Knots with large character varieties Philip Choi, Joan Porti, Seokbeom Yoon Abstract . We study knots whose SL 2 ​ ( ℂ ) \mathrm{SL}_{2}(\mathbb{C}) -character varieties have a component of dimension greater than one. We call such knots 𝒳 \mathcal{X} -large and introduce two diagrammatic constructions that produce 𝒳 \mathcal{X} -large knots. The first construction uses split link diagrams and rational tangle replacements, providing a topological explanation for most 𝒳 \mathcal{X} -large knots observed in knot tables. The second construction is based on braids and orientation-reversing involutions, and is motivated by a detailed analysis of the knot 10 123 10_{123} , also known as the Turk’s head knot T ​ h ​ ( 3 , 5 ) Th(3,5) . In particular, this approach applies to Turk’s head knots T ​ h ​ ( p , q ) Th(p,q) with p p and q q odd, leading us to conjecture that all such knots are 𝒳 \mathcal{X} -large. In doing so, we also present a non-orientable analogue of Thurston’s theorem giving a lower bound on the dimension of character varieties of non-orientable 3-manifolds. 1. Introduction Let K be a knot in S^{3} . Here and throughout the paper, we distinguish knots from links: a knot has a single component, whereas a link has more than one component. We denote by \mathcal{G}(K):=\pi_{1}(S^{3}\setminus K) the knot group of K , and by \mathcal{X}(K) the \mathrm{SL}_{2}(\mathbbm{C}) -character variety of \mathcal{G}(K) , \mathcal{X}(K):=\mathrm{Hom}(\mathcal{G}(K),\mathrm{SL}_{2}(\mathbbm{C}))/\!/_{\mathrm{SL}_{2}(\mathbbm{C})}. Character varieties of knot groups have been extensively studied in the literature. However, most existing results focus on the case in which \mathcal{X}(K) has dimension one. This is partly because, when K is hyperbolic, the geometric representation lies in a one-dimensional component of \mathcal{X}(K) , and partly because many knots (particularly those with few crossings) appear to have one-dimensional character varieties. In this paper, we study knots K for which \dim\mathcal{X}(K)>1 . Here, \dim denotes the (complex) dimension of a component of maximal dimension. We call such knots \mathcal{X} -large . It is well known that if K is \mathcal{X} -large, then K is large, meaning that its complement has a closed essential surface [ 4 ] . However, the converse does not hold: there exist large knots that are not \mathcal{X} -large, such as 8_{16} and 8_{17} [ 12 ] . According to computational data in [ 9 ] , obtained in joint work of the first author with Seonhwa Kim on parabolic representations, there are only three candidates in Rolfsen’s table with at most 10 crossings that can be \mathcal{X} -large: 10_{98} , 10_{99} , and 10_{123} (see Appendix A for details). One motivation of this paper is to determine whether these knots are indeed \mathcal{X} -large and, if so, to provide a topological explanation for this phenomenon, rather than relying on brute-force computations of character varieties. To the author’s knowledge, existing related results (for instance, [ 5 , 13 , 3 ] ) do not apply to these knots. A main goal of the paper is to introduce two new diagrammatic constructions that produce {\mathcal{X}} -large knots. The first construction uses split link diagrams and 2-tangle replacements. We present this construction explicitly in Section 2.2 (Theorem 2.1 ), after explaining the key idea using the knot 10_{98} as a toy example in Section 2.1 . This construction is simple yet powerful: it produces infinitely many \mathcal{X} -large knots from a fixed link diagram. In fact, most \mathcal{X} -large knots observed in our computation arise from this construction. For example, it provides a diagrammatic explanation of why 10_{98} and 10_{99} are \mathcal{X} -large. Unfortunately, at present we do not know of a similar explanation for 10_{123} ; however, our second construction applies to 10_{123} . The second construction uses braids and orientation-reversing involutions. It was motivated by the observation that the knot 10_{123} admits a braid presentation that decomposes into two parts, one of which is conjugate to the other by an orientation-reversing involution (see Figure 6 ). We present a detailed analysis of 10_{123} in Section 3.1 and then generalize this construction in Section 3.2 (Theorem 3.5 ). This construction is applied in particular to Turk’s head knots Th(p,q) with p and q odd, which leads us to conjecture that all such Turk’s head knots are \mathcal{X} -large (Conjecture 3.6 ). We also present a variant of Theorem 3.5 that applies to strongly positive amphichiral knots (Theorem 3.7 ). The proof of Theorem 3.5 requires a non-orientable analogue of Thurston’s theorem giving a lower bound on the dimension of character varieties [ 16 , Theorem 5.6] . Roughly speaking, Thurston’s theorem asserts that under mild assumptions, a (connected) orientable 3-manifold with k boundary tori has a character variety of dimension at least k . For Theorem 3.5 , we require a corresponding statement for non-orientable 3-manifolds with Klein bottle boundary. To the best of the authors’ knowledge, such a non-orientable version does not appear elsewhere in the literature. We therefore postpone the proof of Theorem 3.5 to Section 4 , where we formulate and prove the dimension estimation in the non-orientable setting (Theorem 4.4 ). Using our constructions, we are able to account for all \mathcal{X} -large knots observed in our computations. In particular, as mentioned above, we confirm that 10_{98} , 10_{99} , and 10_{123} are \mathcal{X} -large. Consequently, we obtain the following statement: Up to 10 crossings, the only \mathcal{X} -large knots are 10_{98} , 10_{99} , and 10_{123} . This naturally leads to the question of whether our constructions capture all \mathcal{X} -large knots up to a certain crossing number. However, we do not attempt to address this question in this paper, as determining the list of \mathcal{X} -large knots requires computing full character varieties and their dimensions, which is extremely difficult even for knots with 11 crossings. 2. Rational tangle replacements In this section, we present rational tangle replacements that give rise to \mathcal{X} -large knots. We start with the knot 10_{98} as a toy example and then explain the construction in generality. 2.1. Toy example: the knot 10_{98} Let L be the split link consisting of the knot 3_{1} and the unknot. Here, split means that there exists a 2-sphere separating the unknot from 3_{1} . We fix a diagram of L with a crossing c and two Wirtinger generators g and h around c shown as in Figure 1 (left). Note that the unknot component is colored red. Figure 1. The split link 3_{1}\sqcup O and the knot 10_{98} . Let \rho be a representation of \mathcal{G}(L)=\pi_{1}(S^{3}\setminus L) whose restriction to \mathcal{G}(3_{1}) is irreducible and such that G=\rho(g) and H=\rho(h) are of the form G=\begin{pmatrix}m&1\\ 0&m^{-1}\end{pmatrix},\quad H=\begin{pmatrix}m&0\\ u&m^{-1}\end{pmatrix}\,. (1) Such \rho exists for any non-zero complex numbers m and u , since h is a meridian of the trivial component of the split link L , and g is a meridian of the other non-trivial component. We assume that (u,m) is a solution to the Riley polynomial of 3_{1} ( [ 15 ] ): u+m^{2}-1+m^{-2}=0. (2) This means that the matrices G and H correspond to a non-abelian representation of 3_{1} . More precisely, the knot group of 3_{1} has two generators x and y with one relator xyx=yxy , and the assignment x\mapsto G and y\mapsto H defines a representation, if Equation ( 2 ) is satisfied. Put differently, Equation ( 2 ) implies that GHG=HGH . On the other hand, a small circle around the crossing c meets the diagram of L at four points, corresponding to four Wirtinger generators (two of which coincide). These generators are mapped to G,H,G , and G^{-1}HG , under the representation \rho , as in Figure 2 (left). Since the matrices G and H satisfy GHG=HGH , it follows that HGH^{-1}=G^{-1}HG and HGHG^{-1}H^{-1}=G . Hence, one may replace the crossing c with the rational 2 -tangle shown in Figure 2 (right), keeping the outside of the circle unchanged. This operation yields the knot 10_{98} (see Figure 1 ) and shows that the representation \rho of \mathcal{G}(L) induces a representation of \mathcal{G}(10_{98}) . Figure 2. Replacement of a crossing by a 2-tangle. We now consider matrices A whose conjugation actions fix the matrix G , i.e., AGA^{-1}=G . Explicitly, for a non-zero complex number t , one has A=\begin{pmatrix}t&\frac{t-t^{-1}}{m-m^{-1}}\\ 0&t^{-1}\end{pmatrix}\ \text{ if }m^{2}\neq 1,\quad\text{and}\quad A=\begin{pmatrix}m&t\\ 0&m^{-1}\end{pmatrix}\ \text{ if }m^{2}=1\,. (3) Letting H^{A}:=AHA^{-1} , one has GH^{A}G=H^{A}GH^{A} . Using the matrices H^{A} in place of H , the aforementioned tangle replacement produces representations of \mathcal{G}(L) , each of which induces a representation of \mathcal{G}(10_{98}) . By deforming m and t , we obtain a 2-dimensional family of representations of \mathcal{G}(10_{98}) , and one easily checks that all of these representations have distinct conjugacy classes. This explains that the character variety of 10_{98} has a component of dimension greater than 1. That is, the knot 10_{98} is \mathcal{X} -large. 2.2. General construction In this section, we generalize the tangle replacement discussed in the previous section by using an arbitrary rational 2-tangle diagram instead of the specific one shown in Figure 2 . Given a diagram of a knot or link, replacing a crossing c with a 2-tangle diagram R means removing a small disc containing c and inserting R into it, while keeping the exterior of the disc unchanged; see Figure 3 . The boundary of the disc intersects the diagram at four points. We label two of them by o (resp., u ) if they are connected by the over-pass (resp., under-pass) of c . We then define the c -closure of R as the knot or link obtained from R by joining the two o -points and the two u -points, creating one crossing outside the disc in which the o -points (resp., u -points) are connected by the over-pass (resp., under-pass); see Figure 4 . For example, the c -closure of the 2-tangle diagram in Figure 2 (right) is 3_{1} . Figure 3. Replacement of a crossing by a 2-tangle. Theorem 2.1 . Let L be a split link consisting of a non-trivial knot T and an unknot. Fix a diagram of L such that there is a crossing c at which T intersects with the unknot. Suppose • that a knot K is obtained by replacing the crossing c with a rational 2-tangle diagram R , and • that the c -closure of R is a non-trivial (two-bridge) knot. Then K is \mathcal{X} -large. Proof. Let R_{c} be the two-bridge knot obtained by the c -closure of R . It is proven in [ 15 ] that there is a two-variable polynomial P such that two matrices G and H with the same trace correspond to a non-abelian representation of R_{c} if P(\mathrm{tr}\,G,\mathrm{tr}\,HG)=0 . Precisely, the assignment sending two Wirtinger generators lying above the tangle R to G and H , shown as in Figure 4 , defines a non-abelian representation of \mathcal{G}(R_{c}) if P(\mathrm{tr}\,G,\mathrm{tr}\,HG)=0 . Note that the two Wirtinger generators lying below R are mapped to G^{-1}HG and G under the representation, hence one may replace R with the crossing in Figure 3 (left). Figure 4. The c -closure of a rational 2-tangle R . We now consider the given diagram of the split link L . Let g and h be Wirtinger generators lying above the crossing c that are associated with the over-pass and under-pass of c , respectively. We may assume that h is a meridian of the unknot component of L , and g is a meridian of the other component T ; otherwise, we flip the diagram of L . As L is a split link and as h is a meridian of the unknot component, a representation of \mathcal{G}(L) is determined by a representation of \mathcal{G}(T) together with a single matrix assigned to h , which can be chosen freely. It follows that there exists a representation \rho of \mathcal{G}(L) whose restriction to \mathcal{G}(T) is irreducible and such that G=\rho(g) and H=\rho(h) satisfy \mathrm{tr}\,G=\mathrm{tr}\,H and P(\mathrm{tr}\,G,\mathrm{tr}\,HG)=0 . As explained, such \rho induces a representation of \mathcal{G}(K) by replacing the crossing c with the 2-tangle R . On the other hand, for fixed G , there is a 1-dimensional family of matrices A satisfying AGA^{-1}=G . It follows that one can deform H into H^{A}=AHA^{-1} while keeping G fixed. Using the matrices H^{A} in place of H , we obtain a 1-dimensional family of representations of \mathcal{G}(L) , each of which induces a representation of \mathcal{G}(K) . Since the restriction of \rho to \mathcal{G}(T) is irreducible, there exists an element g^{\prime}\in\mathcal{G}(T) such that the invariant subspace of G^{\prime}=\rho(g^{\prime}) is distinct from that of G . Since varying A changes the trace of G^{\prime}H^{A} , those induced representations of \mathcal{G}(K) have distinct conjugacy classes. The previous argument provides a curve in \mathcal{X}(K) for each value of the trace of G . As the A -polynomial is nontrivial ( [ 6 , 1 , 2 ] ), we obtain arbitrarily many curves in \mathcal{X}(K) . An algebraic variety has finitely many irreducible components, so a component of \mathcal{X}(K) has dimension greater than 1. ∎ The above theorem is effective in producing \mathcal{X} -large knots. For example, by taking T to be the knot 3_{1} , one can deduce that 10_{99} , 11a_{132} , 11a_{157} , and 12a_{923} are \mathcal{X} -large; see Figure 5 . Figure 5. Examples of \mathcal{X} -large knots. 3. Braids and orientation-reversing involutions In this section, we construct \mathcal{X} -large knots from braids and orientation-reversing involutions. We first explain how it works for the knot 10_{123} as a toy example. 3.1. Toy example: the knot 10_{123} Let B_{3} be the braid group of 3 strands with standard generators \sigma_{1}^{\pm 1} and \sigma_{2}^{\pm 1} . Let b=\sigma_{1}\sigma_{2}^{-1}\sigma_{1}\sigma_{2}^{-1}\sigma_{1} and b^{*}=\sigma_{2}^{-1}\sigma_{1}\sigma_{2}^{-1}\sigma_{1}\sigma_{2}^{-1} . Then the closure of the composition bb^{*} is the knot 10_{123} . See Figure 6 ; we draw braids left-to-right, so the composition is horizontal. The braid b is related to b^{*} under the correspondence \sigma_{1}^{\pm 1}\longmapsto\sigma_{2}^{\mp 1}\quad\text{and}\quad\sigma_{2}^{\pm 1}\longmapsto\sigma_{1}^{\mp 1}\,. (4) Viewing braids as mapping classes of the punctured disc relative to the boundary, via the isomorphism B_{3}\cong\mathrm{MCG}(D^{2}\setminus\{p_{1},p_{2},p_{3}\},\ \partial D^{2}), we have b^{\ast}=\tau b\tau where \tau is the reflection that fixes p_{2} and interchanges p_{1} and p_{3} . Figure 6. Braid presentation of the knot 10_{123} . Let g_{i} be the Wirtinger generators on the left of the braid b and h_{i} be those on the right for i=1,2,3 . If each g_{i} is assigned an \mathrm{SL}_{2}(\mathbbm{C}) -matrix G_{i} , then along the braid b , this assignment determines matrices H_{i} corresponding to h_{i} . Regarding the triples (G_{1},G_{2},G_{3}) and (H_{1},H_{2},H_{3}) as representations of the free group F_{3} of rank 3, the map sending (G_{1},G_{2},G_{3}) to (H_{1},H_{2},H_{3}) defines an automorphism of the representation variety of F_{3} . Restricting our attention to the case where (G_{1},G_{2},G_{3}) is irreducible, this induces an automorphism of \mathcal{X}^{\mathrm{irr}}(F_{3}) by sending [G_{1},G_{2},G_{3}] to [H_{1},H_{2},H_{3}] . Here \mathcal{X}^{\mathrm{irr}}(F_{3}) denotes the character variety of irreducible representations of F_{3} , and the square brackets denote the conjugacy classes. Lemma 3.1 . There exists a subvariety U\subset\mathcal{X}^{\mathrm{irr}}(F_{3}) whose dimension is at least 2 and such that each point [G_{1},G_{2},G_{3}]\in U satisfies \mathrm{tr}\,G_{1}=\mathrm{tr}\,G_{2}=\mathrm{tr}\,G_{3},\quad\mathrm{tr}\,G_{1}G_{2}G_{3}\neq\pm 2,\quad\text{and}\quad[H_{1},H_{2},H_{3}]=[G_{3}^{-1},G_{2}^{-1},G_{1}^{-1}]. (5) Proof. The character variety of F_{3} is 6-dimensional, hence the first condition in ( 5 ) defines a 4-dimensional subvariety V . Precisely, its coordinate ring \mathbbm{C}[V] is given by \mathbbm{C}[x,y,z,b,c]/P where P=xyz+x^{2}+y^{2}+z^{2}-b(x+y+z)-c. Here the variables x,y,z,b and c are related to G_{1},G_{2},G_{3} as follows: \displaystyle(x,y,z)=(\mathrm{tr}\,G_{1}G_{2},\ \mathrm{tr}\,G_{1}G_{3},\ \mathrm{tr}\,G_{2}G_{3}), \displaystyle b=\mathrm{tr}\,G_{1}(\mathrm{tr}\,G_{1}+\mathrm{tr}\,G_{1}G_{2}G_{3}),\quad c=4-3(\mathrm{tr}\,G_{1})^{2}-(\mathrm{tr}\,G_{1}G_{2}G_{3})^{2}-(\mathrm{tr}\,G_{1})^{3}\mathrm{tr}\,G_{1}G_{2}G_{3}. See, for instance, [ 8 ] for details. Moreover, a direct computation shows that the braid group B_{3} acts on \mathbbm{C}[V] as follows: \displaystyle\sigma_{1}\cdot(x,y,z,b,c) \displaystyle=(x,\,z,\,b-y-xz,\,b,\,c); \displaystyle\sigma_{1}^{-1}\cdot(x,y,z,b,c) \displaystyle=(x,\,b-z-xy,\,y,\,b,\,c); \displaystyle\sigma_{2}\cdot(x,y,z,b,c) \displaystyle=(y,\,b-x-yz,\,z,\,b,\,c); \displaystyle\sigma_{2}^{-1}\cdot(x,y,z,b,c) \displaystyle=(b-y-xz,\,x,\,z,\,b,\,c). Let (X,Y,Z,b,c):=b\cdot(x,y,z,b,c) so that (X,Y,Z)=(\mathrm{tr}\,H_{1}H_{2},\mathrm{tr}\,H_{1}H_{3},\mathrm{tr}\,H_{2}H_{3}) . We first claim that Y-b+y+xz\notin(X-z,\,Z-x) (6) where the right-hand side is the ideal in \mathbbm{C}[V] generated by X-z and Z-x . To prove this, we quotient \mathbbm{C}[V] by (b,x^{2},y^{2},z^{2}) . A straightforward computation then shows that P=xyz-c and (X,Y,Z)=(y,\,-x-yz,\,z) in the quotient ring. Consequently, X-z=0 and Z-x=0 force x=y=z . However, Y-y=-x-y-yz=-2x\neq 0 , which proves the claim. We now consider the subvariety V_{0} of V defined by x=Z and z=X . Since V is 4-dimensional, each component of V_{0} has dimension at least 2. As the actions of \sigma_{1}^{\pm 1} and \sigma_{2}^{\pm 1} preserve the polynomial P , we have xyz+x^{2}+y^{2}+z^{2}-b(x+y+z)-c=XYZ+X^{2}+Y^{2}+Z^{2}-b(X+Y+Z)-c and thus (Y-y)(Y-b+y+xz)=0 on V_{0} . It follows from ( 6 ) that there exists a point of V_{0} at which Y-b+y+xz\neq 0 . On a small neighborhood U\subset V_{0} of that point, the same inequality holds, which forces Y=y . It follows that (X,Y,Z)=(z,y,x) holds in U . In addition, we may assume that \mathrm{tr}\,G_{1}G_{2}G_{3}\neq\pm 2 in U , as we have not specified c , which can vary \mathrm{tr}\,G_{1}G_{2}G_{3} . Summing up, the following equations holds in U : \displaystyle\mathrm{tr}\,G_{1} \displaystyle=\mathrm{tr}\,G_{2}=\mathrm{tr}\,G_{3}, \displaystyle\mathrm{tr}\,H_{1}H_{2} \displaystyle=\mathrm{tr}\,G_{2}G_{3}=\mathrm{tr}\,G_{3}^{-1}G_{2}^{-1}, \displaystyle\mathrm{tr}\,H_{1}H_{3} \displaystyle=\mathrm{tr}\,G_{1}G_{3}=\mathrm{tr}\,G_{3}^{-1}G_{1}^{-1}, \displaystyle\mathrm{tr}\,H_{2}H_{3} \displaystyle=\mathrm{tr}\,G_{1}G_{2}=\mathrm{tr}\,G_{2}^{-1}G_{1}^{-1}, as well as \mathrm{tr}\,G_{1}G_{2}G_{3}\neq\pm 2 . The equality \mathrm{tr}\,G_{1}=\mathrm{tr}\,G_{2}=\mathrm{tr}\,G_{3} implies \mathrm{tr}\,H_{i}=\mathrm{tr}\,G_{j}=\mathrm{tr}\,G_{j}^{-1} for any i,j , and the fact G_{1}G_{2}G_{3}=H_{1}H_{2}H_{3} implies \mathrm{tr}\,H_{1}H_{2}H_{3}=\mathrm{tr}\,G_{1}G_{2}G_{3}=\mathrm{tr}\,G_{3}^{-1}G_{2}^{-1}G_{1}^{-1}. It follows that two irreducible representations (H_{1},H_{2},H_{3}) and (G_{3}^{-1},G_{2}^{-1},G_{1}^{-1}) have the same character. As irreducible representations are determined by their characters, we conclude that [H_{1},H_{2},H_{3}]=[G_{3}^{-1},G_{2}^{-1},G_{1}^{-1}] holds in U . This completes the proof. ∎ The last equation in ( 5 ) means that H_{1}=AG_{3}^{-1}A^{-1},\quad H_{2}=AG_{2}^{-1}A^{-1},\quad H_{3}=AG_{1}^{-1}A^{-1} (7) for some A\in\mathrm{SL}_{2}(\mathbbm{C}) . We write these compactly as (H_{1},H_{2},H_{3})=A(G_{3}^{-1},G_{2}^{-1},G_{1}^{-1})A^{-1}. (8) Lemma 3.2 . If G_{1}G_{2}G_{3}\neq\pm I , then the matrix A in ( 8 ) satisfies that A^{2}=-I . Proof. If we conjugate each G_{i} by a matrix B , then each H_{i} is also conjugated by the same matrix B . Since Equation ( 8 ) implies that for i=1,2,3 , BH_{i}B^{-1}=(BAB^{-1})\,(BG_{4-i}^{-1}B^{-1})\,(BA^{-1}B^{-1}), (9) hence A is also conjugated by B . In particular, the trace of A remains unchanged. On the other hand, since the relation G_{1}G_{2}G_{3}=H_{1}H_{2}H_{3} holds, one has H_{1}H_{2}H_{3}=A(G_{1}G_{2}G_{3})^{-1}A^{-1}=A(H_{1}H_{2}H_{3})^{-1}A^{-1}\,. (10) Up to conjugation, we may assume that H_{1}H_{2}H_{3} is upper-triangular: it is either \begin{pmatrix}m&0\\ 0&m^{-1}\end{pmatrix}\text{ for }m^{2}\neq 1\quad\text{or}\quad\begin{pmatrix}\pm 1&\alpha\\ 0&\pm 1\end{pmatrix}\text{ for }\alpha\neq 0\,. In any case, a straightforward computation shows that Equation ( 10 ), combined with the fact G_{1}G_{2}G_{3}=H_{1}H_{2}H_{3}\neq\pm I , forces the trace of A to be 0, and we deduce that A^{2}=-I from the Cayley-Hamilton formula. ∎ Recall that b^{\ast}=\tau b\tau is related to b under the correspondence ( 4 ). It follows that if b sends (G_{1},G_{2},G_{3}) to (H_{1},H_{2},H_{3}) , then b^{\ast} sends (G_{3}^{-1},G_{2}^{-1},G_{1}^{-1}) to (H_{3}^{-1},H_{2}^{-1},H_{1}^{-1}) . Therefore, along the composition bb^{\ast} , we have (G_{1},G_{2},G_{3})\overset{b}{\longmapsto}A(G_{3}^{-1},G_{2}^{-1},G_{1}^{-1})A^{-1}\overset{b^{\ast}}{\longmapsto}A^{2}(G_{1},G_{2},G_{3})A^{-2}=(G_{1},G_{2},G_{3}) for [G_{1},G_{2},G_{3}]\in U . Note that the last equality follows from Lemma 3.2 . This proves that each point of U induces a representation of the closure of bb^{\ast} , which is the knot 10_{123} . As U has dimension at least 2, we conclude that 10_{123} is \mathcal{X} -large. Remark 3.3 . The argument in this section applies to any braids b\in B_{3} satisfying the inequality ( 6 ). For example, if we take b=\sigma_{1}\sigma_{2}^{-1}\sigma_{2}^{-1}\sigma_{1}\sigma_{1} , then b^{\ast}=\sigma_{2}^{-1}\sigma_{1}\sigma_{1}\sigma_{2}^{-1}\sigma_{2}^{-1} , and the closure of bb^{\ast} is the knot 10_{99} (which is \mathcal{X} -large, shown as in Section 2 ). Note, however, that the inequality ( 6 ) does not always hold. For instance, it fails for b=\sigma_{1}\sigma_{2}\sigma_{1}\sigma_{2}\sigma_{1} . 3.2. General construction The argument in the previous section for the knot 10_{123} gives rise to the following question. Let B_{n} denote the braid group with n strands. We view braids as mapping classes of the punctured disc relative to the boundary, via the isomorphism B_{n}\cong\mathrm{MCG}(D^{2}\setminus\{p_{1},\dotsc,p_{n}\},\partial D^{2}). The mapping class group \mathrm{MCG}(D^{2}\setminus\{p_{1},\dotsc,p_{n}\}) (not relative to the boundary) acts on B_{n} by conjugation. We are interested in the action of orientation-reversing involutions, namely, in classes \tau\in\mathrm{MCG}(D^{2}\setminus\{p_{1},\dotsc,p_{n}\}) that reverse the orientation and satisfy \tau^{2}=\mathrm{Id} . Question . Let K be a knot obtained by the closure of a braid bb^{*} with b\in B_{n} and b^{*}=\tau b\tau , where \tau\in\mathrm{MCG}(D^{2}\setminus\{p_{1},\dotsc,p_{n}\}) is an orientation-reversing involution. Is K \mathcal{X} -large? We first give a necessary and sufficient condition for K to be a knot. Lemma 3.4 . Let b and b^{\ast}=\tau b\tau be as above. Then the closure of bb^{\ast} is a knot if and only if n is odd and \pi(b\tau) is an n -cycle, where \pi:B_{n}\rightarrow S_{n} is the canonical projection from the braid group to the symmetric group. Proof. Since bb^{\ast}=(b\tau)^{2} , its image under \pi is a square in S_{n} and therefore has even sign. Hence it cannot be an n -cycle when n is even. When n is odd, it is straightforward to see that \pi((b\tau)^{2}) is an n -cycle if and only if \pi(b\tau) is an n -cycle. ∎ Theorem 3.5 below answers the above question under mild assumptions. Its proof is based on a non-orientable version of a dimension estimate for representation varieties proved by Thurston in his notes [ 16 , Theorem 5.6] . More precisely, its proof relies on a dimension estimation for the representation variety of the mapping torus M_{b\tau} of b\tau : M_{b\tau}=(D^{2}\times[0,1]\setminus b)/_{\sim} where the quotient identifies D^{2}\times\{0\} with D^{2}\times\{1\} via \tau . Note that M_{b\tau} is a non-orientable 3-manifold whose boundary consists of two Klein bottles, due to Lemma 3.4 . To the best of the authors’ knowledge, this non-orientable version does not appear elsewhere in the literature. We therefore postpone the proof of Theorem 3.5 to Section 4 , where we formulate and prove the dimension estimation in the non-orientable setting. Theorem 3.5 . Let K be a knot obtained by the closure of a braid bb^{*} with b\in B_{n} and b^{*}=\tau b\tau , where \tau\in\mathrm{MCG}(D^{2}\setminus\{p_{1},\dotsc,p_{n}\}) is an orientation-reversing involution. Suppose that the mapping torus M_{b\tau} of b\tau admits a representation \rho into \mathrm{SL}_{2}(\mathbbm{C}) such that (a) the restriction of \rho to each boundary Klein bottle of M_{b\tau} is irreducible, and (b) the restriction of \rho to the orientable double cover of M_{b\tau} is irreducible. Then K is \mathcal{X} -large. A typical example of an orientation-reversing involution \tau\in\mathrm{MCG}(D^{2}\setminus\{p_{1},\dotsc,p_{n}\}) is the reflection that interchanges p_{i} and p_{n+1-i} for 1\leq i\leq n . In this case, b^{\ast}=\tau b\tau is related to b under the correspondence \sigma_{i}^{\pm 1}\longmapsto\sigma_{n-i}^{\mp 1},\quad i=1,\ldots,n-1. Here \sigma_{1},\ldots,\sigma_{n-1} are the standard generators of the braid group B_{n} . Interesting examples of knots admitting braid presentations of the form bb^{\ast} are Turk’s head knots Th(p,q) with p and q odd. A standard braid presentation of Th(p,q) is \left(\sigma_{1}\,\sigma_{2}^{-1}\cdots\sigma_{p-2}\,\sigma_{p-1}^{-1}\right)^{q}\in B_{p}. Define O=\sigma_{1}\sigma_{3}\cdots\sigma_{p-2},\quad E=\sigma_{2}^{-1}\sigma_{4}^{-1}\cdots\sigma_{p-1}^{-1},\quad b=(OE)^{\frac{q-1}{2}}O. Since \sigma_{i} and \sigma_{j} commute whenever |i-j|>1 , one has O^{\ast}=E , E^{\ast}=O , and b^{\ast}=(EO)^{\frac{q-1}{2}}E . It then follows that the closure of bb^{\ast} is Th(p,q) ; Figure 7 illustrates this decomposition. Motivated by this observation and supported by computational experiments, we propose the following. Conjecture 3.6 . Turk’s head knots Th(p,q) with p and q odd are \mathcal{X} -large. Figure 7. A standard braid presentation of Th(5,3) and the choice of b . Another example of an orientation-reversing involution \tau\in\mathrm{MCG}(D^{2}\setminus\{p_{1},\dotsc,p_{n}\}) is the reflection that fixes each p_{i} and reverses the orientation of a small loop around p_{i} . In this case, b^{\ast}=\tau b\tau is related to b under the correspondence \sigma_{i}^{\pm 1}\longmapsto\sigma_{i}^{\mp 1},\quad i=1,\ldots,n-1. Namely, b^{\ast} is the mirror image of b . This naturally leads us to strongly positive amphichiral knots, which are knots that admit a diagram that is mapped to its mirror image by a rotation of \pi preserving the orientation [ 11 ] . By definition, a strongly positive amphichiral knot is obtained by gluing an oriented tangle and its mirror. To be precise, let \epsilon=(\epsilon_{1},\ldots,\epsilon_{n})\in\{\pm\}^{n} be an n -tuple of signs and \mathrm{Mor}(\epsilon,-\epsilon) denote the set of oriented (n,n) -tangles in D^{2}\times[0,1] such that the orientation at p_{i}\times\{0\} (resp., p_{i}\times\{1\} ) matches with the sign \epsilon_{i} (resp., -\epsilon_{i} ) for 1\leq i\leq n . Note that braids of n strands may be regarded as elements of \mathrm{Mor}(+^{n},-^{n}) . For an oriented tangle t\in\mathrm{Mor}(\epsilon,-\epsilon) , we denote by t^{\ast} its mirror image, obtained by reversing the over/under information at every crossing. Clearly, b^{\ast} also lies in \mathrm{Mor}(\epsilon,-\epsilon) . By gluing the boundaries of t and t^{\ast} naturally in an orientation-preserving way, we obtain an oriented knot or link K=t\cup t^{\ast} , which is, by definition, strongly positive amphichiral. See Figure 8 for examples. We also denote by M_{t\tau} the mapping torus of t\tau : M_{t\tau}:=(D^{2}\times[0,1]\setminus t)/_{\sim} where the quotient identifies D^{2}\times\{0\} with D^{2}\times\{1\} via the reflection \tau . It is a non-orientable 3-manifold, and when K is a knot, its boundary consists of two Klein bottles. Then, by replacing M_{b\tau} with M_{t\tau} in Theorem 3.5 and its proof, we obtain the following. Theorem 3.7 . Let t\in\mathrm{Mor}(\epsilon,-\epsilon) be an oriented tangle where \epsilon is a tuple of signs. Suppose that a knot K is obtained by gluing t to its mirror image t^{\ast}\in\mathrm{Mor}(\epsilon,-\epsilon) along their boundaries, and that M_{t\tau} admits a representation \rho into \mathrm{SL}_{2}(\mathbbm{C}) such that (a) the restriction of \rho to each boundary Klein bottle of M_{t\tau} is irreducible, and (b) the restriction of \rho to the orientable double cover of M_{t\tau} is irreducible. Then the knot K is (by construction strongly positive amphichiral and) \mathcal{X} -large. Figure 8. Symmetric diagrams of 10_{99} and 12a_{1019} . The diagram of 12a_{1019} is taken from [ 10 ] . 4. Dimension estimate à la Thurston In this section, we present a non-orientable analogue of Thurston’s theorem [ 16 , Theorem 5.6] giving a lower bound on the dimension of character varieties. In the following, for a manifold M , the variety of \mathrm{SL}_{2}(\mathbbm{C}) -representations is denoted by \mathcal{R}(M)=\mathrm{Hom}(\pi_{1}(M),\,\mathrm{SL}_{2}(\mathbb{C})) , and the character of \rho\in\mathcal{R}(M) is denoted by \chi_{\rho} . 4.1. Klein bottle Let K^{2} be the Klein bottle. Its fundamental group has the presentation \pi_{1}(K^{2})\cong\langle a,b\mid aba^{-1}=b^{-1}\rangle. (11) Lemma 4.1 . Let \rho\in\mathcal{R}(K^{2}) . Then there are three possibilities up to conjugation: (1) \rho(b)=\pm\left(\begin{smallmatrix}1&0\\ 0&1\end{smallmatrix}\right) and \rho(a) is arbitrary. (2) \rho(b)=\pm\left(\begin{smallmatrix}\lambda&0\\ 0&\lambda^{-1}\end{smallmatrix}\right) with \lambda\neq\pm 1 and \rho(a)=\pm\left(\begin{smallmatrix}0&1\\ -1&0\end{smallmatrix}\right) . (3) \rho(b)=\pm\left(\begin{smallmatrix}1&1\\ 0&1\end{smallmatrix}\right) and \rho(a)=\pm\left(\begin{smallmatrix}i&0\\ 0&-i\end{smallmatrix}\right) . The proof of the lemma is straightforward by considering the possible conjugacy classes for \rho(b) and using the relation aba^{-1}=b^{-1} . Corollary 4.2 . The character variety \mathcal{X}(K^{2}) has three components. Exactly one of them contains irreducible representations, and this component has dimension 1. Proof. Representation in Case (1) of Lemma 4.1 are reducible and yields two component of \mathcal{X}(K^{2}) , characterized by \chi(b)=\pm 2 (and \chi(ab)=\pm\chi(a) ). Representations in Case (3) form a single point, lying in one of the previous components. Finally, representations in Case (2) are irreducible and yield the third component, which is 1-dimensional and characterized by \chi(a)=0 (and \chi(ab)=0 ). ∎ Lemma 4.3 . Let \rho\in\mathcal{R}(K^{2}) . If \rho is irreducible, then it is a smooth point of \mathcal{R}(K^{2}) , and its Zariski tangent space has dimension 4 . Proof. The component R_{0} of \mathcal{R}(K^{2}) containing \rho has dimension 4, because the corresponding component of \mathcal{X}(K^{2}) has dimension 1 and \rho is irreducible. The Zariski tangent space at \rho also has dimension 4; see [ 14 , Proposition 4.1] . As the Zariski tangent space has the same dimension as R_{0} , \rho is a smooth point. ∎ 4.2. Dimension estimate Recall that any (connected) non-orientable manifold M admits the orientable double cover \widetilde{M}\to M , and it fundamental group \pi_{1}(\widetilde{M}) consists precisely of the orientation-preserving elements of \pi_{1}(M) . Theorem 4.4 . Let M be a compact non-orientable 3-manifold whose boundary consists of k Klein bottles K^{2}_{1},\ldots,K^{2}_{k} . Let \rho\in\mathcal{R}(M) be a representation such that (a) the restriction of \rho to each K^{2}_{i} is irreducible, and (b) the restriction of \rho to the orientable double cover \widetilde{M} is irreducible. Then any component X_{0} of \mathcal{X}(M) containing \chi_{\rho} has dimension at least k ; that is, \dim(X_{0})\geq k. Proof. For each boundary Klein bottle K_{i}^{2} , consider an embedded loop \gamma_{i}\subset M with base point in K^{2}_{i} that intersects \partial M^{3} only at its base point. In particular, K_{i}^{2}\cup\gamma_{i}\cong K^{2}\vee S^{1} . We fix such \gamma_{i} so that it is orientation-preserving and the restriction of \rho to \pi_{1}(T^{2}_{i}\cup\gamma_{i}) is irreducible, where T^{2}_{i} is the orientable double cover of K^{2}_{i} . We may assume that the embedded curves \gamma_{1},\dotsc,\gamma_{k} are pairwise disjoint. Consider a compact manifold M^{\prime} obtained from M by drilling out a small neighborhood of each \gamma_{i} . Then its boundary \partial M^{\prime} consists of k non-orientable surfaces of genus 4: \partial M^{\prime}=N_{4,1}\sqcup\dotsb\sqcup N_{4,k}. Using the Poincaré duality with \mathbb{Z}/2\mathbb{Z} -coefficients, we compute that the Euler characteristic of M^{\prime} is equal to -k : \chi(M^{\prime})=\frac{1}{2}\chi(\partial M^{\prime})=\frac{1}{2}\sum_{i=1}^{k}\chi(N_{4,i})=-k. This means that M^{\prime} has the homotopy type of a CW-complex with one vertex, k+1+r edges and r faces. Therefore, \pi_{1}(M^{\prime}) admits a presentation with k+1+r generators and r relations, which yields \dim\mathcal{R}(M^{\prime})\geq 3(k+1+r)-3r=3k+3. (12) Let R_{0} be the component of \mathcal{R}(M) containing \rho and corresponding to the component X_{0} of \mathcal{X}(M) . Viewing \mathcal{R}(M) as a subvariety of \mathcal{R}(M^{\prime}) , we obtain the commutative diagram (after dealing with base points): {\lx@inpgf@ignorespaces R_{0}\subset\mathcal{R}(M)} {\lx@inpgf@ignorespaces\mathcal{R}(M^{\prime})} {\lx@inpgf@ignorespaces\prod_{i}\mathcal{R}(K^{2}_{i}\cup\gamma_{i})} {\lx@inpgf@ignorespaces\prod_{i}\mathcal{R}(N_{4,i})} \scriptstyle{\lx@inpgf@ignorespaces\mathrm{res}} (13) where \mathrm{res} denotes the restriction map, and the horizontal arrows are inclusions. Note that \mathcal{R}(M)=\{\rho\in\mathcal{R}(M^{\prime})\mid\mathrm{res}(\rho)\in\prod_{i}\mathcal{R}(K^{2}_{i}\cup\gamma_{i})\}\,. We next require [ 14 , Lemma 6.12] , which can be regarded as a local version of [ 7 , Proposition 1] . For the reader’s convenience, we recall [ 14 , Lemma 6.12] below. Lemma 4.5 . Let W be a complex (analytic) variety and W^{\prime}\subset W a complex subvariety, both irreducible and smooth. Let X be a variety and f\colon X\to W a holomorphic map. Then each irreducible component Y of f^{-1}(W^{\prime})\subset X has codimension \mathrm{codim}(Y,X)\leq\mathrm{codim}(W^{\prime},W). To apply Lemma 4.5 , one compares the diagram ( 13 ) with the following diagram: {\lx@inpgf@ignorespaces Y\subset f^{-1}(W^{\prime})} {\lx@inpgf@ignorespaces X} {\lx@inpgf@ignorespaces W^{\prime}} {\lx@inpgf@ignorespaces W} \scriptstyle{\lx@inpgf@ignorespaces f} \scriptstyle{\lx@inpgf@ignorespaces f} (14) We verify that the hypotheses of Lemma 4.5 are fulfilled: (1) Each \mathcal{R}(N_{4,i}) is smooth and has dimension 9: By the choice of \gamma_{i} , \rho restricted to the orientable double cover of N_{4,i} is irreducible. Then, by [ 14 , Proposition 4.2] , each \mathcal{R}(N_{4,i}) is smooth and has dimension -3\chi(N_{4,i})+3=9 . (2) \mathcal{R}(K^{2}_{i}\cup\gamma_{i}) is smooth and has dimension 7: This assertion follows from Lemma 4.3 together with the isomorphism \mathcal{R}(K^{2}_{i}\cup\gamma_{i})=\mathcal{R}(\pi_{1}(K^{2}_{i})*\mathbb{Z})\cong\mathcal{R}(K^{2}_{i})\times\mathrm{SL}_{2}(\mathbb{C}). (3) The map \mathcal{R}(K^{2}_{i}\cup\gamma_{i})\hookrightarrow\mathcal{R}(N_{4,i}) is an immersion: The tangent map of this inclusion is the map induced at the level of cocycles Z^{1}(\pi_{1}(K^{2}_{i}\cup\gamma_{i}),\mathfrak{sl}_{2}(\mathbb{C}))\to Z^{1}(\pi_{1}(N_{4,i}),\mathfrak{sl}_{2}(\mathbb{C})). This morphism is injective, because it is induced by a surjection of fundamental groups \pi_{1}(N_{4,i})\twoheadrightarrow\pi_{1}(K^{2}_{i}\cup\gamma_{i}) . Then, by applying Lemma 4.5 , one has \mathrm{codim}(R_{0},\mathcal{R}(M^{\prime}))\leq\sum_{i}\mathrm{codim}(\mathcal{R}(N_{4,i}),\,\mathcal{R}(K^{2}_{i}\cup\gamma_{i}))=2k. Hence, with the inequality ( 12 ), one has \dim(R_{0})\geq\dim(\mathcal{R}(M^{\prime}))-2k\geq 3k+3-2k=k+3. This proves the theorem, because \dim(X_{0})=\dim(R_{0})-3\geq k . ∎ 4.3. Proof of Theorem 3.5 Recall that the mapping torus M_{b\tau}=(D^{2}\times[0,1]\setminus b)/_{\sim} is a non-orientable 3-manifold with Klein bottle boundary. Lemma 3.4 shows that the boundary of M_{b\tau} consists two boundary Klein bottles. Under the assumptions (a) and (b) in the theorem, Theorem 4.4 is applied to M_{b\tau} , implying that the character variety \mathcal{X}(M_{b\tau}) has a component X_{0} of dimension at least 2. Fix a representation \rho of \pi_{1}(M_{b\tau}) with \chi_{\rho}\in X_{0} and let \rho_{i} be the restriction of \rho to D^{2}\times\{i\} minus the punctures for i=0,1 (so that \rho_{i} ’s are representations of a free group). Then \rho_{i} ’s are conjugate to each other. Precisely, one has \rho_{1}=\rho(a)\,\rho_{0}\,\rho(a)^{-1} where a is a loop obtained by the quotient of [0,1] . Note that a is one of the generators of the Klein bottle. Lemma 4.1 shows that \rho(a^{2})=-I and thus \rho induces a representation of \mathcal{G}(K) ; recall that the knot K is the closure of bb^{\ast} . As this works for every representation \rho with \chi_{\rho}\in X_{0} and as X_{0} has dimension at least 2, we conclude that K is \mathcal{X} -large. Appendix A Parabolic representations In this appendix, we discuss parabolic representations, namely, representations that send a meridian to a matrix of trace 2. More precisely, we restrict our attention to the subvariety \mathcal{X}^{par}(K)=\mathrm{tr}\,_{\mu}^{-1}(2)\subset\mathcal{X}(K) where \mathrm{tr}\,_{\mu}:\mathcal{X}(K)\rightarrow\mathbbm{C} denotes the trace function associated with a meridian of a knot K . If K is \mathcal{X} -large, i.e., if \dim\mathcal{X}(K)>1 , then clearly \dim\mathcal{X}^{par}(K)>0 . Therefore, a necessary condition for K to be \mathcal{X} -large is that \mathcal{X}^{par}(K) has positive dimension. In [ 9 ] , \mathcal{X}^{par}(K) is explicitly computed for knots K with at most 12 crossings. Knots K for which \mathcal{X}^{par}(K) has positive dimension are as follows. 10_{98},\,10_{99},\,10_{123},\,11a_{43},\,11a_{44},\,11a_{47},\,11a_{57},\,\ldots. To give a topological explanation for this phenomenon of positive dimension, we modify Theorem 2.1 as follows. Theorem A.1 . Let T be a link with two components T_{1} and T_{2} and L the split link consisting of T together with an unknot O . Assume that at least one of the T_{i} —say T_{1} —is nontrivial and that there is a diagram of L such that • T_{2} meets T_{1} only along a single arc of T_{1} 1 1 1 An arc of T_{1} means an arc of the diagram of T_{1} obtained by ignoring the other components T_{2} and O . , and • there is a crossing c_{i} at which T_{i} meets the unknot O for i=1,2 . Suppose that • a knot K is obtained by replacing the crossing c_{i} with a rational 2 -tangle diagram R_{i} , for i=1,2 , and • the c_{i} -closure of R_{i} is a non-trivial (two-bridge) knot for i=1,2 . Then \mathcal{X}^{par}(K) has positive dimension. Proof. For X\neq\pm I\in\mathrm{SL}_{2}(\mathbbm{C}) , we denote by C(X) the centralizer C(X)=\{A\in\mathrm{SL}_{2}(\mathbbm{C})\mid AX=XA\}. It is well known that C(X) is one-dimensional, and that if \mathrm{tr}\,(X)=\pm 2 , then every element of C(X) also has trace \pm 2 (see, for example, Equation ( 3 )). By assumption, T_{2} intersects T_{1} along a single arc. Let g be the Wirtinger generator corresponding to this arc. This allows us to extend any parabolic representation \rho of \mathcal{G}(T_{1}) to a parabolic representation of \mathcal{G}(T) by sending (\text{every Wirtinger generator on }T_{2})\longmapsto A\in C(\rho(g)), where A\in C(\rho(g)) satisfies \mathrm{tr}\,(A)=2 . Such choices of A form a 1-dimensional family, parameterized by a single variable, say t . Extending the representation further to \mathcal{G}(L)=\mathcal{G}(T\sqcup O) , we may assign an arbitrary trace– 2 matrix to a meridian of the unknot O . This introduces two additional free variables, say x and y ; explicitly, the trace– 2 matrices can be parameterized by x and y as (x,y)\longleftrightarrow\begin{pmatrix}1+xy&x^{2}\\ -y^{2}&1-xy\end{pmatrix}. We now impose the conditions required for replacing each crossing c_{i} by the tangle R_{i} , for i=1,2 . The replacement of c_{1} by R_{1} yields an equation involving only x and y , while the replacement of c_{2} by R_{2} yields an equation involving t , x , and y . It follows that the solution space of these two equations has dimension at least one. Since each solution induces a parabolic representation of K , we conclude that \mathcal{X}^{p}(K) has positive dimension. ∎ Figure 9 presents several examples with 11 crossings that can be obtained from Theorem A.1 where T consists of the trefoil knot 3_{1} together with a linked unknot. It is natural to ask whether these knots are \mathcal{X} -large; we leave this question for future work. Figure 9. Examples of knots K for which \mathcal{X}^{par}(K) has positive dimension.