WorksMathematics
Prime Knots with Augmentation-Rank Gaps at Unbounded Bridge Number
Abstract
We construct explicit braid families containing infinitely many distinct hyperbolic prime knots of each bridge number and each sufficiently large , with complex augmentation rank at most and meridional rank . Alternating-group colorings force meridional generators, while two fixed meridian words let a local rank-one obstruction persist as the braid grows. An oriented diagram isotopy realizes the associated two-cusped exteriors as fillings of a single three-cusped exterior. Rigorous interval certificates establish hyperbolicity of that exterior and the two small-bridge exteriors used in the filling argument.
Introduction
How far can one get in determining the bridge number of a knot using rank-one meridian representations? For a knot , the meridional rank is the least number of positive meridians generating its group, and the bridge number is the least number of maxima in a bridge presentation. The conventional complex augmentation rank uses the cord algebra with and meridian parameter different from . By Cornwell’s correspondence it is the maximum dimension of an irreducible complex representation in which a meridian is diagonalizable with eigenvalues , where [1 Theorem 1.2 and Definition 1.8]. Meridians are conjugate, so this condition holds for every meridian. The inequalities
give two ways to bound bridge number from below. We construct prime knots of arbitrarily large bridge number for which the first bound falls short and the second is an equality.
Read braid words from left to right, from top to bottom, and write for the positive crossing at positions . Let
For , put and embed in on the first four strands. Define
An empty product is the identity.
Theorem 1. For every integer and , is a knot and
For each and each sufficiently large , all sufficiently large give hyperbolic prime knots, and infinitely many distinct knot types occur for that fixed .
Cornwell proved for the four knots , , , and their mirrors. His rank-gap classification assumes crossing number at most ten and bridge number and braid index both equal to three, and its augmentation upper bounds use Gröbner bases of braid-specific equations [1 Corollary 1.12 and Theorem 5.3]. Theorem 1 gives infinite hyperbolic prime families at unbounded bridge number. Gao, Gu, Li and Wang prove for the untwisted Whitehead double of the square knot, using its satellite presentation for the augmentation bound and a finite quotient and cocycle calculation for the meridional bound [3 Theorem 1.3 and Section 6]. Here the rank-one obstruction persists under a family of braid tails, and the diagram isotopy produces hyperbolic examples without taking satellites or connected sums. The theorem gives an upper bound on augmentation rank. Its hyperbolicity cutoffs are existential, and the cutoff in is allowed to depend on .
The algebraic construction uses the same braid action in two ways. The braid after fixes the first and fourth meridians. After conjugating the labels on strands five through by the fourth label, this part of the braid acts by ordinary Hurwitz moves on the remaining entries. These moves preserve their ordered product. In an alternating group we can compute the order of this product explicitly and use it to close up a coloring that needs generators. In a rank-one representation, assuming that both the displacement vectors and their covectors are independent forces two matrix identities inside . Seven signed crossing relations then give a contradiction.
The two filling steps use different peripheral markings. The first filling uses the disk marking of a cusp of the fixed three-cusped exterior. The homeomorphism to the two-cusped parent is unmarked and transfers hyperbolicity. The second filling uses the original disk marking of the circle in that parent.
To obtain the two-cusped parent, put a circle around the strands carrying the Hurwitz moves. Varying amounts to filling that circle. The number of strands through it varies with , so a separate argument is needed to control the exteriors before filling. The diagram isotopy in Section 4 trades each extra threaded curl for a full twist between a knot arc and a circle arc. This identifies all the two-cusped exteriors with fillings of a single three-cusped manifold, to which we apply hyperbolic Dehn filling twice.
Alternating-group colorings
For any group, the Wirtinger rule at a positive braid crossing is ; its inverse is . A tuple of labels at the top of a closed braid defines a homomorphism from the knot group exactly when the crossing rules return the tuple at the bottom. The image is generated by the top labels, because they generate the Wirtinger group.
Put . On the entries at positions , , , its action is
In particular, fixes the fourth label and acts as on . The subgroup of generated by fixes entries one and four and acts by adjacent Hurwitz moves on the tuple
This follows for the last generators because simultaneous conjugation commutes with , and then holds for inverses as well. The product of the entries in (1) is invariant, since has product .
We use permutations composed from right to left. The top labels on symbols are
The first four generate . Indeed, is a three-cycle on . By conjugating that cycle by and , then conjugating the resulting cycles by and , respectively, we obtain the additional supports
These cycles generate . To see the enlargement at each step, an alternating group on an old set of at least three symbols together with a three-cycle on two old symbols and one new symbol generates the alternating group on the union. Conjugations and inverses give both orientations for every pair of old symbols, including when the old set has size three. If the cycle has just one old symbol, conjugate it by an old permutation moving that symbol. The product of the original cycle with the inverse conjugate gives a cycle on two old symbols and one new symbol, after which the other new symbol can also be added. The displayed supports can therefore be adjoined in order. Each for consists of two transpositions with distinct old symbols and distinct new symbols . Conjugate it by an even old permutation fixing and moving , and multiply the two double transpositions. This gives a three-cycle on two old symbols and . Repeat with fixed and moved to adjoin . The full tuple generates .
To check the coloring, let , , . Applying the crossing rule through gives
This four-strand calculation can also be read from the crossing relations in (2). With for , the product of the middle tuple is
In each displayed progression the terms run only up to the indicated endpoint. For the product is the one computed below. For larger each additional factor inserts its two new symbols just before and in the respective cycles. The two long cycles have lengths and , so has order . Reading the cycles gives
The adjacent moves in act on (1) as the left shift that sends the first entry to at the end. After shifts all labels have been conjugated by . After shifts the three displayed identities return the initial middle tuple . Adding shifts has no effect because has order .
Example. At the middle tuple has length three and . Four shifts close the coloring of , and a further shifts act trivially. This gives the exponent on five strands.
The strand permutation of is a -cycle on positions ; its exponent is mod . The permutation of is . The two transpositions insert the two omitted positions into that cycle, giving one knot component. Its group surjects onto by the coloring just constructed. A double transposition moves two independent directions in the natural permutation representation over . The following lower bound uses its two transposition edges. A set of at most double transpositions has a support graph with at most edges on vertices. Each generator preserves the components of the graph, so such a set cannot generate a transitive group. The quotient forces , since every positive meridian is conjugate to a top label. A closed -strand braid gives a bridge presentation with at most maxima, and the Wirtinger relations of a bridge presentation give one generating meridian per bridge. Thus .
The rank-one obstruction
We prove an upper bound for every knot obtained by following with a braid in the subgroup acting on . That subgroup fixes the first and fourth meridian words. These two fixed positions suffice for the rank-one argument.
Lemma 2. Suppose that belongs to and is a knot. Let its group act on a finite-dimensional space over a field so that each meridian has displacement of rank one. Write each meridian as , where and . Then at least one of the spans of all the 's and all the 's has dimension at most , and both dimensions are at most .
Proof. Each of the two spans is already spanned by its top-strand vectors or covectors. The root span of the generating meridians is invariant under those generators and their inverses, and the covector span is invariant under right multiplication by the same matrices. Conjugating a generator therefore introduces neither a new root direction nor a new covector direction.
Suppose both spans have dimension . In particular both families of four vectors and covectors on the first four strands are separately independent. Set
The first two identities hold for every meridian, with one common . When , conjugation by a matrix replaces its pair by . Two pairs for the same nonzero displacement differ by reciprocal scaling.
Here is the crossing trace through . We use arc subscripts on the matrices in this proof. A row represents . The input labels are and the output labels are .
The repeated outer labels express closure, since fixes the first and fourth entries. Each row is obtained by or in the order of the letters in .
Put , , , . Replacing by changes the root by a multiple of the root of and the covector by a multiple of its covector, preserving both independent families. Let and denote the root and covector spans of , respectively. Both have dimension three, and in the pair of one has and . Rows 1 through 7 and row 13 propagate pairs in .
Rows 0 and 8 give for . If have pairs and in , comparing the pairs on both sides gives
for some . The coefficients on give and . The other components give and . Applying to the expression for gives . Thus and . The relation is now . Row 9, together with , gives .
Next, rows 10–12 send to , then , then , where , and . Write the pair of as in , and put , , . Conjugating at each step gives
| root | covector | |
Table 3.
For example ; the next row uses . The last pair lies in , so and . Its remaining coefficients multiply to one, which gives .
Of the eight rows supported in , rows 1–7 suffice for the contradiction. They now satisfy the hypotheses of Lemma 3 and contain the three independent pairs of . This proves the span bound.
Lemma 3. Suppose the rank-one meridian matrices of common nonzero determinant satisfy the seven signed relations in rows 1–7 of (2), with and . Their root and covector spans cannot both have dimension three.
Proof. Reorder the crossings so that the propagation comes first:
Here the rows with over-arcs 3, 4, 2 and output arcs 13, 8, 4, respectively, solve the corresponding prefix relations for the other under-arc. The first five crossings propagate all the matrices from . If both spans have dimension three, the corresponding pairs are separately independent. Set
In the bases and , propagation gives
| 2 | ||
| 3 | ||
| 4 | ||
| 13 |
Table 4.
Lemma 3. Suppose the rank-one meridian matrices of common nonzero determinant satisfy the seven signed relations in rows 1–7 of (2), with and . Their root and covector spans cannot both have dimension three.
Proof. Reorder the crossings so that the propagation comes first:
Here the rows with over-arcs 3, 4, 2 and output arcs 13, 8, 4, respectively, solve the corresponding prefix relations for the other under-arc. The first five crossings propagate all the matrices from . If both spans have dimension three, the corresponding pairs are separately independent. Set
In the bases and , propagation gives
| 2 | ||
| 3 | ||
| 4 | ||
| 13 |
Table 5 (PDF p. 5).
:::*
For the first row, has pair , and . The next two rows conjugate by and , respectively, with , and , . The last row uses .
At the crossing , put and . The first and third coefficients of give
Thus are nonzero. The middle coefficients give and . At the pair has zero first coefficients. Comparing the first coefficients of the pairs of and therefore gives
with . Cancelling yields . This makes , contrary to .
For an irreducible representation in Cornwell’s correspondence, the root span is nonzero and invariant, hence equals . The common kernel of the covectors is the fixed subspace and is zero, so the covectors span . Lemma 2 bounds the dimension by and proves the augmentation inequality in Theorem 1.
Hyperbolicity from two successive fillings
A circle around the moving strands
Set and . The braid relation gives . For instance, conjugation of by gives when . In the step from to , the new letter of is conjugated by , giving because . The earlier letters are unchanged by before conjugating by . It follows that the two braid words
are convenient starting points for a twist on the consecutive strands in positions .
Add strand and append
The last strand closes to an unknot , bounding a disk pierced once by each strand in positions and missing the other strands. One way to see the disk is to factor as , where
Each factor pushes the last puncture once around puncture . The stems in this ordered product run along the same side of the adjacent punctures, so their concatenation bounds an oval around precisely those punctures. Equivalently, in the closure of the last strand, its two opposite crossings across strand make a cancelling detour. The remaining oval passes over the selected strands on the outward trip and under them on the return. This gives the claimed disk in the slab of , including the two outside positions. The knot component of has the permutation already computed in Section 2; the same is true for .
For a disk-bounding circle with this marking, filling along means killing , where is a positive meridian and the preferred disk longitude. A disk twist on the solid torus obtained by removing takes this slope to . Localizing the twist to the slab inserts full twists on the disk-piercing strands. To check the sign, put for the selected meridians and for the circle meridian. Tracing the outward and return crossings of the oval gives the selected outputs and circle output . The return has crossed the whole ordered tuple, which contributes . The two outside crossings conjugate the outside meridian by inverse powers of the initial and final circle labels and cancel after closure. The closure relation is , and is represented by , the boundary of the punctured disk. Filling gives . By the Hurwitz formula, conjugates the whole selected tuple by , so the geometric disk twist inserts in our convention.
Let and , where denotes the link exterior. The preceding disk twist identifies with as exteriors without a fixed peripheral basis, since their knot braids differ by one full twist on the disk. In the original marking of we have
Indeed, the inserted power of is . Conjugating the resulting braid by gives the word defining .
A single exterior for the first filling
Consider the six-strand braid
and the seven-strand suffix . Set . The closure of has two components, a knot component containing and a circle. The fourth and fifth closure arcs lie on these different components. These assignments also follow by tracing the strand permutation of , which fixes position five and cycles the other five positions. The new last component bounds a disk around positions four and five, by the same oval construction as for . Thus has three boundary tori, and filling the new circle along the disk-marked slope gives
Lemma 4. For every , the oriented link is isotopic to , with the knot and circle components identified.
Here is how the isotopy works. After braid conjugation and commutation,
On the knot component alone this is an iterated positive stabilization. In the link, the disk of meets the additional strands, so the corresponding curls cannot simply be removed. First cancel the detour of the circle and the two sides of the conjugating braid. The outer crossings then form a ladder along the circle. Slide each successive triangle through the already formed clasps by two Reidemeister III moves per clasp. At its end a Reidemeister III move exposes a curl, and a Reidemeister I move leaves one more two-crossing positive clasp. Repeating this leaves the diagram with clasps in the band of the fourth and fifth closure arcs of . Appendix A proves the isotopy by specifying the moves, the induction on the ladder, and the identification of that band.
By Lemma 4 and (4),
This homeomorphism is used to transfer hyperbolicity. The second filling (3) uses the original disk marking of . Twisting around either circle can change the expression of the other circle’s longitude, so no common longitude basis is required for the two fillings.
Three finite hyperbolicity certificates
The computation certifies the complete hyperbolic structures of , , , where , use , with its full disk twist, in the literal braid . Table 1 (PDF p. 8) records the sizes of the input diagrams and certified triangulations. The two direct tests for , give explicit bridge numbers covered by the theorem regardless of the threshold obtained from .
| Exterior | Diagram crossings | Cusps | Tetrahedra |
| 32 | 3 | 35 | |
| 32 | 2 | 31 | |
| 43 | 2 | 34 |
Table 1 (PDF p. 8). Literal braid diagrams and complete ideal triangulations for the three hyperbolicity certificates. Crossings are counted in the given projection.
Each exact input binds a planar-diagram code and its components to the braid, checking every crossing port and every closure identification. The exterior of that diagram is constructed in Spherogram 2.4.1 and SnapPy 3.3.2. Its decorated triangulation signature is checked against the retained one with the component marking, and the face pairings and gluing equations of the reloaded triangulation are checked. SnapPy under SageMath 10.9 then applies a Krawczyk interval test at 100 bits to both the fresh exterior and the retained triangulation. For the analytic criterion, a Krawczyk inclusion certifies a solution to a square subsystem of the shape equations in the interval box. The logarithmic edge and cusp equations are checked in the correct branches, and every shape has positive imaginary part. Together with the subsystem reduction, the positivity, gluing and completeness checks give a complete hyperbolic structure on the triangulated exterior [4 Sections 2–4]. Exact inputs, software pins, rational interval bounds and the replay procedure are supplied in the computational supplement. The proof uses these three finite certificates and the written diagram induction.
Proof of the prime-family conclusion
Since is complete and finite-volume hyperbolic, Thurston’s hyperbolic Dehn filling theorem [5 Theorem 5.8.2] applies to the cusp with the other two cusps unfilled. The slopes escape every finite set as grows. Thus is hyperbolic for every sufficiently large , as well as for by the direct certificates. For any such , apply the theorem to on the cusp. The slopes in (3) escape every finite set as grows, so all sufficiently large are hyperbolic. A hyperbolic knot is prime. Indeed, a composite knot group is the amalgam of the nontrivial summand groups over their meridian, and the two summand longitudes lie outside that amalgamating subgroup by peripheral injectivity. They commute with the common meridian and fail to commute with one another by amalgam normal form. The centralizer of a nonidentity element of a torsion-free discrete subgroup of is abelian, as is seen by putting the element in parabolic or loxodromic normal form.
For a fixed such , the lengths of the second filling slopes on an embedded horotorus tend to infinity, since they are the norms of in a flat lattice with . For sufficiently large the fillings are hyperbolic and . The volume inequality of Futer, Kalfagianni and Purcell [2 Theorem 1.1] and strict volume decrease under proper hyperbolic filling [5 Theorem 6.5.6] give
The volumes converge to , and by Mostow–Prasad rigidity each hyperbolic exterior type has one volume. A finite set of exterior types would have a maximum volume strictly below . Thus there are infinitely many knot types for that . Together with the algebraic bounds, this proves Theorem 1.
Computational supplement. The accompanying supplement supplies the exact planar diagrams, decorated triangulations, face pairings, gluing equations, rational interval enclosures, and replay scripts for the hyperbolicity certificates in § 4.3. It also supplies the finite diagram moves and checks of the port conventions used in Appendix § 5. The algebraic and diagram inductions in the article establish the unrestricted parameter statements.
AI assistance and review. ChatGPT (OpenAI) assisted with mathematical arguments, literature checks, exposition, proof review and verification code.
The oriented diagram isotopy
We prove Lemma 4 in the spherical braid projection. The proof fixes the crossings of and keeps track of the two arcs where the new clasps are inserted.
Port convention and initial cancellations
A crossing has ports in counterclockwise order. The even pair is the understrand and the odd pair the overstrand. A positive letter uses inputs and outputs in left-to-right order. A negative letter uses inputs and outputs . On each strand, connect its last output to its first input to close the braid. Write for port of crossing and take port indices modulo four.
Put and number the crossings of from zero. Cancel with , followed by with for , in increasing order. The first pair consists of the two crossings in the circle detour. In each pair the bigon has edges – and –. The signs are opposite, with one branch over at both crossings. These conditions specify Reidemeister II cancellations.
The indexing of the uncancelled crossings and the neighbours in the successive bigons can be obtained as follows. Name the crossings of by . Use for the crossings of , and set , for , the crossings on the descending and returning parts of . The detour cancellation connects to and to . In the resulting first bigon and at each later step , the four outside neighbours of , are
The cancellation connects to and wires the next bigon, with the last step joining to . This describes the cancellations by induction. The listed neighbours are obtained from the port rule by following the consecutive crossings of and the return of . In particular the pairs use no crossing from .
Let , , , . The edges of the remaining diagram outside are specified by their port neighbours in Table 2 (PDF p. 10). The ’s follow the consecutive crossings of , with proceeding to , or to at the end. The two circuits past the circle follow the ’s in order and the ’s in reverse order. At the doubled the two crossings and meet directly. The joins from the cancelled bigons give the attachments from to . With the four unchanged attachments to these traces give the table. All crossings in it are positive.
| Crossing | Port 0 | Port 1 | Port 2 | Port 3 |
| , | or | |||
| , | ||||
| , | ||||
Table 2 (PDF p. 10). Port neighbours after the initial cancellations. The port 2 neighbour is at . Every edge is reciprocal, including the ones meeting .
Two local moves and the induction
Write for the triangle with edges
Consecutive ports make this a face in the oriented spherical rotation system. The projection is connected and these faces are empty disks. For the Reidemeister III moves below the edges have respectively two underports, two overports, and one of each. They are the bottom, top and middle branches required for the move.
First consider the triangle with an existing clasp attached, having edges -- and --. Use the six external endpoints in the first row of the table below, in the cyclic order ; the inputs are . Apply Reidemeister III, first at this triangle and then at . The clasp edges persist, and the other internal edges become –, –, –. Tracing the three strands between the six endpoints gives the second row:
| before | ||||||
| after |
Table 6.
Thus the triangle can pass one clasp relative to all six external endpoints.
For the same triangle at the cap (Figure 1), use the extra edge – and the first row below. The boundary order is , with inputs . The first Reidemeister III move puts on the edge from to , with –, – and a monogon –. Remove the monogon by Reidemeister I. The two crossings left form a positive clasp with edges –, – and the external attachments in the second row:

Figure 1. The signed cap move. The black strand runs from to , and the blue strand from to . Gaps mark the underpasses. All crossings are positive. The type III move on exposes the curl at , and the type I move removes it while fixing the four boundary endpoints.
| before | ||||
| after |
Table 7.
In particular this is an ordinary positive full twist between the two incoming strands. Both local rewrites keep the cyclic boundary orders and orientations.
Now put . At stage , for , set . There are already clasp pairs for . Each has its two clasp edges, and
The moving triangle begins at . At the triangle and cap are in Table 2 (PDF p. 10). At each for , use the first local rewrite with , . The two former attachments of to now attach and to , giving the triangle at . This leaves – in place. At , the cap rewrite removes and gives the clasp .
To see the next stage, track the four ends of the unchanged part of the ladder. At the start of stage , is attached to and to . The first rewrite sends those attachments to and , whether it is at a clasp or at the cap. Later moves in the stage keep them there. The cap rewires – and – to the two new edges from . The next triangle and cap therefore satisfy the same hypotheses. This proves the induction for every .
At its end have been deleted. The diagram consists of the ladder with indices and , and clasp pairs inserted between the two arcs – and –. The first endpoint of each arc now leads into the first clasp (or directly to if ). The two edges from to then lead to the next clasp or to the base ladder. All other edges of the base ladder remain as specified in the table, and the crossings of are fixed.
Identifying the insertion band in
Number the crossings of the literal word from zero. Its fourth and fifth closure arcs run from 24.2 to 0.0 and from 25.1 to 0.3. Cut those two arcs just before crossing zero, leaving a named incoming and outgoing end for each arc. The Reidemeister II pairs (6, 25) and (7, 22) cancel with the four ends open. For the first pair the bigon edges are 6.0–25.2, 6.3–25.3. For the second they are 7.2–22.0, 7.3–22.3 after the first cancellation. The two incoming ends and the fourth outgoing end stay attached to their ports. The fifth outgoing end moves from 25.1 to 5.2. Neither cancellation meets the slab at the cut. The resulting 22-crossing projection, with the arcs reclosed, is the ladder. The port bijection has no rotations and is given by
| crossing of | 23 | 24 | 0 | 1 | 2 | 5 | 4 | 3 |
Table 8.
This is verified by the braid port rule, including the edges of and the four attachments in Table 2 (PDF p. 10). The cut arcs correspond precisely to – and –. The marked ends identify the actual fourth/fifth strand band before the first crossing of , including both boundary orientations. In the closed braid a positive full twist in that band can be written at the end of . Each clasp from the induction has the port pairing of in this band, so the resulting projection is the twice-reduced diagram of . The oriented moves preserve the component containing and the other component. Since , this proves Lemma 4.
References
- [1]C. R. Cornwell, KCH representations, augmentations, and A-polynomials, J. Symplectic Geom. 15 (2017), 983–1017. doi:10.4310/JSG.2017.v15.n4.a2.DOI
- [2]D. Futer, E. Kalfagianni, and J. S. Purcell, Dehn filling, volume, and the Jones polynomial, J. Differential Geom. 78 (2008), 429–464. doi:10.4310/jdg/1207834551.DOI
- [3]H. Gao, S. Gu, Z. Li, and J. Wang, Meridional ranks of iterated Whitehead doubles, preprint, 2026. arXiv:2611.15673v1.arxiv.org/abs/2611.15673
- [4]N. Hoffman, K. Ichihara, M. Kashiwagi, H. Masai, S. Oishi, and A. Takayasu, Verified computations for hyperbolic 3-manifolds, Exp. Math. 25 (2016), 66–78. doi:10.1080/10586458.2015.1029599.DOI
- [5]W. P. Thurston, The Geometry and Topology of Three-Manifolds, Princeton lecture notes, electronic edition 1.1, 2002. Electronic edition.