adopt the convention N 0, 1, 2, . . .. D2p : s, t sp t2 1, tst sp1 . If X is a set acted on by a group G, and is the equivalence relation x gx for all g G, the set of orbits of G on X is denoted X . If M is a move on X, i.e. it species a subset Y of X X such that x1 x2 i x1 and x2 are related by a series of M moves i x1 x2 Y , then the set of equivalence classes of X modulo the equivalence relation is again denoted X . The identity n, ntangle is denoted In . The tensor product and composition of tangle diagrams are given as T1 T2 : T1 T2 T1 T2 : T1 T2

S 3  L surg

K, S 3  L surg

S 3 OK,

A KONTSEVICH INVARIANT FOR COLOURED KNOTS

Concept

Section 0

Example 4 3 0

pColoured Knot

2.1 1 2 1

1
2 Untying Link for a pColoured Knot 2.1 2 1 0

Local Picture for an Untying Link for a pColoured Knot

1
2.3

pColoured (Pointed) NonAssociative Tangle Diagram

, , ,

1 1 0

4.1

, y
2

,
0

pColoured Winding Diagram

ss 4.2

2 2
1

1
ts s x

y 6 pColoured Jacobi Diagram 4.2

1 22

s2 ts s x

pColoured Soft-Skeleton Jacobi Diagram

1
4.2 t s

pColoured Trivalent Jacobi Diagram

5.1

1 s2

ts s

Table 1. Some spaces used in this paper.

DANIEL MOSKOVICH

k : 1, . . . , k . We denote the set of n n matrices over a ring R by Matnn R. The set of n n diagonal matrices over R is denoted Diagnn R. The n n identity matrix is denoted Inn . For square matrices A and B we set 0 A B : A B . 0 As in [30], we write multi-indices as k k1 , k2 , , k .

Acknowledgements. The author would like to sincerely thank Tomotada Ohtsuki for his support and encouragement throughout the writing of this paper. Thanks also to Kazuo Habiro for useful discussions and ideas, to Andrew Kricker for advice and support, and to Gwnal Massuyeau, Dror Bar-Natan, Jim Davis, and Teruhisa e e Kadokami for useful discussions. Finally, special thanks to my wonderful wife Yona Chava for always being there for me. 2. The Untying Link In the following section we express a pcoloured knot as a 1framed link in a handlebody with some information attached satisfying certain conditions and modulo certain Kirby-like moves. This section then is concerned with the bottom arrow of the right commutative diagram in Equation 1.1:

K, S 3  L surg

S 3 K0

2.1. Untying pColoured Knots by Dehn Surgery. Let G be a xed discrete group, and let K, denote a coloured knot. The following denition is useful: Denition 2.1. Let K0 denote the set of equivalence classes of coloured knots K, modulo 1framed surgery by links whose components are unknotted and which represents a homotopy class in ker . Let K, be a given coloured knot and let K0 , 0 be a chosen representative of its equivalence class in K0 . An untying link L for K, is a 1framed link in S 3 K0 whose components are unknotted and which represents a homotopy class in ker such that surgery by L recovers S 3 , K, .

From now on let G D2p with p 3, 5 except when otherwise stated. The coloured knot K, with : 1 S 3 K D2p is called a pcoloured knot, and K is said to be pcolourable with pcolouring . This terminology reects the fact that K, is uniquely characterized by a knot diagram of K whose arcs are coloured by elements of Zp (the cyclic group of order p) such that at least two colours are used, and that at each crossing half the sum of the labels of the under-crossing arcs equals the label of the over-crossing arc modulo p. Figure 2.1 presents two colourings of the 74 knot for dierent values of p. For more about pcolourability we refer the reader to Foxs original paper [13]. In the G D2p case which we are discussing we call chosen representatives of equivalence classes in K0 are called untied pcoloured knots (the author has resisted the temptation to call them pcoloured unknots). By the results of [29], when p 3, 5 (and conjecturally for any odd prime p) they may be chosen to be connect sums of n copies of a p1 , 1 knots as given in Figure 2.1, with n 1, 2, . . . , p. The untied pcoloured knot in the same class in K0 as a pcoloured knot K, is called the untying of K, and is denoted OK, .

A KONTSEVICH INVARIANT FOR COLOURED KNOTS

0 2

0 3

0 2 1 3 1 2 Figure 1. The coloured 74 knot. Figure 2. The coloured 74 knot.

1 5

0 1 2

Figure 3. The knot 31 , 1 .

Figure 4. The knot 51 , 1 .

By [29, Section 6], the number of p1 , 1 knots appearing as connect summands in an untied pcoloured knot which is the untying of a pcoloured knot K, is determined by the coloured untying invariant of K, which is dened as follows: Here C2 denotes the 2fold covering of S branched over K, denotes the Bockstein homomorphism on cohomology, and a denotes the cohomology class in H 1 C2 ; ZpZ determined by .
3

cuK, :

a H 3 C2 ; ZpZ

ZpZ

2.2. A Kirbylike Theorem. Since in the abelian case any crossing change can be realized by 1framed surgery on an unknot representing an element of ker Ab, there are (innitely) many ways of untying any knot by such surgeries. However [16, Theorem 1] tells us that all untying links of K are related by a series of Kirby moves: Kirby I.

L
Kirby II. L1 L2

(Stabilization)

L1

L2

(Handle Sliding)

DANIEL MOSKOVICH

Here denotes disjoint union. For G D2p one further move is required: Kirby IIIp .

Lnew

(Dihedral Circumcision)

where Lnew is any 1-framed component representing an element in ker . Kirby I, Kirby II, and Kirby IIIp together are called the pcoloured Kirby moves. We have the following Kirby-like theorem: Theorem 1. Two links L and L are both untying links for a pcoloured knot K, if and only if L is an untying link for K, and L and L are related by the pcoloured Kirby moves. This is proved in Section 6.1. 2.3. The Local Picture. Before we move onto the next arrow in Equation 1.1 and lift to a covering space, we rst present a local (easier to lift) picture of the data given above. Let L S 3 OK, be an untying link for a pcoloured knot K, . Then we can isotopy L so that it lies inside a ball B S 3 such that B OK, is a genus 1 cuK, solid handlebody in S 3 OK, which lifts to a single deck of any branched covering space of itself (in other words, the holes in the handlebody have no crossings). The example p 3 and cuK, 3 is given below, where we can isotopy any untying link of K, into the dotted box:

We have

B OK,

where N denotes a regular neighbourhood of OK, in S 3 . We denote the holes the components of N B, by H1 , H2 , . . . , H1cuK, . Each Hi has a core which is an arc in OK, . We dene the colouring of Hi to be the colouring of its core, and require (by correctly choosing ) that the colouring of H1 is g1 ZpZ and the colouring of Hj is g2 ZpZ for all j 2. In addition we choose the numbering of the holes so that the cores of H1 and of H2 are in the same prime-summand of OK, . Continuing with the example n 2, we now have the following local picture for L, where the coloured line indicates linkage with possibly more than one arc:

A KONTSEVICH INVARIANT FOR COLOURED KNOTS

g1

g2

(2.1)

This presentation is not unique to see this, contemplate the following picture: g2

g1

(2.2)

and slide the line-segment drawn linking to the hole drawn in the upper-right corner of the picture up and out of the picture and around the knot OK, until it comes back up through the bottom of the picture. We obtain a new link L in which is ambient isotopic in S 3 OK, but not in . Examining the situation more closely, we see that because each prime component of OK, is a twist-knot with p twists, L is related to L as follows (for p 3): g1 g2

(2.3)

L
If p 5 then the altered segment goes one more time around the holes. The move taking L to L (and its inverse) are called moves. It will be useful in what follows to add slicing information to our picture. We dene a slice to be an incompressible disc in whose boundary is contained in B Hi for some i 1, 2, . . . , 1 cuK, . The colour of the slice is dened to be the colour of the hole to which it is adjacent. A maximal non-intersecting family of slices added to our data gives us what we call the sliced local picture. Summarizing, corresponding to the rst three rows of Table 1 we have at our disposal three equivalent presentations for a pcoloured knot:

DANIEL MOSKOVICH

(1) As a pair K, . (2) As an untying link in the complement of an untied pcoloured knot OK, , modulo the pcoloured Kirby moves. (3) As the untying link L pushed into a genus 1 cuK, handlebody, modulo the pcoloured Kirby moves, sliced or not sliced. 3. Lifting the Link Having obtained a 1framed link in the complement of an untied pcoloured knot in place of our original pcoloured knot K, , we now proceed with our plan and lift it to S 3 the irregular branched dihedral cover of S 3 , OK, , 1 , following the right arrow in the right commutative diagram in 1.1: S3 L
1

S 3 OK, L

We then calculate the D2p -covariant linking matrix of L in terms of the linking matrix of L. In this section we take components of L to be oriented. 3.1. The Irregular Branched Dihedral Covering. There are two branched covering spaces of S 3 naturally associated with a pcoloured knot K, . The rst is the regular branched dihedral covering MK, , dened to be the covering space 3 of S branched over K associated to ker. The second is the irregular branched dihedral covering MK, , which is dened to be the covering space of S 3 branched over K associated to 1 S2 where S2 D2p . Note that since the trivial group is an order 2 subgroup of S2 , we have that MK, is a 2fold cover of MK, . Thus as in [7], associated to the graph of subgroups of D2p we have the graph of covering spaces

e
h (3.1) Zp S2 MK, f

MK, j g S3 MK,

D2p

S 3 is a 2fold branched cyclic covering, j : MK, MK, is where g : MK, K, a pfold unbranched covering, h : M MK, is the quotient by the period 2 K, , and f : MK, covering translation of M S 3 is the pfold irregular branched dihedral covering. Irregular branched dihedral covers are an exceptionally rich class of 3manifolds in fact any closed orientable 3manifold is homeomorphic to a 3fold irregular branched dihedral covering space of some knot [18, 26, 32]. A cut-and-paste construction of MK, is given in [7]. We present a surgery construction of this space, based on Section 2. Contemplate the sliced local picture of an untying link L, and let components of L be equipped with an orientation. When we cross a slice, the component in

A KONTSEVICH INVARIANT FOR COLOURED KNOTS

the lift of the link is acted on by a deck transformation in the pfold irregular branched dihedral cover of S 3 OK, by conjugation by the colour of the slice. To picture this, drawn p copies of L one above the other representing copies of L in the irregular branched dihedral cover, and examine how crossing the slices connects on to the other. This is illustrated below for p 3:

L L L
We nd that the lift of the link in 2.1 is one of the following: p 3:

5:

3.2. Equivariant Linking Matrices. In the present section we calculate the D2p equivariant linking matrix of L in terms of the linking matrix of L. We rst assign an orientation and a basepoint to each component of L, a basepoint p for which is disjoint from L, and paths from p to the basepoints of the components of L such that the paths and basepoints all lie within a small ball in . 3.2.1. Link Diagrams in . As before, let be a genus n solid handlebody, viewed as D I for a standard disc with n holes D. Then any link L : S1 has a link diagram in D, unique up to Reidemeister moves. A link diagram of L in D is said to be pointed if the image of each component of L comes equipped with a basepoint distinct from the crossing points. Such a diagram is unique up to Reidemeister moves away from the basepoints. Pointed link diagrams are in bijective correspondence with tangles in , whose endpoints are on standard points on D 0 and on D 1. Every link in determines a pointed link diagram by choosing arbitrary basepoints for its components. To incorporate colouring information into link diagrams and into pointed link diagrams, equip D with n simple non-intersecting curves C1 , C2 , . . . , Cn called gluing sites from each of its holes to the boundary of the disc which are projections of the slices of and have colours induced from the colour of the corresponding slices. Explicitly, for n 2 we can picture D C1 C2 as

10

DANIEL MOSKOVICH

A (pointed or non-pointed) link diagram in D C1 C2 Cn is called a coloured link diagram, and is assumed to be in general position with respect to the gluing sites. 3.2.2. Equivariant Linking Matrices. Let x denote a crossing between components Li and Lj in L, dene lblx Li , Lj as follows. Beginning at the basepoint of a component Li of L, proceed along Li in the direction of its orientation. When we cross a slice we record the colour encountered as an element of D2p (if the colour of the slice is c then record tsc D2p ). Continue until we arrive at a crossing x, then cross over to the component Lj and proceed along Lj in the direction of its orientation recording colours until we reach the basepoint of Lj . We obtain a sequence g1 , g2 , . . . gk D2p . Dene now: Note that lblx Li , Lj lblx Li , Lj :
D

lbl1 Li , Lj . Using this term we dene: x

g1 g2 gk

Wij 2p L :

1sgnx lblx Li, Lj


i, j we

where sgnx denoted the sign of the crossing x. Considering all 1 obtain an n n matrix: W D2p L :

Consider the diamond of covering spaces 3.1, and let L, L, and L be lifts of L to K, , and MK, correspondingly. By choosing a fundamental domain MK, , M with respectto covering translations in each of these coverings, we may choose K1 , . . . , K , K1 , . . . , K , and K1 , . . . , K connected components of the lifts of K1 , . . . , K . The groups of translations are ZpZ, D2p and S2 correspondingly. Choose actions such that a path which starts at p, crosses the slice coloured tsi , and returns to p (without crossing the slice again) is lifted to a path which starts at p and ends at tsi ptsi , tsi p, and tp correspondingly. The lifted links L, L, and L may now be identied with the sets of translates: K1 , tK1 t, tsK1 ts, . . . , tsp1 K1 tsp1 , K2 , . . . , tsp1 K tsp1 K1 , tK1 , . . . , tsp1 K1 , K2 , . . . , tsp1 K K 1 , tK 1 , K 2 , tK 2 , . . . , K , tK

Mat ZD2p This is a Hermitian matrix (that is, W D L is equal to its transpose followed by the involution g g 1 ). The topological meaning of W D L is as follows:
D
2p 2p

Wij 2p L

A KONTSEVICH INVARIANT FOR COLOURED KNOTS

11

We dene equivariant linking matrices for L, L, and L as follows: Linki,j L : Linki,j L :


Linki,j L :
g D2p

gLinkKi , g Kj g gLinkKi , g Kj

g D2p

LinkKi , Kj tLinkKi , tKj

Lemma 3.1. W D2p L is equal to the equivariant linking matrix for L and for and the matrix induced from it by the projection D2p S2 is the equivariant L, linking matrix for L. Proof. Note that by projecting D2p to 1 the linking matrix of L is recovered from W D2p L. To recover the linking matrices of L, L, and L, make the following substitutions: For L: Mat33 Z : Mat ZD2p 1 For L: I33

001 010 100

010 001 100

: Mat ZD2p 1 I66 s

0 1 0 1 10 10

01 10

Mat66 Z I22

0 I 0 0 0 I I 0 0

with I

For L:

: Mat ZD2p 1 I22 t

Mat22 Z

0 1 10

Lemma 3.2. (1) (2) (3)

LinkL LinkL LinkL

LinkL LinkL LinkL

Proof. This follows from looking at the lifts of these matrices to the universal cover. Remark 3.3. Note that the linking matrices of L and of L do not change under the move (recall Section 2.3) because the contribution of the arc which has been slid to these matrices is 1 t1 s s2 sp1 which generates the kernels of and .

12

DANIEL MOSKOVICH

4. A pColoured Diagram Valued Kontsevich Invariant of the Link In the present section we use the information we have gleaned about lifting untying links in the complements of untied p-coloured knots in order to dene a pcoloured diagram valued Kontsevich invariant of the link from which we can then cross nally to the LMO invariant of MK, . We are continuing then with the arrow S3 L
1

S 3 OK, L

In Sections 4.1 and 4.2 we review well-known concepts related to the Kontsevich invariant in a slightly modied setting. Section 4.1 is a straightforward generalization of [16, Section 2.2] presented slightly dierently. Section 4.2 reviews denitions and some basic results concerning Jacobi diagrams in our context, in a form somewhere in-between [20, Section 3] and [16, Section 3.3]. 4.1. pColoured (Pointed) Tangle Diagrams. In this section we decompose a (pointed or unpointed) link diagram in D C1 C2 Cn into diagrams on simpler surfaces as in [16]. Consider the following surfaces in R2 : t 0

b and c contain a distinguished segment of their boundaries called a gluing site marked by an arrow. There are well-dened notions of tangle diagrams and of link diagrams on such surfaces, with boundary points of the diagram lying on standard points of the boundaries of a , b , and c . Consider an m1 , m2 tangle diagram in i . In the pointed case, the corresponding tangle diagrams are also said to be pointed, and images of basepoints of link components are marked on the diagram by associating to it a subset of top basepoints S1 m1 and a subset of bottom basepoints S2 m2 . Graphically, we represent such a diagram as a tangle diagram in i with the kth boundary points at t 0 marked if k S1 , and the lth boundary point at t 1 marked if l S2 . An example of a dihedral based 3, 3tangle diagram with S1 2, 3 and S2 is given below:

b Composition and tensor products of pcoloured (pointed) tangle diagrams are dened as with tangles, with Si Sj and Si Sj denoting Si composed with Sj and Si tensor Sj correspondingly. Composed pcoloured (pointed) tangle diagram are required to match up on their boundaries shapes, images of boundary points, and images of basepoint. For example:

A KONTSEVICH INVARIANT FOR COLOURED KNOTS

13

Ti

Ti1

Remark 4.1. Our concept of a tangle diagram is more general than in [20, 16] because we allow tensor products. That is because in this paper we must make concrete calculations, and allowing tensor products of tangle diagrams reduces the number of associators we must take into account. A pcoloured link diagram in D C1 C2 Cn may be decomposed with respect to some generic height function on D into a sequence of tangle diagrams D E1 , . . . , El1 , F1 , G1 , El1 1 , . . . , E2 , F2 , G2 , . . . , Fn , Gn , Eln1 1 , . . . , Eln

where 0 l1 l2 ln and Ei are tangle diagrams in a , Fi are tangle diagrams in b , and Gi are tangle diagrams in c . In the pointed case, we require in this decomposition that the basepoint of each element be mapped to an pair of endpoints of pointed tangle diagrams (on the top of some Ti and on the bottom of some Ti1 ) which are marked. The slices C1 , C2 , . . . , Cn map to the gluing sites. The pcolouring of a gluing site is the pcolouring of the corresponding slice. A pcoloured (pointed) tangle diagram whose gluing sites are pcoloured is said to be a pcoloured (pointed) tangle diagram. The above decomposition of a pcoloured link diagram into pcoloured tangle diagrams is unique up to the following equivalence relation: Denition 4.2. Regular isotopy is the equivalence relationship between dihedral tangle diagrams generated by the following moves: Move I: Te Te

Te Tf Tg

Tf Tg

Tf Tg

where the glued edge Te must have no images of basepoints on it. Move II: Ta Tb Td Tc

Unit for Tensor Product: If Ti is an empty pcoloured tangle diagram on x and Tj is any pcoloured (based) tangle diagram on y then Ti Tj Tj Ti Tj , for some x, y a, b, c.

14

DANIEL MOSKOVICH

Changing Projections: Described below. To eliminate the basepoints, we must slide them along the link and prove invariance under such slides of whatever it is that we are examining. This corresponds to the following moves: Denition 4.3. Sliding marked endpoints along consecutive tangles is called a basing move. If the gluing site is not crossed this is called a 1 move, otherwise it is called a 2 move. We now attach extra information to the top and bottom endpoints of an m, n tangle, making them what Bar-Natan calls non-associative tangles [3]. The category of non-associative tangles is dened as a category whose objects are words in the free magma on two letters , (we denote the empty word ), whose morphisms are freely generated by the following basic non-associative tangles (A, B, C , ): A denoted . (1) The identity morphism A (2) The associativity morphism ABC AB C and its inverse, denoted and correspondingly. (3) The braiding morphism AB BA and its inverse, denoted and correspondingly. (4) Pair creationannihilation AB (A B) and its inverse, denoted and correspondingly. For a morphism w1 w2 , we call w1 the top boundary word and w2 the bottom boundary word of M . An m, ntangle diagram in a which we call A becomes a non-associative tangle diagram by assigning to A two words w1 , w2 in the free magma on two letters , of lengths m and n correspondingly, along with a decomposition of A into elementary non-associative tangles such that the top boundary word of the decomposition is w1 and the bottom boundary word is w2 . For m, n-tangle diagrams in b we further require that w2 be of the form v1 v2 where the number of letters in v1 corresponds to the number of endpoints of the tangle diagram adjacent to the gluing site and v1 has the standard left bracketing. For c we make the corresponding demand of the top boundary word. The main denition for this section is: Denition 4.4. A presentation of a link diagram A in D is a decomposition of A into 1 cuK, non-associative identity tangles in b , 1 cuK, non-associative identity tangles in c , and any number of elementary non-associative tangles in a such that the boundary words match up and the top boundary word and bottom boundary word of the decomposition are the empty words. A presentation is written as a vector of boundary words above a vector of tangles where the empty words at the top and bottom are omitted. Clearly every link-diagram has a presentation. In the example below (corresponding to the untying link for the 3coloured 74 knot) we abridge and multiply out some factors so the composite tangles are no longer elementary, but the meaning is clear:
M

A KONTSEVICH INVARIANT FOR COLOURED KNOTS

15

, , ,

1 1 0

,
0

We can now complete Denition 4.2 with the nal move: Changing projections:

, ,
g g


g g

is equivalent to

,
g

For pointed link-diagrams, we incorporate the basepoint information into the boundary words by considering ordered triples of a non-associative tangles T together with a subset S1 m and a subset S2 n. Graphically, we represent such a tangles by writing the boundary words as elements in the free magma over four letters , , , rather than over two , , where the ith letter of the top (bottom) word w is marked with a cross if i S1 (if i S2 . For w a word in , ,, and , let w denote w with the crosses forgotten. The generalization to the case of coloured non-associative tangles is straightforward. 4.2. Jacobi Diagrams. In this section we review the Jacobi-diagram terminology which is relevant in our context. For general references for the theory of Jacobi diagrams, see e.g. [2, 31]. Some of denitions and conventions we use here dier from classical usage in particular our division of a skeleton into a hard and soft part and view of a circle as a trivalent graph with no vertices and no edges are non-standard. A uni-trivalent graph is a nite pseudograph (the pseudo tells us that we allow multiple edges and loops) for which each vertex has valence either 1 or 3, and whose trivalent vertices are oriented, i.e. a cyclic order of the three edges around each trivalent vertex is xed. A uni-trivalent graph which has no legs is called a trivalent graph. Convention. A circle is considered a trivalent graph with no vertices and no edges. A skeleton is dened to be a disjoint union of oriented circles ( X ) and oriented segments ( X ), collectively called the hard skeleton; and two sets (X and

16

DANIEL MOSKOVICH

1
1

1
2

1
x Figure 5. A Jacobi diagram.
X ), collectively called the soft skeleton (we refrain from using the standard term colours to avoid confusion). Our main object of study is now dened to be a uni-trivalent graph G whose legs are labeled by points a given skeleton such that no two legs are labeled by the same point in the hard skeleton (although many dierent points may be labeled by the same point in the soft skeleton). This is called a Jacobi Diagram. By denition a circle is a Jacobi diagram with no legs and no arcs. We call this the null Jacobi diagram and denote it . A Jacobi diagram with two legs (labeled x and y) and no internal vertices is called a chord and is denoted x y . An example of a Jacobi diagram is given in Figure 4.2. 1dimensional components of the skeleton are drawn using thick lines, while the uni-trivalent graphs are drawn using dotted lines. Colours are denoted 1 , . . . , n , 1 , . . . , n . Let D denote the free Qmodule of Jacobi diagrams over a skeleton . We dene local moves on elements of D which dier inside a dotted circle as indicated below. The equivalence classes of elements of D modulo the relations AS, IHX, ST U and a further global move IB which we dene below are denoted A. Such a space is called an Aspace.

The AS relation

The IHX relation

The STU relation

The IB (Innitesimal Basing) move takes a Jacobi diagram G to the Jacobi diagram G obtained taking the sum of all ways of attaching all legs labeled i to

A KONTSEVICH INVARIANT FOR COLOURED KNOTS

17

edges adjacent to legs labeled i with a counterclockwise direction, for all i. This is called the xavoured link relation in [10]. An example is given below:

1 1 2
1

3 2

1 1

3 2

1 1

3 2

1 1

1 4

1 4

1 4

1 4

In the same way that the word knot is used in Knot Theory for both equivalence classes of knots (modulo ambient isotopy) and individual embeddings K : S 1 S3 which are representatives of those classes, we shall use our terminology loosely and call both elements in A and elements in D by the same name, Jacobi diagrams. Remark 4.5. Our A is usually called A in the literature.
o

Two maps between spaces of Jacobi diagrams are ubiquitous in the theory of nite-type invariants. The rst is the symmetrization map where Y is any skeleton, which maps a diagram D AX to the average of all ways of placing legs coloured x on an oriented interval labeled x, for all x X, and ignored legs labeled by elements of Y . Its inverse is called the formal PBW map and is denoted by The space of Jacobi diagrams over a given skeleton A is graded by degree, where the degree of a Jacobi diagram is dened as half its number of vertices. In the present note, we shall be interested in two classes of Jacobi diagrams which carry auxiliary structure. An oriented Jacobi diagram whose edges are labeled by nite series of elements in the set ZD2p and an orientation of such edges is called a pcoloured winding diagram. Such labels have been drawn in the literature as boxes with labels inside them and as labeled beads, but we prefer to draw them as triangles on the vertices labeled by the appropriate elements of ZD2p . We think of them as auxiliary bivalent vertices with an distinguished ingoing edge and outgoing edge, and call them coupons following [20]. They are subject to the relations: The Triv relation 1 The OR relation g The Linearity relation gh g X : A
X

X : AX

AX

g 1

g h

hg

18

DANIEL MOSKOVICH

y s s2

2 2
1

1
ts s x Figure coloured diagram. 6. A p winding

1 2
2

s2

ts

Figure 7. A p coloured Jacobi diagram.

The VI relation g g g g

g g Here OR stands for orientation reversing, and V I for vertex invariance. An (unoriented) Jacobi diagram edge-labeled by elements in ZD2p shall be called a pcoloured Jacobi diagram. We may map pcoloured winding diagrams to pcoloured Jacobi diagrams by the threading map. This is dened on coupons labeled by non-trivial elements of D2p by

(4.1)

T hrD2p

k g k

and extended by linearity to edges labeled by elements in the integer group ring ZD2p . Jacobi diagrams (without coupons) and pcoloured winding diagrams are related by the operation of pushing coupons to legs. For the denition of this operation for a general (not necessarily dihedral) group see [16, Section 3.6]. In the dihedral context, for D AX Y , dene DX Xtsi to be the diagram obtained by D by attaching a coupon labeled tsi D2p pointing away from X to each edge adjacent to a leg labeled by an element in the set X. For example: y

1 2

Dx

ts2 x

1
ts2 ts2

1 2

ts2 x

A KONTSEVICH INVARIANT FOR COLOURED KNOTS

19

4.3. An Invariant of pColoured Untying Links. In this subsection we construct a non-commutative analogue of Krickers winding diagram-valued Kontsevich invariant of a link [20]. We begin by recalling the combinatorial denition of the Kontsevich invariant (see for instance [31, Chapter 6]). The Kontsevich invariant is a map which eats non-associative tangles and spits out elements of Aspaces. Consider again the elementary non-associative tangles:

which we assume to be oriented clockwise or down. On these elements we dene the Kontsevich invariant as follows:

Z Z
where

R
1 2

Z Z

1
1

R 1 2

denotes a rational associator, an element of A x,y,z whose existence is known but whose exact form is still shrouded in mystery. See [4, 21, 22]. is the wheels element [8, 12, 35]. It is dened to be exp

n 1

b2n w2n

where b2n are the modied Bernoulli numbers dened by

n 1

b2n x2n

1 log 2

sinh x 2
x 2

R denotes the Jacobi diagram

1 exp 2

w2n are the wheels which are the following Jacobi diagrams

w2

w4

w6

To obtain the Kontsevich invariant of any elementary tangle from the above denition, we make use of the following two operations:

20

DANIEL MOSKOVICH

Comultiplication:

Antipode: . . .
S

1n

. . .

where n is the number of legs adjacent to the hard skeleton component whose orientation we are reversing. For an example of how these operations are used in practice to calculate the Kontsevich invariant, see Section 6.2. Given a presentation of a link diagram in D (Denition 4.4), all that remains is to dene our analogue of the Kontsevich invariant at gluing sites and at images of basepoints. This is done with the help of the following maps: (1) At a gluing site labeled n: Gn

tsn

tsn tsn

For Ti and Ti1 non-associative tangle diagrams on b and on c correspondingly, separated by a gluing site labeled n, and the matching boundary word is factored as w w1 w2 , then: Z D2p T : Z D Ti Iw
2p

(2) Let Iw denote Iw with each crossed strand broken. For instance:
For Ti and Ti1 non-associative tangle diagrams on a and w is the boundary word between them: Z D2p T :

Gn 2 Z D2p Ti1 w

Z D Ti Iw Z D Ti1
2p 2p

Lemma 4.6. Z D2p is a well-dened invariant of a link diagram in D. Proof. The proof that Z D2p does not depend on the choice of bracketings and is preserved by under regular isotopy (Denition 4.2) is the same mutatis mutandis as the proof of [16, Lemma 4.3] and so is omitted. Lemma 4.7. Z D2p respects basing relations (see Denition 4.3). Proof. Identical to [16, Proof of Lemma 4.4]. Lemma 4.8. Z D2p L Z D2p L Z D2p L

A KONTSEVICH INVARIANT FOR COLOURED KNOTS

21

Proof. This follows from the corresponding properties for Z, for G, and for all normalization factors used. Recall that a Jacobi diagram is said to be Xsubstantial for a subset X of its skeleton if it contains no strut components both of whose legs are labeled by points in elements of X. It is said to be substantial if X is the entire skeleton. Proposition 4.9. For any untying link L,

Z D2p L

1 exp

xj
D

Wij 2p L xi

i,j

where R is a series of substantial diagrams. Proof. Our proof identical to the proof of [20, Theorem 3.5.5] and so is omitted. 5. Diagrammatic Integration In the following section we construct a D2p equivariant invariant of the irregular dihedral covering space MK, from the pcoloured winding diagram valued Kontsevich invariant of its surgery presentation L. In this way we realize the nal top arrow of the right commutative diagram in Equation 1.1:
surgL MK,  S 3 L An invariant of MK, which gives a non-commutative analogue of the rational Kontsevich integral should be analogous to the LMO invariant of MK, [36]. The LMO invariant of a 3manifold is dened by performing an operation called LMO integration in [11] on the Kontsevich integral of its surgery presentation. The basic idea of LMO integration is to kill the skeletons of Jacobi diagrams appearing in the Kontsevich integral by gluing legs in a certain way, until we obtain a sum over Q of Jacobi diagrams whose underlying graphs are trivalent. It is straightforward to perform an analogous gluing of legs in the pcoloured setting, and this is what we do in Section 5.1. Proving that the formal sum obtained is indeed an invariant of MK, (and therefore of K, ) is less simple, and this is put o until Section 6.2. If the linking matrix of L is invertible over Z (this occurs if MK, happens to be an integral homology 3sphere) then LMO integration simplies to rhus A integration (also called formal Gaussian integration) see [9, 10, 11]. This is the subject of Section 5.2.

5.1. LMO Integration. Let Z D2p L be the pcoloured winding diagram valued Kontsevich invariant of a pcoloured untying link L for K, which has com ponents, and let Z D2p L denote Z D2p L . In this section we do not need basepoints for L, so we are free to assume that Z D2p L has only oriented circles as its skeleton. When we change L by pcoloured Kirby moves, or when we perform a move, Z D2p L changes. LMO integration may be na vely thought of as a method of obtaining invariance under these moves by gluing legs and introducing relations tailor-made to cancel the changes. There does exist a more conceptual interpretation of LMO integration as outlined in [11], but it is still hand-wavy. We begin by introducing the relations we would like to mod out by (assume that n 2):

22

DANIEL MOSKOVICH

The On relation

2ng
The Pn relation 2n legs

0
All ways of completing with dotted arcs

The ZM relation

t 11 s s2 sp1

A conceptual explanation of the On and Pn relations (as the dimension of some phantom vector space and its symmetric algebra) is given in [11, Section 3.2]. The rst few relations Pn are: The P2 relation

The P3 relation

. . .
For the ZM relation, recall Remark 3.3 which implies that a move corresponds to introducing the element 1t1ss2 sp1 which ought to be degenerate. The next step is to replace solid circle components of the skeleton with dashed graphs. Following [36] we dene a series of elements Tm A1 , 2 , . . . , m as follows. Set T0 T1 0 and set T2 1 2 . For m 3, dene

A KONTSEVICH INVARIANT FOR COLOURED KNOTS

23

where r k1 and T is the Jacobi diagram corresponding to the permutation as follows: 1 2 3 m2 0 m1

1r T m 1 m2 r S is the number of k 1, 2, . . . , m 2 satisfying k


Tm :

m 2

1 2 3 m2 1 2 3 Figure 8. The Jacobi diagram T .

m2

A to To eliminate the skeleton, dene the map 1 : A 1 , 2 , . . . . l be the linear map which replaces a solid circle component of the skeleton with m legs attached to it by Tm with legs glued up in pairs correspondingly:

where if orientations are reversed then we multiply Tm by 1m . The result is taken modulo On . The maps n : A 1 , 2 , . . . . ln A are dened similarly, except that rst we act n 1 times on each solid circle by the comultiplication map 1 , and multiply the result by the normalization factor n! . The rst few values of Tm for m 2 are: T3 1 2 T4 1 6

. . .
Let and denote the number of positive and negative eigenvalues of LinkL respectively.


1 6

Tm

24

DANIEL MOSKOVICH

Denition 5.1. Let L be a pcoloured untying link for a pcoloured knot K, with irregular dihedral cover MK, . Then we dene the pcoloured LMO series:
D Zn 2p ,gp MK, :

n Z U

n Z U

n Z D2p L

where U and U denote the 1 and 1framed unknots respectively. The sum of the pcoloured LMO series is the pcoloured rational Kontsevich invariant : Z D2p ,gp MK, : where Zn 2p
D ,gp

1 Z1

D2p ,gp

D MK, 1 Z2

2p ,gp

MK, 2 MK, .

D MK, i denotes the degree i part of Zn

2p ,gp

Theorem 2. Z D2p ,gp :

pcoloured knots

Agp,0 ZM

is an invariant of pcoloured knots. This theorem is proven in Section 6.2.

5.2. rhus Integral. For the present section we assume that MK, is an integral A homology 3sphere, and that therefore LinkL is invertible over Z. In this case D2p ,gp the denition of Z simplies allowing us to view it as a pcoloured formal Gaussian integral. The simplied version of the LM O integral is called the rhus A Integral. We briey recall the denition of Gaussian integration. Let V be a vector space with Lebesgue measure dv and consider IT eT dv where T is a polynomial V 1 which may be written as a sum of the form T 2 Q P where Q is a nondegenerate quadratic and P is some perturbation. The Fourier transform means Q12 where , denotes the usual that up to homothety we have IT eP , e pairing S V S V C. The diagrammatic analogue of this [9, 10, 11] is based on the formula
1

xj
D

Z D2p L

exp

Wij 2p L xi

i,j

of Proposition 4.9 where the equivariant linking matrix plays the role of the quadratic and R is the perturbation. The analogy to the pairing is given by , : A X A X A which sends a pair of Jacobi diagrams to the sum of all ways of pairing up all Xlabeled legs in the rst argument to all Xlabeled legs in the second argument, and to zero if the number of Xlabeled legs in each argument is dierent. For example:

A KONTSEVICH INVARIANT FOR COLOURED KNOTS

25

x y x z

y z

x
y y z z y

z z

To dene an analogue to Gaussian integration we must invert the equivariant linking matrix. For W D2p L to be invertible if and only if LinkL is invertible over Z, we must consider it over the Cohn localization Rloc of the quotient ring R : ZD2p 1 t1 s sp1 ([16]). It seems an interesting problem to identify Rloc explicitly. One key lemma in the construction is: Lemma 5.2 (Translation Invariance). For any matrix power series M set

Mat R

F x M x : Then:

F x1
n

i1

M1i1 x 1 1i

i2

M1i2 x 2 1i

M1i x 1i

F x dX

F x M x dX

Proof. Our proof is identical to [11, Proof of Proposition 3.1] (see also [20, Proof of Theorem 7.2.1]), and is omitted. We continue by repackaging LMO integration as in [11, Proposition 2.1]: Lemma 5.3.
m

m
G dX :

m where:
1
x X x x

m!

,G
X

Om , Pm1 , ZM

Proof. Compare coecients. Continuing, for a cycle

Sm consider the product


1 xi 1 xi 2 xi m mi 1
m

x1 , x2 , . . . , xm :
detM :

With respect to this product we have the function

1sgn

Mi,i , Mi,2 i , . . . Mi,i

cycle of i 1

Lemma 5.4. detW D2p L is an invariant of K, .

26

DANIEL MOSKOVICH

Proof. Our proof consists of two stages. First we show that det is a Whitehead determinant. Then we prove that any Whitehead determinant of W D2p L is an invariant of K, . Recall that a Whitehead determinant is dened as a function MatR R satisfying the following properties [24]: i) detAB detA detB . ii) deteij r 1 for each elementary transvection eij r (an elementary transvection is the identity matrix plus a matrix which is zero everywhere except in its i, j th coordinate where it is r). iii) detA Im m detA for all m. Properties 2 and 3 are clear from the denition. It remains to prove Property 1. 6. Proofs 6.1. Proof of Theorem 1. Our theorem may be proved in much the same way that the relative version of the Kirby Theorem is proved by Garoufalidis and Kricker [15]. Our notation closely follows that of Roberts [33]. Let L1 and L2 be two untying links for a pcoloured knot K, , and let M denote the ambient manifold S 3 OK, . Pick a base-point for M which avoids the L1 and L2 . Let ML1 and ML2 be dieomorphic manifolds relative to their boundaries, obtained from M by surgery on the framed links L1 and L2 , with basepoints induced by the base-point of M . We can assume that the dieomorphism between them preserves base-points. Our goal is to prove that L1 and L2 are equivalent modulo our expanded set of Kirby moves. Let W0 and W1 be 4manifolds obtained by taking M I and attaching 2handles to the top surface along L0 and L1 . Put in another way, W0 and W1 are cobordisms between M and ML0 and ML1 correspondingly. The boundaries W0 and W1 are homeomorphic (we write this boundary as M M I N where N is homeomorphic to MLi , i 0, 1, following Roberts), and so we can cross the boundaries with I and glue them together using the homeomorphism, to get a closed smooth 4manifold W . See Figure 6.1.

ML0

ML1

W0

W1

Let : 1 W0 1 W1 be the fundamental group of W , and let K , 1 be the (pointed) EilenbergMacLane space associated to this group. We have a

A KONTSEVICH INVARIANT FOR COLOURED KNOTS

27

well-dened map : W K , 1 from W into K , 1 which maps the baseline (the basepoint of M times an interval ending at the basepoints of WL0 or WL1 ) to the base-point of K , 1. By [15, Theorem 4] we know that L1 and L2 are equivalent by stabilization and handle-slides if and only if W 0 H4 . If H4 would vanish as it does in the abelian case, we would be nished and pcoloured circumcision would not be necessary. But already for the simplest case OK, 31 , 1 and L1 L2 this is not the case. But fortunately we may kill this homology group by pcoloured circumcision as follows: For L an untying link for a pcoloured knot, let 3 C, L denote the union of L with a component C in the kernel of the pcolouring which ringed by a 0framed unknot (we call this pcoloured circumcision of L by C). The manifolds ML and M3 C,L have the same homology because all we have done is to add a canceling pair of handles, but is divided by the relation induced by C. In the degenerate case where L1 L2 are empty links, we rename W by W. For simplicity, we assume that L1 , L2 ker both consist of a single component (since cobordisms may be stacked, proving the theorem for this case proves it in general). Here W diers from W by a single 2-handle attached along the curve L1 (or L2 ) [17]. Take C a component parallel to L1 but with opposite framing and orientation, and apply pcoloured circumcision by C to W (C may be chosen to be disjoint from L2 in W ). This will kill the contribution of L1 (and so also of L2 ) to the fundamental group of W , reducing it to the fundamental group of W . . We assume that the coloured untying It remains to prove the theorem for W invariant of K, is one (the proof easily generalizes to the general case). Laying out the p, 2torus knot symmetrically on the boundary of an imaginary unknotted solid torus and looking at it from above (consider Figure 2.1 or the embedding giving explicitly at the beginning of [27]), we obtain a knot diagram with p arcs a0 , . . . , ap1 . Denoting the meridians of these arcs m0 , . . . , mp1 , by the p colouring rule we may assume without loss of generality that mi tsi for i 0, . . . , p 1. The Wirtinger presentation of the knot group of tp, 2 (see [6]) is G: with i ZpZ and

1 x0 , . . . , xp1 xi xi1 x12 xi1 i

u, v u2 v p

x0 x1
1 2

x1 1

x0 x1

Choose C1 to be a curve that goes p times around a0 and a1 . The homotopy class corresponding to C1 is in ker because x0 x1 Zp D2p , and so pcoloured cir cumcision by C1 is a Kirby IIIp move, which takes W to a homeomorphic manifold W1 with fundamental group G1 : u, v u2 v p , v p Zp Z2

To turn W1 into a homeomorphic manifold W2 with fundamental group D2p we perform pcoloured circumcision to introduce the relation uvu1 v 1, or equivalently x1 x2 x1 1 or x2 x2 1. We achieve this by taking C2 to be a curve which 0 0 1

28

DANIEL MOSKOVICH

links to a0 twice and to a1 twice. Again this is in ker as its image under is t2 ts2 1. We conclude by recalling the homology of D2p . Lemma 6.1. For n odd, H4 D2n 0.

Proof. First we calculate the cohomology of D2n as in [1, Lemma IV.6.3]. We have H D2n ; Z2 Z2 e (the polynomial algebra on one generator with coecients in Z2 ), and H D2n ; Zn Zn b2 E eb where E is the exterior algebra, e is as before, and b is the order n Bockstein operation. Combining to get integral cohomology, we obtain:
Z 0

H i D2n

Z2 Z2n

for for for for

i 0, i 2 mod 4, i odd, i 0 mod 4.

Since D2n is nite, Hi D2n is nite abelian for all i 0. The Universal Coefcient Theorem now tells us that H i1 D2n is isomorphic to Hi D2n , which in particular implies the vanishing of H4 D2n . Summarizing, pcoloured circumcision has allowed us to simplify the fundamental group of W until it becomes D2p , whose fourth homology vanishes. This proves our pcoloured version by Garoufalidis and Krickers relative version of the FennRourke Theorem. Remark 6.2. The above proof is deeply unsatisfying as it articially works on the basis of a dicult proof which has since been substantially streamlined and improved. As pointed out by Andrew Kricker, a good proof of our theorem (and the main theorem of [15]) would use the simpler mapping class group methods of [23] and [25]. Remark 6.3. Many thanks to Kazuo Habiro for pointing out that Fenn and Rourkes condition that map the base-line of W to the base-point of K , 1 is necessary for to be well-dened. Consider for instance a closed 4-manifold with fundamental group Z4 , in which case K , 1 is the 4-torus. Without the condition on the basepoints, W would have Z indeterminacy because it could wrap W around the 4-torus any number of times. 6.2. Proof of Theorem 2. 6.2.1. Invariance under orientation change and Kirby moves. The arguments of [36] for proving invariance under change of orientation of a single link component and invariance under Kirby I and Kirby II carry over to our setting, where the slight dierences in conventions between this paper and [36] are handled by the comments in [16, Proof of Theorem 4]. We prove invariance under Kirby IIIp . Let L be an untying link for a pcoloured knot K, , and let L denote the union of L with a 1framed component x ker ringed by a 0framed unknot y. By the magic formula for the Kontsevich invariant for the long Hopf link [12, Theorem 4] (see also [5, 8]) we have:

A KONTSEVICH INVARIANT FOR COLOURED KNOTS

29

(6.1)

Z D2p

C R

Since n L vanishes on Jacobi diagrams which do not have 2n legs on each component, in particular it vanishes for Jacobi diagrams which do not have 2n legs on y. For D a homogenous summand of the RHS of Equation 6.1, we have by [31, Lemma 10.23]:

n D

n D

Pn

where D is D minus y and all legs labeled by points of y, and n1 the number of dotted arcs connecting n D to Pn and n2 the number of dotted connecting Pn to satisfy n1 n2 n. If n1 0 then n D 0 by [36, Lemma 3.3] since each connected diagram in has a trivalent vertex. Thus the only contributions to Z D2p L with legs labeled x or y come from exp x y . These are the same contributions as we would have from L the disjoint union of L and a 0framed Hopf link, and since L and L are related by Kirby I and Kirby II, we have: Z D2p L Z D2p L Z D2p L

6.2.2. Invariance under the Move. Invariance under the move results in a nontrivial way from the ZM relation. Since any coupon marked 1 t1 s sp1 will kill the Jacobi diagram it is on when we LMO integrate, we take the liberty of redening our coupons to be over ZD2p 1 t1 s sp1

R:

To further simplify notation, when there is a 1 t label in the diagram we write 1 s sp1 0 as such a diagram will also vanish when LMO integrated. Similarly when a Jacobi diagram D is in the kernel of LMO integration or is equivalent to a Jacobi diagram with fewer chords, we write D 0. By Section 4.1 without the limitation of generality concerning the possible colourings of slices, we may write the move as

30

DANIEL MOSKOVICH

t t ts t ts

ts ts ts

ts ts ts ts

where there are no associators around the coupons. To see how the pcoloured winding diagram valued Kontsevich invariant changes when the move is performed, we begin by recalling that the associator satises the following two relations (see for instance [31]): The pentagon relation

3 2 1

The hexagon relation

R 1 R 1 R

We now make the following calculation:

A KONTSEVICH INVARIANT FOR COLOURED KNOTS

31

(6.2)

1 2

( ( ( ( ( ( ) ) )

Z D2p
( )

D2p Z

S2 S1 R S 1 1 R S1

) )

12 S1 1 R

12 S1 1 gR g g S1 1 gR 12

with gR :

exp

1 2

x g

and the third equality follows from the hexagon relation. Taking the degree 1 term of the nal diagram gives: g1

12

Based on this we have the following lemma (where we are taking degree 1 terms for ease of notation, but replacing these with exponents does not change the result or the proof): Lemma 6.4. t t

t 1g t 1g

for any g R such that gh for some h R.

32

DANIEL MOSKOVICH

Proof. t t t-1 t g g 1 t t t-1 g g 1 t

We also know that t t ts ts 1 ts ts 1

Remembering that we have s1 valued coupons on both ends of the left skeleton component, this last term equals: t t

A:

1 ts t 1s1

Dene t t

B:

t 1s1 1 ts

Because for p (6.3) while for p 5:

3 we have:

s s2 s s2 2 s s2 1 s s2 s s2 2 s s2 3.

A KONTSEVICH INVARIANT FOR COLOURED KNOTS

33

s s4 s2 s3 s s2 s3 s4 1 s s4 2s s2 s3 2s4 4 s s2 s3 s4 5. the conditions of Lemma 6.4 are fullled for s, for s1 , for s s1 , and for s s1 , and therefore both A B 0 and A B 0. Thus A 0, nishing the proof.
(6.4) References
1. A. Adem and R.J. Milgram, Cohomology of nite groups, second ed., Grundlehren der Mathematischen Wissenschaften, vol. 309, Springer, 2004. 2. D. Bar-Natan, On the Vassiliev knot invariants, Topology 34 (1995), 423472. 3. , Non-associative tangles, Geometric Topology (Proceedings of the Georgia International Topology Conference) (Providence) (W.H. Kazez, ed.), Amer. Math. Soc., International Press, 1997, pp. 139183. 4. , On Associators and the GrothendieckTeichmuller group I, Selecta Math. (N.S.)) 4 (1998), 183212, arXiv:q-alg/9606021. 5. D. Bar-Natan and R. Lawrence, A rational surgery formula for the LMO invariant, Israel J. Math. 140 (2004), 2960, arXiv:math.GT/0007045. 6. G. Burde and H. Zieschang, Knots, second revised and extended ed., de Gruyter Studies in Mathematics, vol. 5, Walter de Gruyter, Berlin and New York, 2003. 7. S.E. Cappell and J.L. Shaneson, Linking numbers in branched covers, Contemporary Mathematics, vol. 35, American Mathematical Society, 1984, pp. 165179. 8. L. Rozansky D. Bar-Natan, S. Garoufalidis and D. P. Thurston, Wheels, wheeling, and the Kontsevich integral of the unknot, Israel J. Math. 119 (2000), 217237, arXiv:q-alg/9703025. , The rhus integral of rational homology 3spheres I: A highly non trivial at conA 9. nection on S 3 , Selecta Math. (N.S.) 8 (2002), 315339, arXiv:q-alg/9706004. 10. , The rhus integral of rational homology 3spheres II: Invariance and universality, A Selecta Math. (N.S.) 8 (2002), 341371, arXiv:math.QA/9801049. , The rhus integral of rational homology 3spheres III: The relation with A 11. the LeMurakamiOhtsuki invariant, Selecta Math. (N.S.) 10 (2004), 305324, arXiv:math.QA/9808013. 12. T. Q. T. Le D. Bar-Natan and D. P. Thurston, Two applications of elementary knot theory to Lie algebras and Vassiliev invariants, Geom. Topol. 7 (2003), no. 1, 131, arXiv:math.QA/0204311. 13. R.H. Fox, A quick trip through knot theory, Topology of 3Manifolds and Related Topics (Georgia 1961) (M.K. Fort, ed.), Prentice-Hall, 1962. 14. S. Garoufalidis and A. Kricker, Finite type invariants of cyclic branched covers, arXiv:math.GT/0107220, 2003. 15. , A surgery view of boundary links, Math. Ann. 327 (2003), 103115, arXiv:math.GT/0205328. 16. , A rational non-commutative invariant of boundary links, Geom. Topol. 8 (2004), 115204, arXiv:math.GT/0105028. 17. R.E. Gompf and A. Stipsicz, 4manifolds and Kirby calculus, American Mathematical Society, Providence, RI, 1999. 18. H.M. Hilden, Every closed orientable 3manifold is a 3fold branched covering space of S 3 , Bull. Amer. Math. Soc. 80 (1974), 12431244. 19. A. Kricker, Branched cyclic covers and nite type invariants, arXiv:math.GT/0003035, 2000. 20. , The lines of the Kontsevich integral and Rozanskys rationality conjecture, arXiv:math.GT/0005284, 2000. 21. V. Kurlin, Compressed Drinfeld associators, J. Algebra 292 (2005), no. 1, 184242, arXiv:math.GT/0408398. 22. T.T.Q. Le, On denominators of the Kontsevich integral and the universal perturbative invariant of 3manifolds, Invent. Math. 135 (1999), 689722. 23. N. Lu, A simple proof of the fundamental theorem of Kirby calculus, Trans. Amer. Math. Soc. 331 (1992), 143156. 24. B.A. Magurn, An algebraic introduction to K-theory, Encyclopedia of Mathematics and its Applications, vol. 87, Cambridge University Press, 2002.

34

DANIEL MOSKOVICH

25. S. Matveev and M. Polyak, A geometric presentation of the surface mapping class group and surgery, Comm. Math. Phys. 160 (1994), 537556. 26. J.M. Montesinos, A representation of closed orientable 3manifolds as 3fold branched coverings of S 3 , Bull. Amer. Math. Soc. 80 (1974), 845846. 27. D. Moskovich, Surgery description of orientationpreserving d2p actions on compact orientable 3manifolds, arXiv:math.GT/0508141, 2005. 28. , Acyclic Jacobi diagrams, arXiv:math.QA/0507351, 2006. , Surgery untying of coloured knots, Algebr. Geom. Topol. 6 (2006), 673697. 29. 30. T. Ohtsuki, A polynomial invariant of integral homology 3spheres, Proc. Math. Camb. Philos. Soc. 117 (1995), 83112. 31. , Quantum invariants, a study of knots, 3manifolds, and their sets, Series on Knots and Everything, vol. 29, World Scientic Publishing Co., 2002. 32. V.V. Prasolov and A.B. Sossinsky, Knots, links, braids and 3-manifolds: An introduction to the new invariants in low-dimensional topology, American Mathematical Society, 1997. 33. J. Roberts, Kirby calculus in manifolds with boundary, Turkish J. Math. 21 (1997), 111117, arXiv:math.GT/9812086. 34. L. Rozansky, A rational structure of generating functions for Vassiliev invariants, Notes accompanying lectures at the Summer School on Quantum Invariants of Knots and Three Manifolds (Joseph Fourier Institute) (C. Lescop, ed.), University of Grenoble, June 1999. 35. D.P. Thurston, Wheeling: A diagrammatic analogue of the Duo isomorphism, Ph.D. thesis, U.C. Berkeley, April 2000, arXiv:math.QA/0006083. 36. J. Murakami T.T.Q. Le and T. Ohtsuki, On a universal perpurbative invariant of 3manifolds, Topology 73 (1998), no. 3, arXiv:q-alg/9512002. Research Institute for Mathematical Sciences, Kyoto University, Kyoto, 606-8502 JAPAN E-mail address: dmoskovich@gmail.com URL: http://www.sumamathematica.com/

Você também pode gostar