Introduction
Let K ⊂ S 3 K \subset S^3 K ⊂ S 3 be a knot and let π K = π 1 ( S 3 ∖ K ) \pi_K = \pi_1(S^3 \setminus K) π K = π 1 ( S 3 ∖ K ) . The meridional rank mr ( K ) \operatorname{mr}(K) mr ( K ) is the minimum number of meridians needed to generate π K \pi_K π K . A bridge presentation gives a meridional generating set, so mr ( K ) ≤ b ( K ) \operatorname{mr}(K) \le b(K) mr ( K ) ≤ b ( K ) , where b ( K ) b(K) b ( K ) denotes the bridge number. The meridional rank conjecture of Cappell–Shaneson asks whether equality holds for every knot. It appears in Kirby’s problem list [27, Problem 1.11] and, in its link formulation. See also [3, Problem 1.18(a)].
The conjecture is known for several classes of knots and links. Boileau–Zieschang proved it for Montesinos links by using their twofold branched covers [10], and Rost–Zieschang proved it for torus links [31 ] . The characterization of knots whose groups can be generated by two meridians is due to Boileau–Zimmermann [11, Corollary 3.3], together with the subsequent orbifold geometrization theorem [9].
Diagrammatic descriptions of bridge number clarify a difficulty in the conjecture. Blair–Kjuchukova–Velazquez–Villanueva [6] proved that the bridge number equals the Wirtinger number, the minimum number of seed strands needed for successive Wirtinger generation over all diagrams. Blair–Kjuchukova–Pfaff allow relations witnessed by certain embedded two-spheres; their plain sphere number also equals the bridge number [5]. To deduce the conjecture from these descriptions, one would still need to pass from an arbitrary meridional generating set to such a system without increasing its size.
For lower bounds, the conjugacy class of a meridian is essential. Ordinary group rank gr ( K ) \operatorname{gr}(K) gr ( K ) allows arbitrary generators and can be much smaller: a nontrivial torus-knot group has rank two, whereas its meridional rank can be arbitrarily large [31 ] . A quotient retaining the meridian class can give a stronger bound. Baader–Blair–Kjuchukova use meridian-to-reflection Coxeter quotients to prove the conjecture for twisted links and arborescent links associated to bipartite trees with nonzero even weights [1]. Baader–Kjuchukova [2, Section 2] use symmetric-group quotients to prove the meridional rank conjecture for a family of twisted Whitehead doubles of connected sums of T ( 2 , 5 ) T(2,5) T ( 2 , 5 ) , with arbitrarily large bridge number. Recent work of Blair–Kjuchukova–Pfaff extends this approach to fishnet families with any prescribed bridge number at least three and arbitrarily large bridge distance relative to that number [4]. The question of what finite quotients can detect is also recorded in [3, Problem 1.18(b)].
Knot contact homology gives a different lower bound. Cornwell [15, Theorems 1.2 and 1.5] relates its augmentations to representations of the knot group and proves
ar ( K ) ≤ mr ( K ) ≤ b ( K ) , (1) \operatorname{ar}(K) \le\operatorname{mr}(K) \le b(K), \tag*{(1)} ar ( K ) ≤ mr ( K ) ≤ b ( K ) , (1)
where ar ( K ) \operatorname{ar}(K) ar ( K ) denotes augmentation rank. In the corresponding microlocal rank one representations, a meridian acts by a diagonalizable rank-one perturbation of the identity. For an irreducible microlocal rank one representation ρ \rho ρ , the images of ρ ( m i ) − I \rho(m_i)-I ρ ( m i ) − I for any meridional generating set { m i } \{m_i\} { m i } span the representation space. Its dimension, therefore, bounds the meridional rank from below. We will vary these representations and their peripheral eigenvalues along an algebraic curve to choose data for successive Whitehead doubles. We recall the definitions in Section 2.1.
We study these ranks under untwisted Whitehead doubling. Two satellite results are particularly relevant. For satellites with braid patterns, Cornwell–Hemminger prove a maximal augmentation-rank multiplication theorem and apply it to the iterated torus knots satisfying their hypotheses [16, Theorem 1.2 and Corollary 1.3]. The Whitehead pattern has winding number zero, so that theorem does not apply directly. Using the action on the tree associated to a satellite decomposition, Boileau–Dutra–Jang–Weidmann control the peripheral intersections of meridional subgroups. They prove the conjecture for the class generated from meridionally tame knots by connected sums and satellites with braid patterns [7, Theorem 1]. This includes every knot whose exterior is a graph manifold [7, Corollary 1]. However, Whitehead doubles of nontrivial knots are not meridionally tame [7, Introduction and Section 6], so this theorem does not directly include Whitehead iteration.
Dutra’s preprint [18, Theorem 1.1] states the meridional rank conjecture for (once) Whitehead doubles of prime algebraically tame knots. Even with this conclusion, iteration would require a new argument: the first Whitehead double is no longer algebraically tame [18, Example 2.8].1 Our contribution is a sheaf-level doubling construction whose iteration is controlled on peripheral curves realized by the companion. By keeping track of both peripheral eigenvalues, we can choose representations for which every required doubling step is defined.
Throughout, Wh ( K ) \operatorname{Wh}(K) Wh ( K ) denotes the untwisted Whitehead double with the pattern and peripheral conventions fixed in § 2.2 . We set Wh 0 ( K ) = K \operatorname{Wh}^{0}(K)=K Wh 0 ( K ) = K and Wh N + 1 ( K ) = Wh ( Wh N ( K ) ) \operatorname{Wh}^{N+1}(K)=\operatorname{Wh}(\operatorname{Wh}^{N}(K)) Wh N + 1 ( K ) = Wh ( Wh N ( K )) . For a nontrivial companion,
b ( Wh ( K ) ) = 2 b ( K ) ; (2) b(\operatorname{Wh}(K))=2b(K); \tag*{(2)} b ( Wh ( K )) = 2 b ( K ) ; (2)
see [18, Section 4]. Thus, an augmentation rank lower bound that doubles at every step proves the meridional rank conjecture for the iterates, provided the initial augmentation rank equals b ( K ) b(K) b ( K ) .
Main results
Our first main result applies to every nontrivial two-bridge knot.
Theorem 1.1. Let K K K be a nontrivial two-bridge knot. For every integer N ≥ 0 N \ge0 N ≥ 0 ,
ar ( Wh N ( K ) ) = mr ( Wh N ( K ) ) = b ( Wh N ( K ) ) = 2 N + 1 . \operatorname{ar}(\operatorname{Wh}^{N}(K))=\operatorname{mr}(\operatorname{Wh}^{N}(K))=b(\operatorname{Wh}^{N}(K))=2^{N+1}. ar ( Wh N ( K )) = mr ( Wh N ( K )) = b ( Wh N ( K )) = 2 N + 1 .
For N = 1 N=1 N = 1 , the meridional-rank equality already follows from the lower bound for prime satellite knots in [8, Corollary 1.6]; see also [7, Introduction]. Theorem 1.1 also provides maximal-rank augmentations and propagates both equalities through every finite number of Whitehead doubles.
Given an irreducible microlocal rank one representation ρ \rho ρ on V V V , we extend ρ ⊕ ρ ∗ \rho\oplus\rho^{*} ρ ⊕ ρ ∗ from the companion group to the satellite group, where ( V ∗ , ρ ∗ ) (V^{*},\rho^{*}) ( V ∗ , ρ ∗ ) is the dual representation. The two distinguished peripheral eigenlines span a two-dimensional subspace of V ⊕ V ∗ V \oplus V^{*} V ⊕ V ∗ , on which we solve the Whitehead-pattern relations explicitly. The remaining subspace is fixed by the two pattern meridians. Under the nonvanishing conditions of Theorem 2.12 , this gives an irreducible 2 dim V 2\dim V 2 dim V -dimensional representation whose meridian still has microlocal rank 1, doubling the resulting lower bound for meridional rank.
To iterate, we compute the output longitude eigenvalue as well, so that the output of one step can be tested against the hypotheses of the next. This gives the algebraic correspondence of § 2.4 . We must then find peripheral values realized by a fixed companion that admits any prescribed finite number of steps.
The difficulty is that the peripheral values realized by a fixed knot might all lie in the exceptional locus. An open dense set in the full peripheral torus does not rule this out. We address this issue using a branch at infinity of the curve. The branch is called Whitehead-iterable branch if the meridian eigenvalue has a pole and the squared longitude eigenvalue either has a pole or has a zero of order greater than that pole order; see Definition 4.1. The key observation is that one branch of the doubling correspondence again satisfies this condition. By Theorem 4.2 , for each N ≥ 0 N \ge0 N ≥ 0 there is a nonempty Zariski-open subset of the curve on which N N N doubling steps can be performed. For two-bridge knots, the ideal-point argument in § 4.2 supplies a curve with an iterable branch whose peripheral values are realized by rank-two augmentations on a Zariski-dense subset. Combining these facts with (1.1) and (1.2) proves Theorem 1.1 .
The same construction provides a comparison with ordinary group rank. We call an irreducible curve in the peripheral torus an actual rank-d d d augmentation peripheral curve if a Zariski-dense subset of its points is realized by rank-d d d augmentations with ϵ ( μ ) ≠ 1 \epsilon(\mu) \ne1 ϵ ( μ ) = 1 . The coordinates are ( μ , λ 2 ) (\mu,\lambda^2) ( μ , λ 2 ) , where λ \lambda λ is the preferred longitude eigenvalue; see Definition 4.4 .
Theorem 1.2. For every knot K K K and every integer N ≥ 0 N \ge0 N ≥ 0 ,
gr ( Wh N ( K ) ) ≤ gr ( K ) + N . \operatorname{gr}(\operatorname{Wh}^{N}(K)) \le\operatorname{gr}(K) + N. gr ( Wh N ( K )) ≤ gr ( K ) + N . If K K K has an actual rank-d d d augmentation peripheral curve with an iterable branch, then
ar ( Wh N ( K ) ) ≥ 2 N d \operatorname{ar}(\operatorname{Wh}^{N}(K)) \ge2^{N}d ar ( Wh N ( K )) ≥ 2 N d for every N ≥ 0 N \ge0 N ≥ 0 . In particular,
gr ( Wh N ( K ) ) ≤ ar ( Wh N ( K ) ) whenever gr ( K ) + N ≤ 2 N d . \operatorname{gr}(\operatorname{Wh}^{N}(K)) \le\operatorname{ar}(\operatorname{Wh}^{N}(K)) \quad\text{whenever} \quad\operatorname{gr}(K) + N \le2^{N}d. gr ( Wh N ( K )) ≤ ar ( Wh N ( K )) whenever gr ( K ) + N ≤ 2 N d . For a nontrivial two-bridge knot, this gives
gr ( Wh N ( K ) ) ≤ N + 2 ≤ ar ( Wh N ( K ) ) = 2 N + 1 . \operatorname{gr}(\operatorname{Wh}^{N}(K)) \le N + 2 \le\operatorname{ar}(\operatorname{Wh}^{N}(K)) = 2^{N+1}. gr ( Wh N ( K )) ≤ N + 2 ≤ ar ( Wh N ( K )) = 2 N + 1 .
The proof is given in § 5 . A complementary lower bound for ordinary group rank was obtained by Gu–Wang–Zou [21, Proposition 1.3]. For every nontrivial knot K K K and N ≥ 1 N \ge1 N ≥ 1 , their estimate and Theorem 1.2 give
N + 1 ≤ gr ( Wh N ( K ) ) ≤ N + gr ( K ) . N + 1 \le\operatorname{gr}(\operatorname{Wh}^{N}(K)) \le N + \operatorname{gr}(K). N + 1 ≤ gr ( Wh N ( K )) ≤ N + gr ( K ) .
Thus ordinary group rank grows linearly, whereas the augmentation and meridional ranks in Theorem 1.1 grow exponentially. The lower bound in [21] uses the hyperbolic pieces of the JSJ decomposition and holds without an augmentation hypothesis. That work also determines ordinary group rank exactly for hyperbolic two-bridge companions; see § 5 .
The augmentation lower bound is not always sharp; this limitation is already present in Cornwell’s work [15]. Our last main result exhibits a further obstruction to iteration: even when the augmentation rank agrees with the bridge number for the companion, it may fail to double. The example nevertheless satisfies the meridional rank conjecture.
Theorem 1.3. Let S = T ( 2 , 3 ) # T ( 2 , 3 ) S = T(2,3)\mathbin{\#}T(2,3) S = T ( 2 , 3 ) # T ( 2 , 3 ) be the square knot. Then
ar ( S ) = mr ( S ) = b ( S ) = 3 , \operatorname{ar}(S) = \operatorname{mr}(S) = b(S) = 3, ar ( S ) = mr ( S ) = b ( S ) = 3 , whereas
ar ( Wh ( S ) ) ≤ 5 < 6 = mr ( Wh ( S ) ) = b ( Wh ( S ) ) . \operatorname{ar}(\operatorname{Wh}(S)) \le5 < 6 = \operatorname{mr}(\operatorname{Wh}(S)) = b(\operatorname{Wh}(S)). ar ( Wh ( S )) ≤ 5 < 6 = mr ( Wh ( S )) = b ( Wh ( S )) .
The augmentation obstruction uses the two trefoil factors of the companion group and their preferred longitudes. The meridional lower bound uses a different representation: an explicit epimorphism π Wh ( S ) ↠ S 14 \pi_{\operatorname{Wh}(S)} \twoheadrightarrow S_{14} π Wh ( S ) ↠ S 14 and its permutation module F 2 14 \mathbb{F}_{2}^{14} F 2 14 . The first cohomology of π W h ( S ) \pi_{\mathrm{Wh}}(S) π Wh ( S ) with coefficients in this module has dimension 14, while the meridian defect has rank 3. A cocycle estimate then rules out generation by five meridians; see § 6.2 for the representation and exact linear calculation. Here the finite quotient supplies the coefficient module for the cohomology of the knot group. This gives a strict gap between augmentation rank and meridional rank. The exact augmentation rank of W h ( S ) \mathrm{Wh}(S) Wh ( S ) and the ranks of further Whitehead doubles of S S S remain undetermined in this paper.
Conventions and organization
All knots are oriented. A meridian is positively oriented, and elements represented by meridians are understood up to conjugacy. We use a preferred zero-linking longitude. Sheaves and augmentations may be defined over any field k \mathbb{k} k , but in the algebraic-curve applications, we take k = C \mathbb{k} = \mathbb{C} k = C . The finite-dimensional permutation module in § 6.2 is explicitly over F 2 \mathbb{F}_2 F 2 .
We use π K : = π 1 ( S 3 ∖ K ) \pi_K := \pi_1(S^3 \setminus K) π K := π 1 ( S 3 ∖ K ) for the knot group throughout. The commutator convention is [ a , b ] = a b a − 1 b − 1 [a,b] = aba^{-1}b^{-1} [ a , b ] = ab a − 1 b − 1 . Products of permutations are composed from right to left, and permutation matrices act on column vectors by g e i = e g ( i ) ge_i = e_{g(i)} g e i = e g ( i ) . The embedding ϕ K \phi_K ϕ K of the Whitehead solid torus into a tubular neighborhood of K K K respects the preferred zero framing. A “two-bridge knot” in the statements below is nontrivial.
§ 2 recalls the representations and augmentations and constructs the doubling correspondence. § 2.5 gives the meridional upper bounds. § 3 recalls the algebraic curves, valuations, and character varieties used to find peripheral data to which the iteration applies. § 4.1 proves the iterable-branch criterion, and § 4.2 applies slope detection to prove Theorem 1.1 . § 5 compares ordinary group rank with augmentation rank. Finally, § 6 proves Theorem 1.3 for the first Whitehead double of the square knot.
Microlocal rank-1 representations and Whitehead doubling
The meridian and the preferred longitude of a knot K K K are denoted by m m m and ℓ \ell ℓ , respectively. We use the specialization of knot contact homology in which the extra variable U U U is set equal to one. For an embedded submanifold A ⊂ S 3 A \subset S^3 A ⊂ S 3 , we denote by Op ( A ) \operatorname{Op}(A) Op ( A ) a fixed sufficiently small open tubular neighborhood of A A A .
Microlocal rank-1 representations and augmentations
We first introduce the notation for the microlocal rank of a representation of the knot group.
Definition 2.1. For a representation ρ : π K → GL ( V ) \rho: \pi_K \to\operatorname{GL}(V) ρ : π K → GL ( V ) , define the microlocal rank to be
μ rk K ( ρ ) : = rank ( ρ ( m ) − I ) , \mu\operatorname{rk}_K(\rho) := \operatorname{rank}(\rho(m)-I), μ rk K ( ρ ) := rank ( ρ ( m ) − I ) , where I I I is the identity matrix of size dim V \dim V dim V . This number is independent of the chosen meridian, since all meridians are conjugate.
We call ρ \rho ρ a μ rk \mu\operatorname{rk} μ rk -one representation if μ rk K ( ρ ) = 1 \mu\operatorname{rk}_K(\rho)=1 μ rk K ( ρ ) = 1 .
A μ rk \mu\operatorname{rk} μ rk -one representation is diagonalizable if there is a basis in which
ρ ( m ) ∼ diag ( μ , 1 , … , 1 ) , μ ≠ 1. \rho(m) \sim\operatorname{diag}(\mu,1,\ldots,1), \qquad\mu\ne1. ρ ( m ) ∼ diag ( μ , 1 , … , 1 ) , μ = 1. A μ rk \mu\operatorname{rk} μ rk -one representation is unipotent if there is a basis in which
ρ ( m ) ∼ ( 1 1 0 1 ) ⊕ I dim V − 2 . \rho(m) \sim\begin{pmatrix} 1 & 1 \\ 0 & 1 \end{pmatrix} \oplus I_{\dim V-2}. ρ ( m ) ∼ ( 1 0 1 1 ) ⊕ I d i m V − 2 .
We recall the sheaf-theoretic reason for this terminology [20, Section 3.2]. Let j : S 3 ∖ K ↪ S 3 j:S^3\setminus K\hookrightarrow S^3 j : S 3 ∖ K ↪ S 3 be the open embedding. A representation ρ : π K → GL ( V ) \rho:\pi_K\to\operatorname{GL}(V) ρ : π K → GL ( V ) determines a local system E ρ \mathcal{E}_{\rho} E ρ on the knot complement. For the underived direct image, the stalk of j ∗ E ρ j_*\mathcal{E}_{\rho} j ∗ E ρ on K K K is the meridian-invariant subspace V ρ ( m ) V^{\rho(m)} V ρ ( m ) . The local gluing map is the inclusion
T : V ρ ( m ) ↪ V . T:V^{\rho(m)}\hookrightarrow V. T : V ρ ( m ) ↪ V .
Its mapping cone is the microlocal stalk, up to a degree shift, in the sense of Kashiwara–Schapira, along the knot conormal torus Λ K \Lambda_K Λ K inside the contact cosphere bundle, i.e. Λ K ⊂ S T ∗ S 3 \Lambda_K\subset ST^*S^3 Λ K ⊂ S T ∗ S 3 is a Legendrian surface. Consequently,
dim H ∙ ( Cone ( T ) ) = dim ( V / V ρ ( m ) ) = rank ( ρ ( m ) − I ) . \dim H^\bullet(\operatorname{Cone}(T))=\dim(V/V^{\rho(m)})=\operatorname{rank}(\rho(m)-I). dim H ∙ ( Cone ( T )) = dim ( V / V ρ ( m ) ) = rank ( ρ ( m ) − I ) .
Thus a μ \mu μ rk-one representation induces a microlocal rank one sheaf microsupported along the conormal. Conversely, up to the equivalence modulo local systems on S 3 S^3 S 3 in [20, Definition 3.8 and Proposition 3.11], a microlocal rank one abelian sheaf comes from one of the three forms: (1) an irreducible diagonalizable μ \mu μ rk-1 representation via the above construction; (2) an irreducible unipotent μ \mu μ rk-1 representation via the above construction; and (3) a local system of rank 1 that is supported on K K K [20]. The first type is generic and the main focus of this paper.
A diagonalizable μ \mu μ rk-one representation is also called a KCH representation by Cornwell [15], where KCH refers to knot contact homology . Next, we explain its background related to knot contact homology and augmentations.
The (framed) knot contact homology is a knot invariant defined via counting holomorphic disks for Λ K \Lambda_K Λ K , whose formula was first discovered by Ng [29 ] . We use the specialization U = 1 U=1 U = 1 of the Legendrian dg algebra A ( Λ K ) \mathcal{A}(\Lambda_K) A ( Λ K ) associated to Λ K \Lambda_K Λ K . The dg algebra A ( Λ K ) \mathcal{A}(\Lambda_K) A ( Λ K ) is semi-freely generated by Reeb chords over Z [ H 1 ( Λ K ) ] ≅ Z [ μ ± 1 , λ ± 1 ] \mathbb{Z}[H_1(\Lambda_K)]\cong\mathbb{Z}[\mu^{\pm1},\lambda^{\pm1}] Z [ H 1 ( Λ K )] ≅ Z [ μ ± 1 , λ ± 1 ] . An augmentation of the dg algebra is a morphism ϵ : A ( Λ K ) → k \epsilon:\mathcal{A}(\Lambda_K)\to\mathbb{k} ϵ : A ( Λ K ) → k which preserves grading, differential, and the unit. The case of the knot conormal has the special property that its generators are concentrated in non-negative degrees. Hence an augmentation descends to the degree 0 homology ϵ : H 0 ( A ( Λ K ) ) → k \epsilon:H_0(\mathcal{A}(\Lambda_K))\to\mathbb{k} ϵ : H 0 ( A ( Λ K )) → k .
The algebra H 0 ( A ( Λ K ) ) H_0(\mathcal{A}(\Lambda_K)) H 0 ( A ( Λ K )) admits a combinatorial model by the framed cord algebra.
Definition 2.2. The framed cord algebra Cord ( K ) \operatorname{Cord}(K) Cord ( K ) is the tensor algebra over Z [ λ ± 1 , μ ± 1 ] \mathbb{Z}[\lambda^{\pm1},\mu^{\pm1}] Z [ λ ± 1 , μ ± 1 ] generated by symbols [ γ ] [\gamma] [ γ ] , γ ∈ π K \gamma\in\pi_K γ ∈ π K , modulo the following relations:
[ e ] = 1 − μ [e]=1-\mu [ e ] = 1 − μ ,
[ m γ ] = [ γ m ] = μ [ γ ] [m\gamma]=[\gamma m]=\mu[\gamma] [ mγ ] = [ γ m ] = μ [ γ ] , [ ℓ γ ] = [ γ ℓ ] = λ [ γ ] [\ell\gamma]=[\gamma\ell]=\lambda[\gamma] [ ℓ γ ] = [ γ ℓ ] = λ [ γ ] ,
[ γ 1 γ 2 ] − [ γ 1 m γ 2 ] = [ γ 1 ] [ γ 2 ] [\gamma_1\gamma_2]-[\gamma_1m\gamma_2]=[\gamma_1][\gamma_2] [ γ 1 γ 2 ] − [ γ 1 m γ 2 ] = [ γ 1 ] [ γ 2 ] .
An augmentation is a unital algebra homomorphism
ϵ : Cord ( K ) ⟶ k . \epsilon:\operatorname{Cord}(K)\longrightarrow\mathbb{k}. ϵ : Cord ( K ) ⟶ k .
Note that ϵ \epsilon ϵ includes a specialization of λ \lambda λ and μ \mu μ to nonzero elements of k \mathbb{k} k .
Augmentations of the framed cord algebra and the above equivalence classes of microlocal rank one abelian sheaves modulo local systems are in bijective correspondence, with the normalization of [20, Theorem 4.17]. This is a theorem under the principle of “augmentation-sheaf correspondence” in the theory of Legendrian invariants. Considering the first and second cases of the sheaf classification, suppose ( ρ , V ) (\rho,V) ( ρ , V ) is a μ \mu μ rk-1 representation; then the associated augmentation ϵ ρ \epsilon_\rho ϵ ρ is given by
ϵ ρ ( [ γ ] ) = tr ( ρ ( γ ) − ρ ( m γ ) ) . \epsilon_\rho([\gamma])=\operatorname{tr}(\rho(\gamma)-\rho(m\gamma)). ϵ ρ ([ γ ]) = tr ( ρ ( γ ) − ρ ( mγ )) .
Conversely, suppose ϵ \epsilon ϵ is an augmentation of the framed cord algebra with ϵ ( μ ) ≠ 1 \epsilon(\mu) \ne1 ϵ ( μ ) = 1 . To reconstruct the μ \mu μ rk-1 representation ( ρ ϵ , V ϵ ) (\rho_\epsilon,V_\epsilon) ( ρ ϵ , V ϵ ) following [20, Proposition 4.11], we choose a set of based meridional generators { m 1 , … , m N } \{m_1,\ldots,m_N\} { m 1 , … , m N } with m 1 = m m_1=m m 1 = m . Since the meridians are conjugate, there exist elements g i ∈ π K g_i \in\pi_K g i ∈ π K such that m i = g i − 1 m 1 g i m_i=g_i^{-1}m_1g_i m i = g i − 1 m 1 g i . Define an N N N by N N N square matrix R ϵ R_\epsilon R ϵ by ( R ϵ ) i j : = ϵ ( [ g i g j − 1 ] ) (R_\epsilon)_{ij}:=\epsilon([g_i g_j^{-1}]) ( R ϵ ) ij := ϵ ([ g i g j − 1 ]) . Let R j R_j R j , j = 1 , … , N j=1,\ldots,N j = 1 , … , N , be the column vectors of R ϵ R_\epsilon R ϵ , and define the vector space
V ϵ : = Span k { R 1 , … , R N } . (3) V_\epsilon:=\operatorname{Span}_{\mathbb{k}}\{R_1,\ldots,R_N\}. \tag*{(3)} V ϵ := Span k { R 1 , … , R N } . (3)
and for any γ ∈ π K \gamma\in\pi_K γ ∈ π K , the ρ ϵ ( γ ) \rho_\epsilon(\gamma) ρ ϵ ( γ ) action is given by
ρ ϵ ( γ ) R j : = ( ϵ ( [ g i γ g j − 1 ] ) ) i = 1 N . \rho_\epsilon(\gamma)R_j:=\left(\epsilon([g_i\gamma g_j^{-1}])\right)_{i=1}^{N}. ρ ϵ ( γ ) R j := ( ϵ ([ g i γ g j − 1 ]) ) i = 1 N .
The resulting representation is indeed μ \mu μ rk-1. For example, by relation 3 in Definition 2.2,
ρ ϵ ( m t ) R j = R j − ϵ ( [ g t g j − 1 ] ) R t . \rho_\epsilon(m_t)R_j=R_j-\epsilon([g_tg_j^{-1}])R_t. ρ ϵ ( m t ) R j = R j − ϵ ([ g t g j − 1 ]) R t .
It follows at once that rank ( ρ ϵ ( m t ) − I ) ≤ 1 \operatorname{rank}(\rho_\epsilon(m_t)-I)\le1 rank ( ρ ϵ ( m t ) − I ) ≤ 1 . Since ( R ϵ ) t t = 1 − ϵ ( μ ) ≠ 0 (R_\epsilon)_{tt}=1-\epsilon(\mu)\ne0 ( R ϵ ) tt = 1 − ϵ ( μ ) = 0 , we have R t ≠ 0 R_t\ne0 R t = 0 and ρ ϵ ( m t ) R t = ϵ ( μ ) R t \rho_\epsilon(m_t)R_t=\epsilon(\mu)R_t ρ ϵ ( m t ) R t = ϵ ( μ ) R t . Thus the rank is exactly one, and the representation is diagonalizable μ \mu μ rk-one. This elementary identity is the point at which the microlocal condition enters the construction.
The dimension of V ϵ V_\epsilon V ϵ is independent of N N N and the choices of meridional generators. Meanwhile, it does depend on the choice of ϵ \epsilon ϵ . Moreover, the representation is irreducible [15, Theorem 1.2].
Definition 2.3. For a knot K K K , define its augmentation rank , meridional rank , and bridge number as follows:
ar ( K ) : = max { dim V ϵ : ϵ : Cord ( K ) → k , ϵ ( μ ) ≠ 1 } , mr ( K ) : = min { r : π K is generated by r meridians } , b ( K ) : = min { r : K admits an r -bridge presentation } . \begin{aligned}
\operatorname{ar}(K)&:=\max\{\dim V_\epsilon:\epsilon:\operatorname{Cord}(K)\to\mathbb{k},\ \epsilon(\mu)\ne1\},\\
\operatorname{mr}(K)&:=\min\{r:\pi_K\text{ is generated by }r\text{ meridians}\},\\
b(K)&:=\min\{r:K\text{ admits an }r\text{-bridge presentation}\}.
\end{aligned} ar ( K ) mr ( K ) b ( K ) := max { dim V ϵ : ϵ : Cord ( K ) → k , ϵ ( μ ) = 1 } , := min { r : π K is generated by r meridians } , := min { r : K admits an r -bridge presentation } .
Via the augmentation-sheaf correspondence, the augmentation rank can also be read as the maximum rank of the irreducible diagonalizable microlocal rank one representations of π K \pi_K π K . The two occurrences of the word rank should not be confused. Microlocal rank records the normal jump along the conormal, whereas the other rank is rank in the usual sense. We immediately have the following inequalities.
Proposition 2.4. For every knot K K K ,
ar ( K ) ≤ mr ( K ) ≤ b ( K ) . \operatorname{ar}(K)\le\operatorname{mr}(K)\le b(K). ar ( K ) ≤ mr ( K ) ≤ b ( K ) .
Proof. By construction (2.1), the maximal value of dim V ϵ \dim V_\epsilon dim V ϵ is no larger than the number of meridional generators. Hence it yields
ar ( K ) ≤ mr ( K ) . \operatorname{ar}(K)\le\operatorname{mr}(K). ar ( K ) ≤ mr ( K ) .
A bridge presentation with b ( K ) b(K) b ( K ) bridges gives a Wirtinger presentation whose bridge meridians generate π K \pi_K π K . Hence
mr ( K ) ≤ b ( K ) . \operatorname{mr}(K)\le b(K). mr ( K ) ≤ b ( K ) .
Let ( ρ , V ) (\rho,V) ( ρ , V ) be a diagonalizable μ \mu μ rk-one representation. Since the distinguished meridian and the preferred longitude commute in the knot group, we can choose a suitable basis such that their representing matrices have the form
ρ ( m ) = ( μ 0 0 0 I ) , ρ ( ℓ ) = ( λ 0 0 0 ∗ ) . \rho(m)=\begin{pmatrix}\mu_0 & 0\\0 & I\end{pmatrix},\qquad\rho(\ell)=\begin{pmatrix}\lambda_0 & 0\\0 & *\end{pmatrix}. ρ ( m ) = ( μ 0 0 0 I ) , ρ ( ℓ ) = ( λ 0 0 0 ∗ ) .
Note that the distinguished μ 0 \mu_0 μ 0 -eigenvector of ρ ( m ) \rho(m) ρ ( m ) is also an eigenvector of ρ ( ℓ ) \rho(\ell) ρ ( ℓ ) , with eigenvalue λ 0 \lambda_0 λ 0 . This is a property that we shall use repetitively.
Finally, the associated augmentation has
ϵ ρ ( μ ) = μ 0 , ϵ ρ ( λ ) = λ 0 , ϵ ρ ( [ γ ] ) = ( 1 − μ 0 ) ρ ( γ ) 11 . \epsilon_\rho(\mu)=\mu_0,\qquad\epsilon_\rho(\lambda)=\lambda_0,\qquad\epsilon_\rho([\gamma])=(1-\mu_0)\rho(\gamma)_{11}. ϵ ρ ( μ ) = μ 0 , ϵ ρ ( λ ) = λ 0 , ϵ ρ ([ γ ]) = ( 1 − μ 0 ) ρ ( γ ) 11 .
Definition 2.5. Following the notation above, we call the pair
( μ 0 , λ 0 ) ∈ ( k × ) 2 (\mu_0,\lambda_0)\in(\mathbb{k}^{\times})^2 ( μ 0 , λ 0 ) ∈ ( k × ) 2 the peripheral data of the μ \mu μ rk-1 representation ( ρ , V ) (\rho,V) ( ρ , V ) or the associated augmentation.
The Whitehead pattern and the satellite group
Let W h : = C ∪ P ⊂ S 3 \mathrm{Wh}:=C\cup P\subset S^3 Wh := C ∪ P ⊂ S 3 be the oriented Whitehead link in standard position, where C C C is the axis component and P P P is the pattern component. Removing a tubular neighborhood of C C C gives the solid torus
V : = S 3 ∖ O p ( C ) , \mathcal{V}:=S^3\setminus\mathcal{O}_p(C), V := S 3 ∖ O p ( C ) ,
and the pattern component becomes a pattern P ⊂ V P\subset\mathcal{V} P ⊂ V . The pattern has winding number zero. Equivalently, its two longitudinal runs travel once around V \mathcal{V} V with opposite orientations. Our peripheral convention on the outer boundary is
m C = ℓ V , ℓ C = m V , m_C=\ell_{\mathcal{V}},\qquad\ell_C=m_{\mathcal{V}}, m C = ℓ V , ℓ C = m V ,
where ( m C , ℓ C ) (m_C,\ell_C) ( m C , ℓ C ) is the peripheral pair of the deleted axis. This fixes the notation in the gluing formulas below.
Definition 2.6. Let P ⊂ V P\subset\mathcal{V} P ⊂ V be the Whitehead pattern fixed above. The untwisted Whitehead double of K K K is the knot
W h ( K ) : = ϕ K ( P ) , \mathrm{Wh}(K):=\phi_K(P), Wh ( K ) := ϕ K ( P ) , where ϕ K \phi_K ϕ K is the zero-framed embedding chosen above. Thus its restriction to the boundary identifies the peripheral curves by
( ϕ K ) ∗ ( m V ) = m K , ( ϕ K ) ∗ ( ℓ V ) = ℓ K , (\phi_K)_*(m_{\mathcal{V}})=m_K,\qquad(\phi_K)_*(\ell_{\mathcal{V}})=\ell_K, ( ϕ K ) ∗ ( m V ) = m K , ( ϕ K ) ∗ ( ℓ V ) = ℓ K , with ℓ K \ell_K ℓ K the preferred longitude of K K K . The clasp sign is determined by our fixed choice of $P.
Orient all braid strands downward and let σ i \sigma_i σ i denote a positive right-handed crossing. One convenient braid representative for the oriented Whitehead link is
B W : = σ 1 σ 2 − 1 σ 1 σ 2 − 1 σ 1 ∈ B 3 , B_W:=\sigma_1\sigma_2^{-1}\sigma_1\sigma_2^{-1}\sigma_1\in B_3, B W := σ 1 σ 2 − 1 σ 1 σ 2 − 1 σ 1 ∈ B 3 ,
where the first strand closes to C C C and the other two strands close to P P P . The four crossings between C C C and P P P have signed sum zero. The unique self-crossing of P P P accounts for the correction from the braid framing to the preferred zero framing.
We now pass to the group of the pattern exterior
E P : = V ∖ O p ( P ) . E_P:=\mathcal{V}\setminus\mathcal{O}_p(P). E P := V ∖ O p ( P ) .
Its fundamental group admits the following presentation [18]:
π 1 ( E P ) = ⟨ x 1 , x 2 , ℓ V | ℓ V x 2 ℓ V − 1 = x 1 , ℓ V ( x 2 x 1 − 1 x 2 − 1 x 1 x 2 − 1 ) ℓ V − 1 = x 2 x 1 − 1 x 2 − 1 ⟩ . (4) \pi_1(E_P)=\left\langle x_1,x_2,\ell_{\mathcal{V}}\ \middle|\ \ell_{\mathcal{V}}x_2\ell_{\mathcal{V}}^{-1}=x_1,\ \ell_{\mathcal{V}}(x_2x_1^{-1}x_2^{-1}x_1x_2^{-1})\ell_{\mathcal{V}}^{-1}=x_2x_1^{-1}x_2^{-1}\right\rangle. \tag*{(4)} π 1 ( E P ) = ⟨ x 1 , x 2 , ℓ V ℓ V x 2 ℓ V − 1 = x 1 , ℓ V ( x 2 x 1 − 1 x 2 − 1 x 1 x 2 − 1 ) ℓ V − 1 = x 2 x 1 − 1 x 2 − 1 ⟩ . (4)
Figure 1. Left: Whitehead pattern. Right: Whitehead double of knot 4 1 4_1 4 1 .
The peripheral meridian on the outer boundary is
m V = x 2 x 1 − 1 x 2 − 1 x 1 . (5) m_{\mathcal V}=x_2x_1^{-1}x_2^{-1}x_1. \tag*{(5)} m V = x 2 x 1 − 1 x 2 − 1 x 1 . (5)
and ⟨ m V , ℓ V ⟩ ≅ Z 2 \langle m_{\mathcal V},\ell_{\mathcal V}\rangle\cong\mathbb{Z}^2 ⟨ m V , ℓ V ⟩ ≅ Z 2 . The meridian of the satellite component is x 1 x_1 x 1 .
Figure 2. Generating loops of the pattern exterior.
Lemma 2.7. With the downward-oriented braid convention fixed above and the peripheral base path chosen at the meridian x 1 x_1 x 1 , the preferred zero-framed longitude of the satellite component is represented by
ℓ W h = x 1 − 1 ℓ V x 1 ℓ V − 1 x 1 − 1 ℓ V − 1 x 1 ℓ V . (6) \ell_{\mathrm{Wh}}=x_1^{-1}\ell_{\mathcal V}x_1\ell_{\mathcal V}^{-1}x_1^{-1}\ell_{\mathcal V}^{-1}x_1\ell_{\mathcal V}. \tag*{(6)} ℓ Wh = x 1 − 1 ℓ V x 1 ℓ V − 1 x 1 − 1 ℓ V − 1 x 1 ℓ V . (6) Proof. Follow the zero-framed push-off from the basepoint at x 1 x_1 x 1 through the two oppositely oriented longitudinal runs of the pattern. The two runs give the successive conjugations by ℓ V \ell_{\mathcal V} ℓ V and ℓ V − 1 \ell_{\mathcal V}^{-1} ℓ V − 1 in (6) . The four axis–pattern crossings have total signed contribution zero.
Removing the single pattern-meridian contribution from the braid framing therefore gives the displayed preferred longitude, with no additional power of m V m_V m V . □
On the pattern exterior, the torus boundary is subject to the following conditions:
ℓ V x 2 ℓ V − 1 = x 1 , m V = x 2 x 1 − 1 x 2 − 1 x 1 , [ ℓ V , m V ] = 1 \ell_V x_2\ell_V^{-1}=x_1,\qquad m_V=x_2x_1^{-1}x_2^{-1}x_1,\qquad[\ell_V,m_V]=1 ℓ V x 2 ℓ V − 1 = x 1 , m V = x 2 x 1 − 1 x 2 − 1 x 1 , [ ℓ V , m V ] = 1
Proposition 2.8. Suppose K K K is a non-trivial knot. The knot group of the Whitehead double can be computed by
π W h ( K ) ≅ π 1 ( E P ) ∗ ⟨ m V , ℓ V ⟩ = ⟨ m K , ℓ K ⟩ π K , \pi_{\mathrm{Wh}(K)}\cong\pi_1(E_P)*_{\langle m_V,\ell_V\rangle=\langle m_K,\ell_K\rangle}\pi_K, π Wh ( K ) ≅ π 1 ( E P ) ∗ ⟨ m V , ℓ V ⟩ = ⟨ m K , ℓ K ⟩ π K , where m V m_V m V is identified with m K m_K m K and ℓ V \ell_V ℓ V with ℓ K \ell_K ℓ K .
Proof. Let X K : = S 3 ∖ Op ( K ) X_K:=S^3\setminus\operatorname{Op}(K) X K := S 3 ∖ Op ( K ) . The satellite exterior is the union of E P E_P E P and X K X_K X K . Their intersection retracts onto the outer boundary torus of V V V . The zero-framed gluing identifies the two ordered peripheral pairs. The assertion follows from Seifert–van Kampen. □
The one-step doubling construction
A μ \mu μ rk-one representation of π W h ( K ) \pi_{\mathrm{Wh}(K)} π Wh ( K ) restricts to a representation of π K \pi_K π K of microlocal rank at most two. The goal of this subsection is to describe the reverse construction: starting with a μ \mu μ rk-one representation of π K \pi_K π K , we first pass to a representation of π K \pi_K π K of microlocal rank two and then extend it across the pattern exterior to obtain a μ \mu μ rk-one representation of π W h ( K ) \pi_{\mathrm{Wh}(K)} π Wh ( K ) .
We first record the algebraic construction used in the doubling argument. Let G G G be a group and let ρ : G → GL ( V ) \rho:G\to\operatorname{GL}(V) ρ : G → GL ( V ) be a finite-dimensional representation. Put V ∗ : = Hom k ( V , k ) V^*:=\operatorname{Hom}_{\mathbb{k}}(V,\mathbb{k}) V ∗ := Hom k ( V , k ) . The dual representation ( ρ ∗ , V ∗ ) (\rho^*,V^*) ( ρ ∗ , V ∗ ) is defined by ( ρ ∗ ( g ) φ ) ( v ) : = φ ( ρ ( g − 1 ) v ) (\rho^*(g)\varphi)(v):=\varphi(\rho(g^{-1})v) ( ρ ∗ ( g ) φ ) ( v ) := φ ( ρ ( g − 1 ) v ) , g ∈ G g\in G g ∈ G , φ ∈ V ∗ \varphi\in V^* φ ∈ V ∗ , v ∈ V v\in V v ∈ V .
Definition 2.9. Define the self-dual double associated to ρ \rho ρ to be
ρ # : = ρ ⊕ ρ ∗ : G ⟶ GL ( V ⊕ V ∗ ) . \rho^\#:=\rho\oplus\rho^*:G\longrightarrow\operatorname{GL}(V\oplus V^*). ρ # := ρ ⊕ ρ ∗ : G ⟶ GL ( V ⊕ V ∗ ) .
Let ρ : π K ⟶ GL ( V ) \rho:\pi_K\longrightarrow\operatorname{GL}(V) ρ : π K ⟶ GL ( V ) be an n n n -dimensional diagonalizable μ \mu μ rk-one representation. Choose a distinguished meridian eigenvector e ∈ V e\in V e ∈ V . Since m K m_K m K and ℓ K \ell_K ℓ K commute, it is also an eigenvector for the longitude. Let ( μ , λ ) (\mu,\lambda) ( μ , λ ) , μ ≠ 1 \mu\ne1 μ = 1 be the peripheral data, namely
ρ ( m K ) e = μ e , ρ ( ℓ K ) e = λ e . \rho(m_K)e=\mu e,\qquad\rho(\ell_K)e=\lambda e. ρ ( m K ) e = μ e , ρ ( ℓ K ) e = λ e .
Define V 0 : = ker ( ρ ( m K ) − I ) V_0:=\ker(\rho(m_K)-I) V 0 := ker ( ρ ( m K ) − I ) , so that V = k e ⊕ V 0 V=\mathbb{k}e\oplus V_0 V = k e ⊕ V 0 , and let e ∗ ∈ V ∗ e^*\in V^* e ∗ ∈ V ∗ be determined by e ∗ ( e ) : = 1 e^*(e):=1 e ∗ ( e ) := 1 and e ∗ ∣ V 0 : = 0 e^*|_{V_0}:=0 e ∗ ∣ V 0 := 0 . We consider its self-dual double ( ρ # , V ⊕ V ∗ ) (\rho^\#,V\oplus V^*) ( ρ # , V ⊕ V ∗ ) from Definition 2.9.
Since ρ ( ℓ K ) \rho(\ell_K) ρ ( ℓ K ) preserves V 0 V_0 V 0 , the vector e ∗ e^* e ∗ is a common eigenvector for the dual peripheral action, with eigenvalues μ − 1 \mu^{-1} μ − 1 and λ − 1 \lambda^{-1} λ − 1 . Define
U : = Span { e , e ∗ } , W : = ker ( ρ # ( m K ) − I ) . U:=\operatorname{Span}\{e,e^*\},\qquad W:=\ker(\rho^\#(m_K)-I). U := Span { e , e ∗ } , W := ker ( ρ # ( m K ) − I ) .
They form a splitting of the total space, namely:
V ⊕ V ∗ = U ⊕ W . V\oplus V^*=U\oplus W. V ⊕ V ∗ = U ⊕ W .
In the ordered basis ( e , e ∗ ) (e,e^*) ( e , e ∗ ) of U U U , put
M U : = ρ # ( m K ) ∣ U = ( μ 0 0 μ − 1 ) , L U : = ρ # ( ℓ K ) ∣ U = ( λ 0 0 λ − 1 ) . M_U:=\left.\rho^\#(m_K)\right|_U=
\begin{pmatrix}
\mu& 0\\
0 & \mu^{-1}
\end{pmatrix},
\qquad
L_U:=\left.\rho^\#(\ell_K)\right|_U=
\begin{pmatrix}
\lambda& 0\\
0 & \lambda^{-1}
\end{pmatrix}. M U := ρ # ( m K ) U = ( μ 0 0 μ − 1 ) , L U := ρ # ( ℓ K ) U = ( λ 0 0 λ − 1 ) .
Thus ρ # ( m K ) − I \rho^\#(m_K)-I ρ # ( m K ) − I vanishes on W W W and has rank two on U U U . Moreover, both matrices have determinant one.
We are ready to explain the motivation of the self-dual double construction. Under the satellite gluing, the companion meridian m K m_K m K is identified with the commutator m V m_V m V in (5) . Consequently, a necessary condition for a representation of π K \pi_K π K to extend over the pattern exterior is that the image of m K m_K m K have determinant one. The original μ \mu μ rk-1 representation does not satisfy this condition, since det ρ ( m K ) = μ ≠ 1 \det\rho(m_K)=\mu\ne1 det ρ ( m K ) = μ = 1 . Passing to ρ # = ρ ⊕ ρ ∗ \rho^\#=\rho\oplus\rho^* ρ # = ρ ⊕ ρ ∗ replaces the distinguished eigenvalue μ \mu μ by the reciprocal pair μ , μ − 1 \mu,\mu^{-1} μ , μ − 1 , removing this determinant obstruction.
Our next goal is to extend the self-dual double representation into a μ \mu μ rk-1 representation of the Whitehead pattern. Namely, we aim to find 2-by-2 matrices X 1 X_1 X 1 and X 2 X_2 X 2 that satisfy the following boundary conditions:
L U X 2 L U − 1 = X 1 , M U = X 2 X 1 − 1 X 2 − 1 X 1 . (7) L_U X_2 L_U^{-1}=X_1,\qquad M_U=X_2X_1^{-1}X_2^{-1}X_1. \tag*{(7)} L U X 2 L U − 1 = X 1 , M U = X 2 X 1 − 1 X 2 − 1 X 1 . (7)
Assume X 1 : = ( a b c d ) X_1:=\begin{pmatrix}a&b\\c&d\end{pmatrix} X 1 := ( a c b d ) . The microlocal condition requires rank ( X 1 − I ) = 1 \operatorname{rank}(X_1-I)=1 rank ( X 1 − I ) = 1 . Then, det ( X 1 − I ) = 0 \det(X_1-I)=0 det ( X 1 − I ) = 0 gives b c = ( a − 1 ) ( d − 1 ) bc=(a-1)(d-1) b c = ( a − 1 ) ( d − 1 ) . On the chart b ≠ 0 b\ne0 b = 0 , d ≠ 1 d\ne1 d = 1 , we can normalize the ( 1 , 2 ) (1,2) ( 1 , 2 ) entry to d − 1 d-1 d − 1 , and the ( 2 , 1 ) (2,1) ( 2 , 1 ) entry then becomes a − 1 a-1 a − 1 .
Therefore, we seek a solution of the form
X 1 = ( a d − 1 a − 1 d ) , X 2 = L U − 1 X 1 L U = ( a η − 1 ( d − 1 ) η ( a − 1 ) d ) , X_1=\begin{pmatrix}a&d-1\\a-1&d\end{pmatrix},\qquad X_2=L_U^{-1}X_1L_U=\begin{pmatrix}a&\eta^{-1}(d-1)\\\eta(a-1)&d\end{pmatrix}, X 1 = ( a a − 1 d − 1 d ) , X 2 = L U − 1 X 1 L U = ( a η ( a − 1 ) η − 1 ( d − 1 ) d ) ,
where η : = λ 2 \eta:=\lambda^2 η := λ 2 . Let μ ~ : = det X 1 = a + d − 1 \widetilde{\mu}:=\det X_1=a+d-1 μ := det X 1 = a + d − 1 be the determinant. For μ ~ ∉ { 0 , 1 } \widetilde{\mu}\notin\{0,1\} μ ∈ / { 0 , 1 } , the matrix X 1 X_1 X 1 is invertible and diagonalizable μ \mu μ rk-1. The same holds for X 2 X_2 X 2 , since it is conjugate to X 1 X_1 X 1 . It remains to choose a , d a,d a , d so that the second equation in (2.5) holds and μ ~ ∉ { 0 , 1 } \widetilde{\mu}\notin\{0,1\} μ ∈ / { 0 , 1 } . For the resulting Whitehead double Wh ( K ) \operatorname{Wh}(K) Wh ( K ) , μ ~ \widetilde{\mu} μ will be the distinguished meridian eigenvalue.
Lemma 2.10. Let ( μ , λ ) (\mu,\lambda) ( μ , λ ) be the peripheral data of the diagonalizable μ \mu μ rk-1 representation ρ \rho ρ . Suppose ( μ , λ ) (\mu,\lambda) ( μ , λ ) satisfy the following conditions:
λ 2 ≠ 1 , λ 2 μ ≠ − 1 , λ 4 μ 2 − λ 4 μ + 4 λ 2 μ − μ + 1 ≠ 0. (8) \lambda^2\ne1,\qquad\lambda^2\mu\ne-1,\qquad\lambda^4\mu^2-\lambda^4\mu+4\lambda^2\mu-\mu+1\ne0. \tag*{(8)} λ 2 = 1 , λ 2 μ = − 1 , λ 4 μ 2 − λ 4 μ + 4 λ 2 μ − μ + 1 = 0. (8) Then there exist a , d ∈ k a,d\in\mathbb{k} a , d ∈ k for which X 2 X 1 − 1 X 2 − 1 X 1 = M U X_2X_1^{-1}X_2^{-1}X_1=M_U X 2 X 1 − 1 X 2 − 1 X 1 = M U and μ ~ ∉ { 0 , 1 } \widetilde{\mu}\notin\{0,1\} μ ∈ / { 0 , 1 } .
Proof. We set r : = ( a − 1 ) ( d − 1 ) r:=(a-1)(d-1) r := ( a − 1 ) ( d − 1 ) . Then equation (2.5) is equivalent to X 2 X 1 − 1 = M U X 1 − 1 X 2 X_2X_1^{-1}=M_UX_1^{-1}X_2 X 2 X 1 − 1 = M U X 1 − 1 X 2 . Direct multiplication gives
X 2 X 1 − 1 = 1 μ ~ ( a d − η − 1 r a ( d − 1 ) ( η − 1 − 1 ) d ( a − 1 ) ( η − 1 ) a d − η r ) , X_2X_1^{-1}=\frac{1}{\widetilde{\mu}}\begin{pmatrix}ad-\eta^{-1}r&a(d-1)(\eta^{-1}-1)\\d(a-1)(\eta-1)&ad-\eta r\end{pmatrix}, X 2 X 1 − 1 = μ 1 ( a d − η − 1 r d ( a − 1 ) ( η − 1 ) a ( d − 1 ) ( η − 1 − 1 ) a d − η r ) ,
M U X 1 − 1 X 2 = 1 μ ~ ( μ ( a d − η r ) μ d ( d − 1 ) ( η − 1 − 1 ) μ − 1 a ( a − 1 ) ( η − 1 ) μ − 1 ( a d − η − 1 r ) ) . M_UX_1^{-1}X_2=\frac{1}{\widetilde{\mu}}\begin{pmatrix}\mu(ad-\eta r)&\mu d(d-1)(\eta^{-1}-1)\\\mu^{-1}a(a-1)(\eta-1)&\mu^{-1}(ad-\eta^{-1}r)\end{pmatrix}. M U X 1 − 1 X 2 = μ 1 ( μ ( a d − η r ) μ − 1 a ( a − 1 ) ( η − 1 ) μ d ( d − 1 ) ( η − 1 − 1 ) μ − 1 ( a d − η − 1 r ) ) .
Note that a ≠ 1 a\ne1 a = 1 and d ≠ 1 d\ne1 d = 1 . Otherwise, say a = 1 a=1 a = 1 (resp. d = 1 d=1 d = 1 ), then X 1 X_1 X 1 and X 2 X_2 X 2 are upper (resp. lower) triangular with the diagonal entries { 1 , d } \{1,d\} { 1 , d } (resp. { a , 1 } \{a,1\} { a , 1 } ), so their commutator has diagonal entries { 1 , 1 } \{1,1\} { 1 , 1 } , which is a contradiction to μ ≠ 1 \mu\ne1 μ = 1 . Since η ≠ 1 \eta\ne1 η = 1 , comparison of the ( 1 , 2 ) (1,2) ( 1 , 2 ) and ( 1 , 1 ) (1,1) ( 1 , 1 ) entries yields
a = μ d , (9) a=\mu d, \tag*{(9)} a = μ d , (9)
η ( 1 − μ ) a d + ( μ η 2 − 1 ) r = 0. (10) \eta(1-\mu)ad+(\mu\eta^2-1)r=0. \tag*{(10)} η ( 1 − μ ) a d + ( μ η 2 − 1 ) r = 0. (10)
We consider two possibilities depending on the value of μ \mu μ . Case 1, μ ≠ − 1 \mu\ne-1 μ = − 1 . By μ ~ = a + d − 1 \widetilde{\mu}=a+d-1 μ = a + d − 1 and equation (2.7), we have d = μ ~ + 1 μ + 1 d=\frac{\widetilde{\mu}+1}{\mu+1} d = μ + 1 μ + 1 , and subsequently a = μ ( μ ~ + 1 ) μ + 1 a=\frac{\mu(\widetilde{\mu}+1)}{\mu+1} a = μ + 1 μ ( μ + 1 ) , r = ( μ μ ~ − 1 ) ( μ ~ − μ ) ( μ + 1 ) 2 r=\frac{(\mu\widetilde{\mu}-1)(\widetilde{\mu}-\mu)}{(\mu+1)^2} r = ( μ + 1 ) 2 ( μ μ − 1 ) ( μ − μ ) . Substitution into equation (2.8), followed by elimination of denominators, gives
η μ ( 1 − μ ) ( μ ~ + 1 ) 2 + ( μ η 2 − 1 ) ( μ μ ~ − 1 ) ( μ ~ − μ ) = 0. \eta\mu(1-\mu)(\widetilde{\mu}+1)^2+(\mu\eta^2-1)(\mu\widetilde{\mu}-1)(\widetilde{\mu}-\mu)=0. η μ ( 1 − μ ) ( μ + 1 ) 2 + ( μ η 2 − 1 ) ( μ μ − 1 ) ( μ − μ ) = 0.
Reorganizing by the powers of μ ~ \widetilde{\mu} μ , we have
P η , μ ( μ ~ ) : = ( μ ( η − 1 ) ( η μ + 1 ) ( μ ~ 2 + 1 ) + ( 1 + μ 2 + 2 η μ − 2 η μ 2 − η 2 μ − η 2 μ 3 ) μ ~ ) = 0. (11) P_{\eta,\mu}(\widetilde{\mu}) := \bigl(\mu(\eta-1)(\eta\mu+1)(\widetilde{\mu}^2+1)+(1+\mu^2+2\eta\mu-2\eta\mu^2-\eta^2\mu-\eta^2\mu^3)\widetilde{\mu}\bigr)=0. \tag*{(11)} P η , μ ( μ ) := ( μ ( η − 1 ) ( η μ + 1 ) ( μ 2 + 1 ) + ( 1 + μ 2 + 2 η μ − 2 η μ 2 − η 2 μ − η 2 μ 3 ) μ ) = 0. (11)
Case 2, μ = − 1 \mu=-1 μ = − 1 . Equations (9) give a = − d a=-d a = − d and μ ~ = − 1 \widetilde{\mu}=-1 μ = − 1 . Equation (10) then becomes
( η − 1 ) 2 d 2 = η 2 + 1. (\eta-1)^2d^2=\eta^2+1. ( η − 1 ) 2 d 2 = η 2 + 1.
Since η ≠ 1 \eta\ne1 η = 1 , we may choose d d d with d 2 = ( η 2 + 1 ) / ( η − 1 ) 2 d^2=(\eta^2+1)/(\eta-1)^2 d 2 = ( η 2 + 1 ) / ( η − 1 ) 2 and then set a : = − d a:=-d a := − d . Meanwhile, P η , − 1 ( μ ~ ) = ( η − 1 ) 2 ( μ ~ + 1 ) 2 P_{\eta,-1}(\widetilde{\mu})=(\eta-1)^2(\widetilde{\mu}+1)^2 P η , − 1 ( μ ) = ( η − 1 ) 2 ( μ + 1 ) 2 , so μ ~ = − 1 \widetilde{\mu}=-1 μ = − 1 is also a solution of P η , − 1 ( μ ~ ) = 0 P_{\eta,-1}(\widetilde{\mu})=0 P η , − 1 ( μ ) = 0 .
Combining the two cases, the equation X 2 X 1 − 1 X 2 − 1 X 1 = M U X_2X_1^{-1}X_2^{-1}X_1=M_U X 2 X 1 − 1 X 2 − 1 X 1 = M U reduces to P η , μ ( μ ~ ) = 0 P_{\eta,\mu}(\widetilde{\mu})=0 P η , μ ( μ ) = 0 . Under the hypotheses λ 2 ≠ 1 \lambda^2\ne1 λ 2 = 1 , λ 2 μ ≠ − 1 \lambda^2\mu\ne-1 λ 2 μ = − 1 , the quadratic and constant coefficients of P η , μ ( μ ~ ) P_{\eta,\mu}(\widetilde{\mu}) P η , μ ( μ ) are nonzero. Hence, it is a genuine quadratic polynomial.
Recall that the representation is diagonalizable μ \mu μ rk-one, which requires μ ~ ≠ 1 \widetilde{\mu}\ne1 μ = 1 . We compute
P η , μ ( 1 ) = − ( μ − 1 ) ( η 2 μ 2 − η 2 μ + 4 η μ − μ + 1 ) . P_{\eta,\mu}(1)=-(\mu-1)(\eta^2\mu^2-\eta^2\mu+4\eta\mu-\mu+1). P η , μ ( 1 ) = − ( μ − 1 ) ( η 2 μ 2 − η 2 μ + 4 η μ − μ + 1 ) .
Since μ ≠ 1 \mu\ne1 μ = 1 , the third hypothesis ensures that neither root is 1. Finally, when μ ≠ − 1 \mu\ne-1 μ = − 1 , a root of P η , μ ( μ ~ ) = 0 P_{\eta,\mu}(\widetilde{\mu})=0 P η , μ ( μ ) = 0 determines the desired values of a , d a,d a , d ; when μ = − 1 \mu=-1 μ = − 1 , we use the choice of d d d made above and set a = − d a=-d a = − d .
Remark 2.11. The three conditions in (8) rule out the following cases:
If λ 2 = 1 \lambda^2=1 λ 2 = 1 , then L U = ± I L_U=\pm I L U = ± I , which gives X 2 = X 1 X_2=X_1 X 2 = X 1 , and M U = I M_U=I M U = I .
If λ 2 μ = − 1 \lambda^2\mu=-1 λ 2 μ = − 1 and λ 2 ≠ 1 \lambda^2\ne1 λ 2 = 1 , then the quadratic and constant terms of P η , μ P_{\eta,\mu} P η , μ become zero, and the only solution is μ ~ = 0 \widetilde{\mu}=0 μ = 0 . In other words, X 1 X_1 X 1 is not invertible, and it cannot define a representation.
If λ 4 μ 2 − λ 4 μ + 4 λ 2 μ − μ + 1 = 0 \lambda^4\mu^2-\lambda^4\mu+4\lambda^2\mu-\mu+1=0 λ 4 μ 2 − λ 4 μ + 4 λ 2 μ − μ + 1 = 0 , and λ 2 ≠ 1 \lambda^2\ne1 λ 2 = 1 , λ 2 μ ≠ − 1 \lambda^2\mu\ne-1 λ 2 μ = − 1 , then μ ~ = 1 \widetilde{\mu}=1 μ = 1 . The matrix X 1 X_1 X 1 is unipotent μ \mu μ rk-one but not diagonalizable μ \mu μ rk-one.
Theorem 2.12. Let ρ : π K → GL ( V ) \rho:\pi_K\to\operatorname{GL}(V) ρ : π K → GL ( V ) be an n n n -dimensional irreducible diagonalizable μ \mu μ rk-one representation. Suppose its peripheral data ( μ , λ ) (\mu,\lambda) ( μ , λ ) satisfy (8) in Lemma 2.10. Then there exists a 2 n 2n 2 n -dimensional irreducible diagonalizable μ \mu μ rk-one representation
ρ ~ : π Wh ( K ) ⟶ GL ( V ⊕ V ∗ ) . \widetilde{\rho}:\pi_{\operatorname{Wh}(K)}\longrightarrow\operatorname{GL}(V\oplus V^*). ρ : π Wh ( K ) ⟶ GL ( V ⊕ V ∗ ) .
Proof. Recall the decomposition V ⊕ V ∗ = U ⊕ W V\oplus V^*=U\oplus W V ⊕ V ∗ = U ⊕ W . The matrices X 1 , X 2 X_1,X_2 X 1 , X 2 from Lemma 2.10 act on U U U . Extend both by the identity on W W W , and denote them by X ~ 1 , X ~ 2 \widetilde{X}_1,\widetilde{X}_2 X 1 , X 2 . Define ρ ~ \widetilde{\rho} ρ on the Whitehead pattern by
ρ ~ ( m V ) : = ρ # ( m K ) , ρ ~ ( ℓ V ) : = ρ # ( ℓ K ) , ρ ~ ( x i ) : = X ~ i ( i = 1 , 2 ) . \widetilde{\rho}(m_V):=\rho^\#(m_K),\qquad\widetilde{\rho}(\ell_V):=\rho^\#(\ell_K),\qquad\widetilde{\rho}(x_i):=\widetilde{X}_i\quad(i=1,2). ρ ( m V ) := ρ # ( m K ) , ρ ( ℓ V ) := ρ # ( ℓ K ) , ρ ( x i ) := X i ( i = 1 , 2 ) .
On the knot exterior, define ρ ~ ∣ π K = ρ # ∣ π K \left.\widetilde{\rho}\right|_{\pi_K}=\left.\rho^\#\right|_{\pi_K} ρ ∣ π K = ρ # π K . By van Kampen, this is a well-defined 2 n 2n 2 n -dimensional representation of π Wh ( K ) \pi_{\operatorname{Wh}(K)} π Wh ( K ) .
To see the microlocal rank, we choose the meridian x 1 x_1 x 1 of Wh ( K ) \operatorname{Wh}(K) Wh ( K ) . By construction, X ~ 1 \widetilde{X}_1 X 1 is the identity on W W W , and on U U U , it is straightforward to check that e + e ∗ e+e^* e + e ∗ is an eigenvector of eigenvalue μ ~ ≠ 1 \widetilde{\mu}\ne1 μ = 1 , and ( d − 1 ) e + ( 1 − a ) e ∗ (d-1)e+(1-a)e^* ( d − 1 ) e + ( 1 − a ) e ∗ is an eigenvector with the other eigenvalue, 1. Therefore, ρ ~ \widetilde{\rho} ρ is a diagonalizable μ \mu μ rk-one representation.
Finally, it remains to prove irreducibility and identify the augmentation rank. Since ρ \rho ρ is irreducible, so is ρ ∗ \rho^* ρ ∗ . Consequently, the two nonzero invariant orbit spans satisfy
Span { ρ ( g ) e : g ∈ π K } = V , Span { ρ ∗ ( g ) e ∗ : g ∈ π K } = V ∗ . \operatorname{Span}\{\rho(g)e:g\in\pi_K\}=V,\qquad\operatorname{Span}\{\rho^*(g)e^*:g\in\pi_K\}=V^*. Span { ρ ( g ) e : g ∈ π K } = V , Span { ρ ∗ ( g ) e ∗ : g ∈ π K } = V ∗ .
Let v 1 : = e + e ∗ v_1 := e + e^* v 1 := e + e ∗ be the distinguished right μ ~ \widetilde{\mu} μ -eigenvector of X ~ 1 \widetilde{X}_1 X 1 . Then the distinguished μ ~ \widetilde{\mu} μ -eigenvector of X ~ 2 \widetilde{X}_2 X 2 is v 2 : = λ − 1 e + λ e ∗ v_2 := \lambda^{-1}e + \lambda e^* v 2 := λ − 1 e + λ e ∗ . In matrix form, they are column vectors v 1 = ( 1 , 1 ) t v_1 = (1,1)^t v 1 = ( 1 , 1 ) t , v 2 = ( λ − 1 , λ ) t v_2 = (\lambda^{-1},\lambda)^t v 2 = ( λ − 1 , λ ) t . For any g ∈ π K g \in\pi_K g ∈ π K , the action of ρ ~ ( g ) \widetilde{\rho}(g) ρ ( g ) on the eigenvectors v 1 , v 2 v_1,v_2 v 1 , v 2 is
ρ ~ ( g ) v 1 = ρ ( g ) e ⊕ ρ ∗ ( g ) e ∗ , ρ ~ ( g ) v 2 = λ − 1 ρ ( g ) e ⊕ λ ρ ∗ ( g ) e ∗ . \widetilde{\rho}(g)v_1 = \rho(g)e \oplus\rho^*(g)e^*, \qquad\widetilde{\rho}(g)v_2 = \lambda^{-1}\rho(g)e \oplus\lambda\rho^*(g)e^*. ρ ( g ) v 1 = ρ ( g ) e ⊕ ρ ∗ ( g ) e ∗ , ρ ( g ) v 2 = λ − 1 ρ ( g ) e ⊕ λ ρ ∗ ( g ) e ∗ .
Because λ 2 ≠ 1 \lambda^2 \ne1 λ 2 = 1 , we have
λ ρ ~ ( g ) v 1 − ρ ~ ( g ) v 2 λ − λ − 1 = ρ ( g ) e ⊕ 0 , ρ ~ ( g ) v 2 − λ − 1 ρ ~ ( g ) v 1 λ − λ − 1 = 0 ⊕ ρ ∗ ( g ) e ∗ . \frac{\lambda\widetilde{\rho}(g)v_1-\widetilde{\rho}(g)v_2}{\lambda-\lambda^{-1}} = \rho(g)e \oplus0,\qquad\frac{\widetilde{\rho}(g)v_2-\lambda^{-1}\widetilde{\rho}(g)v_1}{\lambda-\lambda^{-1}} = 0 \oplus\rho^*(g)e^*. λ − λ − 1 λ ρ ( g ) v 1 − ρ ( g ) v 2 = ρ ( g ) e ⊕ 0 , λ − λ − 1 ρ ( g ) v 2 − λ − 1 ρ ( g ) v 1 = 0 ⊕ ρ ∗ ( g ) e ∗ .
Hence the right distinguished meridian orbit spans V ⊕ V ∗ V \oplus V^* V ⊕ V ∗ .
Similarly, regard the row vector w 1 : = ( a − 1 , d − 1 ) ∈ U ∗ w_1 := (a-1,d-1) \in U^* w 1 := ( a − 1 , d − 1 ) ∈ U ∗ as a covector on V ⊕ V ∗ V \oplus V^* V ⊕ V ∗ by extending it by zero on W W W . Direct multiplication gives w 1 X 1 = ( ( a − 1 ) ( a + d − 1 ) , ( d − 1 ) ( a + d − 1 ) ) = μ ~ w 1 w_1X_1 = ((a-1)(a+d-1),(d-1)(a+d-1)) = \widetilde{\mu}w_1 w 1 X 1 = (( a − 1 ) ( a + d − 1 ) , ( d − 1 ) ( a + d − 1 )) = μ w 1 . Since X 2 = L U − 1 X 1 L U X_2 = L_U^{-1}X_1L_U X 2 = L U − 1 X 1 L U , the row vector w 2 : = w 1 L U = ( λ ( a − 1 ) , λ − 1 ( d − 1 ) ) w_2 := w_1L_U = (\lambda(a-1),\lambda^{-1}(d-1)) w 2 := w 1 L U = ( λ ( a − 1 ) , λ − 1 ( d − 1 )) satisfies w 2 X 2 = μ ~ w 2 w_2X_2 = \widetilde{\mu}w_2 w 2 X 2 = μ w 2 .
We claim that a ≠ 1 a \ne1 a = 1 and d ≠ 1 d \ne1 d = 1 . If a = 1 a=1 a = 1 , then (9) gives d = μ − 1 d=\mu^{-1} d = μ − 1 , and (10) becomes η ( 1 − μ ) μ − 1 = 0 \eta(1-\mu)\mu^{-1}=0 η ( 1 − μ ) μ − 1 = 0 , a contradiction. Similarly, d = 1 d=1 d = 1 also leads to a contradiction. Thus both coordinates of w 1 w_1 w 1 are nonzero. Since λ 2 ≠ 1 \lambda^2 \ne1 λ 2 = 1 , the same separation calculation in ( V ⊕ V ∗ ) ∗ = V ∗ ⊕ V ∗ ∗ (V \oplus V^*)^* = V^* \oplus V^{**} ( V ⊕ V ∗ ) ∗ = V ∗ ⊕ V ∗∗ , together with (5) , shows that the left distinguished meridian orbit spans ( V ⊕ V ∗ ) ∗ (V \oplus V^*)^* ( V ⊕ V ∗ ) ∗ .
Since w 1 ( v 1 ) = μ ~ − 1 w_1(v_1)=\widetilde{\mu}-1 w 1 ( v 1 ) = μ − 1 , we normalize it to v 1 ∗ : = w 1 / ( μ ~ − 1 ) v_1^* := w_1/(\widetilde{\mu}-1) v 1 ∗ := w 1 / ( μ − 1 ) , so as to pair with v 1 v_1 v 1 . The preceding arguments imply
Span { ρ ~ ( h ) v 1 : h ∈ π Wh ( K ) } = V ⊕ V ∗ , Span { ρ ~ ∗ ( h ) ( v 1 ∗ ) : h ∈ π Wh ( K ) } = ( V ⊕ V ∗ ) ∗ . (12) \operatorname{Span}\{\widetilde{\rho}(h)v_1:h\in\pi_{\operatorname{Wh}(K)}\}=V\oplus V^*,\qquad\operatorname{Span}\{\widetilde{\rho}^*(h)(v_1^*):h\in\pi_{\operatorname{Wh}(K)}\}=(V\oplus V^*)^*. \tag*{(12)} Span { ρ ( h ) v 1 : h ∈ π Wh ( K ) } = V ⊕ V ∗ , Span { ρ ∗ ( h ) ( v 1 ∗ ) : h ∈ π Wh ( K ) } = ( V ⊕ V ∗ ) ∗ . (12)
Let S ⊂ V ⊕ V ∗ S \subset V \oplus V^* S ⊂ V ⊕ V ∗ be ρ ~ \widetilde{\rho} ρ -invariant. If ( X ~ 1 − I ) S ≠ 0 (\widetilde{X}_1-I)S \ne0 ( X 1 − I ) S = 0 , then S S S contains im ( X ~ 1 − I ) = k v 1 \operatorname{im}(\widetilde{X}_1-I)=\mathbb{k}v_1 im ( X 1 − I ) = k v 1 , and the first equality in (12) gives S = V ⊕ V ∗ S=V\oplus V^* S = V ⊕ V ∗ . If ( X ~ 1 − I ) S = 0 (\widetilde{X}_1-I)S=0 ( X 1 − I ) S = 0 , then S S S lies in the 1 1 1 -eigenspace of X ~ 1 \widetilde{X}_1 X 1 , which v 1 ∗ v_1^* v 1 ∗ annihilates. Invariance of S S S implies that every ρ ~ ∗ ( h ) ( v 1 ∗ ) \widetilde{\rho}^*(h)(v_1^*) ρ ∗ ( h ) ( v 1 ∗ ) also annihilates S S S , and the second equality in (12) gives S = 0 S=0 S = 0 . Thus ρ ~ \widetilde{\rho} ρ is irreducible. □ \square □
Corollary 2.13. Let K K K be a nontrivial knot admitting an n n n -dimensional irreducible diagonalizable μ \mu μ rk-one representation whose peripheral data ( μ , λ ) (\mu,\lambda) ( μ , λ ) satisfy (8) in Lemma 2.10. If n = b ( K ) n=b(K) n = b ( K ) , then
ar ( Wh ( K ) ) = mr ( Wh ( K ) ) = b ( Wh ( K ) ) = 2 n . \operatorname{ar}(\operatorname{Wh}(K))=\operatorname{mr}(\operatorname{Wh}(K))=b(\operatorname{Wh}(K))=2n. ar ( Wh ( K )) = mr ( Wh ( K )) = b ( Wh ( K )) = 2 n . Proof. By Theorem 2.12, there is a 2 n 2n 2 n -dimensional irreducible diagonalizable μ \mu μ rk-one representation ρ ~ \widetilde{\rho} ρ of π Wh ( K ) \pi_{\operatorname{Wh}(K)} π Wh ( K ) . Use the eigenvectors v 1 v_1 v 1 and w 1 w_1 w 1 from the proof of the theorem. After enlarging the meridional generating set if necessary, the matrix of the augmentation induced by ρ ~ \widetilde{\rho} ρ factors as
R i j = ( 1 − μ ~ ) ( v 1 ∗ ρ ~ ( g i ) ) ( ρ ~ ( g j − 1 ) v 1 ) . (13) R_{ij}=(1-\widetilde{\mu})(v_1^*\widetilde{\rho}(g_i))(\widetilde{\rho}(g_j^{-1})v_1). \tag*{(13)} R ij = ( 1 − μ ) ( v 1 ∗ ρ ( g i )) ( ρ ( g j − 1 ) v 1 ) . (13) Equation (12) says that the covectors and vectors in (13) span ( V ⊕ V ∗ ) ∗ (V\oplus V^*)^* ( V ⊕ V ∗ ) ∗ and V ⊕ V ∗ V\oplus V^* V ⊕ V ∗ , respectively. Hence, the first factor has column rank 2 n 2n 2 n and the second has row rank 2 n 2n 2 n . Since μ ~ ≠ 1 \widetilde{\mu}\ne1 μ = 1 , their product has rank 2 n 2n 2 n . Consequently,
ar ( Wh ( K ) ) ≥ 2 n . \operatorname{ar}(\operatorname{Wh}(K))\ge2n. ar ( Wh ( K )) ≥ 2 n . The Whitehead pattern has geometric index two. A minimal bridge presentation of K K K gives a doubled bridge presentation of Wh ( K ) \operatorname{Wh}(K) Wh ( K ) , while the satellite bridge inequality gives the reverse bound. Since K K K is nontrivial, these two bounds apply and give
b ( Wh ( K ) ) = 2 b ( K ) = 2 n ; b(\operatorname{Wh}(K))=2b(K)=2n; b ( Wh ( K )) = 2 b ( K ) = 2 n ;
see [18, Section 4]. Proposition 2.4 now yields
2 n ≤ ar ( Wh ( K ) ) ≤ mr ( Wh ( K ) ) ≤ b ( Wh ( K ) ) = 2 n , 2n \le\operatorname{ar}(\operatorname{Wh}(K)) \le\operatorname{mr}(\operatorname{Wh}(K)) \le b(\operatorname{Wh}(K)) = 2n, 2 n ≤ ar ( Wh ( K )) ≤ mr ( Wh ( K )) ≤ b ( Wh ( K )) = 2 n ,
and all three ranks are equal to 2 n 2n 2 n .
Proposition 2.14. The peripheral data of ρ ~ : π Wh ( K ) → GL ( V ⊕ V ∗ ) \tilde{\rho}\colon\pi_{\operatorname{Wh}(K)} \to\operatorname{GL}(V \oplus V^{*}) ρ ~ : π Wh ( K ) → GL ( V ⊕ V ∗ ) are the following:
(1) If μ ≠ − 1 \mu\ne-1 μ = − 1 , then
( μ ~ = a + d − 1 , λ ~ = μ + λ 2 − μ ~ − μ λ 2 μ ~ μ λ 2 − μ μ ~ − λ 2 μ ~ + 1 ) . (14) \left(\tilde{\mu}=a+d-1,\ \tilde{\lambda}=\frac{\mu+\lambda^{2}-\tilde{\mu}-\mu\lambda^{2}\tilde{\mu}}{\mu\lambda^{2}-\mu\tilde{\mu}-\lambda^{2}\tilde{\mu}+1}\right). \tag*{(14)} ( μ ~ = a + d − 1 , λ ~ = μ λ 2 − μ μ ~ − λ 2 μ ~ + 1 μ + λ 2 − μ ~ − μ λ 2 μ ~ ) . (14) where a , d a,d a , d are the choices made in the construction.
(2) If μ = − 1 \mu=-1 μ = − 1 , then a = − d a=-d a = − d , and
( μ ~ = − 1 , λ ~ = η + 1 + η − 1 + ( η − η − 1 ) d ) . (15) \left(\tilde{\mu}=-1,\ \tilde{\lambda}=\eta+1+\eta^{-1}+(\eta-\eta^{-1})d\right). \tag*{(15)} ( μ ~ = − 1 , λ ~ = η + 1 + η − 1 + ( η − η − 1 ) d ) . (15) where d d d is the choice made in the construction.
Proof. The meridian of Wh ( K ) \operatorname{Wh}(K) Wh ( K ) is x 1 x_1 x 1 , and X 1 v 1 = μ ~ v 1 X_1v_1=\tilde{\mu}v_1 X 1 v 1 = μ ~ v 1 . By (6) , the action of the longitude on U U U is
ρ ~ ( ℓ Wh ) ∣ U = X 1 − 1 L U X 1 L U − 1 X 1 − 1 L U − 1 X 1 L U . \tilde{\rho}(\ell_{\operatorname{Wh}})|_U=X_1^{-1}L_UX_1L_U^{-1}X_1^{-1}L_U^{-1}X_1L_U. ρ ~ ( ℓ Wh ) ∣ U = X 1 − 1 L U X 1 L U − 1 X 1 − 1 L U − 1 X 1 L U .
Because v 1 = e + e ∗ v_1=e+e^* v 1 = e + e ∗ is the μ ~ \tilde{\mu} μ ~ -eigenvector of X 1 X_1 X 1 , it must be an eigenvector for ρ ~ ( ℓ Wh ) \tilde{\rho}(\ell_{\operatorname{Wh}}) ρ ~ ( ℓ Wh ) . A straightforward calculation yields
ρ ~ ( ℓ Wh ) v 1 = η ( a − 1 ) + ( d − 1 ) ( a − 1 ) + η ( d − 1 ) v 1 . (16) \tilde{\rho}(\ell_{\operatorname{Wh}})v_1=\frac{\eta(a-1)+(d-1)}{(a-1)+\eta(d-1)}v_1. \tag*{(16)} ρ ~ ( ℓ Wh ) v 1 = ( a − 1 ) + η ( d − 1 ) η ( a − 1 ) + ( d − 1 ) v 1 . (16)
Note the denominator ( a − 1 ) + η ( d − 1 ) (a-1)+\eta(d-1) ( a − 1 ) + η ( d − 1 ) is never zero. Otherwise, suppose ( a − 1 ) + η ( d − 1 ) = 0 (a-1)+\eta(d-1)=0 ( a − 1 ) + η ( d − 1 ) = 0 . Together with a = μ d a=\mu d a = μ d , this gives ( μ + η ) d = 1 + η (\mu+\eta)d=1+\eta ( μ + η ) d = 1 + η . If μ + η = 0 \mu+\eta=0 μ + η = 0 , then η = − 1 \eta=-1 η = − 1 and hence μ = 1 \mu=1 μ = 1 , which is a contradiction. Therefore, d = 1 + η μ + η d=\frac{1+\eta}{\mu+\eta} d = μ + η 1 + η . Substituting this expression and a = μ d a=\mu d a = μ d into (10) , we obtain
0 = η ( 1 − μ ) a d + ( μ η 2 − 1 ) ( a − 1 ) ( d − 1 ) = η ( 1 − μ ) ( 1 + μ η ) 2 ( μ + η ) 2 . 0=\eta(1-\mu)ad+(\mu\eta^{2}-1)(a-1)(d-1)=\frac{\eta(1-\mu)(1+\mu\eta)^{2}}{(\mu+\eta)^{2}}. 0 = η ( 1 − μ ) a d + ( μ η 2 − 1 ) ( a − 1 ) ( d − 1 ) = ( μ + η ) 2 η ( 1 − μ ) ( 1 + μ η ) 2 .
This yields either η = 0 \eta=0 η = 0 , μ = 1 \mu=1 μ = 1 , or μ η = − 1 \mu\eta=-1 μ η = − 1 , which is a contradiction.
Next, we shall solve a a a and d d d in terms of μ \mu μ and λ \lambda λ (or η \eta η ). There are two possibilities.
Case 1, μ ≠ − 1 \mu\ne-1 μ = − 1 . From a = μ d a=\mu d a = μ d and μ ~ = a + d − 1 \tilde{\mu}=a+d-1 μ ~ = a + d − 1 , we have a − 1 = ( μ μ ~ − 1 ) / ( μ + 1 ) a-1=(\mu\tilde{\mu}-1)/(\mu+1) a − 1 = ( μ μ ~ − 1 ) / ( μ + 1 ) and d − 1 = ( μ ~ − μ ) / ( μ + 1 ) d-1=(\tilde{\mu}-\mu)/(\mu+1) d − 1 = ( μ ~ − μ ) / ( μ + 1 ) . Substituting these identities into (2.14) gives the desired (2.12).
Case 2, μ = − 1 \mu=-1 μ = − 1 . Then a = − d a=-d a = − d , μ ~ = − 1 \tilde{\mu}=-1 μ ~ = − 1 , and ( η − 1 ) 2 d 2 = η 2 + 1 (\eta-1)^{2}d^{2}=\eta^{2}+1 ( η − 1 ) 2 d 2 = η 2 + 1 . Substitution into (2.14) gives
λ ~ = η + 1 + ( η − 1 ) d η + 1 − ( η − 1 ) d = η + 1 + η − 1 + ( η − η − 1 ) d . \tilde{\lambda}=\frac{\eta+1+(\eta-1)d}{\eta+1-(\eta-1)d}=\eta+1+\eta^{-1}+(\eta-\eta^{-1})d. λ ~ = η + 1 − ( η − 1 ) d η + 1 + ( η − 1 ) d = η + 1 + η − 1 + ( η − η − 1 ) d .
This proves the desired (2.13).
Iteration and the algebraic peripheral correspondence
The main conclusion of the preceding section is a rank-doubling construction. Starting from an n n n -dimensional irreducible diagonalizable μ \mu μ rk-one representation of π K \pi_K π K , under suitable conditions, we obtain a 2 n 2n 2 n -dimensional irreducible diagonalizable μ \mu μ rk-one representation of π W h ( K ) \pi_{\mathrm{Wh}(K)} π Wh ( K ) . If the construction can be repeated N N N times, it produces a 2 N n 2^Nn 2 N n -dimensional representation of π W h N ( K ) \pi_{\mathrm{Wh}^N(K)} π Wh N ( K ) , and hence ar ( W h N ( K ) ) ≥ 2 N n \operatorname{ar}(\mathrm{Wh}^N(K)) \geq2^Nn ar ( Wh N ( K )) ≥ 2 N n .
Most of the constructions work for any field k \mathbb{k} k , but we eventually take k = C \mathbb{k} = \mathbb{C} k = C for the geometric application. An affine algebraic set is a common zero locus of polynomials, with its Zariski topology; it need not be irreducible. We write k [ A ] \mathbb{k}[A] k [ A ] for its coordinate ring. A regular function on an open subset is locally a quotient of polynomials with a nonvanishing denominator. An algebraic set is irreducible if it is nonempty and not a union of two proper closed subsets. An irreducible component is a maximal irreducible closed subset. A nonempty open subset of an irreducible algebraic set is dense, and a morphism is dominant if its image is dense. Dimension is the maximum length of a strict chain of nonempty irreducible closed subsets. A zero-dimensional irreducible algebraic set is a point, and a curve is irreducible and one-dimensional. See [23, Chapter I, Sections 1–4].
A subset is constructible if it is a finite union of locally closed subsets. Chevalley’s theorem says that the image of a constructible set under a morphism is constructible [19, Chapter 1, Theorem 4.91 and Observation 4.92]. We shall also use the elementary consequence that a dense constructible subset of an irreducible algebraic set contains a nonempty open subset. These conventions apply to the peripheral tori and correspondences below.
Let ( k × ) 2 (\mathbb{k}^{\times})^2 ( k × ) 2 be the algebraic torus with coordinates ( μ , η = λ 2 ) (\mu,\eta= \lambda^2) ( μ , η = λ 2 ) . Recall from § 2.3 that the Whitehead doubling construction requires ( μ , η ) ∈ U (\mu,\eta) \in\mathcal{U} ( μ , η ) ∈ U , where
U : = { ( μ , η ) ∈ ( k × ) 2 : ( μ − 1 ) ( η − 1 ) ( μ η + 1 ) ( μ 2 η 2 − μ η 2 + 4 μ η − μ + 1 ) ≠ 0 } . \mathcal{U} := \left\{(\mu,\eta) \in(\mathbb{k}^{\times})^2 : (\mu-1)(\eta-1)(\mu\eta+1)(\mu^2\eta^2-\mu\eta^2+4\mu\eta-\mu+1) \ne0\right\}. U := { ( μ , η ) ∈ ( k × ) 2 : ( μ − 1 ) ( η − 1 ) ( μ η + 1 ) ( μ 2 η 2 − μ η 2 + 4 μ η − μ + 1 ) = 0 } .
Here the first constraint is to ensure that ρ ( m K ) \rho(m_K) ρ ( m K ) is diagonalizable, and the remaining three constraints are the hypotheses (8) used in Lemma 2.10, Theorem 2.12 and Corollary 2.13, with reasons explained in Remark 2.11. As we repeat the doubling construction, the same conditions must hold for the output peripheral data at every stage. We begin with the definition for the loci where N N N iterations are allowed.
Definition 2.15. Define a length-N N N sequence consisting of
( ( μ j , η j ) 0 ≤ j ≤ N , ( a j , d j ) 0 ≤ j < N ) ∈ ( k × ) 2 N + 2 × k 2 N , \left((\mu_j,\eta_j)_{0\leq j\leq N},(a_j,d_j)_{0\leq j<N}\right) \in(\mathbb{k}^{\times})^{2N+2} \times\mathbb{k}^{2N}, ( ( μ j , η j ) 0 ≤ j ≤ N , ( a j , d j ) 0 ≤ j < N ) ∈ ( k × ) 2 N + 2 × k 2 N , satisfying the following conditions:
doubling constraints: for 0 ≤ j < N 0 \leq j < N 0 ≤ j < N ,
μ j ≠ 1 , η j ≠ 1 , η j μ j ≠ − 1 , η j 2 μ j 2 − η j 2 μ j + 4 η j μ j − μ j + 1 ≠ 0 ; \mu_j \ne1,\qquad\eta_j \ne1,\qquad\eta_j\mu_j \ne-1,\qquad\eta_j^2\mu_j^2-\eta_j^2\mu_j+4\eta_j\mu_j-\mu_j+1 \ne0; μ j = 1 , η j = 1 , η j μ j = − 1 , η j 2 μ j 2 − η j 2 μ j + 4 η j μ j − μ j + 1 = 0 ; internal relations: for 0 ≤ j < N 0 \leq j < N 0 ≤ j < N ,
a j = μ j d j , η j ( 1 − μ j ) a j d j + ( μ j η j 2 − 1 ) ( a j − 1 ) ( d j − 1 ) = 0. a_j = \mu_jd_j,\qquad\eta_j(1-\mu_j)a_jd_j+(\mu_j\eta_j^2-1)(a_j-1)(d_j-1)=0. a j = μ j d j , η j ( 1 − μ j ) a j d j + ( μ j η j 2 − 1 ) ( a j − 1 ) ( d j − 1 ) = 0. inductive relations: for 0 ≤ j < N 0 \leq j < N 0 ≤ j < N ,
μ j + 1 : = a j + d j − 1 , η j + 1 : = ( η j ( a j − 1 ) + ( d j − 1 ) ( a j − 1 ) + η j ( d j − 1 ) ) 2 . \mu_{j+1} := a_j+d_j-1,\qquad\eta_{j+1} := \left(\frac{\eta_j(a_j-1)+(d_j-1)}{(a_j-1)+\eta_j(d_j-1)}\right)^2. μ j + 1 := a j + d j − 1 , η j + 1 := ( ( a j − 1 ) + η j ( d j − 1 ) η j ( a j − 1 ) + ( d j − 1 ) ) 2 .
Let C N C_N C N be the set of length-N N N sequences. Define the projection
pr : ( k × ) 2 N + 2 × k 2 N ⟶ ( k × ) 2 , ( ( μ j , η j ) , ( a j , d j ) ) ⟼ ( μ 0 , η 0 ) . \operatorname{pr} : (\mathbb{k}^{\times})^{2N+2} \times\mathbb{k}^{2N} \longrightarrow(\mathbb{k}^{\times})^2,
\qquad
((\mu_j,\eta_j), (a_j,d_j)) \longmapsto(\mu_0,\eta_0). pr : ( k × ) 2 N + 2 × k 2 N ⟶ ( k × ) 2 , (( μ j , η j ) , ( a j , d j )) ⟼ ( μ 0 , η 0 ) .
Define S N : = pr ( C N ) S_N := \operatorname{pr}(C_N) S N := pr ( C N ) to be the image. Note that S 0 = ( k × ) 2 S_0 = (\mathbb{k}^{\times})^2 S 0 = ( k × ) 2 and S 1 = U S_1 = \mathcal{U} S 1 = U . Moreover, by truncating a length-( N + 1 ) (N+1) ( N + 1 ) sequence, we have
S N + 1 ⊆ S N . S_{N+1} \subseteq S_N. S N + 1 ⊆ S N .
A length-N N N sequence records the compatible data required to carry out N N N successive doubling steps. Suitable initial peripheral data would start the iteration.
Proposition 2.16. Let ρ 0 : π K → G L ( V 0 ) \rho_0 : \pi_K \to\mathrm{GL}(V_0) ρ 0 : π K → GL ( V 0 ) be an n n n -dimensional irreducible diagonalizable μ \mu μ rk-one representation with peripheral data ( μ 0 , λ 0 ) (\mu_0,\lambda_0) ( μ 0 , λ 0 ) . Suppose ( μ 0 , λ 0 2 ) ∈ S N (\mu_0,\lambda_0^2) \in S_N ( μ 0 , λ 0 2 ) ∈ S N . Then for each 0 ≤ j ≤ N 0 \leq j \leq N 0 ≤ j ≤ N , there exists a 2 j n 2^j n 2 j n -dimensional irreducible diagonalizable μ \mu μ rk-one representation
ρ j : π W h j ( K ) ⟶ G L ( V j ) . \rho_j : \pi_{\mathrm{Wh}^j(K)} \longrightarrow\mathrm{GL}(V_j). ρ j : π Wh j ( K ) ⟶ GL ( V j ) .
Proof. Choose a length-N N N sequence ( ( μ j , η j ) 0 ≤ j ≤ N , ( a j , d j ) 0 ≤ j < N ) ((\mu_j,\eta_j)_{0\leq j\leq N},(a_j,d_j)_{0\leq j<N}) (( μ j , η j ) 0 ≤ j ≤ N , ( a j , d j ) 0 ≤ j < N ) representing ( μ 0 , λ 0 2 ) ∈ S N (\mu_0,\lambda_0^2) \in S_N ( μ 0 , λ 0 2 ) ∈ S N . For 0 ≤ j < N 0 \leq j < N 0 ≤ j < N , use the block determined by a j , d j a_j,d_j a j , d j in the construction of Theorem 2.12, and define V j + 1 : = V j ⊕ V j ∗ V_{j+1}:=V_j\oplus V_j^* V j + 1 := V j ⊕ V j ∗ . The doubling constraints allow the construction at every stage, while the inductive relations and the peripheral formula give μ j + 1 = a j + d j − 1 \mu_{j+1}=a_j+d_j-1 μ j + 1 = a j + d j − 1 and λ j + 1 = η j + 1 \lambda_{j+1}=\eta_{j+1} λ j + 1 = η j + 1 . Induction therefore gives irreducible diagonalizable μ \mu μ rk-one representations
ρ j : π W h j ( K ) ⟶ G L ( V j ) , 0 ≤ j ≤ N . \rho_j:\pi_{\mathrm{Wh}^j(K)}\longrightarrow\mathrm{GL}(V_j), \qquad0\leq j\leq N. ρ j : π Wh j ( K ) ⟶ GL ( V j ) , 0 ≤ j ≤ N .
Since dim V j + 1 = 2 dim V j \dim V_{j+1}=2\dim V_j dim V j + 1 = 2 dim V j , we have dim V j = 2 j n \dim V_j=2^j n dim V j = 2 j n . □
Corollary 2.17. Let K K K be nontrivial, and let ρ 0 : π K → G L ( V 0 ) \rho_0:\pi_K\to\mathrm{GL}(V_0) ρ 0 : π K → GL ( V 0 ) be an n n n -dimensional irreducible diagonalizable μ \mu μ rk-one representation with peripheral data ( μ 0 , λ 0 ) (\mu_0,\lambda_0) ( μ 0 , λ 0 ) . If b ( K ) = n b(K)=n b ( K ) = n and ( μ 0 , λ 0 2 ) ∈ S N (\mu_0,\lambda_0^2)\in S_N ( μ 0 , λ 0 2 ) ∈ S N , then
ar ( W h j ( K ) ) = mr ( W h j ( K ) ) = b ( W h j ( K ) ) = 2 j n , 0 ≤ j ≤ N . (17) \operatorname{ar}(\mathrm{Wh}^j(K))=\operatorname{mr}(\mathrm{Wh}^j(K))=b(\mathrm{Wh}^j(K))=2^j n,\qquad0\leq j\leq N. \tag*{(17)} ar ( Wh j ( K )) = mr ( Wh j ( K )) = b ( Wh j ( K )) = 2 j n , 0 ≤ j ≤ N . (17)
Proof. This proof iterates the argument in Corollary 2.13 N N N times. Namely, Proposition 2.16 yields ar ( W h j ( K ) ) ≥ 2 j n \operatorname{ar}(\mathrm{Wh}^j(K))\geq2^j n ar ( Wh j ( K )) ≥ 2 j n . Iterating the doubled bridge presentation gives b ( W h j ( K ) ) ≤ 2 j b ( K ) = 2 j n b(\mathrm{Wh}^j(K))\leq2^j b(K)=2^j n b ( Wh j ( K )) ≤ 2 j b ( K ) = 2 j n . Together with Proposition 2.4, these bounds give
2 j n ≤ ar ( W h j ( K ) ) ≤ mr ( W h j ( K ) ) ≤ b ( W h j ( K ) ) ≤ 2 j n , 2^j n\leq\operatorname{ar}(\mathrm{Wh}^j(K))\leq\operatorname{mr}(\mathrm{Wh}^j(K))\leq b(\mathrm{Wh}^j(K))\leq2^j n, 2 j n ≤ ar ( Wh j ( K )) ≤ mr ( Wh j ( K )) ≤ b ( Wh j ( K )) ≤ 2 j n ,
which proves the claimed equalities. □
Example 2.18. We now consider the case when K = 3 1 K=3_1 K = 3 1 is the right-handed trefoil, which has bridge index 2. A routine computation shows that we can express its knot group as
π K = ⟨ a , b ∣ a b a = b a b ⟩ . \pi_K=\langle a,b\mid aba=bab\rangle. π K = ⟨ a , b ∣ aba = bab ⟩ . Furthermore, we can identify m K = a m_K=a m K = a and ℓ K = ( a b ) 3 a − 6 \ell_K=(ab)^3a^{-6} ℓ K = ( ab ) 3 a − 6 . We can write down a concrete representation
ρ : π K ⟶ G L ( C 2 ) \rho:\pi_K\longrightarrow\mathrm{GL}(\mathbb{C}^2) ρ : π K ⟶ GL ( C 2 ) as
ρ ( a ) = ( 2 1 0 1 ) and ρ ( b ) = ( 1 0 − 2 2 ) . \rho(a)=\begin{pmatrix}2&1\\0&1\end{pmatrix}
\qquad\text{and}\qquad
\rho(b)=\begin{pmatrix}1&0\\-2&2\end{pmatrix}. ρ ( a ) = ( 2 0 1 1 ) and ρ ( b ) = ( 1 − 2 0 2 ) . One can verify it is indeed a diagonalizable μ \mu μ rk-one representation with peripheral data ( μ 0 , λ 0 ) = ( 2 , − 1 8 ) (\mu_0,\lambda_0)=(2,-\frac{1}{8}) ( μ 0 , λ 0 ) = ( 2 , − 8 1 ) . At each iteration, there are two choices of μ \mu μ , which are roots of the
quadratic polynomial in (11) . Since the two distinct candidates are inverse to each other, we always take the one whose norm is greater than 1. This provides an iteration ( μ n , λ n 2 ) → ( μ n + 1 , λ n + 1 2 ) (\mu_n,\lambda_n^2) \to(\mu_{n+1},\lambda_{n+1}^2) ( μ n , λ n 2 ) → ( μ n + 1 , λ n + 1 2 ) provided ( μ n , λ n 2 ) ∈ S 1 (\mu_n,\lambda_n^2)\in S_1 ( μ n , λ n 2 ) ∈ S 1 . A short Python script prints the peripheral data of the first five iterations as follows. With some effort, one could show that the initial data ( 2 , 1 64 ) (2,\frac{1}{64}) ( 2 , 64 1 )
Table 1. Peripheral data for first 5 iterations, choosing μ n > 1 \mu_n>1 μ n > 1 at every step. Values are rounded to 16 significant digits.
lies in S N S_N S N for all N N N . The final remark is to note that this infinite iteration stays “bounded” in contrast to the iterable branches discussed in § 4.1 .
Note that the above discussion does not involve the consideration of a specific knot K K K . The sets S N S_N S N depend only on the Whitehead pattern, not on the companion knot. To apply the construction to a given companion K K K for a prescribed number N N N of steps, one must find a representation of the required dimension whose peripheral data satisfy ( μ , λ 2 ) ∈ S N (\mu,\lambda^2)\in S_N ( μ , λ 2 ) ∈ S N . We next show that each S N S_N S N contains a Zariski-open dense subset of the ambient peripheral torus. This statement does not by itself imply that S N S_N S N meets a prescribed peripheral curve of K K K . The next proposition shows an essential property of S N S_N S N , which asserts enough general points.
Proposition 2.19. For every N ≥ 0 N\geq0 N ≥ 0 , the set S N S_N S N is constructible and contains a Zariski-open dense subset of ( k × ) 2 (\mathbb{k}^{\times})^2 ( k × ) 2 . Moreover, since k = C \mathbb{k}=\mathbb{C} k = C is uncountable, ⋂ N ∈ N S N \bigcap_{N\in\mathbb{N}}S_N ⋂ N ∈ N S N is Zariski dense.
To prove this proposition, we first consider the correspondence defined by the doubling construction. Consider two tori ( k × ) ( μ , η ) 2 (\mathbb{k}^{\times})^2_{(\mu,\eta)} ( k × ) ( μ , η ) 2 and ( k × ) ( μ ~ , λ ~ ) 2 (\mathbb{k}^{\times})^2_{(\widetilde{\mu},\widetilde{\lambda})} ( k × ) ( μ , λ ) 2 . Let π 1 , π 2 \pi_1,\pi_2 π 1 , π 2 be the projections from the product to the first and second components. We also define π ~ 2 \widetilde{\pi}_2 π 2 by
π ~ 2 ( μ , η , μ ~ , λ ~ ) : = ( μ ~ , λ ~ 2 ) \widetilde{\pi}_2(\mu,\eta,\widetilde{\mu},\widetilde{\lambda}):=(\widetilde{\mu},\widetilde{\lambda}^{2}) π 2 ( μ , η , μ , λ ) := ( μ , λ 2 )
These maps fit into the commutative diagram
( k × ) ( μ , η ) 2 × ( k × ) ( μ ~ , λ ~ ) 2 → π 1 ( k × ) ( μ , η ) 2 π 2 ↓ π ~ 2 ↑ ( k × ) ( μ ~ , λ ~ ) 2 → λ ~ = λ 2 ( k × ) ( μ ~ , η ~ ) 2 \begin{CD}
(\mathbb{k}^{\times})^2_{(\mu,\eta)} \times(\mathbb{k}^{\times})^2_{(\widetilde{\mu},\widetilde{\lambda})} @>{\pi_1}>> (\mathbb{k}^{\times})^2_{(\mu,\eta)} \\
@V{\pi_2}VV @. @A{\widetilde{\pi}_2}AA \\
(\mathbb{k}^{\times})^2_{(\widetilde{\mu},\widetilde{\lambda})} @>>{\widetilde{\lambda}=\lambda^2}> (\mathbb{k}^{\times})^2_{(\widetilde{\mu},\widetilde{\eta})}
\end{CD} ( k × ) ( μ , η ) 2 × ( k × ) ( μ , λ ) 2 π 2 ↓ ⏐ ( k × ) ( μ , λ ) 2 π 1 λ = λ 2 ( k × ) ( μ , η ) 2 ( k × ) ( μ , η ) 2 π 2 ⏐ ↑
We formulate a correspondence between ( μ , η ) (\mu,\eta) ( μ , η ) and ( μ ~ , λ ~ ) (\widetilde{\mu},\widetilde{\lambda}) ( μ , λ ) from the iteration procedure. Recall from (11) that μ ~ \widetilde{\mu} μ is the solution of P η , μ ( μ ~ ) = 0 P_{\eta,\mu}(\widetilde{\mu})=0 P η , μ ( μ ) = 0 , and from (14) that λ \lambda λ is a rational polynomial. Let δ ( μ , η , μ ~ ) : = μ η − μ μ ~ − η μ ~ + 1 \delta(\mu,\eta,\widetilde{\mu}):=\mu\eta-\mu\widetilde{\mu}-\eta\widetilde{\mu}+1 δ ( μ , η , μ ) := μ η − μ μ − η μ + 1 be the denominator, then we can rewrite the expression as a polynomial δ λ ~ + μ η μ ~ − μ − η + μ ~ = 0 \delta\widetilde{\lambda}+\mu\eta\widetilde{\mu}-\mu-\eta+\widetilde{\mu}=0 δ λ + μ η μ − μ − η + μ = 0 . Then we define:
W ^ : = { ( μ , η , μ ~ , λ ~ ) : P η , μ ( μ ~ ) = 0 , δ λ ~ + μ η μ ~ − μ − η + μ ~ = 0 } ⊂ ( k × ) ( μ , η ) 2 × ( k × ) ( μ ~ , λ ~ ) 2 . (18) \widehat{W}:=\left\{(\mu,\eta,\widetilde{\mu},\widetilde{\lambda}):P_{\eta,\mu}(\widetilde{\mu})=0,\ \delta\widetilde{\lambda}+\mu\eta\widetilde{\mu}-\mu-\eta+\widetilde{\mu}=0\right\}\subset(\mathbb{k}^{\times})^2_{(\mu,\eta)}\times(\mathbb{k}^{\times})^2_{(\widetilde{\mu},\widetilde{\lambda})}. \tag*{(18)} W := { ( μ , η , μ , λ ) : P η , μ ( μ ) = 0 , δ λ + μ η μ − μ − η + μ = 0 } ⊂ ( k × ) ( μ , η ) 2 × ( k × ) ( μ , λ ) 2 . (18)
Note that P η , μ ( μ ~ ) = 0 P_{\eta,\mu}(\widetilde{\mu})=0 P η , μ ( μ ) = 0 is generically quadratic, namely
P η , μ ( μ ~ ) = A μ ~ 2 + B μ ~ + A , P_{\eta,\mu}(\widetilde{\mu})=A\widetilde{\mu}^{2}+B\widetilde{\mu}+A, P η , μ ( μ ) = A μ 2 + B μ + A ,
where
A = μ ( η − 1 ) ( η μ + 1 ) , A=\mu(\eta-1)(\eta\mu+1), A = μ ( η − 1 ) ( η μ + 1 ) ,
B = 1 + μ 2 + 2 η μ − 2 η μ 2 − η 2 μ − η 2 μ 3 . B=1+\mu^{2}+2\eta\mu-2\eta\mu^{2}-\eta^{2}\mu-\eta^{2}\mu^{3}. B = 1 + μ 2 + 2 η μ − 2 η μ 2 − η 2 μ − η 2 μ 3 .
The discriminant is
B 2 − 4 A 2 = ( μ − 1 ) ( μ + 1 ) 2 ( μ η 2 − 1 ) ( μ 2 η 2 − μ η 2 + 4 μ η − μ + 1 ) . (19) B^{2}-4A^{2}=(\mu-1)(\mu+1)^{2}(\mu\eta^{2}-1)(\mu^{2}\eta^{2}-\mu\eta^{2}+4\mu\eta-\mu+1). \tag*{(19)} B 2 − 4 A 2 = ( μ − 1 ) ( μ + 1 ) 2 ( μ η 2 − 1 ) ( μ 2 η 2 − μ η 2 + 4 μ η − μ + 1 ) . (19)
Under the conditions μ ≠ − 1 \mu\ne-1 μ = − 1 and μ η 2 ≠ 1 \mu\eta^{2}\ne1 μ η 2 = 1 , the discriminant is non-zero. Hence, we introduce the Zariski-open subset
U ′ : = { ( μ , η ) ∈ U : μ ≠ − 1 , μ η 2 ≠ 1 } ⊂ U ⊂ ( k × ) ( μ , η ) 2 . \mathcal{U}' := \{(\mu,\eta)\in\mathcal{U}:\mu\ne-1,\ \mu\eta^{2}\ne1\}\subset\mathcal{U}\subset(\mathbb{k}^{\times})^{2}_{(\mu,\eta)}. U ′ := {( μ , η ) ∈ U : μ = − 1 , μ η 2 = 1 } ⊂ U ⊂ ( k × ) ( μ , η ) 2 .
Finally, we introduce
W : = Closure ( W ^ ∩ π 1 − 1 ( U ′ ) ) ∩ W ^ . \mathcal{W}:=\operatorname{Closure}(\widehat{\mathcal{W}}\cap\pi_{1}^{-1}(\mathcal{U}'))\cap\widehat{\mathcal{W}}. W := Closure ( W ∩ π 1 − 1 ( U ′ )) ∩ W .
This is the correspondence that we will work with.
Proposition 2.20. We have the following properties:
(1) W \mathcal{W} W is irreducible.
(2) π 1 ∣ W : W → ( k × ) ( μ , η ) 2 \pi_{1}|_{\mathcal{W}}:\mathcal{W}\to(\mathbb{k}^{\times})^{2}_{(\mu,\eta)} π 1 ∣ W : W → ( k × ) ( μ , η ) 2 is dominant and generically two-to-one.
(3) π 2 ∣ W : W → ( k × ) ( μ ~ , λ ~ ) 2 \pi_{2}|_{\mathcal{W}}:\mathcal{W}\to(\mathbb{k}^{\times})^{2}_{(\widetilde{\mu},\widetilde{\lambda})} π 2 ∣ W : W → ( k × ) ( μ , λ ) 2 is dominant and generically at most two-to-one.
(4) π ~ 2 ∣ W : W → ( k × ) ( μ ~ , η ~ ) 2 \widetilde{\pi}_{2}|_{\mathcal{W}}:\mathcal{W}\to(\mathbb{k}^{\times})^{2}_{(\widetilde{\mu},\widetilde{\eta})} π 2 ∣ W : W → ( k × ) ( μ , η ) 2 is also dominant.
Proof. (1) Over the Zariski-open dense subset U ′ \mathcal{U}' U ′ , P η , μ ( μ ~ ) = 0 P_{\eta,\mu}(\widetilde{\mu})=0 P η , μ ( μ ) = 0 has two distinct solutions. Moreover, the discriminant B 2 − 4 A 2 B^{2}-4A^{2} B 2 − 4 A 2 is not a square in k ( μ , η ) \mathbb{k}(\mu,\eta) k ( μ , η ) , and P η , μ P_{\eta,\mu} P η , μ is irreducible over k ( μ , η ) \mathbb{k}(\mu,\eta) k ( μ , η ) . Indeed, to see the non-squareness, the factor ( μ − 1 ) (\mu-1) ( μ − 1 ) has order 1. The other factors ( μ + 1 ) (\mu+1) ( μ + 1 ) , ( μ η 2 − 1 ) (\mu\eta^{2}-1) ( μ η 2 − 1 ) , ( μ 2 η 2 − μ η 2 + 4 μ η − μ + 1 ) (\mu^{2}\eta^{2}-\mu\eta^{2}+4\mu\eta-\mu+1) ( μ 2 η 2 − μ η 2 + 4 μ η − μ + 1 ) do not contain any power of ( μ − 1 ) (\mu-1) ( μ − 1 ) , because when evaluated at μ = 1 \mu=1 μ = 1 , they equal 2 2 2 , ( η 2 − 1 ) (\eta^{2}-1) ( η 2 − 1 ) , 4 η 4\eta 4 η respectively, none of which is zero.
The coefficients of P η , μ ( μ ~ ) P_{\eta,\mu}(\widetilde{\mu}) P η , μ ( μ ) have no common factors, making it primitive. By Gauss’s lemma, P η , μ ( μ ~ ) P_{\eta,\mu}(\widetilde{\mu}) P η , μ ( μ ) is also irreducible over k [ μ , η ] \mathbb{k}[\mu,\eta] k [ μ , η ] . Once η , μ , μ ~ \eta,\mu,\widetilde{\mu} η , μ , μ are fixed, λ ~ \widetilde{\lambda} λ is uniquely determined as a rational polynomial. In particular, over the Zariski-open dense locus δ ≠ 0 \delta\ne0 δ = 0 , the variety is irreducible. The exceptional case, since we transformed the equation of λ ~ \widetilde{\lambda} λ into a polynomial, is that both the numerator and denominator equal zero, i.e. μ η μ ~ − μ − η + μ ~ = 0 \mu\eta\widetilde{\mu}-\mu-\eta+\widetilde{\mu}=0 μ η μ − μ − η + μ = 0 , μ η − μ μ ~ − η μ ~ + 1 = 0 \mu\eta-\mu\widetilde{\mu}-\eta\widetilde{\mu}+1=0 μ η − μ μ − η μ + 1 = 0 . The solutions are μ = μ ~ = 1 \mu=\widetilde{\mu}=1 μ = μ = 1 and η \eta η can be anything. However, this is excluded from U \mathcal{U} U , and certainly excluded from U ′ \mathcal{U}' U ′ . Altogether, we proved that W \mathcal{W} W is irreducible.
(2) Since U ′ \mathcal{U}' U ′ is open dense in ( k × ) ( μ , η ) 2 (\mathbb{k}^{\times})^{2}_{(\mu,\eta)} ( k × ) ( μ , η ) 2 , π 1 ∣ W \pi_{1}|_{\mathcal{W}} π 1 ∣ W is dominant. A generic fiber has two choices of μ ~ \widetilde{\mu} μ and a corresponding λ ~ \widetilde{\lambda} λ .
(3) To study the property of π 2 \pi_{2} π 2 , we need to express η , μ \eta,\mu η , μ in terms of μ ~ , λ ~ \widetilde{\mu},\widetilde{\lambda} μ , λ . First, with μ , μ ~ , λ ~ \mu,\widetilde{\mu},\widetilde{\lambda} μ , μ , λ fixed, η \eta η is uniquely determined by
η = μ μ ~ λ ~ + μ − μ ~ − λ ~ μ μ ~ + μ λ ~ − μ ~ λ ~ − 1 . \eta=\frac{\mu\widetilde{\mu}\widetilde{\lambda}+\mu-\widetilde{\mu}-\widetilde{\lambda}}{\mu\widetilde{\mu}+\mu\widetilde{\lambda}-\widetilde{\mu}\widetilde{\lambda}-1}. η = μ μ + μ λ − μ λ − 1 μ μ λ + μ − μ − λ .
Substituting in P η , μ ( μ ~ ) = 0 P_{\eta,\mu}(\widetilde{\mu})=0 P η , μ ( μ ) = 0 and clearing denominators, we get ( μ − 1 ) ( μ + 1 ) 2 Q μ ~ , λ ~ ( μ ) = 0 (\mu-1)(\mu+1)^{2}Q_{\widetilde{\mu},\widetilde{\lambda}}(\mu)=0 ( μ − 1 ) ( μ + 1 ) 2 Q μ , λ ( μ ) = 0 , where
Q μ ~ , λ ~ ( μ ) = μ ~ ( μ ~ λ ~ + 1 ) 2 ( μ 2 + 1 ) + ( − μ ~ 4 λ ~ 2 + μ ~ 4 λ ~ − 2 μ ~ 3 λ ~ − μ ~ 2 λ ~ 2 − 2 μ ~ 2 λ ~ − μ ~ 2 − 2 μ ~ λ ~ + λ ~ − 1 ) μ Q_{\widetilde{\mu},\widetilde{\lambda}}(\mu)=\widetilde{\mu}(\widetilde{\mu}\widetilde{\lambda}+1)^{2}(\mu^{2}+1)+(-\widetilde{\mu}^{4}\widetilde{\lambda}^{2}+\widetilde{\mu}^{4}\widetilde{\lambda}-2\widetilde{\mu}^{3}\widetilde{\lambda}-\widetilde{\mu}^{2}\widetilde{\lambda}^{2}-2\widetilde{\mu}^{2}\widetilde{\lambda}-\widetilde{\mu}^{2}-2\widetilde{\mu}\widetilde{\lambda}+\widetilde{\lambda}-1)\mu Q μ , λ ( μ ) = μ ( μ λ + 1 ) 2 ( μ 2 + 1 ) + ( − μ 4 λ 2 + μ 4 λ − 2 μ 3 λ − μ 2 λ 2 − 2 μ 2 λ − μ 2 − 2 μ λ + λ − 1 ) μ
For generic values of ( μ ~ , λ ~ ) (\widetilde{\mu},\widetilde{\lambda}) ( μ , λ ) , this is a quadratic polynomial. Moreover, since the quadratic and constant coefficients are equal, the two solutions multiply to 1. In particular, they are inverse to each other and non-zero. Therefore, π 2 ∣ W : W → ( k × ) ( μ ~ , λ ~ ) 2 \left.\pi_2\right|_{\mathcal{W}}:\mathcal{W}\to(\mathbb{k}^{\times})^2_{(\widetilde{\mu},\widetilde{\lambda})} π 2 ∣ W : W → ( k × ) ( μ , λ ) 2 is dominant and generically at most two-to-one.
(4) Since π ~ 2 \widetilde{\pi}_2 π 2 is the composite of π 2 \pi_2 π 2 with the finite dominant map ( μ ~ , λ ~ ) ↦ ( μ ~ , λ ~ 2 ) (\widetilde{\mu},\widetilde{\lambda})\mapsto(\widetilde{\mu},\widetilde{\lambda}^2) ( μ , λ ) ↦ ( μ , λ 2 ) , π ~ 2 \widetilde{\pi}_2 π 2 is also dominant.
We next identify the part of the correspondence on which the one-step construction is nondegenerate. Define
W s t e p : = { w = ( μ , η , μ ~ , λ ~ ) ∈ W : π 1 ( w ) ∈ U , δ ( μ , η , μ ~ ) ≠ 0 } . (20) \mathcal{W}_{\mathrm{step}}:=\left\{w=(\mu,\eta,\widetilde{\mu},\widetilde{\lambda})\in\mathcal{W}:\pi_1(w)\in\mathcal{U},\ \delta(\mu,\eta,\widetilde{\mu})\ne0\right\}. \tag*{(20)} W step := { w = ( μ , η , μ , λ ) ∈ W : π 1 ( w ) ∈ U , δ ( μ , η , μ ) = 0 } . (20)
Observe that over W s t e p \mathcal{W}_{\mathrm{step}} W step , we have μ ≠ − 1 \mu\ne-1 μ = − 1 . Indeed, if μ = − 1 \mu=-1 μ = − 1 , then η ≠ 1 \eta\ne1 η = 1 and the equation
P η , − 1 ( μ ~ ) = ( η − 1 ) 2 ( μ ~ + 1 ) 2 P_{\eta,-1}(\widetilde{\mu})=(\eta-1)^2(\widetilde{\mu}+1)^2 P η , − 1 ( μ ) = ( η − 1 ) 2 ( μ + 1 ) 2
forces μ ~ = − 1 \widetilde{\mu}=-1 μ = − 1 . But then δ ( − 1 , η , − 1 ) = 0 \delta(-1,\eta,-1)=0 δ ( − 1 , η , − 1 ) = 0 , contrary to the definition of W s t e p \mathcal{W}_{\mathrm{step}} W step .
We also note that over W s t e p \mathcal{W}_{\mathrm{step}} W step , given the input peripheral data ( μ , η ) (\mu,\eta) ( μ , η ) , the pairs ( a , d ) (a,d) ( a , d ) and ( μ ~ , λ ~ ) (\widetilde{\mu},\widetilde{\lambda}) ( μ , λ ) determine each other in a unique way. Namely, by Proposition 2.14, we have
μ ~ = a + d − 1 , λ ~ = η ( a − 1 ) + ( d − 1 ) ( a − 1 ) + η ( d − 1 ) . \widetilde{\mu}=a+d-1,\qquad\widetilde{\lambda}=\frac{\eta(a-1)+(d-1)}{(a-1)+\eta(d-1)}. μ = a + d − 1 , λ = ( a − 1 ) + η ( d − 1 ) η ( a − 1 ) + ( d − 1 ) .
And conversely, we can solve ( a , d ) (a,d) ( a , d ) as
a = μ ( μ ~ + 1 ) μ + 1 , d = μ ~ + 1 μ + 1 . a=\frac{\mu(\widetilde{\mu}+1)}{\mu+1},\qquad d=\frac{\widetilde{\mu}+1}{\mu+1}. a = μ + 1 μ ( μ + 1 ) , d = μ + 1 μ + 1 .
We now iterate the correspondence with the above mild restrictions. Define
G 0 : = ( k × ) 2 , G N + 1 : = π 1 ( W s t e p ∩ π ~ 2 − 1 ( G N ) ) . \mathcal{G}_0:=(\mathbb{k}^{\times})^2,\qquad\mathcal{G}_{N+1}:=\pi_1\left(\mathcal{W}_{\mathrm{step}}\cap\widetilde{\pi}_2^{-1}(\mathcal{G}_N)\right). G 0 := ( k × ) 2 , G N + 1 := π 1 ( W step ∩ π 2 − 1 ( G N ) ) .
It follows from the inductive construction that a point ( μ 0 , η 0 ) (\mu_0,\eta_0) ( μ 0 , η 0 ) lies in G N \mathcal{G}_N G N if and only if there exist w j ∈ W s t e p w_j\in\mathcal{W}_{\mathrm{step}} w j ∈ W step for 0 ≤ j < N 0\le j<N 0 ≤ j < N and { ( μ j , η j ) } 1 ≤ j ≤ N \{(\mu_j,\eta_j)\}_{1\le j\le N} {( μ j , η j ) } 1 ≤ j ≤ N such that
π 1 ( w j ) = ( μ j , η j ) , π ~ 2 ( w j ) = ( μ j + 1 , η j + 1 ) , 0 ≤ j < N . \pi_1(w_j)=(\mu_j,\eta_j),\qquad\widetilde{\pi}_2(w_j)=(\mu_{j+1},\eta_{j+1}),\qquad0\le j<N. π 1 ( w j ) = ( μ j , η j ) , π 2 ( w j ) = ( μ j + 1 , η j + 1 ) , 0 ≤ j < N .
Therefore, by truncation, we find that
G N + 1 ⊆ G N . \mathcal{G}_{N+1}\subseteq\mathcal{G}_N. G N + 1 ⊆ G N .
Moreover, given the data of ( μ j , η j ) (\mu_j,\eta_j) ( μ j , η j ) , we can solve ( a j , d j ) (a_j,d_j) ( a j , d j ) from ( μ j + 1 , η j + 1 ) (\mu_{j+1},\eta_{j+1}) ( μ j + 1 , η j + 1 ) . Hence, we can inductively define the length-N N N -sequence. This yields
G N ⊆ S N . (21) \mathcal{G}_N\subseteq\mathcal{S}_N. \tag*{(21)} G N ⊆ S N . (21)
Proposition 2.21. For every N ≥ 0 N\ge0 N ≥ 0 , the set G N \mathcal{G}_N G N is constructible and contains a Zariski-open dense subset of ( k × ) 2 (\mathbb{k}^{\times})^2 ( k × ) 2 .
Proof. Since W ~ ∩ π 1 − 1 ( U ′ ) ⊂ W s t e p \widetilde{\mathcal{W}}\cap\pi_1^{-1}(\mathcal{U}')\subset\mathcal{W}_{\mathrm{step}} W ∩ π 1 − 1 ( U ′ ) ⊂ W step , we have that W s t e p \mathcal{W}_{\mathrm{step}} W step is a Zariski-open dense subset of W \mathcal{W} W . We prove by induction. The base case N = 0 N=0 N = 0 is clear. Suppose G N \mathcal{G}_N G N is constructible and contains a dense open subset U N ⊂ ( k × ) 2 \mathcal{U}_N\subset(\mathbb{k}^{\times})^2 U N ⊂ ( k × ) 2 . The set W s t e p ∩ π ~ 2 − 1 ( G N ) \mathcal{W}_{\mathrm{step}}\cap\widetilde{\pi}_2^{-1}(\mathcal{G}_N) W step ∩ π 2 − 1 ( G N ) is constructible, and Chevalley’s theorem further shows that its image under π 1 \pi_1 π 1 , which is G N + 1 \mathcal{G}_{N+1} G N + 1 , is also constructible. By Proposition 2.20, π ~ 2 \widetilde{\pi}_2 π 2 is dominant and W \mathcal{W} W is irreducible. Therefore, W s t e p ∩ π ~ 2 − 1 ( U N ) ⊂ W \mathcal{W}_{\mathrm{step}}\cap\widetilde{\pi}_2^{-1}(\mathcal{U}_N)\subset\mathcal{W} W step ∩ π 2 − 1 ( U N ) ⊂ W is Zariski-open dense. Since π 1 \pi_1 π 1 is dominant, its image under π 1 \pi_1 π 1 is also Zariski-open dense. Therefore, G N + 1 \mathcal{G}_{N+1} G N + 1 is dense and constructible, which completes the induction. ∎
Proof of Proposition 2.19. Following Definition 2.15, C N C_N C N is defined by a finite collection of polynomials and non-vanishing conditions, hence it is locally closed and further constructible. By the Chevalley’s theorem, its projection S N = pr ( C N ) S_N = \operatorname{pr}(C_N) S N = pr ( C N ) is also constructible. By Proposition 2.21, the set G N G_N G N contains a Zariski-open dense subset. Since G N ⊂ S N G_N \subset S_N G N ⊂ S N , S N S_N S N also contains a Zariski-open dense subset.
Finally, we show that ⋂ N ∈ N S N \bigcap_{N\in\mathbb{N}}S_N ⋂ N ∈ N S N is Zariski dense. For each N N N , we choose a non-zero polynomial f N f_N f N such that its principal open subset U N : = { f N ≠ 0 } ∩ ( k × ) 2 ⊂ S N U_N := \{f_N \ne0\}\cap(\mathbb{k}^{\times})^2 \subset S_N U N := { f N = 0 } ∩ ( k × ) 2 ⊂ S N . We claim that ⋂ N U N \bigcap_N U_N ⋂ N U N is Zariski dense. It suffices to show that for any principal open subset { g ≠ 0 } \{g\ne0\} { g = 0 } , the intersection { f N ≠ 0 } ∩ ( ⋂ U N ) ≠ ∅ \{f_N\ne0\}\cap(\bigcap U_N)\ne\varnothing { f N = 0 } ∩ ( ⋂ U N ) = ∅ . Consider any element h ∈ { g , f 0 , f 1 , … } h\in\{g,f_0,f_1,\ldots\} h ∈ { g , f 0 , f 1 , … } . Let Z ( h ) ⊂ k × Z(h)\subset\mathbb{k}^{\times} Z ( h ) ⊂ k × be the subset of μ 0 \mu_0 μ 0 such that h ( μ 0 , η ) = 0 h(\mu_0,\eta)=0 h ( μ 0 , η ) = 0 with respect to η \eta η . Then Z ( h ) Z(h) Z ( h ) is finite, since for h = ∑ i > 0 h i ( μ ) η i h=\sum_{i>0}h_i(\mu)\eta^i h = ∑ i > 0 h i ( μ ) η i , each h i h_i h i has finitely many roots. Note that Z ( g ) ∪ ( ⋃ N Z ( f N ) ) Z(g)\cup(\bigcup_N Z(f_N)) Z ( g ) ∪ ( ⋃ N Z ( f N )) is countable. Since k = C \mathbb{k}=\mathbb{C} k = C is uncountable, we can choose μ 0 ∈ k × ∖ ( Z ( g ) ∪ ( ⋃ N Z ( f N ) ) ) \mu_0\in\mathbb{k}^{\times}\setminus(Z(g)\cup(\bigcup_N Z(f_N))) μ 0 ∈ k × ∖ ( Z ( g ) ∪ ( ⋃ N Z ( f N ))) , such that h ( μ 0 , η ) h(\mu_0,\eta) h ( μ 0 , η ) is non-zero. Iterate the argument for the second variable, we find η 0 \eta_0 η 0 such that neither g ( μ 0 , η 0 ) g(\mu_0,\eta_0) g ( μ 0 , η 0 ) nor any f N ( μ 0 , η 0 ) f_N(\mu_0,\eta_0) f N ( μ 0 , η 0 ) vanished. This completes the proof. □
Meridional bounds
We next compare the representations constructed in § 2.4 with meridional generating sets. The upper bound below does not require an augmentation or a condition on its peripheral data.
Proposition 2.22. For every knot K K K and every integer N ≥ 0 N\ge0 N ≥ 0 ,
mr ( Wh ( K ) ) ≤ 2 mr ( K ) , mr ( Wh N ( K ) ) ≤ 2 N mr ( K ) . \operatorname{mr}(\operatorname{Wh}(K))\le2\operatorname{mr}(K),\qquad\operatorname{mr}(\operatorname{Wh}^{N}(K))\le2^{N}\operatorname{mr}(K). mr ( Wh ( K )) ≤ 2 mr ( K ) , mr ( Wh N ( K )) ≤ 2 N mr ( K ) . Proof. Let r = mr ( K ) r=\operatorname{mr}(K) r = mr ( K ) . Choose meridional generators m i = h i m K h i − 1 m_i=h_i m_K h_i^{-1} m i = h i m K h i − 1 , 1 ≤ i ≤ r 1\le i\le r 1 ≤ i ≤ r , with h 1 = 1 h_1=1 h 1 = 1 . In the satellite presentation of Proposition 2.8, we have
m V = x 2 x 1 − 1 x 2 − 1 x 1 = m K , ℓ V = ℓ K , ℓ V x 2 ℓ V − 1 = x 1 . m_V=x_2x_1^{-1}x_2^{-1}x_1=m_K,\qquad\ell_V=\ell_K,\qquad\ell_Vx_2\ell_V^{-1}=x_1. m V = x 2 x 1 − 1 x 2 − 1 x 1 = m K , ℓ V = ℓ K , ℓ V x 2 ℓ V − 1 = x 1 . In particular, both x 1 x_1 x 1 and x 2 x_2 x 2 are meridians of Wh ( K ) \operatorname{Wh}(K) Wh ( K ) . Let H H H be the subgroup generated by the 2 r 2r 2 r meridians
u i = h i x 1 h i − 1 , v i = h i x 2 h i − 1 , 1 ≤ i ≤ r . u_i=h_i x_1h_i^{-1},\qquad v_i=h_i x_2h_i^{-1},\qquad1\le i\le r. u i = h i x 1 h i − 1 , v i = h i x 2 h i − 1 , 1 ≤ i ≤ r . Here elements of the companion group are understood through their images in the satellite group. For each i i i ,
v i u i − 1 v i − 1 u i = h i ( x 2 x 1 − 1 x 2 − 1 x 1 ) h i − 1 = m i . v_i u_i^{-1}v_i^{-1}u_i=h_i(x_2x_1^{-1}x_2^{-1}x_1)h_i^{-1}=m_i. v i u i − 1 v i − 1 u i = h i ( x 2 x 1 − 1 x 2 − 1 x 1 ) h i − 1 = m i . Thus H H H contains the image of the companion group, and hence ℓ V \ell_V ℓ V . It also contains x 1 = u 1 x_1=u_1 x 1 = u 1 and x 2 = v 1 x_2=v_1 x 2 = v 1 . These elements generate the pattern group. The satellite presentation therefore gives H = π Wh ( K ) H=\pi_{\operatorname{Wh}(K)} H = π Wh ( K ) , proving the first inequality. Induction gives the second. The same presentation argument applies when K K K is the unknot; it does not require the companion homomorphism to be injective. □
Together with Proposition 2.4 , this gives the unconditional bounds
ar ( Wh N ( K ) ) ≤ mr ( Wh N ( K ) ) ≤ 2 N mr ( K ) . (22) \operatorname{ar}(\operatorname{Wh}^{N}(K))\le\operatorname{mr}(\operatorname{Wh}^{N}(K))\le2^{N}\operatorname{mr}(K). \tag*{(22)} ar ( Wh N ( K )) ≤ mr ( Wh N ( K )) ≤ 2 N mr ( K ) . (22)
The next result gives a sufficient condition for equality. Recall the chain-admitting set S N S_N S N from Definition 2.15 and the paragraph following it.
Theorem 2.23. Let K K K be a knot and set r = mr ( K ) r=\operatorname{mr}(K) r = mr ( K ) . Suppose that K K K admits a rank-r r r augmentation ϵ \epsilon ϵ with ϵ ( μ ) ≠ 1 \epsilon(\mu)\ne1 ϵ ( μ ) = 1 and
( ϵ ( μ ) , ϵ ( λ ) ) 2 ∈ S N . \left(\epsilon(\mu),\epsilon(\lambda)\right)^2\in S_N. ( ϵ ( μ ) , ϵ ( λ ) ) 2 ∈ S N .
Then
ar ( Wh N ( K ) ) = mr ( Wh N ( K ) ) = 2 N mr ( K ) . \operatorname{ar}(\operatorname{Wh}^{N}(K))=\operatorname{mr}(\operatorname{Wh}^{N}(K))=2^{N}\operatorname{mr}(K). ar ( Wh N ( K )) = mr ( Wh N ( K )) = 2 N mr ( K ) .
Proof. The augmentation–representation correspondence recalled in § 2.1 gives an irreducible r r r -dimensional diagonalizable μ \mu μ rk-one representation with the prescribed peripheral data. Proposition 2.16 then gives ar ( Wh N ( K ) ) ≥ 2 N r \operatorname{ar}(\operatorname{Wh}^{N}(K)) \ge2^{N}r ar ( Wh N ( K )) ≥ 2 N r . Combining this with (22) proves the result. □
Remark 2.24. The condition on the peripheral point is essential. In § 6 , we show that the square knot S S S satisfies ar ( S ) = mr ( S ) = b ( S ) = 3 \operatorname{ar}(S)=\operatorname{mr}(S)=b(S)=3 ar ( S ) = mr ( S ) = b ( S ) = 3 , whereas ar ( Wh ( S ) ) ≤ 5 \operatorname{ar}(\operatorname{Wh}(S))\leq5 ar ( Wh ( S )) ≤ 5 . Thus maximal augmentation rank for the companion alone does not suffice.
Algebraic curves and character varieties
Theorem 2.23 reduces the meridional-rank problem to finding an augmentation whose peripheral point lies in S N S_N S N . To make this choice for every N N N , we use a curve of peripheral values. We first recall how a point at infinity of such a curve determines the orders of poles and zeros. We then record the character-variety constructions used to obtain the curve from a boundary slope. Throughout this section, the ground field is C \mathbb{C} C , and we use the algebraic conventions of § 2.4 .
Curves, ideal points, and valuations
For an irreducible algebraic set A A A , we denote its field of rational functions by C ( A ) \mathbb{C}(A) C ( A ) . A rational map is defined on a nonempty open subset, not necessarily on its whole source; two representatives define the same map if they agree on their common domain. A birational map restricts to an isomorphism between nonempty open subsets. We write Dom ( φ ) \operatorname{Dom}(\varphi) Dom ( φ ) for the largest open subset on which a rational map φ \varphi φ is defined.
Definition 3.1. Suppose C C C is an affine algebraic curve. A smooth projective model ( C ~ , ϕ ~ ) (\widetilde{C},\widetilde{\phi}) ( C , ϕ ) of C C C is a smooth projective curve C ~ \widetilde{C} C together with a birational map ϕ ~ : C ~ ⇢ C \widetilde{\phi}:\widetilde{C}\dashrightarrow C ϕ : C ⇢ C . A point P ∈ C ~ P\in\widetilde{C} P ∈ C not in the domain of ϕ ~ \widetilde{\phi} ϕ is called an ideal point of C C C .
Every affine algebraic curve has a smooth projective model [23, Chapter I, Corollary 6.11]. The birational map identifies C ( C ) \mathbb{C}(C) C ( C ) with C ( C ~ ) \mathbb{C}(\widetilde{C}) C ( C ) ; we use this identification for rational functions throughout. The following lemma is important to lift valuation-related results from the PSL 2 ( C ) \operatorname{PSL}_{2}(\mathbb{C}) PSL 2 ( C ) setup to the SL 2 ( C ) \operatorname{SL}_{2}(\mathbb{C}) SL 2 ( C ) setup.
Lemma 3.2. Suppose f : C → D f:C\to D f : C → D is a morphism between two affine algebraic curves so that D = im ( f ) D=\operatorname{im}(f) D = im ( f ) . Suppose ( C ~ , ϕ ~ C ) (\widetilde{C},\widetilde{\phi}_C) ( C , ϕ C ) and ( D ~ , ϕ ~ D ) (\widetilde{D},\widetilde{\phi}_D) ( D , ϕ D ) are the respective smooth projective models. Then there exists a unique surjective morphism f ~ : C ~ → D ~ \widetilde{f}:\widetilde{C}\to\widetilde{D} f : C → D such that as rational maps we have ϕ ~ D ∘ f ~ = f ∘ ϕ ~ C \widetilde{\phi}_D\circ\widetilde{f}=f\circ\widetilde{\phi}_C ϕ D ∘ f = f ∘ ϕ C .
Proof. The existence of a dominant morphism f ~ : C ~ → D ~ \widetilde{f}:\widetilde{C}\to\widetilde{D} f : C → D with the identity ϕ ~ D ∘ f ~ = f ∘ ϕ ~ C \widetilde{\phi}_D\circ\widetilde{f}=f\circ\widetilde{\phi}_C ϕ D ∘ f = f ∘ ϕ C is established in [23, Chapter I, Corollary 6.12]. The meaning of “dominant” is precisely what we assumed (D = im ( f ~ ) D=\operatorname{im}(\widetilde{f}) D = im ( f ) ). [19, Chapter 1, Theorem 4.100] says that C ~ \widetilde{C} C is complete, and hence by [19, Chapter 1, Theorem 4.98] the image of f ~ \widetilde{f} f is closed, and must be all of D ~ \widetilde{D} D . □
Let P ∈ C ~ P\in\widetilde{C} P ∈ C . Denote by O C ~ , P \mathcal{O}_{\widetilde{C},P} O C , P the local ring of rational functions regular near P P P , and by m C ~ , P \mathfrak{m}_{\widetilde{C},P} m C , P its maximal ideal of functions vanishing at P P P . Since C ~ \widetilde{C} C is a smooth curve, this local ring is a discrete valuation ring [23, Chapter I, Theorems 5.1 and 6.2A]. In particular, m C ~ , P \mathfrak{m}_{\widetilde{C},P} m C , P has a generator s s s , and every nonzero f ∈ C ( C ~ ) f \in\mathbb{C}(\widetilde{C}) f ∈ C ( C ) has a unique expression f = s k h f=s^{k}h f = s k h with k ∈ Z k \in\mathbb{Z} k ∈ Z and h h h a unit in O C ~ , P \mathcal{O}_{\widetilde{C},P} O C , P . The integer k k k is independent of the choice of s s s .
Definition 3.3. A generator s s s of m C ~ , P \mathfrak{m}_{\widetilde{C},P} m C , P is a local parameter at P P P . For f = s k h f=s^{k}h f = s k h as above, its valuation at P P P is ν P ( f ) = k \nu_{P}(f)=k ν P ( f ) = k , and its pole order is
pole P ( f ) = max { 0 , − ν P ( f ) } . \operatorname{pole}_{P}(f)=\max\{0,-\nu_{P}(f)\}. pole P ( f ) = max { 0 , − ν P ( f )} . We set ν P ( 0 ) = + ∞ \nu_{P}(0)=+\infty ν P ( 0 ) = + ∞ . Thus zeros have positive valuation and poles have negative valuation.
The next lemma explains why the inequalities between valuations used later are preserved when we pass to a covering curve.
Lemma 3.4. Suppose ψ : C ~ → C ~ ′ \psi:\widetilde{C}\to\widetilde{C}' ψ : C → C ′ is a surjective morphism between two smooth projective curves. Let P ∈ C ~ P\in\widetilde{C} P ∈ C and Q = ψ ( P ) Q=\psi(P) Q = ψ ( P ) . Let s Q s_{Q} s Q be the local parameter at Q Q Q as in Definition 3.3 . Write
e ( P / Q ) = ν P ( ψ ∗ ( s Q ) ) . e(P/Q)=\nu_{P}\left(\psi^{*}(s_{Q})\right). e ( P / Q ) = ν P ( ψ ∗ ( s Q ) ) . Then the following statements are true:
The value e ( P / Q ) e(P/Q) e ( P / Q ) is a positive integer and independent of the choice of the local parameter s Q s_{Q} s Q .
For any nonzero rational function f f f on C ~ ′ \widetilde{C}' C ′ , we have
ν P ( ψ ∗ f ) = e ( P / Q ) ν Q ( f ) . \nu_{P}\left(\psi^{*}f\right)=e(P/Q)\nu_{Q}(f). ν P ( ψ ∗ f ) = e ( P / Q ) ν Q ( f ) . Consequently, we have
pole P ( ψ ∗ f ) = e ( P / Q ) pole Q ( f ) . \operatorname{pole}_{P}\left(\psi^{*}f\right)=e(P/Q)\operatorname{pole}_{Q}(f). pole P ( ψ ∗ f ) = e ( P / Q ) pole Q ( f ) . Proof. Since by construction s Q ( Q ) = 0 s_{Q}(Q)=0 s Q ( Q ) = 0 , we have ( ψ ∗ s Q ) ( P ) = 0 (\psi^{*}s_{Q})(P)=0 ( ψ ∗ s Q ) ( P ) = 0 . [23, Chapter I, Theorem 4.4] implies that ψ ∗ s Q \psi^{*}s_{Q} ψ ∗ s Q is not identically 0 0 0 , and hence e ( P / Q ) = ν P ( ψ ∗ ( s Q ) ) e(P/Q)=\nu_{P}(\psi^{*}(s_{Q})) e ( P / Q ) = ν P ( ψ ∗ ( s Q )) is a positive integer. It is clearly independent of the choice of s Q s_{Q} s Q : two different choices differ by the multiplication of a function nonzero at Q Q Q whose pullback is nonzero at P P P , so the valuation does not change. For any nonzero rational function f f f , write f = s Q k ⋅ u f=s_{Q}^{k}\cdot u f = s Q k ⋅ u . Then we have
ψ ∗ ( f ) = ( ψ ∗ s Q ) k ⋅ ( ψ ∗ u ) . \psi^{*}(f)=\left(\psi^{*}s_{Q}\right)^{k}\cdot\left(\psi^{*}u\right). ψ ∗ ( f ) = ( ψ ∗ s Q ) k ⋅ ( ψ ∗ u ) . The former has valuation k ⋅ e ( P / Q ) k\cdot e(P/Q) k ⋅ e ( P / Q ) and the latter is nonzero at P P P . Hence we are done. □
We also need the analytic interpretation of a local parameter. The convergence in the following statement is used in § 4.1 .
Lemma 3.5. [33 Chapter 2, Sections 2.1–2.3] and [34, Chapter 8, Section 1.1] . Let C ~ \widetilde{C} C be a smooth projective curve over C \mathbb{C} C , let P ∈ C ~ P\in\widetilde{C} P ∈ C , and let s s s be a local parameter at P P P . In the complex topology, there are a neighborhood V V V of P P P and ε > 0 \varepsilon>0 ε > 0 such that
s ∣ V : V ⟶ D ε : = { z ∈ C : ∣ z ∣ < ε } s|_{V}:V\longrightarrow D_{\varepsilon}:=\{z\in\mathbb{C}:|z|<\varepsilon\} s ∣ V : V ⟶ D ε := { z ∈ C : ∣ z ∣ < ε } is biholomorphic and sends P P P to 0 0 0 . We may take V V V inside a Zariski-open neighborhood on which s s s is regular. For every nonzero f ∈ C ( C ~ ) f\in\mathbb{C}(\widetilde{C}) f ∈ C ( C ) , after possibly shrinking this disc,
f ∘ ( s ∣ V ) − 1 ( z ) = ∑ j = k ∞ c j z j , k = ν P ( f ) , c k ≠ 0 , f\circ(s|_{V})^{-1}(z)=\sum_{j=k}^{\infty}c_{j}z^{j},\qquad k=\nu_{P}(f),\quad c_{k}\ne0, f ∘ ( s ∣ V ) − 1 ( z ) = j = k ∑ ∞ c j z j , k = ν P ( f ) , c k = 0 , where the series converges on the punctured disc. The amount of shrinking may depend on $f.
Indeed, smooth complex algebraic curves are locally complex manifolds, and algebraic local parameters give analytic coordinates, as explained in [33 Chapter 2, Section 2.3] . Regular functions are holomorphic in these coordinates. Writing f = s k h f=s^k h f = s k h and expanding the holomorphic unit h h h gives the asserted Laurent expansion. For any fixed finite collection of rational functions, one disc may be chosen for all of them by shrinking it once.
Let P P P now be an ideal point of C C C on its smooth projective model. The birational map ϕ ~ \widetilde{\phi} ϕ is regular and injective on a nonempty Zariski-open subset of C ~ \widetilde{C} C , whose complement is finite [33 Chapter I, Example 1.32 and Theorem 1.19] . We can therefore choose V V V above so that ϕ ~ \widetilde{\phi} ϕ is regular and injective on V ∖ { P } V \setminus\{P\} V ∖ { P } .
Definition 3.6. Suppose C C C is an affine algebraic curve, and ( C ~ , ϕ ~ ) (\widetilde{C},\widetilde{\phi}) ( C , ϕ ) is its smooth projective model. A Puiseux branch is the choice of an ideal point P ∈ C ~ P \in\widetilde{C} P ∈ C and a local parameter s s s at P P P , together with suitable V V V and ε \varepsilon ε supplied by the above lemma such that V V V contains only one ideal point, namely P P P , and ϕ ~ \widetilde{\phi} ϕ is injective on V ∖ { P } V\setminus\{P\} V ∖ { P } . Then we obtain the Puiseux parametrization
ϕ ~ ∘ ( s ∣ V ∖ { P } ) − 1 : D ε ∖ { 0 } → C . \widetilde{\phi}\circ(s|_{V\setminus\{P\}})^{-1}:D_{\varepsilon}\setminus\{0\}\to C. ϕ ∘ ( s ∣ V ∖ { P } ) − 1 : D ε ∖ { 0 } → C . Through s ∣ V : V → D ε s|_V:V\to D_{\varepsilon} s ∣ V : V → D ε , we may treat s s s as the coordinate on D ε D_{\varepsilon} D ε . We treat the restriction
ϕ ~ ∘ ( s ∣ V ∖ { P } ) − 1 ∣ D ε ′ ∖ { 0 } : D ε ′ ∖ { 0 } → C \widetilde{\phi}\circ(s|_{V\setminus\{P\}})^{-1}|_{D_{\varepsilon'}\setminus\{0\}}:D_{\varepsilon'}\setminus\{0\}\to C ϕ ∘ ( s ∣ V ∖ { P } ) − 1 ∣ D ε ′ ∖ { 0 } : D ε ′ ∖ { 0 } → C for 0 < ε ′ < ε 0<\varepsilon'<\varepsilon 0 < ε ′ < ε as the same Puiseux parametrization as the original.
This parametrization is injective and holomorphic. In an affine embedding of C C C , its coordinate functions have the convergent Laurent expansions of Lemma 3.5. A proper closed subset of an irreducible curve is finite. An infinite subset is therefore dense. If it is constructible, it contains a nonempty open subset by the fact recalled in § 2.4 . We use this fact to pass from branches to open subsets in § 4.1 .
Character varieties and peripheral trace functions
For a knot K ⊂ S 3 K\subset S^3 K ⊂ S 3 , define
R S L 2 ( C ) ( K ) = Hom ( π K , S L 2 ( C ) ) . R_{\mathrm{SL}_2(\mathbb{C})}(K)=\operatorname{Hom}(\pi_K,\mathrm{SL}_2(\mathbb{C})). R SL 2 ( C ) ( K ) = Hom ( π K , SL 2 ( C )) .
A finite presentation of π K \pi_K π K realizes this as an affine algebraic set: the matrices assigned to the generators satisfy the determinant-one conditions and the equations given by the relators. Inversion is polynomial on S L 2 ( C ) \mathrm{SL}_2(\mathbb{C}) SL 2 ( C ) .
Definition 3.7. The character of ρ ∈ R S L 2 ( C ) ( K ) \rho\in R_{\mathrm{SL}_2(\mathbb{C})}(K) ρ ∈ R SL 2 ( C ) ( K ) is the function χ ρ ( g ) = tr ( ρ ( g ) ) \chi_\rho(g)=\operatorname{tr}(\rho(g)) χ ρ ( g ) = tr ( ρ ( g )) on π K \pi_K π K . For g ∈ π K g\in\pi_K g ∈ π K , the associated trace function is t g ( ρ ) = tr ( ρ ( g ) ) t_g(\rho)=\operatorname{tr}(\rho(g)) t g ( ρ ) = tr ( ρ ( g )) on the representation set. We denote by T S L 2 ( C ) ( K ) T_{\mathrm{SL}_2(\mathbb{C})}(K) T SL 2 ( C ) ( K ) the subalgebra of C [ R S L 2 ( C ) ( K ) ] \mathbb{C}[R_{\mathrm{SL}_2(\mathbb{C})}(K)] C [ R SL 2 ( C ) ( K )] generated by these functions.
Note it is straightforward to check that all t g t_g t g are regular functions. We now construct the character variety X S L 2 ( C ) ( K ) X_{\mathrm{SL}_2(\mathbb{C})}(K) X SL 2 ( C ) ( K ) of K K K . Observe that the group S L 2 ( C ) \mathrm{SL}_2(\mathbb{C}) SL 2 ( C ) acts on R S L 2 ( C ) ( K ) R_{\mathrm{SL}_2(\mathbb{C})}(K) R SL 2 ( C ) ( K ) by conjugation: for an element h ∈ S L 2 ( C ) h\in\mathrm{SL}_2(\mathbb{C}) h ∈ SL 2 ( C ) and ρ ∈ R S L 2 ( C ) ( K ) \rho\in R_{\mathrm{SL}_2(\mathbb{C})}(K) ρ ∈ R SL 2 ( C ) ( K ) , define
( h ⋅ ρ ) ( g ) = h ⋅ ρ ( g ) ⋅ h − 1 , ∀ g ∈ π K (h\cdot\rho)(g)=h\cdot\rho(g)\cdot h^{-1},\quad\forall g\in\pi_K ( h ⋅ ρ ) ( g ) = h ⋅ ρ ( g ) ⋅ h − 1 , ∀ g ∈ π K
Since S L 2 ( C ) \mathrm{SL}_2(\mathbb{C}) SL 2 ( C ) is reductive,2 it follows from [13, Theorem 1.24]3 that C [ R S L 2 ( C ) ( K ) ] S L 2 ( C ) \mathbb{C}[R_{\mathrm{SL}_2(\mathbb{C})}(K)]^{\mathrm{SL}_2(\mathbb{C})} C [ R SL 2 ( C ) ( K ) ] SL 2 ( C ) is finitely generated and defines an affine quotient with a surjective regular quotient map.
Definition 3.8. The S L 2 ( C ) \mathrm{SL}_2(\mathbb{C}) SL 2 ( C ) character variety of K K K is the affine quotient
X S L 2 ( C ) ( K ) = R S L 2 ( C ) ( K ) / / S L 2 ( C ) . X_{\mathrm{SL}_2(\mathbb{C})}(K)=R_{\mathrm{SL}_2(\mathbb{C})}(K)//\mathrm{SL}_2(\mathbb{C}). X SL 2 ( C ) ( K ) = R SL 2 ( C ) ( K ) // SL 2 ( C ) . Its quotient map τ \tau τ identifies its coordinate ring with the invariant ring:
τ ∗ : C [ X S L 2 ( C ) ( K ) ] ⟶ ≅ C [ R S L 2 ( C ) ( K ) ] S L 2 ( C ) . \tau^*: \mathbb{C}[X_{\mathrm{SL}_2(\mathbb{C})}(K)] \overset{\cong}{\longrightarrow} \mathbb{C}[R_{\mathrm{SL}_2(\mathbb{C})}(K)]^{\mathrm{{SL}}_2(\mathbb{C})}. τ ∗ : C [ X SL 2 ( C ) ( K )] ⟶ ≅ C [ R SL 2 ( C ) ( K ) ] SL 2 ( C ) . The term character variety does not require this affine algebraic set to be irreducible.
Lemma 3.9 (Sikora [35 Proposition 2] ). Suppose K ⊂ S 3 K \subset S^3 K ⊂ S 3 is a knot. Then we have
C [ X S L 2 ( C ) ( K ) ] ≅ C [ R S L 2 ( C ) ( K ) ] S L 2 ( C ) = T S L 2 ( C ) ( K ) \mathbb{C}[X_{\mathrm{SL}_2(\mathbb{C})}(K)] \cong\mathbb{C}[R_{\mathrm{SL}_2(\mathbb{C})}(K)]^{\mathrm{SL}_2(\mathbb{C})}=T_{\mathrm{SL}_2(\mathbb{C})}(K) C [ X SL 2 ( C ) ( K )] ≅ C [ R SL 2 ( C ) ( K ) ] SL 2 ( C ) = T SL 2 ( C ) ( K ) where T S L 2 ( C ) ( K ) T_{\mathrm{SL}_2(\mathbb{C})}(K) T SL 2 ( C ) ( K ) is the trace algebra as in Definition 3.7 .
Lemma 3.11. Suppose K ⊂ S 3 K \subset S^3 K ⊂ S 3 is a knot. Then there is a canonical bijection between the set X S L 2 ( C ) ( K ) X_{\mathrm{SL}_2(\mathbb{C})}(K) X SL 2 ( C ) ( K ) and the set of all characters { χ ρ } ρ ∈ R S L 2 ( C ) ( K ) \{\chi_\rho\}_{\rho\in R_{\mathrm{SL}_2(\mathbb{C})}(K)} { χ ρ } ρ ∈ R SL 2 ( C ) ( K ) .
Proof. Since τ \tau τ is surjective, every point is τ ( ρ ) \tau(\rho) τ ( ρ ) for some representation ρ \rho ρ . Two representations have the same image under τ \tau τ precisely when all invariant regular functions agree on them. By Lemma 3.9 , this is equivalent to the equality of all their traces, that is, to equality of their characters. □
With the above lemma at hand, we may write an element of the character variety X S L 2 ( C ) ( K ) X_{\mathrm{SL}_2(\mathbb{C})}(K) X SL 2 ( C ) ( K ) as χ ρ \chi_\rho χ ρ . In particular, we have the following definition.
Definition 3.12. Suppose K ⊂ S 3 K \subset S^3 K ⊂ S 3 is a knot. Use the fixed peripheral system ( m K , ℓ K ) (m_K,\ell_K) ( m K , ℓ K ) with the preferred zero-linking longitude. We regard m K m_K m K and ℓ K \ell_K ℓ K as commuting elements in the knot group π K \pi_K π K . For ( p , q ) ∈ Z 2 (p,q)\in\mathbb{Z}^2 ( p , q ) ∈ Z 2 , we define
I p , q = ( t m K p ℓ K q ) 2 − 4. I_{p,q}=\left(t_{m_K^p\ell_K^q}\right)^2-4. I p , q = ( t m K p ℓ K q ) 2 − 4. Note the trace functions are regular functions on X S L 2 ( C ) ( K ) X_{\mathrm{SL}_2(\mathbb{C})}(K) X SL 2 ( C ) ( K ) and so is I p , q I_{p,q} I p , q .
PSL2 ( C ) _2(\mathbb{C}) 2 ( C ) character variety
By definition P S L 2 ( C ) = S L 2 ( C ) / { ± 1 } \mathrm{PSL}_2(\mathbb{C})=\mathrm{SL}_2(\mathbb{C})/\{\pm1\} PSL 2 ( C ) = SL 2 ( C ) / { ± 1 } , and there is an algebraic group isomorphism P S L 2 ( C ) ≅ S O 3 ( C ) \mathrm{PSL}_2(\mathbb{C})\cong\mathrm{SO}_3(\mathbb{C}) PSL 2 ( C ) ≅ SO 3 ( C ) . This isomorphism provides us with the facts we will need to construct the character variety: P S L 2 ( C ) \mathrm{PSL}_2(\mathbb{C}) PSL 2 ( C ) is reductive and R P S L 2 ( C ) ( K ) R_{\mathrm{PSL}_2(\mathbb{C})}(K) R PSL 2 ( C ) ( K ) is indeed an affine algebraic set (cf. [25, Section 2]).
Definition 3.13. Suppose K ⊂ S 3 K \subset S^3 K ⊂ S 3 is a knot. Define
R P S L 2 ( C ) ( K ) = Hom ( π K , P S L 2 ( C ) ) . R_{\mathrm{PSL}_2(\mathbb{C})}(K)=\operatorname{Hom}(\pi_K,\mathrm{PSL}_2(\mathbb{C})). R PSL 2 ( C ) ( K ) = Hom ( π K , PSL 2 ( C )) . Define its P S L 2 ( C ) \mathrm{PSL}_2(\mathbb{C}) PSL 2 ( C ) character variety X P S L 2 ( C ) ( K ) X_{\mathrm{PSL}_2(\mathbb{C})}(K) X PSL 2 ( C ) ( K ) to be the affine algebraic set arising from a construction similar to that in § 3.2 , which has coordinate ring
C [ X P S L 2 ( C ) ( K ) ] ≅ C [ R P S L 2 ( C ) ( K ) ] P S L 2 ( C ) , \mathbb{C}[X_{\mathrm{PSL}_2(\mathbb{C})}(K)]\cong\mathbb{C}[R_{\mathrm{PSL}_2(\mathbb{C})}(K)]^{\mathrm{PSL}_2(\mathbb{C})}, C [ X PSL 2 ( C ) ( K )] ≅ C [ R PSL 2 ( C ) ( K ) ] PSL 2 ( C ) , and admits a surjective regular map τ ^ : R P S L 2 ( C ) ( K ) → X P S L 2 ( C ) ( K ) \hat{\tau}:R_{\mathrm{PSL}_2(\mathbb{C})}(K)\to X_{\mathrm{PSL}_2(\mathbb{C})}(K) τ ^ : R PSL 2 ( C ) ( K ) → X PSL 2 ( C ) ( K ) .
Definition 3.14. Suppose K ⊂ S 3 K \subset S^3 K ⊂ S 3 is a knot. For ρ ∈ R P S L 2 ( C ) ( K ) \rho\in R_{\mathrm{PSL}_2(\mathbb{C})}(K) ρ ∈ R PSL 2 ( C ) ( K ) define the character associated to it as
χ ρ : π K → C \chi_\rho:\pi_K \to\mathbb{C} χ ρ : π K → C where χ ρ ( g ) = tr ( ρ ( g ) ) 2 \chi_\rho(g)=\operatorname{tr}(\rho(g))^2 χ ρ ( g ) = tr ( ρ ( g ) ) 2 . Note tr ( ρ ( g ) ) \operatorname{tr}(\rho(g)) tr ( ρ ( g )) itself is not defined. More precisely, we write tr ( ρ ( g ) ) 2 = tr ( A ) 2 \operatorname{tr}(\rho(g))^2=\operatorname{tr}(A)^2 tr ( ρ ( g ) ) 2 = tr ( A ) 2 where A A A is a lift of ρ ( g ) \rho(g) ρ ( g ) in S L 2 ( C ) \mathrm{SL}_2(\mathbb{C}) SL 2 ( C ) . This lift is determined up to a sign but taking the square resolves the ambiguity. Similarly, for an element g ∈ π K g\in\pi_K g ∈ π K we define the square trace function associated to g g g as t ^ g ( ρ ) = tr ( ρ ( g ) ) 2 \hat{t}_g(\rho)=\operatorname{tr}(\rho(g))^2 t ^ g ( ρ ) = tr ( ρ ( g ) ) 2 . Clearly, t ^ g \hat{t}_g t ^ g are P S L 2 ( C ) \mathrm{PSL}_2(\mathbb{C}) PSL 2 ( C ) invariant and hence become regular functions on X P S L 2 ( C ) ( K ) X_{\mathrm{PSL}_2(\mathbb{C})}(K) X PSL 2 ( C ) ( K ) .
Lemma 3.15 (Heusener–Porti [25, Theorem 1.3 and Proposition 2.2]). The square trace functions t ^ g \hat{t}_g t ^ g for all g ∈ π K g\in\pi_K g ∈ π K generate C [ R P S L 2 ( C ) ( K ) ] P S L 2 ( C ) \mathbb{C}[R_{\mathrm{PSL}_2(\mathbb{C})}(K)]^{\mathrm{PSL}_2(\mathbb{C})} C [ R PSL 2 ( C ) ( K ) ] PSL 2 ( C ) . Consequently, there is a (canonical) bijection between the set of all characters and the set X P S L 2 ( C ) ( K ) X_{\mathrm{PSL}_2(\mathbb{C})}(K) X PSL 2 ( C ) ( K ) , which identifies χ ρ \chi_\rho χ ρ with τ ^ ( ρ ) \hat{\tau}(\rho) τ ^ ( ρ ) for any ρ ∈ R P S L 2 ( C ) ( K ) \rho\in R_{\mathrm{PSL}_2(\mathbb{C})}(K) ρ ∈ R PSL 2 ( C ) ( K ) .
For the comparison below, we use finite morphisms between affine algebraic sets, without requiring these sets to be irreducible. We need to align the terminology in this paper with that in [19]: (1) Affine algebraic sets as defined in § 2.4 are algebraic varieties in [19, Chapter 1, Observations 4.19 and 4.39]. (2) A regular map between two affine algebraic sets as in Section 2.4 is a morphism between algebraic varieties as in {citep}[19, Chapter 1, Definitions 3.28 and 4.37, Observation 3.29, and Corollary 3.38]. (3) Such a map is finite in the sense of [19, Definition 4.74] if the coordinate ring of its source is finitely generated as a module over the coordinate ring of its target via pullback; the global coordinate rings then provide the required trivial affine cover.
Lemma 3.16. Suppose K ⊂ S 3 K\subset S^3 K ⊂ S 3 is a knot. Then there is a surjective regular map (cf. § 2.4 ) ψ : X S L 2 ( C ) ( K ) → X P S L 2 ( C ) ( K ) \psi:X_{\mathrm{SL}_2(\mathbb{C})}(K)\to X_{\mathrm{PSL}_2(\mathbb{C})}(K) ψ : X SL 2 ( C ) ( K ) → X PSL 2 ( C ) ( K ) such that the coordinate ring C [ X S L 2 ( C ) ( K ) ] \mathbb{C}[X_{\mathrm{SL}_2(\mathbb{C})}(K)] C [ X SL 2 ( C ) ( K )] is a finitely generated C [ X P S L 2 ( C ) ( K ) ] \mathbb{C}[X_{\mathrm{PSL}_2(\mathbb{C})}(K)] C [ X PSL 2 ( C ) ( K )] -module via the pullback
ψ ∗ : C [ X P S L 2 ( C ) ( K ) ] → C [ X S L 2 ( C ) ( K ) ] . \psi^*:\mathbb{C}[X_{\mathrm{PSL}_2(\mathbb{C})}(K)]\to\mathbb{C}[X_{\mathrm{SL}_2(\mathbb{C})}(K)]. ψ ∗ : C [ X PSL 2 ( C ) ( K )] → C [ X SL 2 ( C ) ( K )] .
Proof. First note that for the knot group π K \pi_K π K , we have H 2 ( π K ; Z / 2 Z ) = 0 H^2(\pi_K;\mathbb{Z}/2\mathbb{Z})=0 H 2 ( π K ; Z /2 Z ) = 0 (cf. the discussion before [12, Lemma 3.3]), and hence any P S L 2 ( C ) \mathrm{PSL}_2(\mathbb{C}) PSL 2 ( C ) representation lifts to an S L 2 ( C ) \mathrm{SL}_2(\mathbb{C}) SL 2 ( C ) representation (cf. [25, Section 4]). Then Heusener–Porti [25, Proposition 4.2] supply the identification of X P S L 2 ( C ) ( K ) X_{\mathrm{PSL}_2(\mathbb{C})}(K) X PSL 2 ( C ) ( K ) with the GIT quotient
X P S L 2 ( C ) ( K ) ≅ X S L 2 ( C ) ( K ) / / H 1 ( π K ; Z / 2 Z ) . X_{\mathrm{PSL}_2(\mathbb{C})}(K)\cong X_{\mathrm{SL}_2(\mathbb{C})}(K)//H^1(\pi_K;\mathbb{Z}/2\mathbb{Z}). X PSL 2 ( C ) ( K ) ≅ X SL 2 ( C ) ( K ) // H 1 ( π K ; Z /2 Z ) .
Note that H 1 ( π K ; Z / 2 Z ) = Hom ( π K , Z / 2 Z ) = Z / 2 Z = { ± 1 } H^1(\pi_K;\mathbb{Z}/2\mathbb{Z})=\operatorname{Hom}(\pi_K,\mathbb{Z}/2\mathbb{Z})=\mathbb{Z}/2\mathbb{Z}=\{\pm1\} H 1 ( π K ; Z /2 Z ) = Hom ( π K , Z /2 Z ) = Z /2 Z = { ± 1 } . So we obtain a regular map
ψ : X S L 2 ( C ) ( K ) → X P S L 2 ( C ) ( K ) \psi:X_{\mathrm{SL}_2(\mathbb{C})}(K)\to X_{\mathrm{PSL}_2(\mathbb{C})}(K) ψ : X SL 2 ( C ) ( K ) → X PSL 2 ( C ) ( K )
whose pullback provides an identification
ψ ∗ : C [ X P S L 2 ( C ) ( K ) ] ≅ C [ X S L 2 ( C ) ( K ) ] { ± 1 } . \psi^*:\mathbb{C}[X_{\mathrm{PSL}_2(\mathbb{C})}(K)]\cong\mathbb{C}[X_{\mathrm{SL}_2(\mathbb{C})}(K)]^{\{\pm1\}}. ψ ∗ : C [ X PSL 2 ( C ) ( K )] ≅ C [ X SL 2 ( C ) ( K ) ] { ± 1 } .
Then C [ X S L 2 ( C ) ( K ) ] \mathbb{C}[X_{\mathrm{SL}_2(\mathbb{C})}(K)] C [ X SL 2 ( C ) ( K )] is indeed a finitely generated module over C [ X S L 2 ( C ) ( K ) ] { ± 1 } \mathbb{C}[X_{\mathrm{SL}_2(\mathbb{C})}(K)]^{\{\pm1\}} C [ X SL 2 ( C ) ( K ) ] { ± 1 } : let t g 1 , … , t g n t_{g_1},\ldots,t_{g_n} t g 1 , … , t g n be a set of S L 2 ( C ) \mathrm{SL}_2(\mathbb{C}) SL 2 ( C ) trace functions that generate C [ X S L 2 ( C ) ( K ) ] \mathbb{C}[X_{\mathrm{SL}_2(\mathbb{C})}(K)] C [ X SL 2 ( C ) ( K )] as a C \mathbb{C} C -algebra. Then we know t g i 2 ∈ C [ X S L 2 ( C ) ( K ) ] { ± 1 } t_{g_i}^2\in\mathbb{C}[X_{\mathrm{SL}_2(\mathbb{C})}(K)]^{\{\pm1\}} t g i 2 ∈ C [ X SL 2 ( C ) ( K ) ] { ± 1 } , and hence t g 1 ϵ 1 ⋯ t g n ϵ n t_{g_1}^{\epsilon_1}\cdots t_{g_n}^{\epsilon_n} t g 1 ϵ 1 ⋯ t g n ϵ n for all ϵ i ∈ { 0 , 1 } \epsilon_i\in\{0,1\} ϵ i ∈ { 0 , 1 } generate C [ X S L 2 ( C ) ( K ) ] \mathbb{C}[X_{\mathrm{SL}_2(\mathbb{C})}(K)] C [ X SL 2 ( C ) ( K )] as a C [ X S L 2 ( C ) ( K ) ] { ± 1 } \mathbb{C}[X_{\mathrm{SL}_2(\mathbb{C})}(K)]^{\{\pm1\}} C [ X SL 2 ( C ) ( K ) ] { ± 1 } -module. □ \square □
Remark 3.17. We now provide a concrete description of the map ψ \psi ψ constructed above. From Lemma 3.11 and Lemma 3.15, we can identify points in the character variety with characters. If ρ ∈ R S L 2 ( C ) ( K ) \rho\in R_{\mathrm{SL}_2(\mathbb{C})}(K) ρ ∈ R SL 2 ( C ) ( K ) is an S L 2 ( C ) \mathrm{SL}_2(\mathbb{C}) SL 2 ( C ) representation, it naturally induces a P S L 2 ( C ) \mathrm{PSL}_2(\mathbb{C}) PSL 2 ( C ) representation ρ ^ \widehat{\rho} ρ by composing with the natural quotient map SL 2 ( C ) → PSL 2 ( C ) \operatorname{SL}_{2}(\mathbb{C}) \to\operatorname{PSL}_{2}(\mathbb{C}) SL 2 ( C ) → PSL 2 ( C ) . Then in terms of characters, ψ ( χ ρ ) = χ ρ ^ \psi(\chi_{\rho})=\chi_{\widehat{\rho}} ψ ( χ ρ ) = χ ρ .
Corollary 3.18. Suppose D ⊂ X PSL 2 ( C ) ( K ) D \subset X_{\operatorname{PSL}_{2}(\mathbb{C})}(K) D ⊂ X PSL 2 ( C ) ( K ) is a one-dimensional irreducible Zariski-closed subset. Then there exists a one-dimensional irreducible Zariski-closed subset C ⊂ X SL 2 ( C ) ( K ) C \subset X_{\operatorname{SL}_{2}(\mathbb{C})}(K) C ⊂ X SL 2 ( C ) ( K ) such that
ψ ∣ C : C ⟶ D \left.\psi\right|_{C}: C \longrightarrow D ψ ∣ C : C ⟶ D is a surjective morphism.
Proof. Let C 1 , ⋯ , C n C_{1},\cdots,C_{n} C 1 , ⋯ , C n be the irreducible components of ψ − 1 ( D ) \psi^{-1}(D) ψ − 1 ( D ) (by [23, Chapter I, Corollary 1.6] there are finitely many). By Lemma 3.16 and the convention above, ψ \psi ψ is finite. Thus, we apply [19, Chapter 1, Theorem 4.93] and conclude that ψ \psi ψ is a closed map, so each ψ ( C i ) \psi(C_{i}) ψ ( C i ) is a closed subset of D D D . By the definition of irreducibility in Section 2.4, there exists an index i i i such that ψ ( C i ) = D \psi(C_{i})=D ψ ( C i ) = D . Hence take C = C i C=C_{i} C = C i . It remains to show that C C C is indeed a curve. By [19, Chapter 1, Theorem 4.93], ψ ∣ C \left.\psi\right|_{C} ψ ∣ C is a finite map. Then [19, Chapter 1, Lemma 4.95] implies that C C C must be a curve. □
Ohtsuki's result
In this section, we summarize key ingredients from Ohtsuki [30 ] . Let K K K be a 2-bridge knot, then its fundamental group has a presentation of the form
π K = ⟨ u , v ∣ w u = v w ⟩ \pi_{K}=\langle u,v\mid wu=vw\rangle π K = ⟨ u , v ∣ w u = v w ⟩
for a suitable word w w w in u u u and v v v . Ohtsuki defines R R R to be the dimension ≥ 4 \ge4 ≥ 4 components of the variety of all PSL 2 ( C ) \operatorname{PSL}_{2}(\mathbb{C}) PSL 2 ( C ) representations, and X X X to be the set of all conjugacy classes of elements in R R R . For a generic element [ ρ ] ∈ X [\rho]\in X [ ρ ] ∈ X , he defined
τ = 2 tr ( U ~ V ~ ) − ( tr U ~ ) 2 ( tr U ~ ) 2 − 4 and ξ = − 2 ( tr ( U ~ ) ) 2 − 4 . \tau=\frac{2\operatorname{tr}(\widetilde{U}\widetilde{V})-(\operatorname{tr}\widetilde{U})^{2}}{(\operatorname{tr}\widetilde{U})^{2}-4}\quad\text{and}\quad\xi=\frac{-2}{(\operatorname{tr}(\widetilde{U}))^{2}-4}. τ = ( tr U ) 2 − 4 2 tr ( U V ) − ( tr U ) 2 and ξ = ( tr ( U ) ) 2 − 4 − 2 .
Here we take U = ρ ( u ) U=\rho(u) U = ρ ( u ) , V = ρ ( v ) V=\rho(v) V = ρ ( v ) and SL 2 ( C ) \operatorname{SL}_{2}(\mathbb{C}) SL 2 ( C ) -lifts U ~ \widetilde{U} U and V ~ \widetilde{V} V which are conjugate to each other. Ohtsuki showed that these coordinates identify the set of generic irreducible points (i.e., conjugacy classes) of X X X with a Zariski-open subset C ( K ) C(K) C ( K ) of the one-dimensional algebraic set in C 2 \mathbb{C}^{2} C 2 cut out by z ( τ , ξ ) = 0 z(\tau,\xi)=0 z ( τ , ξ ) = 0 . The following lemma is implicit in his paper.
Lemma 3.20. For any element g ∈ π K g\in\pi_{K} g ∈ π K , the map t ^ g : C ( K ) → C \widehat{t}_{g}:C(K)\to\mathbb{C} t g : C ( K ) → C defined by
t ^ g ( [ ρ ] ) = ( tr ( ρ ( g ) ) ) 2 \widehat{t}_{g}([\rho])=(\operatorname{tr}(\rho(g)))^{2} t g ([ ρ ]) = ( tr ( ρ ( g )) ) 2 is regular on C ( K ) C(K) C ( K ) .
The above lemma is not explicitly proved, but is essentially a corollary to [30 Lemma 1.2] with proof similar to that of [30 Lemma 4.9] . As a corollary, Ohtsuki stated that functions
I p , q O h ( [ ρ ] ) = ( tr ( ρ ( m K p ℓ K q ) ) ) 2 − 4. I^{O h}_{p,q}([\rho])=\left(\operatorname{tr}(\rho(m_{K}^{p}\ell_{K}^{q}))\right)^{2}-4. I p , q O h ([ ρ ]) = ( tr ( ρ ( m K p ℓ K q )) ) 2 − 4.
for all ( p , q ) ∈ Z 2 (p,q)\in\mathbb{Z}^{2} ( p , q ) ∈ Z 2 are all regular on C ( K ) C(K) C ( K ) . The key ingredient is the following.
Lemma 3.21. Suppose K ⊂ S 3 K \subset S^3 K ⊂ S 3 is a 2-bridge knot and r ≠ 0 r \ne0 r = 0 is the boundary slope of a non-closed oriented essential (i.e., incompressible, boundary-incompressible, and non-boundary-parallel) properly embedded orientable surface in the knot complement, then there exists an irreducible component X 0 X_0 X 0 of C ( K ) \mathcal{C}(K) C ( K ) , a smooth projective model X ~ 0 \widetilde{X}_0 X 0 with a point Q ∈ X ~ 0 Q \in\widetilde{X}_0 Q ∈ X 0 , and ϵ 0 ∈ { 1 , 2 } \epsilon_0 \in\{1,2\} ϵ 0 ∈ { 1 , 2 } such that for any ( p , q ) ∈ Z 2 (p,q) \in\mathbb{Z}^2 ( p , q ) ∈ Z 2 , we have
pole Q ( I p , q O h ∣ X 0 ) = ϵ 0 ⋅ ∣ p − r q ∣ . \operatorname{pole}_Q\left(I^{O_h}_{p,q}\big|_{X_0}\right)=\epsilon_0\cdot\lvert p-rq\rvert. pole Q ( I p , q O h X 0 ) = ϵ 0 ⋅ ∣ p − r q ∣ .
Proof. By Hatcher–Thurston’s classification (cf. [24, Theorem 1]), as recalled in [30, Section 2], there is a continued fraction $[n_j]$, with $\lvert n_j\rvert \ge 2$ for every $j$, such that $r=N_{[n_j]}$ (cf. [30 Section 2] ` for definition). The unique continued fraction expansion [ n j ] [n_j] [ n j ] whose coefficients are all even has corresponding N [ n j ] = 0 N_{[n_j]}=0 N [ n j ] = 0 . Since r ≠ 0 r\ne0 r = 0 , in an expansion [ n j ] [n_j] [ n j ] such that r = N [ n j ] r=N_{[n_j]} r = N [ n j ] , some n j n_j n j is odd. Consequently, choosing each residue k j ∈ Z / ( n j ) k_j\in\mathbb{Z}/(n_j) k j ∈ Z / ( n j ) to be the class of 1 gives a tuple allowed by Ohtsuki’s Theorem 5.1. Thus we can apply that theorem and the pole-order formula immediately preceding Proposition 5.2 and take X 0 X_0 X 0 , Q Q Q and ϵ 0 ∈ { 1 , 2 } \epsilon_0\in\{1,2\} ϵ 0 ∈ { 1 , 2 } as desired. □
Two-bridge knots and their Whitehead doubles
Whitehead-iterable branches
For every finite N N N , the set S N S_N S N of Section 2.4 contains a dense open subset of the peripheral torus. We now give a condition ensuring that a curve of actual augmentation points meets these sets. The condition is expressed in terms of valuations on a smooth projective model, using the terminology of the preceding section.
We work over C \mathbb{C} C and write
T = ( C ∗ ) ( μ , η ) 2 , η = λ 2 . T=(\mathbb{C}^*)^2_{(\mu,\eta)},\qquad\eta=\lambda^2. T = ( C ∗ ) ( μ , η ) 2 , η = λ 2 .
We use S N S_N S N and G N G_N G N for the sets already defined in Section 2.4; in particular, G N ⊆ S N G_N\subseteq S_N G N ⊆ S N .
Definition 4.1. Let C ⊂ T \mathcal{C}\subset T C ⊂ T be an irreducible algebraic curve. We say that C \mathcal{C} C has a Whitehead-iterable branch (and iterable branch for short), if C \mathcal{C} C admits an ideal point P P P with ν P ( μ ) = − p \nu_P(\mu)=-p ν P ( μ ) = − p and ν P ( η ) = q \nu_P(\eta)=q ν P ( η ) = q for ( p , q ) ∈ Z 2 (p,q)\in\mathbb{Z}^2 ( p , q ) ∈ Z 2 such that
p > 0 and (either q > p or q < 0 ) . (23) p>0\quad\text{and}\quad\text{(either }q>p\text{ or }q<0\text{)}. \tag*{(23)} p > 0 and (either q > p or q < 0 ) . (23)
Now for each point ( μ , η ) (\mu,\eta) ( μ , η ) , we have one-step correspondence ( μ ~ , λ ~ ) (\widetilde{\mu},\widetilde{\lambda}) ( μ , λ ) as in (18) . Since μ ~ \widetilde{\mu} μ is defined via a quadratic equation, in general it is not a rational function on the original C \mathcal{C} C , so the valuation ν \nu ν is not directly applicable for μ ~ \widetilde{\mu} μ . There are two routes to settle this issue: (1) Take a suitable (branched) double cover of C \mathcal{C} C so that ( μ ~ , λ ~ ) (\widetilde{\mu},\widetilde{\lambda}) ( μ , λ ) are well defined rational functions, and work with valuations of them, and (2) switch to use complex analytical topology. The two routes will lead to equivalent arguments, and we adopt (2).
As in Definition 3.6, in a small enough complex-analytical neighborhood of the ideal point P P P of C \mathcal{C} C , we have a local Puiseux parameter s s s . Thus we introduce the following notation for a local (near P P P ) meromorphic function f f f with at most a pole at s = 0 s=0 s = 0 : one can express such an f f f that is not identically zero as
f = s r ⋅ h ( s ) f=s^r\cdot h(s) f = s r ⋅ h ( s )
where r ∈ Z r\in\mathbb{Z} r ∈ Z and h h h is holomorphic with h ( 0 ) ≠ 0 h(0)\ne0 h ( 0 ) = 0 . Then we write o ( f ) = r \mathfrak{o}(f)=r o ( f ) = r . Also define o ( 0 ) = + ∞ \mathfrak{o}(0)=+\infty o ( 0 ) = + ∞ . Note μ \mu μ and η \eta η restrict to meromorphic functions near s = 0 s=0 s = 0 , and the above valuation, via Lemma 3.5, can be restated as
o ( μ ) = − p and o ( η ) = q . \mathfrak{o}(\mu)=-p\quad\text{and}\quad\mathfrak{o}(\eta)=q. o ( μ ) = − p and o ( η ) = q .
Furthermore, we have the following properties for o \mathfrak{o} o :
o ( c ) = 0 \mathfrak{o}(c)=0 o ( c ) = 0 for c ∈ C ∗ c\in\mathbb{C}^{*} c ∈ C ∗ .
o ( f ⋅ g ) = o ( f ) + o ( g ) \mathfrak{o}(f\cdot g)=\mathfrak{o}(f)+\mathfrak{o}(g) o ( f ⋅ g ) = o ( f ) + o ( g ) .
o ( f + g ) ≥ min { o ( f ) , o ( g ) } \mathfrak{o}(f+g)\geq\min\{\mathfrak{o}(f),\mathfrak{o}(g)\} o ( f + g ) ≥ min { o ( f ) , o ( g )} , with equality whenever o ( f ) ≠ o ( g ) \mathfrak{o}(f)\ne\mathfrak{o}(g) o ( f ) = o ( g ) .
Theorem 4.2. Let C ⊂ T C\subset T C ⊂ T be an irreducible algebraic curve with an iterable branch. For every integer N ≥ 0 N\geq0 N ≥ 0 , the intersection C ∩ S N C\cap S_N C ∩ S N (cf. § 2.4 ) contains a nonempty Zariski-open subset of C ~ \widetilde{C} C .
Proof. Let P ∈ C ~ P\in\widetilde{C} P ∈ C be the ideal point specified by the iterable branch, and let u u u be a local parameter at P P P . On a sufficiently small parameter, write η ( u ) = u q h ( u ) 2 \eta(u)=u^qh(u)^2 η ( u ) = u q h ( u ) 2 , where h h h is holomorphic near 0 0 0 and h ( 0 ) ≠ 0 h(0)\ne0 h ( 0 ) = 0 . On the covering disk given by u = s 2 u=s^2 u = s 2 , define
μ 0 ( s ) : = μ ( s 2 ) , λ 0 ( s ) : = s q h ( s 2 ) , η 0 ( s ) : = λ 0 ( s ) 2 = η ( s 2 ) . \mu_0(s):=\mu(s^2),\qquad\lambda_0(s):=s^qh(s^2),\qquad\eta_0(s):=\lambda_0(s)^2=\eta(s^2). μ 0 ( s ) := μ ( s 2 ) , λ 0 ( s ) := s q h ( s 2 ) , η 0 ( s ) := λ 0 ( s ) 2 = η ( s 2 ) .
Throughout this proof, o \mathfrak{o} o denotes the order at s = 0 s=0 s = 0 on this covering disk. Thus
− o ( μ 0 ) = 2 p , o ( η 0 ) = 2 q . -\mathfrak{o}(\mu_0)=2p,\qquad\mathfrak{o}(\eta_0)=2q. − o ( μ 0 ) = 2 p , o ( η 0 ) = 2 q .
We will establish the following claim by induction via the correspondence W \mathcal{W} W introduced in § 2.4 .
Claim. For any N ∈ Z ≥ 0 N\in\mathbb{Z}_{\geq0} N ∈ Z ≥ 0 , there exists ϵ N > 0 \epsilon_N>0 ϵ N > 0 such that on D ϵ N = { z ∈ C ∣ ∣ z ∣ < ϵ N } D_{\epsilon_N}=\{z\in\mathbb{C}\mid|z|<\epsilon_N\} D ϵ N = { z ∈ C ∣ ∣ z ∣ < ϵ N } , there exist meromorphic functions μ N ( s ) , λ N ( s ) \mu_N(s),\lambda_N(s) μ N ( s ) , λ N ( s ) with only possible pole at s = 0 s=0 s = 0 ,
η N : = λ N 2 , p N : = − o ( μ N ) , and q N : = o ( η N ) \eta_N:=\lambda_N^2,\quad p_N:=-\mathfrak{o}(\mu_N),\quad\text{and}\quad q_N:=\mathfrak{o}(\eta_N) η N := λ N 2 , p N := − o ( μ N ) , and q N := o ( η N )
such that the following conditions hold.
p N + 1 = p N and q N + 1 = { 2 ( p N + q N ) q N > p N , 2 q N q N < 0. p_{N+1}=p_N\quad\text{and}\quad q_{N+1}=
\begin{cases}
2(p_N+q_N) & q_N>p_N,\\
2q_N & q_N<0.
\end{cases} p N + 1 = p N and q N + 1 = { 2 ( p N + q N ) 2 q N q N > p N , q N < 0.
As a consequence, all ( p N , q N ) (p_N,q_N) ( p N , q N ) satisfy (4.1).
We have ( μ N ( s ) , η N ( s ) , μ N + 1 ( s ) , λ N + 1 ( s ) ) ∈ W s t e p (\mu_N(s),\eta_N(s),\mu_{N+1}(s),\lambda_{N+1}(s))\in\mathcal{W}_{\mathrm{step}} ( μ N ( s ) , η N ( s ) , μ N + 1 ( s ) , λ N + 1 ( s )) ∈ W step (cf. (2.18)) for all 0 < ∣ s ∣ < min { ϵ N , ϵ N + 1 } 0<|s|<\min\{\epsilon_N,\epsilon_{N+1}\} 0 < ∣ s ∣ < min { ϵ N , ϵ N + 1 } . As a consequence, for all N ∈ Z ≥ 0 N\in\mathbb{Z}_{\geq0} N ∈ Z ≥ 0 , ( μ 0 ( s ) , η 0 ( s ) ) ∈ C ∩ G N (\mu_0(s),\eta_0(s))\in C\cap G_N ( μ 0 ( s ) , η 0 ( s )) ∈ C ∩ G N for all 0 < ∣ s ∣ < min { ϵ 0 , … , ϵ N } 0<|s|<\min\{\epsilon_0,\ldots,\epsilon_N\} 0 < ∣ s ∣ < min { ϵ 0 , … , ϵ N } .
We first note that the above claim suffices to prove the theorem. Indeed, this claim guarantees that there are infinitely many points of C ∩ G N C\cap G_N C ∩ G N . By Proposition 2.21, this intersection is constructible. An infinite constructible subset of an irreducible algebraic curve contains a nonempty Zariski-open subset. Finally, the fact that G N ⊆ S N G_N\subseteq S_N G N ⊆ S N as in (2.19) gives the assertion.
We now proceed to prove the claim. Induction starts with ( μ 0 ( s ) , η 0 ( s ) ) (\mu_0(s),\eta_0(s)) ( μ 0 ( s ) , η 0 ( s )) . Since p 0 = 2 p p_0=2p p 0 = 2 p and q 0 = 2 q q_0=2q q 0 = 2 q , the pair ( p 0 , q 0 ) (p_0,q_0) ( p 0 , q 0 ) satisfies (4.1). Assume μ n ( s ) , η n ( s ) \mu_n(s),\eta_n(s) μ n ( s ) , η n ( s ) have been constructed with the desired properties in the claim. We now build μ n + 1 \mu_{n+1} μ n + 1 and η n + 1 \eta_{n+1} η n + 1 . Pointwise, μ n + 1 ( s ) \mu_{n+1}(s) μ n + 1 ( s ) is a solution to the equation
A n ( s ) z 2 + B n ( s ) z + A n ( s ) = 0. A_n(s)z^2+B_n(s)z+A_n(s)=0. A n ( s ) z 2 + B n ( s ) z + A n ( s ) = 0.
where A n A_n A n and B n B_n B n are derived from (2.9) by inserting μ n ( s ) \mu_n(s) μ n ( s ) and η n ( s ) \eta_n(s) η n ( s ) . Explicitly, we have
A n = − μ n ( η n − 1 ) ( μ n η n + 1 ) , A_n=-\mu_n(\eta_n-1)(\mu_n\eta_n+1), A n = − μ n ( η n − 1 ) ( μ n η n + 1 ) ,
B n = μ n 3 η n 2 + 2 μ n 2 η n − μ n 2 + μ n η n 2 − 2 μ n η n − 1. B_n=\mu_n^3\eta_n^2+2\mu_n^2\eta_n-\mu_n^2+\mu_n\eta_n^2-2\mu_n\eta_n-1. B n = μ n 3 η n 2 + 2 μ n 2 η n − μ n 2 + μ n η n 2 − 2 μ n η n − 1.
They are obviously meromorphic on a sufficiently small disk. A direct computation of the order yields
Table 2. o \mathfrak{o} o of building blocks of A n A_n A n and B n B_n B n . “Leading” terms for B n B_n B n are bolded.
As a result, we have
o ( A n B n ) = { ( − p n ) − ( − 2 p n ) = p n q n > p n , ( 2 q n − 2 p n ) − ( 2 q n − 3 p n ) = p n q n < 0. \mathfrak{o}\left(\frac{A_n}{B_n}\right)=
\begin{cases}
(-p_n)-(-2p_n)=p_n & q_n>p_n,\\
(2q_n-2p_n)-(2q_n-3p_n)=p_n & q_n<0.
\end{cases} o ( B n A n ) = { ( − p n ) − ( − 2 p n ) = p n ( 2 q n − 2 p n ) − ( 2 q n − 3 p n ) = p n q n > p n , q n < 0.
As a conclusion, we obtain an equation
A n B n z 2 + z + A n B n = 0 \frac{A_n}{B_n}z^2+z+\frac{A_n}{B_n}=0 B n A n z 2 + z + B n A n = 0
with holomorphic coefficients. At s = 0 s=0 s = 0 , we must have z = 0 z=0 z = 0 and derivative of the left-hand-side at z = 0 z=0 z = 0 is 1 ≠ 0 1\ne0 1 = 0 . Thus the holomorphic implicit function theorem applies and there exists a local (non-constant) holomorphic function z n + 1 ( s ) z_{n+1}(s) z n + 1 ( s ) satisfying the equation. We then take μ n + 1 ( s ) = z n + 1 − 1 ( s ) \mu_{n+1}(s)=z_{n+1}^{-1}(s) μ n + 1 ( s ) = z n + 1 − 1 ( s ) , as a meromorphic function for sufficiently small ∣ s ∣ |s| ∣ s ∣ . From Vieta’s theorem, we know
z n + 1 ( s ) + μ n + 1 ( s ) = − B n ( s ) A n ( s ) . z_{n+1}(s)+\mu_{n+1}(s)=-\frac{B_n(s)}{A_n(s)}. z n + 1 ( s ) + μ n + 1 ( s ) = − A n ( s ) B n ( s ) .
Hence we conclude (note we must have o ( μ n + 1 ) < 0 < o ( z n + 1 ) \mathfrak{o}(\mu_{n+1})<0<\mathfrak{o}(z_{n+1}) o ( μ n + 1 ) < 0 < o ( z n + 1 ) )
o ( μ n + 1 ) = o ( z n + 1 + μ n + 1 ) = o ( B n A n ) = − p n . \mathfrak{o}(\mu_{n+1})=\mathfrak{o}(z_{n+1}+\mu_{n+1})=\mathfrak{o}\left(\frac{B_n}{A_n}\right)=-p_n. o ( μ n + 1 ) = o ( z n + 1 + μ n + 1 ) = o ( A n B n ) = − p n .
η n + 1 = λ n + 1 2 \eta_{n+1}=\lambda_{n+1}^2 η n + 1 = λ n + 1 2 is uniquely determined by μ n \mu_n μ n , η n \eta_n η n , and μ n + 1 \mu_{n+1} μ n + 1 via (2.16). However, we need a different formula to perform the computation.
Recall, as in the proof of Lemma 2.10, we have
d n = 1 + μ n + 1 1 + μ n and a n = μ n d n . d_n=\frac{1+\mu_{n+1}}{1+\mu_n}\quad\text{and}\quad a_n=\mu_nd_n. d n = 1 + μ n 1 + μ n + 1 and a n = μ n d n .
Definition 2.15 item (2) implies an identity
( μ n η n 2 − 1 ) ( a n − 1 ) ( d n − 1 ) = η n ( μ n − 1 ) a n d n . (24) (\mu_n\eta_n^2-1)(a_n-1)(d_n-1)=\eta_n(\mu_n-1)a_nd_n. \tag*{(24)} ( μ n η n 2 − 1 ) ( a n − 1 ) ( d n − 1 ) = η n ( μ n − 1 ) a n d n . (24)
Direct computation shows
o ( d n ) = 0 , o ( a n ) = − p n , o ( μ n − 1 ) = − p n , o ( a n − 1 ) = − p n , \mathfrak{o}(d_n)=0,\quad\mathfrak{o}(a_n)=-p_n,\quad\mathfrak{o}(\mu_n-1)=-p_n,\quad\mathfrak{o}(a_n-1)=-p_n, o ( d n ) = 0 , o ( a n ) = − p n , o ( μ n − 1 ) = − p n , o ( a n − 1 ) = − p n ,
and
o ( μ n η n 2 − 1 ) = { 0 q n > p n , 2 q n − p n q n < 0. \mathfrak{o}(\mu_n\eta_n^2-1)=
\begin{cases}
0 & q_n>p_n,\\
2q_n-p_n & q_n<0.
\end{cases} o ( μ n η n 2 − 1 ) = { 0 2 q n − p n q n > p n , q n < 0.
Hence we conclude
o ( d n − 1 ) = { q n − p n q n > p n , − q n q n < 0. \mathfrak{o}(d_n-1)=
\begin{cases}
q_n-p_n & q_n>p_n,\\
-q_n & q_n<0.
\end{cases} o ( d n − 1 ) = { q n − p n − q n q n > p n , q n < 0.
Definition 2.15 item (3) implies
η n + 1 = ( η n ( a n − 1 ) + ( d n − 1 ) ( a n − 1 ) + η n ( d n − 1 ) ) 2 . (25) \eta_{n+1}=\left(\frac{\eta_n(a_n-1)+(d_n-1)}{(a_n-1)+\eta_n(d_n-1)}\right)^2. \tag*{(25)} η n + 1 = ( ( a n − 1 ) + η n ( d n − 1 ) η n ( a n − 1 ) + ( d n − 1 ) ) 2 . (25)
A direct computation for the denominator shows that o ( ( a n − 1 ) + η n ( d n − 1 ) ) = − p n \operatorname{o}((a_n-1)+\eta_n(d_n-1))=-p_n o (( a n − 1 ) + η n ( d n − 1 )) = − p n for both cases. We now focus on the numerator. For the case q n < 0 q_n<0 q n < 0 , we compute o ( η n ( a n − 1 ) + ( d n − 1 ) ) = q n − p n \operatorname{o}(\eta_n(a_n-1)+(d_n-1))=q_n-p_n o ( η n ( a n − 1 ) + ( d n − 1 )) = q n − p n . For the case q n > p n q_n>p_n q n > p n , Equation (24) further implies
η n ( a n − 1 ) + ( d n − 1 ) = η n ( a n − 1 ) + η n ( μ n − 1 ) a n d n ( μ n η n 2 − 1 ) ( a n − 1 ) = η n ( μ n η n 2 − 1 ) ( a n − 1 ) 2 + a n μ n d n − a n d n ( μ n η n 2 − 1 ) ( a n − 1 ) = η n μ n η n 2 ( a n − 1 ) 2 − a n 2 + 2 a n − 1 + a n μ n d n − a n d n ( μ n η n 2 − 1 ) ( a n − 1 ) = (recall a n = μ n d n ) = η n μ n η n 2 ( a n − 1 ) 2 + a n ( 1 + 1 − d n ) − 1 ( μ n η n 2 − 1 ) ( a n − 1 ) . \begin{aligned}
\eta_n(a_n-1)+(d_n-1)
&=\eta_n(a_n-1)+\frac{\eta_n(\mu_n-1)a_nd_n}{(\mu_n\eta_n^2-1)(a_n-1)}\\
&=\eta_n\frac{(\mu_n\eta_n^2-1)(a_n-1)^2+a_n\mu_nd_n-a_nd_n}{(\mu_n\eta_n^2-1)(a_n-1)}\\
&=\eta_n\frac{\mu_n\eta_n^2(a_n-1)^2-a_n^2+2a_n-1+a_n\mu_nd_n-a_nd_n}{(\mu_n\eta_n^2-1)(a_n-1)}\\
&\mathrel{\phantom{=}} \text{(recall }a_n=\mu_nd_n\text{)}
=\eta_n\frac{\mu_n\eta_n^2(a_n-1)^2+a_n(1+1-d_n)-1}{(\mu_n\eta_n^2-1)(a_n-1)}.
\end{aligned} η n ( a n − 1 ) + ( d n − 1 ) = η n ( a n − 1 ) + ( μ n η n 2 − 1 ) ( a n − 1 ) η n ( μ n − 1 ) a n d n = η n ( μ n η n 2 − 1 ) ( a n − 1 ) ( μ n η n 2 − 1 ) ( a n − 1 ) 2 + a n μ n d n − a n d n = η n ( μ n η n 2 − 1 ) ( a n − 1 ) μ n η n 2 ( a n − 1 ) 2 − a n 2 + 2 a n − 1 + a n μ n d n − a n d n = (recall a n = μ n d n ) = η n ( μ n η n 2 − 1 ) ( a n − 1 ) μ n η n 2 ( a n − 1 ) 2 + a n ( 1 + 1 − d n ) − 1 .
Hence we conclude o ( η n ( a n − 1 ) + ( d n − 1 ) ) = q n \operatorname{o}(\eta_n(a_n-1)+(d_n-1))=q_n o ( η n ( a n − 1 ) + ( d n − 1 )) = q n . In summary, we conclude
q n + 1 = o ( η n + 1 ) = { 2 ( q n + p n ) q n > p n , 2 q n q n < 0. q_{n+1}=\operatorname{o}(\eta_{n+1})=
\begin{cases}
2(q_n+p_n) & q_n>p_n,\\
2q_n & q_n<0.
\end{cases} q n + 1 = o ( η n + 1 ) = { 2 ( q n + p n ) 2 q n q n > p n , q n < 0.
Finally, it remains to show that ( μ n ( s ) , η n ( s ) , μ n + 1 ( s ) , λ n + 1 ( s ) ) ∈ W s t e p (\mu_n(s),\eta_n(s),\mu_{n+1}(s),\lambda_{n+1}(s))\in\mathcal{W}_{\mathrm{step}} ( μ n ( s ) , η n ( s ) , μ n + 1 ( s ) , λ n + 1 ( s )) ∈ W step when ∣ s ∣ |s| ∣ s ∣ is small enough and non-zero. This amounts to checking several non-vanishing conditions as in (19) and (20) . The expression δ ( μ n , η n , μ n + 1 ) \delta(\mu_n,\eta_n,\mu_{n+1}) δ ( μ n , η n , μ n + 1 ) can be expressed as the denominator of λ n + 1 \lambda_{n+1} λ n + 1 in (25) times − ( μ n + 1 ) -(\mu_n+1) − ( μ n + 1 ) , which has order − 2 p n -2p_n − 2 p n and hence non-zero for small enough ∣ s ∣ |s| ∣ s ∣ . Remaining terms are:
η n − 1 ≠ 0 , μ n ± 1 ≠ 0 , μ n η n + 1 ≠ 0 , μ n 2 η n 2 − μ n η n 2 + 4 μ n η n − μ n + 1 ≠ 0 , μ n η n 2 − 1 ≠ 0. \eta_n-1\ne0,\quad\mu_n\pm1\ne0,\quad\mu_n\eta_n+1\ne0,\quad\mu_n^2\eta_n^2-\mu_n\eta_n^2+4\mu_n\eta_n-\mu_n+1\ne0,\quad\mu_n\eta_n^2-1\ne0. η n − 1 = 0 , μ n ± 1 = 0 , μ n η n + 1 = 0 , μ n 2 η n 2 − μ n η n 2 + 4 μ n η n − μ n + 1 = 0 , μ n η n 2 − 1 = 0.
This can be verified by direct computation:
Table 3. o \operatorname{o} o of the five terms in the non-vanishing conditions.
In summary, each has finite order in both cases, and hence must be non-zero when ∣ s ∣ |s| ∣ s ∣ is positive but small enough. We thus complete the induction and conclude the proof of the claim and hence the theorem. □
Remark 4.3. The passage to the covering disk u = s 2 u=s^2 u = s 2 ensures the existence of the initial meromorphic λ 0 \lambda_0 λ 0 . Although η \eta η is meromorphic in the original parameter u u u , its order q q q may be odd, whereas the square of a meromorphic function has even order. After the substitution u = s 2 u=s^2 u = s 2 , the function η 0 ( s ) = η ( s 2 ) \eta_0(s)=\eta(s^2) η 0 ( s ) = η ( s 2 ) has order 2 q 2q 2 q and hence admits a meromorphic square root on a sufficiently small disk. If q q q is even, this covering substitution can be omitted. This applies, for example, to the branch constructed for a 2-bridge knot in Theorem 4.6 , since q = r p q=rp q = r p , where r r r is an even integer.
Definition 4.4. Let K K K be a knot. An irreducible curve C ⊂ T C \subset T C ⊂ T is an actual rank-d d d augmentation peripheral curve for K K K if it contains a Zariski-dense subset C a c t C^{\mathrm{act}} C act such that every point of C a c t C^{\mathrm{act}} C act is of the form
( μ , η ) = ( ϵ ( μ ) , ϵ ( λ ) 2 ) (\mu,\eta) = (\epsilon(\mu),\epsilon(\lambda)^2) ( μ , η ) = ( ϵ ( μ ) , ϵ ( λ ) 2 ) for a rank-d d d augmentation ϵ \epsilon ϵ of K K K with ϵ ( μ ) ≠ 1 \epsilon(\mu) \ne1 ϵ ( μ ) = 1 .
Under the augmentation–sheaf correspondence, these points are ( μ , λ 2 ) (\mu,\lambda^2) ( μ , λ 2 ) , where ( μ , λ ) (\mu,\lambda) ( μ , λ ) are the peripheral data of irreducible d d d -dimensional diagonalizable μ \mu μ rk-one representations. The word “actual” distinguishes these points from points introduced only by taking a Zariski closure.
Corollary 4.5. Suppose that K K K has an actual rank-d d d augmentation peripheral curve with an iterable branch. Then, for every N ≥ 0 N \ge0 N ≥ 0 ,
ar ( Wh N ( K ) ) ≥ 2 N d . \operatorname{ar}(\operatorname{Wh}^{N}(K)) \ge2^{N}d. ar ( Wh N ( K )) ≥ 2 N d . If d = b ( K ) d=b(K) d = b ( K ) , then
ar ( Wh N ( K ) ) = mr ( Wh N ( K ) ) = b ( Wh N ( K ) ) = 2 N b ( K ) . \operatorname{ar}(\operatorname{Wh}^{N}(K))=\operatorname{mr}(\operatorname{Wh}^{N}(K))=b(\operatorname{Wh}^{N}(K))=2^{N}b(K). ar ( Wh N ( K )) = mr ( Wh N ( K )) = b ( Wh N ( K )) = 2 N b ( K ) . Proof. By Theorem 4.2, C ∩ S N C \cap S_N C ∩ S N contains a nonempty open subset of C C C . This subset meets C a c t C^{\mathrm{act}} C act . Choose a rank-d d d augmentation giving such a point and apply Proposition 2.16. This proves the lower bound. When d = b ( K ) d=b(K) d = b ( K ) , Corollary 2.17 gives all three equalities. □
The actual augmentation point chosen in this proof may depend on N N N . The result does not require a single point to realize all the iterations.
Two-bridge knots admit iterable branches
Throughout this section, we take K K K to be a 2-bridge knot. The goal of this section is to establish the following theorem
Theorem 4.6. Any 2-bridge knot admits an actual rank-2 augmentation peripheral curve as in Definition 4.4 with an iterable branch as in Definition 4.1.
Proof of Theorem 1.1. The combination of Theorem 4.6 and Corollary 4.5 directly implies
ar ( Wh N ( K ) ) = mr ( Wh N ( K ) ) = b ( Wh N ( K ) ) = 2 N + 1 \operatorname{ar}(\operatorname{Wh}^{N}(K))=\operatorname{mr}(\operatorname{Wh}^{N}(K))=b(\operatorname{Wh}^{N}(K))=2^{N+1} ar ( Wh N ( K )) = mr ( Wh N ( K )) = b ( Wh N ( K )) = 2 N + 1
for every N ≥ 0 N \ge0 N ≥ 0 . □
The rest of the section is dedicated to the proof of Theorem 4.6. We need several preparations. First, a properly embedded surface S ⊂ E ( K ) = S 3 − int ( N ( K ) ) S \subset E(K)=S^3-\operatorname{int}(N(K)) S ⊂ E ( K ) = S 3 − int ( N ( K )) is said to be essential if it is incompressible, boundary-incompressible, and non-boundary-parallel. We have the following existence result.
Lemma 4.7. Suppose K ⊂ S 3 K \subset S^3 K ⊂ S 3 is a 2-bridge knot. Then there exists an oriented essential non-closed properly embedded surface S S S in its knot complement and ∂ S \partial S ∂ S has slope r r r which is an even non-zero integer.
Proof. First, [28 Theorem 1] shows that the diameter of the set of boundary slopes is positive for a 2-bridge knot K K K . Since a minimal genus Seifert surface has boundary slope 0, there must exist a non-zero boundary slope r r r . Furthermore, [28 Section 1] indicates that such an r r r can be picked to be an even integer. □
Proposition 4.8. Suppose K ⊂ S 3 K \subset S^3 K ⊂ S 3 is a nontrivial 2-bridge knot. Then there exist a nonzero even boundary slope r r r , a one-dimensional irreducible Zariski-closed subset C ′ ⊂ X S L 2 ( C ) ( K ) C' \subset X_{\mathrm{SL}_2(\mathbb{C})}(K) C ′ ⊂ X SL 2 ( C ) ( K ) , an ideal point Q Q Q of the smooth projective model C ~ ′ \widetilde{C}' C ′ of C ′ C' C ′ , and a positive rational number ε \varepsilon ε , such that for any ( p , q ) ∈ Z 2 (p,q) \in\mathbb{Z}^2 ( p , q ) ∈ Z 2 we have
pole Q ( I p , q ∣ C ′ ) = ε ⋅ ∣ p − r q ∣ . \operatorname{pole}_{Q}(I_{p,q}|_{C'}) = \varepsilon\cdot|p-rq|. pole Q ( I p , q ∣ C ′ ) = ε ⋅ ∣ p − r q ∣. Here, I p , q I_{p,q} I p , q is a regular function on X S L 2 ( C ) ( K ) X_{\mathrm{SL}_2(\mathbb{C})}(K) X SL 2 ( C ) ( K ) introduced in Definition 3.12, and we view I p , q ∣ C ′ I_{p,q}|_{C'} I p , q ∣ C ′ as a rational function on C ~ ′ \widetilde{C}' C ′ .
Proof. We first show that there exists a 1-dimensional irreducible Zariski-closed subset D ⊂ X P S L 2 ( C ) ( K ) D \subset X_{\mathrm{PSL}_2(\mathbb{C})}(K) D ⊂ X PSL 2 ( C ) ( K ) , and an ideal point Q Q Q of its smooth projective model D ~ \widetilde{D} D , such that there exists a positive integer ε ′ \varepsilon' ε ′ and for any ( p , q ) ∈ Z 2 (p,q) \in\mathbb{Z}^2 ( p , q ) ∈ Z 2 , we have
pole Q ( I p , q ∣ D ) = ε ′ ⋅ ∣ p − r q ∣ . \operatorname{pole}_{Q}(I_{p,q}|_{D}) = \varepsilon' \cdot|p-rq|. pole Q ( I p , q ∣ D ) = ε ′ ⋅ ∣ p − r q ∣.
We proceed with the non-zero even boundary slope r r r obtained from Lemma 4.7. Lemma 3.21 then applies and produces a quasi-affine curve X 0 X_0 X 0 with a point Q Q Q in its smooth projective model X ~ 0 \widetilde{X}_0 X 0 and the pole order formula
pole Q ( I p , q O h ∣ X 0 ) = ε 0 ⋅ ∣ p − r q ∣ . (26) \operatorname{pole}_{Q}(I^{Oh}_{p,q}|_{X_0}) = \varepsilon_0 \cdot|p-rq|. \tag*{(26)} pole Q ( I p , q O h ∣ X 0 ) = ε 0 ⋅ ∣ p − r q ∣. (26)
However, we still need a transition from Ohtsuki’s space X X X to our P S L 2 ( C ) \mathrm{PSL}_2(\mathbb{C}) PSL 2 ( C ) character variety. Ohtsuki defined elements in X X X as conjugation classes of representations ρ ∈ R P S L 2 ( C ) ( K ) \rho\in R_{\mathrm{PSL}_2(\mathbb{C})}(K) ρ ∈ R PSL 2 ( C ) ( K ) , and our map τ ^ : R P S L 2 ( C ) ( K ) → X P S L 2 ( C ) ( K ) \widehat{\tau}: R_{\mathrm{PSL}_2(\mathbb{C})}(K) \to X_{\mathrm{PSL}_2(\mathbb{C})}(K) τ : R PSL 2 ( C ) ( K ) → X PSL 2 ( C ) ( K ) defined as in Definition 3.13 with a chosen set of square trace functions generating the coordinate ring naturally induces a map, which, by abusing notation, we still denote by τ ^ : X 0 → X P S L 2 ( C ) ( K ) \widehat{\tau}: X_0 \to X_{\mathrm{PSL}_2(\mathbb{C})}(K) τ : X 0 → X PSL 2 ( C ) ( K ) . One core identity that is straightforward from the construction is that
I p , q O h = I p , q ∘ τ ^ I^{Oh}_{p,q} = I_{p,q} \circ\widehat{\tau} I p , q O h = I p , q ∘ τ
We claim that the map τ ^ \widehat{\tau} τ is regular. Indeed, by Lemma 3.15, the coordinate ring C [ X P S L 2 ( C ) ( K ) ] \mathbb{C}[X_{\mathrm{PSL}_2(\mathbb{C})}(K)] C [ X PSL 2 ( C ) ( K )] is generated by square trace functions. Lemma 3.20 implies that all square trace functions are regular on X 0 X_0 X 0 .
Next, take D = τ ^ ( X 0 ) D = \widehat{\tau}(X_0) D = τ ( X 0 ) to be the Zariski-closure of the image inside X P S L 2 ( C ) ( K ) X_{\mathrm{PSL}_2(\mathbb{C})}(K) X PSL 2 ( C ) ( K ) . It is irreducible since X 0 X_0 X 0 is. [25, Corollary 3.6] states that the irreducible locus X P S L 2 ( C ) i r r ( K ) ⊂ X P S L 2 ( C ) ( K ) X^{\mathrm{irr}}_{\mathrm{PSL}_2(\mathbb{C})}(K) \subset X_{\mathrm{PSL}_2(\mathbb{C})}(K) X PSL 2 ( C ) irr ( K ) ⊂ X PSL 2 ( C ) ( K ) is Zariski open. Taking D i r r = X P S L 2 ( C ) i r r ( K ) ∩ D D^{\mathrm{irr}} = X^{\mathrm{irr}}_{\mathrm{PSL}_2(\mathbb{C})}(K) \cap D D irr = X PSL 2 ( C ) irr ( K ) ∩ D , it is open in D D D . Thus, τ ^ − 1 ( D i r r ) \widehat{\tau}^{-1}(D^{\mathrm{irr}}) τ − 1 ( D irr ) is open, and we show that it is not empty in X 0 X_0 X 0 . Suppose, on the contrary, that every element [ ρ ] ∈ X 0 [\rho] \in X_0 [ ρ ] ∈ X 0 is reducible. Then for any [ ρ ] ∈ X 0 [\rho] \in X_0 [ ρ ] ∈ X 0 , τ ^ ( [ ρ ] ) = χ ρ \widehat{\tau}([\rho]) = \chi_\rho τ ([ ρ ]) = χ ρ is a reducible character. [25, Remark 3.7] implies that the same χ ρ \chi_\rho χ ρ can also be represented by an abelian representation ρ ′ \rho' ρ ′ : χ ρ = χ ρ ′ \chi_\rho= \chi_{\rho'} χ ρ = χ ρ ′ . Yet for abelian ρ ′ \rho' ρ ′ , ρ ′ ( ℓ K ) = [ ± I 2 ] \rho'(\ell_K) = [\pm I_2] ρ ′ ( ℓ K ) = [ ± I 2 ] . This makes I r , 1 = I r , 0 I_{r,1} = I_{r,0} I r , 1 = I r , 0 , and pulling back, we conclude I r , 1 O h ∣ X 0 = I r , 0 O h ∣ X 0 I^{Oh}_{r,1}|_{X_0} = I^{Oh}_{r,0}|_{X_0} I r , 1 O h ∣ X 0 = I r , 0 O h ∣ X 0 but Equation (26) suggests
pole Q ( I r , 1 O h ) = 0 and pole Q ( I r , 0 O h ) ≠ 0. \operatorname{pole}_{Q}(I^{Oh}_{r,1}) = 0 \quad\text{and} \quad\operatorname{pole}_{Q}(I^{Oh}_{r,0}) \ne0. pole Q ( I r , 1 O h ) = 0 and pole Q ( I r , 0 O h ) = 0.
This provides a contradiction.
As a result, V = τ ^ − 1 ( D i r r ) V = \widehat{\tau}^{-1}(D^{\mathrm{irr}}) V = τ − 1 ( D irr ) is a non-empty open subset of X 0 X_0 X 0 . [25, Lemma 3.15] then shows that τ ^ \widehat{\tau} τ is injective on V V V . To summarize, we obtain a rational map τ ^ : X 0 → D \widehat{\tau}: X_0 \to D τ : X 0 → D which is injective on a non-empty Zariski-open subset. A standard argument shows that τ ^ \widehat{\tau} τ is birational. See, for example, [22, Exercise 7.8 and paragraph right before Exercise 7.10].
As a consequence, X ~ 0 \widetilde{X}_0 X 0 also serves as a smooth projective model for D D D . Write the birational map ϕ D : X ~ 0 ⇢ D \phi_D: \widetilde{X}_0 \dashrightarrow D ϕ D : X 0 ⇢ D . We now argue that Q Q Q is also an ideal point for D D D : if not, Q ∈ Dom ( ϕ D ) Q \in\operatorname{Dom}(\phi_D) Q ∈ Dom ( ϕ D ) . Yet I 1 , 0 I_{1,0} I 1 , 0 is a regular function on all of X P S L 2 ( C ) ( K ) X_{\mathrm{PSL}_2(\mathbb{C})}(K) X PSL 2 ( C ) ( K ) , which means it is regular on a neighborhood of ϕ ~ D ( Q ) \widetilde{\phi}_D(Q) ϕ D ( Q ) and its pullback ϕ ~ D ∗ I 1 , 0 \widetilde{\phi}_D^*I_{1,0} ϕ D ∗ I 1 , 0 is regular in a neighborhood of Q ∈ X ~ 0 Q \in\widetilde{X}_0 Q ∈ X 0 : this implies that pole Q ( I 1 , 0 ) = 0 \operatorname{pole}_Q(I_{1,0}) = 0 pole Q ( I 1 , 0 ) = 0 , which contradicts Equation (26) . Thus, the curve D D D and its ideal point Q Q Q are the desired data on X P S L 2 ( C ) ( K ) X_{\mathrm{PSL}_2(\mathbb{C})}(K) X PSL 2 ( C ) ( K ) , i.e., for any ( p , q ) ∈ Z 2 (p,q) \in\mathbb{Z}^2 ( p , q ) ∈ Z 2 , we have
pole Q ( I p , q ∣ D ) = ε ′ ⋅ ∣ p − r q ∣ . \operatorname{pole}_Q(I_{p,q}|_D) = \varepsilon' \cdot|p-rq|. pole Q ( I p , q ∣ D ) = ε ′ ⋅ ∣ p − r q ∣.
We can then lift D D D to the desired curve C ′ ⊂ X S L 2 ( C ) ( K ) C' \subset X_{\mathrm{SL}_2(\mathbb{C})}(K) C ′ ⊂ X SL 2 ( C ) ( K ) via Corollary 3.18 . The ideal point and associated valuation are liftable as well by Lemma 3.2 and Lemma 3.4 . This completes the proof. □
For a 2-bridge knot K K K , let C ′ ⊂ X S L 2 ( C ) ( K ) C' \subset X_{\mathrm{SL}_2(\mathbb{C})}(K) C ′ ⊂ X SL 2 ( C ) ( K ) be the curve supplied by Proposition 4.8 , and use the fixed peripheral system ( m K , ℓ K ) (m_K,\ell_K) ( m K , ℓ K ) . We now proceed to construct the desired curve C ⊂ T C \subset T C ⊂ T (with iterable branch) to establish Theorem 4.6 . The basic idea is to twist an S L 2 ( C ) \mathrm{SL}_2(\mathbb{C}) SL 2 ( C ) representation to obtain a μ \mu μ rk-one representation, though a detailed construction shall pass to what we call the eigenvalue cover E E E of C ′ C' C ′ . A conceptual diagram is the following.
C ′ ← π E → ζ C ⊂ T . C' \xleftarrow{\ \pi\ } E \xrightarrow{\ \zeta\ } C \subset T. C ′ π E ζ C ⊂ T .
Next, we construct E E E . First, Culler–Shalen [17, Proof of Corollary 1.4.5] show that characters from irreducible representations form a Zariski-open subset of X S L 2 ( C ) ( K ) X_{\mathrm{SL}_2(\mathbb{C})}(K) X SL 2 ( C ) ( K ) . Hence on C ′ C' C ′ they form a possibly empty open subset U U U . We argue that this is nonempty. Otherwise all characters come from reducible S L 2 ( C ) \mathrm{SL}_2(\mathbb{C}) SL 2 ( C ) representations. Such a representation can be written in upper-triangular form, and its diagonal part is an abelian representation with the same character. Yet for an abelian representation ℓ K \ell_K ℓ K is sent to I 2 I_2 I 2 , so I r , 1 = I r , 0 I_{r,1}=I_{r,0} I r , 1 = I r , 0 , violating the pole order formula in Proposition 4.8 . Thus U ⊂ C ′ U \subset C' U ⊂ C ′ is a non-empty and hence dense open subset of C ′ C' C ′ .
Define X ⊂ C ′ × ( C ∗ ) 2 X \subset C' \times(\mathbb{C}^*)^2 X ⊂ C ′ × ( C ∗ ) 2 by the equations
M 2 − t m K M + 1 = 0 L 2 − t ℓ K L + 1 = 0 ( M L ) 2 − t m K ℓ K ( M L ) + 1 = 0 (27) \begin{aligned}
M^2-t_{m_K}M+1&=0\\
L^2-t_{\ell_K}L+1&=0\\
(ML)^2-t_{m_K\ell_K}(ML)+1&=0 \tag*{(27)}
\end{aligned} M 2 − t m K M + 1 L 2 − t ℓ K L + 1 ( M L ) 2 − t m K ℓ K ( M L ) + 1 = 0 = 0 = 0 (27)
X X X is an affine algebraic set, since the trace functions are polynomial functions on X S L 2 ( C ) ( K ) X_{\mathrm{SL}_2(\mathbb{C})}(K) X SL 2 ( C ) ( K ) and hence on C ′ C' C ′ . The projection π : X → C ′ \pi:X\to C' π : X → C ′ is surjective: m K m_K m K and ℓ K \ell_K ℓ K commute in π K \pi_K π K by assumption, so for a representation ρ \rho ρ we can always take the eigenvalues of ρ ( m K ) \rho(m_K) ρ ( m K ) and ρ ( ℓ K ) \rho(\ell_K) ρ ( ℓ K ) on a common eigenline to satisfy Equation (27) . Note the third equation requires the compatibility between M M M and L L L . The defining equations also imply that C [ X ] \mathbb{C}[X] C [ X ] is a finitely generated module over C [ C ′ ] \mathbb{C}[C'] C [ C ′ ] via the pullback π ∗ \pi^* π ∗ : four generators are enough: 1 1 1 , M M M , L L L , M L ML M L . As a result, the proof of Corollary 3.18 applies again: there exists an irreducible Zariski-closed subset E ⊂ X E \subset X E ⊂ X such that π ( E ) = C ′ \pi(E)=C' π ( E ) = C ′ . C [ E ] \mathbb{C}[E] C [ E ] is a quotient of C [ X ] \mathbb{C}[X] C [ X ] , so it is also a finitely generated module over C [ C ′ ] \mathbb{C}[C'] C [ C ′ ] via pullback. Then E E E is itself an algebraic curve. We have the following property of E E E .
Lemma 4.9. Let K ⊂ S 3 K \subset S^3 K ⊂ S 3 be a 2-bridge knot. Let r r r be a non-zero even boundary slope arising from Lemma 4.7 , C ′ ⊂ X S L 2 ( C ) ( K ) C' \subset X_{\mathrm{SL}_2(\mathbb{C})}(K) C ′ ⊂ X SL 2 ( C ) ( K ) be the algebraic curve arising from Proposition 4.8 and π : E → C ′ \pi:E\to C' π : E → C ′ be the eigenvalue cover of C ′ C' C ′ together with regular functions M M M and L L L as above. Then E E E admits an ideal point Q ′ Q' Q ′ such that
ν Q ′ ( M ) ≠ 0 , r ν Q ′ ( M ) + ν Q ′ ( L ) = 0. (28) \nu_{Q'}(M)\ne0,\quad r\nu_{Q'}(M)+\nu_{Q'}(L)=0. \tag*{(28)} ν Q ′ ( M ) = 0 , r ν Q ′ ( M ) + ν Q ′ ( L ) = 0. (28)
Proof. Let ( E ~ , ϕ E ) (\widetilde{E},\phi_E) ( E , ϕ E ) be a smooth projective model of E E E as in Definition 3.1. From Lemma 3.2 , we obtain a compatible surjective morphism π ~ : E ~ → C ~ \widetilde{\pi}:\widetilde{E}\to\widetilde{C} π : E → C . Take Q ′ ∈ π ~ − 1 ( Q ) Q'\in\widetilde{\pi}^{-1}(Q) Q ′ ∈ π − 1 ( Q ) . We argue that Q ′ Q' Q ′ must be an ideal point of E E E . Otherwise, ϕ E ( Q ′ ) ∈ E \phi_E(Q')\in E ϕ E ( Q ′ ) ∈ E . The function I 1 , 0 I_{1,0} I 1 , 0 is regular on C ′ C' C ′ , and hence π ∗ I 1 , 0 \pi^*I_{1,0} π ∗ I 1 , 0 is regular on E E E . Further pulling it back, ϕ ~ E ∗ π ∗ I 1 , 0 \widetilde{\phi}_E^*\pi^*I_{1,0} ϕ E ∗ π ∗ I 1 , 0 is a regular function on a Zariski-open neighborhood of Q ′ Q' Q ′ so it has pole order 0 0 0 . On the other hand, Lemma 3.4 implies that
pole Q ′ ( π ∗ I p , q ) = pole Q ′ ( π ~ ∗ I p , q ) = ε 1 ⋅ ∣ p − r q ∣ , ε 1 > 0. (29) \operatorname{pole}_{Q'}(\pi^*I_{p,q})=\operatorname{pole}_{Q'}(\widetilde{\pi}^*I_{p,q})=\varepsilon_1\cdot\lvert p-rq\rvert,\qquad\varepsilon_1>0. \tag*{(29)} pole Q ′ ( π ∗ I p , q ) = pole Q ′ ( π ∗ I p , q ) = ε 1 ⋅ ∣ p − r q ∣ , ε 1 > 0. (29)
This provides a contradiction to the compatibility of π ~ \widetilde{\pi} π by taking p = 1 p=1 p = 1 and q = 0 q=0 q = 0 . Hence we conclude that Q ′ Q' Q ′ remains an ideal point of E E E . Now we further explore ν Q ′ ( M ) \nu_{Q'}(M) ν Q ′ ( M ) and ν Q ′ ( L ) \nu_{Q'}(L) ν Q ′ ( L ) . Taking p = 1 p=1 p = 1 and q = 0 q=0 q = 0 in Equation (29) , together with π ∗ I 1 , 0 = ( M − M − 1 ) 2 \pi^*I_{1,0}=(M-M^{-1})^2 π ∗ I 1 , 0 = ( M − M − 1 ) 2 , shows that ν Q ′ ( M ) ≠ 0 \nu_{Q'}(M)\ne0 ν Q ′ ( M ) = 0 . Hence
E 0 = { e ∈ E : M ( e ) 2 ≠ 1 } E_0=\{e\in E:M(e)^2\ne1\} E 0 = { e ∈ E : M ( e ) 2 = 1 }
is a nonempty open subset of E E E .
The third equation in (27) ensures that M M M and L L L belong to the same peripheral eigenline on E 0 E_0 E 0 . To see this, fix e ∈ E 0 e\in E_0 e ∈ E 0 and a representation of the character π ( e ) \pi(e) π ( e ) . Write α = M ( e ) \alpha=M(e) α = M ( e ) . In a basis of meridian eigenvectors, the meridian and longitude matrices are diag ( α , α − 1 ) \operatorname{diag}(\alpha,\alpha^{-1}) diag ( α , α − 1 ) and diag ( β , β − 1 ) \operatorname{diag}(\beta,\beta^{-1}) diag ( β , β − 1 ) , respectively. The second equation allows L ( e ) = β L(e)=\beta L ( e ) = β or β − 1 \beta^{-1} β − 1 . In the latter case, the third equation implies
( α − α − 1 ) ( β − β − 1 ) = 0. (\alpha-\alpha^{-1})(\beta-\beta^{-1})=0. ( α − α − 1 ) ( β − β − 1 ) = 0.
Since α 2 ≠ 1 \alpha^2\ne1 α 2 = 1 , this forces β = β − 1 \beta=\beta^{-1} β = β − 1 , so either choice is compatible. It follows on E 0 E_0 E 0 , and hence on E E E , that
π ∗ I p , q = ( M p L q − M − p L − q ) 2 ( ( p , q ) ∈ Z 2 ) . \pi^*I_{p,q}=(M^pL^q-M^{-p}L^{-q})^2\qquad((p,q)\in\mathbb{Z}^2). π ∗ I p , q = ( M p L q − M − p L − q ) 2 (( p , q ) ∈ Z 2 ) .
Taking p = r p=r p = r and q = 1 q=1 q = 1 in Equation (29) shows that r ν Q ′ ( M ) + ν Q ′ ( L ) = 0 r\nu_{Q'}(M)+\nu_{Q'}(L)=0 r ν Q ′ ( M ) + ν Q ′ ( L ) = 0 . Thus we obtain the desired relation (28) .
Proof of Theorem 4.6. Let K K K be a 2-bridge knot, r r r be a non-zero even integral boundary slope as in Lemma 4.7 , ( m K , ℓ K ) (m_K,\ell_K) ( m K , ℓ K ) be chosen peripheral data, and C ′ ⊂ X S L 2 ( C ) ( K ) C'\subset X_{\mathrm{SL}_2(\mathbb{C})}(K) C ′ ⊂ X SL 2 ( C ) ( K ) be an irreducible algebraic curve with ideal point Q Q Q with special pole order formula supplied by Proposition 4.8 . Let π : E → C ′ \pi:E\to C' π : E → C ′ be the eigenvalue cover constructed as above, together with regular functions M M M and L L L , and E E E 's smooth projective model ( E ~ , ϕ E ) (\widetilde{E},\phi_E) ( E , ϕ E ) with ideal point Q ′ ∈ E ~ Q'\in\widetilde{E} Q ′ ∈ E and valuation data as in Lemma 4.9 .
We now construct the actual rank-2 augmentation peripheral curve C ⊂ T C\subset T C ⊂ T as follows. Take σ = 1 \sigma=1 σ = 1 if ν Q ′ ( M ) < 0 \nu_{Q'}(M)<0 ν Q ′ ( M ) < 0 and σ = − 1 \sigma=-1 σ = − 1 otherwise. We define a map ζ : E → T = ( C ∗ ) ( μ , η ) 2 \zeta:E\to T=(\mathbb{C}^*)^2_{(\mu,\eta)} ζ : E → T = ( C ∗ ) ( μ , η ) 2 by μ = M 2 σ \mu=M^{2\sigma} μ = M 2 σ and η = L 2 σ \eta=L^{2\sigma} η = L 2 σ . Let C ⊂ T C\subset T C ⊂ T be the Zariski closure of im ζ \operatorname{im}\zeta im ζ . Since E E E is irreducible, so is C C C . The dimension theorem of Ferrer Santos–Rittatore [19, Chapter 1, Theorem 4.84] then implies that C C C has dimension at most 1 1 1 . Dimension 0 0 0 implies that C C C must be a single point (see § 2.4 ). If C C C were a point, then the map ζ \zeta ζ would be constant. Yet (viewing μ \mu μ as a regular function on E E E ) ν Q ′ ( μ ) = 2 σ ⋅ ν Q ′ ( M ) < 0 \nu_{Q'}(\mu)=2\sigma\cdot\nu_{Q'}(M)<0 ν Q ′ ( μ ) = 2 σ ⋅ ν Q ′ ( M ) < 0 so ζ \zeta ζ is not constant. Consequently, C C C is an affine algebraic curve. Take its smooth projective model ( C ~ , ϕ C ) (\widetilde{C},\phi_C) ( C , ϕ C ) . Lemma 3.2 produces a compatible surjective morphism E ~ → C ~ \widetilde{E}\to\widetilde{C} E → C . Taking P P P to be the image of Q ′ Q' Q ′ under this morphism, Lemma 3.4 now implies that there exists a positive ε \varepsilon ε such that
− p = ν P ( μ ) = ε ⋅ 2 σ ⋅ ν Q ′ ( M ) < 0. -p=\nu_P(\mu)=\varepsilon\cdot2\sigma\cdot\nu_{Q'}(M)<0. − p = ν P ( μ ) = ε ⋅ 2 σ ⋅ ν Q ′ ( M ) < 0.
The same lemma gives
q = ν P ( η ) = ε ⋅ 2 σ ⋅ ν Q ′ ( L ) , q p = r . q=\nu_P(\eta)=\varepsilon\cdot2\sigma\cdot\nu_{Q'}(L),\qquad\frac{q}{p}=r. q = ν P ( η ) = ε ⋅ 2 σ ⋅ ν Q ′ ( L ) , p q = r .
The pole of the regular function μ \mu μ at P P P shows that P P P is an ideal point of C C C . Since p > 0 p>0 p > 0 and r ∉ [ 0 , 1 ] r\notin[0,1] r ∈ / [ 0 , 1 ] , either q > p q>p q > p or q < 0 q<0 q < 0 .
As a consequence, with Lemma 3.5, we conclude that C C C has an iterable branch as defined in Definition 4.1. It remains to show that C C C provides an actual rank-two augmentation peripheral curve, as in Definition 4.4.
First, note there is a dense open subset U ⊂ C ′ ⊂ X S L 2 ( C ) ( K ) U\subset C'\subset X_{\mathrm{SL}_2(\mathbb{C})}(K) U ⊂ C ′ ⊂ X SL 2 ( C ) ( K ) consisting of characters of irreducible representations. Pulling it back via the surjective π \pi π , we obtain an open dense subset V ′ = π − 1 ( U ) ⊂ E V'=\pi^{-1}(U)\subset E V ′ = π − 1 ( U ) ⊂ E . We restrict V ′ V' V ′ to the nonempty open subset where M , L M,L M , L belong to a common peripheral eigenline, and further exclude the closed subsets M = ± 1 M=\pm1 M = ± 1 . Call the resulting subset V V V . It is still open, and note that E E E is not fully contained in M = ± 1 M=\pm1 M = ± 1 since ν Q ′ ( M ) ≠ 0 \nu_{Q'}(M)\ne0 ν Q ′ ( M ) = 0 by Lemma 4.9. Hence we conclude that V ⊂ E V\subset E V ⊂ E is an open and dense subset.
For e ∈ V e\in V e ∈ V , choose an irreducible representation ρ e \rho_e ρ e with character π ( e ) \pi(e) π ( e ) , and write α = M ( e ) \alpha=M(e) α = M ( e ) and β = L ( e ) \beta=L(e) β = L ( e ) . Let ab : π K → Z \operatorname{ab}:\pi_K\to\mathbb{Z} ab : π K → Z be the abelianization with ab ( m K ) = 1 \operatorname{ab}(m_K)=1 ab ( m K ) = 1 . Define the scalar twist ρ e σ : π K → G L ( C ) \rho_e^\sigma:\pi_K\to\mathrm{GL}(\mathbb{C}) ρ e σ : π K → GL ( C ) by
ρ e σ ( g ) = α σ ab ( g ) ρ e ( g ) . \rho_e^\sigma(g)=\alpha^{\sigma\operatorname{ab}(g)}\rho_e(g). ρ e σ ( g ) = α σ ab ( g ) ρ e ( g ) .
After this scalar twist, the two eigenvalues of the meridian are 1 1 1 and α 2 σ \alpha^{2\sigma} α 2 σ , and hence an irreducible S L 2 ( C ) \mathrm{SL}_2(\mathbb{C}) SL 2 ( C ) representation yields an irreducible two-dimensional diagonalizable μ \mu μ rk-one representation. Since ab ( ℓ K ) = 0 \operatorname{ab}(\ell_K)=0 ab ( ℓ K ) = 0 , the longitude eigenvalue on the distinguished meridian eigenline is β σ \beta^\sigma β σ . The augmentation–representation correspondence therefore realizes
( μ ( e ) , η ( e ) ) = ( α 2 σ , β 2 σ ) (\mu(e),\eta(e))=(\alpha^{2\sigma},\beta^{2\sigma}) ( μ ( e ) , η ( e )) = ( α 2 σ , β 2 σ )
as an actual rank-two augmentation peripheral point.
To conclude the proof, it remains to show that ζ ( V ) \zeta(V) ζ ( V ) contains an open and dense subset of C C C . Now the fact that V V V is dense in E E E implies that ζ ( V ) \zeta(V) ζ ( V ) is dense in ζ ( E ) \zeta(E) ζ ( E ) , and hence dense in C = ζ ( E ) C=\zeta(E) C = ζ ( E ) . Thus the constructibility theorem of Ferrer Santos–Rittatore [19, Theorem 4.9] applies, and the proof is complete. □
Corollary 4.10. Let K K K be a nontrivial knot with mr ( K ) = 2 \operatorname{mr}(K)=2 mr ( K ) = 2 . Then, for every N ≥ 0 N\ge0 N ≥ 0 ,
ar ( W h N ( K ) ) = mr ( W h N ( K ) ) = b ( W h N ( K ) ) = 2 N + 1 . \operatorname{ar}(\mathrm{Wh}^{N}(K))=\operatorname{mr}(\mathrm{Wh}^{N}(K))=b(\mathrm{Wh}^{N}(K))=2^{N+1}. ar ( Wh N ( K )) = mr ( Wh N ( K )) = b ( Wh N ( K )) = 2 N + 1 . Proof. The two-meridian characterization of two-bridge knots [11, Corollary 3.3], with the orbifold theorem [9], implies that K K K is two-bridge. Apply Theorem 1.1. □
Ordinary group rank
Having proved the two-bridge result, we compare it with ordinary group rank. Here generators are unrestricted: gr ( K ) \operatorname{gr}(K) gr ( K ) is the minimum cardinality of a generating set of π K \pi_K π K .
Proposition 5.1. For every knot K K K and every N ≥ 0 N\ge0 N ≥ 0 ,
gr ( W h ( K ) ) ≤ gr ( K ) + 1 , gr ( W h N ( K ) ) ≤ gr ( K ) + N . \operatorname{gr}(\mathrm{Wh}(K))\le\operatorname{gr}(K)+1,\qquad\operatorname{gr}(\mathrm{Wh}^{N}(K))\le\operatorname{gr}(K)+N. gr ( Wh ( K )) ≤ gr ( K ) + 1 , gr ( Wh N ( K )) ≤ gr ( K ) + N .
Proof. The relation ℓ V x 2 ℓ V − 1 = x 1 \ell_Vx_2\ell_V^{-1}=x_1 ℓ V x 2 ℓ V − 1 = x 1 in (4) gives x 2 = ℓ V − 1 x 1 ℓ V x_2=\ell_V^{-1}x_1\ell_V x 2 = ℓ V − 1 x 1 ℓ V . Thus x 1 x_1 x 1 and ℓ V \ell_V ℓ V generate the pattern group. In the satellite group, ℓ V = ℓ K \ell_V=\ell_K ℓ V = ℓ K belongs to the image of the companion group. It follows that π W h ( K ) \pi_{\mathrm{Wh}(K)} π Wh ( K ) is generated by this image and x 1 x_1 x 1 . This proves the first inequality, and induction gives the second. □
For comparison, Gu–Wang–Zou [21, Proposition 1.3] prove that
gr ( W h N ( K ) ) ≥ N + 1 \operatorname{gr}(\mathrm{Wh}^{N}(K)) \ge N+1 gr ( Wh N ( K )) ≥ N + 1
for every nontrivial knot K K K and N ≥ 1 N\ge1 N ≥ 1 . Each Whitehead layer contributes a hyperbolic piece to the JSJ decomposition, and their argument uses these pieces to bound the rank from below. Combining this with Proposition 5.1 , we obtain
N + 1 ≤ gr ( W h N ( K ) ) ≤ N + gr ( K ) , N ≥ 1. N+1\le\operatorname{gr}(\mathrm{Wh}^{N}(K))\le N+\operatorname{gr}(K),\qquad N\ge1. N + 1 ≤ gr ( Wh N ( K )) ≤ N + gr ( K ) , N ≥ 1.
In particular, for a fixed nontrivial K K K ,
lim N → ∞ gr ( W h N ( K ) ) N + 1 = 1. \lim_{N\to\infty}\frac{\operatorname{gr}(\mathrm{Wh}^{N}(K))}{N+1}=1. N → ∞ lim N + 1 gr ( Wh N ( K )) = 1.
This asymptotic formula also appears in [21, Corollary 2.5], where the upper bound is expressed using tunnel number.
When K K K is hyperbolic with tunnel number one, [21, Corollary 2.6] gives the exact value gr ( W h N ( K ) ) = N + 2 \operatorname{gr}(\mathrm{Wh}^{N}(K))=N+2 gr ( Wh N ( K )) = N + 2 for N ≥ 1 N\ge1 N ≥ 1 . In particular, for a hyperbolic two-bridge knot, Theorem 1.1 gives the comparison
gr ( W h N ( K ) ) = N + 2 < 2 N + 1 = ar ( W h N ( K ) ) = mr ( W h N ( K ) ) , N ≥ 1. \operatorname{gr}(\mathrm{Wh}^{N}(K))=N+2<2^{N+1}=\operatorname{ar}(\mathrm{Wh}^{N}(K))=\operatorname{mr}(\mathrm{Wh}^{N}(K)),\qquad N\ge1. gr ( Wh N ( K )) = N + 2 < 2 N + 1 = ar ( Wh N ( K )) = mr ( Wh N ( K )) , N ≥ 1.
The exact ordinary-rank formula here uses the hyperbolic hypothesis; we do not assert it for every two-bridge companion.
Corollary 5.2. Suppose K K K has a rank-d d d augmentation ϵ \epsilon ϵ with ϵ ( μ ) ≠ 1 \epsilon(\mu)\ne1 ϵ ( μ ) = 1 and peripheral point in S N S_N S N . If 2 N d ≥ gr ( K ) + N 2^Nd\ge\operatorname{gr}(K)+N 2 N d ≥ gr ( K ) + N , then
gr ( W h N ( K ) ) ≤ ar ( W h N ( K ) ) . \operatorname{gr}(\mathrm{Wh}^{N}(K))\le\operatorname{ar}(\mathrm{Wh}^{N}(K)). gr ( Wh N ( K )) ≤ ar ( Wh N ( K )) .
Proof. Proposition 2.16 gives ar ( W h N ( K ) ) ≥ 2 N d \operatorname{ar}(\mathrm{Wh}^{N}(K))\ge2^Nd ar ( Wh N ( K )) ≥ 2 N d , whereas Proposition 5.1 gives gr ( W h N ( K ) ) ≤ gr ( K ) + N \operatorname{gr}(\mathrm{Wh}^{N}(K))\le\operatorname{gr}(K)+N gr ( Wh N ( K )) ≤ gr ( K ) + N . The numerical hypothesis compares these bounds. □
For a rank-two augmentation the numerical condition is gr ( K ) ≤ 2 N + 1 − N \operatorname{gr}(K)\le2^{N+1}-N gr ( K ) ≤ 2 N + 1 − N . In particular it is automatic when gr ( K ) = 2 \operatorname{gr}(K)=2 gr ( K ) = 2 and N ≥ 1 N\ge1 N ≥ 1 , provided the required peripheral point lies in S N S_N S N . The existence of such a point, for each N N N , is supplied by an actual rank-two augmentation peripheral curve with an iterable branch; see the proof of Corollary 4.5 .
Proof of Theorem 1.2. The unconditional upper bound is Proposition 5.1 . Under the actual-curve hypothesis, Corollary 4.5 gives ar ( W h N ( K ) ) ≥ 2 N d \operatorname{ar}(\mathrm{Wh}^{N}(K))\ge2^Nd ar ( Wh N ( K )) ≥ 2 N d for every N ≥ 0 N\ge0 N ≥ 0 . Consequently,
gr ( W h N ( K ) ) ≤ gr ( K ) + N ≤ 2 N d ≤ ar ( W h N ( K ) ) \operatorname{gr}(\mathrm{Wh}^{N}(K))\le\operatorname{gr}(K)+N\le2^Nd\le\operatorname{ar}(\mathrm{Wh}^{N}(K)) gr ( Wh N ( K )) ≤ gr ( K ) + N ≤ 2 N d ≤ ar ( Wh N ( K ))
whenever the middle inequality holds. Finally, a nontrivial two-bridge knot has a two-generator nonabelian knot group, so gr ( K ) = 2 \operatorname{gr}(K)=2 gr ( K ) = 2 . Theorem 1.1 and N + 2 ≤ 2 N + 1 N+2\le2^{N+1} N + 2 ≤ 2 N + 1 yield the final assertion. □
Remark 5.3. For a fixed knot with the actual-curve hypothesis and d > 0 d>0 d > 0 , the numerical condition in Theorem 1.2 holds for all sufficiently large N N N . This conclusion depends on the curve hypothesis; the linear upper bound alone does not imply gr ( W h N ( K ) ) ≤ ar ( W h N ( K ) ) \operatorname{gr}(\mathrm{Wh}^{N}(K))\le\operatorname{ar}(\mathrm{Wh}^{N}(K)) gr ( Wh N ( K )) ≤ ar ( Wh N ( K )) .
The first Whitehead double of the square knot
In this section, we prove Theorem 1.3. Let
S : = T ( 2 , 3 ) # T ( 2 , 3 ) ‾ S := T(2,3) \# \overline{T(2,3)} S := T ( 2 , 3 ) # T ( 2 , 3 )
be the square knot. It has
ar ( S ) = mr ( S ) = b ( S ) = 3 \operatorname{ar}(S) = \operatorname{mr}(S) = b(S) = 3 ar ( S ) = mr ( S ) = b ( S ) = 3
where ar ( S ) = 2 + 2 − 1 = 3 \operatorname{ar}(S)=2+2-1=3 ar ( S ) = 2 + 2 − 1 = 3 is given by the formula in [15, Theorem 1.9], and b ( S ) = 2 + 2 − 1 = 3 b(S)=2+2-1=3 b ( S ) = 2 + 2 − 1 = 3 is by [32 ] .
For its untwisted Whitehead double Wh ( S ) \operatorname{Wh}(S) Wh ( S ) , Dutra’s bridge formula gives b ( Wh ( S ) ) = 2 b ( S ) = 6 b(\operatorname{Wh}(S))=2b(S)=6 b ( Wh ( S )) = 2 b ( S ) = 6 [18, Section 4]. In § 6.1 , we show that ar ( Wh ( S ) ) ≤ 5 \operatorname{ar}(\operatorname{Wh}(S))\leq5 ar ( Wh ( S )) ≤ 5 , and in § 6.2 , we then use a finite quotient and group cohomology to show that its mr ( Wh ( S ) ) = 6 \operatorname{mr}(\operatorname{Wh}(S))=6 mr ( Wh ( S )) = 6 .
Augmentation rank of Wh(S)
The knot group of the square knot is
π S = ⟨ a , b , c ∣ a b a = b a b , a c a = c a c ⟩ , \pi_S=\langle a,b,c\mid aba=bab,\ aca=cac\rangle, π S = ⟨ a , b , c ∣ aba = bab , a c a = c a c ⟩ ,
where the peripheral subgroup is generated by m = a , ℓ = ( a b ) 3 ( a c ) − 3 m=a,\ \ell=(ab)^3(ac)^{-3} m = a , ℓ = ( ab ) 3 ( a c ) − 3 . See e.g. [36 Section 2.2] for such a presentation. By Proposition 2.8,
π Wh ( S ) = ⟨ x 1 , x 2 , a , b , c , ℓ | a = x 2 x 1 − 1 x 2 − 1 x 1 , ℓ x 2 ℓ − 1 = x 1 , [ a , ℓ ] = 1 , a b a = b a b , a c a = c a c , ℓ = ( a b ) 3 ( a c ) − 3 ⟩ (30) \pi_{\operatorname{Wh}(S)}=
\left\langle x_1,x_2,a,b,c,\ell\ \middle|\
\begin{aligned}
a&=x_2x_1^{-1}x_2^{-1}x_1,\quad\ell x_2\ell^{-1}=x_1,\quad[a,\ell]=1,\\
aba&=bab,\quad aca=cac,\quad\ell=(ab)^3(ac)^{-3}
\end{aligned}
\right\rangle \tag*{(30)} π Wh ( S ) = ⟨ x 1 , x 2 , a , b , c , ℓ a aba = x 2 x 1 − 1 x 2 − 1 x 1 , ℓ x 2 ℓ − 1 = x 1 , [ a , ℓ ] = 1 , = bab , a c a = c a c , ℓ = ( ab ) 3 ( a c ) − 3 ⟩ (30)
Proposition 6.1. Over C \mathbb{C} C , ar ( Wh ( S ) ) ≤ 5 \operatorname{ar}(\operatorname{Wh}(S))\leq5 ar ( Wh ( S )) ≤ 5 .
Proof. We know ar ( Wh ( S ) ) ≤ b ( Wh ( S ) ) = 6 \operatorname{ar}(\operatorname{Wh}(S))\leq b(\operatorname{Wh}(S))=6 ar ( Wh ( S )) ≤ b ( Wh ( S )) = 6 . We shall prove by contradiction and assume ar ( Wh ( S ) ) = 6 \operatorname{ar}(\operatorname{Wh}(S))=6 ar ( Wh ( S )) = 6 . By definition, there exists a 6-dimensional irreducible diagonalizable μ r k − 1 \mu\mathrm{rk}-1 μ rk − 1 representation
φ : π Wh ( S ) ⟶ GL ( V ) . \varphi:\pi_{\operatorname{Wh}(S)}\longrightarrow\operatorname{GL}(V). φ : π Wh ( S ) ⟶ GL ( V ) . Let X i = φ ( x i ) X_i=\varphi(x_i) X i = φ ( x i ) , A = φ ( a ) A=\varphi(a) A = φ ( a ) etc. Let W i : = im ( X i − I ) W_i:=\operatorname{im}(X_i-I) W i := im ( X i − I ) , W a : = im ( A − I ) W_a:=\operatorname{im}(A-I) W a := im ( A − I ) etc. Since ( φ , V ) (\varphi,V) ( φ , V ) is μ r k − 1 \mu\mathrm{rk}-1 μ rk − 1 , each of W i W_i W i has dimension 1. Since a , b , c a,b,c a , b , c are conjugate, their microlocal ranks are equal. Moreover, in the doubling construction, each of them is the composite of two meridians of Wh ( S ) \operatorname{Wh}(S) Wh ( S ) , therefore
s : = dim W a = dim W b = dim W c ≤ 2. s:=\dim W_a=\dim W_b=\dim W_c\leq2. s := dim W a = dim W b = dim W c ≤ 2. Define W 12 = W 1 + W 2 W_{12}=W_1+W_2 W 12 = W 1 + W 2 , and define W a b c = W a + W b + W c W_{abc}=W_a+W_b+W_c W ab c = W a + W b + W c .
We consider the subspace W : = W 12 + W a b c ⊂ V W:=W_{12}+W_{abc}\subset V W := W 12 + W ab c ⊂ V . Each of X i X_i X i preserves every subspace containing W 12 W_{12} W 12 , hence preserving W W W . The matrices A , B , C A,B,C A , B , C preserve W a b c W_{abc} W ab c and act as the identity on V / W a b c V/W_{abc} V / W ab c , hence they also preserve W W W . Since ℓ \ell ℓ is a word of a , b , c a,b,c a , b , c , its image preserves W W W as well. Altogether, we see π Wh ( S ) \pi_{\operatorname{Wh}(S)} π Wh ( S ) preserves W W W . By irreducibility, we have dim V = dim W \dim V=\dim W dim V = dim W . Therefore,
6 = dim V = dim W ≤ dim W 12 + dim W a b c ≤ 2 + 3 s . 6=\dim V=\dim W\leq\dim W_{12}+\dim W_{abc}\leq2+3s. 6 = dim V = dim W ≤ dim W 12 + dim W ab c ≤ 2 + 3 s . Since s ≤ 2 s\leq2 s ≤ 2 , it is only possible with s = 2 s=2 s = 2 . Now W a ⊂ W 12 W_a\subset W_{12} W a ⊂ W 12 and both spaces have dimension two, so W a = W 12 W_a=W_{12} W a = W 12 . Consequently W W W itself is invariant under X 1 , X 2 X_1,X_2 X 1 , X 2 as well as the companion group. Irreducibility and dimensions give the direct decomposition
V = W a ⊕ W b ⊕ W c . V=W_a\oplus W_b\oplus W_c. V = W a ⊕ W b ⊕ W c .
By choosing a suitable basis, we have
A = ( N A 12 A 13 0 I 0 0 0 I ) , B = ( I 0 0 B 21 N C 21 0 0 I ) , C = ( I 0 0 0 I 0 C 31 C 32 N ) . A=\begin{pmatrix}N&A_{12}&A_{13}\\0&I&0\\0&0&I\end{pmatrix},\qquad B=\begin{pmatrix}I&0&0\\B_{21}&N&C_{21}\\0&0&I\end{pmatrix},\qquad C=\begin{pmatrix}I&0&0\\0&I&0\\C_{31}&C_{32}&N\end{pmatrix}. A = N 0 0 A 12 I 0 A 13 0 I , B = I B 21 0 0 N 0 0 C 21 I , C = I 0 C 31 0 I C 32 0 0 N .
where N ∈ GL 2 ( C ) N\in\operatorname{GL}_{2}(\mathbb{C}) N ∈ GL 2 ( C ) . Direct block multiplication in A B A = B A B ABA=BAB A B A = B A B gives
( N + A 12 B 21 ) [ N − I , A 12 , A 13 ] = 0 , ( B 21 A 12 + N ) [ B 21 , N − I , C 21 ] = 0. (N+A_{12}B_{21})[N-I,A_{12},A_{13}]=0,\qquad(B_{21}A_{12}+N)[B_{21},N-I,C_{21}]=0. ( N + A 12 B 21 ) [ N − I , A 12 , A 13 ] = 0 , ( B 21 A 12 + N ) [ B 21 , N − I , C 21 ] = 0.
The matrices [ N − I , A 12 , A 13 ] [N-I,A_{12},A_{13}] [ N − I , A 12 , A 13 ] , [ B 21 , N − I , C 21 ] [B_{21},N-I,C_{21}] [ B 21 , N − I , C 21 ] have rank two because A , B A,B A , B have μ rk = 2 \mu\operatorname{rk}=2 μ rk = 2 . Therefore
A 12 B 21 = B 21 A 12 = − N . A_{12}B_{21}=B_{21}A_{12}=-N. A 12 B 21 = B 21 A 12 = − N .
Similarly, the relation A C A = C A C ACA=CAC A C A = C A C gives
A 13 C 31 = C 31 A 13 = − N . A_{13}C_{31}=C_{31}A_{13}=-N. A 13 C 31 = C 31 A 13 = − N .
With these identities, it is a routine calculation to find
L ∣ W a = ( A B ) 3 ( A C ) − 3 ∣ W a = − N 3 ( − N 3 ) − 1 = I . L|_{W_a}=(AB)^3(AC)^{-3}|_{W_a}=-N^3(-N^3)^{-1}=I. L ∣ W a = ( A B ) 3 ( A C ) − 3 ∣ W a = − N 3 ( − N 3 ) − 1 = I .
Finally, since A = X 2 X 1 − 1 X 2 − 1 X 1 A=X_2X_1^{-1}X_2^{-1}X_1 A = X 2 X 1 − 1 X 2 − 1 X 1 , we have W a ⊂ W 1 + W 2 W_a\subset W_1+W_2 W a ⊂ W 1 + W 2 . By dimension count, we must have W a = W 1 ⊕ W 2 W_a=W_1\oplus W_2 W a = W 1 ⊕ W 2 . However, the relation L X 2 L − 1 = X 1 LX_2L^{-1}=X_1 L X 2 L − 1 = X 1 yields W 2 = L − 1 W 1 W_2=L^{-1}W_1 W 2 = L − 1 W 1 . Since L ∣ W a = I L|_{W_a}=I L ∣ W a = I , this implies W 1 = W 2 W_1=W_2 W 1 = W 2 , a contradiction.
This concludes the proof by contradiction, and we prove the assertion. □
Meridional rank of Wh(S S S )
We use group cohomology to give an estimate for mr ( Wh ( S ) ) \operatorname{mr}(\operatorname{Wh}(S)) mr ( Wh ( S )) . We recall the basic notions; see [14, 26] for general references. Let G G G be a group and V V V be a left G G G -module. Functorially, H n ( G , V ) : = R n Hom Z [ G ] ( Z , V ) H^n(G,V):=R^n\operatorname{Hom}_{\mathbb{Z}[G]}(\mathbb{Z},V) H n ( G , V ) := R n Hom Z [ G ] ( Z , V ) . Alternatively, the bar resolution gives a concrete cochain complex
C n ( G , V ) = Map ( G n , V ) , C^n(G,V)=\operatorname{Map}(G^n,V), C n ( G , V ) = Map ( G n , V ) ,
where the maps are set theoretic, and the differential is given by
( d f ) ( g 1 , … , g n + 1 ) = g 1 ⋅ f ( g 2 , … , g n + 1 ) + ∑ i = 1 n ( − 1 ) i f ( g 1 , … , g i g i + 1 , … , g n + 1 ) + ( − 1 ) n + 1 f ( g 1 , … , g n ) . (df)(g_1,\ldots,g_{n+1})=g_1\cdot f(g_2,\ldots,g_{n+1})+\sum_{i=1}^{n}(-1)^i f(g_1,\ldots,g_i g_{i+1},\ldots,g_{n+1})+(-1)^{n+1}f(g_1,\ldots,g_n). ( df ) ( g 1 , … , g n + 1 ) = g 1 ⋅ f ( g 2 , … , g n + 1 ) + i = 1 ∑ n ( − 1 ) i f ( g 1 , … , g i g i + 1 , … , g n + 1 ) + ( − 1 ) n + 1 f ( g 1 , … , g n ) .
We note two special cases. For n = 0 n=0 n = 0 ,
H 0 ( G , V ) = V G : = { v ∈ V : ρ ( g ) v = v for every g ∈ G } . H^0(G,V)=V^G:=\{v\in V:\rho(g)v=v\text{ for every }g\in G\}. H 0 ( G , V ) = V G := { v ∈ V : ρ ( g ) v = v for every g ∈ G } .
For n = 1 n=1 n = 1 , the 1-cocycle maps are given by crossed homomorphisms:
Z 1 ( G ; V ) = { f : G → V : f ( g h ) = f ( g ) + ρ ( g ) f ( h ) for all g , h ∈ G } , Z^1(G;V)=\{f:G\to V:f(gh)=f(g)+\rho(g)f(h)\text{ for all }g,h\in G\}, Z 1 ( G ; V ) = { f : G → V : f ( g h ) = f ( g ) + ρ ( g ) f ( h ) for all g , h ∈ G } ,
because ( d f ) ( g , h ) = g f ( h ) − f ( g h ) + f ( g ) = 0 (df)(g,h)=gf(h)-f(gh)+f(g)=0 ( df ) ( g , h ) = g f ( h ) − f ( g h ) + f ( g ) = 0 . The 1-coboundary maps are given by principal crossed homomorphisms.
B 1 ( G , V ) = { f v : v ∈ V , f v ( g ) = ρ ( g ) v − v } , B^1(G,V)=\{f_v:v\in V,\ f_v(g)=\rho(g)v-v\}, B 1 ( G , V ) = { f v : v ∈ V , f v ( g ) = ρ ( g ) v − v } ,
because ( d v ) ( g ) = g v − v (dv)(g)=gv-v ( d v ) ( g ) = g v − v .
Lemma 6.2. Let G G G be generated by r ≥ 1 r\geq1 r ≥ 1 conjugates of an element m m m , and let ρ : G → GL ( V ) \rho:G\to\operatorname{GL}(V) ρ : G → GL ( V ) be a representation on a finite-dimensional vector space over a field F F F . Then
dim Z 1 ( G , V ) − dim V ≤ ( r − 1 ) μ rk ( ρ ) (31) \dim Z^1(G,V)-\dim V\leq(r-1)\mu\operatorname{rk}(\rho) \tag*{(31)} dim Z 1 ( G , V ) − dim V ≤ ( r − 1 ) μ rk ( ρ ) (31)
Proof. For any 1 1 1 -cocycle f : G → V f:G\to V f : G → V , and any g ∈ G g\in G g ∈ G , apply f f f to the identity e G = g g − 1 e_G=gg^{-1} e G = g g − 1 , we get 0 = f ( g ) + ρ ( g ) f ( g − 1 ) 0=f(g)+\rho(g)f(g^{-1}) 0 = f ( g ) + ρ ( g ) f ( g − 1 ) . Therefore f ( g − 1 ) = − ρ ( g − 1 ) f ( g ) f(g^{-1})=-\rho(g^{-1})f(g) f ( g − 1 ) = − ρ ( g − 1 ) f ( g ) .
Let m 1 = m m_1=m m 1 = m and m i = g i − 1 m g i m_i=g_i^{-1}mg_i m i = g i − 1 m g i for 2 ≤ i ≤ r 2\leq i\leq r 2 ≤ i ≤ r . We have
f ( m i ) = f ( g i − 1 m g i ) = f ( g i − 1 ) + ρ ( g i − 1 ) f ( m g i ) = − ρ ( g i − 1 ) f ( g i ) + ρ ( g i − 1 ) ( f ( m ) + ρ ( m ) f ( g i ) ) = ρ ( g i − 1 ) f ( m ) + ρ ( g i − 1 ) ( ρ ( m ) − I ) f ( g i ) = ρ ( g i − 1 ) f ( m ) + ( ρ ( m i ) − I ) ρ ( g i − 1 ) f ( g i ) . \begin{aligned}
f(m_i)&=f(g_i^{-1}mg_i)=f(g_i^{-1})+\rho(g_i^{-1})f(mg_i) \\
&=-\rho(g_i^{-1})f(g_i)+\rho(g_i^{-1})(f(m)+\rho(m)f(g_i)) \\
&=\rho(g_i^{-1})f(m)+\rho(g_i^{-1})(\rho(m)-I)f(g_i) \\
&=\rho(g_i^{-1})f(m)+(\rho(m_i)-I)\rho(g_i^{-1})f(g_i).
\end{aligned} f ( m i ) = f ( g i − 1 m g i ) = f ( g i − 1 ) + ρ ( g i − 1 ) f ( m g i ) = − ρ ( g i − 1 ) f ( g i ) + ρ ( g i − 1 ) ( f ( m ) + ρ ( m ) f ( g i )) = ρ ( g i − 1 ) f ( m ) + ρ ( g i − 1 ) ( ρ ( m ) − I ) f ( g i ) = ρ ( g i − 1 ) f ( m ) + ( ρ ( m i ) − I ) ρ ( g i − 1 ) f ( g i ) .
A cocycle is determined by its values on the generating set. Once f ( m ) f(m) f ( m ) is fixed, the possible value of f ( m i ) f(m_i) f ( m i ) belongs to an affine space with direction im ( ( ρ ( m i ) − I ) ) \operatorname{im}((\rho(m_i)-I)) im (( ρ ( m i ) − I )) , the dimension of which equals to the microlocal rank. Therefore, the map
Z 1 ( G , V ) ⟶ V ⊕ ⨁ i = 2 r im ( I − ρ ( m i ) ) , Z^1(G,V)\longrightarrow V\oplus\bigoplus_{i=2}^{r}\operatorname{im}(I-\rho(m_i)), Z 1 ( G , V ) ⟶ V ⊕ i = 2 ⨁ r im ( I − ρ ( m i )) ,
f ⟼ ( f ( m ) , ( f ( m i ) − ρ ( g i − 1 ) f ( m ) ) i = 2 r ) , f\longmapsto\left(f(m),(f(m_i)-\rho(g_i^{-1})f(m))_{i=2}^{r}\right), f ⟼ ( f ( m ) , ( f ( m i ) − ρ ( g i − 1 ) f ( m ) ) i = 2 r ) ,
is injective. Thus dim Z 1 ( G ; V ) ≤ dim V + ( r − 1 ) μ rk ( ρ ) \dim Z^1(G;V)\leq\dim V+(r-1)\mu\operatorname{rk}(\rho) dim Z 1 ( G ; V ) ≤ dim V + ( r − 1 ) μ rk ( ρ ) , the inequality follows. On the other hand, there is a short exact sequence 0 → H 0 ( G ; V ) → V → B 1 ( G ; V ) → 0 0\to H^0(G;V)\to V\to B^1(G;V)\to0 0 → H 0 ( G ; V ) → V → B 1 ( G ; V ) → 0 . This implies dim V = dim B 1 ( G ; V ) + dim H 0 ( G ; V ) \dim V=\dim B^1(G;V)+\dim H^0(G;V) dim V = dim B 1 ( G ; V ) + dim H 0 ( G ; V ) .
Remark 6.3. The left hand side of the inequality is in fact the truncated negative Euler number dim H 1 ( G , V ) − dim H 0 ( G , V ) \dim H^1(G,V)-\dim H^0(G,V) dim H 1 ( G , V ) − dim H 0 ( G , V ) . Indeed, there is a short exact sequence 0 → H 0 ( G , V ) → V → B 1 ( G ; V ) → 0 0\to H^0(G,V)\to V\to B^1(G;V)\to0 0 → H 0 ( G , V ) → V → B 1 ( G ; V ) → 0 . This implies dim V = dim B 1 ( G ; V ) + dim H 0 ( G ; V ) \dim V=\dim B^1(G;V)+\dim H^0(G;V) dim V = dim B 1 ( G ; V ) + dim H 0 ( G ; V ) . Therefore,
dim Z 1 ( G , V ) − dim V = dim H 1 ( G , V ) − dim H 0 ( G , V ) . \dim Z^1(G,V)-\dim V=\dim H^1(G,V)-\dim H^0(G,V). dim Z 1 ( G , V ) − dim V = dim H 1 ( G , V ) − dim H 0 ( G , V ) .
Let S 14 S_{14} S 14 be the permutation group of 14 14 14 elements. We define a group homomorphism ϕ : π W h ( S ) → S 14 \phi:\pi_{\mathrm{Wh}(S)}\to S_{14} ϕ : π Wh ( S ) → S 14 by
ϕ ( x 1 ) = ( 1 2 ) ( 3 4 ) ( 5 6 ) , ϕ ( x 2 ) = ( 1 7 ) ( 3 8 ) ( 5 9 ) , ϕ ( a ) = ( 1 7 2 ) ( 3 8 4 ) ( 5 9 6 ) , ϕ ( b ) = ( 1 13 7 ) ( 3 5 10 ) ( 4 11 9 ) , ϕ ( c ) = ( 1 3 10 ) ( 2 12 8 ) ( 5 14 9 ) , ϕ ( ℓ ) = ( 1 6 8 ) ( 2 9 3 ) ( 4 7 5 ) ( 10 12 11 ) . (32) \begin{aligned}
\phi(x_1)&=(1\,2)(3\,4)(5\,6),\\
\phi(x_2)&=(1\,7)(3\,8)(5\,9),\\
\phi(a)&=(1\,7\,2)(3\,8\,4)(5\,9\,6),\\
\phi(b)&=(1\,13\,7)(3\,5\,10)(4\,11\,9),\\
\phi(c)&=(1\,3\,10)(2\,12\,8)(5\,14\,9),\\
\phi(\ell)&=(1\,6\,8)(2\,9\,3)(4\,7\,5)(10\,12\,11). \tag*{(32)}
\end{aligned} ϕ ( x 1 ) ϕ ( x 2 ) ϕ ( a ) ϕ ( b ) ϕ ( c ) ϕ ( ℓ ) = ( 1 2 ) ( 3 4 ) ( 5 6 ) , = ( 1 7 ) ( 3 8 ) ( 5 9 ) , = ( 1 7 2 ) ( 3 8 4 ) ( 5 9 6 ) , = ( 1 13 7 ) ( 3 5 10 ) ( 4 11 9 ) , = ( 1 3 10 ) ( 2 12 8 ) ( 5 14 9 ) , = ( 1 6 8 ) ( 2 9 3 ) ( 4 7 5 ) ( 10 12 11 ) . (32)
It is straightforward to check these assignments satisfy the relators in π W h ( S ) \pi_{\mathrm{Wh}(S)} π Wh ( S ) , hence ϕ \phi ϕ is well defined.
Proposition 6.4. ϕ \phi ϕ is surjective.
Proof. Observe that
ϕ ( x 1 b c ) = ( 1 6 5 14 3 4 11 9 10 13 7 2 12 8 ) \phi(x_1bc)=(1\,6\,5\,14\,3\,4\,11\,9\,10\,13\,7\,2\,12\,8) ϕ ( x 1 b c ) = ( 1 6 5 14 3 4 11 9 10 13 7 2 12 8 )
is a cyclic permutation of order 14 14 14 , and
ϕ ( ( x 2 b ) 7 ) = ( 1 13 ) \phi((x_2b)^7)=(1\,13) ϕ (( x 2 b ) 7 ) = ( 1 13 )
is a simple transposition. The distance between 1 1 1 and 13 13 13 in the cyclic permutation is 9 9 9 , which is coprime to 14 14 14 . Therefore these two elements generate S 14 S_{14} S 14 , proving surjectivity. □
Now let V = F 2 14 V=\mathbb{F}_{2}^{14} V = F 2 14 . Define the representation ρ ϕ \rho_{\phi} ρ ϕ by the composition
ρ ϕ : π W h ( S ) → ϕ S 14 → G L ( V ) , \rho_{\phi}\colon\pi_{\mathrm{Wh}(S)} \xrightarrow{\phi} S_{14} \to GL(V), ρ ϕ : π Wh ( S ) ϕ S 14 → G L ( V ) ,
where the second map is the permutation representation, defined by permuting the basis vectors.
Proposition 6.5. ( ρ ϕ , V ) (\rho_{\phi},V) ( ρ ϕ , V ) has microlocal rank 3 and
dim Z 1 ( π W h ( S ) , V ) = 27. (33) \dim Z^{1}(\pi_{\mathrm{Wh}(S)},V)=27. \tag*{(33)} dim Z 1 ( π Wh ( S ) , V ) = 27. (33)
Proof. We compute by brute force. Let e i = ( 0 , … , 0 , 1 , 0 , … , 0 ) e_i=(0,\ldots,0,1,0,\ldots,0) e i = ( 0 , … , 0 , 1 , 0 , … , 0 ) where the i i i -th coordinate is 1. Then the standard basis for V V V is { e 1 , … , e 14 } \{e_1,\ldots,e_{14}\} { e 1 , … , e 14 } . Since ϕ ( x 1 ) = ( 12 ) ( 34 ) ( 56 ) \phi(x_1)=(12)(34)(56) ϕ ( x 1 ) = ( 12 ) ( 34 ) ( 56 ) , three simple transpositions. We can compute
im ( ρ ϕ ( x 1 ) − I ) = Span { e 1 + e 2 , e 3 + e 4 , e 5 + e 6 } . \operatorname{im}(\rho_{\phi}(x_1)-I)=\operatorname{Span}\{e_1+e_2,e_3+e_4,e_5+e_6\}. im ( ρ ϕ ( x 1 ) − I ) = Span { e 1 + e 2 , e 3 + e 4 , e 5 + e 6 } .
Hence, the microlocal rank is 3.
For the cocycles, recall that π W h ( S ) \pi_{\mathrm{Wh}(S)} π Wh ( S ) is generated by 6 generators, relabeled as s i s_i s i for i = 1 , … , 6 i=1,\ldots,6 i = 1 , … , 6 , and 6 relators, denoted by R i R_i R i . To name a function f f f , we need to specify f ( s i ) f(s_i) f ( s i ) for all i = 1 , … , 6 i=1,\ldots,6 i = 1 , … , 6 . Each f ( s i ) f(s_i) f ( s i ) is a linear combination of 14 basis vectors. Therefore, f f f is specified by a vector of 84 = 6 × 14 84=6\times14 84 = 6 × 14 binary coordinates. We can choose { ( s i , e j ) 1 ≤ i ≤ 6 , 1 ≤ j ≤ 14 } \{(s_i,e_j)_{1\leq i\leq6,1\leq j\leq14}\} {( s i , e j ) 1 ≤ i ≤ 6 , 1 ≤ j ≤ 14 } to be the basis of this vector space U U U .
To be well-defined, f f f must satisfy the cocycle condition for each relator R i R_i R i . For example, take R 2 R_2 R 2 to be e π W h ( S ) = ℓ x 2 ℓ − 1 x 1 − 1 e_{\pi_{\mathrm{Wh}(S)}}=\ell x_2\ell^{-1}x_1^{-1} e π Wh ( S ) = ℓ x 2 ℓ − 1 x 1 − 1 . Apply f f f to both sides, we get Applying the cocycle f f f to e π W h ( S ) = ℓ x 2 ℓ − 1 x 1 − 1 e_{\pi_{\mathrm{Wh}(S)}}=\ell x_2\ell^{-1}x_1^{-1} e π Wh ( S ) = ℓ x 2 ℓ − 1 x 1 − 1 , we obtain
0 = f ( ℓ ) + ρ ϕ ( ℓ ) f ( x 2 ) + ρ ϕ ( ℓ x 2 ) f ( ℓ − 1 ) + ρ ϕ ( ℓ x 2 ℓ − 1 ) f ( x 1 − 1 ) = f ( ℓ ) + ρ ϕ ( ℓ ) f ( x 2 ) − ρ ϕ ( ℓ x 2 ℓ − 1 ) f ( ℓ ) − ρ ϕ ( ℓ x 2 ℓ − 1 x 1 − 1 ) f ( x 1 ) = ( I − ρ ϕ ( x 1 ) ) f ( ℓ ) + ρ ϕ ( ℓ ) f ( x 2 ) − f ( x 1 ) . \begin{aligned}
0&=f(\ell)+\rho_{\phi}(\ell)f(x_2)+\rho_{\phi}(\ell x_2)f(\ell^{-1})+\rho_{\phi}(\ell x_2\ell^{-1})f(x_1^{-1})\\
&=f(\ell)+\rho_{\phi}(\ell)f(x_2)-\rho_{\phi}(\ell x_2\ell^{-1})f(\ell)-\rho_{\phi}(\ell x_2\ell^{-1}x_1^{-1})f(x_1)\\
&=(I-\rho_{\phi}(x_1))f(\ell)+\rho_{\phi}(\ell)f(x_2)-f(x_1).
\end{aligned} 0 = f ( ℓ ) + ρ ϕ ( ℓ ) f ( x 2 ) + ρ ϕ ( ℓ x 2 ) f ( ℓ − 1 ) + ρ ϕ ( ℓ x 2 ℓ − 1 ) f ( x 1 − 1 ) = f ( ℓ ) + ρ ϕ ( ℓ ) f ( x 2 ) − ρ ϕ ( ℓ x 2 ℓ − 1 ) f ( ℓ ) − ρ ϕ ( ℓ x 2 ℓ − 1 x 1 − 1 ) f ( x 1 ) = ( I − ρ ϕ ( x 1 )) f ( ℓ ) + ρ ϕ ( ℓ ) f ( x 2 ) − f ( x 1 ) .
This is a system of dim V = 14 \dim V=14 dim V = 14 linear equations. Solutions to this system is the kernel of a 14 × 84 14\times84 14 × 84 matrix. Since there are 6 relators, we obtain a total of 14 × 6 = 84 14\times6=84 14 × 6 = 84 equations, which form a matrix of size 84 × 84 84\times84 84 × 84 .
For clarity, we describe how the matrix is assembled. Let
( s 1 , … , s 6 ) = ( x 1 , x 2 , a , b , c , ℓ ) , M i = ρ ϕ ( s i ) , (s_1,\ldots,s_6)=(x_1,x_2,a,b,c,\ell),\qquad M_i=\rho_{\phi}(s_i), ( s 1 , … , s 6 ) = ( x 1 , x 2 , a , b , c , ℓ ) , M i = ρ ϕ ( s i ) ,
where M i e j = e ϕ ( s i ) ( j ) M_i e_j=e_{\phi(s_i)(j)} M i e j = e ϕ ( s i ) ( j ) . We write the six relators as
R 1 = a − 1 x 2 x 2 − 1 x 1 − 1 x 1 , R_1=a^{-1}x_2x_2^{-1}x_1^{-1}x_1, R 1 = a − 1 x 2 x 2 − 1 x 1 − 1 x 1 ,
R 2 = ℓ x 2 ℓ − 1 x 1 − 1 , R_2=\ell x_2\ell^{-1}x_1^{-1}, R 2 = ℓ x 2 ℓ − 1 x 1 − 1 ,
R 3 = a ℓ a − 1 ℓ − 1 , R_3=a\ell a^{-1}\ell^{-1}, R 3 = a ℓ a − 1 ℓ − 1 ,
R 4 = a b a b − 1 a − 1 b − 1 , R_4=abab^{-1}a^{-1}b^{-1}, R 4 = aba b − 1 a − 1 b − 1 ,
R 5 = a c a c − 1 a − 1 c − 1 , R_5=acac^{-1}a^{-1}c^{-1}, R 5 = a c a c − 1 a − 1 c − 1 ,
R 6 = ℓ − 1 ( a b ) 3 ( c − 1 a − 1 ) 3 . R_6=\ell^{-1}(ab)^3(c^{-1}a^{-1})^3. R 6 = ℓ − 1 ( ab ) 3 ( c − 1 a − 1 ) 3 .
In the following pseudocode, D [ r , i ] D[r,i] D [ r , i ] denotes the 14 × 14 14\times14 14 × 14 block in block row r r r and block column i i i . All additions are over F 2 \mathbb{F}_2 F 2 , and powers in the relators are expanded before reading the letters.
D := zero matrix of size 84 x 84
for r = 1,...,6:
P := identity matrix of size 14
for each letter s_i^epsilon in R_r, from left to right:
if epsilon = -1:
P := P M_i^(-1)
D[r,i] := D[r,i] + P
if epsilon = 1:
P := P M_i
return D
At the start of each letter, P P P is the image of the preceding prefix. The order of the updates follows from the cocycle identity and f ( s i − 1 ) = − M i − 1 f ( s i ) f(s_i^{-1}) = -M_i^{-1} f(s_i) f ( s i − 1 ) = − M i − 1 f ( s i ) ; the minus sign disappears over F 2 \mathbb{F}_2 F 2 . Thus the r r r -th block row records f ( R r ) = 0 f(R_r)=0 f ( R r ) = 0 . This gives the matrix displayed below.
( ( 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ⋮ ) ) \left(
\begin{pmatrix}
0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&1&0&0&0&0&1&0&0&0&0&0&0&0&0&0&0&0&0&0&1&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0\\
1&0&0&0&0&0&1&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&1&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0\\
0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&1&0&0&0&1&0&0&0&0&0&0&0&0&0&0&0&0&0&1&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0\\
0&0&1&0&0&0&0&1&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&1&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0\\
0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&1&0&0&1&0&0&0&0&0&0&0&0&0&0&0&0&0&1&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0\\
0&0&0&0&1&0&0&0&1&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&1&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0&0\\
\vdots
\end{pmatrix}
\right) 0 1 0 0 0 0 ⋮ 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
The set of cocycles Z 1 ( π W h ( S ) , V ) Z^1(\pi_{Wh(S)},V) Z 1 ( π W h ( S ) , V ) is the kernel of this matrix. Using a computer program, we find that
dim Z 1 ( π W h ( S ) , V ) = 27. \dim Z^1(\pi_{Wh(S)},V)=27. dim Z 1 ( π W h ( S ) , V ) = 27.
Proof of Theorem 1.3. We know that b ( W h ( S ) ) = 6 b(\mathrm{Wh}(S))=6 b ( Wh ( S )) = 6 , and that ar ( W h ( S ) ) ≤ 5 \operatorname{ar}(\mathrm{Wh}(S))\leq5 ar ( Wh ( S )) ≤ 5 by Proposition 6.1. If π W h ( S ) \pi_{Wh(S)} π W h ( S ) is generated by r r r meridians, Lemma 6.2 and Proposition 6.5 imply
27 − 14 = 13 ≤ 3 ( r − 1 ) . 27-14=13\leq3(r-1). 27 − 14 = 13 ≤ 3 ( r − 1 ) .
Hence r > 5 r>5 r > 5 . Together with the upper bound b = 6 b=6 b = 6 , this implies mr ( W h ( S ) ) = 6 \operatorname{mr}(\mathrm{Wh}(S))=6 mr ( Wh ( S )) = 6 .
Acknowledgements
The authors would like to thank Yu Zhao for some helpful conversations.
The authors acknowledge the use of OpenAI’s ChatGPT during the development of this work. This project began in October 2025, motivated in part by an interest in exploring how AI tools could support mathematical research. Initially, GPT assisted with writing SageMath code, implementing computations in SnapPy, and conducting computational experiments. It was later used to explore proof strategies, develop arguments, prepare an initial draft, and assist in checking the code, computations, and proposed proofs.
Throughout this process, the authors supplied the relevant literature, many of the initial mathematical ideas, detailed prompts, and directed the successive stages of experimentation, proof development, and revision. The initial AI-assisted draft was subsequently substantially rewritten by the authors, who reorganized the manuscript and revised the exposition and presentation of the proofs. Further checks and revisions were carried out through mathematical discussions among the authors and additional interactions with these tools. The AI-assisted checks were part of this review process, not an independent certification of the proofs. The authors take full responsibility for the mathematical content, computations, references, and final presentation.
HG is supported by the National Key R&D Program of China 2023YFA1010500 and NSFC Grant No. 12625107. SG is supported by NSFC Grant No. 12201102.
(H. Gao) Yau Mathematical Sciences Center, Tsinghua University, Beijing, 100084 & Beijing Institute of Mathematical Sciences and Applications, Beijing, 101408, China Email address: gaohonghao@tsinghua.edu.cn
(S. Gu) Department of Mathematics, Northeastern University, Shenyang, Liaoning, China, 110004 Email address: shijiegutop@gmail.com
(Z. Li) State Key Laboratory of Mathematical Sciences, Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing 100190, China Email address: zhenkun@amss.ac.cn
(J. Wang) State Key Laboratory of Mathematical Sciences, Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing 100190, China Email address: jian.wang.4@amss.ac.cn