Toroidal homology spheres and SU(2)-representations
TTOROIDAL INTEGER HOMOLOGY THREE-SPHERES HAVEIRREDUCIBLE SU (2) -REPRESENTATIONS TYE LIDMAN, JUANITA PINZ ´ON-CAICEDO, AND RAPHAEL ZENTNER
Abstract.
We prove that if an integer homology three-sphere contains an em-bedded incompressible torus, then its fundamental group admits irreducible SU (2)-representations. Our methods use instanton Floer homology, and in particularthe surgery exact triangle, holonomy perturbations, and a non-vanishing resultdue to Kronheimer-Mrowka, as well as results about surgeries on cables due toGordon. Introduction
The fundamental group is one of the most powerful invariants to distinguish closedthree-manifolds. In fact, by Perelman’s proof of Thurston’s Geometrization conjec-ture [28, 30, 29], fundamental groups determine closed, orientable three-manifoldsup to orientations of the prime factors and up to the indeterminacy arising fromlens spaces. Prominently, the three-dimensional Poincar´e conjecture, a special caseof Geometrization, characterizes S as the unique closed, simply-connected three-manifold. For a three-manifold with non-trivial fundamental group, it is then usefulto quantify the non-triviality of the fundamental group. Since the Geometrizationtheorem implies that three-manifolds have residually finite fundamental groups [17],this non-triviality can be measured by representations to finite groups. However,there is not a finite group G such that every three-manifold group has a non-trivialhomomorphism to G . Therefore, a more uniform measurement of non-triviality canbe found in the following conjecture: Conjecture 1 (Kirby Problem 3.105(A), [18]) . If Y is a closed, connected, three-manifold other than S , then π ( Y ) admits a non-trivial SU (2) -representation. Note that this conjecture is equivalent to the statement that the fundamentalgroups of all integer homology three-spheres other than S admit irreducible SU (2)-representations. Indeed, every three-manifold whose first homology group is non-zero admits non-trivial abelian representations to SU (2). Moreover, lens spacesare examples of manifolds that admit non-trivial SU (2)-representations of theirfundamental groups, but no irreducible ones. There are also three-manifolds withnon-abelian fundamental group which do not admit irreducible representations [26].However, for representations of perfect groups to SU (2), non-triviality is equivalentto irreducibility. a r X i v : . [ m a t h . G T ] J a n TYE LIDMAN, JUANITA PINZ ´ON-CAICEDO, AND RAPHAEL ZENTNER
For comparison, the third author showed in [36] that Conjecture 1 is true if onereplaces SU (2) with SL ( C ). The reader may also relate Conjecture 1 with char-acterizing the three-manifolds with simplest instanton or Heegaard Floer homolo-gies. One side of the L-space conjecture predicts that every prime integer homologythree-sphere other than S and the Poincar´e homology three-sphere admits a co-orientable taut foliation. This fact, together with the gauge-theoretic methods usedby Kronheimer-Mrowka in [22], would then imply Conjecture 1.There are many families of integer homology three-spheres for which Conjecture 1has been established, such as those which are Seifert fibered (although the methodsgo back to Fintushel-Stern [12], this can be found explicitly in [32, Theorem 2.1]),branched double covers of non-trivial knots with determinant 1 [7, Theorem 3.1]and [35, Corollary 9.2], 1 /n -surgeries on non-trivial knots in S [21], those that arefilled by a Stein manifold which is not a homology ball [1], or for splicings of knotsin S [36].It follows again from Geometrization that there are three (non-disjoint) typesof prime integer homology three-spheres: Seifert fibered, hyperbolic, and toroidalones. We remark that although some toroidal integer homology three-spheres areSeifert fibered, they are never hyperbolic. The third author established that if allhyperbolic integer homology three-spheres have irreducible SU (2)-representations,then Conjecture 1 holds in general. While we are unable to complete the remainingstep in this program, we confirm the existence of SU (2)-representations for toroidalinteger homology three-spheres. Theorem 1.1.
Let Y be a toroidal integer homology three-sphere. Then π ( Y ) admits an irreducible SU (2) -representation. A proof of Theorem 1.1 could be obtained by showing that toroidal integer homol-ogy three-spheres have non-trivial instanton Floer homology. Although we expectthe latter to be true (see [18, Problem 3.106]), we do not prove it in this article. Ourproof of Theorem 1.1 instead relies on holonomy perturbations in a manner similarto the proof of [36, Theorem 8.3]. If Y is a toroidal integer homology three-sphere,then Y can be viewed as a splice of knots K i in integer homology three-spheres Y i for i = 1 , Y i has an irreducible SU (2)-representation, then there is a π -surjective map from Y to Y i and we canpull back to an irreducible SU (2)-representation for Y . If not, then we will studythe image of the space of representations of the knot exterior Y i \ N ( K i ) ◦ in thecharacter variety for the boundary torus (i.e. in the pillowcase). Here, N ( K i ) de-notes a closed tubular neighborhood of K i , and N ( K i ) ◦ denotes its interior. Similarto the case of non-trivial knots in S , if Y i has no irreducible representations, we willshow that the image in the pillowcase contains a suitably essential loop. The loopsfor the two exteriors will have a non-trivial intersection, and therefore the splicedmanifold Y will admit an irreducible SU (2)-representation. OROIDAL HOMOLOGY SPHERES AND SU (2)-REPRESENTATIONS 3 Theorem 1.1 gives a simpler proof of [36, Theorem 9.4] since it avoids the use ofa finiteness result of Boileau-Rubinstein-Wang.
Corollary 1.2 (Theorem 9.4, [36]) . Every integer homology three-sphere other than S has an irreducible SL ( C ) -representation of its fundamental group.Proof. By the remarks above we have to consider three cases: Seifert fibered, hy-perbolic, and toroidal integer homology three-spheres. Let Y be an integer ho-mology three-sphere other than S . If Y is hyperbolic, it admits an irreducible SL ( C )-representation by lifting the holonomy representation to P SL ( C ) [8]. If Y is Seifert fibered, then π ( Y ) admits an irreducible SU (2)-representation by [32,Theorem 2.1]. If Y is toroidal, the result now follows from Theorem 1.1. (cid:3) In order to generalize the holonomy perturbation machinery developed by thethird author from non-trivial knots in S , we will need to establish a non-vanishingresult which may be of independent interest. Theorem 1.3.
Let J be a knot in an integer homology three-sphere Y such that theexterior of J is irreducible and boundary-incompressible. Suppose that I ∗ ( Y ) = 0 .Then, I w ∗ ( Y ( J )) (cid:54) = 0 . Here, and throughout this article, I ∗ denotes Floer’s original version of instantonFloer homology and I w ∗ denotes instanton Floer homology for an admissible SO (3)-bundle with second Stiefel-Whitney class w . (Note that Y ( J ) admits only one suchbundle.)The proof of Theorem 1.3 is a combination of (1) Kronheimer-Mrowka’s non-vanishing result for instanton Floer homology of three-manifolds with a taut suturedmanifold hierarchy [19], (2) the surgery exact triangle in instanton Floer homology,and (3) Gordon’s description of surgery on cable knots [16]. The argument is similarto Kronheimer-Mrowka’s proof of Property P [22].While Theorem 1.3 itself may not be particularly interesting, it does lead to thefollowing corollary, whose analogue in Heegaard Floer homology has been establishedby Ni [27, p.1144] and Conway and Tosun [6]. The proof of the corollary appears inSection 2 below.
Corollary 1.4.
Let Y (cid:54) = S be an integer homology three-sphere which bounds aMazur manifold. Then, I ∗ ( Y ) (cid:54) = 0 , and hence π ( Y ) admits an irreducible SU (2) -representation. Recall that Baldwin-Sivek prove that if an integer homology three-sphere Y bounds a Stein domain with non-trivial homology, then π ( Y ) admits an irreducible SU (2)-representation [1, Theorem 1.1]. In light of Conjecture 1, the following con-jecture would be a natural extension of their work: Conjecture 2. If Y (cid:54) = S is an integer homology three-sphere which bounds a Steininteger homology ball, then π ( Y ) admits an irreducible SU (2) -representation. TYE LIDMAN, JUANITA PINZ ´ON-CAICEDO, AND RAPHAEL ZENTNER
Since Stein domains admit handlebody decompositions with no three-handles [11],Corollary 1.4 proves this conjecture for the boundaries of Stein integer homologyballs with the simplest possible handle decompositions.Theorem 1.1 also has two simple corollaries. The first one is obtained by consid-ering branched covers over satellite knots in S . Remarkably, its proof requires nouse of gauge theory, beyond our main result. Its proof appears in Section 5 below. Corollary 1.5.
Let K be a prime, satellite knot in S . Conjecture 1 holds for anynon-trivial cyclic branched cover of K . To obtain the second corollary, define a graph manifold integer homology three-sphere to be a closed, orientable three-manifold whose torus decomposition has nohyperbolic pieces. As discussed above, the fundamental groups of Seifert integerhomology three-spheres other than S admit irreducible SU (2)-representations, andhence we obtain: Corollary 1.6.
Let Y be a graph manifold integer homology three-sphere other than S . Then π ( Y ) admits an irreducible SU (2) -representation. A first alternate proof of this corollary can be obtained by noting that everyinteger homology three-sphere other than S which is a graph manifold can berealized as the branched double cover of a non-trivial (arborescent) knot in S , see[3]. A second alternate proof can be obtained by noting that every prime graphmanifold integer homology three-sphere Y other than S or Σ(2 , ,
5) admits a co-orientable taut foliation by [2, Corollary 0.3]. This implies that I ∗ ( Y ) (cid:54) = 0, and thisin turn implies that there exists an irreducible SU (2)-representation. On the otherhand, the binary dodecahedral group is well-known to admit two conjugacy classesof irreducible representations, completing the proof. Note that, unlike for Seifertinteger homology three-spheres, the Casson invariant of a non-trivial graph manifoldcan be zero. For example, the three-manifold Y obtained as the splice of two copiesof the exterior of the right handed trefoil has trivial Casson invariant [15, 4]. Outline.
In Section 2 we establish the main technical result Theorem 1.3 whosestrategy also leads us to prove Corollary 1.4 about Mazur manifolds. In Section 3,we review the pillowcase construction and prove Theorem 1.1 in subsection 3.3,using a technical result about invariance under holonomy perturbations in instantonFloer homology reviewed in Section 4. The material in Section 4 is mostly known(or at least folklore knowledge) and can be found elsewhere, but the reader mightappreciate our synthesis of the role of holonomy perturbations and our sketch ofinvariance in order to follow more easily through the proof of our main results. InSection 5, we prove Corollary 1.5. Some authors impose additional constraints, such as primeness or a non-trivial torusdecomposition.
OROIDAL HOMOLOGY SPHERES AND SU (2)-REPRESENTATIONS 5 Acknowledgements.
Tye Lidman was partially supported by NSF grant DMS-1709702 and a Sloan Fellowship. Juanita Pinz´on-Caicedo is grateful to the MaxPlanck Institute for Mathematics in Bonn for its hospitality and financial supportwhile a portion of this work was prepared for publication. She was partially sup-ported by NSF grant DMS-1664567, and by Simons Foundation Collaboration grant712377. Raphael Zentner is grateful to the DFG for support through the Heisenbergprogram. We would also like to thank John Baldwin, Paul Kirk, and Tom Mrowkafor helpful discussions.2.
Instanton Floer homology of 0-surgery
In this section we rely solely on formal properties of instanton Floer homologyto prove Theorem 1.3 regarding the instanton Floer homology of 0-surgeries, andCorollary 1.4 regarding the instanton Floer homology of integer homology three-spheres that bound Mazur manifolds. More concrete aspects of instanton Floerhomology groups, in particular those regarding perturbations, appear in Section 4but in this section we wish to place the focus on the usefulness of formal propertiesfor purposes of computations.We consider instanton Floer homology for admissible bundles, as introduced byFloer [14]. For integer homology three-spheres, this is the trivial SU (2)-bundle over Y . For three-manifolds with positive first Betti number, this is an SO (3)-bundle P → Y such that there is a surface Σ ⊆ Y on which the second Stiefel-Whitneyclass w := w ( P ) evaluates non-trivially, that is, such that (cid:104) w ( P ) , [Σ] (cid:105) (cid:54) = 0. Theinstanton Floer homology group is defined as a version of Morse homology of theChern-Simons function on the space of connections on the admissible bundle [14, 9].It is denoted by I ∗ ( Y ) for the trivial bundle on integer homology three-spheres, andit is denoted by I w ∗ ( Y ) for SO (3)-bundles P → Y with w ( P ) = w . We remark herethat for an integer homology three-sphere, the trivial connection is isolated and isthe unique reducible connection (up to gauge equivalence). In the other cases, theadmissibility condition ensures that there are no reducible flat connections on thebundle.In the case of a knot K in an integer homology three-sphere Y , there is a uniqueadmissible bundle on the 0-surgery Y ( K ), because H ( Y ( K ); Z / ∼ = Z /
2. There-fore, the instanton Floer homology group I w ( Y ( K )) is defined without ambiguity. Proposition 2.1.
Instanton Floer homology satisfies the following properties:(1) For Y an integer homology three-sphere and any n ∈ Z , the three-manifolds Y /n ( K ) , Y / ( n +1) ( K ) , and Y ( K ) fit into an exact triangle I ∗ ( Y /n ( K )) (cid:39) (cid:39) I ∗ ( Y / ( n +1) ( K )) (cid:54) (cid:54) I w ∗ ( Y ( K )) . (cid:111) (cid:111) TYE LIDMAN, JUANITA PINZ ´ON-CAICEDO, AND RAPHAEL ZENTNER (2) If M is an irreducible three-manifold with b ( M ) = 1 , then I w ∗ ( M ) (cid:54) = 0 .(3) For Y an integer homology three-sphere, if π ( Y ) admits no irreducible SU (2) -representations, then I ∗ ( Y ) = 0 .(4) I w ∗ ( S × S ) = 0 .Proof. The surgery exact triangle in (2.1(1)) is originally due to Floer [14, Theorem2.4] with details given in [5, Theorem 2.5]. The non-triviality result in (2.1(2)) isprecisely [20, Theorem 7.21]. Next, (2.1(3)) follows from [13, Theorem 1], since if π ( Y ) admits no irreducible SU (2)-representations, then the generating set for theinstanton Floer chain groups is empty. Finally, (2.1(4)) follows from (2.1(3)) and(2.1(1)), by considering the surgery exact triangle for surgery on the unknot in S .Alternatively, this follows from the definition of I w ∗ (see Section 4), since π ( S × S )admits no representations to SO (3) which do not lift to SU (2). (cid:3) We will be particularly interested in integer homology three-spheres whose fun-damental groups do not admit irreducible SU (2)-representations. We therefore es-tablish the following definition. Definition 2.2.
An integer homology three-sphere Y is SU (2) -cyclic if every SU (2)-representation of π ( Y ) is trivial.Notice that Conjecture 1 states that S is the only SU (2)-cyclic integer homologythree-sphere.Having stated the above formal properties of instanton Floer homology, the proofsof Theorem 1.3 and Corollary 1.4 now follow easily.2.1. Non-vanishing of Instanton Floer Homology.
In this subsection we il-lustrate the way the formal properties from Proposition 2.1 can be used to showthat the instanton homology groups are non-zero in two cases: (1) three-manifoldsobtained as 0-surgery along knots in SU (2)-cyclic integer homology three-sphereswhose exterior is irreducible and boundary incompressible, and (2) three-manifoldsother than S obtained as the boundary of a Mazur manifold. Proof of Theorem 1.3.
We assume I w ∗ ( Y ( K )) is trivial and argue by contradiction.By Proposition 2.1(1) the three-manifolds Y /n ( K ), Y / ( n +1) ( K ), and Y ( K ) fit to-gether in an exact triangle I ∗ ( Y /n ( K )) (cid:39) (cid:39) I ∗ ( Y / ( n +1) ( K )) (cid:54) (cid:54) I w ∗ ( Y ( K )) . (cid:111) (cid:111) The assumption I w ∗ ( Y ( K )) = 0 implies that there is an isomorphism I ∗ ( Y / ( n +1) ( K )) ∼ = I ∗ ( Y /n ( K )) for each n ∈ Z . OROIDAL HOMOLOGY SPHERES AND SU (2)-REPRESENTATIONS 7 In particular, if n = 0 then I ∗ ( Y ( K )) ∼ = I ∗ ( Y ) = 0 thus showing that for all n ∈ Z , I ∗ ( Y /n ( K )) = 0 . (1)Now, a result of Gordon [16, Lemma 7.2] shows that Y / ( K ) is diffeomorphic to Y ( K , ), where K , is the (2 , K (See Figure 5 for an example of K , ). This together with Equation (1) implies I ∗ ( Y ( K , )) = 0. An iteration of anexact triangle as in Proposition 2.1(1) for surgeries along K , gives I w ∗ ( Y ( K , )) =0. We now consider a decomposition of Y ( K , ) that includes the knot exterior of K in Y . Denote by C , a closed curve that lies in the boundary of a “small” solid torus S × ∂D / ⊂ S × D , and representing the class 2[ S ] + [ ∂D / ] in H ( S × ∂D / ).Notice that the 0-framing of K , in Y induces the framing on C , determined bythe curve λ in ∂N ( C , ) that represents the class 2[ S ] in H ( S × ∂D ) (see [16, pg.692]). Therefore, the manifold Y ( K , ) can be expressed as the union of the knotexterior Y \ N ( K ), and the result of Dehn surgery on S × D along the curve C , with framing given by λ . By hypothesis the knot exterior Y \ N ( K ) is irreducibleand boundary-incompressible, and by [16, Lemma 7.2] the 0-surgery along the curve C , is a Seifert fibred space with incompressible boundary. Hence Y ( K , ) is anirreducible closed three-manifold with first Betti number equal to 1 and with trivialinstanton Floer homology. However, this contradicts Proposition 2.1(2), which saysthat I w ∗ ( Y ( K , )) (cid:54) = 0. (cid:3) Next, consider integer homology three-spheres that bound a Mazur manifold,that is, a four-manifold that admits a handle decomposition in terms of exactly one0-handle, one 1-handle, and one 2-handle as in Figure 1. Then we have:
Proof of Corollary 1.4. If Y bounds a Mazur manifold, then there exists a knot J in Y such that Y ( J ) = S × S . Moreover, if I ∗ ( Y ) = 0, a combination of thesurgery exact triangle from Proposition 2.1(1) and the computation I w ∗ ( S × S ) = 0from Proposition 2.1(4) shows once again that I ∗ ( Y / ( J )) = 0. The same argumentused above in the proof of Theorem 1.3 then gives I ∗ ( Y ( J , )) = 0. However, itis easy to see that the exterior of a knot in S × S which generates homology iseither irreducible and boundary-incompressible or a solid torus. The latter casecorresponds to Y = S , so by assumption we have that Y ( J , ) is irreducible with b = 1. But this contradicts Proposition 2.1(2). (cid:3) The pillowcase alternative
In this section we recall the relevant background on SU (2)-character varieties andgeneralize work of the third author [36] to prove Theorem 1.1, our main result. TYE LIDMAN, JUANITA PINZ ´ON-CAICEDO, AND RAPHAEL ZENTNER n Figure 1.
A Mazur manifold with one two-handle attached withframing given by n for some n ∈ N .3.1. The pillowcase.
Given a connected manifold M , we denote by R ( M ) = Hom( π ( M ) , SU (2)) /SU (2)the space of SU (2)-representations of its fundamental group, up to conjugation.We will write R ( M ) ∗ for the subset of irreducible representations. For example,the space R ( T ) is identified with the pillowcase , an orbifold homeomorphic toa two-dimensional sphere with four corner points. To see this, notice that since π ( T ) ∼ = Z is abelian, the image of any representation ρ : π ( T ) → SU (2) iscontained in a maximal torus subgroup of SU (2). Up to conjugation, this torus canbe identified with the circle group consisting of matrices of the form (cid:20) e iθ e − iθ (cid:21) for θ ∈ [0 , π ]. Thus, if we denote the generators of π ( T ) ∼ = Z by m and l , then,again after conjugation, a representation ρ ∈ R ( T ) is determined by ρ ( m ) = (cid:20) e iα e − iα (cid:21) and ρ ( l ) = (cid:20) e iβ e − iβ (cid:21) , and hence we can associate to ρ a pair ( α, β ) ∈ [0 , π ] × [0 , π ]. However, conjugationof ρ by the element (cid:20) − (cid:21) gives rise to the representation associated to the pair(2 π − α, π − β ). This is the only ambiguity, however, as can be seen using the factthat the trace of an element in SU (2) determines its conjugacy class. Therefore R ( T ) is isomorphic to the quotient of the fundamental domain [0 , π ] × [0 , π ] byidentifications on the boundary as indicated in Figure 2.If we have a three-manifold M with torus boundary, then the inclusion i : T ∼ = ∂M (cid:44) → M induces a map i ∗ : R ( M ) → R ( T ) by restricting a representation to theboundary. For instance, if K is a knot in a three-manifold Y , then the three-manifold Y ( K ) := Y \ N ( K ) ◦ obtained by removing the interior of a tubular neighborhood OROIDAL HOMOLOGY SPHERES AND SU (2)-REPRESENTATIONS 9 R ( K ) R ( K ) P Q P Q
Figure 2.
The gluing pattern for obtaining the pillowcase from arectangle, and the image of the representation variety R ( K ) of thetrefoil in the pillowcase. N ( K ) of K from Y , is a three-manifold with boundary a two-dimensional torus.Figure 2 shows the image of R ( S ( K )) when K is the right handed trefoil in S ,once in the pillowcase, and once in the fundamental domain [0 , π ] × [0 , π ]. Here weuse the convention that the first coordinate corresponds to ρ ( m K ), where m K is ameridian to the knot K , and the second coordinate corresponds to ρ ( l K ), where l K is a longitude of the knot K .For a knot K in a three-manifold Y there is a well-defined notion of meridian m K , and if the knot is nullhomologous, there is a well-defined notion of longitude l K . In particular, this is the case for any knot K in an integer homology three-sphere Y . In what follows, we will use the notation R ( K ) := R ( Y ( K )) if it is clearwhich integer homology three-sphere Y we have in mind, and we will stick to theabove convention of the coordinates in R ( T ) corresponding to the meridian andlongitude of K . With these conventions, all abelian representations in R ( K ) mapunder i ∗ to the thick red line { β = 0 mod 2 π Z } ‘at the bottom’ of the pillowcase R ( T ). Indeed, l K is a product of commutators in the fundamental group of theknot complement, so an abelian representation necessarily maps l K to the identity.Furthermore, for any α ∈ [0 , π ] we can find an abelian representation of R ( K ) whoserestriction to R ( T ) corresponds to ( α, a := { α = 0 mod 2 π Z } and a π := { α = π mod 2 π Z } , we obtain a cylinder C = [0 , π ] × R / π Z . In the gluing patternof Figure 2 this means that we do not perform the identifications along the fourindicated vertical boundary lines. Our main goal is to prove Theorem 3.5 below, which asserts that if K is a knotin an SU (2)-cyclic integer homology three-sphere whose 0-surgery has non-trivialinstanton homology, then the image of R ( K ) in the pillowcase contains a homo-logically non-trivial embedded closed curve in the cylinder C . In order to deriveTheorem 1.1 from this, we need a more refined statement, namely, that there is ahomologically non-trivial embedded closed curve in i ∗ ( R ( K )) that is disjoint from aneighborhood of the two lines a and a π . Notice that for a knot in S , there are norepresentations with ρ ( l K ) (cid:54) = id and ρ ( m K ) = ± id. This is because the fundamen-tal group of a knot complement in S is normally generated by the meridian of theknot. In particular, there are no representations in i ∗ ( R ( K )) that have coordinates( α, β ) with β (cid:54) = 0, and α = 0 or α = π . In [36, Proposition 8.1], it is shown thatthe image of R ( K ) ∗ , the subset of irreducible representations in R ( K ), in fact staysoutside a neighborhood of these two lines. We begin with a generalization of thisfact. Lemma 3.1.
Let K be a knot in an SU (2) -cyclic integer homology three-sphere Y .There is a neighborhood of the lines { α = 0 mod 2 π Z } and { α = π mod 2 π Z } inthe pillowcase which is disjoint from the image of R ( K ) ∗ .Proof. Suppose by contradiction that the image of R ( K ) ∗ intersects every neighbor-hood of the lines { α = 0 mod 2 π Z } and { α = π mod 2 π Z } . If that was the case,then we could find a sequence of elements in R ( K ) ∗ whose image under i ∗ convergesto a point on one of the two lines. By the compactness of R ( K ), the limit is theimage of a representation ρ : π ( Y ( K )) → SU (2) sending every meridional curve µ to ±
1. We first claim that ρ must be a central representation (and hence reducible),and so its image under i ∗ can only be (0 ,
0) or ( π, ρ ( µ ) = 1, then ρ : π ( Y ( K )) → SU (2) is really a representation of π ( Y ).Since Y is assumed to be SU (2)-cyclic, then the representation is trivial and there-fore ρ ( λ ) = 1. As a consequence, if the limit of elements in R ( K ) ∗ is an elementof the line { α = 0 mod 2 π Z } , then it is the point (0 ,
0) in the pillowcase. Next,consider the case that ρ ( µ ) = −
1. If the representation ρ is irreducible, then weobtain an irreducible representation (cid:101) ρ : π ( Y ) → SO (3). The obstruction to liftingan SO (3) representation into an SU (2)-representation is an element of H ( Y ; Z / Y is an integer homology three-sphere, the obstruction vanishes and (cid:101) ρ would lift to an irreducible representation to SU (2), contradicting the fact that Y is SU (2)-cyclic. Therefore, a representation ρ : π ( Y ( K )) → SU (2) satisfying ρ ( µ ) = − H ( Y ( K )). Be-cause λ is trivial in H ( Y ( K )), we see that ρ is the central representation sending µ to − λ to 1, and this corresponds to the point ( − π,
0) in the pillowcase. Allof this shows that if a sequence of elements in i ∗ R ( K ) converges to a point on thelines { α = 0 mod 2 π Z } and { α = π mod 2 π Z } , then the limit point is a centralrepresentation. For notation, we will call these representations ρ ± for the sign ofthe image of µ . OROIDAL HOMOLOGY SPHERES AND SU (2)-REPRESENTATIONS 11 Now, it remains to show that the points (0 ,
0) and ( π,
0) cannot be limits ofirreducible representations. We remark here that this fact does not require that Y is SU (2)-cyclic. Let Γ = π ( Y ( K )). A result of Weil [33] expanded in [25, Chapter2] shows that T ρ R ( K ) corresponds to H (Γ; su (2) ad ◦ ρ ). This group is identified withthe first cohomology group (with twisted coefficients) of a K (Γ , H (Γ; su (2) ad ◦ ρ ) = H ( Y ( K ); su (2) ad ◦ ρ ) andso T ρ R ( K ) = H ( Y ( K ); su (2) ad ◦ ρ ). Next, since each representation ρ ± is central,then ad ◦ ρ ± is the trivial representation and so H (cid:0) Y ( K ); su (2) ad ◦ ρ ± (cid:1) = H (cid:0) Y ( K ); R (cid:1) ∼ = R . This shows that the tangent space to R ( K ) at ρ ± is three-dimensional. Since weobtain three dimensions of freedom by abelian representations near ρ ± in R ( K ), theentire tangent space to R ( K ) consists of tangent vectors to abelian representationsand so there cannot be irreducible representations near ρ ± , completing the proof. (cid:3) Essential curves in the pillowcase.
In this section, we relate the instantonFloer homology of 0-surgery on a knot to the image of the character variety of theknot exterior in the pillowcase. This will be the key step in the proof of Theorem 1.1,found at the end of this subsection.We next establish some notation, following Kronheimer-Mrowka in [21], that willbe useful in the proof of our next theorem.
Definition 3.2.
For a subset L ⊆ R ( T ), we denote by R ( K | L ) the set of elements[ ρ ] ∈ R ( K ) such that [ i ∗ ρ ] ∈ L . Theorem 3.3.
Let K be a knot in an integer homology three-sphere Y , and assumethat the instanton Floer homology of the -surgery is non-vanishing, I w ∗ ( Y ( K )) (cid:54) = 0 .Then any topologically embedded path from P = (0 , π ) to Q = ( π, π ) in the associatedpillowcase has an intersection point with the image of R ( K ) . Before proving the theorem, we point out that this generalizes [36, Theorem 7.1],from knots in S to knots in general integer homology three-spheres. The maindifference in the argument compared to [36, Theorem 7.1] is that here we makeuse of the non-trivial instanton Floer homology of the 0-surgery in an essentialway, which is exploited through its connection with holonomy perturbations of theChern-Simons functional. The arguments of the third author in [36] instead useholonomy perturbations of a moduli space which computes the Donaldson invariantsof a closed 4-manifold containing the 0-surgery as a hypersurface. In that case, thenon-vanishing result builds on the existence of a taut foliation on S ( K ) for a non-trivial knot K . In the case at hand, we do not know whether Y ( K ), the 0-surgeryon a knot K in the integer homology three-sphere Y , admits a taut foliation. Proof.
Suppose by contradiction that there is a continuous embedded path c from P to Q such that its image is disjoint from i ∗ ( R ( K )) ⊆ R ( T ). (We will not distinguish between paths and their image for the remainder of this proof.) In otherwords, R ( K | c ) is empty. In particular, we may suppose that c is disjoint from thebottom line { β = 0 } of the pillowcase R ( T ), since any element of this line liesin the image of i ∗ . Since the image i ∗ ( R ( K )) is compact, there is a neighborhood U ⊆ R ( T ) of the image of c in R ( T ) which is still disjoint from i ∗ ( R ( K )). Since R ( K | c ) is empty, for c (cid:48) sufficiently close to c , R ( K | c (cid:48) ) is empty as well.Associated to a three-manifold and admissible bundle, we consider two objects:the Chern-Simons functional and holonomy perturbations of the Chern-Simons func-tional. These are described in detail in Section 4, in particular Sections 4.2 and 4.3,but their definition is not needed for the proof. Given a three-manifold Z with ad-missible bundle represented by w and a holonomy perturbation Ψ, let R w Ψ ( Z ) denotethe set of critical points of the Chern-Simons functional perturbed by Ψ. By Theo-rem 4.4 below (which is essentially a synthesis of [36, Theorem 4.2 and Proposition5.3]), there exists a path c (cid:48) arbitrarily close to c and a (holonomy) perturbation Ψof the Chern-Simons functional such that R w Ψ ( Y ( K )) is a double cover of R ( K | c (cid:48) ).Therefore, R w Ψ ( Y ( K )) is empty, so computing Morse homology with respect to thisperturbation of the Chern-Simons functional produces a trivial group. However,Theorem 4.5 below asserts that computing Morse homology with respect to the par-ticular perturbation Ψ produces a group isomorphic to I w ∗ ( Y ( K )), which is non-zeroby assumption. Therefore, we obtain a contradiction. (cid:3) Remark . Although [36, Proposition 5.3] is only stated for knots in S , thearguments used in its proof apply for a knot in an arbitrary SU (2)-cyclic integerhomology three-sphere.If we combine the constraint that Y is SU (2)-cyclic with the assumption that I w ∗ ( Y ( K )) is non-trivial, then we obtain the following generalization of [36, Theorem7.1], which will be the last step before the proof of our main theorem. Theorem 3.5. (Pillowcase alternative) Suppose Y is an SU (2) -cyclic integer ho-mology three-sphere. Suppose K is a knot in Y such that the 0-surgery Y ( K ) hasnon-trivial instanton Floer homology I w ∗ ( Y ( K )) , where w is the non-zero class in H ( Y ( K ); Z / ∼ = Z / . Then the image i ∗ ( R ( Y ( K ))) in the cut-open pillowcase C = [0 , π ] × ( R / π Z ) contains a topologically embedded curve which is homologicallynon-trivial in H ( C ; Z ) ∼ = Z .Proof. The hypothesis implies that the lines { (0 , β ) ∈ R ( T ) | β (cid:54) = 0 } and { ( π, β ) ∈ R ( T ) | β (cid:54) = 0 } have empty intersection with i ∗ ( R ( Y ( K ))). The conclusion thenfollows from Theorem 3.3 together with the Alexander duality argument of [36,Lemma 7.3]. (cid:3) Main Result.
In this subsection we prove that if an integer homology three-sphere contains an embedded incompressible torus, then the fundamental group ofthe homology three-sphere admits irreducible SU (2)-representations. To derive our OROIDAL HOMOLOGY SPHERES AND SU (2)-REPRESENTATIONS 13 i ∗ R ( K ) QP Figure 3.
This is a hypothetical image of a representation variety i ∗ ( R ( K )) of a knot K in an integer homology three-sphere Y . Thehomology three-sphere Y is assumed to satisfy I w ∗ ( Y ( K )) (cid:54) = 0 andassumed to not be SU (2)-cyclic. As a consequence, i ∗ ( R ( K )) inter-sects every path joining P and Q as in Theorem 3.3, but it does notcontain a curve which is homologically non-trivial in the cut-openpillowcase C = [0 , π ] × ( R / π Z ). This hypothetical example thusillustrates that the SU (2)-cyclic assumption is necessary in Theo-rem 3.5.result we first recall that we can realize a toroidal integer homology three-sphereas a splice, as in [10, Proof of Corollary 6.2]. We then study the image of the twoknot exteriors in the pillowcase of the incompressible torus. With this in mind, weinclude the following definition. Definition 3.6.
Let K ⊂ Y and K ⊂ Y be oriented knots in oriented integerhomology three-spheres. For i = 1 ,
2, denote by µ i , λ i ⊂ ∂N ( K i ) a meridian andlongitude for K i in Y i . Form a three-manifold Y as( Y \ N ( K ) ◦ ) ∪ h ( Y \ N ( K ) ◦ ) , where h : ∂N ( K ) → ∂N ( K ) identifies µ with λ , and λ with µ . The manifold Y is called the splice of Y and Y along knots Y and K .Let Y be an integer homology three-sphere and let T be a two-dimensional torusembedded in Y in such manner that its normal bundle is trivial. A simple applicationof the Mayer-Vietoris sequence shows that Y \ N ( T ) ◦ has two connected components M , M , and that each component has the same homology groups as S . The “halflives, half dies” principle shows that for each i = 1 , α i , β i ) forthe peripheral subgroup of ∂M i such that β i is nullhomologous in M i . Therefore, if Y i denotes the union of M i and a solid torus S × D in such a way that the curve { } × ∂D gets identified with α i , then Y i is an integer homology three-sphere.Moreover, since T is incompressible in Y , then the core of the solid torus in Y i is a non-trivial knot K i . In other words, every toroidal integer homology three-spherecan be expressed as a splice of non-trivial knots K and K in integer homologythree-spheres Y and Y .With all of this in place, we are ready to prove our main result. Proof of Theorem 1.1.
Realize Y as a splice ( Y \ N ( K ) ◦ ) ∪ h ( Y \ N ( K ) ◦ ), with K , K non-trivial knots. Suppose first that Y i \ N ( K i ) ◦ is reducible, in other words,that Y i \ N ( K i ) ◦ = Q i Z i \ N ( J i ) ◦ ) where Q i , Z i are integer homology three-spheres and J i ⊂ Z i has irreducible and boundary-incompressible exterior. As aconsequence of Van-Kampen’s theorem, there exists a surjection π ( Y i \ N ( K i ) ◦ ) → π ( Z i \ N ( J i ) ◦ ), and this surjection induces a π -surjection from Y to the splice of( Z , J ) and ( Z , J ). Thus, our proof reduces to the case when Y is the splice oftwo knots with irreducible and boundary-incompressible exteriors, which we assumefrom now on.Next, by the Seifert–van Kampen theorem, the pieces of the decomposition fitinto the following commutative diagram π ( Y \ N ( K ) ◦ ) (cid:42) (cid:42) π ( T ) (cid:52) (cid:52) (cid:42) (cid:42) π ( Y ) π ( Y \ N ( K ) ◦ ) (cid:52) (cid:52) and since each Y i \ N ( K i ) ◦ is a homology circle, there exists a π -surjection from Y to each Y i . Therefore, our proof reduces further to the case when both Y and Y are SU (2)-cyclic since an irreducible representation for Y i gives rise to one for Y .To recap, the previous two paragraphs allow us to assume that Y is the spliceof ( Y , K ), ( Y , K ) with each Y i an SU (2)-cyclic homology three-sphere, and each K i ⊂ Y i a knot with irreducible and boundary-incompressible exterior. Then, asa consequence of Proposition 2.1(3) we have that each Y i has trivial instantonFloer homology. Moreover, since each Y i \ N ( K i ) ◦ is irreducible and boundary-incompressible, Theorem 1.3 shows that the instanton Floer homology of 0-surgeryon Y i along K i is non-zero. Therefore, the hypotheses of both Theorem 3.5 andLemma 3.1 hold, and the proof now follows exactly as in [36, Proof of Theorem8.3(i)] with [36, Theorem 7.1] and [36, Proposition 8.1(ii)] replaced by Theorem 3.5and Lemma 3.1 respectively. (cid:3) Review of instanton Floer homology and holonomy perturbations
We start this section with a disclaimer: We do not claim to prove any originalor new result in this section. However, we review instanton Floer homology andholonomy perturbations to the extent which is necessary in order to understandthe proof of our main results above. For instance, Section 4.3 below contains a
OROIDAL HOMOLOGY SPHERES AND SU (2)-REPRESENTATIONS 15 P Q
Figure 4.
Let Y be the three-manifold obtained as the splice oftwo copies of the exterior of a right handed trefoil, and let T bethe incompressible torus given as the intersection of the two knotexteriors. The figure shows the image of each copy of R ( T , ) ∗ in thepillowcase. Note that any representation of the splice correspondingto an intersection of the red and blue curves is irreducible.synthesis of the third author’s results about holonomy perturbations from [36] whichwe hope the reader unfamiliar with this reference will appreciate. Section 4.5 belowcontains a result about invariance under holonomy perturbations in the context ofan admissible bundle with non-trivial second Stiefel-Whitney class, together witha sketch of proof. Again, this result is already contained in [13] and [9], but bylooking up these references it may not be immediately clear whether these resultsapply verbatim in our situation.The proof of Theorem 3.3 relies on a non-vanishing result of an instanton Floerhomology group I w ∗ , Φ ( Y ( K )), computed with suitable perturbation terms Φ of theChern-Simons function. We will review the construction of these perturbation termsbelow, which are built from the holonomy along families of circles, parametrized byembedded surfaces. The critical points of the complex underlying the homologygroup I w ∗ , Φ ( Y ( K )) will have a clear interpretation in terms of intersections of therepresentation variety R ( K ) with certain deformations of the path given by thestraight line { β = π } in the pillowcase, resulting as the representation variety of theboundary of the exterior of K in Y as before.On the other hand, Theorem 1.3 yields a non-vanishing result for I w ∗ ( Y ( K )),defined in the usual way, and in particular without the above class of perturbationterms. We can therefore complete the proof from the fact that the two instanton Floer homology groups, I w ∗ ( Y ( K )) and I w ∗ , Φ ( Y ( K )), are isomorphic, and we sketchthe proof of this below. Remark . In the construction of both I w ∗ ( Y ( K )) and I w ∗ , Φ ( Y ( K )) there are typi-cally perturbation terms involved for the sake of transversality. These can be chosenas small as one likes, in a suitable sense. We will omit these auxiliary perturbationsfrom our notation. The perturbations labeled by the terms Φ, however, will have aclear geometric purpose, and the discussion below will focus on these.4.1. The Chern-Simons function.
For details on the holonomy perturbations weuse we refer the reader to Donaldson’s book [9], Floer’s orginal article [14], and thethird author’s article [36].If we deal with an admissible SO (3)-bundle F → Y over a three-manifold Y with second Stiefel-Whitney class w , we may suppose that it arises from an U (2)-bundle E → Y as its adjoint bundle su ( E ), see for instance [9, Section 5.6]. Then w = w ( E ) ≡ c ( E ) mod 2. The space of SO (3)-connections on F is then naturallyisomorphic to the space of U (2)-connections on E that induce a fixed connection θ in the determinant line bundle det( E ), which we will suppress from notation.When dealing with functoriality properties, it is more accurate to consider w tobe an embedded 1-manifold which is Poincar´e dual to w ( E ) = w ( F ), see [23].We will fix a reference connection A on E and consider the Chern-Simons func-tion CS : A → R A (cid:55)→ (cid:90) Y tr(2 a ∧ ( F A ) + a ∧ d A a + 13 a ∧ [ a ∧ a ]) , defined on the affine space A of connections A in E which induce θ in det( E ), andwhere we have written A = A + a with a ∈ Ω ( Y ; su ( E )). The term F A denotesthe curvature of a connection A , and ( F A ) denotes its trace-free part, and d A denotes the exterior derivative associated to a connection A . We denote by G thegroup of bundle automorphisms of E which have determinant 1. The Chern-Simonsfunction induces a circle-valued function CS : B → R / Z on the space B = A / G ofconnections modulo gauge equivalence, and the instanton Floer homology I w ∗ ( Y ) isthe Morse homology, in a suitable sense, of the Chern-Simons function CS. To carrythis out, one has to deal with a suitable grading on the critical points, which willonly be a relative Z / OROIDAL HOMOLOGY SPHERES AND SU (2)-REPRESENTATIONS 17 prove independence of the various choices involved, and in particular the Riemannianmetric and the perturbation terms required for transversality.One may also deal with orientations, but we do not need this in our situation,where Z / Review of holonomy perturbations.
To set up the perturbation of theChern-Simons function we are using, we need to introduce some notation. Let χ : SU (2) → R be a class function, that is, a smooth conjugation invariant function.Any element in SU (2) is conjugate to a diagonal element, and hence there is a2 π -periodic even function g : R → R such that χ (cid:18)(cid:20) e it e − it (cid:21)(cid:19) = g ( t ) (2)for all t ∈ R . Furthermore, let Σ be a compact surface with boundary, and let µ be a real-valued two-form which has compact support in the interior of Σ and with (cid:82) Σ µ = 1. Let ι : Σ × S → Y be an embedding. Let N ⊆ Y be a codimension-zerosubmanifold containing the image of ι , and such that the bundle E is trivializedover N in such a way that the connection θ in det( E ) induces the trivial productconnection in the determinant line bundle of our trivialization of E over N . Thismeans that connections in A can be understood as SU (2)-connections in E whenrestricted to N .Associated to this data, we can define a functionΦ : A → R which is invariant under the action of the gauge group G . For z ∈ Σ, we denoteby ι z : S → Y the circle t (cid:55)→ ι ( z, t ). A connection A ∈ A provides an SU (2)-connection over the image of ι . The holonomy Hol ι z ( A ) of A around the loop ι z (with variable starting point) is a section of the bundle of automorphisms of E withdeterminant 1 over the loop. Since χ is a class function, χ (Hol ι z ( A )) is well-defined.We can therefore define Φ( A ) = (cid:90) Σ χ (Hol ι z ( A )) µ ( z ) , (3)and this function is invariant under the action of the gauge group G . It dependson the data ( ι, χ, µ ) and a trivialization of the bundle over a codimension-zero sub-manifold N , but we will omit the latter from notation.We will have to work with a finite sequence of such embeddings, all supportedin a submanifold N of codimension zero over which the bundle E → N is trivial.For some n ∈ N , let ι k : S × Σ k → N ⊆ Y be a sequence of embeddings for k = 0 , . . . , n − ι k is disjoint from the interiorof the image of ι l for k (cid:54) = l . We also suppose class functions χ k : SU (2) → R corresponding to even, 2 π -periodic functions g k : R → R as above to be chosen, for k = 0 , . . . , n −
1, and we assume that µ k is a two-form on Σ k with support in theinterior of Σ k and integral 1. Just as in the case of (3), this data determines a finitesequence of functions Φ k : A → R , k = 0 , . . . , n − , and we are interested in the Morse homology of the functionCS + Ψ : B → R / Z , where Ψ = n − (cid:88) k =0 Φ k . (4) Definition 4.2.
We denote by R w Ψ ( Y ) the space of critical points [ A ] ∈ A / G ofthe function CS + Ψ : B → R / Z , where Ψ is specified by the holonomy perturbationdata { ι k , χ k } as above.If the holonomy perturbation data Ψ is chosen in a way such that R w Ψ ( Y ) doesnot contain equivalence classes of connections [ A ] such that A is reducible, then theconstruction for defining a Floer homology I w Ψ ( Y ) with generators given by criticalpoints of the perturbed Chern-Simons function CS+Ψ, and with differentials definedfrom negative gradient flow lines, goes through in the same way as in [9, 14]. Thiswill require additional small perturbations in order to make the critical points non-degenerate and in order to obtain transversality for the moduli spaces of flow-lines.In fact, we really have not done anything new compared to the constructions inthese references since the same perturbations already appear there for the sake ofobtaining transversality of the moduli spaces involved in the construction. The onlyslight difference is that in Floer’s work, the surfaces Σ k appearing in the definitionof the embeddings ι k are always chosen to be disks, whereas those used in the proofof Theorem 3.3 above, i.e. in [36, Theorem 4.2 and Proposition 5.3], the surfacesΣ k are all annuli.More specifically, for completeness, we recall a bit more on the implementationof the holonomy perturbations used in the work of the third author as needed inthe previous section for studying Y ( K ). Given a smoothly embedded path c from P = (0 , π ) to Q = ( π, π ) avoiding (0 ,
0) and ( π, φ t througharea-preserving maps of the pillowcase R ( T ) such that φ maps the straight line c := { β = π } from P to Q to the path c , and such that φ t fixes the four corner pointsof the pillowcase. Theorem 3.3. of [36] states that isotopies through area preservingmaps can be C -approximated by isotopies through finitely many shearing maps.For details on shearing maps we refer the reader to [36, Sections 2 and 3]. Theessential relationship is outlined in the following subsection, which we include forthe sake of clarity and completeness of our exposition.4.3. Review of holonomy perturbations and shearing maps.
We denote by R ( N ) the space of flat SU (2)-connections in the trivial SU (2)-bundle over N = S × Σ up to gauge equivalence, where Σ = S × I = S × [0 ,
1] is an annulus. Thetwo inclusion maps i − : S × ( S × { } ) → N and i + : S × ( S × { } ) → N induce OROIDAL HOMOLOGY SPHERES AND SU (2)-REPRESENTATIONS 19 restriction maps r − , r + : R ( N ) → R ( T ) to the representation varieties of the twoboundary tori, which are pillowcases. In this situation, we have that both r − and r + are homeomorphisms, and under the natural identification of these tori we have r − = r + .Now if χ is a class function as in Equation (2) above then instead of the flatnessequation F A = 0 for connections A on the trivial bundle over N , one may considerthe equation F A = χ (cid:48) (Hol l ( A )) µ, (5)where l = S × pt denote “longitudes” in N , where Hol l ( A ) is the holonomy of A along longitudes parametrised by points in Σ, and where χ (cid:48) : SU (2) → su (2) is thetrace dual of the derivative dχ of χ , and where µ is a 2-form with compact support inthe interior of Σ and (cid:82) Σ µ = 1. It can then be proved that Hol l ( A ) does not dependon the choice of longitude, and that solutions A of this equation are reducible, see[5, Lemma 4], and also [36, Proposition 2.1].If we denote by R χ ( N ) the solutions of Equation (5) up to gauge equivalence, thenwe still have two restriction maps r ± : R χ ( N ) → R ( T ). However, in this situationwe have the following relationship: Proposition 4.3.
The two restriction maps r ± are homeomorphisms and fit into acommutative diagram R χ ( N ) r − (cid:122) (cid:122) r + (cid:36) (cid:36) R ( T ) φ χ (cid:47) (cid:47) R ( T ) , (6) where φ is a shearing map that relates to χ as follows:If we write m − = { pt } × S × { } and m + = { pt } × S × { } for “meridians”given by the boundaries of Σ in { pt } × Σ , and if Hol m ± ( A ) = (cid:20) e iβ ± e − iβ ± (cid:21) , and Hol l ( A ) = (cid:20) e iα e − iα (cid:21) , (7) which we may suppose up to gauge equivalence, then we have φ χ (cid:18) αβ − (cid:19) = (cid:18) αβ − + f ( α ) (cid:19) , (8) where f : R → R is the derivative of the function g appearing in Equation (2).Here, ( α, β ± ) determine points in R ( T ) determined by Hol m ± ( A ) and Hol l ( A ) asin Equation (7) above. Equation (6) is essentially proved in [5, Lemma 4], and a proof also appears in[36, Proposition 2.1].Of course, one can iterate this construction: One may choose a finite collection ofdisjoint embeddings ι k : S × Σ into a closed three-manifold Y , and class functions χ k . The embeddings may chosen to be “parallel” in that the image of ι k correspondsto S × ( S × [ k, k + 1]) ⊆ S × ( S × [0 , n ]) ⊆ Y , but the role of “meridian” and“longitude” may be chosen arbitrarily in an SL ( Z ) worth of possible choices. Inthis case the restriction maps to the two boundary components of S × ( S × [0 , n ]) inthe diagram analogous to Equation (6) will be related by a composition of shearingmaps.4.4. Holonomy perturbations and the pillowcase.
The main application ofholonomy perturbations we have in mind is stated as Theorem 4.4 below. To putit into context, note first that for a non-trivial bundle, the critical space of theChern-Simons function R w ( Y ( K )) is a double cover of R ( K | c ), where c is thestraight line from (0 , π ) to ( π, π ) in the pillowcase, see [36, Proposition 5.1]. Ifwe choose holonomy perturbations associated to some data { ι k , χ k } n − k =0 as above,where the image of ι k corresponds to S × ( S × [ k, k + 1]) ⊆ S × ( S × [0 , n ]) ⊆ Y in a collar neighborhood of the Dehn filling torus in Y ( K ), then repeated use ofProposition 4.3 above will imply that for the holonomy perturbation Ψ determinedby the data { ι k , χ k } n − k =0 , the critical space of R w Ψ ( Y ( K )) will correspond to R ( K | c (cid:48) ),where c (cid:48) is the image of c under a composition of shearing maps φ n − ◦ · · · ◦ φ ,with “directions” determined by the embeddings ι k . (In Equation (8) we are dealingwith a shearing in direction (cid:18) (cid:19) , but we can pick any direction in Z .)The main point of [36, Theorem 4.2] is that the area-preserving maps of thepillowcase obtained by composition of shearing maps is C -dense in the space of allarea-preserving maps of the pillowcase, and this yields the following result. Theorem 4.4 (Theorem 4.2 and Proposition 5.3, [36]) . Let K be a knot in an SU (2) -cyclic integer homology three-sphere Y . Let c be an embedded path from (0 , π ) to ( π, π ) missing the other orbifold points of the pillowcase. Then, there exists anarbitrarily close path c (cid:48) and a holonomy perturbation Ψ along disjoint embeddings of S × ( S × I ) parallel to the boundary of a neighborhood of K such that R w Ψ ( Y ( K )) double-covers R ( K | c (cid:48) ) . (To see that we get a double-cover here we refer the reader to [36, Remark 1.2]).We only stress the fact that we must assume there are no reducible connectionsin R w Ψ ( Y ), since the presence of such solutions will result in a failure of the transver-sality arguments involved in the discussion.4.5. Invariance of instanton Floer homology.
The instanton Floer homologygroups I w ( Y ) and I w Ψ ( Y ), the latter being defined under the additional assumptionthat R w Ψ ( Y ) does not contain reducible connections, depend on additional data thatwe have already suppressed from notation, notably the choice of a Riemannianmetric on Y and holonomy perturbations just as defined above in order to achievetransversality. More explicitly, holonomy perturbations have already been implicitin the definition of instanton Floer homology unless the critical points of CS had OROIDAL HOMOLOGY SPHERES AND SU (2)-REPRESENTATIONS 21 been non-degenerate at the start and the moduli space defining the flow lines hadbeen cut out transversally. In Floer’s original work [14], and elaborated in moredetail in Donaldson’s book [9], invariance under the choice of Riemannian metricand the choice of holonomy perturbations follows from a more general concept,namely the functoriality of instanton Floer homology under cobordisms. See alsothe discussion in [23, Section 3.8.] Theorem 4.5 (Invariance under holonomy perturbations) . Suppose that the spaceof critical points R w Ψ ( Y ) of the perturbed Chern-Simons function 4 appearing in Def-inition 4.2 above does not contain equivalence classes of reducible connections. Thenthe associated instanton Floer homology groups I w ∗ ( Y ) and I w ∗ , Ψ ( Y ) are isomorphic.Sketch of Proof. The proof of this statement is standard, so we will describe a chainmap determining the isomorphism on homology and outline the ideas along whichthe result is proved.Slightly more generally, suppose we are dealing with a smooth map [0 , → C ∞ ( A , R ), s (cid:55)→ Γ( s ). We may suppose that this map is constant near 0 and 1. TheFloer differential counts flow lines of the Chern-Simons function, possibly suitablyperturbed. Instead of doing this, we may also consider the downward gradientflow equation of the time-dependent function CS +Γ( s ), where we extend Γ( s ) to amap ( −∞ , ∞ ) → C ∞ ( A , R ) which is constant Γ(0) on ( −∞ ,
0] and constant Γ(1)on [1 , ∞ ). If we are given critical points ρ of CS +Γ(0) and ρ of CS +Γ(1) of thesame index, then we consider a zero-dimensional moduli space M ρ ,ρ of connections A = { A ( t ) } t on E → R × Y of finite L -norm (inducing θ on det( E ), pulled backto R × Y ), such that the equation dAdt = − grad(CS +Γ( t ))( A ( t )) (9)holds on R × Y , where grad denots the L -gradient, and A limits to ρ and ρ in thelimit t → ±∞ , respectively. Finally, we also require that the moduli space M ρ ,ρ is cut out transversally.We require that the addition of the term − grad(Γ( t ))( A ( t )) to the gradient flowEquation (9) for the Chern-Simons function does not alter the linearized deformationtheory for A , see for instance [9, Sections 3 and 4]. Furthermore, we have to requirethat the Uhlenbeck compactification goes through with the perturbation we havein mind. It is shown in [9, Section 5.5] that both hold for the function Γ builtfrom holonomy perturbations as described in Equation (10). One essential featureis that the holonomy perturbation term appearing in the flow equation is uniformlybounded.A suitable interpolation between the holonomy perturbation data Γ(0) = 0 andΓ(1) = Ψ for Ψ as in Equation (4) is given, for instance, by the following formula. Suppose Ψ is determined by data { ι k , χ k } n − k =0 . Then for t ∈ [ kn , k +1 n ] we defineΓ( t ) = k − (cid:88) l =0 Φ l + β ( t − k/n )Φ k (10)for any k ∈ { , . . . , n − } . Here β : [0 , n ] → [0 ,
1] is a smooth function which is 0 ina neighborhood of 0 and 1 in a neighborhood of n .Now the moduli space M ρ ,ρ does not contain any reducibles, because if it did,then the limits ρ and ρ in R w ( Y ) and R w Ψ ( Y ), respectively, would also be reducible,and by our assumption and the setup for instanton Floer homology for admissiblebundles, this does not occur.One defines a linear map ζ : C w ( Y ) → C w Ψ ( Y ) of the underlying chain complexessuch that the “matrix entry” corresponding to the elements ρ ∈ C w ( Y ) and ρ ∈ C w Ψ ( Y ) is given by the signed count of the moduli space M ρ ,ρ , where the sign isdetermined in the usual way by the choice of a homology orientation . That ζ isa chain map follows from analyzing the compactification of suitable 1-dimensionalmoduli spaces, making use of Uhlenbeck compactification – no bubbling can occurhere due to the dimension of the moduli space – and the chain convergence discussedin [9, Section 5.1], together with suitable glueing results.That different interpolations yield chain homotopic chain maps follows from study-ing the compactification of ( − ζ defines a chain homotopy equivalence follows from the functoriality prop-erty: One may consider a further path Γ (cid:48) : [1 , → C ∞ ( A , R ) such that Γ (cid:48) (1) = Γ(1),similar as above. This defines a corresponding chain map ζ (cid:48) : C w Ψ → C w Γ (cid:48) (2) . On theother hand, one may concatenate the path Γ( t ) and the path Γ (cid:48) ( t ) and build acorresponding interpolation Γ (cid:48)(cid:48) : [0 , → C ∞ ( A , R ), resulting in a chain map ζ (cid:48)(cid:48) as above. A neck stretching argument then shows that ζ (cid:48)(cid:48) and ζ (cid:48) ◦ ζ are chainhomotopy equivalent, and hence induce the same maps on homology. In our sit-uation we take Γ (cid:48) (2) to be 0, meaning that this defines again the “unperturbed”chain complex C w ( Y ) (which, again, may contain some perturbations for the sakeof regularity omitted in our notation). One may finally interpolate between Γ (cid:48)(cid:48) andthe 0-term along a 1-dimensional family. Analyzing again the compactification ofsuitable ( − ζ (cid:48)(cid:48) and the identity. (cid:3) Remark . There is some confusion about invariance under “small” and “large”holonomy perturbations in the field. If one is given holonomy perturbation datafor which the underlying space of critical points and moduli spaces defining thedifferentials are already cut out transversally, then for small enough perturbationsthe same will still hold, and the resulting chain complexes will be isomorphic. This isdue to the fact that the condition of being cut out transversally is an open condition,
OROIDAL HOMOLOGY SPHERES AND SU (2)-REPRESENTATIONS 23 - Figure 5.
Left: The pattern representing the (2,1)-cable satelliteoperation. Right: The (2,1)-cable for the right handed trefoil. Theextra twisting appears as a consequence of the requirement that alongitude in S × D maps to the canonical longitude of the trefoil.expressed as the surjectivity of the deformation operators involved in the linearizedequation together with the Coulomb gauge fixing.If on the other hand, one is given a situation where the critical points and theunperturbed moduli spaces are not cut out transversally, then one needs to perturb,and even if these perturbations are chosen “small”, the resulting chain complexes willin general not be isomorphic but only chain homotopy equivalent. In this situation,the proof of invariance is really the same as proving the invariance under “large”perturbations, and already present in [14] and [9].5. Branched covers of prime satellite knots
In this section, we prove Corollary 1.5, establishing the existence of a non-trivial SU (2) representation for cyclic branched covers of prime satellite knots. We beginwith a definition of satellite knots. Definition 5.1.
Let P ⊂ S × D be an oriented knot in the solid torus. Consideran orientation-preserving embedding h : S × D → S whose image is a tubularneighborhood of a knot K so that S × {∗ ∈ ∂D } is mapped to the canonicallongitude of K . The knot h ( P ) is called the satellite knot with pattern P and companion K , and is denoted P ( K ). The winding number of the satellite is definedto be the algebraic intersection number of P with {∗} × D . See Figure 5 for anexample. Corollary 1.5.
Let K be a prime, satellite knot in S and let Σ( K ) be any non-trivial cyclic cover of S branched over K . Then π (Σ( K )) admits a non-trivial SU (2) representation.Proof. Let K be a prime satellite knot in S . If Σ( K ) is not an integer homologythree-sphere, then there is a non-trivial abelian representation. In the case whenΣ( K ) is an integer homology three-sphere, then by Theorem 1.1 it suffices to show that Σ( K ) is toroidal. Write K = P ( J ) and observe that if Σ( K ) is the d -foldcover of S branched over P ( J ), then there is a decomposition of Σ ( P ( J )) as theunion of Σ( S × D , P ), the d -fold cover of S × D branched over P , and a d -foldcovering space of the knot complement S \ N ( J ). The isomorphism type of thislatter covering space depends only on the greatest common divisor between d andthe winding number of S × D , see for example [31] or [24, pg. 220]. Since theexterior of J has incompressible boundary, the same is true of any cover. Therefore,we just need to show that Σ( D × S , P ) has incompressible boundary. We claimthe following. Let P be a non-trivial pattern knot in D × S which does notcorrespond to a connect-sum and which is not contained in an embedded B . Thenfor any cyclic branched cover over P , Σ( D × S , P ) has incompressible boundary.This claim is standard and proved in the lemma below for completeness. (cid:3) Lemma 5.2.
Let P be a non-trivial pattern knot in D × S which does not corre-spond to a connect-sum and which is not contained in an embedded B . Then forany cyclic branched cover over P , M = Σ( D × S , P ) has incompressible boundary.Proof. Suppose that γ is an essential loop on ∂M , which is nullhomotopic in M .Let G denote the group of covering transformations of M and consider the actionof G on the boundary. We first claim that γ can be isotoped on the boundary suchthat for each g ∈ G , either g ( γ ) ∩ γ = ∅ or g ( γ ) = γ . Of course, we only need torestrict to the subgroup of elements which fix the boundary component containing γ setwise. Further, since γ bounds in M , it is easy to see that the homology class in ∂M is fixed by all such elements. Because every finite group action on the torus isequivalent to the quotient of an affine action of the plane, the claim easily follows.Now, because of this claim, and because the curve γ is disjoint from the lift of P ,the equivariant Dehn’s lemma [34] implies that there exists a disk D in M bounding γ such that for all g , either g ( D ) ∩ D = ∅ or g ( D ) = D , and furthermore, D istransverse to the lift of the branch set. Consider the (possibly disconnected) surfaceΣ = (cid:83) g ∈ G g ( D ). Then, Σ /G is a collection of disks in D × S and Σ → Σ /G isa branched cover (although some components of Σ may have trivial branch locus).Furthermore, each component of the boundary of Σ /G is an essential curve on theboundary of the solid torus. For homology reasons, it is necessarily a meridionalcurve on the solid torus and each component of Σ /G is a meridional disk. (Thecomponents cannot have any other topology, since a disk can only cover/branchcover another disk.) Now, if any component of Σ /G does not intersect P , then wecan cut D × S along one of these disks, and see that P is contained in B and wehave a contradiction. If some component of Σ /G does intersect P , it must intersectin exactly one point, since a disk cannot be such a cyclic branched cover over a diskwith more than one branch point. (Here we are using that the branch points allcorrespond to intersections of P with the disk.) In other words, P is the patternfor a connect-sum, and again we have a contradiction. This proves the claim andcompletes the proof of the lemma. (cid:3) OROIDAL HOMOLOGY SPHERES AND SU (2)-REPRESENTATIONS 25 References [1] J. A. Baldwin and S. Sivek. Stein fillings and SU(2) representations.
Geom. Topol. , 22(7):4307–4380, 2018.[2] M. Boileau and S. Boyer. Graph manifold Z -homology 3-spheres and taut foliations. J. Topol. ,8(2):571–585, 2015.[3] F. Bonahon and L. C. Siebenmann. The characteristic toric splitting of irreducible compact3-orbifolds.
Math. Ann. , 278(1-4):441–479, 1987.[4] S. Boyer and A. Nicas. Varieties of group representations and Casson’s invariant for rationalhomology 3-spheres.
Trans. Amer. Math. Soc. , 322(2):507–522, 1990.[5] P. J. Braam and S. K. Donaldson. Floer’s work on instanton homology, knots and surgery. In
The Floer memorial volume , volume 133 of
Progr. Math. , pages 195–256. Birkh¨auser, Basel,1995.[6] J. Conway and B. Tosun. Mazur-type manifolds with L -space boundary. Math. Res. Lett. ,27(1):35–42, 2020.[7] C. Cornwell, L. Ng, and S. Sivek. Obstructions to Lagrangian concordance.
Algebr. Geom.Topol. , 16(2):797–824, 2016.[8] M. Culler. Lifting representations to covering groups.
Adv. in Math. , 59(1):64–70, 1986.[9] S. K. Donaldson.
Floer homology groups in Yang-Mills theory , volume 147 of
Cambridge Tractsin Mathematics . Cambridge University Press, Cambridge, 2002. With the assistance of M.Furuta and D. Kotschick.[10] E. Eftekhary. Bordered Floer homology and existence of incompressible tori in homologyspheres.
Compos. Math. , 154(6):1222–1268, 2018.[11] Y. Eliashberg. Topological characterization of Stein manifolds of dimension > Internat. J.Math. , 1(1):29–46, 1990.[12] R. Fintushel and R. J. Stern. Instanton homology of Seifert fibred homology three spheres.
Proc. London Math. Soc. (3) , 61(1):109–137, 1990.[13] A. Floer. An instanton-invariant for 3-manifolds.
Comm. Math. Phys. , 118(2):215–240, 1988.[14] A. Floer. Instanton homology and Dehn surgery. In
The Floer memorial volume , volume 133of
Progr. Math. , pages 77–97. Birkh¨auser, Basel, 1995.[15] S. Fukuhara and N. Maruyama. A sum formula for Casson’s λ -invariant. Tokyo J. Math. ,11(2):281–287, 1988.[16] C. M. Gordon. Dehn surgery and satellite knots.
Trans. Amer. Math. Soc. , 275(2):687–708,1983.[17] J. Hempel. Residual finiteness for 3-manifolds. In
Combinatorial group theory and topology(Alta, Utah, 1984) , volume 111 of
Ann. of Math. Stud. , pages 379–396. Princeton Univ. Press,Princeton, NJ, 1987.[18] R. Kirby. Problems in Low-Dimensional Topology, 1995.[19] P. Kronheimer and T. Mrowka. Knots, sutures, and excision.
J. Differential Geom. , 84(2):301–364, 2010.[20] P. Kronheimer and T. Mrowka. Knots, sutures, and excision.
J. Differential Geom. , 84(2):301–364, 2010.[21] P. B. Kronheimer and T. S. Mrowka. Dehn surgery, the fundamental group and SU(2).
Math.Res. Lett. , 11(5-6):741–754, 2004.[22] P. B. Kronheimer and T. S. Mrowka. Witten’s conjecture and property P.
Geom. Topol. ,8:295–310, 2004.[23] P. B. Kronheimer and T. S. Mrowka. Knot homology groups from instantons.
J. Topol. ,4(4):835–918, 2011. [24] C. Livingston and P. Melvin. Abelian invariants of satellite knots. In
Geometry and topology(College Park, Md., 1983/84) , volume 1167 of
Lecture Notes in Math. , pages 217–227. Springer,Berlin, 1985.[25] A. Lubotzky and A. R. Magid. Varieties of representations of finitely generated groups.
Mem.Amer. Math. Soc. , 58(336):xi+117, 1985.[26] K. Motegi. Haken manifolds and representations of their fundamental groups in SL(2 , C ). Topology Appl. , 29(3):207–212, 1988.[27] Y. Ni. Nonseparating spheres and twisted Heegaard Floer homology.
Algebr. Geom. Topol. ,13(2):1143–1159, 2013.[28] G. Perelman. The entropy formula for the Ricci flow and its geometric applications, 2002.[29] G. Perelman. Finite extinction time for the solutions to the Ricci flow on certain three-manifolds, 2003.[30] G. Perelman. Ricci flow with surgery on three-manifolds, 2003.[31] H. Seifert. On the homology invariants of knots.
Quart. J. Math., Oxford Ser. (2) , 1:23–32,1950.[32] S. Sivek and R. Zentner. A menagerie of SU (2)-cyclic 3-manifolds, 2019. arXiv:1910.13270.[33] A. Weil. Remarks on the cohomology of groups. Ann. of Math. (2) , 80:149–157, 1964.[34] S.-T. Yau and W. H. Meeks, III. The equivariant loop theorem for three-dimensional manifoldsand a review of the existence theorems for minimal surfaces. In
The Smith conjecture (NewYork, 1979) , volume 112 of
Pure Appl. Math. , pages 153–163. Academic Press, Orlando, FL,1984.[35] R. Zentner. A class of knots with simple SU (2)-representations. Selecta Math. (N.S.) ,23(3):2219–2242, 2017.[36] R. Zentner. Integer homology 3-spheres admit irreducible representations in SL(2 , C ). DukeMath. J. , 167(9):1643–1712, 2018.
Department of Mathematics, North Carolina State University, Raleigh, NC 27607
Email address : [email protected] University of Notre Dame, Department of Mathematics, Notre Dame, IN 46556,USA.
Email address : [email protected] Fakult¨at f¨ur Mathematik, Universit¨at Regensburg, 93040 Regensburg, Germany
Email address ::