12.07.2015 Views

An equivariant Casson invariant of knots in homology ... - CiteSeerX

An equivariant Casson invariant of knots in homology ... - CiteSeerX

An equivariant Casson invariant of knots in homology ... - CiteSeerX

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

<strong>An</strong> <strong>equivariant</strong> <strong>Casson</strong> <strong><strong>in</strong>variant</strong> <strong>of</strong> <strong>knots</strong> <strong>in</strong> <strong>homology</strong> spheres ∗Julien Marché †March 21, 2005AbstractFrom the construction <strong>of</strong> the <strong>Casson</strong> <strong><strong>in</strong>variant</strong> <strong>of</strong> <strong>homology</strong> spheres us<strong>in</strong>g <strong>in</strong>tersection onconfiguration spaces, we propose a construction <strong>of</strong> an <strong>equivariant</strong> <strong>Casson</strong> <strong><strong>in</strong>variant</strong> for a knotK homologous to 0 <strong>in</strong> rational <strong>homology</strong> sphere M. Our construction is adapted from C.Lescop ([7]) and use the same ideas <strong>in</strong> an <strong>equivariant</strong> sett<strong>in</strong>g. We show that the <strong><strong>in</strong>variant</strong>we obta<strong>in</strong> is similar to the 2-loop part <strong>of</strong> the rational Kontsevich <strong>in</strong>tegral, and may be equal.We check that it codes the <strong>Casson</strong> <strong><strong>in</strong>variant</strong>s <strong>of</strong> the cyclic ramified cover<strong>in</strong>gs <strong>of</strong> M along K<strong>in</strong> exactly the same way as found by S. Garoufalidis and A. Kricker for the 2-loop <strong><strong>in</strong>variant</strong>([2]).1 IntroductionDeal<strong>in</strong>g with universal f<strong>in</strong>ite type <strong><strong>in</strong>variant</strong>s <strong>of</strong> l<strong>in</strong>ks <strong>in</strong> 3-manifolds, two approaches are usedessentially, both l<strong>in</strong>ked with perturbative Chern-Simons theory. The first one uses <strong>in</strong>tegrationon configuration spaces <strong>of</strong> po<strong>in</strong>ts ly<strong>in</strong>g <strong>in</strong> a prescribed way on the l<strong>in</strong>k and the 3-manifold. Theidea <strong>of</strong> the construction is rather elegant but the actual computation <strong>of</strong> the <strong><strong>in</strong>variant</strong>s are verytechnical (see for <strong>in</strong>stance [5, 6, 7]). The second approach is based on the Kontsevich <strong>in</strong>tegral.Although computations are difficult <strong>in</strong> this sett<strong>in</strong>g too, the comb<strong>in</strong>atorics <strong>of</strong> the Kontsevich<strong>in</strong>tegral make it a more tractable <strong><strong>in</strong>variant</strong>. By <strong>in</strong>stance, formulas for <strong><strong>in</strong>variant</strong>s <strong>of</strong> lens spaces,some Seifert fibered spaces and torus <strong>knots</strong> can be achieved [1, 10]. We will focus on a property<strong>of</strong> f<strong>in</strong>ite type <strong><strong>in</strong>variant</strong>s <strong>of</strong> a knot <strong>in</strong> a 3-manifold called rationality which were shown us<strong>in</strong>g theKontsevich <strong>in</strong>tegral (see [3]).Let us expla<strong>in</strong> this briefly: for any knot K homologous to 0 <strong>in</strong> a rational <strong>homology</strong> sphere M,the (unwheeled) Kontsevich <strong>in</strong>tegral is a series <strong>of</strong> uni-trivalent graphs with rational coefficientshav<strong>in</strong>g the follow<strong>in</strong>g form: Z # (M, K) = exp( ∑ i>0 z i). In this formula z i is a series <strong>of</strong> connectedgraphs hav<strong>in</strong>g first Betti number (or loop number) equal to i. The first series z 1 is directlyl<strong>in</strong>ked to the Alexander polynomial ∆ <strong>of</strong> the pair (M, K) and hence is well understood. Therationality theorem states that each z i for i > 1 is coded by a f<strong>in</strong>ite sum <strong>of</strong> diagrams with loopnumber i with rational expressions <strong>of</strong> the form P (t)∆(t)attached to the edges (see [3]). We call thissum zirat (M, K). For a fixed <strong>in</strong>teger i, this <strong><strong>in</strong>variant</strong> is a k<strong>in</strong>d <strong>of</strong> polynomial <strong><strong>in</strong>variant</strong>, and codesa big part <strong>of</strong> the Kontsevich <strong>in</strong>tegral. This gives a good motivation for comput<strong>in</strong>g it, start<strong>in</strong>gwith the 2-loop part z2 rat (M, K).∗ keywords: f<strong>in</strong>ite type <strong><strong>in</strong>variant</strong>s, configuration space, Kontsevich <strong>in</strong>tegral, rationality, <strong>Casson</strong> <strong><strong>in</strong>variant</strong>.2000 Mathematics Subject Classification: 57M27.† Institut de Mathématiques de Jussieu, Équipe “Topologie et Géométries Algébriques” Case 7012, UniversitéParis VII, 75251 Paris CEDEX 05, France. e-mail: marche@math.jussieu.fr1


This happens to be a difficult but promett<strong>in</strong>g task because this so-called 2-loop part <strong>of</strong> theKontsevich <strong>in</strong>tegral has some strik<strong>in</strong>g topological properties. Among them, we can mentionbeautiful formulas for cyclic branched cover<strong>in</strong>gs <strong>of</strong> M along K ([2]), surgery formulas alongclaspers([3, 9]), simple cabl<strong>in</strong>g formulas for degree 0 cabl<strong>in</strong>gs or toric cabl<strong>in</strong>gs ([11, 12]), and adegree <strong>in</strong>equality relat<strong>in</strong>g the degree <strong>of</strong> the 2-loop part to the genus <strong>of</strong> the knot when M is S 3([13]).A bad feature <strong>of</strong> this <strong><strong>in</strong>variant</strong> is its <strong>in</strong>tricate def<strong>in</strong>ition: to obta<strong>in</strong> the rational form <strong>of</strong> theKontsevich <strong>in</strong>tegral, one has to use a surgery presentation <strong>of</strong> the pair (M, K), and an <strong>equivariant</strong>version <strong>of</strong> the Kontsevich <strong>in</strong>tegral together with an <strong>equivariant</strong> version <strong>of</strong> formal gaussian<strong>in</strong>tegration.The attempt <strong>of</strong> this paper is to give a simpler, geometric def<strong>in</strong>ition <strong>of</strong> this <strong><strong>in</strong>variant</strong>. This isjust an attempt as we will construct a new <strong><strong>in</strong>variant</strong> and just conjecture that it co<strong>in</strong>cide with the2-loop part <strong>of</strong> the Kontsevich <strong>in</strong>tegral, but we will give good reasons to believe this conjecture.The idea for the def<strong>in</strong>ition <strong>of</strong> this <strong><strong>in</strong>variant</strong> is to go back to configuration space <strong>in</strong>tegrals.When construct<strong>in</strong>g the first non-trivial <strong><strong>in</strong>variant</strong> <strong>of</strong> a rational <strong>homology</strong> sphere M, i.e the <strong>Casson</strong><strong><strong>in</strong>variant</strong>, one can replace the <strong>in</strong>tegral <strong>of</strong> some differential form by an <strong>in</strong>tersection <strong>of</strong> 3 suitable4-cycles <strong>in</strong> the configuration space <strong>of</strong> 2 po<strong>in</strong>ts <strong>in</strong> M. This was expla<strong>in</strong>ed to us by C. Lescop([7]). We can adapt this construction to the <strong>equivariant</strong> sett<strong>in</strong>g, consider<strong>in</strong>g cycles with twistedcoefficients.In order to give an understandable plan <strong>of</strong> the paper, let us present briefly this construction:we will give it <strong>in</strong> much more details <strong>in</strong> the article. Let K be an oriented knot homologous to0 <strong>in</strong> a rational <strong>homology</strong> sphere M. The complement <strong>of</strong> a tubular neighborhood <strong>of</strong> K <strong>in</strong> Mis a manifold N satisfy<strong>in</strong>g H 1 (N, Z) = Z. Let us fix a trivialization on N which is standard<strong>in</strong> some way on the boundary <strong>of</strong> N. Then, we consider the compactified configuration spaceC 2 (N) <strong>of</strong> 2 po<strong>in</strong>ts <strong>in</strong> N. If these po<strong>in</strong>ts are very close to each other or if one <strong>of</strong> them is <strong>in</strong> theboundary <strong>of</strong> N, we can def<strong>in</strong>e a vector <strong>in</strong> S 2 which is <strong>in</strong> some sense the direction from the firstpo<strong>in</strong>t to the second. For almost any po<strong>in</strong>t <strong>in</strong> S 2 , the preimage <strong>of</strong> this po<strong>in</strong>t is a 3-cycle <strong>in</strong> theboundary <strong>of</strong> C 2 (N) with twisted coefficents which bounds a 4-cycle <strong>in</strong> C 2 (N). By <strong>in</strong>tersect<strong>in</strong>gthree such twisted 4-cycles we obta<strong>in</strong> an object which we identify with a 2-loop diagram. Wenote it <strong>Casson</strong>(M, K). More precisely, we obta<strong>in</strong> the follow<strong>in</strong>g theorem.Let (M, K) be a pair with Alexander polynomial ∆. We note A = ⊗ 3Z[t ±1 ] Z[t±1 ,1∆(t) ] wherethe r<strong>in</strong>g Z[t ±1 ] acts diagonally on the tensor product. The group S 3 acts on A by permut<strong>in</strong>gthe factors and the group Z/2Z acts by replac<strong>in</strong>g t by t −1 simultaneously <strong>in</strong> each term <strong>of</strong> thetensor product. Let Res r : A → Q be the map def<strong>in</strong>ed by the formula below for all <strong>in</strong>tegers rsuch that ∆ does not vanish on r-th roots <strong>of</strong> unity:Res r f ⊗ g ⊗ h = 1 r∑ω r =η r =δ r =ωηδ=1f(ω)g(η)h(δ).Theorem. For all pairs (M, K) there is a well-def<strong>in</strong>ed element <strong>of</strong> A obta<strong>in</strong>ed as before andnoted <strong>Casson</strong>(M, K). This element is <strong><strong>in</strong>variant</strong> through the action <strong>of</strong> S 3 × Z/2 and conta<strong>in</strong>s the<strong>Casson</strong> <strong><strong>in</strong>variant</strong> <strong>of</strong> the cyclic branched covers <strong>of</strong> M along K <strong>of</strong> order r noted Σ r (M, K) <strong>in</strong> thefollow<strong>in</strong>g way.λ(Σ r (M, K)) = 1 8 σ r(M, K) + 1 6 Res r <strong>Casson</strong>(M, K).In this formula, σ r (M, K) is the sum <strong>of</strong> the <strong>equivariant</strong> signatures <strong>of</strong> (M, K) at r roots <strong>of</strong> unity.2


Remark 1.1. • Such a formula were proven <strong>in</strong> [2] with the <strong><strong>in</strong>variant</strong> <strong>Casson</strong>(M, K) replacedby z2 rat (M, K). This is a good reason to believe that we have <strong>in</strong>deed the identity<strong>Casson</strong>(M, K) = z2 rat (M, K) but this is not a pro<strong>of</strong> because there are non-trivial 2-loopdiagrams hav<strong>in</strong>g zeros r-residues for all r, for <strong>in</strong>stance t 2 ⊗ t ⊗ 1 − t ⊗ 1 ⊗ 1 ∈ A.• The motivation <strong>of</strong> this construction was to give a simple, geometric construction <strong>of</strong> the2-loop part <strong>of</strong> the Kontsevich <strong>in</strong>tegral: as a bonus, our construction has another advantage:the <strong><strong>in</strong>variant</strong> <strong>Casson</strong>(M, K) has obviously “<strong>in</strong>tegral” coefficients, which is not clear for the2-loop part <strong>of</strong> the Kontsevich <strong>in</strong>tegral. For <strong>in</strong>stance, if we can identify <strong>Casson</strong>(M, K) with(M, K), we will prove Rozansky’s <strong>in</strong>tegrality conjecture about 2-loop polynomial (see[14]).z rat2Aknowledgments: I would like to thank S. Garoufalidis, C. Lescop, G. Masbaum, T. Ohtsuki,and P. Vogel for <strong>in</strong>spir<strong>in</strong>g many ideas <strong>of</strong> this article and for stimulat<strong>in</strong>g discussions.Contents1 Introduction 12 Construction <strong>of</strong> the <strong>equivariant</strong> <strong>Casson</strong> <strong><strong>in</strong>variant</strong> 32.1 Configuration space <strong>of</strong> 2 po<strong>in</strong>ts and its boundary . . . . . . . . . . . . . . . . . . 32.2 Def<strong>in</strong>ition <strong>of</strong> the cycle Σ a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42.3 The cycle Σ a bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62.4 Construction <strong>of</strong> the <strong><strong>in</strong>variant</strong> . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.5 About trivializations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 Recover<strong>in</strong>g <strong>Casson</strong> <strong><strong>in</strong>variant</strong>s 113.1 Forgett<strong>in</strong>g coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113.2 The lift map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113.3 A formula for <strong>Casson</strong>(Σ r (M, K)) . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 Construction <strong>of</strong> the <strong>equivariant</strong> <strong>Casson</strong> <strong><strong>in</strong>variant</strong>2.1 Configuration space <strong>of</strong> 2 po<strong>in</strong>ts and its boundaryLet us fix once for all an oriented banded knot K homologous to 0 <strong>in</strong> a rational <strong>homology</strong> sphereM. We suppose that K has self-l<strong>in</strong>k<strong>in</strong>g 0. By blow<strong>in</strong>g up the knot <strong>in</strong> M, we obta<strong>in</strong> a 3-manifoldN whose boundary is identified with the unit normal bundle <strong>of</strong> K <strong>in</strong> M. Imitat<strong>in</strong>g the article[7], we want to identify a neigbourhood <strong>of</strong> the boundary <strong>of</strong> N with some subset <strong>of</strong> R 3 .We will describe this subset very explicitely: let S 1 be the unit circle <strong>of</strong> C embedded <strong>in</strong> thestandard way <strong>in</strong> C × R = R 3 , Z be the axis {0} × R and C be the circle Z ∪ {∞} <strong>in</strong> R 3 ∪ {∞}.Consider the map θ : R 3 \ Z → [0, 1) def<strong>in</strong>ed <strong>in</strong> the follow<strong>in</strong>g way: any po<strong>in</strong>t M <strong>in</strong> R 3 \ Z hasa unique projection onto the circle S 1 which we call f(M). Then, we set θ(M) = ||M−f(M)||||M+f(M)||3


where || · || denotes the euclidian norm. This map extends cont<strong>in</strong>uously to C by sett<strong>in</strong>g θ(x) = 1for all x <strong>in</strong> C. Moreover the level surfaces <strong>of</strong> θ are tori for parameters <strong>in</strong> (0, 1), and we haveS 1 = θ −1 (0) and C = θ −1 (1).Let U be the solid torus θ −1 ([ 1 4 , 1]) <strong>in</strong> S3 and ϕ be an orientation preserv<strong>in</strong>g embedd<strong>in</strong>g <strong>of</strong> U<strong>in</strong> M mapp<strong>in</strong>g C to K such that a parallel copy <strong>of</strong> C is sent to the prefered longitude <strong>of</strong> K.Then we haveM = U ∪ θ −1 ([ 1 4 , 3 4 ]) T M,Kwhere we have set T M,K = M \ ϕ(θ −1 (( 3 4 , 1])). The manifold T M,K is isomorphic to N and thechoice <strong>of</strong> ϕ is unique up to isotopy. We will say that ϕ is a “good” embedd<strong>in</strong>g. Our furtherconstructions are isotopy <strong><strong>in</strong>variant</strong>s, hence the precise choice <strong>of</strong> good embedd<strong>in</strong>g does not matter.Def<strong>in</strong>ition 2.1. Let us fix a pair (M, K) and a good embedd<strong>in</strong>g ϕ : U → M. A trivialization <strong>of</strong>M \ K standard along the knot K is a trivialization <strong>of</strong> M \ K which co<strong>in</strong>cide with the standardtrivialization <strong>of</strong> R 3 on ϕ(U \ C). In formulas, we have τ| ϕ(U\C) = ϕ ∗ τ R 3.In section 2.7, we will show that such a trivialization always exists.A direct computation shows that we have H 1 (M \K, Z) = Z. The generator <strong>of</strong> this co<strong>homology</strong>group is the l<strong>in</strong>k<strong>in</strong>g number with K and hence depends on an orientation <strong>of</strong> K. It is representedby a map f : M \ K → S 1 which is unique up to homotopy. If we choose a good embedd<strong>in</strong>gϕ : U → M, we can suppose that the restriction <strong>of</strong> f ◦ ϕ to U \ C is exactly the previousprojection map f from R 3 \ Z to S 1 .As the pair (M, K) is fixed once for all, we simplify the notations and set T = T M,K . Let ∆ bethe diagonal <strong>in</strong> T × T . We call C 2 (T ) the manifold with corners obta<strong>in</strong>ed by blow<strong>in</strong>g up ∆. Letus fix a trivialization τ standard along the knot K. We construct a new map p : ∂C 2 (T ) → S 2 .The set ∂T is identified with the level surface θ −1 ( 3 4 ). The boundary <strong>of</strong> C 2(T ) is decomposed<strong>in</strong>to three manifolds: the unit normal bundle <strong>of</strong> ∆ noted U∆, the spaces T × ∂T and ∂T × T .• For y ∈ ∂T and x ∈ T , we set x ′ = (1 − ε(θ(x)))f(x) + ε(θ(x))x where ε is a smooth mapfrom [0, 1] to itself which is equal to 0 on [0, 1 3 ] and to 1 on [ 2 3, 1]. In that way, the po<strong>in</strong>tx ′ always belong to R 3 .Then, we set p(x, y) =y−x′||y−x ′ ||and p(y, x) =• For a po<strong>in</strong>t (x, v, w) <strong>in</strong> U∆, we setx′ −y||x ′ −y|| .p(x, v, w) = τ x(w − v)||τ x (w − v)|| .Lemma 2.2. The map p : ∂C 2 (T ) → S 2 is well-def<strong>in</strong>ed and cont<strong>in</strong>uous.Pro<strong>of</strong>. The pro<strong>of</strong> is clear as the 3 def<strong>in</strong>itions <strong>of</strong> p co<strong>in</strong>cide when the arguments are both closeto the boundary <strong>of</strong> T . The figure 1 may help visualiz<strong>in</strong>g the map p.2.2 Def<strong>in</strong>ition <strong>of</strong> the cycle Σ aIn this section, we use the map p to def<strong>in</strong>e twisted 3-cycles <strong>in</strong> ∂C 2 (T ).4


UZ∂TS 1CFigure 1: Identification <strong>of</strong> the boundary <strong>of</strong> the knotThe map f × f gives a map from C 2 (T ) to S 1 × S 1 . Hence, <strong>in</strong> this section, we will considerer<strong>homology</strong> with coefficients <strong>in</strong> a Z[t ±11 , t±1 2 ]-module. Let E be the Z[t±1 1 , t±1 2 ]-moduleZ[t ±11 , t±1 2 ]/(t 1t 2 − 1). It will be helpful to keep <strong>in</strong> m<strong>in</strong>d the follow<strong>in</strong>g <strong>in</strong>terpretation: an element<strong>of</strong> the cover<strong>in</strong>g space <strong>of</strong> C 2 (T ) associated to E is a couple po<strong>in</strong>ts (x, y) <strong>in</strong> T with a <strong>homology</strong>class <strong>of</strong> path from f(x) to f(y) <strong>in</strong> S 1 .For all regular values a <strong>of</strong> p <strong>in</strong> S 2 , we have a codimension 2 submanifold Σ a <strong>of</strong> ∂C 2 (T ). We wantto show that for some values <strong>of</strong> a, the submanifold Σ a is a 3-cycle <strong>in</strong> ∂C 2 (T ) with coefficients<strong>in</strong> E.Consider the cover<strong>in</strong>g π : S 1 ×R → S 1 ×S 1 def<strong>in</strong>ed by π(x, t) = (x, exp(it)x). The submanifoldΣ a is an element <strong>of</strong> H 3 (∂C 2 (T ), E) if and only if it lifts to this cover<strong>in</strong>g, <strong>in</strong> other words, thereis a map ˜f which makes the follow<strong>in</strong>g diagram commute: f×fΣ a efS 1 × RπS 1 × S 1Lemma 2.3. Let n be the vector (0, 0, 1) ∈ S 2 ⊂ R 3 . There is a disc D around n <strong>in</strong> S 2 suchthat for any regular value a <strong>of</strong> p <strong>in</strong> D ∪ −D, the map f : Σ a → S 1 × S 1 lifts through π.Pro<strong>of</strong>. Fix a regular value a close to n. Suppose we have some element (x, y) ∈ Σ a . If x and yare very close, then f(x) and f(y) are very close and there is no difficulty to lift the diagonal<strong>of</strong> S 1 through π. But when x and y are not close, it means that p is close to the standard mapy−x||y−x||(x, y) ↦→ . Geometrically, we see that if the vector p(x, y) is close to n, then f(x) and f(y)are close to each other. It means that we can globally lift Σ a through π. Of course, <strong>in</strong> order toidentify Σ a with an element <strong>of</strong> H 3 (∂C 2 (T ), E), we have to specify which lift we choose, but thereis a prefered choice: as f(x) and f(y) are close to each other on the circle, we choose the pathfrom f(x) to f(y) which stays <strong>in</strong> the <strong>in</strong>terval [f(x), f(y)]. We see that the situation is exactlythe same close to −n, hence the lemma is proven.5


In that way, we obta<strong>in</strong> the follow<strong>in</strong>g short exact sequence.0 H 2 (C 2 (T )) H 3 (∂C 2 (T )) H 3 (C 2 (T )) 0∼ sH 3 (U∆, ∂)H 3 (U∆)(2)Our aim is to show that the cycle Σ a maps to 0 <strong>in</strong> H 3 (C 2 (T )). We will use for that the<strong>in</strong>tersection product 〈, 〉 : H 3 (∂C 2 (T )) ⊗ H 3 (∂C 2 (T )) → H 1 (∂C 2 (T ), F ⊗ Z F ) p 1∗→ H 1 (T, Z) ⊗ Λ.Thanks to the exact sequence (2) and the section s, the space H 3 (∂C 2 (T )) is naturally splitted<strong>in</strong> two parts noted H 2 (C 2 (T )) and H 3 (U∆). For dimensional reasons, the <strong>in</strong>tersection formrestricted to these spaces is 0. Let us exam<strong>in</strong>e the pair<strong>in</strong>g between H 2 (C 2 (T )) and H 3 (U∆).We have seen that the Z factor <strong>of</strong> the sequence (1) is represented by the product <strong>of</strong> two disjo<strong>in</strong>tSeifert surfaces. Geometrically, this class cannot <strong>in</strong>tersect with H 3 (U∆). This proves that thepair<strong>in</strong>g only depends on the image <strong>of</strong> the first factor <strong>in</strong> H 3 (U∆, ∂U∆) = Λ. This pair<strong>in</strong>g isdef<strong>in</strong>ed by the formula 〈1, α〉 = α.Let us expla<strong>in</strong> how Σ a appears <strong>in</strong> this decomposition: by def<strong>in</strong>ition <strong>of</strong> Σ a as a preimage <strong>of</strong>a by p, it must have self-<strong>in</strong>tersection 0. Suppose Σ a maps to f + x. Then, f is obta<strong>in</strong>ed by<strong>in</strong>tersect<strong>in</strong>g Σ a on a small sphere centered at x ∈ T , hence we have f = 1. The self-<strong>in</strong>tersection<strong>of</strong> Σ is then 2x which must be 0.This proves that Σ a bounds a 4-cycle <strong>in</strong> H 4 (C 2 (T ), ∂C 2 (T ); F ). This cycle is unique because<strong>of</strong> the vanish<strong>in</strong>g <strong>of</strong> H 4 (C 2 (T )).2.4 Construction <strong>of</strong> the <strong><strong>in</strong>variant</strong>Let D be the disc <strong>of</strong> lemma 2.3 and a, b, c be three regular values <strong>of</strong> p <strong>in</strong> D ∪ −D. Then, Σ a , Σ band Σ c are three disjo<strong>in</strong>t 3-cycles <strong>in</strong> H 3 (∂C 2 (T ), F ). Each 3-cycle bounds a 4-cycle ∆ a , ∆ b , ∆ c<strong>in</strong> C 2 (T ), unique up to a 5-boundary. As they have disjo<strong>in</strong>t boundaries, we can form their<strong>in</strong>tersection. We get an element <strong>of</strong> H 0 (C 2 (T ), ⊗ 3Z F ). We compute H 0(C 2 (T ), ⊗ 3Z F ) = ⊗ 3A Λ.In the last tensor product, the action <strong>of</strong> A is supposed to be diagonal.For convenience, let us give a geometric <strong>in</strong>terpretation <strong>of</strong> this <strong>in</strong>tersection. Let a, b, c be threeregular values <strong>of</strong> p <strong>in</strong> D, Σ i and ∆ i the correspond<strong>in</strong>g cycles. Let f : T → S 1 be the usualrepresentant <strong>of</strong> the generator <strong>of</strong> H 1 (T, Z). A po<strong>in</strong>t <strong>in</strong> the <strong>in</strong>terior <strong>of</strong> C 2 (T ) is represented bya 4-tuple (x, y, γ, g) where x and y are two dist<strong>in</strong>ct po<strong>in</strong>ts <strong>of</strong> T , γ is a homotopy class <strong>of</strong> path<strong>in</strong> S 1 from f(x) to f(y) and g is an element <strong>of</strong> Λ. We identify the 4-tuple (x, y, tγ, g) with the4-tuple (x, y, γ, tg) where tγ is the path γ with an extra loop around the circle.Then, at each triple <strong>in</strong>tersection po<strong>in</strong>t <strong>of</strong> ∆ a , ∆ b and ∆ c , we get three 4-tuples such as(x, y, γ a , g a ), (x, y, γ b , g b ) and (x, y, γ c , g c ). By glu<strong>in</strong>g these three paths <strong>in</strong> x and y we obta<strong>in</strong>a graph Θ with edges decorated by elements <strong>of</strong> Λ. More formally, we obta<strong>in</strong> an element <strong>of</strong>Λ. We will call A this space.⊗ 3ALemma 2.5. The element ∆ a ∩∆ b ∩∆ c is <strong>in</strong>dependent <strong>of</strong> the choice <strong>of</strong> regular values <strong>in</strong> D∪−D.It only depends on the pair (M, K) and the homotopy class <strong>of</strong> trivialization τ and we note it<strong>Casson</strong>(M, K, τ). Let S be the automorphism <strong>of</strong> Λ send<strong>in</strong>g t on t −1 . The group G = Z/2Z × S 3acts on A by the formula S k σ · x 1 ⊗ x 2 ⊗ x 3 = S k x σ(1) ⊗ S k x σ(2) ⊗ S k x σ(3) . The element<strong>Casson</strong>(M, K, τ) is G-<strong><strong>in</strong>variant</strong>.7


Pro<strong>of</strong>. Suppose we have two triples <strong>of</strong> regular values <strong>of</strong> p <strong>in</strong> D ∪ −D, namely a, b, c and a ′ , b ′ , c ′ .Associated to these values, there are 6 3-cycles Σ a , Σ b , Σ c and Σ ′ a, Σ ′ b , Σ′ c. We note as usual∆ a , ∆ b , ∆ c the 4-cycles bound<strong>in</strong>g the Σ i <strong>in</strong> C 2 (T ). The idea <strong>of</strong> the pro<strong>of</strong> is to def<strong>in</strong>e cycles ∆ ′ iby add<strong>in</strong>g cobordisms W i between ∆ i and ∆ ′ i <strong>in</strong> ∂C 2(T ) × [0, 1]. This is possible if and only ifthey represent the same <strong>homology</strong> class <strong>in</strong> H 3 (∂C 2 (T ), F ). From the lemma 2.4, we now thatall cycles Σ bound <strong>in</strong> C 2 (T ) and hence come from H 4 (C 2 (T ), ∂C 2 (T ); F ). In the exact sequence(1), we have seen that all cycles Σ <strong>in</strong> the middle term map to 1, hence, they all differ by amultiple <strong>of</strong> the generator η <strong>of</strong> the first space, that is a product <strong>of</strong> two disjo<strong>in</strong>t Seifert surfaces.Such a multiple has to be 0 for the follow<strong>in</strong>g reason: consider a parallel <strong>in</strong> the boundary <strong>of</strong> T ,and γ the surface made <strong>of</strong> pair <strong>of</strong> po<strong>in</strong>ts ly<strong>in</strong>g on the parallel. We check immediately that η<strong>in</strong>tersect γ <strong>in</strong> precisely one po<strong>in</strong>t whereas the cycles Σ never <strong>in</strong>tersect γ as they are made <strong>of</strong>pairs <strong>of</strong> po<strong>in</strong>ts <strong>in</strong> almost vertical position. This f<strong>in</strong>ally proves that all cycles Σ are homologousto each other.Let W i be a cobordism between the Σ i and the Σ ′ i . The difference between the <strong>in</strong>tersectioncomputed from the Σ ′ i and from the Σ i is equal to the triple <strong>in</strong>tersection <strong>of</strong> the W i . But we candecompose the <strong>in</strong>tersection W a ∩ W b ∩ W c <strong>in</strong> the follow<strong>in</strong>g sum: W a ∩ Σ b × [0, 1] ∩ Σ c × [0, 1] +Σ ′ a × [0, 1] ∩ W b ∩ Σ c × [0, 1] + Σ ′ a × [0, 1] ∩ Σ ′ b × [0, 1] ∩ W c. As all cycles Σ are disjo<strong>in</strong>t, be<strong>in</strong>gpreimages <strong>of</strong> dist<strong>in</strong>ct po<strong>in</strong>ts <strong>of</strong> S 2 , we see that this triple <strong>in</strong>tersection vanishes.We can now conclude the lemma: as the construction <strong>of</strong> the <strong><strong>in</strong>variant</strong> <strong>Casson</strong>(M, K, τ) doesnot depend on the choice <strong>of</strong> regular values <strong>in</strong> D, it only depends on the homotopy class <strong>of</strong> τ,hence <strong>Casson</strong>(M, K, τ) is actually an <strong><strong>in</strong>variant</strong> <strong>of</strong> the triple (M, K, τ).It is clear that the element <strong>Casson</strong>(M, K, τ) is S 3 <strong><strong>in</strong>variant</strong> because the action <strong>of</strong> the permutationcorresponds to a permutation <strong>of</strong> the regular values a, b, c and the <strong><strong>in</strong>variant</strong> does not depend<strong>of</strong> the choice <strong>of</strong> a triple.We use the same idea to prove the S-<strong>in</strong>variance <strong>of</strong> <strong>Casson</strong>(M, K, τ).Then, let us consider the automorphism S <strong>of</strong> C 2 (T ) send<strong>in</strong>g (x, y) to (y, x). This automorphismchanges orientation. We check that p(S(x, y)) = p(y, x) = −p(x, y) on the boundary. Whenwe compute the <strong>Casson</strong> <strong><strong>in</strong>variant</strong> by tak<strong>in</strong>g po<strong>in</strong>ts close to −n <strong>in</strong>stead <strong>of</strong> n, we change thepo<strong>in</strong>ts a, b, c to −a, −b, −c and then the cycles <strong>in</strong>to SΣ a , SΣ b , SΣ c with opposite orientations.As we have seen above, the <strong>in</strong>tersection rema<strong>in</strong>s the same, hence we have ∆ a ∩ ∆ b ∩ ∆ c =S∆ a ∩ S∆ b ∩ S∆ c . The map S changes the orientation <strong>of</strong> the cycles and <strong>of</strong> the ambient space,hence it does not change the sign <strong>of</strong> the <strong>in</strong>tersection. Us<strong>in</strong>g the map S act<strong>in</strong>g on A, we obta<strong>in</strong>S <strong>Casson</strong>(M, K, τ) = <strong>Casson</strong>(M, K, τ). We have f<strong>in</strong>ally proven that <strong>Casson</strong>(M, K, τ) is welldef<strong>in</strong>edand G-<strong><strong>in</strong>variant</strong>.2.5 About trivializationsThe def<strong>in</strong>ition <strong>of</strong> <strong>Casson</strong>(M, K, τ) depends on the choice <strong>of</strong> the trivialization τ up to homotopy.In this part, we will adapt many considerations <strong>of</strong> [7].First, let us show that such a trivialization always exists. <strong>An</strong>y 3-manifold has a trivialization,hence let us fix an arbitrary one τ : T M → R 3 respect<strong>in</strong>g orientations. We consider themonodromy map π 1 (M) → π 1 (SO(3)) = Z/2. This map is a group homomorphism with values<strong>in</strong> a commutative group, hence it factors through H 1 (M, Z). As the knot K is supposed to behomologous to 0, its monodromy has to be trivial.8


Let us exam<strong>in</strong>e this trivialization <strong>in</strong> the decomposition M = U ∪ T . It differs from thestandard trivialization <strong>of</strong> U \ ∞ by composition with a map G : U \ ∞ → SO(3). Moreover,both trivializations have trivial monodromies along a generator <strong>of</strong> π 1 (U \ ∞), hence the mapG <strong>in</strong>duces a trivial map on fundamental groups. As π 2 (SO(3)) = 0 and U is homotopic to aglu<strong>in</strong>g <strong>of</strong> S 1 and S 2 <strong>in</strong> one po<strong>in</strong>t, the map G is contractible. This implies that we can deformthe trivialization τ on U \ ∞ such that it co<strong>in</strong>cides with the standard trivialization <strong>of</strong> R 3 , whichis the def<strong>in</strong>ition <strong>of</strong> a trivialization <strong>of</strong> M \ K standard along the knot.Next, fix the pair (M, K) and the good diffeomorphism ϕ : U → M. Two trivializations <strong>of</strong>T standard along the knot differ by a map G : (T, ∂T ) → (SO(3), 1). Let us consider the setH <strong>of</strong> all homotopy class <strong>of</strong> maps from (M, K) to (SO(3), 1). This is a group us<strong>in</strong>g the productstructure <strong>of</strong> SO(3).By adapt<strong>in</strong>g results <strong>of</strong> [7], lemma 2.32 p.28, we obta<strong>in</strong> the follow<strong>in</strong>g lemma:Lemma 2.6. The group H is abelian and the map deg : H ⊗ Q → Q is an isomorphism.Pro<strong>of</strong>. Let G be an element <strong>of</strong> H, and consider the map <strong>in</strong>duced by G on fundamental groups.It gives an element <strong>of</strong> hom(π 1 (M, K), Z/2) = H 1 (M, K; Z/2). Moreover, the map H →H 1 (M, K; Z/2) is a group homomorphism. This map fits <strong>in</strong> the follow<strong>in</strong>g exact sequence:deg0 Z iH H 1 (M, K; Z/2) 0The element ρ = i(1) is constructed <strong>in</strong> the follow<strong>in</strong>g way: choose a po<strong>in</strong>t x <strong>in</strong>side T and fixa small ball B around x. Identify<strong>in</strong>g B with the complement <strong>of</strong> 1 <strong>in</strong> S 3 , we deduce from thestandard cover<strong>in</strong>g map S 3 → SO(3) a map B → SO(3) send<strong>in</strong>g the boundary <strong>of</strong> B on 1 andhav<strong>in</strong>g degree 2. By extend<strong>in</strong>g this map on T by one, we obta<strong>in</strong> a element <strong>of</strong> H called ρ <strong>of</strong>degree 2. Hence, the map deg ◦i is the multiplication by 2. The fact that the above sequence isexact is a consequence <strong>of</strong> the cell decomposition <strong>of</strong> the pair (M, K).The subgroup [H, H] maps to 0 on H 1 (M, K; Z/2) and is a subgroup <strong>of</strong> Z <strong>of</strong> degree 0. Hence,it is 0 and H is a commutative group. This f<strong>in</strong>ally proves that the map deg2: H ⊗ Q → Q is anisomorphism.We will use the standard map p 1 from the set <strong>of</strong> trivializations <strong>of</strong> rational <strong>homology</strong> ballsto Z given <strong>in</strong> [7]. Here, we def<strong>in</strong>e p 1 (M, K, τ) as the p 1 <strong><strong>in</strong>variant</strong> <strong>of</strong> the trivialization τ onB = T ∪ (U ∩ B(0, 2)), where B(0, 2) is a closed ball <strong>of</strong> radius 2 centered at 0.Briefly, it is def<strong>in</strong>ed <strong>in</strong> the follow<strong>in</strong>g way: fix a 4-manifold W bound<strong>in</strong>g B × {0} ∪ B(0, 2) ×{1} ∪ ∂B(0, 2) × [0, 1]. Us<strong>in</strong>g the trivialization τ <strong>of</strong> M together with the tangent vector <strong>of</strong> thecomponent [0, 1], we can construct a trivialization <strong>of</strong> the tangent space <strong>of</strong> W restricted to itsboundary. The obstruction <strong>of</strong> extend<strong>in</strong>g the complexified trivialization <strong>of</strong> T W ⊗ C to W is bydef<strong>in</strong>ition p 1 (W, τ)[W, ∂W ]. We set p 1 (τ) = p 1 (W, τ) − 3σ(W ). This <strong>in</strong>teger does not depend onW .We will use the follow<strong>in</strong>g identity proved <strong>in</strong> [7]:Fix a triple (M, K, τ): for all maps G : (T, ∂T ) → (SO(3), 1), we have p 1 (M, K, Gτ) −p 1 (M, K, τ) = −2 deg(G).Our next task is to show that the same k<strong>in</strong>d <strong>of</strong> formula appears for the <strong><strong>in</strong>variant</strong> <strong>Casson</strong>(M, K, τ).Precisely, we will show the follow<strong>in</strong>g lemma.9


Lemma 2.7. Let Θ be the element 1 ⊗ 1 ⊗ 1 ∈ A. Fix a triple (M, K, τ) and map G : (T, ∂T ) →(SO(3), 1), we have <strong>Casson</strong>(M, K, Gτ) − <strong>Casson</strong>(M, K, τ) = − 1 2deg GΘ.Pro<strong>of</strong>. Fix a triple (M, K, τ). First, we show that the map from H to A def<strong>in</strong>ed by G ↦→<strong>Casson</strong>(M, K, Gτ) − <strong>Casson</strong>(M, K, τ) is a group homomorphism. Let G be an element <strong>of</strong> H, Σ aa cycle <strong>in</strong> H 3 (∂C 2 (T ), F ) associated to the trivialization τ and Σ ′ a the cycle associated to thetrivialization Gτ. We remark that these cycles are homologous. Indeed, as they differ only onthe submanifold U∆ <strong>of</strong> ∂C 2 (T ), their difference is <strong>in</strong> H 3 (U∆, F ). But they both map to 0 <strong>in</strong>H 3 (C 2 (T ), F ) which is isomorphic to H 3 (∂C 2 (T ), F ) through the <strong>in</strong>clusion. This f<strong>in</strong>ally impliesthat they are homologous.We can compute the <strong><strong>in</strong>variant</strong> <strong>Casson</strong>(M, K, Gτ) <strong>in</strong> the follow<strong>in</strong>g way. Consider the manifoldC 2 + (T ) obta<strong>in</strong>ed by thicken<strong>in</strong>g the component U∆ <strong>of</strong> the boundary <strong>of</strong> C 2(T ). Precisely,we have C 2 + (T ) = C 2(T ) ∪ U∆ × [0, 1] where we identify the manifold U∆ <strong>in</strong> the first factorwith U∆ × {0} and for all x ∈ ∂U∆ we identifiy {x} × [0, 1] with {x}. This new manifoldis obviously diffeomorphic to C 2 (T ). We call ∂ 1 C 2 + (T ) and ∂ 2C 2 + (T ) the submanifolds correspond<strong>in</strong>grespectively to U∆ × {0} and U∆ × {1}. We consider three cycles Σ a , Σ b , Σ c as ly<strong>in</strong>g<strong>in</strong> ∂ 1 C 2 + (T ) and Σ′ a, Σ ′ b , Σ′ c as ly<strong>in</strong>g <strong>in</strong> ∂ 2 + C 2(T ). As Σ i is homologous to Σ ′ i , we can f<strong>in</strong>d threecobordisms Π i <strong>in</strong> U∆×[0, 1] from Σ i to Σ ′ i . Geometrically, this implies that we have the identity<strong>Casson</strong>(M, K, Gτ)−<strong>Casson</strong>(M, K, τ) = Π a ·Π b ·Π c . Let us take another element G ′ <strong>of</strong> the groupH. We f<strong>in</strong>d some 4-cycles Π ′ i which def<strong>in</strong>e cobordisms from Σ i to Σ G′i , the cycles def<strong>in</strong>ed with G ′ τ.to Σ GG′iBy apply<strong>in</strong>g the map G to the manifold U∆×[0, 1], we obta<strong>in</strong> a cobordism GΠ ′ i from ΣG iwhich has the same triple <strong>in</strong>tersection. This proves that Π a ∪ GΠ ′ a · Π b ∪ GΠ ′ b · Π c ∪ GΠ ′ c is equalto Π a · Π b · Π c + Π ′ a · Π ′ b · Π′ c. Hence, the map send<strong>in</strong>g G to <strong>Casson</strong>(M, K, Gτ) − <strong>Casson</strong>(M, K, τ)is a group homomorphism.The above map take its values <strong>in</strong> A which has no torsion, therefore it is completely determ<strong>in</strong>edby its value on a generator <strong>of</strong> H ⊗ Q, for <strong>in</strong>stance the element ρ.Let us construct the cobordisms Π i <strong>in</strong> such a situation. As ρ has support <strong>in</strong> a small ball, wecan suppose that the cobordisms Π i are trivial except <strong>in</strong> that ball. Let B be the standard ball<strong>in</strong> R 3 with radius 2π. The map ρ from B to SO(3) sends a po<strong>in</strong>t x at distance d from the orig<strong>in</strong>to the rotation <strong>of</strong> axis x and angle d. Call x, y, z be the coord<strong>in</strong>ates <strong>of</strong> R 3 . We use the samenames for the correspond<strong>in</strong>g directions <strong>in</strong> S 2 . We have the follow<strong>in</strong>g situation: there are threecycles Σ x , Σ y and Σ z <strong>in</strong> B × S 2 which are respectively equal to B × {x}, B × {y} and B × {z}and three 3-cycles Σ ′ x, Σ ′ y and Σ ′ z obta<strong>in</strong>ed from the others by apply<strong>in</strong>g the map ρ. The cyclesΣ i and Σ ′ i co<strong>in</strong>cide on ∂B × S2 , and then def<strong>in</strong>e a cycle <strong>in</strong> ∂B × S 2 × [0, 1]. We know thereare cobordisms Π i from Σ i to Σ ′ i and we need to compute their <strong>in</strong>tersection. Let us add some<strong>in</strong>termediary steps:(Σ x , Σ y , Σ z ) → (Σ x , Σ y , Σ ′ z) → (Σ x , Σ ′ y, Σ ′ z) → (Σ ′ x, Σ ′ y, Σ ′ z)When two cycles are unchanged <strong>in</strong> some step, we use the trivial cobordism. As the cycles Σ iare disjo<strong>in</strong>t, and so are the cycles Σ ′ i , only the middle term may give some contribution. Hencewe need to compute the follow<strong>in</strong>g <strong>in</strong>tersection Σ x × [0, 1] ∩ Π y ∩ Σ ′ z × [0, 1].From the def<strong>in</strong>tion <strong>of</strong> ρ, we see that Σ x ∩ Σ ′ z is equal to C × {x} where C is a circle <strong>in</strong> Bly<strong>in</strong>g <strong>in</strong> the plane x = z. We also compute that Σ ′ y ∩ Σ x is equal to D × {x} where D is a circlely<strong>in</strong>g <strong>in</strong> the plane x = y. Hence, generically, Σ x × [0, 1] ∩ Π y is a 2-cycle <strong>in</strong> B × {x} × [0, 1]bound<strong>in</strong>g D × {x} × {1}. The triple <strong>in</strong>tersection that we have computed is hence the l<strong>in</strong>k<strong>in</strong>gnumber between C and D which is -1.10


From the lemma 2.7, we obta<strong>in</strong> the follow<strong>in</strong>g proposition:Proposition 2.8. Let (M, K) be a pair and τ be a trivialization standard along the knot. Thequantity <strong>Casson</strong>(M, K) = <strong>Casson</strong>(M, K, τ) − 1 4 p 1(τ)Θ ∈ A does not depend on the choice <strong>of</strong>trivialization τ.3 Recover<strong>in</strong>g <strong>Casson</strong> <strong><strong>in</strong>variant</strong>sFrom now we have not shown anyth<strong>in</strong>g about the non-triviality <strong>of</strong> the <strong><strong>in</strong>variant</strong> <strong>Casson</strong>(M, K).This will come from the fact that our construction is a generalization <strong>of</strong> Lescop’s construction.3.1 Forgett<strong>in</strong>g coefficientsWe recall that for a pair (M, K) with normalized Alexander polynomial ∆, we have constructedthe <strong><strong>in</strong>variant</strong> <strong>Casson</strong>(M, K) as an element <strong>of</strong> A = ⊗ 3Z[t ±1 ] Z[t±1 , 1 ∆]. The evaluation at 1 fromZ[t ±1 , 1 ∆ ] to Z[ 1∆(1) ] <strong>in</strong>duces a map Res 11 : A → Z[∆(1)] called evaluation.Proposition 3.1. For all pairs (M, K), we have Res 1 <strong>Casson</strong>(M, K) = 6 <strong>Casson</strong>(M).Pro<strong>of</strong>. This is a consequence <strong>of</strong> Lescops articles [7, 8]. By surgery arguments, she showed thatthe first term <strong>of</strong> the <strong><strong>in</strong>variant</strong> Z(M) she constructs is equal to 1 2λ(M) where λ is the <strong>Casson</strong><strong><strong>in</strong>variant</strong>. When we forget coefficients, our construction reduces strictly to her construction, upto the order <strong>of</strong> the automorphism group <strong>of</strong> the graph C which is 12. This proves the assertion.Hence, the <strong><strong>in</strong>variant</strong> <strong>Casson</strong>(M, K) conta<strong>in</strong>s at least the <strong>Casson</strong> <strong><strong>in</strong>variant</strong> <strong>of</strong> the underly<strong>in</strong>gmanifold. We will show <strong>in</strong> this section a formula relat<strong>in</strong>g the <strong>Casson</strong> <strong><strong>in</strong>variant</strong> <strong>of</strong> a pair (M, K)and the <strong>Casson</strong> <strong><strong>in</strong>variant</strong> <strong>of</strong> its cyclic branched cover<strong>in</strong>gs. It will prove that <strong>Casson</strong>(M, K) andz 2 (M, K) are both generat<strong>in</strong>g functions for the <strong>Casson</strong> <strong><strong>in</strong>variant</strong> <strong>of</strong> cyclic branched cover<strong>in</strong>gs <strong>of</strong>M along K. This gives a good motivation to believe that these <strong><strong>in</strong>variant</strong>s are actually equal.3.2 The lift mapLet us give an algebraic def<strong>in</strong>ition <strong>of</strong> the map Lift r given <strong>in</strong> [2].Consider a pair (M, K) with Alexander polynomial ∆. For any <strong>in</strong>teger r, we can considerthe cyclic branched cover<strong>in</strong>g <strong>of</strong> M along K <strong>of</strong> order r. We note Σ r (M, K) the pair made <strong>of</strong>this manifold and the preimage <strong>of</strong> K. It may happen that the new manifold is not a rational<strong>homology</strong> sphere. This occurs exactly when the polynomial ∆ vanishes on some r-th root <strong>of</strong>unity.From now, we suppose that this situation does not occur: we will say that the pair (M, K) isr-regular. We can compute the Alexander polynomial <strong>of</strong> the new pair Σ r (M, K) by remark<strong>in</strong>gthat its Alexander module is precisely H 1 (M \ K, A) where we replace the action <strong>of</strong> t <strong>in</strong> A bythe action <strong>of</strong> θ = t r .We have the follow<strong>in</strong>g situation. The r<strong>in</strong>g A r = Z[θ ±1 ] is <strong>in</strong>cluded <strong>in</strong> A. By add<strong>in</strong>g a primitiver-th root <strong>of</strong> unity ξ <strong>in</strong> A, we obta<strong>in</strong> a Galois extension A r ⊂ A[ξ] with Galois group Z/r. Thismap extends to A r [ 1 ∆ r] → A[ξ, 1 ∆ ] where we have set ∆ r(θ) = N(∆) = ∏ i ∆(ξi t).11


Hence, the Alexander polynomial <strong>of</strong> Σ r (M, K) is just the norm <strong>of</strong> ∆.We can def<strong>in</strong>e the Lift r map easily <strong>in</strong> this sett<strong>in</strong>g. Let A be the space ⊗ 3A A[ 1 ∆] where A actsdiagonally on the three copies <strong>of</strong> A[ 1 ∆]. This space has a natural algebra structure given by theformula a ⊗ b ⊗ c · a ′ ⊗ b ′ ⊗ c ′ = aa ′ ⊗ bb ′ ⊗ cc ′ .By def<strong>in</strong><strong>in</strong>g A r = ⊗ 3A rA r [ 1 ∆ r] we obta<strong>in</strong> a new algebra with an <strong>in</strong>clusion map A r → A. Byadd<strong>in</strong>g the r-th root ξ to A, we obta<strong>in</strong> a Galois extension with Galois group (Z/r) 3 /(Z/r). Inthese notations, the map Lift r is just the map 1 r Tr : A → A r.Let us see how to compute it more concretely. Let x = P ∆ ⊗ Q ∆ ⊗ R ∆be an element <strong>of</strong> A. As thepolynomial ∆ r is the norm <strong>of</strong> ∆, there is a polynomial G ∈ A such that ∆ r = ∆G. We computex = GP∆ r⊗ GQ∆ r⊗ GR∆ r. The element ∆ r lies <strong>in</strong> A r and hence is not affected by the trace. Therema<strong>in</strong><strong>in</strong>g elements are comb<strong>in</strong>ation <strong>of</strong> monomials like t a ⊗ t b ⊗ t c and moreover, we can supposethat c is 0. The trace <strong>of</strong> such elements is precisely Tr t a ⊗ t b ⊗ 1 = ∑ i,j (ξi t) a ⊗ (ξ j t) b ⊗ 1 =r 2 δ r|a δ r|b t a ⊗ t b ⊗ 1 = r 2 θ a/r ⊗ θ b/r ⊗ 1. This formula gives us a direct way for comput<strong>in</strong>g theLift r map.3.3 A formula for <strong>Casson</strong>(Σ r (M, K))We prove <strong>in</strong> this section a formula relat<strong>in</strong>g the <strong>Casson</strong> <strong><strong>in</strong>variant</strong> <strong>of</strong> (M, K) and the <strong>Casson</strong><strong><strong>in</strong>variant</strong> <strong>of</strong> Σ r (M, K) which is completely analogous to the formula <strong>of</strong> [2].First, we recall that there is a well def<strong>in</strong>ed <strong><strong>in</strong>variant</strong> σ(M, K) : S 1 → Z called signature whichis a locally constant map jump<strong>in</strong>g at roots <strong>of</strong> the Alexander polynomial. For any <strong>in</strong>teger r, wewrite σ r (M, K) = ∑ ω r =1σ(M, K)(ω).Proposition 3.2. For any <strong>in</strong>teger r and any r-regular pair (M, K), we have<strong>Casson</strong>(Σ r (M, K)) = 3 4 σ r(M, K)Θ + Lift r <strong>Casson</strong>(M, K).Pro<strong>of</strong>. Fix an <strong>in</strong>teger r and a r-regular pair (M, K). We also choose a trivialization τ <strong>of</strong> Mstandard along the knot K. Let T be the manifold obta<strong>in</strong>ed as usual from (M, K) by remov<strong>in</strong>ga regular neighbourhood <strong>of</strong> the knot and T r the correspond<strong>in</strong>g manifold for Σ r (M, K). There isa cover<strong>in</strong>g map π r : Σ r (M, K) → (M, K). In the standard identifications <strong>of</strong> the boundaries <strong>of</strong> Tand T r with subsets <strong>of</strong> R 3 , we can suppose that π r is the map from C × R to itself def<strong>in</strong>ed by theformula π r (z, t) = (z r , t). The problem is that the trivialization π ∗ rτ is not standard along theknot <strong>of</strong> the pair Σ r (M, K). Precisely, the difference is described by a map from ∂T r to SO(3)with values <strong>in</strong> rotations around the z-axis. Let f be the map from T to S 1 used to construct themap p, by compos<strong>in</strong>g the trivialization π ∗ rτ with the rotation G(x) <strong>of</strong> angle −rf(x) around theaxis z, we obta<strong>in</strong> a trivialization <strong>of</strong> T r standard along the knot. We use it to def<strong>in</strong>e a map p r ,cycles Σ r i and ∆r i and then the <strong><strong>in</strong>variant</strong> <strong>Casson</strong>(Σr (M, K), Gπ ∗ rτ). We would like to comparethis <strong><strong>in</strong>variant</strong> with the follow<strong>in</strong>g one:Let ∆ a ′, ∆ b ′, ∆ c ′ be the twisted 4-cycles <strong>in</strong> C 2 (T ) whose triple <strong>in</strong>tersection give <strong>Casson</strong>(M, K, τ).We obta<strong>in</strong> correspond<strong>in</strong>g cycles <strong>in</strong> C 2 (T r ) by tak<strong>in</strong>g the preimage by π r × π r <strong>of</strong> the cycles ∆ i ′and <strong>in</strong>tersect<strong>in</strong>g them. We remark that although there is no projection from C 2 (T r ) to C 2 (T ),there is a well def<strong>in</strong>ed preimage <strong>of</strong> a twisted cycle. Let us call π ∗ r(Σ i ′) the cycles <strong>of</strong> ∂C 2 (T r )obta<strong>in</strong>ed <strong>in</strong> this lift<strong>in</strong>g procedure.As <strong>in</strong> lemma 2.5, to identify the <strong>in</strong>tersections ∆ r a ∩∆ r b ∩∆r c with π ∗ r(∆ a ′)∩π ∗ r(∆ b ′)∩π ∗ r(∆ c ′), itis sufficient to show that the cycles Σ r a, Σ r b , Σr c, π ∗ rΣ a ′, π ∗ rΣ b ′, π ∗ rΣ c ′ are homologous and disjo<strong>in</strong>t.12


It is clear that all these cycles are homologous. To ensure that they are disjo<strong>in</strong>t, it is sufficientto take regular values a, b, c, a ′ , b ′ , c ′ hav<strong>in</strong>g dist<strong>in</strong>ct scalar product with n, as we have the identityp(π r x, π r y) · n = p r (x, y) · n.We now compute <strong>Casson</strong>(Σ r (M, K), Gπ ∗ rτ) from the second def<strong>in</strong>ition.Let us look closely how the preimage map works on twisted 0-cycles <strong>of</strong> C 2 (T ): a 0-cycle is apair <strong>of</strong> dist<strong>in</strong>ct po<strong>in</strong>ts x, y <strong>in</strong> T together with a path from f(x) to f(y) <strong>in</strong> S 1 and a coefficient1∆(t) , where ∆(t) is the Alexander polynomial. As before, we note ∆ r the norm <strong>of</strong> ∆, which is apolynomial <strong>in</strong> θ = t r satisfy<strong>in</strong>g ∆ r = ∆G. The coefficient 1 ∆ may be written G ∆ r, and 1 ∆ ris fixedby the lift<strong>in</strong>g map, be<strong>in</strong>g a polynomial <strong>in</strong> θ. Once we have taken a preimage x r <strong>of</strong> x <strong>in</strong> T r , thepolynomial G gives a comb<strong>in</strong>ation <strong>of</strong> paths from f(x r ) to a po<strong>in</strong>t f(y r ) for some y r ∈ T r . Whenwe divide it by ∆ r , and take the sum over all choices <strong>of</strong> x r , we obta<strong>in</strong> precisely the lift map fromC 0 (C 2 (T ), F ) → C 0 (C 2 (T r ), F r ). It is now easy to describe what happens when we take a triple<strong>in</strong>tersection <strong>of</strong> the preimages <strong>of</strong> ∆ a , ∆ b and ∆ c . Fix a triple <strong>in</strong>tersection po<strong>in</strong>t <strong>of</strong> the ∆ i . Itcorresponds to two po<strong>in</strong>ts x, y <strong>in</strong> C 2 (T ) with three paths from f(x) to f(y) with coefficients. Weneed to do the lift operation for the three paths and to take the rema<strong>in</strong><strong>in</strong>g <strong>in</strong>tersection po<strong>in</strong>ts.It means that the po<strong>in</strong>t x r is the same for the three lifts (and we have r choices), and the targetpo<strong>in</strong>ts y r <strong>in</strong> T r need to co<strong>in</strong>cide. This is precidely the def<strong>in</strong>ition <strong>of</strong> the map Lift r .This prove the formula <strong>Casson</strong>(Σ r (M, K), π ∗ rτ) = Lift r <strong>Casson</strong>(M, K, τ). When we normalizethe trivializations, we have the follow<strong>in</strong>g correction factor:<strong>Casson</strong>(Σ r (M, K)) = − 1 4 (p 1(Gπ ∗ rτ) − rp 1 (τ)] + Lift r <strong>Casson</strong>(M, K)The equality p 1 (Gπ ∗ rτ) − rp 1 (τ) = 3σ r (M, K) is classical (see for <strong>in</strong>stance [4]). Let us sketchthe pro<strong>of</strong> briefly: fix a triple (M, K, τ) and a 4-manifold W bound<strong>in</strong>g M. We suppose that Whas signature 0, such that p 1 (τ) is exactly the obstruction <strong>of</strong> extend<strong>in</strong>g some trivialization <strong>of</strong>the complexified tangent bundle <strong>of</strong> M restricted to its boundary. Take a Seifert surface S <strong>of</strong> K<strong>in</strong> M, and push it <strong>in</strong>to W . By tak<strong>in</strong>g the cycl<strong>in</strong>g cover<strong>in</strong>g <strong>of</strong> W over S <strong>of</strong> order r, we obta<strong>in</strong>a manifold W r bound<strong>in</strong>g Σ r (M, K). The obstruction <strong>of</strong> extend<strong>in</strong>g the trivialization Gπ ∗ rτ ismultiplied by r, but the manifold W r has no longer signature 0 but σ r (M, K) (see [4]). Hence,from the def<strong>in</strong>ition <strong>of</strong> the p 1 <strong><strong>in</strong>variant</strong>, one has p 1 (Gπ ∗ rτ) = rp 1 (τ) − 3σ r (M, K). Hence thecorrection factor <strong>in</strong> the formula above is precisely 3 4 σ r(M, K) as announced.References[1] D. Bar-Natan and R. Lawrence. A rational surgery formula for the LMO <strong><strong>in</strong>variant</strong>.arXiv:math.GT/0007045, to appear <strong>in</strong> Israel J. Math.[2] S. Garoufalidis and A. Kricker. F<strong>in</strong>ite type <strong><strong>in</strong>variant</strong>s <strong>of</strong> cyclic branched covers. Topology,43:1247–1283, 2004. arXiv:math.GT/0107220.[3] S. Garoufalidis and A. Kricker. A rational noncommutative <strong><strong>in</strong>variant</strong> <strong>of</strong> boundary l<strong>in</strong>ks.Geom. Topol., 8:115–204, 2004. arXiv:math.GT/0105028.[4] L. H. Kauffman. On <strong>knots</strong>, volume 115 <strong>of</strong> <strong>An</strong>nals <strong>of</strong> Mathematics Studies. Pr<strong>in</strong>cetonUniversity Press, Pr<strong>in</strong>ceton, NJ, 1987.13


[5] M. Kontsevich. Feynman diagrams and low-dimensional topology. In First EuropeanCongress <strong>of</strong> Mathematics, Vol. II (Paris, 1992), volume 120 <strong>of</strong> Progr. Math., pages 97–121. Birkhäuser, 1994.[6] G. Kuperberg and D. P. Thurston. Perturbative 3-manifold <strong><strong>in</strong>variant</strong>s by cut-and-pastetopology. UC Davis Math 1999-36.[7] C. Lescop. On the Kontsevich-Kuperberg-Thurston construction <strong>of</strong> a configuration-space<strong><strong>in</strong>variant</strong> for rational <strong>homology</strong> 3-spheres. Prepublication Institut Fourier 655.[8] C. Lescop. Splitt<strong>in</strong>g formulae for the Kontsevich-Kuperberg-Thurston <strong><strong>in</strong>variant</strong> <strong>of</strong> rational<strong>homology</strong> 3-spheres. Prepublication Institut Fourier 656.[9] J. Marché. Surgery on a s<strong>in</strong>gle clasper and the 2-loop part <strong>of</strong> the Kontsevich <strong>in</strong>tegral.arXiv:math.GT/0410272.[10] J. Marché. A computation <strong>of</strong> the Kontsevich <strong>in</strong>tegral <strong>of</strong> torus <strong>knots</strong>. Alg. Geom. Topol.,4:1155–1175, 2004.[11] J. Marché. Sur l’<strong>in</strong>tégrale de Kontsevich des nœuds dans les variétés de dimension 3. PhDthesis, Université Paris 7, 2004.[12] T. Ohtsuki. A cabl<strong>in</strong>g formula for the 2-loop polynomial <strong>of</strong> <strong>knots</strong>. arXiv:math.GT/0310216.[13] T. Ohtsuki. communication privée, 2003.[14] L. Rozansky. A rationality conjecture about Kontsevich <strong>in</strong>tegral <strong>of</strong> <strong>knots</strong> and its implicationsto the structure <strong>of</strong> the colored Jones polynomial. Topology Appl., 127:47–76, 2003.14

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!