Arxiv:2104.13592V1 [Math.GT] 28 Apr 2021 Covers [Wat10]

KHOVANOV HOMOLOGY AND STRONG INVERSIONS

ARTEM KOTELSKIY, LIAM WATSON, AND CLAUDIUS ZIBROWIUS

Abstract. There is a one-to-one correspondence between strong inversions on knots in the three-sphere and a special class of four-ended tangles. We compute the reduced Khovanov homology of such tangles for all strong inversions on knots with up to 9 crossings, and discuss these computations in the context of earlier work by the second author. In particular, we provide a counterexample to [Wat17, Conjecture 29] as well as a reﬁnement of and additional evidence for [Wat17, Conjecture 28].

The Brieskorn spheres Σ(2, q, 2nq ∓ 1) may be obtained by Dehn surgery on a torus knot in the three-sphere, namely, these are the integer homology spheres 3 ( ) where is the S±1/n T2,q T2,q positive (2, q) torus knot. These homology spheres admit Seifert ﬁbrations, with base orbifold S2(2, q, 2nq ∓ 1). Denoting by Σ(A, b) the two-fold branched cover of A with branch set b, each of these manifolds admits two descriptions as a two-fold branched cover:

3 ( ) ∼ Σ( 3 ) ∼ Σ( 3 (± 1 )) S±1/n T2,q = S ,Tq,∗2qn 1 = S , τ n ∓ This construction might be best termed as classical; for our purposes it is helpful to review the notation introduced in [Wat10]. The first of these two-fold branched covers results from an involution on the Seifert fibred space that preserves an orientation on the fibres—we refer to this as the Seifert involution. The second of these arises from the Montesinos involution, which reverses an orientation on the fibres; the branch set in question arises from the Montesinos trick, that is, by first constructing Figure 1. The branch 3 a tangle (B , τ) over which the exterior of T2,q is realized as a two-fold set τ(+1) when q = 5 branched cover. We review the construction below as it is central to 1 our enumeration of tangles. In particular, the branch set τ(± n ) is an explicit Montesinos link, which despite the notation depends on q. Generically, these two branch sets are distinct knots. For example, when q = 5 and n = 1 we have 3 ∼ 3 ∼ 3 S+1(T2,5) = Σ(S ,T5∗,9) = Σ(S , τ(+1)) and it can be calculated that

dim Kh(g T5∗,9) = 57 > 15 = dim Kh(g τ(+1)) where the reduced Khovanov homology is taken over the two-element ﬁeld F. (The Montesinos branch set is shown in Figure1, and the construction of the tangle in this case is reviewed in Figure2.) This example serves to answer a question due to Ozsváth in the negative: the total dimension of the mod 2 reduced Khovanov homology is not an invariant of two-fold branched arXiv:2104.13592v1 [math.GT] 28 Apr 2021 covers [Wat10]. This paper focusses on tangles admitting knot exteriors as two-fold branched covers, and the immersed curves that arise as the reduced Khovanov invariants of these tangles. Coeﬃcients are restricted to F throughout. 1. Strong inversions A knot K in S3 is invertible if it admits an isotopy exchanging a choice of orientation on K with the reverse of this choice. A strong inversion is an inversion realized by an involution

AK is supported by an AMS-Simons travel grant. LW is supported by an NSERC discovery/accelerator grant.

1 2 ARTEM KOTELSKIY, LIAM WATSON, AND CLAUDIUS ZIBROWIUS of the three-sphere. Note that invertible knots which are not strongly invertible exist [Har80], but that when restricting to hyperbolic knots the two symmetries are equivalent. Following Sakuma [Sak86], given a knot K and a strong inversion h, we will call the pair (K, h) a strongly invertible knot. Strongly invertible knots (K, h) and (K0, h0) are equivalent if there exists an 3 orientation preserving homeomorphism f on S for which f(K) = K0 (so that K and K0 are 1 equivalent knots) and h = f − ◦ h0 ◦ f. Alternatively, a strong inversion on a given knot may be viewed as the conjugacy class of an order-two element of the mapping class group of the exterior of the knot; this mapping class group is known as the symmetry group of the knot K. For hyperbolic knots, this group is a subgroup of a dihedral group; see [Sak86, Proposition 3.1]. A strong inversion gives rise to an involution on a knot’s exterior with one-dimensional fixed point set meeting the boundary in exactly 4 points. Note that, according to the Smith conjecture, given an involution on the three-sphere with one-dimensional fixed point set, the fixed point set is unknotted; as a result the fixed point set of the strong inversion on the exterior is a pair of unknotted arcs. Taking the quotient of the order two action on the knot exterior gives rise to a three ball h 3 3 B = (S r ν(K))/h, and the image of the fixed point set of the strong inversion gives rise to a pair of properly embedded arcs τ = Im(Fix(h)). Therefore, given a strongly invertible knot, the knot exterior may be µ viewed as a two-fold branched cover over a four-ended tangle τ in a 3 3 ∼ 3 three ball: h y (S ,K) so that S r ν(K) = Σ(B , τ). We refer to the tangle T = (B3, τ) as the associated quotient tangle to a given knot with strong inversion; see Figure2 for an explicit example. Figure 2. The cinqfoil exterior as a two-fold In this context, the natural notion of equivalence on tangles is homeo- branched cover morphism of the pair (B3, τ) where the boundary is fixed only set-wise. Note that this is, of course, more flexible than the requirement that the boundary be fixed point-wise as is perhaps more common when considering tangle diagrams. In order to be clear about the distinction, we will refer to this latter as a framed tangle. We remark that [Wat10, Definition 3] gives a third notion of equivalence by introducing sutured tangles. This object will not play an explicit role here, however, consulting Figure2 the reader may find that sutures are a helpful tool for tracking the image of the knot meridian µ in the quotient. From this point of view, there is a distinguished trivial tangle obtained as the quotient of a solid torus, that is, the associated quotient tangle for the trivial knot. The equivalence class of this tangle is the rational tangle. We will have need for choices of representatives Qp/q as p 1 described in Figure3: For any q ∈ QP , there is a framed tangle diagram Qp/q. This choice will allow us to make use of the Montesinos trick: 3 ( ) ∼ Σ( 3 ( p )) Sp/q K = S , τ q p p p where q -surgery along the knot K corresponds to the knot τ( q ) obtained by gluing the − q - rational tangle Q p/q to the tangle τ as in Figure 3e. −

Q p/q τ −

p (a) Q∞ (b) Q0 (c) Q1 (d) Q1/3 (e) τ( q )

Figure 3. Examples of framed rational tangles (a-d) and the rational closure of a four-ended tangle τ (e) KHOVANOV HOMOLOGY AND STRONG INVERSIONS 3

µ µ

λ + n1µ λ + n2µ

Figure 4. Two strong inversions on the knot 946 (top row) together with the associate quotient tangle for each (bottom row). In order to present minimal crossing diagrams for the associated quotient tangles, this is the one place where we present something other than the preferred tangle representative: The ﬁrst symmetry gives rise to a tangle T1 shown with framing n1 = 2, while the second symmetry gives rise to a tangle T2 with framing n2 = −6.

In particular, we are fixing a preferred representative for our associated quotient tangle once and for all: Given (K, h), this is the tangle T = (B3, τ) such that the rational closure τ(0) corresponds to surgery along the Seifert longitude and τ(∞) corresponds to S3 (K) =∼ S3. The latter implies that τ(∞) is the unknot, as observed above. In fact, there is a bijection∞ between equivalence classes of non-trivial strongly invertible knots (K, h) and tangles (B3, τ) for which τ(∞) is unknotted (this follows from [GL89, Theorem 2]; compare [Wat17, Proposition 9]). As such, associated quotient tangles provide an invariant of strong inversions. Strong inversions give a means of enumerating interesting tangles, and we use this strategy below. Observe that, given a hyperbolic knot admitting a pair of distinct strong inversions h1 and , by Thurston’s hyperbolic Dehn filling theorem, 3 ( ) is hyperbolic for all but finitely h2 Sp/q K p many slopes q . Moreover Σ( 3 ( p )) ∼ 3 ( ) ∼ Σ( 3 ( p )) S , τ1 q = Sp/q K = S , τ2 q p p and, generically, the branch sets τ1( q ) and τ2( q ) are distinct knots. It is a striking fact that for all K with fewer than 9 crossings p p dim Kh(g τ1( q )) = dim Kh(g τ2( q )) that is, finding a negative answer to Ozsváth’s question in the hyperbolic setting is surprisingly difficult. This observation follows from Theorem 3.1 below. A range of examples of strong inversions and associated quotient tangles are considered in [Wat17]. We expand this list in a systematic way below, and highlight in particular the knot 946 in the Rolfsen table; see Figure4. It is worth noting that these tangles distinguish the strong inversions in question: Ignoring one strand, the second tangle contains 819 as a subknot, while the only subknots in the first tangle are trefoils. In the context of the discussion above, it seems natural to try and articulate a relative version of Ozsváth’s question. In particular, are there Khovanov-type tangle invariants that are invariants of the knot K independent on the chosen strong inversion h? We will see that 946 dashes any hope of this (see Counterexample 3.5) and, furthermore, provides hyperbolic counterexamples to Ozsváth’s original question (see Remark 3.2). Interestingly, 946 also makes a star appearance in the recent work of Boyle and Issa [BI21]. In particular, this knot is slice, but it has non-zero equivariant four-genus; see [BI21, Figure 14] and the surrounding discussion. 4 ARTEM KOTELSKIY, LIAM WATSON, AND CLAUDIUS ZIBROWIUS

Kh(f T1)

h Kh(f T2)

BN(g T1)

T1 T2

Figure 5. Illustrated on the left is the trefoil with strong inversion, highlighting the decomposition of the unknot into tangles T1 and T2, where T2 is the preferred representative of the associated quotient tangle to the strong inversion. The corresponding intersection picture is shown on the right, where all of the relevant information has been projected to the front face of the sphere: The Lagrangian Floer homology HF(BN(f T1), Kh(f T2)) = F recovers the one-dimensional reduced Khovanov homology of the unknot.

2. Review of the tangle invariants Khg and BNg Let T be a four-ended tangle in the three-ball B3. In earlier work we interpreted Bar-Natan’s tangle invariant [[T ]]/l in terms of isotopy classes of immersed curves in the four-punctured sphere 3 ∂B r ∂T [KWZ19, BN05]. After choosing a distinguished tangle end ∗ of T , the construction outputs two curve-valued invariants Kh(g T ) and BN(g T ):

[BN05] [KWZ19] 2 3 T 7−−−−→ [[T ]]/l 7−−−−−→ Kh(g T ), BN(g T ) # S4, = ∂B r ∂T ∗ Strictly speaking, these immersed curves come equipped with local systems. However, in this paper, we will suppress this subtlety, since over F the local systems are trivial in all known examples. The curve invariants enjoy the following gluing property [KWZ19, Theorem 1.9]: Given a decomposition of a knot K = T1 ∪ T2 into four-ended tangles, the reduced Khovanov homology of the knot is recovered via Lagrangian Floer homology: ∼ (1) Kh(g K) = HF(m(BN(g T1)), Kh(g T2)) where m is the map identifying the two four-punctured spheres. An example illustrating this gluing formula is given in Figure5. Let us recall some basic facts about Kh(g T ) and BN(g T ) from [KWZ19, Section 6]: Both invariants may consist of multiple components. A component is either an immersion of a circle or an immersion of an interval. In the ﬁrst case we call a component compact; in the second, non-compact. The invariant BN(g T ) contains exactly one non-compact component (unless the tangle T contains some closed component, which is not the case in the present context), whereas Kh(g T ) consists only of compact components. 2 These curves become easier to manage when considered in a certain covering space of S4, , namely the planar cover that factors through the toroidal two-fold cover: ∗ 2 2 2 2 (R r Z ) → (T r 4pt) → S4, ∗ 2 2 2 Given an immersed curve c # S4, , denote by c˜ # R r Z an arbitrary lift of c. A curve c is ∗p 1 p called linear if there exists some q ∈ QP such that for every open neighbourhood U of q in 1 QP , there is a curve cU homotopic to c with the property that the slopes c˜U0 (t) of the lift c˜U p 1 are contained in U. If there exists such a q ∈ QP , it is unique, and we call it the slope of c. KHOVANOV HOMOLOGY AND STRONG INVERSIONS 5

˜sn(0) 2n punctures z }| {

n punctures ˜rn(0) z }| {

Figure 6. Curves {˜sn(0)}n>1 and {˜rn(0)}n>1

We call a linear curve c rational if there exists some neighbourhood of the special puncture ∗ 2 of S4, that cU can be chosen to avoid for all U simultaneously. Otherwise, we call c special. ∗ Figure6 shows some examples of lifts of linear curves of slope 0. The curves ˜sn(0) define >0 a family of special curves sn(0), n ∈ Z ; the curves ˜rn(0) define a family of rational curves >0 p 1 p p p rn(0), n ∈ Z . For every q ∈ QP , we define the linear curves sn( q ) and rn( q ) of slope q as the images of sn(0) and rn(0), respectively, under the action of q r p s 2 ∼ considered as an element the mapping class group Mod(S4, ) = PSL(2, Z), where qs − pr = 1. ∗ Example 2.1. The curve Kh(g T2) from Figure5 consists of the special component s2(∞) and the rational component r1(4). Furthermore, BN(g T1) = {a(∞)} is the red vertical arc, while Kh(g T1) = {r1(∞)} consists of the figure-eight curve which lies in a small neighbourhood of this arc. More generally, justifying the terminology, naturality of the invariants under the mapping p p class group action implies that BN(g Qp/q) = {a( q )} and Kh(g Qp/q) = {r1( q )}—so, rational tangles have rational invariants.

Computations suggest that the geography of components of Kh(g T ) is highly contrained: (1) Every component of Kh(g T ) is linear. p p 1 (2) Every special component of Kh(g T ) is equal to sn( q ) for some n > 1 and q ∈ QP . p p 1 (3) Every rational component of Kh(g T ) is equal to rn( q ) for some q ∈ QP . We expect to establish these restrictions in the near future [KWZ21]; in the context of Heegaard Floer homology, analogous properties are known for the tangle invariant HFT(T ) [Zib19]. Note, however, that components of the invariant BN(g T ) are known to be much more complicated even when restricting to only compact components. We conclude this section with a computation of the Lagrangian Floer homology of a simple arc with the special curves sn(∞), which will play a central role in the next section.

Deﬁnition 2.2. Deﬁne Vκ as the four-dimensional bigraded vector space supported in δ-grading 0 and quantum gradings −5, −1, +1, and +5.

Lemma 2.3. Let a(0)= BN(g ) and s2(∞) as in Figure8. Then 1 1 ∼ 0 2 2 HF(a(0), s2(∞)) = q δ− h Vκ More generally, for any positive integer n, n M 1 1 ∼ 4(2i n 1) 2 2 +2(2i n 1) HF(a(0), s2n(∞)) = q − − δ− h − − Vκ i=1

In both cases, the isomorphism holds as absolutely bigraded vector spaces if s2n(∞) is symmetrically bigraded in the sense of Figure7.

Proof. The total dimension of HF(a(0), s2n(∞)) is 4n, since the minimal number of intersection points between the two curves is 4n. To compute the bigrading on the Lagrangian Floer homology, we need to recall some facts from [KWZ19, Section 7]. We will focus on the case n = 1, the computation for n > 1 is similar. 6 ARTEM KOTELSKIY, LIAM WATSON, AND CLAUDIUS ZIBROWIUS

5 S 4 D 2 S2 0 D 2 S 3 − 3 − 2 − 1 0 1 2 − − − D D 3 S 2 D 0 S2 2 D 4 S 5 − 2 − 1 0 1 2 3 − −

Figure 7. The symmetrically bigraded type D structure corresponding to the special curve s2n(∞) for n = 1. The type D structures for n > 1 look similar. The total number of generators in idempotent is equal to 8n. The absolute bigrading is ﬁxed by requiring that the minimal and maximal homological and quantum gradings are ±(2n + 1) and ±(4n + 1), respectively.

-2 -4 0 2 SD -1 -2 0 1

SD 0 -2 2 4 0 -1 1 2

5 − 1 3 1 − 1 1 q δ 2 h q δ 2 h

3 2 -3 5 -2 3 0 0 -5 -3

−5 − 1 −2 −1 − 1 0 q δ 2 h q δ 2 h

Figure 8. Computing the gradings associated with the four-dimensional vector space HF(a(0), s2(∞)), according to Lemma 2.3

The invariants BN(g T ) and Kh(g T ) are topological interpretations of algebraic invariants Д(T ) and Д1(T ), respectively, which are type D structures over the algebra B. For example, the curve s2(∞), which is the blue curve on the left of Figure8, corresponds to the type D structure in 0 Figure7; the arc a(0) corresponds to the one-generator complex [ 0]. To translate between 2 these two viewpoints, we need to fix a parametrization of the four-punctured sphere S4, , which is indicated by the two grey dashed arcs in Figure8. Briefly, intersection points of∗ these arcs with a given curve give rise to the generators of the corresponding type D structure and paths between intersection points determine the differential; for more details, see [KWZ19, Example 1.6]. The generators of the type D structures carry a bigrading, that is a quantum grading q and a homological grading h. We indicate these gradings on generators • and the q corresponding intersection points by super- and subscripts like so: •h. Then the bigrading of an intersection point • generating HF(a(0), s2(∞)) is computed by considering a path on 0 a(0) ∪ s2(∞) which starts at x = 0, turns right at the intersection point • and ends at the nearest intersection point y of s2(∞) with the parametrization. These paths are illustrated on 2 the right of Figure8. Each of these paths in S4, is homotopic relative to the parametrizing arcs to some path γ on the boundary of the special∗ puncture; these homotopies are indicated on the right of Figure8 by the shaded disks. We set q(γ) =0 , −1, −2, depending on whether the path γ is constant, equal to S, or equal to D, respectively. (More generally, these paths correspond to some algebra elements in B, whose quantum gradings define q(γ).) Then h(•) = h(y) − h(x) and q(•) = q(y) − q(x) + q(γ) 1 Finally, we compute the δ-grading from the identity δ + h = 2 q. KHOVANOV HOMOLOGY AND STRONG INVERSIONS 7

? ? ? 31 −4 1 84 −4 9 91 −16 4 918 −8 20 937 0 22 ? ? ? ? 41 0 2 85 8 10 92 −4 7 919 0 20 938 −8 28 ? ? ? 51 −8 2 86 −4 11 93 12 9 920 −8 20 939 4 27 ? ? ? ? ? 52 −4 3 87 4 11 94 −8 10 921 4 21 940 −4 37 ? ? ? 61 0 4 88 0 12 95 4 11 922 4 21 941 0 24 ? ? ? ? 62 −4 5 89 0 12 96 −12 13 923 −8 22 942 0 4 ? ? 63 0 6 810 4 13 97 −8 14 924 0 22 943 8 6 ? ? 71 −12 3 811 −4 13 98 −4 15 925 −4 23 944 0 8 ? ? ? ? 72 −4 5 812 0 14 99 −12 15 926 4 23 945 −4 11 ? ? ? ? 1 73 8 6 813 0 14 910 8 16 927 0 24 946 0 4 ? ? ? ? 2 74 4 7 814 4 15 911 8 16 928 −4 25 946 −4 7 ? ? ? 75 −8 8 815 −8 16 912 −4 17 929 −4 25 947 4 13 ? ? ? 76 −4 9 816 −4 17 913 8 18 930 0 26 948 −4 13 ? ? ? ? 77 0 10 818 0 22 914 0 18 931 4 27 949 8 12 ? ? 81 0 6 819 8 2 915 4 19 934 0 34 ? ? ? 82 −8 8 820 0 4 916 12 19 935 −4 13 ? ? ? 83 0 8 821 −4 7 917 −4 19 936 8 18

Table 1. The ungraded reduced Khovanov homology of tangles TK,h associated with strong inversions h on knots K with up to 9 crossings. The ﬁrst columns specify the strong inversions, the second columns specify the slope k of the single rational component r1(k) in Kh(f TK,h), and the third columns give the number n of special components, all of which are equal to s2(∞).

3. The Khovanov homology of strong inversions A strongly invertible knot (K, h) gives rise to a four-ended associated quotient tangle T = 3 (B , τ). (When we have need for it, we will use the notation TK,h to remember the dependence on the strongly invertible knot.) We use the same preferred framing as in Section1; in particular, τ(∞) is the unknot. This imposes strong restrictions on the curve invariant Kh(g T ), namely, it implies that Kh(g τ(∞)) = HF(BN(g ), Kh(g T )) = HF(a(∞), Kh(g T )) = F In other words, Kh(g T ) is homotopic to a multicurve which intersects the arc a(∞) only once, as is the case for the example Kh(g T2) from Figure5. In the light of the geography restrictions, this means that Kh(g T ) should be a union of special components of slope ∞ together with a single rational component of an integer slope. In fact, with the help of a computer [Zib21], we show the following: Theorem 3.1. Let h be a strong inversion on a knot K ⊂ S3 with at most 9 crossings. Then

Kh(g TK,h) = r1(k) ∪ s2(∞) ∪ · · · ∪ s2(∞) | {z } n >0 for some integers k ∈ 4Z and n ∈ Z . Table1 shows the ungraded invariants for all pairs (K, h) from Theorem 3.1; the bigraded invariants are listed in Section4. Note that, in order to match Sakuma’s table [Sak86], the ﬁrst row in Table1 describes the left-hand trefoil and its rational component has slope −4, while Figure5 depicts the right-hand trefoil and its rational component has slope 4. Among tabulated knots through 9 crossings, there are 57 knots that admit two distinct strong inversions. For all but one of these knots with two distinct strong inversions, the ungraded invariants Kh(g T ) agree; these pairs of strong inversions are indicated in Table1 by the superscript ?. 8 ARTEM KOTELSKIY, LIAM WATSON, AND CLAUDIUS ZIBROWIUS

1 2 Remark 3.2. The knot 946 is the only knot in this collection whose strong inversions 946 and 946 can be distinguished by their ungraded invariants Kh(g T ), as shown in the two highlighted rows in Table1. Here, we follow the same numbering convention for strong inversions as in [Sak86]. 1 Note that both the slope k and the number n of special components s2(∞) distinguish 946 and 2 946. With this example in hand, one can check that dim Kh(g τ1(n)) 6= dim Kh(g τ2(n)) by applying 3 ∼ 3 the pairing formula (1), despite the fact that Σ(S , τ1(n)) = Σ(S , τ2(n)) by construction. p Calculations for dim Kh(g τi( q )) with i = 1, 2 can be obtained as well with a little more patience.

As stated in Section2, we expect Kh(g T ) to contain only linear curves; it would be interesting to give a geometric interpretation for the slopes of these curves. For quotient tangles TK,h of strong inversions (K, h), the existence of the unknot closure implies that the slope of any special curve is fixed and the slope of the rational component is an integer. As we can see from the second column in Table1, this integer may be non-zero; in other words, the slope of the rational component of Kh(TK,h) need not agree with the slope given by the rational longitude of the knot K. Our computations raise the following questions: Questions 3.3. Is there a geometric/topological meaning of the slope of the rational component of Kh(g TK,h)? Is the slope always divisible by 4? We now focus on the special components: Definition 3.4. Given a knot with a strong inversion (K, h), let s(K, h) be the set of special components of Kh(g TK,h). Let a(0) = BN(g ) as in Figure8. Then define κ(K, h) := HF(a(0), s(K, h)) One can show that κ(K, h) agrees as a relatively bigraded invariant with an invariant of the same name that was defined by the second author as a certain finite-dimensional quotient of the inverse limit of lim Kh(g τ(n)) associated with the maps Kh(g τ(n + 1)) → Kh(g τ(n)) induced by resolving a crossing←− [Wat17].

Counterexample 3.5. Returning to the pair of strong inversions on the knot 946, we calculate that dim κ(K, h1) = 16 < 28 = dim κ(K, h2) contrary to [Wat17, Conjecture 29]. This conjecture was originally posed in the hope that a relative variant of Ozsváth’s question might help to explain the examples described in [Wat10], which depend heavily on the Seifert structure. Comparing with Remark 3.2, it is worth noting a simple relationship between κ(K, h) and Kh(g τ(n)) when restricting to the preferred framing: According to Deﬁnition 3.4,

dim Kh(g τ(n)) = dim κ(K, h) + dim HF(a(n), r1(k)) where k is as in Theorem 3.1. Note that the second summand is the dimension of the reduced Khovanov homology of the (2, n − k)-torus link. On the other hand, certain structural properties appear to persist: Conjecture 3.6 ([Wat17, Conjecture 28]). For any strong inversion (K, h), the vector space κ(K, h) is a direct sum of copies of Vκ. In particular, dim κ(K, h) is divisible by 4.

By Lemma 2.3, the vector space Vκ is precisely the Lagrangian Floer homology of the arc a(0) and the special curve s2(∞). So a strong inversion (K, h) satisﬁes Conjecture 3.6 if all special components of Kh(g TK,h) are equal to s2(∞) up to a grading shift. In particular, the calculations summarized in Table1 verify Conjecture 3.6 for all strong inversions on tabulated knots through 9 crossings. We remark that Conjecture 3.6 implies that the slope of the rational component of Kh(g TK,h) should be divisible by 4 (Questions 3.3, second part). This can be seen as follows. First note KHOVANOV HOMOLOGY AND STRONG INVERSIONS 9 that 3 3 |H1(Σ(S , τ(0)))| = |H1(S0 (K))| = ∞ hence det(τ(0)) = 0. The determinant of a knot can also be computed by evaluating the Jones polynomial at −1, which agrees with the Euler characteristic of Khovanov homology with respect to the δ-grading:

0 = det(τ(0)) = |V (−1)| = |χδKh(g τ(0))| = |χδ HF(a(0), Kh(g T ))| This means that the δ-graded intersections of a(0) with special components cancel out intersections with the rational component. Conjecture 3.6 implies that intersections with special components come in groups of 4 concentrated in a single δ-grading, and so the number of intersections with the rational component—also concentrated in a single δ-grading—should be divisible by 4. Due to Lemma 2.3, the following is a reﬁnement of Conjecture 3.6:

Conjecture 3.7. Given a strongly invertible knot (K, h), any special component of Kh(g TK,h) is equal to s2n(∞) for some n > 0. In particular, all special components have even length.

The special components in all examples that we have tabulated are equal to s2(∞), but there exist strongly invertible knots (K, h) for which Kh(g TK,h) contains special components s2n(∞) with n 6= 1. As we can see from Table1, the ungraded invariant κ(K, h) cannot tell all strong inversions apart. However, the absolutely bigraded invariant κ(K, h) can distinguish any pair of strong inversion for knots up to 9 crossings. Question 3.8 ([Wat17, Question 19]). Is there any hyperbolic knot K ⊂ S3 with two distinct strong inversions that cannot be distinguished by the absolutely bigraded invariant κ? The restriction to hyperbolic is important here, and was omitted in error in the statement of [Wat17, Question 19]. Note that the question is not interesting for torus knots, as these admit unique strong inversions. However, it is possible to generate satellite knots with strong inversions whose associated quotient tangles differ by mutation on a sub-tangle. Owing to the insensitivity of Khovanov homology to mutation, such symmetries cannot be separated. Remark 3.9. We have focussed entirely on four-ended tangles admitting an unknot closure. We expect Conjecture 3.7 to hold more generally: For any four-ended tangle T , any special p p 1 component of Kh(g T ) should be equal to s2n( q ) for some q ∈ QP and n > 0. Remark 3.10. In this paper, we have worked exclusively over the field F of two elements; all of the above questions should be read with this coefficient system in place. We conclude with a comment about other fields of coefficients. Unlike knot Floer homology, Khovanov homology is known to behave very differently over fields of different characteristic. Many of the conjectures above fail if we work over a field different from F. For example, consider the second strong 2 inversion 7 on the knot 74 in Sakuma’s table [Sak86]. Its invariant Kh(g T 2 ; /3) consists of a 4 74 Z single rational component r1(6), two special components s2(∞), but also five special components s1(∞). So the corresponding version of Conjecture 3.6 over Z/3 is false. Also note that the slope of the rational component is no longer divisible by 4.

4. A table of invariants In the following, we list the absolutely bigraded invariants κ for all strong inversions whose underlying knots have at most 9 crossings. We do this as follows: We subdivide the drawing plane into a grid and associate with each point on the plane a bigrading, namely the homological grading is equal to the x-coordinate and the δ-grading is equal to the y-coordinate. Moving a point right and upwards increases both gradings. For each pair (K, h), we can write

M ai bi κ(K, h) = h δ Vκ i 10 ARTEM KOTELSKIY, LIAM WATSON, AND CLAUDIUS ZIBROWIUS since all special components of the tangle invariants have length 2. Then, for each i, we place a grey box centred at the point (ai, bi). If multiple boxes line up at the same bigrading, we add a label which indicates the number of such boxes. Finally, the absolute bigrading is speciﬁed by the two numbers in the bottom left corner, which indicate the absolute coordinate of the bottom-and-left-most intersection point of the grid: The top-left number is the δ-grading, the bottom-right number is equal to the homological grading. For example, 1 2.5 5.5 6.5 4.5 κ(41) = h δ Vκ ⊕ h δ Vκ The superscripts ? = 1, 2 indicate the strong inversions, using the same numbering convention as in [Sak86]. We follow the grading conventions in [KWZ19]; when comparing these computations to [Wat17], note that our δ-grading has the opposite sign, and our quantum grading is twice as large.

Remark 4.1. There is a mistake in the diagram for 943 in [Sak86], where the diagram given is an alternating knot of determinant 43 and Rasmussen’s s-invariant ±2. The only such knots with 6 9 crossings sharing these properties are 921 and 922, according to knotinfo [LM21]. By comparing the κ-invariant of the quotients, we see that Sakuma’s diagram shows 922; The actual knot 943 can be obtained from this diagram by changing the two crossings in the centre of the diagram, close to the two points on the knot that stay ﬁxed under the strong inversion. Its determinant is 13 and s = ±4. The only other knot of no more than 9 crossings with these properties is 73, but its strong inversions have diﬀerent κ than the ones we compute for 943.

31 4 2 1 72 1 41 12 4 2 2 2 2 72 41 3 −6 0 −7 71 51 4 3 −2 −7 0 51 2 2 7 2 8 3 −5 2 −3 2 5 1 2 −1 7 −7 4 −3 2 1 61 2 8 74 2 −11 62 −14 1 −2 0 71 2 5 2 8 1 −2 6 2 2 4 −2 2 75 3 2 7 62 2 2 2 71 2 2 6 2 1 2 4 63 −6 −6 −11 2 2 76 2 62 2 3 −7 3 −15 2 1 2 2 71 8 7 2 7 2 −6 −15 KHOVANOV HOMOLOGY AND STRONG INVERSIONS 11

2 2 2 1 2 3 77 2 89 −1 2 −7 2 82 3 2 9 2 1 −2 81 −3 2 12 8 2 4 2 10 −6 −8 82 1 −6 −9 1 2 2 811 2 2 1 2 8 2 2 4 2 8 −10 2 2 82 3 82 11 3 −2 −7 −10 −15

1 1 2 2 83 812 3 2 2 0 4 −14 −6 2 82 2 2 3 3 812 2 2 −4 −8 0 −8

1 2 2 84 81 2 2 13 2 0 −6 2 −19 2 2 8 2 2 4 4 2 3 2 −2 813 3 −6 81 3 2 −15 5 −6 2 −1 1 4 2 814 3 2 −11 85 −15 3 2 2 2 82 3 2 14 2 1 3 86 −11 −2 2 −19 −7 1 3 4 2 2 815 3 2 86 2 2 −2 8 −15 −2 2 1 2 3 2 87 815 2 3 −2 2 0 −2 −18 2 87 2 2 −1 2 3 2 −3 816 2 2 2 1 3 2 −7 88 −18 −1 2 −6 3 2 81 3 4 2 18 3 2 82 2 8 2 −6 −18 −12 −19 12 ARTEM KOTELSKIY, LIAM WATSON, AND CLAUDIUS ZIBROWIUS

2 3 91 2 3 2 2 4 3 8 2 2 818 8 2 3 2 1 2 2 2 2 819 −5 98 2 3 2 −13 820 −23 −1 −4 91 4 3 1 9 2 8 2 2 7 21 −2 −1 −11 2 92 2 3 2 9 2 2 821 2 8 −2 −6 −7 1 3 91 8 910 3 −2 −11 2 −11 2 1 2 3 2 92 910 2 −11 −11 16 2 1 2 2 2 911 2 2 92 −1 −1 2 −9 −3 1 9 2 2 2 3 9 2 3 2 −15 11 2 −13 −11 −9 92 3 −5 2 2 2 2 0 91 2 2 12 2 2 2 4 1 −14 94 4 2 2 3 −18 912 2 3 2 2 1 2 0 9 2 4 3 −4 1 3 2 913 3 2 −7 2 1 −3 95

−5 2 3 2 9 3 13 3 2 −16 95 −19 −11 −14 2 91 2 2 1 14 2 2 9 2 2 2 2 6 2 2 12 0 2 2 2 96 3 3 92 3 2 3 14 2 3 2 −10 −6 −14 1 2 97 2 1 3 2 2 915 4 2 2 12 −1 2 −2 −6

92 3 3 7 2 3 −7 KHOVANOV HOMOLOGY AND STRONG INVERSIONS 13

2 2 2 2 3 2 9 2 3 923 3 3 15 2 2 2 3 −13 2 −12 −10 3 3 2 1 4 3 2 5 2 916 924 2 4 −11 2 −5 −7 −12 2 2 92 2 2 3 5 16 3 3 925 3 4 −2 −3 2 −8

1 3 2 1 3 3 917 2 3 2 926 2 5 3 2 −2 −2 2 −10 −3 2 2 3 3 2 3 2 917 2 2 9 2 5 2 26 3 −2 −10 −11 −19 2 2 1 3 918 3 3 1 3 2 927 2 5 2 2 2 4 8 −10 7 2 92 4 4 2 18 4 2 92 3 5 2 −2 27 3 4 −15 −3 −4 91 3 2 19 2 4 2 1 3 4 −2 2 928 4 4 −7 3 2 −7 −19 2 2 2 3 919 2 3 2 4 2 2 2 92 4 3 28 3 6 2 −7 −18 1 2 9 3 4 2 2 20 3 3 8 4 4 −2 929 2 3 2 2 −7 3 −18 92 2 3 20 2 3 5 3 4 2 930 4 5 3 2 2 −6 −19 1 2 3 921 3 3 2 2 1 4 3 1 931 2 6 3 2 3 −11 2 4 2 −19 921 2 4 2 −7 2 3 2 −7 92 3 4 31 2 4 2 2 −2 2 4 2 −6 922 2 4 −6 −12 3 2 3 5 2 934 4 5 3 3 2 1 2 4 923 −6 2 4 −22 11 2 14 ARTEM KOTELSKIY, LIAM WATSON, AND CLAUDIUS ZIBROWIUS

2 1 2 2 935 941 2 3 2 2 3 11 6 2 2

2 942 9 −2 35 2 3 2 3 −7 943 3 2 −5 936 2 4 2 −8 −7 2 −1 944 2 2 −2 2 −4 1 2 3 937 2 3 2 2 2 945 2 2 −6 −22 −1 −15 2 2 3 3 91 937 2 4 46 −2 −1 −6 2 −15 92 3 46 938 4 5 8 2 4 2 2 6 −2 2 947 2 3 3 2 −6 2 4 2 −11 939 3 3 2 1 2 2 −11 948 3 −22 3 −6 4 3 1 3 6 3 2 940 4 4 92 2 2 3 2 48 2 −7 1 −22 1

2 2 3 2 6 3 949 2 2 5 5 940 −11 2 4 2 −11 1 −2

Acknowledgements. The computer program [Zib21] used in this work represents a major overhaul to an earlier piece of software, developed as part of a UBC undergraduate research project with Gurkeerat Chhina.

References [BI21] Keegan Boyle and Ahmad Issa. Equivariant 4-genera of strongly invertible and periodic knots, 2021. ArXiv preprint 2101.05413. [BN05] Dror Bar-Natan. Khovanov’s homology for tangles and cobordisms. Geom. Topol., 9:1443–1499, 2005. ArXiv: math/0410495v2. [GL89] Cameron McA. Gordon and John Luecke. Knots are determined by their complements. J. Amer. Math. Soc., 2(2):371–415, 1989. [Har80] Richard Hartley. Knots and involutions. Math. Z., 171(2):175–185, 1980. [KWZ19] Artem Kotelskiy, Liam Watson, and Claudius Zibrowius. Immersed curves in Khovanov homology, 2019. ArXiv preprint 1910.14584. [KWZ21] Artem Kotelskiy, Liam Watson, and Claudius Zibrowius. Thin links and Conway spheres, 2021. In preparation. [LM21] Charles Livingston and Allison H. Moore. Knotinfo: Table of knot invariants. https://knotinfo. math.indiana.edu, February 2021. KHOVANOV HOMOLOGY AND STRONG INVERSIONS 15

[Sak86] Makoto Sakuma. On strongly invertible knots. In Algebraic and topological theories (Kinosaki, 1984), pages 176–196. Kinokuniya, Tokyo, 1986. [Wat10] Liam Watson. A remark on Khovanov homology and two-fold branched covers. Paciﬁc J. Math., 245(2):373–380, 2010. [Wat17] Liam Watson. Khovanov homology and the symmetry group of a knot. Adv. Math., 313:915–946, 2017. [Zib19] Claudius Zibrowius. On symmetries of peculiar modules; or, δ-graded link Floer homology is mutation invariant, 2019. ArXiv preprint 1909.04267. [Zib21] Claudius Zibrowius. kht++, a program for computing Khovanov invariants for links and tangles. URL: https://cbz20.raspberryip.com/code/khtpp/docs/, April 2021.

Department of Mathematics, Indiana University Email address: [email protected]

Department of Mathematics, University of British Columbia Email address: [email protected]

Faculty of Mathematics, University of Regensburg Email address: [email protected]