TOPICS IN KNOT THEORY: ON GENERALIZED CROSSING CHANGES AND THE ADDITIVITY OF THE TURAEV GENUS By Cheryl Lyn Jaeger Balm A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Mathematics – Doctor of Philosophy 2013 ABSTRACT TOPICS IN KNOT THEORY: ON GENERALIZED CROSSING CHANGES AND THE ADDITIVITY OF THE TURAEV GENUS By Cheryl Lyn Jaeger Balm We first study cosmetic crossing changes and cosmetic generalized crossing changes in knots of genus one, satellite knots, and knots obtained via twisting operations on standardly embedded tori in the knot complement. As a result, we find obstructions to the existence of cosmetic generalized crossing changes in several large families of knots. We then study Turaev surfaces and use decomposing spheres to analyze the additivity of the Turaev genus for the summands of composite knots with Turaev genus one. ACKNOWLEDGMENTS I would like to thank my advisor, Effie Kalfagianni, for all her help and support. I am grateful to my dad Timothy Balm, my husband Thomas Jaeger and my dear friends Daniel Smith and Sara Vredevoogd for always having faith in me. I would also like to thank Bob Bell, Scott Carter, Chris Cornwell, Ron Fintushel, Stefan Friedl, Chris Hays, Teena Gerhardt Hedden, Matt Hedden, Mark Powell, Ben Schmidt and Luke Williams. iii TABLE OF CONTENTS LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii KEY TO SYMBOLS AND ABBREVIATIONS . . . . . . . . xi Chapter 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Generalized crossing changes . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Additivity of the Turaev genus . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 7 Chapter 2 Preliminaries . . . . . . . 2.1 Surfaces and 3-manifolds . . . . . 2.2 Seifert surfaces and the Alexander 2.3 Alternating knots . . . . . . . . . 2.4 Classification of knots . . . . . . 2.5 Twisted knots and braids . . . . . . . . . . . . . . . . . . . polynomial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 11 13 15 17 18 Chapter 3 Crossing changes and genus-one knots 3.1 Crossing disks and Seifert surface . . . . . . . . . 3.2 Obstructing cosmetic crossings in genus-one knots 3.3 Crossing changes and double branched covers . . 3.4 S-equivalence of Seifert matrices . . . . . . . . . . 3.5 Pretzel knots . . . . . . . . . . . . . . . . . . . . 3.6 Genus-one knots with low crossing number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 21 23 26 29 33 34 Chapter 4 Crossing changes and embedded tori 4.1 Essential tori . . . . . . . . . . . . . . . . . . . 4.2 Preliminary results . . . . . . . . . . . . . . . . 4.3 Obstrucing cosmetic crossings in satellite knots 4.4 Cosmetic crossings and twisting operations . . . 4.5 Another obstruction in satellite knots . . . . . . 4.6 Whitehead doubles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 37 38 42 51 56 57 Chapter 5 On the additivity of the Turaev genus 5.1 Constructing the Turaev surface . . . . . . . . . . 5.2 Turaev surfaces and decomposing spheres . . . . . 5.3 Decomposing spheres in standard position . . . . 5.4 Composite knots with Turaev genus one . . . . . 5.5 Turaev genus one and additivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 60 61 65 69 76 iv . . . . . . BIBLIOGRAPHY . . . . . . . . . . . . . v . . . . . . . . . . . . . . . . . 94 LIST OF TABLES Table 3.1 Genus-one knots with at most 12 crossings. . . . . . . . . . . . . . . 35 Table 5.1 All possible configurations of C− and the bubbles of T− from Theorem 5.5.9. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 vi LIST OF FIGURES Figure 1.1 An example of a crossing change in a knot diagram. . . . . . . . . . 2 Figure 1.2 The crossing on the left has sign +1 and the crossing on the right has sign −1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 A positive crossing is changed to a negative crossing by inserting one full twist in the positive (right-hand) direction. . . . . . . . . . . . . 3 Figure 1.4 Examples of nugatory crossing circles. . . . . . . . . . . . . . . . . . 4 Figure 1.5 The A and B smoothings of a crossing. . . . . . . . . . . . . . . . . 8 Figure 1.6 A saddle at a crossing of P in the construction of the Turaev surface TP , where the projection P is shown in gray. . . . . . . . . . . . . . 8 Figure 1.3 Figure 2.1 Possible decomposition sites for a composite alternating link diagram. 16 Figure 2.2 The four-valent graph P . . . . . . . . . . . . . . . . . . . . . . . . . 16 Figure 2.3 A follow-swallow companion torus for a composite knot. . . . . . . . 17 Figure 2.4 (A) The pattern knot for any positive Whitehead double D+ (K, n). (B) The knot D+ (K, −2) for the left-handed trefoil K. . . . . . . . 18 Figure 2.5 The generators σ1 and σ2 of B3 . . . . . . . . . . . . . . . . . . . . . 19 Figure 2.6 A fibered 5-braid K. By inserting q full twists at the shaded meridional disk, one obtains a twisted fibered knot K3,q . . . . . . . . . . . 20 Figure 3.1 The crossing arc α = F ∩ D. . . . . . . . . . . . . . . . . . . . . . . 22 Figure 3.2 A genus-one surface F with generators a1 and a2 of H1 (F ) and a non-separating arc α. . . . . . . . . . . . . . . . . . . . . . . . . . . 24 vii Figure 3.3 On the left is a diagram for P (p, q, r) with p, q and r positive. Note that each of p, q and r denote the total number of crossings in the corresponding twist region. On the right is the pretzel knot P (3, 3, −3). 33 Figure 4.1 A knotted 3-ball B inside of a solid torus with disks D, D ⊂ ∂B. . . 41 Figure 4.2 An example of an unknotted torus containing a crossing circle L which bounds a crossing disk for the knot K ∼ K#U . . . . . . . . . . . . . = 43 On the left is the solid torus V , cut into two solid tori by the annulus A. On the right is a diagram depicting the construction of X from the proof of Theorem 4.3.1. . . . . . . . . . . . . . . . . . . . . . . . 47 Figure 5.1 The construction of the ribbon graph GP from a knot projection P . 61 Figure 5.2 A bubble at a crossing of P containing a saddle-shaped disk where the decomposing sphere A meets the interior of the bubble. . . . . . 62 (A) A projection P illustrating the alternating property. (B) The corresponding 4-valent graph G. . . . . . . . . . . . . . . . . . . . . 64 The rectangle R and the surgery curve µ from the proof of Lemma 5.3.3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 If C ⊂ C+ bounds a disk on T+ and has w+ (C) = P S, then C must contribute a curve to C− , which violates Lemma 5.3.3. . . . . . . . . 69 Square depictions of T+ and T− as seen from M+ . The thicker lines are components of C± and the cirlces are bubbles on T± with overand undercrossings shown. . . . . . . . . . . . . . . . . . . . . . . . 71 The curves C− , C and C+ from Cases 1 and 2 of the proof of Corollary 5.4.4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 Overstrands and understrands of K on T± , denoted by o and u, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 Figure 5.9 The components of C− and C+ from Theorem 5.5.5. . . . . . . . . . 79 Figure 5.10 The decomposition of K = Ka #Kb from Theorem 5.5.5. The shaded regions contain alternating projections of subtangles of Ka and Kb . . 80 Figure 4.3 Figure 5.3 Figure 5.4 Figure 5.5 Figure 5.6 Figure 5.7 Figure 5.8 viii Figure 5.11 Figure 5.12 Figure 5.13 Figure 5.14 Figure 5.15 Figure 5.16 Figure 5.17 The A- and B-disks of T locally at the annulus Λ before and after the decomposition of K in Theorem 5.5.5. . . . . . . . . . . . . . . . 81 (A) An example of the annulus Λa from Theorem 5.5.6, which, in this example, meets six bubbles. The overcrossings of the a-bubbles are labeled as ai . The b-bubbles are labeled as Bi . (B) A diagram of Ka which is contained in Λa away from the undercrossings of the b-bubbles, which are labeled as bi . Each of the rectangular regions contains an alternating subtangle of Ka , and the ai are now the connecting arcs of the diagram. (C) The induced alternating projection of Ka . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 The components of C− and the regions of T− − η(C− ) from Theorem 5.5.7. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 (A) An example of the annulus Λb and the disk D from Theorem 5.5.7. The a-bubbles, other than B1 and B2 , are labeled as Ai , and the overcrossings of the b-bubbles are labeled as bi . (B) A diagram of Kb contained (mostly) in Λb ∪D. The bi are now connecting arcs of the diagram, and each of the rectangular regions contains an alternating subtangle. The ai and ˜i correspond to the overcrossings of the Ai b and Bi , respectively. (C) The induced alternating projection of Kb . . 84 (A) An example of the punctured annulus Λa from Theorem 5.5.7. The overcrossings of the a-bubbles, other than B1 and B2 , are labeled as ai . (B) A projection of Ka contained (mostly) in Λa . The ai are now connecting arcs in the diagram, as are α and the overcrossings of the Bi , which are labeled ˜i . Each of the rectangular regions contains b an alternating subtangle. (C) A diagram for Ka . . . . . . . . . . . . 85 (A) A simplification of the diagram from Figure 5.15(C). (B) A diagram for the all-A smoothing of Ka . The thicker lines are each Seifert circles. The lighter line segments represent crossings from the knot diagram which will be edges in the ribbon graph. The shaded gray region contains additional Seifert circles and edges from the alternating portion of the knot diagram. (C) A genus-one ribbon graph for Ka . The rectangular region contains a planar sub-ribbon graph. The thicker gray lines each represent two or more planar edges connecting the vertices a, b and c to vertices in the planar portion of the graph. 86 The curves of C± from Corollary 5.5.8 with the bubbles of T± . . . . 86 ix Figure 5.18 Figure 5.19 An example of the annuli of T+ − η(C+ ) from Corollary 5.5.8. The corresponding diagram of Kb is alternating, and the diagram of Ka has Turaev genus 1 by the arguments of Figure 5.16. . . . . . . . . . 87 (A) The configuration of T− from the first row of Table 5.1. (B) A diagram of Ka Kb on the torus given by this coniguration. The shaded boxes contain (possibly trivial) alternating sub-diagrams. (C) The resulting diagram of Ka Kb on R2 , which shows that Ka and Kb are alternating knots. . . . . . . . . . . . . . . . . . . . . . . . . 90 x KEY TO SYMBOLS AND ABBREVIATIONS A decomposing sphere which is preferred or in standard position Bp the pth braid group C± , C± A ∩ T or A ∩ T for decomposing spheres A, A and a Turaev surface T D+ (K, n) Whitehead double of K with n full twists and a positive clasp D− (K, n) Whitehead double of K with n full twists and a negative clasp ∆(·, ·) minimal geometric instersection number ∆K (t) Alexander polynomial of K ∆2 p central element of Bp given by one full twist ∂ boundary G ribbon graph g(·) genus galt (·) alternating genus gDA (·) disk-alternating genus gT (·) Turaev genus int(M ) M − ∂M K a knot K class of knots which are known not to admit cosmetic generalized crossing changes K(q) = KL (q) the knot obtained from K via an order-q generalized crossing change at the crossing circle L Kn = Kn,V twist knot of K xi Kp,q twist braid obtained via q full twists on p strands of the closed braid K lk(·, ·) linking number M (q) the 3-manifold obtained from S 3 by order-q Dehn surgery along a crossing circle L M± handlebody bounded by T± lying outside of the Turaev surface T ML S 3 − η(L) for a knot or link L η(·) closed regular neighborhood P projection of a knot corresponding to a Turaev surface Sn n − dimensional sphere T a Haken system T± Turaev surface T with T ∩ {bubbles} replaced by the upper/lower hemispheres of the bubbles τ (·) Haken number U unknot V± S 3 − M± w(·, ·) winding number w± (·) cyclic word which records the saddles and punctures of a curve of C± wrap(·, ·) wrapping number χ(·) Euler characteristic YK double branched cover of S 3 branched over K Z[t±1 ] the ring Laurent polynomials with integer coefficients Zm = Z/mZ cyclic abelian group of order m := defined to be equal xii ∼ = isomorphic . = equivalent up to multiplication by a unit # connect sum ∼ S − equivalent matrices ≈ congruent matrices ·|X a function restricted to a domain X · ˆ · closure of a manifold closure of a braid xiii Chapter 1 Introduction The primary goal of this thesis is to investigate relationships between knots and surfaces in R3 and use these relationships to explore properties of knots. Much of the work done here has been motivated by the nugatory crossing conjecture, Problem 1.58 on Kirby’s list [1], which asks when a crossing change in a knot diagram does not change the underlying knot type. Another major goal of this work is to study the Turaev genus, a fairly new knot invariant first defined in 2008 by Dasbach, Futer, Kalfagianni, Lin and Stoltzfus [13]. We are especially interested in deciding whether the Turaev genus is additive under the operation of connect summing. Throughout this work, we will be considering oriented tame knots and surfaces smoothly embedded in R3 ⊂ S 3 , even when they are not explicitly stated as such. 1.1 Generalized crossing changes A fundamental open question in knot theory is the question of when a crossing change on an oriented knot changes the isotopy class of the knot. Traditionally, a crossing change is defined in terms of a knot diagram by changing the over- and understrand of a single crossing in the diagram, as illustrated in Figure 1.1. We would like to study the effects of crossing changes without restricting ourselves to any particular diagram of the knot in question. To this end, we define a crossing disk for an 1 Figure 1.1: An example of a crossing change in a knot diagram. Figure 1.2: The crossing on the left has sign +1 and the crossing on the right has sign −1. oriented knot K ⊂ S 3 to be an embedded disk D ⊂ S 3 such that K intersects int(D) twice with zero algebraic intersection number. In other words, K meets D exactly twice and in opposite directions. A crossing circle is the boundary of a crossing disk. A crossing change in K can be achieved by inserting a full twist in the appropriate direction in K at a crossing disk D. More precisely, choose an orientation on K and let C be a crossing in some diagram of K. Regardless of the choice of orientation, C is either a positive or negative crossing, as defined in Figure 1.2. Let L be a crossing circle for K which encircles the crossing C. If C has sign ε, where ε = ±1, then changing the sign of C to −ε is equivalent to inserting one full twist of sign ε at L, where a twist is positive if it is a right-hand twist and negative if it is a left-hand twist. This process is illustrated in Figure 1.3. Throughout this paper we will let η(·) denote the closed regular neighborhood of a man2 Figure 1.3: A positive crossing is changed to a negative crossing by inserting one full twist in the positive (right-hand) direction. ifold and ML := S 3 − η(L) for any knot or link L. Given a knot K, MK is a manifold with torus boundary ∂η(K). Given s ∈ Q, a slope-s Dehn filling of MK is obtained by gluing a solid torus T to ∂MK so that the meridian of T has slope s on ∂MK . Similarly, s-Dehn surgery on S 3 at K is obtained by removing a neighborhood of K from S 3 and performing a slope-s Dehn filling of MK . Inserting one full twist of sign ε at a crossing circle L is equivalent to performing (−ε)-Dehn surgery on S 3 at L. (See Chapter 6 of [33] for details.) More generally, if we perform (−1/q)-Dehn surgery along the crossing circle L for some q ∈ Z − {0}, we twist K q times at the crossing circle in question. We will call this an order-q generalized crossing change denoted by KL (q), or simply K(q) if there is no chance of confusion regarding the crossing circle L. Note that if q is positive, then we give K q righthand twists when we perform (−1/q)-surgery, and if q is negative, we give K q left-hand twists. Definition 1.1.1. We will use the term crossing change to refer to a crossing change in the diagrammatic sense, which is equivalent to a generalized crossing change of order ±1, and the term generalized crossing change to refer to such changes of any order q = 0. We are interested in studying generalized crossing changes which do not change the 3 Figure 1.4: Examples of nugatory crossing circles. underlying knot. To this end, we have the following definition. Definition 1.1.2. A crossing of K and its corresponding crossing circle L are called nugatory if L bounds an embedded disk in S 3 −η(K). Examples of nugatory crossing circles are shown in Figure 1.4. Clearly a generalized crossing change of any order at a nugatory crossing of K yields a knot isotopic to K. A generalized crossing change on K is called cosmetic if the crossing change yields a knot isotopic to K and is performed at a crossing of K which is not nugatory. We will also use the term cosmetic to refer to the corresponding crossing circle and the crossing itself. The following question, often referred to as the nugatory crossing conjecture, is Problem 1.58 on Kirby’s list [1]. Problem 1.1.3. Does there exist a knot K which admits a cosmetic crossing change? Conversely, if a crossing change on a knot K yields a knot isotopic to K, must the crossing be nugatory? Since a crossing change is the same as an order-(±1) generalized crossing change, one can ask the following stronger question concerning cosmetic generalized crossing changes. Problem 1.1.4. Does there exist a knot K which admits a cosmetic generalized crossing change of any order? 4 Problem 1.1.3 was answered for the unknot by Scharlemann and Thompson when they showed that the unknot admits no cosmetic crossing changes in [37] using work of Gabai [14]. The proof in [37] can be easily generalized to show that the unknot admits no cosmetic generalized crossing changes of any order. It has been shown by Kalfagianni that the answer to Problem 1.1.4 is no for fibered knots [18] and by Torisu that the answer is no for 2-bridge knots [43]. Torisu also reduces Problem 1.1.4 to the case where K is a prime knot in [43]. More precisely, suppose that the knot K is composite and admits a cosmetic generalized crossing change. Then the corresponding crossing disk meets exactly one summand of K. In Chapters 3 and 4 we will use crossing circles to find obstructions to cosmetic generalized crossing changes in several families of knots. Chapter 3 is based on results obtained jointly by the author with Friedl, Kalfagianni and Powell in [4], with the main difference being that many of the results in [4] have been generalized here to include cosmetic generalized crossing changes and not just crossing changes of order-(±1). The results of Chapter 4 first appeared in work by the author in [3] and joint work with Kalfagianni in [5]. In Chapter 3 we focus on genus-one knots. Two of the main results of this chapter are the following. Theorem 3.2.1. Let K be a genus-one knot. If K admits a cosmetic generalized crossing change, then K is algebraically slice. In particular, there is a linear polynomial f (t) ∈ Z[t] . such that the Alexander polynomial of K is of the form ∆K (t) = f (t)f (t−1 ). Theorem 3.3.1. Let K be a genus-one knot and let YK denote the double branched cover of S 3 branched over K. If K admits a cosmetic crossing change, then the homology group H1 (YK ; Z) is a finite cyclic group. These results allow us to find obstructions to cosmetic generalized crossing changes in 5 certain pretzel knots and knots with low crossing number, including those in the following theorem and corollaries. Theorem 3.6.2. Let K be a genus-one knot that has a diagram with at most 12 crossings. Then K admits no cosmetic crossing changes. Corollary 3.6.3. Let K be a genus-one knot that has a diagram with at most 12 crossings. If K admits an order-q cosmetic generalized crossing change, then one of the following must be true. 1. K = 946 and q = 3n for some n ∈ Z 2. K = 11n139 Corollary 3.5.1. The pretzel knot P (p, q, r) with p, q and r odd does not admit a cosmetic generalized crossing change of any order if pq + qr + pr = −m2 , for every m ∈ Z. In Chapter 4 we turn our attention to potential cosmetic generalized crossing changes in satellite knots and other knots embedded in solid tori. Our first main result of this chapter is the following. Theorem 4.3.1. Suppose K is a satellite knot which admits a cosmetic generalized crossing change of order q with |q| ≥ 6. Then K admits a pattern knot K which also has an order-q cosmetic generalized crossing change. This leads us to a very nice corollary. Corollary 4.3.7. If there exists a knot admitting a cosmetic generalized crossing change of order q with |q| ≥ 6, then there must be such a knot which is hyperbolic. 6 We then go on to study knots contained in standardly embedded solid tori, including twist knots and fibered m-braids. This leads to the following two results, in which K denotes the class of knots which are known not to admit cosmetic generalized crossing changes of any order, and Kn,V and Kp,q are defined in Definitions 2.5.1 and 2.5.2, respectively. Theorem 4.4.4. Let K ∈ K be contained in a standardly embedded solid torus V with w(K , V ) = wrap(K , V ) ≥ 3. Then, for every n ∈ Z, the twist knot K n,V does not admit a cosmetic generalized crossing change of any order. Corollary 4.4.5. Let K be a fibered m-braid with m ≥ 3. Then for every 3 ≤ p ≤ m and q ∈ Z, there is no twisted fibered braid Kp,q which admits a cosmetic generalized crossing change of any order. We close Chapter 4 with several results that obstruct cosmetic generalized crossing changes in certain classes of Whitehead doubles, including the following. Corollary 4.6.3. Let K be a prime knot that is not a cable knot. Then no Whitehead double of K admits a cosmetic generalized crossing change of any order. 1.2 Additivity of the Turaev genus Another question of great interest in knot theory is whether the Turaev genus of a knot is additive under the operation of connect sum. The Turaev genus is a generalization of the concept of alternating knots. In [13], the construction of a Turaev surface for a knot K ⊂ S 3 is given as follows. Let P be a projection of K onto R2 such that P has no nugatory crossings. At each crossing of P we can consider the A and B smoothings of the crossing, as defined in Figure 7 A B Figure 1.5: The A and B smoothings of a crossing. Figure 1.6: A saddle at a crossing of P in the construction of the Turaev surface TP , where the projection P is shown in gray. 1.5. Thicken the projection plane to R2 × [−1, 1] so that P lies in R2 × {0}. Away from any crossings, thicken P to P × [−1, 1]. In a neighborhood of each crossing, insert a saddle so that the boundary circles in R2 × {1} correspond to the all-A smoothing of P and the boundary circles in R2 × {−1} correspond to the all-B smoothing. (See Figure 1.6.) Finally, cap off each boundary component with a disk to obtain a Turaev surface TP for K. Lemma 4.1 of [13] shows that for all P , TP is unknotted in S 3 , and Lemma 4.5 of [13] shows that K has an alternating projection P on TP coming directly from P and the construction of TP . Further, P has no nugatory crossings on TP in the sense that there is no simple closed curve on TP which meets P exactly once at a crossing of P and bounds a disk on TP . Definition 1.2.1. The Turaev genus of a knot K is given by gT (K) := min{g(TP ) | TP is a Turaev surface coming form a projection P of K} 8 where g(·) denotes the genus of a surface. If gT (K) = 0, then K has an alternating projection on the sphere S 2 , which clearly can happen if and only if K is alternating. Note that although a knot K has an alternating projection on any Turaev surface for K, the Turaev genus of K is not the same as the minimal genus Heegard splitting of S 3 on which K has an alternating projection, which is called the alternating genus of K, since any Turaev surface for K must be obtained through the procedure described above and hence satisfies properties besides simply admitting an alternating projection. In particular, gT (K) is bounded below by the alternating genus of K. Although it is very natural to ask whether the Turaev genus is additive in the sense that gT (K1 #K2 ) = gT (K1 ) + gT (K2 ), very little is known toward answering this question. Abe has shown in [2] that gT (K1 #K2 ) = gT (K1 ) + gT (K2 ) if K1 and K2 are both in a class of knots called adequate knots by using the heavy machinery of Khovanov homology. As a first step towards answering the question of the additivity of the Turaev genus for all knots, we will study composite knots of Turaev genus one. Suppose K is such a knot and T is a torus which is a Turaev surface for K. If K = K1 #K2 , then there is a 2-sphere A which meets K exactly twice and decomposes K into K1 and K2 . Using arguments similar to those used by Menasco in [29], we can add a bubble to T at each crossing of P , as shown in Figure 5.2. W can then construct T+ and T− from T by replacing each disk of T inside a bubble by the upper and lower hemisphere of the bubble, respectively. By analyzing the intersections of A ∩ T± , we can study the Turaev genera of the summands K1 and K2 . Section 5.5 is devoted to such analysis when A and A ∩ T± are especially nice, as in the following corollary, where C− := A ∩ T− and galt and gDA are defined in Definition 5.5.3. 9 Corollary 5.5.8. Let K be a composite knot with gT (K) = 1 and let A be a preferred decomposing sphere for K with respect to the genus-one Turaev surface T such that A decomposes K into Ka #Kb . If no bubble of T meets A more than once, then either gT (Ka )+gT (Kb ) = 1 or |C− | = 1 and, up to relabelling, Ka is alternating and galt (Kb ) = gDA (K) = 1. The question of the additivity of the Turaev genus is especially interesting because of the role played by the Turaev genus in studying knot homologies. In 2000, Khovanov introduced an important new knot invariant refining the Jones polynomial, which is now referred to as Khovanov homology [20]. Then, in 2003, Ozsv´th, Szab´ and Rasmussen defined knot a o Floer homology, a knot invariant which categorifies the Alexander polynomial [31, 35]. For a knot K, both the width of the knot Floer homology and the width of the reduced Khovanov homology of K, denoted wHF (K) and wKh (K), respectively, are bounded above by the inequality gT (K) ≥ w∗ (K) − 1. (See [24, 12, 27] for details.) We will say a knot is thin if its width in both of these homologies is one. The class of quasi-alternating knots was defined in [32], and all knots in this class are known to be thin [26]. There is an infinite family of quasi-alternating knots with Turaev genus one [11]. For these knots, gT (K) − w∗ (K) = 0 and the above inequality is not sharp. This leads one to ask whether gT (K) − w∗ (K) can be arbitrarily large for one or both of these homology theories. In particular, are there quasi-alternating knots with arbitrarily high Turaev genus? Since the connect sum of quasi-alternating knots is quasi-alternating, if the Turaev genus is shown to be additive, this would give a family of knots {Kn } such that gT (Kn ) − w∗ (Kn ) = n − 1 for each positive integer n. 10 Chapter 2 Preliminaries 2.1 Surfaces and 3-manifolds Throughout this work we will be using orientable surfaces embedded in 3-manifolds to study properties of knots. Hence all knots, links, surfaces and 3-manifolds will be assumed to be orientable, even if they are not specifically stated as such. A 3-manifold M is called irreducible if any 2-sphere S 2 ⊂ M bounds a 3-ball on at least one side. Otherwise, we say M is reducible. A surface Σ ⊂ M is compressible if there is a disk D ⊂ M smoothly embedded such that D ∩ Σ = ∂D (where ∂ denotes the boundary of a manifold) and ∂D does not bound a disk in Σ. If no such compressing disk exists, then Σ is incompressible. Let C be a simple closed curve on some surface Σ and let A ⊂ Σ be an open annular neighborhood of C. A Dehn twist of Σ at C is a diffeomorphism φ : Σ → Σ such that φ is the identity map on Σ − A and φ|A is given by a full twist around C. More specifically, if A ∼ S 1 × (0, 1) is given the coordinates (eiθ , x), then φ|A : (eiθ , x) → (ei(θ+2πx) , x). = Given a closed surface T ⊂ M , by M cut along T we mean the 3-manifold M − η(T ). Similarly, given a surface Σ and another surface F ⊂ M with F ∩ Σ = ∂F , we can surger Σ at F by removing an open neighborhood of ∂F from Σ and gluing 2 copies of F to Σ along the newly created boundary components of Σ − η(∂F ). Much of the work in Chapter 4 utilizes tori in knot complements in S 3 . A torus T in a 3-manifold M is called essential if it is incompressible and is not boundary parallel, or 11 peripheral, in M . The manifold M is atoroidal if it does not contain any embedded essential tori. Definition 2.1.1. Given a compact, irreducible, orientable 3-manifold M , let T be a collection of disjointly embedded, pairwise non-parallel, essential tori in M , which we will call an essential torus collection for M . By Haken’s Finiteness Theorem (Lemma 13.2 of [17]) the number τ (M ) := max{|T | | T is an essential torus collection for M } is well-defined and finite, where |T | denotes the number of tori in T . We will call such a collection T with |T | = τ (M ) a Haken system for M . Note that any essential torus T ⊂ M is part of some Haken system T . A torus T ⊂ S 3 always bounds a solid torus on at least one side. If T cuts S 3 into two solid tori, then we say T is standardly embedded or unknotted in S 3 . Let V be a solid torus bounded by T . A meridian of T is a simple closed curve µ ⊂ T such that µ bounds a disk in V called a meridional disk. A longitude of T is a simple closed curve λ ⊂ T such that λ generates H1 (V ) ∼ Z. Let C be the core of the solid torus V . Then a longitude λ of T is = said to be canonical or standard if lk(λ, C) = 0. The torus T is unknotted if and only if a standard longitude on T bounds a disk in the complement of T . Given a connected surface Σ, define χ− (Σ) := max(0, −χ(Σ)), where χ(·) denotes the Euler characteristic. If Σ := n i=1 Σi is disconnected, then χ− (Σ) := n i=1 χ− (Σi ). Given a 3-manifold with boundary M and an element a ∈ H2 (M, ∂M ), we define the Thurston norm x : H2 (M, ∂M ) → Z by x(a) := min{χ− (Σ) | a = [Σ] ∈ H2 (M, ∂M )}. We say a surface (Σ, ∂Σ) ⊂ (M, ∂M ) is Thurston norm-minimizing if χ− (Σ) = x([Σ]). 12 An orientable manifold is called a Haken manifold if it is a compact, irreducible 3manifold that contains an orientable, incompressible surface. The following well-known result of Gabai will be used in Chapters 3 and 4. Theorem 2.1.2 (Gabai, Corollary 2.4 of [14]). Let M be a Haken manifold whose boundary is a nonempty union of tori. Let Σ be a Thurston norm-minimizing surface representing an element of H2 (M, ∂M ) and let P be a component of ∂M such that P ∩ Σ = ∅. Then with at most one exception (up to isotopy) Σ remains Thurston norm-minimizing in each manifold M (s) obtained by a slope-s Dehn filling of M along P . In particular Σ remains incompressible in all but at most one manifold obtained by a Dehn filling of P . 2.2 Seifert surfaces and the Alexander polynomial Given a knot K ⊂ S 3 , and Seifert surface for K is an orientable surface F ⊂ S 3 such that ∂F = K. The genus of K, denoted g(K) is then the minimal genus of a Seifert surface for K. Given a Seifert surface F , we can choose a basis {ai } for the first homology group H1 (F ) := H1 (F ; Z) where each generator ai is represented by an oriented simple closed curve ai on F . By choosing an orientation on F , we can define ai + to be the curve ai pushed ¯ ¯ ¯ slightly off F in the positive normal direction. The Seifert matrix V associated to F and {ai } is then given by Vi,j := lk(ai , aj + ). ¯ ¯ Definition 2.2.1. The Alexander polynomial of a knot K is a polynomial in Z[t±1 ] given . . by ∆K (t) = det(tV − V T ) where V is a Seifert matrix for K and = denotes equality up to multiplication by a unit in Z[t±1 ]. The knot determinant det(K) := |∆K (−1)|. The Alexander polynomial ∆K (t) is a well-defined invariant of K, but, in general, a knot K does not have a unique Seifert matrix. 13 Definition 2.2.2. Two integral square matrices V and V are S-equivalent, denoted V ∼ V , if V is obtained from V by a finite sequence of the following moves or their inverses. 1. Replacing V by P V P T for some integral unimodular matrix P . 2. A column expansion, where we replace the n × n matrix V with an (n + 2) × (n + 2) matrix of the form       V         u1 · · · un   0 ··· 0 0 0   . .  . .  . .    0 0     0 0    1 0 where u1 , . . . , un ∈ Z. 3. A row expansion, which is defined analogously to the column expansion, with the roles of rows and columns reversed. A Seifert matrix for a knot K is not unique since it depends on the choices of the Seifert surface F , the basis {ai } and the orientations. However, any two Seifert matrices for K are always S-equivalent. If K has a unique minimal-genus Seifert surface F , up to isotopy, and g(F ) = g, then a (2g ×2g) Seifert matrix for K depends only on the choice of basis for H1 (F ) and orientations. In this situation, any two (2g × 2g) Seifert matrices V and V for K are congruent, so V = P V P T for some integral unimodular matrix P , and we write V ≈ V . Given a Seifert surface F for a knot K, the Seifert form associated to F is the map ¯ θ : H1 (F ) × H1 (F ) → Z is given by θ(α, β) := lk(¯ , β + ), where α and β are simple closed α ¯ ¯ curves on F representing α and β, respectively. (See [23] for details.) 14 Definition 2.2.3. A knot K is called algebraically slice if it admits a Seifert surface F such that the Seifert form θ : H1 (F ) × H1 (F ) → Z vanishes on a half-dimensional summand of H1 (F ). This half-dimensional summand is called a metabolizer of H1 (F ). If F has genus one, then the existence of a metabolizer for H1 (F ) is equivalent to the existence of an essential oriented simple closed curve on F that has zero self-linking number. . If K is algebraically slice, then ∆K (t) is of the form ∆K (t) = f (t)f (t−1 ), where f (t) ∈ Z[t] is a linear polynomial. 2.3 Alternating knots An alternating knot diagram for a knot K is a projection P such that, as we traverse K, each time we come to a crossing of P , we alternate between meeting an overstrand and an understrand. A knot is alternating if it has an alternating diagram. Since the Turaev genus is a generalization of this property, the following proposition will be helpful in Chapter 5. Proposition 2.3.1. Given knots K1 and K2 , K1 #K2 is alternating if and only if both K1 and K2 are. Proof. One direction holds obviously since the connect sum of alternating diagrams is an alternating diagram. For the other direction, let K := K1 #K2 and assume K is alternating. Let P be a projection of K on R2 . We will call P a composite diagram if there is a simple closed curve S 1 ⊂ R2 meeting P transversely exactly twice and not at crossings such that each of the components of R2 − η(S 1 ) contains a nontrivial summand of K. If P is not composite, then P is called a prime diagram. In [29], Menasco shows that for any alternating knot K with an alternating projection P such that P has no nugatory crossings, K is composite if and only if P is a composite 15 (A) (C) (B) (D) Figure 2.1: Possible decomposition sites for a composite alternating link diagram. v1 v2 v3 e1 e2 vn e0 R en v0 Figure 2.2: The four-valent graph P . diagram. Let P be such a composite alternating diagram for K, and let K and K be two summands of K which are visible in P . At the site where we decompose K into K and K , the projection must look locally like one of the possibilities shown in Figure 2.1. In each of the Figures 2.1(C) and 2.1(D), K and K are alternating. Consider Figure 2.1(A) and let K be the summand on the left. We will denote this projection of K by P . Thinking of P as a four-valent graph with over/under information at each of the vertices, let e0 be the edge of P which was created in decomposing K, and let R be the bounded region of R2 − η(P ) which is adjacent to e0 . (See Figure 2.2.) Label the other edges of P adjacent to R as e1 , ..., en , in order, and the vertices of P such that each ei is adjacent to vi and vi+1 (mod n) . Since each of the ei for 1 ≤ i ≤ n is an edge of P , it must be alternating. This causes a contradiction, since e0 contributes an undercrossing at both v0 and vn . Hence Figure 2.1(A) cannot occur. An identical argument applies to Figure 2.1(B), so K and K must 16 Figure 2.3: A follow-swallow companion torus for a composite knot. be alternating. We can iterate this argument to show that each prime summand of K is alternating and hence so are K1 and K2 . 2.4 Classification of knots In [42], Thurston classified all knots in S 3 as being either hyperbolic knots, torus knots, or satellite knots. Hyperbolic knots are those which admit a complete hyperbolic metric on their complements in S 3 . The torus knot Tp,q is the knot embedded on an unknotted torus with slope p/q with gcd(p, q) = 1. Definition 2.4.1. A knot K is a satellite knot if MK := S 3 − η(K) contains a knotted torus T which is essential in MK and K is contained in the solid torus V bounded by T in S 3 . Such a torus T is called a companion torus for K. Further, there exists a homeomorphism f : (V , K ) → (V, K), called the satellite map, where V is an unknotted solid torus in S 3 and K is contained in int(V ). The knot K is called a pattern knot for the satellite knot K. Any composite knot K = K1 #K2 is a satellite knot. This can be seen by choosing K1 as the pattern knot and K2 as the core of the companion torus, as illustrated in Figure 2.3. Such a companion torus for a composite knot is called a follow-swallow torus. Another well-known class of satellite knots are the Whitehead doubles. The n-twisted, 17 D+ (K, 0) K T T (A) (B) Figure 2.4: (A) The pattern knot for any positive Whitehead double D+ (K, n). (B) The knot D+ (K, −2) for the left-handed trefoil K. positively-clasped Whitehead double of K, denoted D+ (K, n) is obtained using the pattern knot K in Figure 2.4(A), where the core of the companion torus for D+ (K, n) is K and the canonical longitude of the unknotted torus T is sent to the longitude γn of T , where lk(K, γn ) = n. The n-twisted, negatively-clasped Whitehead double of K, written D− (K, n) is defined similarly, except both of the crossings in the pattern knot K are changed from positive to negative. This class of knots is the focus of Section 4.6. Another family of knots discussed in this work is that of cable knots. This class of knots contains all torus knots as well as satellite knots obtained from torus knots in the following way. Let K := Tp,q be a torus knot sitting on the unknotted torus T . By pushing K slightly into the solid torus bounded by T , K can be thought of as a pattern knot and the resulting satellite knot is a cable knot. 2.5 Twisted knots and braids Let K be a knot in a solid torus V . The winding number, w(K, V ), is the minimal algebraic intersection number of K with a meridional disk of V . Similarly, the wrapping number, wrap(K, V ), is the minimal geometric intersection of K with a meridional disk. If V is 18 σ1 = σ2 = Figure 2.5: The generators σ1 and σ2 of B3 . knotted, then K is a satellite knot. If V is unknotted, we have the following definition. Definition 2.5.1. Let K be a knot that is geometrically essential in a solid torus V , so every meridional disk of ∂V meets K at least once. Given n ∈ Z, let Kn,V be the knot obtained from K via n full twists along a meridional disk D ⊂ V such that the geometric intersection of D with K is at least two and and cannot be reduced by isotopy. We call such a disk a twisting disk, and Kn,V is called a twist knot of K. When there is no danger of confusion, we will simply write Kn instead of Kn,V . p−1 For p > 0, let Bp denote the p-string braid group. Then Bp is generated by {σi }i=1 where σi is given by a single crossing between the ith and (i + 1)th strand of a p-braid, and we will let ∆2 denote the full twist braid, which is the central element in Bp (See [6] for p details.) For example, the generators of B3 are shown in Figure 2.5, and ∆2 = (σ1 σ2 )3 . 3 ˆ Given a braid β, we will let β denote the knot or link obtained as the standard closure of β. ˆ Note that β depends only on the conjugacy class of β in Bp . Definition 2.5.2. A fibered m-braid is a closed braid on m strands which is also a fibered knot in the sense that the complement of the knot in S 3 fibers over S 1 . Given a fibered m-braid K with 2 ≤ p ≤ m and q ∈ Z, a new knot can be obtained by inserting a copy of 2q ∆p (q full twists) into p strings of K. We will call such a knot a twisted fibered braid and denote it by Kp,q . Note that the knot Kp,q is not unique and depends on the choice of the p strings which are twisted. 19 Figure 2.6: A fibered 5-braid K. By inserting q full twists at the shaded meridional disk, one obtains a twisted fibered knot K3,q . Twisted fibered braids are a family of twist knots, as illustrated in Figure 2.6. A wellstudied class of twisted fibered braids is produced from torus knots by inserting full twists along a number of strands. We call such knots twisted torus knots (see [8, 10]). n Definition 2.5.3. A braid β = σi 1 σi 2 · · · σin ⊂ Bp , where each i ∈ Z, is called homoge1 2 neous if both of the following hold. 1. Every σi occurs at least once for each 1 ≤ i ≤ p − 1. 2. The exponent of σi is of the same sign for each occurance of σi in β. In particular, a braid β is positive if β is homogeneous and all exponents are positive. The closure of a homogeneous braid is known to be fibered by [41], so by adding full twists to closed homogenous braids we may obtain broad classes of twisted fibered braids. 20 Chapter 3 Crossing changes and genus-one knots Recall that throughout this paper, ML := S 3 − η(L) for any knot or link L. We first prove the following lemma, which will be used several times in this chapter and the next. Lemma 3.0.4. Let K be a knot, and let L a crossing circle for K. Suppose that MK∪L is reducible. Then L is nugatory. Proof. An essential 2-sphere in MK∪L must separate η(K) and η(L). Thus if MK∪L is reducible, then L lies in a 3-ball in S 3 disjoint from K. Since L is unknotted, L bounds a disc in the complement of K. 3.1 Crossing disks and Seifert surface Let K be an oriented knot and L = ∂D be a crossing circle for K. Since the linking number of L and K is zero, K bounds a Seifert surface in the complement of L. Let F be a Seifert surface that is of minimal genus among all such Seifert surfaces in the complement of L. Since F is incompressible, after an isotopy we may assume that the closed components of F ∩ D are homotopically essential in D − η(K). In particular, suppose there is a component of F ∩ D which bounds a disk in D − η(K) and let C be an innermost such component, so C bounds a disk ∆1 ⊂ D − (η(K) ∪ F ). Since F is incompressible, C also bounds a disk ∆2 ⊂ F and ∆1 ∪ ∆2 ∼ S 2 bounds a 3-ball B ⊂ S 3 − F . By isotoping ∆2 through B = we can remove C from F ∩ D. Hence we may assume each simple closed curve of F ∩ D is 21 K D α F Figure 3.1: The crossing arc α = F ∩ D. parallel in D − η(K) to ∂D and, by further isotopy, we can arrange so that F ∩ D contains no closed components. Any component of F ∩ D which has boundary must have its endpoints at K ∩ D, so F ∩ D must now consist of a single arc α that is properly embedded in F , as illustrated in Figure 3.1. The surface F gives rise to a Seifert surface F (q) for KL (q) by twisting F q times at α. Proposition 3.1.1. Suppose that L is an order-q cosmetic crossing circle for a knot K. Let F be a minimal-genus Seifert surface for K in the complement of L. Then F and F (q) are Seifert surfaces of minimal genus for K and K(q), respectively, in S 3 . Proof. If L is nugatory, then L bounds a disc in the complement of F and the conclusion is clear. Suppose L is cosmetic. By Lemma 3.0.4, MK∪L is irreducible. We can consider the surface F properly embedded in MK∪L so that F is disjoint from ∂η(L) ⊂ ∂M . The assumptions on irreducibility of MK∪L and on the genus of F imply that the foliation machinery of Gabai [14] applies. In particular, F is Thurston norm minimizing in MK∪L . The manifolds MK and MK(q) are obtained by Dehn fillings of MK∪L along ∂η(L). By Theorem 2.1.2, F can fail to remain Thurston norm minimizing (i.e. genus minimizing) in 22 at most one of MK and MK(q) . Since we have assumed that L is cosmetic, MK and MK(q) are homeomorphic (by an orientation-preserving homeomorphism). Thus F remains taut in both of MK and MK(q) . This implies that F and F (q) are Seifert surfaces of minimal genus for K and K(q), respectively. By Proposition 3.1.1, a generalized crossing change in a knot K that produces an isotopic knot corresponds to a properly embedded arc α on a minimal-genus Seifert surface F of K. This leads to the following. Lemma 3.1.2. Let F be a minimal-genus Seifert surface for a knot K, and let α ⊂ F be an embedded arc corresponding to a crossing circle L of K. If α is inessential on F , then L is nugatory. Proof. Recall that α is the intersection of a crossing disc D with F , where ∂D = L. Since α is inessential, it separates F into two pieces, one of which is a disc E. Consider D as properly embedded in a regular neighborhood η(F ) of the surface F . The boundary of a regular neighborhood of E in η(F ) is a 2-sphere S that contains the crossing disc D. Then S − int(D) is a disk bounded by the crossing circle L with its interior disjoint from the knot K = ∂F . 3.2 Obstructing cosmetic crossings in genus-one knots The following results use Seifert matrices and the Alexander polynomial to find obstructions to genus-one knots admitting a cosmetic generalized crossing changes. Theorem 3.2.1. Let K be a genus-one knot. If K admits a cosmetic generalized crossing change, then K is algebraically slice. In particular, there is a linear polynomial f (t) ∈ Z[t] 23 α a1 a2 Figure 3.2: A genus-one surface F with generators a1 and a2 of H1 (F ) and a non-separating arc α. . such that the Alexander polynomial of K is of the form ∆K (t) = f (t)f (t−1 ). Proof. Let K(q) be a knot that is obtained from K by a cosmetic generalized crossing change at a crossing disk D. By Proposition 3.1.1, there is a genus-one Seifert surface F such that D intersects F in a properly embedded arc α ⊂ F . Let F (q) denote the result of F after the order-q generalized crossing change at α. Since L = ∂D is cosmetic, by Lemma 3.1.2, α is essential. Further, since the genus of F is one, α is non-separating. We can find a simple closed curve a1 on F that intersects α exactly once. Let a2 be another simple closed curve so that a1 and a2 intersect exactly once, a2 ∩ α = ∅, and the homology classes of a1 and a2 form a basis for H1 (F ) ∼ Z ⊕ Z. Note = that {a1 , a2 } forms a corresponding basis of H1 (F (q)). (See Figure 3.2.) The Seifert matrices of F and F (q) with respect to these bases are   a b  V :=   c d   a + q b  and Vq :=  , c d 24 respectively, where a, b, c, d ∈ Z. The Alexander polynomials of K and K(q) are given by . ∆K (t) = ad(1 − t)2 − (b − ct)(c − tb) (3.1) . ∆K(q) (t) = (a + q)d(1 − t)2 − (b − ct)(c − tb). (3.2) and . Since K ∼ K(q), we must have ∆K (t) = ∆K(q) (t), which easily implies that d = lk(a2 , a+ ) = = 2 0. Hence K is algebraically slice and . ∆K (t) = (b − ct)(c − tb) = (−t)(b − ct)(b − ct−1 ) . = (b − ct)(b − ct−1 ). (3.3) . Setting f (t) := b − ct we obtain ∆K (t) = f (t)f (t−1 ), as desired. Note that since |b − c| is the intersection number between a1 and a2 , by suitable orientation choices, we may assume that c = b + 1. As a corollary of Theorem 3.2.1, we have the following. Corollary 3.2.2. Let K be a genus-one knot. If det(K) is not the square of an integer, then K admits no cosmetic generalized crossing change of any order. Proof. Suppose that K admits a cosmetic generalized crossing change. By Theorem 3.2.1, . ∆K (t) = f (t)f (t−1 ) where f (t) ∈ Z[t] is a linear polynomial. Thus, if K admits a cosmetic generalized crossing change, we have det(K) = |∆K (−1)| = [f (−1)]2 . Recall that any two Seifert matrices for a knot K are S-equivalent as defined in Definition 2.2.2. Hence the following corollary is an immediate consequence of the proof of Theorem 25 3.2.1. Corollary 3.2.3. Let K be a genus-one knot. If K admits an order-q cosmetic generalized   b  a crossing change, then K has a Seifert matrix of the form   which is S-equivalent b+1 0   a + q b  to  . b+1 0 3.3 Crossing changes and double branched covers In this section we primarily restrict our attention to crossing changes, i.e. generalized crossing changes of order ±1. We derive obstructions to cosmetic crossing changes in terms of the homology of the double branched cover of S 3 branched over the knot K. More specifically, we will prove the following. Theorem 3.3.1. Let K be a genus-one knot and let YK denote the double branched cover of S 3 branched over K. If K admits a cosmetic crossing change, then the homology group H1 (YK ; Z) is a finite cyclic group. To prove Theorem 3.3.1 we need the following lemma. Here, given m ∈ Z, we denote by Zm := Z/mZ the cyclic abelian group of order |m|. Lemma 3.3.2. If H denotes the abelian group given by the presentation H∼ = c1 , c2 2xc1 + (2y + 1)c2 = 0 (2y + 1)c1 = 0 then we have the following. 1. H ∼ 0, if y = 0 or y = −1. = 26 , 2. H ∼ Zd ⊕ Z (2y+1)2 , if y = 0, −1 and d := gcd(2x, 2y + 1) with 1 ≤ d ≤ |2y + 1|. = d Proof. If y = 0 or y = −1, clearly we have H ∼ {0}. Suppose that y = 0, −1 and set = d := gcd(2x, 2y + 1) with 1 ≤ d ≤ |2y + 1|. Then there are integers A and B such that 2x = dA, 2y + 1 = dB, and gcd(A, B) = 1. Let α and β be such that αA + βB = 1. Since      dA dB   α β   is a presentation matrix of H and   is invertible over Z, we get that dB 0 −B A       α β   dA dB  d dαB    =  is also a presentation matrix for H. So 2 −B A dB 0 0 −dB c1 , c2 dc1 + dαBc2 = 0 H∼ = dB 2 c 2 (3.4) =0 Now letting c3 = c1 + αBc2 , we have c2 , c3 dc3 = 0 H∼ = dB 2 c 2 (3.5) =0 Hence H ∼ Zd ⊕ ZdB 2 = Zd ⊕ Z (2y+1)2 . = d With this lemma, we are now ready to prove Theorem 3.3.1. Proof of Theorem 3.3.1. Suppose that a genus-one knot K admits a cosmetic crossing change yielding an isotopic knot K . The proof of Theorem 3.2.1 shows that K and K admit Seifert matrices of the form     a  V :=  b   a+ε b   and V :=   b+1 0 b+1 0 27 (3.6) respectively, where a, b ∈ Z and ε = 1 or −1 according to whether K is obtained via (−1)or (+1)-Dehn surgery, respectively. In particular we have . . ∆K (t) = ∆K (t) = b(b + 1)(t2 + 1) − (b2 + (b + 1)2 t). (3.7) Presentation matrices for H1 (YK ) and H1 (YK ) are given by    2b + 1  2a 2a + 2 T V +VT =  and V + (V ) =  2b + 1 0 2b + 1  2b + 1 , 0 (3.8) respectively, as shown in [36]. It follows that Lemma 3.3.2 applies to both H1 (YK ) and H1 (YK ). By that lemma, H1 (YK ) is either cyclic or H1 (YK ) ∼ Zd ⊕ Z (2b+1)2 , with b = = d 0, −1 and gcd(2a, 2b + 1) = d where 1 < d ≤ |2b + 1|. Similarly, H1 (YK ) is either cyclic or H1 (YK ) ∼ Zd ⊕ Z (2b+1)2 , with gcd(2a + 2ε, 2b + 1) = d where 1 < d ≤ |2b + 1|. Since = d K and K are isotopic, we have H1 (YK ) ∼ H1 (YK ). One can easily verify this can only = happen in the case that gcd(2a, 2b + 1) = gcd(2a + 2ε, 2b + 1) = 1 in which case H1 (YK ) is cyclic. It is known that for an algebraically slice knot of genus one every minimal-genus Seifert surface F contains a metabolizer. After completing the metabolizer to a basis of H1 (F ) we have a Seifert matrix V as in equation (3.6) above. Corollary 3.3.3. Let K be an algebraically slice knot of genus one. Suppose that a genus  b  a one Seifert surface of K contains a metabolizer leading to a Seifert matrix V =   b+1 0 so that b = 0, −1 and gcd(2a, 2b+1) = 1. Then K cannot admit a cosmetic crossing change. Proof. Let d = gcd(2a, 2b + 1). As in the proof of Theorem 3.3.1, we use Lemma 3.3.2 28 to conclude that H1 (YK ) ∼ Zd ⊕ Z (2b+1)2 and hence is not cyclic unless d = 1. Now the = d conclusion follows by Theorem 3.3.1. We also have the following corollary with regard to cosmetic generalized crossing changes. Corollary 3.3.4. Let K be an algebraically slice knot of genus one. Suppose that a genus  b  a one Seifert surface of K contains a metabolizer leading to a Seifert matrix V =   b+1 0 so that b = 0, −1. Let d := gcd(2a, 2b + 1). Then K cannot admit an order-q cosmetic generalized crossing change if d does not divide q. Proof. First suppose V is the matrix coming from the basis defined in the proof of Theorem 3.2.1. Again we use Lemma 3.3.2 to see that H1 (YK ) ∼ Zd ⊕ Z (2b+1)2 and H1 (YK(q) ) ∼ = = d Zd ⊕ Z (2b+1)2 where d := gcd(2a + 2q, 2b + 1). Hence H1 (YK ) ∼ H1 (YK(q) ) if and only = d if d = d . Since 2b + 1 is odd, d divides a and can only be a divisor of 2(a + q) if it also   b  a divides q. Hence for any Seifert matrix V for K of the form V =  , if K admits b+1 0 an order-q cosmetic generalized crossing change, then d := gcd(2a, 2b + 1) divides q. 3.4 S-equivalence of Seifert matrices In light of the last section, it is natural to ask the following question.     b  a + q b   a Problem 3.4.1. Given a, b, q ∈ Z, when is  ∼ ? b+1 0 b+1 0 A first observation is that if b = 0 or −1, then the two given matrices given in Problem 3.4.1 are congruent and, in particular, S-equivalent. We therefore restrict ourselves to matrices with non-zero determinant. In terms of Seifert matrices, this is equivalent to considering . knots of genus one whose Alexander polynomial ∆K (t) = det(V − tV T ) is non-trivial. 29 Proposition 3.4.2 is an auxiliary algebraic result about congruences of Seifert matrices. As an application of of this proposition, we prove cosmetic crossing changes do not exist in genus-one knots with non-trivial Alexander polynomial and with a minimal-genus Seifert surface which, up to isotopy, is unique.     b b  a  c Proposition 3.4.2. Suppose that the matrices   ≈   are congruent b+1 0 b+1 0 over Z, where a, b, c ∈ Z. Then there is an integer n such that a + n(2b + 1) = c. Proof. To begin, we suppose that an integral congruence exists as hypothesized. That is, suppose that there exist integers x, y, z, t such that       b b  x z   c x y   a .    = b+1 0 z t b+1 0 y t (3.9) The left hand side multiplies out to give    b xza + yz(b + 1) + xtb  c  .  = 2 a + zt(2b + 1) b+1 0 xza + xt(b + 1) + zyb z  x2 a + xy(2b + 1) (3.10) By the bottom right entry of equation (3.10), we must consider each of the following three cases. • z=0 • z = 0 and a = 0 • z = 0 and a = 0 30 First, if z = 0, then equation (3.10) becomes  x2 a + (2b + 1)xy   xt(b + 1)    xtb  c b = . 0 b+1 0 (3.11) We need x = t = 1 or x = t = −1 for the top right and bottom left entries to be correct. Then setting n = xy proves the proposition in this case. Now, suppose z = 0 and a = 0. Then by the bottom right entry of equation (3.10), zt(2b + 1) = 0. Hence equation (3.10) becomes     b (2b + 1)xy yz(b + 1)  c .  = b+1 0 zyb 0 (3.12) The equations zyb = b + 1 and zy(b + 1) = b imply that b2 = (b + 1)2 , which has no integral solutions. Finally, suppose z, a = 0. Then z = −t(2b + 1)/a, which we substitute into equation (3.10), to yield     −t(b + 1)k/a  c b  xk  =  −tbk/a 0 b+1 0 (3.13) where k := ax + y(2b + 1). The equations (−tk/a)(b + 1) = b and (−tk/a)b = b + 1 imply again that (b + 1)2 = b2 since neither t nor k can be 0. Since this does not have integral solutions, we also rule out this case. Hence the only congruences possible are those in the statement of the proposition, which occur when z = 0 and x = t = ±1. If K is a knot with, up to isotopy, a unique minimal-genus Seifert surface, then the Seifert matrix corresponding to that surface only depends on the choice of basis for the first 31 homology. Put differently, the integral congruence class of the Seifert matrix corresponding to the unique minimal-genus Seifert surface is an invariant of the knot K. As a consequence of Proposition 3.4.2, we have the following theorem. Theorem 3.4.3. Let K be a genus-one knot with a unique minimal-genus Seifert surface, . and suppose K admits a cosmetic crossing change. Then ∆K (t) = 1. Proof. Let K be a genus-one knot with a unique minimal-genus Seifert surface, which admits a cosmetic crossing change (i.e. a cosmetic generalized crossing change of order-(±1)). It follows from Corollary 3.2.3 and from the paragraph preceding the statement of this theorem     b a + ε b   a that K admits a Seifert matrix   for some  which is congruent to  b+1 0 b+1 0 ε ∈ {−1, 1}. For b = 0, −1, Proposition 3.4.2 precludes such congruences from being possible. If b = 0 or −1, then the Alexander polynomial is 1. We also have the following corollary regarding cosmetic generalized crossing changes and unique genus-one Siefert surfaces. Corollary 3.4.4. Let K be a genus-one knot with a unique minimal-genus Seifert surface and suppose K admits an order-q cosmetic generalized crossing change. Then K has a Seifert   b  a matrix V :=   such that (2b + 1) divides q. b+1 0 Proof. Given such a K, it again follows from Corollary 3.2.3 that K admits a Seifert matrix     b  a a + q b  V :=   which is S-equivalent to  . If b = 0 or −1, |2b + 1| = 1 and the b+1 0 b+1 0 result follows immediately. For b = 0, −1, Proposition 3.4.2 implies that there exists some n ∈ Z such that a + n(2b + 1) = a + q and hence q = n(2b + 1). 32 p q r Figure 3.3: On the left is a diagram for P (p, q, r) with p, q and r positive. Note that each of p, q and r denote the total number of crossings in the corresponding twist region. On the right is the pretzel knot P (3, 3, −3). 3.5 Pretzel knots Let K be a three string pretzel knot P (p, q, r) with p, q and r odd as defined in Figure 3.3. The knot determinant of K is given by det(K) = |pq + qr + pr| (see [23]), and if K is non-trivial, then it has genus one. It is known that K is algebraically slice if and only if pq + qr + pr = −m2 , for some odd m ∈ Z [22]. Corollary 3.5.1. The pretzel knot P (p, q, r) with p, q and r odd does not admit a cosmetic generalized crossing change of any order if pq + qr + pr = −m2 , for every m ∈ Z. Proof. This follows immedaitely from Theorem 3.2.1 and the discussion above. Corollary 3.5.2. The knot P (p, q, r) with p, q and r odd does not admit a cosmetic crossing change (i.e. a cosmetic generalized crossing change of order ±1) if either of the following is true. 1. q + r = 0 and gcd(p, q) = 1 2. p + q = 0 and gcd(p, r) = 1 33 Proof. There is a genus-one surface for P (p, q, r) for which a Seifert matrix is V(p,q,r) :=   1  p + q q + 1  . (See Example 6.9 of [23].) Suppose that q + r = 0. If gcd(p, q) = 1, 2 q−1 q+r then gcd(p + q, q) = 1 and the conclusion in Case 1 follows by Corollary 3.3.3. The proof for Case 2 is identical. 3.6 Genus-one knots with low crossing number The purpose of this section is to study potential cosmetic generalized crossing changes in genus-one knots with up to 12 crossings. We will need the following theorem of Trotter. Theorem 3.6.1 (Trotter, Corollary 4.7 of [44]). Let V be a Seifert matrix with | det(V )| a prime or 1. Then any matrix which is S-equivalent to V is congruent to V over Z. With this in mind, we can prove the following. Theorem 3.6.2. Let K be a genus-one knot that has a diagram with at most 12 crossings. Then K admits no cosmetic crossing changes. Proof. Table 1, obtained from KnotInfo [9], gives the 23 knots of genus one with at most 12 crossings and the values of their determinants. We observe that there are four knots with square determinants. These are 61 , 946 , 103 and 11n139 , which are all known to be algebraically slice. Thus, Corollary 3.2.2 excludes cosmetic crossings for all but these four knots. Now 61 and 103 are 2-bridge knots, so by [43] they do not admit cosmetic crossing changes. The knot K = 946 is isotopic to the pretzel knot P (3, 3, −3) of Figure 3.3. By Corollary 3.5.2, this knot admits no cosmetic crossing changes. The only remaining knot from Table 1 is the knot K = 11n139 . This knot is isotopic to the pretzel knot P (−5, 3, −3). There is therefore a genus-one Seifert surface for K for which 34 K 31 41 52 61 72 74 81 83 det(K) 3 5 7 9 11 15 13 17 K 92 95 935 946 101 103 11a247 11a343 det(K) 15 23 27 9 17 25 19 31 K 11a362 11a363 11n139 11n141 12a803 12a1287 12a1166 - det(K) 39 35 9 21 21 37 33 - Table 3.1: Genus-one knots with at most 12 crossings.   −1 2 a Seifert matrix is V :=   as described in Section 3.5. Since | det(V )| = 2 is prime, 1 0 by Corollary 3.2.3 and Theorem 3.6.1 it suffices to show that V is not integrally congruent to two matrices of the forms     b  a a + 1 b  V1 :=   and V2 :=  . b+1 0 b+1 0 If this were true, V1 and V2 would be integrally congruent to each other, which cannot . happen by Proposition 3.4.2 unless b = 0 or −1. In this case, ∆K (t) = 1, but the Alexander . polynomial of 11n139 is ∆11n139 (t) = 2 − 5t + 2t2 . We also have the following corollary regarding cosmetic generalized crossing changes. Corollary 3.6.3. Let K be a genus-one knot that has a diagram with at most 12 crossings. If K admits an order-q cosmetic generalized crossing change, then one of the following must be true. 1. K = 946 and q = 3n for some n ∈ Z 2. K = 11n139 35 Proof. By the proof of Theorem 3.6.2, the only such knots which could admit a cosmetic generalized crossing change of any order are 946 and 11n139 . If K = 946 = P (3, 3, −3), then     3 2  6 3 K has a Seifert matrix  . Hence H1 (YK ) has the presentation matrix  . Thus 1 0 3 0 by Lemma 3.3.2, H1 (Y ) ∼ Z3 ⊕ Z3 , and by Theorem 3.3.1, 3 divides q. = K 36 Chapter 4 Crossing changes and embedded tori 4.1 Essential tori In this chapter we will continue to investigate potential cosmetic generalized crossing changes by studying the interaction of the crossing circle L with embedded tori in the complement of K. We will continue using the notation of Chapter 3. The results of this chapter can also be found in [3, 5]. Fix a knot K and let L be a crossing circle for K. Let M (q) denote the 3-manifold obtained from MK∪L via a Dehn filling of slope (−1/q) along ∂η(L). (Recall that MK∪L := S 3 − η(K ∪ L).) So, for q ∈ Z − {0}, M (q) = MK(q) , and M (0) = MK . We will sometimes use K(0) to denote K ⊂ S 3 when we want to be clear that we are considering K ⊂ S 3 rather than K ⊂ ML . Suppose there is some q ∈ Z for which KL (q) is a satellite knot. Then there is a companion torus T for KL (q) and, by definition, T is essential in M (q). This essential torus T must occur in one of the following two ways. Definition 4.1.1. Let T ⊂ M (q) be an essential torus. We say T is Type 1 if T can be isotoped into MK∪L ⊂ M (q). Otherwise, we say T is Type 2. If T is Type 2, then we may isotope T so that it is the image of a punctured torus (P, ∂P ) ⊂ (MK∪L , ∂η(L)) and each component of ∂P has slope (−1/q) on ∂η(L). 37 In general, let L be any knot or link in S 3 and let Σ be a boundary component of ML . If (P, ∂P ) ⊂ (ML , Σ) is a punctured torus and each component of ∂P is homotopically essential on Σ, then every component of ∂P has the same slope on Σ, which we call the boundary slope of P . Suppose C1 and C2 are two non-separating simple closed curves (or boundary slopes) on a torus Σ. Let si be the slope of Ci on Σ, and let [Ci ] denote the isotopy class of Ci for i = 1, 2. Then ∆(s1 , s2 ) is the minimal geometric intersection number of [C1 ] and [C2 ]. If si is the rational slope (1/qi ) for some qi ∈ Z for i = 1, 2, then ∆(s1 , s2 ) = |q1 − q2 |. (See [15] for more details.) Note that we consider ∞ = (1/0) to be a rational slope. Gordon [15] proved the following theorem relating the boundary slopes of punctured tori in link complements. In fact, Gordon proved a more general result, but we state the theorem here only for the case which we will need later in Section 4.3. Theorem 4.1.2 (Gordon, Theorem 1.1 of [15]). Let L be a knot or link in S 3 and let Σ be a boundary component of ML . Suppose (P1 , ∂P1 ) and (P2 , ∂P2 ) are punctured tori in (ML , Σ) such that the boundary slope of Pi on Σ is si for i = 1, 2. Then ∆(s1 , s2 ) ≤ 5. 4.2 Preliminary results The goal of Section 4.3 is to prove Theorem 4.3.1, which says that any satellite knot K with an order-q cosmetic generalized crossing change with |q| ≥ 6 admits a pattern knot K with a cosmetic generalized crossing change of the same order. To do this we will need a few preliminary results, which we present in this section. Suppose K is a knot contained in a solid torus V ⊂ S 3 . We call K geometrically essential (or simply essential ) in V if every meridional disk of V meets K at least once. With this in 38 mind, we have the following lemma of Kalfagianni and Lin. Lemma 4.2.1 (Kalfagianni and Lin, Lemma 4.6 of [19]). Let V ⊂ S 3 be a knotted solid torus such that K ⊂ int(V) is a knot which is geometrically essential in V , so wrap(V, K) = 0. Suppose K has a crossing disk D with D ⊂ int(V). If K is isotopic to K(q) in S 3 , then K(q) is also geometrically essential in V . Further, if K is not the core of V , then K(q) is also not the core of V . Proof. Suppose, by way of contradiction, that K(q) is not essential in V . Then there is a 3-ball B ⊂ V such that K(q) ⊂ B. This means that the winding number w(K(q), V ) = 0. Let S1 be a Seifert surface for K which is of minimal genus in ML , where L = ∂D. We may isotope S1 so that S1 ∩ D consists of a single curve α as in Section 3.1. Then twisting S1 q times at α gives rise to a Seifert surface S2 for K(q). By Proposition 3.1.1, S1 and S2 are minimal-genus Seifert surfaces in S 3 for K and K(q), respectively. Since w(K(q), V ) = 0, S1 ∩ ∂V = S2 ∩ ∂V is homologically trivial in ∂V . For i = 1, 2, we can surger Si along disks and annuli in ∂V which are bounded by curves in Si ∩ ∂V to get new minimal-genus Seifert surfaces Si ⊂ int(V). Then S2 is incompressible and V is irreducible, so we can isotope S2 into int(B). Hence α and therefore D can also be isotoped into int(B). But then K must not be essential in V , which is a contradiction. Finally, if K is not the core of V , then ∂V is a companion torus for the satellite knot K since V is knotted by assumption. Since a satellite knot cannot be isotopic to the core of its companion torus, K(q) cannot be the core of V . The following lemma and its proof are of the same flavor as Lemma 4.2.1 and will be used in Section 4.4. 39 Lemma 4.2.2. Let K be a satellite knot, T be a companion torus for K, and V be the solid torus bounded by T in S 3 . Suppose that w(K, V ) = 0 and that there are no essential annuli in S 3 − V . Finally, suppose that K admits a cosmetic generalized crossing change of order-q, and let L be the corresponding crossing circle. Then we can isotope L so that L and a crossing disk bounded by L both lie in V . Proof. Let K and L be as in the statement of the lemma. Let S be a minimal-genus Seifert surface for K in ML , and let D be the crossing disk for K which is bounded by L. We may isotope S so that S ∩ D is a single embedded arc α. Then performing (−1/q)-surgery at L twists both K and S at α, producing a surface S(q) ⊂ M (q) which is a Seifert surface for K(q). Again, by Proposition 3.1.1, S and S(q) are minimal-genus Seifert surfaces in S 3 for K and K(q), respectively. We may isotope D to be a 2-dimensional neighborhood of α which is orthogonal to S. So if S ⊂ int(V ), then D ⊂ V and there is nothing more to show. Assume that S ⊂ V , and let C := S ∩ T . We may isotope S so that C is a collection of simple closed curves which are essential in both S and T . Since w(K, T ) = 0, C must be homologically trivial in T , where each component of C is given the orientation induced by S. Hence C bounds a collection of annuli in T which we will denote by A0 . Let S0 := S − (S ∩ V ). Suppose that χ(S0 ) < 0, where χ(·) denotes the Euler characteristic. We may create S ∗ from S by replacing S0 by A0 , isotoped slightly if necessary so that each component of A0 is disjoint. Then S ∗ is a Seifert surface for K, and χ(S ∗ ) > χ(S) since χ(A0 ) = 0. This contradicts the fact that S is a minimal-genus Seifert surface for K, so it must be that χ(S0 ) ≥ 0. Since S0 contains no closed component, and no component of C bounds a disk in S, we conclude that S0 consists of annuli. By assumption, there are no essential annuli in S 3 − V , so each component of S0 must 40 D D Figure 4.1: A knotted 3-ball B inside of a solid torus with disks D, D ⊂ ∂B. be boundary parallel in S 3 − V . Thus we can isotope S0 so that S ⊂ V , and therefore D can be isotoped into V as well. Before moving on to Theorem 4.3.1 and its corollaries, we state the following results of Motegi [30] (see also [39]) and McCullough [28] which we will need in Section 4.3. Lemma 4.2.3 (Motegi, Lemma 2.3 of [30]). Let K be a knot embedded in S 3 and let V1 and V2 be knotted solid tori in S 3 such that the embedding of K is essential in Vi for i = 1, 2. Then there is an ambient isotopy φ : S 3 → S 3 leaving K fixed such that one of the following holds. 1. ∂V1 ∩ φ(∂V2 ) = ∅. 2. There exist meridian disks D and D for both V1 and V2 such that some component of V1 cut along (D D ) is a knotted 3-ball in some component of V2 cut along (D D ). By a knotted 3-ball, we mean a ball B for which there is no isotopy which takes B to the standardly embedded 3-ball while leaving D and D fixed. (See Figure 4.1.) Theorem 4.2.4 (McCullough, Theorem 1 of [28]). Suppose M is a compact, orientable 3manifold that admits a homeomorphism which restricts to Dehn twists on the boundary of M along a simple closed curve in C ⊂ ∂M . Then C bounds a disk in M . 41 4.3 Obstrucing cosmetic crossings in satellite knots The goal of this section is to prove the following theorem and its corollaries. Theorem 4.3.1. Suppose K is a satellite knot which admits a cosmetic generalized crossing change of order q with |q| ≥ 6. Then K admits a pattern knot K which also has an order-q cosmetic generalized crossing change. We begin with the following lemma. Lemma 4.3.2. Let K be a prime satellite knot with a cosmetic crossing circle L of order q. Then at least one of the following must be true. 1. M (q) contains no Type 2 tori 2. |q| ≤ 5 Proof. Suppose M (q) contains a Type 2 torus. We claim that M (0) must also contain a Type 2 torus. Assuming this is true, M (0) and M (q) each contain a Type 2 torus and hence there are punctured tori (P0 , ∂P0 ) and (Pq , ∂Pq ) in (M, ∂η(L)) such that P0 has boundary slope ∞ = (1/0) and Pq has boundary slope (−1/q) on ∂η(L). Then, by Theorem 4.1.2, ∆(∞, −1/q) = |q| ≤ 5, as desired. Thus it remains to show that there is a Type 2 torus in M (0). Let M := MK∪L . Since L is not nugatory, Lemma 3.0.4 implies that M is irreducible and hence the Haken number τ (M ) is well-defined. First assume that τ (M ) = 0. Since K is a satellite knot, M (0) must contain an essential torus, and it cannot be Type 1. Hence M (0) contains a Type 2 torus. Now suppose that τ (M ) > 0 and let T be an essential torus in M . Then T bounds a solid torus V ⊂ S 3 . Let ext(V) denote S 3 − V . If K ⊂ ext(V), then L must be essential in V . If 42 L K Figure 4.2: An example of an unknotted torus containing a crossing circle L which bounds a crossing disk for the knot K ∼ K#U . = V is knotted, then either L is the core of V or L is a satellite knot with companion torus ∂V . This contradicts the fact that L is unknotted. Hence T is an unknotted torus. By definition, L bounds a crossing disk D. Since D meets K twice, D ∩ ext(V ) = ∅. We may assume that D has been isotoped (rel boundary) to minimize the number of components in D ∩ T . Since an innermost component of D − (T ∩ D) is a disk and L is essential in the unknotted solid torus V , D ∩ T consists of standard longitudes on the unknotted torus T . Hence D ∩ ext(V ) consists of either one disk which meets K twice, or two disks which each meet K once. In the first case, L is isotopic to the core of V , which contradicts T being essential in M . In the latter case, the linking number lk(K, V ) = ±1. So K can be considered as the trivial connect sum K#U , where U is the unknot, and the crossing change at L takes place in the unknotted summand U . (See Figure 4.2.) The unknot does not admit cosmetic crossing changes of any order by [37], so KL (q) ∼ K#K where K ∼ U . This contradicts the fact = = that KL (q) ∼ K. Hence, we may assume that T is knotted and K is contained in the solid = torus V bounded by T . If L ⊂ ext(V) and cannot be isotoped into V , then D ∩ T has a component C that is both 43 homotopically non-trivial in D − η(K) and not parallel to ∂D. So C must encircle exactly one of the two points of K ∩ D. This means that wrap(K, V ) = ±1. Since T cannot be boundary parallel in M , K is not the core of T , and hence T is a follow-swallow torus for K and K is composite. But this contradicts the assumption that K is prime. Hence we may assume that L and, in fact, D are contained in int(V). Since V is knotted and D ⊂ int(V), Lemma 4.2.1 implies that if T is a companion torus for K(0), then T is also a companion torus for K(q). This means every Type 1 torus in M (0) is also a Type 1 torus in M (q). Since K(0) and K(q) are isotopic, τ (M (0)) = τ (M (q)). By assumption, M (q) contains a Type 2 torus, which must give rise to a Type 2 torus in M (0), as desired. The following is an immediate corollary of Lemma 4.3.2. Corollary 4.3.3. Let K be a prime satellite knot with a cosmetic crossing circle L of order q with |q| ≥ 6. Then τ (MK∪L ) > 0. We are now ready to prove Theorem 4.3.1. Proof of Theorem 4.3.1. Let K be a satellite knot as in the statement of the theorem. Let L be a crossing circle bounding a crossing disk D which corresponds to a cosmetic generalized crossing change of order q. Let M := MK∪L . If K is a composite knot, then Torisu [43] showed that the crossing change in question must occur within one of the summands of K = K1 #K2 , say K1 . We may assign to K the follow-swallow companion torus T , where the core of T is isotopic to K2 . Then the patten knot corresponding to T is K1 and the theorem holds. Now assume K is prime. By Corollary 4.3.3, τ (M ) > 0. Let T be an essential torus in M and let V ⊂ S 3 be the solid torus bounded by T in S 3 . As shown in the proof of 44 Lemma 4.3.2, V is knotted in M and D can be isotoped to lie in int(V). This means T is a companion torus for the satellite link K ∪ L. Let K ∪ L be a pattern link for K ∪ L corresponding to T . So there is an unknotted solid torus V ⊂ S 3 such that (K ∪ L ) ⊂ V and there is a homeomorphism f : (V , K , L ) → (V, K, L). Let T be a Haken system for M such that T ∈ T . We will call a torus J ∈ T innermost with respect to K if M cut along T has a component C such that ∂C contains ∂η(K) and a copy of J. In other words, J ∈ T is innermost with respect to K if there are no other tori in T separating J from η(K). Choose T to be innermost with respect to K. Let W := V − η(K ∪ L). We first wish to show that W is atoroidal. By way of contradiction, suppose that there is an essential torus F ⊂ W . Then F bounds a solid torus in V which we will denote by F . Since T is innermost with respect to K, either F is parallel to T in M , or K ⊂ V − F . By assumption, F is essential in W and hence not parallel to T ⊂ ∂W . So K ⊂ V − F and, since F is incompressible, L ⊂ F . By the arguments of the proof of Lemma 4.3.2, F is unknotted and we may assume that D has been isotoped to minimize |D ∩ F |. Then D ∩ (S 3 − F ) consists of either one disk which meets K twice or two disks each meeting K once. The first case contradicts the fact that F is essential in M . In the second case, we may consider K as K#U as in Figure 4.2, where U is the unknot, and arrive at a contradiction as in the proof of Lemma 4.3.2 since U admits no cosmetic generalized crossing changes. Hence W is indeed atoroidal, and W := V − η(K ∪ L ) must be atoroidal as well. To finish the proof, we must consider two cases, depending on whether T is compressible in V − η(K(q)). Case 1: K(q) is essential in V . We wish to show that there is an isotopy Φ : S 3 → S 3 such that Φ(K(q)) = K(0) and 45 Φ(V ) = V . First, suppose K(q) is the core of T . By Lemma 4.2.1, K is also the core of T . Since L is cosmetic, there is an ambient isotopy ψ : S 3 → S 3 taking K(q) to K(0). Since K(q) and K(0) are both the core of V , we may choose ψ so that ψ(V ) = V and let Φ := ψ. If K(q) is not the core of T , then T is a companion torus for K(q). Since K(0) = (K(q))L (−q), we may apply Lemma 4.2.1 to K(q) to see that T is also a companion torus for K(0). Again, there is an ambient isotopy ψ : S 3 → S 3 taking K(q) to K(0) such that V and ψ(V ) are both solid tori containing K(0) = ψ(K(q)) ⊂ S 3 . If ψ(V ) = V , we once more let Φ := ψ. If ψ(V ) = V , we may apply Lemma 4.2.3 to V and ψ(V ). If part (2) of Lemma 4.2.3 were satisified, then ψ(V ) ∩ V would give rise to a knotted 3-ball contained in either V or ψ(V ). This contradicts the fact that W , and hence ψ(W ), are atoroidal. Hence part (1) of Lemma 4.2.3 holds, and there is an isotopy φ : S 3 → S 3 fixing K(0) such that (φ ◦ ψ)(T ) ∩ T = ∅. Let Φ := (φ ◦ ψ) : S 3 → S 3 . Recall that by Lemma 4.3.2, M (q) contains no Type 2 tori. Hence T remains innermost with respect to K(q) in S 3 and therefore Φ(T ) is also innermost with respect to K(0). Either T ⊂ S 3 − Φ(V ) or Φ(T ) ⊂ S 3 − V . In either situation, the fact that T and Φ(T ) are innermost implies that T and Φ(T ) are in fact parallel in MK . So, after an isotopy which fixes K(0) ⊂ S 3 , we may assume that Φ(V ) = V . Now let h := (f −1 ◦ Φ ◦ f ) : V → V . Note that h preserves the canonical longitude of ∂V (up to sign). Since h maps K (q) to K (0), K (q) and K (0) are isotopic in S 3 . So either L gives an order-q cosmetic generalized crossing change for the pattern knot K , or L is a nugatory crossing circle for K . Suppose L is nugatory. Then L bounds a crossing disk D and another disk D ⊂ MK . We may assume D ∩ D = L . Let A := D ∪ (D ∩ V ). Since ∂V is incompressible in V − η(K ), by surgering along components of of D ∩ ∂V which bound disks or cobound annuli in ∂V , we may assume A is a properly embedded annulus in V and each component 46 η(K (0)) ∩ V1 K (0) V1 L L A V2 V2 X V Figure 4.3: On the left is the solid torus V , cut into two solid tori by the annulus A. On the right is a diagram depicting the construction of X from the proof of Theorem 4.3.1. of A ∩ ∂V is a longitude of V . Since L ⊂ V , we can extend the homeomorphism h on V ⊂ S 3 to a homeomorphism H on all of S 3 . Since V is unknotted, let C be the core of the solid torus S 3 − int(V ). We may assume that H fixes C. Since D ∪ D gives the same (trivial) connect sum decomposition of K (0) ∼ K (q) and H preserves canonical longitudes = on ∂V , we may assume H(D ) is isotopic to D and H(D ) is isotopic to D . In fact, this isotopy may be chosen so that H(C) and H(V ) remain disjoint throughout the isotopy and H(V ) = V still holds after the isotopy. Thus, we may assume h(A) = A and A cuts V into two solid tori V1 and V2 , as shown in Figure 4.3. We now consider two subcases, depending on how h acts on V1 and V2 . Subcase 1.1: h : Vi → Vi for i = 1, 2. Up to ambient isotopy, we may assume the following. 1. K (q) ∩ V1 = K (0) ∩ V1 2. K (q) ∩ V2 is obtained from K (0) ∩ V2 via q full twists at L Let X be the 3-manifold obtained from V2 − η(V2 ∩ K (0)) by attaching to A ⊂ ∂V2 a thickened neighborhood of ∂η(K (0)) ∩ V1 . (See Figure 4.3.) Then h|X fixes X away 47 from V2 and acts on X ∩ V2 by twisting ∂η(K (0)) ⊂ ∂X q times at L . Hence there is a homeomorphism from X to h(X) given by q Dehn twists at L ⊂ ∂X. So, by Theorem 4.2.4, L bounds a disk in X ⊂ (V − η(K )). But this means L bounds a disk in (V − η(K)) ⊂ MK and hence L is nugatory, contradicting our initial assumptions. Subcase 1.2: h maps V1 → V2 and V2 → V1 . Again, we may assume the following. 1. K (q) ∩ V1 = K (0) ∩ V2 2. K (q) ∩ V2 is obtained from K (0) ∩ V1 via q full twists at L This time we construct X from V1 −η(V1 ∩K (0)) by attaching a thickened neighborhood of ∂η(K (0)) ∩ V2 to A ⊂ ∂V1 . Then the arguments of Subcase 1.1 once again show that L must have been nugatory, giving a contradiction. Hence, in Case 1, we have a pattern knot K for K admitting an order-q cosmetic generalized crossing change, as desired. Case 2: T is compressible in V − η(K(q)). In this case, K(q) is contained in a 3-ball B ⊂ V . Since K(q) is not essential in V , by Lemma 4.2.1, K(0) is also not essential in V , and K(0) = f (K (0)) can be isotoped to K(q) = f (K (q)) via an isotopy contained in the 3-ball B ⊂ V . This means that, once again, K (0) is isotopic to K (q) in S 3 , and either L gives an order-q cosmetic generalized crossing change for the pattern knot K or L is a nugatory crossing circle for K . Applying the arguments of each of the subcases in Case 1, we see that L cannot be nugatory, and hence K is a pattern knot for K admitting an order-q cosmetic generalized crossing change. Theorem 4.3.1 gives obstructions to when cosmetic generalized crossing changes can occur 48 in satellite knots. This leads us to several useful corollaries, including Corollaries 4.3.4 and 4.3.7, which concern hyperbolic knots, and Corollary 4.3.5, which addresses torus knots. Corollary 4.3.4. Let K be a satellite knot admitting a cosmetic generalized crossing change of order q with |q| ≥ 6. Then K admits a pattern knot K which is hyperbolic. Proof. Applying Theorem 4.3.1, repeatedly if necessary, we know K admits a pattern knot K which is not a satellite knot and which also admits an order-q cosmetic generalized crossing change. Kalfagianni has shown that fibered knots do not admit cosmetic generalized crossing changes of any order [18], and it is well-known that all torus knots are fibered. Hence, by Thurston’s classification of knots [42], K must be hyperbolic. Corollary 4.3.5. Suppose K is a torus knot. Then no prime satellite knot with pattern K admits an order-q cosmetic generalized crossing change with |q| ≥ 6. Proof. Let K be a prime satellite knot which admits a cosmetic generalized crossing change of order q with |q| ≥ 6. By way of contradiction, suppose T is a companion torus for K corresponding to a pattern torus knot K . Since K is prime, Lemma 4.3.2 implies that T is Type 1 and hence corresponds to a torus in M := MK∪L , which we will also denote by T . If T is not essential in M , then T must to be parallel in M to ∂η(L). But then T would be compressible in MK , which cannot happen since T is a companion torus for K ⊂ S 3 and is thus essential in MK . So T is essential in M and there is a Haken system T for M with T ∈ T . Since the pattern knot K is a torus knot and hence not a satellite knot, T must be innermost with respect to K. Then the arguments in the proof of Theorem 4.3.1 show that K admits an order-q cosmetic generalized crossing change. However, torus knots are fibered and hence admit no cosmetic generalized crossing changes of any order, giving us our desired contradiction. 49 Note that if K is a torus knot which lies on the surface of the unknotted solid torus V , then (K , V ) is a pattern for a satellite knot which is, by definition, a cable knot. Since any cable of a fibered knot is fibered, it was already known by [18] that these knots do not admit cosmetic generalized crossing changes. However, Corollary 4.3.5 applies not only to cables of non-fibered knots, but also to pattern torus knots embedded in any unknotted solid torus V and hence gives us a new class of knots which do not admit a cosmetic generalized crossing change of order q with |q| ≥ 6. The proof of Corollary 4.3.5 leads us to the following. Corollary 4.3.6. Let K be a knot such that g(K ) = 1 and K is not a satellite knot. If there is a prime satellite knot K such that K is a pattern knot for K and K admits a cosmetic generalized crossing change of order q with |q| ≥ 6, then K is hyperbolic and algebraically slice. Proof. By Corollary 4.3.5 and its proof, K is hyperbolic and admits a cosmetic generalized crossing change of order q. Then by Lemma 3.2.1, K is algebraically slice. Finally, the following corollary summarizes the progress we have made in this section towards answering Problem 1.1.4. Corollary 4.3.7. If there exists a knot admitting a cosmetic generalized crossing change of order q with |q| ≥ 6, then there must be such a knot which is hyperbolic. Proof. Suppose there is a knot K with a cosmetic generalized crossing change of order q with |q| ≥ 6. Since K cannot be a fibered knot, K is not a torus knot, and either K itself is hyperbolic, or K is a satellite knot. If K is a satellite knot, then by Corollary 4.3.4 and its proof, K admits a pattern knot K which is hyperbolic and has an order-q cosmetic generalized crossing change. 50 4.4 Cosmetic crossings and twisting operations Throughout this section, let K denote the class of knots which are known not to admit cosmetic generalized crossing changes. As noted in Section 1.1, K contains all fibered knots, 2-bridge knots and genus-one, algebraically non-slice knots. Torisu shows in [43] that the connect sum of two or more knots in K is also in K. Let V be a solid torus standardly embedded in S 3 , and let K be a knot that is geometrically essential in V . Recall from Definition 2.5.1 that given n ∈ Z, the knot Kn,V is the image of K under the nth power of a meridional Dehn twist of V and is called a twist knot of K. When there is no danger of confusion, we will simply write Kn instead of Kn,V . Our first lemma regarding twist knots is similar to Lemmas 4.2.1 and 4.2.2. Lemma 4.4.1. Let V be a solid torus standardly embedded in S 3 , let K be a knot that is geometrically essential in the interior of V , and let K := Kn be a twist knot of K for some n ∈ Z. Suppose that K admits a cosmetic generalized crossing change of order q, and let L be the corresponding crossing circle. Then we can isotope L so that L and a crossing disk bounded by L lie in V . Proof. Let S be a minimal genus Seifert surface for K in ML , and let D be the crossing disk for K which is bounded by L. As in Section 3.1, we may isotope S so that S ∩ D is a single embedded arc α. Then performing (−1/q)-surgery at L twists both K and S at α, producing a surface S(q) ⊂ M (q) which is a Seifert surface for K(q). By Proposition 3.1.1, S and S(q) are minimal-genus Seifert surfaces in S 3 for K(0) and K(q), respectively. Let W be the solid torus S 3 − V . Since S is minimal-genus, each component of S ∩ W is incompressible in W and therefore is either a disk or an annulus which is parallel to an annulus in T := ∂V . We can isotope S to remove the annular components of S ∩ W , so we 51 may assume that each component C ∈ T ∩ S bounds a disk DC ⊂ S in the complement of V . For each such disk, the intersection α ∩ DC is a collection of properly embedded arcs in DC . Each of these arcs can be isotoped onto ∂(DC ) ⊂ T by an isotopy on DC rel ∂α and then isotoped slightly further into int(V ). By starting with an outermost arc of α ∩ DC , we can successively isotope each arc of α ∩ DC in such a way and such that α remains an embedded arc on S throughout the isotopies until α ⊂ int(V ). Since we may assume that L lies in small neighborhood of α, this process brings L and D into V , as desired. The following lemma follows from the proof of Theorem 4.3.1 and discusses the interplay of nugatory generalized crossing changes and twisting operations. Lemma 4.4.2. Let K be a prime knot that is essential in a standardly embedded solid torus V , and let K be a twist knot of K . Consider the twisting homeomorphism f : (V , K ) −→ (V, K), where V := f (V ). Let L be a crossing circle for K that lies in V , and let L := f (L ) be the corresponding crossing circle for K in V . Suppose that there is a diffeomorphism h : V → V that takes the canonical longitude of V to itself and such that h(K (0)) = K (q). Then, if L is nugatory for K in S 3 , L is also nugatory for K in S 3 . Proof. Suppose L is nugatory. Then L bounds a crossing disk D and another disk D in the complement of K . As in the proof of Theorem 4.3.1, we may assume D ∩ D = L and D ∪ (D ∩ V ) contains a properly embedded annulus (A, ∂A) ⊂ (V , ∂V ) such that each component of ∂A is a standard longitude of V . By the arguments in the proof of Theorem 4.3.1, we may assume h(A) = A and A cuts V into two solid tori, V1 and V2 , as shown in Figure 4.3. Then the conclusion of the lemma holds by considering Cases 1 and 2 in the proof of Theorem 4.3.1. Before getting to the main theorem of this section, we state a result of Shibuya which 52 will be needed later. See [40] for a proof. Theorem 4.4.3 (Shibuya, Theorem 3.10 of [21]). Let K be a knot in S 3 and let V be a standardly embedded solid torus containing K such that wrap(K, V ) = w(K, V ) ≥ 3. Then Kn,V ∼ Km,V implies m = n. = We are now ready to prove the following. Theorem 4.4.4. Let K ∈ K be contained in a standardly embedded solid torus V with w(K , V ) = wrap(K , V ) ≥ 3. Then, for every n ∈ Z, the twist knot K n,V does not admit a cosmetic generalized crossing change of any order. Proof. Suppose that for some K ∈ K there is an embedding of K into a standardly embedded solid torus V as in the statement of the theorem, and that for some n ∈ Z, the twist knot K := K n,V admits an order-q cosmetic crossing change corresponding to a crossing circle L. That is, K(0) and K(q) are isotopic in S 3 . Let f : V −→ V := f (V ) denote the twisting homeomorphism bringing K to K. By Lemma 4.4.1, we may isotope L into V . Now L pulls back, via f , to a crossing circle L of K in V , and the generalized crossing change on K pulls back to a generalized crossing change on K . Let K (q) := K (q) denote the result of this crossing change on K . L Since K is essential in V , by Lemma 4.2.1, K(q) is also essential in V . As in the proof of Theorem 4.3.1, there is an orientation-preserving diffeomorphism φ : S 3 −→ S 3 that brings K = K(0) to K(q). Since V is an unknotted solid torus in S 3 , φ|V is given by a meridional twist on V of some order m ∈ Z. By Theorem 4.4.3, the hypotheses of the theorem imply that m = 0. Thus φ|V must take canonical longitudes to canonical longitudes, preserving orientation. Hence, we may assume that φ fixes V . 53 Let h := (f −1 ◦ φ ◦ f ) : V → V . Then h maps K to K (q), and hence K and K (q) are isotopic in S 3 . So either L gives an order-q cosmetic generalized crossing change for K , or L is a nugatory crossing circle for K . Since K ∈ K, L has to be nugatory. By Lemma 4.4.2 and the fact that h maps the canonical longitude of V to itself, L is also nugatory for K=K n,V , which contradicts our assumption that L is cosmetic. This gives rise to the following corollary. Corollary 4.4.5. Let K be a fibered m-braid with m ≥ 3. Then for every 3 ≤ p ≤ m and q ∈ Z, there is no twisted fibered braid Kp,q which admits a cosmetic generalized crossing change of any order. Proof. Consider a fibered braid K and let Kp,q be a twisted fibered braid obtained from 2q K by inserting ∆p , where 3 ≤ p ≤ m. Consider a disk D that intersects K only in the strings to be twisted, exactly once for each string. Let B := D × [0, 1] be a neighborhood of D that meets K in exactly p unknotted arcs corresponding to the p points of D ∩ K. Let B be a 3-ball that engulfs the part of K outside B and such that B ∩ B = D × {0, 1}. Then V := B ∪ B is a solid torus containing K, and D is a meridional disk of V . (See, for example, Figure 2.6.) Clearly, Kp,q is the result of K under an order-q Dehn twist of V along ∂D. Thus Kp,q is a twist knot of K, and the conclusion follows from Theorem 4.4.4. Knots that are closures of braids on three strands fit into the setting of Corollary 4.4.5 to give us the following corollary. Recall that the braid group B3 has generators σ1 and σ2 as defined in Figure 2.5, and ∆2 := (σ1 σ2 )3 generates the center of B3 . Corollary 4.4.6. Let K be a knot that can be represented as the closure of a 3-braid. Then K admits no cosmetic crossing changes of any order. 54 Proof. Suppose that knot K = w, the closure of some braid w ∈ B3 . By Schreier [38], w is ˆ conjugate to a braid in exactly one of the following forms. p −q p −q 1. ∆2k σ1 1 σ2 1 · · · σ1 s σ2 s , where k ∈ Z and pi , qi , and s are all positive integers p 2. ∆2k σ1 for some k, p ∈ Z 3. ∆2k σ1 σ2 for some k ∈ Z 4. ∆2k σ1 σ2 σ1 for some k ∈ Z 5. ∆2k σ1 σ2 σ1 σ2 for some k ∈ Z This form is unique up to cyclic permutation of the word following ∆2k . Since we are concerned with knots only and cases (2) and (4) will produce links for all k, p ∈ Z, we need not consider these cases. We will call braids of the form (1) generic. In this case, we first notice that an alternating p −q p −q braid of the form σ1 1 σ2 1 · · · σ1 s σ2 s is homogeneous and hence its closure is fibered by Stallings [41]. Thus, if the closure of a generic braid is a knot, then it is a twisted fibered braid. In case (3), w is also a twisted fibered braid since the closure of σ1 σ2 is the unknot ˆ and hence fibered. Finally, the closure of the braid σ1 σ2 σ1 σ2 represents the trefoil, which is also a fibered knot. Thus the closure of a braid of type (5) is also a twisted fibered braid. Since in all cases the twisting occurs on 3 strings, Corollary 4.4.5 applies to give the desired conclusion. Remark 4.4.7. A weaker form of Corollary 4.4.6 was derived by Wiley in [46] where he shows that a closed 3-braid diagram cannot admit a cosmetic crossing change. 55 4.5 Another obstruction in satellite knots In Section 4.3 we saw that any prime satellite knot with pattern a non-satellite in K does not admit a cosmetic generalized crossing change of order greater than 5. In this section, we restrict ourselves to satellite knots with winding number zero and obtain the following result. Theorem 4.5.1. Let C be a prime knot that is not a cable knot and let V be a standardly embedded solid torus in S 3 . Let K be a non-satellite knot in K such that K is geometrically essential in the interior of V with w(K , V ) = 0. Then any knot that is a satellite of C with pattern (K , V ) admits no cosmetic generalized crossing change of any order. Proof. Let (K , V ) be as in the statement of the theorem, and consider the satellite map f : (V , K ) → (V, K) with C = core(V ) and T := ∂V . Suppose that K admits an order-q cosmetic crossing change, and let D be the corresponding crossing disk with L = ∂D. By Lemma 2 of [25] and the assumption that C is not a cable knot, there are no essential annuli in S 3 − V . Hence, by Lemma 4.2.2, we may assume D ⊂ V , so T is also a companion torus for the satellite link K ∪ L. Now K ∪ L is a pattern link for K ∪ L with the satellite map f : (V , K , L ) → (V, K, L) as above. We will show that L is an order-q cosmetic crossing circle for K , which is a contradiction since K ∈ K. From here, the proof is similar to those of Theorem 4.3.1 and Lemma 4.4.2. Since L is cosmetic, M := MK∪L is irreducible. Consider a Haken system T for M with T ∈ T . Since K is not a satellite knot, T is innermost in T with respect to K. As shown in the proof of Theorem 4.3.1, W := V − η(K ∪ L) and W := V − η(K ∪ L ) are atoroidal. If K(q) is not geometrically essential in V , then, by Lemma 4.2.1, K(0) is also not 56 essential in V . But this contradicts V being a companion for K, so K(q) must be essential in V . Hence T is a companion torus for both K(q) and K(0). Since L is cosmetic, there is an ambient isotopy ψ : S 3 → S 3 taking K(q) to K(0) such that V and ψ(V ) are both solid tori containing K(0) = ψ(K(q)) ⊂ S 3 . By Lemma 4.2.3 and the fact that W is atoroidal, there is another isotopy φ : S 3 → S 3 fixing K(0) such that (φ ◦ ψ)(T ) ∩ T = ∅. Let Φ := (φ ◦ ψ) : S 3 → S 3 . Since T is innermost with respect to K, Φ(T ) ∩ T = ∅ implies T and Φ(T ) are parallel in MK . So, after an isotopy which fixes K(0) ⊂ S 3 , we may assume that Φ(V ) = V . Let h := (f −1 ◦ Φ ◦ f ) : V → V . Then h maps K (q) to K (0), and hence K (q) and K (0) are isotopic in S 3 . So either L gives an order-q cosmetic generalized crossing change for the pattern knot K , or L is a nugatory crossing circle for K . Since K ∈ K, L has to be nugatory. By Lemma 4.4.2, L is nugatory for K, which contradicts our assumption that L is cosmetic. 4.6 Whitehead doubles Given a knot K recall that D+ (K, n) is the n-twisted Whitehead double of K with a positive clasp and D− (K, n) is the n-twisted Whitehead double of K with a negative clasp as shown in Figure 2.4. Since non-trivial Whitehead doubles have genus one and the Whitehead double D± (K, n) is a satellite knot for all K ∼ U , where U is the unknot, we may apply the = results of Chapters 3 and 4 to find obstructions to Whitehead doubles admitting cosmetic generalized crossing changes. Corollary 4.6.1. Given a knot K, the Whitehead double D+ (K, n) does not admit a cosmetic generalized crossing change of any order if either n < 0 or |n| is odd. Similarly 57 D− (K, n) does not admit a cosmetic generalized crossing change if either n > 0 or |n| is odd. Proof. A Seifert surface of D+ (K, n) is obtained by plumbing an n-twisted annulus with   −1 0  core K and a Hopf band, and gives rise to a Seifert matrix Vn :=   as shown in −1 n Example 6.8 of [23]. Thus the Alexander polynomial is of the form . ∆n := ∆D+ (K,n) (t) = −n(t2 + 1) + (1 + 2n)t (4.1) Suppose that D+ (K, n) admits an order-q cosmetic generalized crossing change. Then ∆n should be of the form shown in equation (3.7). Comparing the leading coefficients in equations (3.7) and (4.1), we obtain |n| = |b(b + 1)|, which implies that |n| must be even. Thus we have shown that if |n| is odd then D+ (K, n) admits no cosmetic generalized crossing changes. Suppose now that n < 0. Since the Seifert matrix Vn depends only on n and not on K, the knot D+ (K, n) is S-equivalent to D+ (U, n). This is a positive knot in the sense that all the crossings in the standard diagram of D+ (U, n) are positive. Hence D+ (K, n) has non-zero signature by [34]. Therefore D+ (K, n) is not algebraically slice and, by Theorem 3.2.1, it cannot admit a cosmetic generalized crossing change. A similar argument holds for D− (K, n). The following result applies only to cosmetic generalized crossing changes of order ±1. Corollary 4.6.2. If K is not a cable knot, then D± (K, n) admits no cosmetic crossing changes for every n = 0. 58 Proof. Suppose that K is not a cable knot. Then by [45], for every n = 0 the Whitehead doubles D± (K, n) have unique Seifert surfaces of minimal genus. By equation (4.1), ∆n = 1, and the conclusion follows by Theorem 3.4.3. By adding the assumption that K is prime, we can generalize Corollary 4.6.2 to include cosmetic generalized crossing changes of any order and the Whitehead doubles D± (K, 0). Corollary 4.6.3. Let K be a prime knot that is not a cable knot. Then no Whitehead double of K admits a cosmetic generalized crossing change of any order. Proof. Since all Whitehead doubles admit a pattern (U, V ) where U is the unknot and w(U, V ) = 0, the result follows immediately from Theorem 4.5.1. 59 Chapter 5 On the additivity of the Turaev genus 5.1 Constructing the Turaev surface Besides the definition given in Section 1.2, the Turaev genus of a knot can also be defined in a graph-theoretic way. To do this, consider the all-A smoothing of a projection P . This is the collection of simple closed curves in R2 , which we will refer to as the state circles of the all-A smoothing. For each crossing of P , attach an edge to the all-A smoothing which goes from the state circle at one side of the crossing to the state circle at the other side of the crossing. This gives us a trivalent graph PA . (See Figure 5.1.) From PA , we will construct a ribbon graph GP . A ribbon graph is a graph with a cyclic ordering on the adjacent edges at each vertex. To obtain GP from PA , we collapse each of the state circles to a vertex. If a state circle C is contained within an even number of other state circles in PA , then the cyclic ordering at the vertex of GP corresponding to C is the same as the ordering of the edges meeting C in PA . If C is contained within an odd number of other state circles, then the cyclic ordering at the vertex corresponding to C is the opposite of the ordering of the edges meeting C in PA . We can thicken each of the vertices of GP and think of them as disks. Likewise we can thicken the edges of GP and think of them as untwisted bands or “ribbons”. Then GP becomes an embedded surface with boundary in S 3 . It is known that every boundary component of this surface bounds an embedded disk D2 ⊂ S 3 [7], so we can cap off each 60 =⇒ P =⇒ PA GP Figure 5.1: The construction of the ribbon graph GP from a knot projection P . boundary component to obtain a closed surface Σ ⊂ S 3 . This surface is identical to the Turaev surface TP . Further, GP can be embedded in Σ = TP , and this embedding can be used to recover the alternating projection P ⊂ TP from Section 1.2. (See [13] for details.) Note that GP being planar, TP ∼ S 2 , and P being alternating projection are all equiv= alent statements. It is also obvious from either construction of the Turaev surface that for any two knots K1 and K2 , gT (K1 #K2 ) ≤ gT (K1 ) + gT (K2 ). 5.2 Turaev surfaces and decomposing spheres Let K be a composite knot, let T ⊂ S 3 be a genus-n Turaev surface for K (not necessarily of minimum genus), and let P be the corresponding alternating projection of K on T . Suppose A ⊂ S 3 is a twice-punctured 2-sphere decomposing K into K1 #K2 . Following [29], we position K so that K lies on T , except near each crossing of P, where K lies on the boundary of a bubble. (See Figure 5.2.) Let T+ be T with each disk of T inside a bubble replaced by the upper hemisphere of that bubble. Likewise, we define T− to be T with each disk of T inside a bubble replaced by the lower hemisphere of that bubble. Here we are choosing an orientation on T , so that upper and lower correspond to the direction of an outward or inward normal vector on T , respectively. 61 bubble saddle K T Figure 5.2: A bubble at a crossing of P containing a saddle-shaped disk where the decomposing sphere A meets the interior of the bubble. Definition 5.2.1. A decomposing sphere A is generic if all of the following hold. 1. A meets T+ and T− transversely. 2. The punctures of A do not meet any bubbles. 3. Each time A meets the interior of a bubble, it does so in a saddle-shaped disk as shown in Figure 5.2. Note that any decomposing sphere can be isotoped so that it is generic with respect to any given Turaev surface, as shown in [29]. Notation 5.2.2. Throughout this chapter, we will use the following notation. 1. M+ will denote the genus-n handlebody bounded by T+ lying outside of T and M− will denote the genus-n handlebody bounded by T− lying inside of T . So M± can be thought of as the exterior of T± . 2. C± := A ∩ T± and C± := A ∩ T± for generic decomposing spheres A and A . 62 3. V± := S 3 − M± , so V± is the handlebody inside of T± , whereas M± is the handlebody outside of T± . Note that all bubbles are contained in both V+ and V− . Let C be a component of C± for some generic decomposing sphere A, so C is a simple closed curve on T± . Choose an arbitrary orientation for C, and let N := η(C) ∩ T± . By traversing C along the chosen orientation, we can say an object in N is to the right or to the left of C. While the concepts of right and left are dependent upon the chosen orientation for C, we can discuss two objects being on the same side or opposite sides of C without having to specify an orientation. Proposition 5.2.3 (Alternating property). Let A be a generic decomposing sphere with respect to a Turaev surface T , and let C be a component of C± such that C meets two bubbles B1 and B2 in succession. If the two arcs of K ∩ T± in B1 and B2 lie on the same side of C, then C must cross K an odd number of times between meeting B1 and B2 . Similarly, if the two arcs of K ∩ T± in B1 and B2 lie on the opposite side of C, then C must cross K an even number of times between meeting B1 and B2 . The alternating property was first stated in Section 2 of [29] in the case where T is a sphere. An example of the alternating property is shown in Figure 5.3(A). Note that this proposition is called the alternating property since it is only true because the projection P of K on T is alternating. Proof of the alternating property. Let P be the alternating projection of K on T coming from the construction of T , and let G be the underlying four-valent graph. From the construction of T , we see that T − η(G) consists disks, and each edge of G is adjacent to two distinct disks of T − η(G). Since P is an alternating projection, we can orient each of the edges of G 63 G B1 e3 v1 P e2 B2 e1 e4 v2 C C (B) (A) Figure 5.3: (A) A projection P illustrating the alternating property. (B) The corresponding 4-valent graph G. from overcrossing to undercrossing, as in Figure 5.3(B). This gives a consistent orientation on the boundary of each disk of T − η(G). Suppose C is a component of C± and, for concreteness, orient C so that the arc of K ∩ T± in B1 is to the left of C. Let v1 and v2 be the vertices of G corresponding to B1 and B2 , respectively. Project C onto T , so C crosses an edge e1 ⊂ G adjacent to v1 . Let e1 , . . . , en be the (not necessarily distinct) edges of G met by C between B1 and B2 in order, so e1 is adjacent to v1 and en is adjacent to v2 . Note that e1 must be oriented towards v1 and en must be oriented towards v2 . Between two consecutive edges ei and ei+1 , C crosses a single disk of T − η(G). Hence ei and ei+1 must have opposite algebraic intersection number with C, and each consecutive edge of G met by C crosses C in the opposite direction of the edge before it. Suppose that the arc of K ∩ T± in B2 is also to the left of C. Since en is oriented towards v2 , this means both e1 and en cross C from right to left. Hence n must be odd. Likewise, if the arc of K ∩ T± in B2 is to the right of C, then e1 and crosses C from right to left and en crosses C from left to right and hence n is even. 64 5.3 Decomposing spheres in standard position For each component C ⊂ C± , let w± (C) be a cyclic word in P (puncture) and S (saddle) which records, in order, the intersections of C with K and with bubbles, respectively. Note that w± (C) depends on a choice of orientation for C, but this will not matter for us. By the proof of the alternating property, w± (C) has even length for all C ⊂ C± . With this notation, we have the following lemma of Hayashi. Lemma 5.3.1 (Hayashi, Lemma 2.1(i, ii) of [16]). Let T be a Turaev surface for a composite knot K and let M := M+ M− . Then any generic decomposing sphere A can be isotoped so that each of the following holds. 1. Every component C ⊂ C± satisfies w± (C) = ∅. 2. Each component of A ∩ M is incompressible in M . Proof. By construction of T , the corresponding projection P is such that T − η(P) consists of disks. Hence, if there exists some C ⊂ C± with w± (C) = ∅, C must bound a disk D1 ⊂ T . Let C be an innermost such curve on T , so D1 ∩ A = C. Since A meets K exactly twice, C also bounds a disk D2 ⊂ A − η(K), and D1 ∪ D2 bounds a ball in S 3 − η(K). We can then isotope D2 through this ball to D1 and then off of T , removing C from C± . Similarly, suppose R is a compressible component of A ∩ M and let ∆1 ⊂ M be a compression disk for R. Since R is connected, R and ∆1 are both contained in the same component of M , say M+ . Hence ∆1 ∩ K = ∅ and ∂∆1 also bounds a disk ∆2 ⊂ A − η(K). Again, ∆1 ∪ ∆2 bounds a ball in S 3 − η(K), so we can isotope ∆2 through this ball to ∆1 , replacing R with a disk, which is incompressible. This isotopy may eliminate some other components of A ∩ M , but it will not introduce any new components, so we may iterate until all components of A ∩ M are incompressible in M . 65 Lemma 5.3.1 gives us the following corollary. Corollary 5.3.2. Let A be a generic decomposing sphere with respect to a Turaev surface T . Then A can be isotoped so that every component of C± which bounds a disk on T± also bounds a disk in A ∩ M± which is a peripheral disk in M± . Proof. Let C ⊂ C± be an innermost curve bounding a disk D ⊂ T± , so D ∩ A = C. By isotoping D slightly into M± we get a disk D ⊂ M± with D ∩ A = ∂D . By Lemma 5.3.1, D cannot be a compression disk for the component of A ∩ M± containing ∂D . Hence this component of A ∩ M± must be a disk peripheral to T± . Since the corollary now holds when int(D) ∩ A = ∅, we can complete the proof by inducting on |int(D) ∩ A|. Let Cn ⊂ C± bound a disk Dn ⊂ T± such that |int(Dn ) ∩ A| = n. For each curve C ⊂ int(Dn ) ∩ A we may assume by induction that C bounds a peripheral disk in A∩M± which we will call an interior disk. Isotope these interior disks (rel boundary) so that they lie just above T± but are all still disjoint from one another. Push Dn slightly into M± , as before, until we get a disk Dn ⊂ M± that is disjoint from each of the interior disks, with Dn ∩ A = ∂Dn . As in the base case, Dn cannot be a compression disk for the component of A ∩ M± containing ∂Dn , so this component of A ∩ M± must also be a peripheral disk. The following lemma is similar to Lemma 1 of [29] and Lemma 2.1(iii) of [16]. Lemma 5.3.3. Let A be a generic decomposing sphere with respect to a Turaev surface T . Then A can be replaced by a (non-trivial) generic decomposing sphere A so that if some component C ⊂ C± meets the same bubble B at two different arcs γ1 and γ2 , then for each component α ⊂ C − (γ1 ∪ γ2 ) there is no disk D ⊂ T± such that ∂D consists of α and an arc in α ⊂ (B ∩ T ). 66 γ1 λ R µ D β⊂A γ2 Figure 5.4: The rectangle R and the surgery curve µ from the proof of Lemma 5.3.3. Proof. Suppose such a D ⊂ T+ exists, noting that an identical proof holds for T− . Let D be an innermost such disk on T+ . Let H := B ∩ T+ and λ := K ∩ H. Then there is a rectangle R ⊂ H whose boundary consists of γ1 , γ2 and two arcs in ∂H. (See Figure 5.4.) Since D is innermost, A ∩ R ⊂ ∂R. We now consider two cases, depending on whether or not λ ⊂ R. Case 1 : λ ⊂ R. Since A does not meet the interior of R, γ1 and γ2 must belong to the same saddle σ of A ∩ B. There is a band β := η(α) ∩ M+ ∩ A connecting γ1 to γ2 . By Corollary 5.3.2, we may assume that any component of C+ contained in D bounds a disk in A ∩ M+ which is boundary-parallel in M+ . Hence there is a simple closed curve µ ⊂ β ∪ σ such that µ is isotopic in S 3 − (A ∪ η(K)) to a meridian of η(K). Hence, µ bounds a disk in S 3 − A that meets K exactly once. Surgering A along this disk splits A into two spheres, at least one of which still decomposes K nontrivially. Let A be this decomposing sphere. By isotoping A slightly so that the new puncture no longer meets B, A is generic we have removed γ1 and γ2 from C+ . Case 2 : λ ⊂ R There is still a band β := η(α) ∩ M+ ∩ A connecting γ1 to γ2 , but this time the γi each belong to different saddles σ1 and σ2 in B. Since D is innermost, σ1 and σ2 are adjacent in B. Hence there is a simple closed curve µ ⊂ β ∪ R such that µ bounds a disk ∆ ⊂ M+ − A. 67 Thicken this disk slightly to ∆ × [0, 1] with (∂∆ ) × [0, 1] still contained in β ∪ R, and let ∆ denote this thickened disk. Then ∆ together with the region inside B bounded by σ1 and σ2 is isotopic to a 3-ball in S 3 − (A ∪ η(K)). We can isotope A through this 3-ball to eliminate σ1 and σ2 . Note that while the sphere A from Lemma 5.3.3 gives a nontrivial decomposition of K, it may no longer be the original decomposition given by A. Definition 5.3.4. If A is a generic decomposing sphere satisfying the conditions in Lemmas 5.3.1 and 5.3.3 and Corollary 5.3.2, we will say that A is in standard position with respect to a Turaev surface T . We now have the following lemma, which is similar in spirit to Lemma 2.4 of [16]. Lemma 5.3.5. Let A be a decomposing sphere in standard position with respect to a Turaev surface T . If C± contains a curve C which bounds a disk in T± , then w± (C) = P 2 or P SP S. Proof. Let C be a component of, say, C+ which bounds a disk in T+ . (Again, an identical argument holds for T− .) Choose C such that C is an innermost such curve on T+ , so C bounds a disk D ⊂ T+ − A . We know w+ (C) = ∅ by Lemma 5.3.1, and w+ (C) = P by the construction of T . So either w+ (C) = P 2 or C meets some bubble B. Suppose the latter, and let H := B ∩ T+ and λ := K ∩ H. If λ were contained in D, then C would have to meet B on both sides of λ and would hence violate Lemma 5.3.3. Hence D does not contain the overcrossing of any bubble met by C and, by the alternating property, C cannot meet two bubbles consecutively without first crossing a puncture. Thus w+ (C) = P S or P SP S. If w+ (C) = P S, by passing to T− we see that Lemma 5.3.3 is violated, as ilustrated in Figure 5.5. 68 =⇒ C T− T+ Figure 5.5: If C ⊂ C+ bounds a disk on T+ and has w+ (C) = P S, then C must contribute a curve to C− , which violates Lemma 5.3.3. 5.4 Composite knots with Turaev genus one Throughout this section we restrict our attention to the case where the Turaev surface is a torus and K is a composite knot with gT (K) = 1. To start, fix a genus-one Turaev surface T and the corresponding alternating projection P of K on T . Since any generic decomposing sphere gives rise to a decomposing sphere which is in standard position with respect to T , let A be such a sphere in standard position such that A gives the decomposition K = Ka #Kb . Since we have chosen an orientation on T ⊂ S 3 , T has a well-defined meridian and longitude. Namely, a meridian of T is a homologically nontrivial simple closed curve which bounds a disk in the solid torus inside of T and a longitude of T is such a curve which bounds a disk in the solid torus in the exterior of T . Define a longitude of T± to be a generator of H1 (V± ), as defined in Notation 5.2.2. So a longitude of T+ projects to a longitude of T and a longitude of T− projects to a meridian of T . This also gives a well-defined meridian on T± . If there is a component C ⊂ C− which is separating on T− , then Lemma 5.3.5 tells us that this curve is the unique such separating curve among the components of C− and w− (C) = P 2 or P SP S. If w− (C) = P 2 , then C is also a component of C+ and A decomposes K along C into an alternating summand and a summand which has T as a Turaev surface. Hence, 69 since gT (K) = 1 by assumption, gT (Ka ) + gT (Kb ) = 1. If w− (C) = P SP S, then C+ contains no curves which are separating on T+ . Then C+ consists of a collection of parallel, non-separating curves on T+ , and these curves are either meridians or longitudes of T+ . Lemma 5.4.1. Let A be a decomposing sphere for K in standard position with respect to the genus-one Turaev surface T and suppose that C+ contains no curves which are separating on T+ . Then C+ must consist of longitudes of T+ . Proof. By way of contradiction, suppose C+ consists of meridians of T+ . Since A is a sphere, it cannot meet T+ in a single meridian, so |C+ | ≥ 2. Let C be a component of C+ that is innermost in A , so C bounds a disk D ⊂ A − C+ . Since C is a meridian of T+ , D must be a component of A ∩ V+ . There are at least two such innermost curves on A , so we may choose C such that C meets a bubble B. Let σ be a saddle of B met by C. Then σ ⊂ D and C meets H := B ∩ T+ on both sides of σ. Let γ1 and γ2 be the arcs of C ∩ σ and let α be an arc in σ connecting γ1 and γ2 . The arc α cuts D into 2 disks, D1 and D2 . Since C = ∂D is a meridian of T+ , either ∂D1 or ∂D2 must bound a disk in T+ (after isotoping α through B to H). But then this disk violates Lemma 5.3.3, so we have our desired contradiction. In light of Lemma 5.4.1, we have the following definition. Definition 5.4.2. A decomposing sphere A is a preferred decomposing sphere with respect to a Turaev surface T if A is in standard position and C+ consists of longitudes of T+ . Note that if A is a preferred decomposing sphere and C− contains a curve C which is separating on T , then w− (C) = P SP S. Further, if C− contains any curves which are not separating on T− , then the proof of Lemma 5.4.1 shows that these curves must be longitudes of T− . This gives us the following corollary. 70 bubbles C− C+ P P T+ T− Figure 5.6: Square depictions of T+ and T− as seen from M+ . The thicker lines are components of C± and the cirlces are bubbles on T± with over- and undercrossings shown. Corollary 5.4.3. Let A be a preferred decomposing sphere for K with respect to a genus-one Turaev surface T . Then each component of A ∩ M± is a disk. Proof. This follows from Lemmas 5.3.1 and 5.4.1 and Corollary 5.3.2. We now consider T+ as a square with sides identified in the usual way to depict a torus as a moduli space so that the longitudes of C+ appear as horizontal lines on T+ . (See Figure 5.6.) The curves of C− consist of a single separating curve and/or multiple longitudes on T− . If we are viewing both T+ and T− from outside T (i.e. from M+ ), then when we pass from the square depiction of T+ to T− , any longitudes of T− in C− appear as (lines isotopic to) vertical lines. We will use this depiction to prove the following. Lemma 5.4.4. Let A be a preferred decomposing sphere for K with respect to the genus-one Turaev surface T . Then there is a component C− ⊂ C− which meets K twice. Further, if C− is not separating on T− , then P 2 ⊂ w− (C− ). Proof. Consider a component C+ ⊂ C+ which meets a puncture P and let C− be the corresponding component of C− which meets the same puncture. If w+ (C+ ) = P 2 S i for some i ≥ 0, then clearly w− (C− ) also contains P 2 . 71 B1 B B B2 C (A) =⇒ C− P B1 B2 B1 P B2 B2 B1 C (B) C+ =⇒ C− C+ Figure 5.7: The curves C− , C and C+ from Cases 1 and 2 of the proof of Corollary 5.4.4. Suppose w+ (C+ ) = P 2 S i and fix a square depiction of T+ . Since C− will cross P horizontally on the induced square depiction of T− , we may orient C− so it is travelling upward just to the left of P . (Again, see Figure 5.6.) Applying the alternating property to C+ , we see that C− must then be traveling downward just to the right of P . Since C− is a simple closed curve on T− , C− must change direction again, from upward to downward or downward to upward. By the alternating property once more, this can only happen if C− meets the other puncture of A . Since P 2 ⊂ w+ (C+ ), w− (C− ) contains the sequence SP S. Let B1 and B2 be the bubbles corresponding to the two occurrences of S in this sequence and let λi := K ∩ Bi ∩ T− for i = 1, 2. By the alternating property, λ1 and λ2 both lie on the same side of C− . For concreteness, assume that the λi are both to the left of C− , and C− meets B1 just before meeting B2 . (See Figure 5.7.) Let C be the component of C− immediately to the left of C− , so C also meets both B1 and B2 , and orient C in the same direction as C− . We must now consider three cases, depending on which side of C λ1 and λ2 lie on, hoping to draw a contradiction in each case. Case 1: Both λ1 and λ2 lie to the right of C, as shown in Figure 5.7(A). 72 By the alternating property and the fact that C− meets K twice, C does not meet K and there must be some bubble B which meets C after B1 and before B2 , with the overcrossing of B (with respect to T− ) to the left of C. Any bubble which meets C with its overcrossing to the right of C must also meet C− , so B is the only bubble met by C after B1 and before B2 . But then C+ must meet B twice in a manner which contradicts Lemma 5.3.3. Case 2: The λi are on opposite sides of C, as shown in Figure 5.7(B). For concreteness, suppose λ1 is to the right of C and λ2 is to the left, noting that an identical argument holds if the roles are reversed. Since C meets no punctures, C must meet an even number of bubbles between B1 and B2 . As before, any bubble which meets C with its overcrossing to the right of C also meets C− . Hence C does not meet any other bubbles as it passes from B1 to B2 . But then C+ again contradicts Lemma 5.3.3. Case 3: Both λ1 and λ2 lie to the left of C. In this case, C must meet a bubble B after B1 but before B2 with the overstrand of B to the right of C. But then B also meets C− after B1 and before B2 , which cannot happen. For the next lemma, we will need the following definition. Definition 5.4.5. Given a knot K with Turaev surface T and decomposing sphere A which is in standard position with respect to T , let a(A , T ) := min{|w± (C)| | C is a component of C± which is not separating on T± } Note that a(A , T ) is always a positive, even integer. With this in mind we have the following lemma, which is adapted from Theorem 1.3 of [16]. Lemma 5.4.6. Let K be a composite knot with a genus-one Turaev surface T and let A be 73 a preferred decomposing sphere for K with respect to T . Then a(A , T ) = 2. Proof. Fix a preferred decomposing sphere A for K and let a := a(A , T ). We will use an Euler characteristic argument to show that a = 2. Let l := |C+ |, m := |C− | and s be the number of saddles where A meets a bubble of T . Let M := M+ M− . By Corollary 5.4.3, A ∩ M consists of disks. So A is completely comprised of disks corresponding to the components of A ∩M± and the saddles of A . Hence A has a cellular decomposition consisting of 4s 0-cells, 6s 1-cells and (s + m + l) 2-cells, and χ(A ) = m + l − s. First suppose that C− contains a curve C− which is separating on T− . Since A is preferred, w− (C− ) = P SP S. Since C− ⊂ C− , s is bounded below by both s≥ a(m − 1) + 2 2 and s ≥ al − 2 . 2 (5.1) This gives −s ≤ − am a − 2 − 2 2 and l ≤ 2s 2 + . a a (5.2) Hence we have 2 = χ(A ) = m − s + l 2s 2 + a a 2 2 =m−s 1− + ≤ m− a a ≤m−s+ =m 1− ≤− = a 2 1− 2 a + by (5.2) am a − 2 − 2 2 1− 2 a + 2 a by (5.2) a 4 a a 4 − 2 + = −m −2 + −2+ 2 a 2 2 a a a 4 −2 + −2+ 2 2 a since − m ≤ −1 4 a 74 So a ≤ 2. Since a must be positive and even, a = 2 as desired. If there is no component of C− which is separating on T− , then m is even and at least 2. In this case, the inequality (5.1) becomes s≥ am − 2 2 and s ≥ al − 2 . 2 (5.3) This gives −s ≤ − am −1 2 and l ≤ 2s 2 + a a (5.4) Hence we have 2 = χ(A ) = m − s + l ≤m−s+ =m−s =m 1− = −m 2s 2 + a a a−2 2 am + ≤ m− −1 a a 2 a 2 a−2 a + by (5.4) a−2 a + 2 a by (5.4) a−2 2 + a a a a − 2 + 1 ≤ −2 −2 +1 2 2 since − m ≤ −2 = −a + 5 (5.5) So a ≤ 3 and again, since a is positive and even, we conclude a = 2. This gives us the following corollary. Corollary 5.4.7. Let K be a composite knot with a genus-one Turaev surface T and let A be a preferred decomposing sphere for K with respect to T . If C− has no component which is separating on T− , then there is some curve C ⊂ C± with w± (C) = S 2 . 75 Proof. By Lemma 5.4.6, there is a C ⊂ C± such that |w± (C)| = 2. First suppose C ⊂ C− . If C meets K, then by Lemma 5.4.4, P 2 ⊂ w− (C) and hence w− (C) = P 2 . But then C is also a component of C+ and hence must bound a disk on T , which is a contradiction. Hence w− (C) = S 2 . If C ⊂ C+ , by the above argument w+ (C) = P 2 . Suppose w+ (C) = P S. and let C− be the component of C− which meets the same puncture as C. Then P 2 ⊂ w− (C− ) and, again by Lemma 5.4.4, C− must bound a disk on T− , which contradicts our initial assumptions. Hence w+ (C) = S 2 , as desired. 5.5 Turaev genus one and additivity The goal of this section is to study the following question. Problem 5.5.1. If K = K1 #K2 and gT (K) = 1, is it true that gT (K1 ) + gT (K2 ) = 1? Let K = K1 #K2 . By Proposition 2.3.1, gT (K) = 0 if and only if gT (K1 ) = gT (K2 ) = 0. This and the fact that the Turaev genus is sub-additive immediately gives us the converse of Problem 5.5.1. That is, if gT (K1 ) + gT (K2 ) = 1, then gT (K) = 1. Hence, a positive answer to Problem 5.5.1 would give us additivity of the Turaev genus in the sense that gT (K1 #K2 ) = gT (K1 ) + gT (K2 ) for composite knots of Turaev genus one. The results of this chapter will primarily be stated in terms of preferred decomposing spheres. The following proposition justifies this choice. Proposition 5.5.2. Let K be a composite knot with a genus-one Turaev surface T . Assume that for any decomposing sphere A which is preferred with respect to T , if A gives the decomposition K = Ka #Kb , then gT (Ka ) + gT (Kb ) = 1. Then gT (K) = 1 and for any decomposition K = K1 #K2 , gT (K1 ) + gT (K2 ) = 1. 76 Proof. The fact the gT (K) = 1 follows immediately from the preceding discussion. Let A be a decomposing sphere for K which gives the decomposition K = K1 #K2 . We may assume that A is generic with respect to T . Surger A as in results of Section 5.3 to obtain a decomposing sphere A which is in standard position with respect to T . Let K = Ka #Kb be the decomposition given by A . If C± contains a curve C which is separating on T± with w± (C) = P 2 , then clearly gT (Ka ) + gT (Kb ) = 1. Otherwise, by choice of orientation, we may assume that A is preferred, and gT (Ka ) + gT (Kb ) = 1 by assumption. To show that gT (K1 ) + gT (K2 ) = 1, we induct on the number of summands n in the prime decomposition of K. If n = 2, then A must give the same decomposition as A and there is nothing further to show. For n > 2, if A does not decompose K into K1 #K2 , then A was obtained from A by surgering A (perhaps repeatedly) into two spheres, one of which was A . Hence A contains a summand of either K1 or K2 . For concreteness, say Ka is a summand of K1 . So Kb = K1b #K2b , where K1 = Ka #K1b and K2 = K2b . If gT (Ka ) = 1 and gT (Kb ) = 0, then gT (K1b ) = gT (K2b ) = 0 and hence gT (K1 ) = 1 and gT (K2 ) = 0. If gT (Ka ) = 0 and gT (Kb ) = 1, then, by induction, gT (K1b ) + gT (K2b ) = 1. If gT (K1b ) = 1, then gT (K1 ) = 1 and gT (K2 ) = 0. Otherwise, gT (K2b ) = gT (K2 ) = 1 and gT (K1 ) = 0. Throughout the rest of this section, K will denote a composite knot of Turaev genus one, and A will denote a preferred decomposing sphere for K with respect to some genus-one Turaev surface T . We will first study the cases when |C− | ≤ 3. To state our first result, we need the following definitions. Definition 5.5.3. The alternating genus of a knot K, denoted galt (K), is the minimum genus of an unknotted surface on which K has an alternating projection. The disk-alternating 77 u o o o u u u o Figure 5.8: Overstrands and understrands of K on T± , denoted by o and u, respectively. genus of K, denoted gDA (K), is the minimum genus of an unknotted surface Σ on which K has an alternating projection P such that Σ − η(P) consists entirely of disks. It is clear from this definition that galt (K) ≤ gDA (K) ≤ gT (K) for any knot K. Also, K is alternating if and only if galt (K) = gDA (K) = gT (K) = 0. Definition 5.5.4. Each segment of K between two crossings of the projection P on a Turaev surface T consists of two strands: one overstrand and one understrand, as illustrated in Figure 5.8 Hence each bubble of T corresponds to exactly two overstrands and two understrands of K on T± ∩ T . Theorem 5.5.5. Let K be a composite knot with gT (K) = 1 and let A be a preferred decomposing sphere for K with respect to the genus-one Turaev surface T such that A decomposes K into Ka #Kb . If |C− | = 1, then, up to relabelling of the summands, Ka is alternating and galt (Kb ) = gD (K) = 1. Proof. Since gT (K) = 1, Ka and Kb cannot both be alternating. Since C− consists of a single curve C− , F := A ∩ V− must be a disk and C− = ∂F is a separating curve on T− . Since A is preferred, w− (C− ) = P SP S. Since C− = C− , the two arcs of saddles met by C− must actually belong to the same saddle, and hence C− meets exactly one bubble B. Further, by applying Lemma 5.3.3 and the fact that C+ consists of longitudes, we see that 78 Λ C2 C1 C− T+ = ∂V+ T− = ∂V− Figure 5.9: The components of C− and C+ from Theorem 5.5.5. each component of C− − B can be made into a meridian of T− using an arc contained in B ∩ T , as illustrated in Figure 5.9. Hence C+ consists of exactly two components, C1 and C2 , with w+ (C1 ) = w+ (C2 ) = SP . Let Λ ⊂ T+ be the annulus bounded by C1 ∪ C2 with the overstrand of H+ := B ∩ T+ contained in Λ. Since A decomposes K into Ka #Kb , one of these summands, say Ka , is contained in Λ. The other summand, Kb , is contained in Λb := T+ − Λ, other than at the undercrossing of K at B, which is part of Kb but passes under Λ. (See Figure 5.10.) The two overstrands of B are contained in Λ and the two understrands of B are in Λb . Hence at each of the punctures where K meets ∂Λ, if we traverse K from Λ to Λb , we must be going from an overcrossing to an undercrossing, so we may decompose K into Ka #Kb as in Figure 5.10. This gives an alternating projection of Ka , as desired. We also have a projection of Kb on a torus which is alternating, so galt (Kb ) = 1. It remains to show that the complement of Kb on this torus consists of disks. To do this, recall that in the construction of the original Turaev surface T we glued disks to a thickening of a diagram of K in R2 , and these disks came from the all-A and all-B smoothings of the diagram. So T − η(P) consists of disks which we will call A-disks and B-disks, depending on which smoothing type they originally came from. Project C1 and C2 onto T . Then each of the Ci pass through exactly one A-disk 79 u o o u Λ Λb B T+ u o o u Ka Kb Figure 5.10: The decomposition of K = Ka #Kb from Theorem 5.5.5. The shaded regions contain alternating projections of subtangles of Ka and Kb . and one B-disk. Denote these disks by Ai and Bi , respectively. (See Figure 5.11.) When we remove Ka from T , we have the effect of removing from T all of the A- and B-disks which meet Λ, other than the Ai and Bi . We then merge B1 and B2 into one region of T − η(Kb ) which spans Λ − (A1 ∪ A2 ). Denote this region by B. Now B is a disk unless B1 = B2 , which cannot happen since, by construction, P has no nugatory crossings on T . Since nothing in Λb was changed when Ka was removed, T − η(Kb ) consists of disks, as desired. Theorem 5.5.6. Let K be a composite knot with gT (K) = 1 and let A be a preferred decomposing sphere for K with respect to the genus-one Turaev surface T such that A decomposes K into Ka #Kb . Then |C− | = 2. Proof. Suppose the |C− | = 2 and let C1 and C2 be the two components of C− . Then C1 and C2 are longitudes of T− . Without loss of generality, P 2 ⊂ w− (C1 ) by Lemma 5.4.4. Note 80 additional A- and B-disks A1 Λ B1 B A1 Λ A2 A2 B2 =⇒ K Kb Figure 5.11: The A- and B-disks of T locally at the annulus Λ before and after the decomposition of K in Theorem 5.5.5. that w− (C2 ) = ∅, and every saddle met by C2 also meets C1 . Now A cuts T− into exactly two annuli and, as before, Ka and Kb are each contained in one of these annuli except at the undercrossings corresponding to bubbles met by C− . Let Λa be the annulus containing Ka away from these undercrossings and likewise define Λb with respect to Kb . We will argue that both Ka and Kb are, in fact, alternating knots, in which case we have the contradiction that K did not have Turaev genus one to begin with. Focusing first on Λa , note that, by the alternating property, each component of C− meets an even number of bubbles, and every other bubble which meets C− has its overcrossing contained in Λa . (See Figure 5.12(A).) We will refer to these bubbles with their overcrossings in Λa as a-bubbles and to the other bubbles which meet C− (with their overcrossings contained in Λb ) as b-bubbles. The a-bubbles cut Λa into rectangles, each of which contain a (possibly trivial) alternating subtangle of Ka . The overcrossings of the a-bubbles connect each of these rectangles together and hence will be called connecting arcs. (See Figure 5.12(B)). 81 B3 a3 b3 P2 a3 u o o a1 =⇒ Λa B2 a2 (A) B1 b2 o o u u o a1 u u u a2 u o o o u =⇒ b1 (B) o u u u o o u o (C) Figure 5.12: (A) An example of the annulus Λa from Theorem 5.5.6, which, in this example, meets six bubbles. The overcrossings of the a-bubbles are labeled as ai . The b-bubbles are labeled as Bi . (B) A diagram of Ka which is contained in Λa away from the undercrossings of the b-bubbles, which are labeled as bi . Each of the rectangular regions contains an alternating subtangle of Ka , and the ai are now the connecting arcs of the diagram. (C) The induced alternating projection of Ka . The undercrossings of the b-bubbles, which are part of Ka and pass behind the rectangles of Λa in the original projection of K, can each be isotoped to cross under a connecting arc, giving an alternating projection of Ka as shown in Figure 5.12(C). By reversing the roles of the a- and b-bubbles, the same argument can be applied to Λb to get an alternating projection of Kb , as well. Theorem 5.5.7. Let K be a composite knot with gT (K) = 1 and let A be a preferred decomposing sphere for K with respect to the genus-one Turaev surface T such that A decomposes K into Ka #Kb . If |C− | = 3, then gT (Ka ) + gT (Kb ) = 1. Proof. Let C1 and C2 be the components of C− which do not meet K. Note that the third component C− ⊂ C− must be separating on T− with w− (C− ) = P SP S. The curves of C− cut T− into three surfaces: a disk D bounded by C− , a punctured annulus Λa with ∂Λa = C− , and an annulus Λb with ∂Λb = C1 ∪ C2 . (See Figure 5.13.) Note that if C− meets the same bubble B twice, then, when we pass to T+ , the components of C− − B create 82 Λb Λa C1 B1 D C− C2 B2 T− Figure 5.13: The components of C− and the regions of T− − η(C− ) from Theorem 5.5.7. either separating curves or meridians in T+ , which cannot happen. So C− meets two distinct bubbles, B1 and B2 . By the alternating property and the fact that C1 meets no punctures, w− (C1 ) = S 2n for some n ∈ Z. Further, by Lemma 5.3.3, C1 cannot meet the same bubble more than once. So C1 meets n distinct bubbles whose overcrossings are contained in Λa and another n distinct bubbles whose overcrossings are contained in Λb . If B1 and B2 were both to meet C1 , then C2 would meet the same n bubbles as C1 with overcrossings contained in Λb but only (n − 2) bubbles with overcrossings contained in Λa . This cannot happen, so, up to relabeling, we may assume B1 meets C1 and B2 meets C2 . Note also that since no component of C− can meet any bubble more than once, each bubble which meets A can contain only one saddle. As in the proof of Theorem 5.5.6, after relabelling the summands of Ka #Kb if necessary, Ka is contained in Λa away from the undercrossings of the b-bubbles (which are part of Ka but pass under Λb ). Similarly, Kb is contained in Λb ∪ D away from the undercrossings of the a-bubbles. Note that B1 and B2 are both a-bubbles. The b-bubbles cut Λb into rectangles which contain alternating subtangles of Kb , as shown in Figure 5.14. The overcrossings of the b-bubbles are connecting arcs for these rectangles. 83 b2 A2 a2 A1 u b2 a2 u oo u Λb b3 b1 =⇒ B1 C− o o u b1 b3 u oo ˜1 b o uou D B2 (A) ˜2 b (B) u o a1 u =⇒ ˜1 b o o u uo o u a1 u u u o o u o ˜2 b (C) Figure 5.14: (A) An example of the annulus Λb and the disk D from Theorem 5.5.7. The a-bubbles, other than B1 and B2 , are labeled as Ai , and the overcrossings of the b-bubbles are labeled as bi . (B) A diagram of Kb contained (mostly) in Λb ∪ D. The bi are now connecting arcs of the diagram, and each of the rectangular regions contains an alternating subtangle. The ai and ˜i correspond to the overcrossings of the Ai and Bi , respectively. (C) b The induced alternating projection of Kb . The disk D also contains an alternating subtangle of Kb , which is connected to one of the rectangles of Λb via the undercrossings of B1 and B2 . As in the proof of Theorem 5.5.6, the undercrossings of the a-bubbles (including B1 and B2 ) can be isotoped to give an alternating diagram for Kb . Hence gT (Kb ) = 0. Similarly, the a-bubbles (including B1 and B2 ) cut Λa into rectangles, two of which, R1 and R2 , are adjacent to C− . (See Figure 5.15.) Each of the rectangles of Λa − {a-bubbles} contains an alternating subtangle of Ka , and these rectangles are connected to each other via the overcrossings of the a-bubbles. Note that there are three connecting arcs between R1 and R2 : two coming from the overcrossings of B1 and B2 , and another coming from the unknotted arc α ⊂ D that was created in the decomposition of K. Each of the rectangles of Λa − {a-bubbles} is also connected to itself via the undercrossing of some b-bubble. These undercrossings can be isotoped under the connecting arcs to give a diagram for Ka as shown 84 a2 a2 a1 =⇒ Λa R1 R2 B2 C− α B1 (A) o a1 ˜1 b α u o u =⇒ u o o uo u u o o uo u ˜2 b (B) (C) Figure 5.15: (A) An example of the punctured annulus Λa from Theorem 5.5.7. The overcrossings of the a-bubbles, other than B1 and B2 , are labeled as ai . (B) A projection of Ka contained (mostly) in Λa . The ai are now connecting arcs in the diagram, as are α and the overcrossings of the Bi , which are labeled ˜i . Each of the rectangular regions contains an b alternating subtangle. (C) A diagram for Ka . in Figures 5.15(C) and 5.16(A). Using this diagram, we get an associated ribbon graph of genus one, as shown in Figure 5.16(C). Hence gT (Ka ) = 1, as desired. Before moving on to the cases when |C− | ≥ 4, we have the following corollary. Corollary 5.5.8. Let K be a composite knot with gT (K) = 1 and let A be a preferred decomposing sphere for K with respect to the genus-one Turaev surface T such that A decomposes K into Ka #Kb . If no bubble of T meets A more than once, then either gT (Ka )+gT (Kb ) = 1 or |C− | = 1 and, up to relabelling, Ka is alternating and galt (Kb ) = gDA (K) = 1. Proof. If |C− | is even, then C− has no curve which is separating on T− and, by Corollary 5.4.7, there is a component C ∗ ⊂ C± with w± (C ∗ ) = S 2 . Since C− has no component which is separating on T− , we may choose the orientation on T so that C ∗ ⊂ C+ . Then, since each bubble contains at most one saddle, |C− | = 2, which cannot happen by Theorem 5.5.6. Suppose |C− | is odd, so there is a curve C− ⊂ C− which is separating on T− with w− (C− ) = P SP S. Since no bubble contains more than one saddle, each longitude C ⊂ C− 85 alternating planar =⇒ ouououou =⇒ c c b a a (A) b (C) (B) Figure 5.16: (A) A simplification of the diagram from Figure 5.15(C). (B) A diagram for the all-A smoothing of Ka . The thicker lines are each Seifert circles. The lighter line segments represent crossings from the knot diagram which will be edges in the ribbon graph. The shaded gray region contains additional Seifert circles and edges from the alternating portion of the knot diagram. (C) A genus-one ribbon graph for Ka . The rectangular region contains a planar sub-ribbon graph. The thicker gray lines each represent two or more planar edges connecting the vertices a, b and c to vertices in the planar portion of the graph. T+ T− Figure 5.17: The curves of C± from Corollary 5.5.8 with the bubbles of T± . has w− (C) = S 2n where |C+ | = 2n. Likewise, C+ has two components C1 and C2 with w+ (Ci ) = P S 2m+1 and all other curves C ⊂ C+ have w+ (C) = S 2m where |C− | = 2m + 1. (See Figure 5.17.) By Corollary 5.4.6, there is a curve C ∗ ⊂ C± with |w± (C ∗ )| = 2. First suppose C ∗ ⊂ C+ . If w+ (C ∗ ) = P S, then m = 0 and |C− | = 1, so the result holds by Theorem 5.5.5. If w+ (C ∗ ) = S 2 , then m = 1 and |C− | = 3, so the result holds by Theorem 5.5.7. Hence we may assume C ∗ ⊂ C− and w− (C ∗ ) = S 2 , so |C+ | = 2. Then C+ cuts T+ into annuli Λa and Λb as in the proof of Theorem 5.5.6 and these annuli give rise to diagrams for the summands 86 Λa =⇒ u o u ou o u u ou o o Ka Λb o u o u =⇒ o u u u o o u o Kb Figure 5.18: An example of the annuli of T+ − η(C+ ) from Corollary 5.5.8. The corresponding diagram of Kb is alternating, and the diagram of Ka has Turaev genus 1 by the arguments of Figure 5.16. Ka and Kb of K with gT (Ka ) + gT (Kb ) = 1, as illustrated in Figure 5.18. In light of this corollary, what remains to be studied are decompositions given by preferred spheres such that |C− | ≥ 4 and there is some bubble on the Turaev surface meeting this decomposing sphere multiple times. The proof of Theorem 5.5.9, which restricts our attention to the case when |C− | = 4, employs an Euler characteristic argument which may, in the future, be helpful in studying the general case of |C− | ≥ 4. Let T be a genus-one Turaev surface for a knot K and let A be a preferred decomposing sphere for K with respect to T such that C− contains no curve which bounds a disk on T− . By Corollary 5.4.3, each component of C± contributes one disk to A ∩ M± , and each of these disks are meridional disks in M± . These disks cut M± into a collection of 3-balls. We will 87 call the 3-balls of M− − η(A ) drums and those of M+ − (A ) caps. Let B be some bubble of T which meets A in n ≥ 1 saddles. These saddles cut B into (n + 1) 3-balls which we will denote by bi for 0 ≤ i ≤ n. Recall that V+ = M− ∪ {bubbles}. Let b0 be the component of B −η(A ) which is adjacent to M− in V+ and let bi+1 be adjacent to bi in B for 0 ≤ i ≤ n − 1. Note that V+ − η(A ) has exactly two components, which we will denote by Y1 and Y2 . Half of the drums of M− − η(A ) are contained in Y1 and the other half are in Y2 , and we will regard these as the 0-cells of the Yi . In V+ − η(A ), b0 is part of some drum and the other bi for 1 ≤ i ≤ n each contribute a 1-cell to some Yi . Hence Y1 Y2 has a cellular decomposition with 0-cells corresponding to drums of M− and 1-cells corresponding to the saddles of A . Then S 3 − η(A ) consists of two 3-balls, X1 and X2 , where each Xi is obtained from Yi by attaching exactly half the caps of M+ , which are the 2-cells of Xi . Hence for X := X1 2 = χ(X) = |C− | − s + |C+ | X2 , (5.6) where s is the total number of saddles in A ∩ {bubbles}. For a bubble B which meets n saddles, bn meets just one cap of M+ and, for i = 0 or n, bi meets exactly two caps. We will call bn a 1-bridge and the other bi for i = 0, n will be called 2-bridges. For each bridge bi , we will say that bi has length i. Note that a bridge of length i connects two drums δ1 , δ2 ⊂ M− by passing over 2i − 1 drums between δ1 and δ2 . Let ∆i , βi , Γi and κi denote the number of drums, 1-bridges, 2-bridges and caps in Xi , 1 respectively, for i = 1, 2. Then ∆i = 1 |C− |, κi = 2 |C+ | and we have 2 1 = χ(Xi ) = ∆i + κi − βi − Γi 88 (5.7) for i = 1, 2. Theorem 5.5.9. Let K be a composite knot with gT (K) = 1 and let A be a preferred decomposing sphere for K with respect to the genus-one Turaev surface T . If there exists C ⊂ C− with w− (C) = S 2 , then |C− | = 4. Note that, by Lemma 5.4.6, there is some component C ⊂ C± with w± (C) = S 2 , so we are simply choosing the orientation on T such that C ⊂ C− . Proof. Assume |C− | = 4 and let C ⊂ C− have w− (C) = S 2 . Since C+ consists of longitudes of T+ , C must meet two distinct bubbles. Hence, by Lemma 5.3.3, each bubble meets each component of C− at most once and therefore contains at most 2 saddles. Since every component of C+ meets at least one of the bubbles meeting C, |C+ | ≤ 8. Hence ∆i = 2 and κi ≤ 4 for i = 1, 2. (In general, if we assume w− (C) = S 2 for some C ⊂ C− , then κi ≤ 2∆i .) By Corollary 5.5.8, some bubble of T meets A more than once, so, without loss of generality, we may assume that Γ2 > 1. The number and length of the 1-bridges in X completely determines the number of caps and 2-bridges. To see this, notice that each cap of Xi is attached to Yi along a gluing circle which meets either 1 bridge of length 2 or 2 bridges each of length 1. Hence each 1-bridge b1 ⊂ Xi of length 1 corresponds to half of a cap and no 2-bridges in Xi , and b1 contributes − 1 to χ(Xi ). Likewise, each 1-bridge b2 ⊂ X1 of 2 length 2 corresponds to 1 2-bridge (of length 1) in X2 and 1 cap in each of X1 and X2 . This j means the 1-bridges of length 2 contribute nothing to χ(X1 ) or χ(X2 ). Letting βi denote the number of 1-bridges of length j in Xi for i = 1, 2, we have 1 1 1 = χ(Xi ) = ∆i − βi . 2 89 (5.8) (B) (A) (C) Figure 5.19: (A) The configuration of T− from the first row of Table 5.1. (B) A diagram of Ka Kb on the torus given by this coniguration. The shaded boxes contain (possibly trivial) alternating sub-diagrams. (C) The resulting diagram of Ka Kb on R2 , which shows that Ka and Kb are alternating knots. 1 Since ∆i = 2, we see that βi = 2 for i = 1, 2. 2 1 Note that β1 = Γ2 . This gives β1 − Γ2 = β1 = 2. An identical argument shows β2 − Γ1 = 2. Hence 1 1 β1 + β2 − Γ1 − Γ2 = β1 + β2 = 4. (5.9) Equation (5.7) applied to χ(X) gives β1 + β2 + Γ1 + Γ2 = 2 + κ1 + κ2 . (5.10) Combining equations (5.9) and (5.10) with the fact that κ1 = κ2 ≤ 4, we have 2(β1 + β2 ) − 4 ≤ 10 (5.11) 2 2 2 1 1 This means β1 + β2 ≤ 7 and hence β1 + β2 ≤ 3 with β1 = Γ2 > 0 and β1 = β2 = 2. In light of the alternating property, this leaves exactly 6 possible configurations for the bubbles of T with respect to C− , which are shown in Table 5.1. Since some curve C− ⊂ C− has P 2 ⊂ w− (C− ) by Lemma 5.3.5, each of these configurations decomposes K into 2 disjoint, 90 2 β1 2 β2 1 0 1 1 2 0 2 1 3 0 Possible configurations of T− Table 5.1: All possible configurations of C− and the bubbles of T− from Theorem 5.5.9. 91 unlinked summands. See, for example, Figure 5.19. Hence K is alternating, contradicting the fact that gT (K) = 1. 92 BIBLIOGRAPHY 93 BIBLIOGRAPHY [1] Problems in low-dimensional topology, Geometric topology (Athens, GA, 1993) (Rob Kirby, ed.), AMS/IP Stud. Adv. Math., vol. 2, Amer. Math. Soc., Providence, RI, 1997, pp. 35–473. MR 1470751 [2] Tetsuya Abe, The Turaev genus of an adequate knot, Topology Appl. 156 (2009), no. 17, 2704–2712. MR 2556029 (2011a:57008) [3] Cheryl Balm, Generalized crossing changes in satellite knots, Preprint: arXiv:1210.0885 [math.GT]. [4] Cheryl Balm, Stefan Friedl, Efstratia Kalfagianni, and Mark Powell, Cosmetic crossings and seifert matrices, Communications in Analysis and Geometry 20 (2012), no. 2, 235– 254. [5] Cheryl Balm and Efstratia Kalfagianni, Cosmetic crossings of twisted knots, Preprint: arXiv:1301.6369 [math.GT]. [6] Joan S. Birman, Braids, links, and mapping class groups, Princeton University Press, Princeton, N.J., 1974, Annals of Mathematics Studies, No. 82. MR 0375281 (51 #11477) [7] B´la Bollob´s and Oliver Riordan, A polynomial invariant of graphs on orientable sure a faces, Proc. London Math. Soc. (3) 83 (2001), no. 3, 513–531. MR 1851080 (2002f:05056) [8] Patrick J. Callahan, John C. Dean, and Jeffrey R. Weeks, The simplest hyperbolic knots, J. Knot Theory Ramifications 8 (1999), no. 3, 279–297. [9] Jae Choon Cha and Charles Livingston, Knotinfo: Table of knot invariants, 2011, http://www.indiana.edu/ knotinfo. [10] Abhijit Champanerkar, David Futer, Ilya Kofman, Jessica Purcell, and Walter Neumann, Volume bounds for generalized twisted torus links, Mathematical Research Letters 18 (2011), no. 6, 1097–1120. [11] Abhijit Champanerkar and Ilya Kofman, Twisting quasi-alternating links, Proc. Amer. Math. Soc. 137 (2009), no. 7, 2451–2458. MR 2495282 (2010b:57005) 94 [12] Abhijit Champanerkar, Ilya Kofman, and Neal Stoltzfus, Graphs on surfaces and Khovanov homology, Algebr. Geom. Topol. 7 (2007), 1531–1540. MR 2366169 (2008m:57005) [13] Oliver T. Dasbach, David Futer, Efstratia Kalfagianni, Xiao-Song Lin, and Neal W. Stoltzfus, The Jones polynomial and graphs on surfaces, J. Combin. Theory Ser. B 98 (2008), no. 2, 384–399. MR 2389605 (2009d:57020) [14] David Gabai, Foliations and the topology of 3-manifolds. II, J. Differential Geom. 26 (1987), no. 3, 461–478. [15] C. McA. Gordon, Boundary slopes of punctured tori in 3-manifolds, Trans. Amer. Math. Soc. 350 (1998), no. 5, 1713–1790. MR 1390037 (98h:57032) [16] Chuichiro Hayashi, Links with alternating diagrams on closed surfaces of positive genus, Math. Proc. Cambridge Philos. Soc. 117 (1995), no. 1, 113–128. MR 1297898 (95m:57015) [17] John Hempel, 3-manifolds, AMS Chelsea Publishing, Providence, RI, 2004, Reprint of the 1976 original. MR 2098385 (2005e:57053) [18] Efstratia Kalfagianni, Cosmetic crossing changes of fibered knots, J. Reine Angew. Math. (Crelle’s Journal) 2012, no. 669, 151–164. [19] Efstratia Kalfagianni and Xiao-Song Lin, Knot adjacency, genus and essential tori, Pacific J. Math. 228 (2006), no. 2, 251–275. MR 2274520 (2007k:57010) [20] Mikhail Khovanov, A categorification of the Jones polynomial, Duke Math. J. 101 (2000), no. 3, 359–426. MR 1740682 (2002j:57025) [21] Masaharu Kouno, Kimihiko Motegi, and Tetsuo Shibuya, Twisting and knot types, J. Math. Soc. Japan 44 (1992), no. 2, 199–216. [22] Jerome P. Levine, Invariants of knot cobordism, Invent. Math. 8 (1969), 98–110; addendum, ibid. 8 (1969), 355. MR 0253348 (40 #6563) [23] W. B. Raymond Lickorish, An introduction to knot theory, Graduate Texts in Mathematics, vol. 175, Springer-Verlag, New York, 1997. MR MR1472978 (98f:57015) [24] Adam M. Lowrance, On knot Floer width and Turaev genus, Algebr. Geom. Topol. 8 (2008), no. 2, 1141–1162. MR 2443110 (2010f:57036) 95 [25] Herbert C. Lyon, Simple knots with unique spanning surfaces, Topology 13 (1974), 275–279. [26] Ciprian Manolescu and Peter Ozsv´th, On the Khovanov and knot Floer homologies a of quasi-alternating links, Proceedings of G¨kova Geometry-Topology Conference 2007, o G¨kova Geometry/Topology Conference (GGT), G¨kova, 2008, pp. 60–81. MR 2509750 o o (2010k:57029) [27] Vassily O. Manturov, Minimal diagrams Preprint:arXiv:math/0501393 [math.GT]. of classical and virtual links, [28] Darryl McCullough, Homeomorphisms which are Dehn twists on the boundary, Algebr. Geom. Topol. 6 (2006), 1331–1340 (electronic). MR 2253449 (2007f:57043) [29] W. Menasco, Closed incompressible surfaces in alternating knot and link complements, Topology 23 (1984), no. 1, 37–44. MR 721450 (86b:57004) [30] Kimihiko Motegi, Knotting trivial knots and resulting knot types, Pacific J. Math. 161 (1993), no. 2, 371–383. MR 1242205 (94h:57012) [31] Peter Ozsv´th and Zolt´n Szab´, Holomorphic disks and knot invariants, Adv. Math. a a o 186 (2004), no. 1, 58–116. MR 2065507 (2005e:57044) [32] , On the Heegaard Floer homology of branched double-covers, Adv. Math. 194 (2005), no. 1, 1–33. MR 2141852 (2006e:57041) [33] V. V. Prasolov and A. B. Sossinsky, Knots, links, braids and 3-manifolds, Translations of Mathematical Monographs, vol. 154, American Mathematical Society, Providence, RI, 1997, An introduction to the new invariants in low-dimensional topology, Translated from the Russian manuscript by Sossinsky [Sosinski˘ MR 1414898 (98i:57018) ı]. [34] J´zef H. Przytycki, Positive knots have negative signature, Bull. Polish Acad. Sci. Math. o 37 (1989), no. 7-12, 559–562 (1990). MR 1101920 (92a:57010) [35] Jacob Andrew Rasmussen, Floer homology and knot complements, ProQuest LLC, Ann Arbor, MI, 2003, Thesis (Ph.D.)–Harvard University. MR 2704683 [36] Dale Rolfsen, Knots and links, Publish or Perish Inc., Berkeley, Calif., 1976, Mathematics Lecture Series, No. 7. MR 0515288 (58 #24236) 96 [37] Martin Scharlemann and Abigail Thompson, Link genus and the Conway moves, Comment. Math. Helv. 64 (1989), no. 4, 527–535. MR 1022995 (91b:57006) [38] Otto Schreier, Uber die gruppen aabb = 1, Abh. Math. Sem. Univ. Hamburg 3 (1924), no. 167-169. [39] Horst Schubert, Knoten und vollringe, Acta Mathematica 90 (1953), 131–286, 10.1007/BF02392437. [40] Tetsuo Shibuya, On the cobordism of compound knots which are T -congruent, Kobe J. Math. 2 (1985), no. 1, 71–74. [41] John R. Stallings, Constructions of fibred knots and links, Algebraic and geometric topology (Proc. Sympos. Pure Math., Stanford Univ., Stanford, Calif., 1976), Part 2, Proc. Sympos. Pure Math., XXXII, Amer. Math. Soc, Providence, R.I., 1978, pp. 55–60. [42] William P. Thurston, Three-dimensional manifolds, Kleinian groups and hyperbolic geometry, Bull. Amer. Math. Soc. (N.S.) 6 (1982), no. 3, 357–381. MR 648524 (83h:57019) [43] Ichiro Torisu, On nugatory crossings for knots, Topology Appl. 92 (1999), no. 2, 119– 129. MR 1669827 (2000b:57015) [44] Hale F. Trotter, On S-equivalence of Seifert matrices, Invent. Math. 20 (1973), 173–207. MR 0645546 (58 #31100) [45] Wilbur Whitten, Isotopy types of knot spanning surfaces, Topology 12 (1973), 373–380. MR 0372842 (51 #9046) [46] Chad Wiley, Nugatory crossings in closed 3-braid diagrams, ProQuest LLC, Ann Arbor, MI, 2008, Thesis (Ph.D.)–University of California, Santa Barbara,. MR 2712035 97