Michele Capovilla-Searle, Keiko Kawamuro, and Rebecca Sorsen

BIRMAN-KO-LEE LEFT CANONICAL FORM AND ITS
APPLICATIONS
MICHELE CAPOVILLA-SEARLE, KEIKO KAWAMURO, AND REBECCA SORSEN

arXiv:2304.00064v1 [math.GT] 31 Mar 2023
Abstract. Using Birman, Ko, and Lee’s left canonical form of a braid,
we characterize almost strongly quasipositive braids and give estimates
of the fractional Dehn twist coefficient.
Contents
1. Introduction 2
2. Canonical Factors via Diagram 5
2.1. Band Generators 5
2.2. Non-Crossing Partition Diagrams 6
2.3. Canonical Factors 9
3. Partial Ordering on CnFct(Bn ) 9
3.1. Operations and ∗ on CnFct(Bn ). 10
3.2. Proof of Theorem 3.4 15
4. Partially Ordered Set (Wds(β), ⇒) 21
5. The Left Canonical Form 26
5.1. Definition of the Left Canonical Form 26
5.2. Left Canonical Form Algorithm 27
5.3. Super Summit Set and The Left Canonical Form 28
6. Detection of SQP and ASQP braids 29
7. The Bennequin Inequality and LCF(β) 31
8. The Negative Band Number and Reduction of LCF(β) 34
8.1. Review of Reduction Operation 34
8.2. Bounds of nb(β) in Terms of inf(β) and sup(β) 36
9. Relations Between inf(β) and nb(β) for β ∈ SSS(β) 38
10. Fractional Dehn Twist Coefficient 41
1
2 MICHELE CAPOVILLA-SEARLE, KEIKO KAWAMURO, AND REBECCA SORSEN
References 46
1. Introduction
The braid groups Bn for n = 1, 2, 3, . . . were originally introduced by

Emil Artin in [1]. The following presentation is known as the standard
presentation of Bn with the Artin generators σ1 , . . . , σn−1 .

σi σj = σj σi , |i − j| > 1
Bn = σ1 , . . . , σn−1 σ σ σ = σ σ σ , i = 1, . . . , n − 2

i i+1 i i+1 i i+1
In the standard presentation of Bn , Artin generators only swap adjacent

strands. In the band generator presentation, any two strands are allowed to
swap. For each i, j with 1 ≤ i < j ≤ n, define
aij = (σj−1 σj−2 · · · σi+1 )σi (σj−1 · · · σj−2 σi+1 )−1 .
Here, aij is the element in Bn which swaps the ith and jth strand, while leav-
ing all other strands fixed. In addition, the two strands being interchanged
must lie in front of all fixed strands. This gives us the band generator
presentation (or Birman-Ko-Lee presentation) [3] as follows:

ajk aij = aij aik = aik ajk , i < j < k
Bn = aij a a = a a , i < j < k < l

ij kl kl ij
Word problems and conjugacy problems have been central problems in the
study of braids. Given two n-braid words w and w0 determining whether w =
w0 the same braid element in Bn is called the word problem, and determining
whether w = vw0 v −1 for some v ∈ Bn i.e., w and w0 are conjugate, is called
the conjugacy problem.
These problems have been solved by a number of people, including Artin
[1], Garside [9], Elrifai and Morton [7], Xu [25], Kang, Ko, and Lee [15], and
Birman, Ko and Lee [3]. The latter three papers used band generators to
solve these problems.
Birman-Ko-Lee’s left canonical form, LCF(β), was used to solve the word
problem using band generator techniques. Namely, Birman, Ko and Lee,
generalizing the earlier work by Xu [25] and Kang, Ko and Lee [15], proved
that for n-braid words w and w0 , we have w = w0 in Bn if and only if
LCF(w) = LCF(w0 ). They also solved the conjugacy problem using the left
canonical form. See Theorem 5.2 for the definition of LCF(β).
Birman, Ko, and Lee’s definition of LCF(β) is algebraic. In this paper, we
study LCF(β) using non-crossing partition diagrams. These diagrams can
be found in the literature, including the book [5] by Dehornoy, Dynnikov,
Rolfsen and Wiest, as well as a paper by Calvez and Wiest [4].
BIRMAN-KO-LEE LEFT CANONICAL FORM AND ITS APPLICATIONS 3
We will give an outline of the paper and state some of our main results.
In §2, we start with reviewing band generators and non-crossing partition
diagrams. We then define canonical factors that play key role in LCF(β) and
relate canonical factors with non-crossing partition diagrams.
In §3, we introduce a partial ordering ≺ on the set of canonical factors
of Bn , denoted CnFct(Bn ). The definition of ≺ is algebraic. We give a
graphical interpretation.
Theorem 3.4. For canonical factors A and B we have A ≺ B if and only
if their convex hulls satisfy cvh(A) ⊂ cvh(B).
To this end, we introduce a binary operation on the set CnFct(Bn ). As
a biproduct, we obtain a useful corollary:
Corollary 3.19. For every canonical factor A ∈ CnFct(Bn ) there exists a
B ∈ CnFct(Bn ) such that A B = AB = δ.
In §4, using the partial ordering ≺ in § 3, we introduce another ordering ⇒
on the set Wds(β) of braid words representing the braid β. We give detailed
examples how the ordering works in the braid group B4 .
In §5, we study the left canonical form. The set Wds(β) is an infinite set;
however, Birman, Ko, and Lee’s theorem [3] states that with respect to the
ordering ⇒, there is a unique maximal element in Wds(β). This unique max-
imal element is called the left canonical form of β and is denoted by LCF(β).
We also review Kang, Ko and Lee’s algorithm [15] for the left canonical form
and compute examples using non-crossing partition diagrams.
In §6, we begin to discuss the notion of positivity of knots and links. In
this case, positivity refers to the property that all crossings of a link have the
same sign. In the literature, multiple notions of positivity for braids have
been studied, including positive (P) braids that is the monoid generated by
Artin generators {σi }n−1
i=1 , quasipositive (QP) braids that is the monoid nor-
mally generated by the Artin generators, and strongly quasipositive (SQP)
braids that is the monoid Bn+ generated by the positive band generators ai,j .
It is also interesting to study the notion of almost positivity. For example,
if we allow one crossing to be negative, does this change any of the properties
of positivity? We call such braids almost strongly quasipositive (ASQP). In
[10], Hamer, Ito, and Kawamuro discussed properties and relations among
the various notions of positivity and almost positivity.
These positivity notions are related by inclusions:
QP ⊂ AQP ⊂ Bn
∪ ∪
P ⊂ SQP = Bn+ ⊂ ASQP
The infimum of a braid β, denoted by inf(β), is an invariant of the braid

and can be directly read from LCF(β) (cf. Definition 5.1 and Theorem 5.2).
We discuss the invariant inf(β) for SQP and ASQP braids. For SQP braids,
we obtain the following if and only if statement.
Theorem 6.3. An n-braid β ∈ Bn is strongly quasipositive if and only if
inf(β) ≥ 0.
For ASQP braids, we obtain an if and only if statement in Theorem 9.4
discussed below.
In §7, we study the negative band number nb(β). It is the minimal number
of negative bands for β ∈ Bn , i.e.,

β is represented by a word in band
nb(β) = min k
generators containing k negative bands
The invariant nb(β) is an upper bound of the defect of the Bennequin in-
equality [13].
1
(−χ(K) − SL(K)) ≤ nb[β]
2
where K is the knot (or link) type of the braid closure β̂, χ(K) is the
maximal Euler characteristic of K, and SL(K) is the maximal self-linking
number for K. We give two estimates of the negative band number nb(β) in
terms of the left canonical form. The first one is an upper bound of −nb(β).
The second one, a lower bound is given in §8.
Theorem 7.3. Let β ∈ Bn with nb(β) ≥ 1. Then
0 < − inf(β) = | inf(β)| ≤ nb(β).
In § 8, we review the reduction operation introduced by Kang, Ko, and

Lee [15] to solve the shortest word problem for 4-braids. We use the reduc-
tion operation to give another estimate of −nb(β) that is complemental to
Theorem 7.3 as follows:
Theorem 8.7. Let β be an n-braid. If inf(β) < 0 then
nb(β) ≤ (n − 2)| inf(β)| − min{0, sup(β)}.
Moreover, the equality holds when n = 3 and we have
nb(β) = | inf(β)| − min{0, sup(β)}.
In §9, we discuss more relations between inf(β) and nb(β) for a specific
super summit element β ∈ SSS(β). We characterize strictly ASQP braids in
terms of the left canonical form.
Theorem 9.4. A braid β ∈ Bn with n ≤ 4 is conjugate to a strictly
almost strongly quasipositive braid if and only if every super summit element
β 0 ∈ SSS(β) has inf(β 0 ) = −1 and LCF(β 0 ) contains a canonical factor of

word length n − 2.
In the above theorem, the restriction on the braid index n = 3, 4 is only
required for the only-if (⇒) direction. The statement of the if-direction (⇐)
holds for general n.
In Section §10, we apply the left canonical form of a braid β to study its
fractional Dehn twist coefficient (denoted FDTC). Following the description
of the fractional Dehn twist coefficient in [14] and [12], we first develop tools
to easily compute the FDTC of a braid β. Then, applying the left canonical
form, we obtain a bound on the FDTC of β.
Theorem 10.7. Suppose LCF(β) = δ r A1 · · · Ak then
inf(β) r r+k sup(β)
= ≤ c(β) ≤ = .
n n n n
Acknowledgements. The authors would like to thank Juan Gonzalez-

Meneses for useful conversations during a workshop held at ICERM
Brown University in April 2022.
2. Canonical Factors via Diagram
In this section we review band generators of the braid group Bn and relate
canonical factors of Bn to non-crossing partition diagrams of an n-punctured
disk Dn .
2.1. Band Generators. In the traditional presentation of Bn , Artin gen-

erators only swap adjacent strands. In the band generator presentation, any
two strands are now allowed to swap. For each i, j with 1 ≤ i < j ≤ n,
define
aij = (σj−1 σj−2 · · · σi+1 )σi (σj−1 · · · σj−2 σi+1 )−1 .
Here, aij is the element in Bn which swaps the ith and jth strand, while leav-
ing all other strands fixed. In addition, the two strands being interchanged
must lie in front of all fixed strands. This gives us the band generator
presentation (or Birman-Ko-Lee presentation) [3] as follows:

ajk aij = aij aik = aik ajk , i < j < k
Bn = aij a a = a a , i < j < k < l
(2.1)
ij kl kl ij
The set of band generators of Bn is denoted by BGen(Bn ). Band generators

ai,j are often called positive bands as each creates a positive crossing in the
braid diagram, versus their inverses a−1i,j are called negative bands.
In B4 , following the notations in [15], the band generators are renamed

as:
a1 := a1,2 , a2 := a2,3 , a3 := a3,4 , a4 := a1,4 , b1 := a1,3 , b2 := a2,4 .
Visually, these six band generators are as follows. The braid strands are
numbered from 1 (bottom strand) to 4 (top strand). The band generators
Figure 1. Band generators for B4
a1 , . . . , a4 , b1 , b2 are positive bands and their inverses a−1 −1 −1 −1

1 , . . . , a4 , b1 , b2
are negative bands.
2.2. Non-Crossing Partition Diagrams. In this subsection, we review

non-crossing partition diagrams. They have been introduced in the book [5]
by Dehornoy, Dynnikov, Rolfsen, and Wiest, and Calvez and Wiest [4] use
diagrams to study 4-braids.
Recall that B4 can be viewed as the mapping class group of the four
times punctured disk D4 . We will parameterize D4 as the unit disk in C
with punctures at
1 3πi 1 πi
P1 = e− 4 , P2 = e− 4 ,
2 2
1 7πi 1 5πi
P3 = e− 4 , P4 = e− 4 .
2 2
P4 P3
P1 P2
Note that we placed the punctures in a counterclockwise direction, while in

[5] and [4] punctures are placed clockwise.
Similarly, we define an n punctured disk Dn . Punctures are labeled

P1 , . . . , Pn counterclockwise and P1 and Pn are separated by the half-line
from the origin with angle π.
The braid strands can be thought of as {P1 , . . . Pn } × [0, 1] in the cylinder
D4 × [0, 1] where the k-th strand is {Pk } × [0, 1]. The braid diagram can
be thought as a view of the cylinder Dn × [0, 1] ⊂ C × R from the position
(−1, 0.5) ∈ C × R.
The line segment Pi Pj × {0.5} in the cylinder can be seen in the braid
diagram as an arc over the braid strands i + 1, i + 2, . . . , j − 1. This gives an
idea to pictorially denote the positive band ai,j by the line segment Pi Pj in
Dn . In fact under the identification of the braid group Bn and the mapping
class group of n-punctured disk Dn , the braid element ai,j corresponds to a
positive half Dehn-twist (counterclockwise) along the line segment Pi Pj .
With this idea in mind, we associate a certain braid word, which will be
called a canonical factor in Definition 2.8 below, with a diagram. These
diagrams have been used in the study of non-crossing partitions.
Definition 2.1. The diagram of n dots (with no edges) represents the iden-
tity element e of the braid group Bn . When n = 4 we define:
( ) := e.
Definition 2.2. (Edge, 2-gon) The diagram of n dots with an edge con-
necting two points Pi and Pj (i < j) represents the band generator ai,j ∈ Bn .
When n = 4 we define:
( ) := a1 ( ) := a2
( ) := a3 ( ) := a4
( ) := b1 ( ) := b2 .
With the new notation we have:

BGen(B4 ) = {a1 , a2 , a3 , a4 , b1 , b2 }
= {( ), ( ), ( ), ( ), ( ), ( )}.
Remark 2.3. In the following, we will define 3-gons, 4-gons, . . . so that an

n-gon corresponds to a length n − 1 word. This justifies viewing a single
edge connecting two vertices (Definition 2.2) as a 2-gon.
Definition 2.4. (Triangle, 3-gon) Consider a triangle with vertices Pi , Pj , Pk ∈
Dn . Denote the band generator corresponding to each edge clockwise by
α0 , α1 , α2 ∈ BGen(Bn ). For example, in the triangle contained in the disk
D4 , the three edges are α0 = a1 =( ), α1 = b1 =( ), and α2 = a2 =( ).
By the second relation for the band generator presentation, Eq.(2.1), for Bn ,
we know that α0 α1 = α1 α2 = α2 α0 holds. Therefore, we let the triangle
represent the length 2 braid α0 α1 = α1 α2 = α2 α0 .
For B4 , we have four different triangles where each admits three band
generator factorizations:
( ) = a2 a1 = ( )( )=( )( )=( )( )
( ) = a3 a2 = ( )( )=( )( )=( )( )
( ) = a4 a3 = ( )( )=( )( )=( )( )
( ) = a1 a4 = ( )( )=( )( )=( )( ).
Definition 2.5. (Square, 4 gon) Consider a square with vertices Pi , Pj , Pk , P` ∈

Dn . Denote the band generator corresponding to each edge clockwise by
α0 , α1 , α2 , α3 ∈ BGen(Bn ). By the band generator presentation for Bn , we
know α0 α1 α2 = α1 α2 α3 = α2 α3 α0 = α3 α0 α1 . We may define the square to
represent the length 3 braid element.
In B4 , there is a unique square:
This square coincides with the fundamental element δ := a3 a2 a1 of B4 . It is

known [15] that δ admits 12 different factorizations in BGen(B4 ) as follows:
a3 a2 a1 a4 a3 a2 a1 a4 a3 a2 a1 a4 b1 a2 a4 a1 b1 a4 (2.2)
a3 b1 a2 a1 a3 b1 b2 a1 a3 a2 b2 a1 a4 b2 a3 a2 a4 b2
Note that δ 4 = ∆2 a full twist, thus δ 4 generates the center of the braid
group B4 .
We can generalize the above construction to a k-gon.

Definition 2.6. (k-gon) Let 2 ≤ k ≤ n. A k-gon in Dn , whose vertices

oriented clockwise, represents a length k − 1 word reading consecutive k − 1
edges of the k-gon clockwise.
Definition 2.7. A disjoint union of polygons represents the product of
braids corresponding to the polygons.
For B4 , we have two such diagrams that represent a1 a3 and a2 a4 .
( ) := a1 a3
( ) := a2 a4 .
By the disjointedness condition, each of these diagrams admits two factor-

izations in BGen(B4 ) as follows:
( )=( )( )=( )( )
( )=( )( )=( )( ).
2.3. Canonical Factors. In this section, we define canonical factors that

play an essential role in the left canonical form of a braid.
An n-braid is positive if it is represented by a word in positive band
generators. Let Bn+ denote the monoid of positive n-braids.
Definition 2.8. For two braid words V, W ∈ Bn , we say V ≤ W if W =
P V Q for some (possibly empty) positive words P, Q ∈ Bn+ . Elements in
the set {W ∈ Bn | e ≤ W ≤ δ} are called canonical factors and the set is
denoted by CnFct(Bn ).
In [3, Corollary 3.5], Birman, Ko, and Lee showed that the cardinality
of the set CnFct(Bn ) is the nth Catalan number Cn = (2n)!/n!(n + 1)!. In
their proof, the following theorem is implicit.
Theorem 2.9. The set CnFct(Bn ) is in a one-to-one correspondence with
the set of noncrossing partitions of n elements.
For example, there are 14 canonical factors in B4 .

CnFct(B4 ) = {e, a1 , a2 , a3 , a4 , b1 , b2 , a2 a1 , a3 a2 , a4 a3 , a1 a4 , a1 a3 , a2 a4 , δ}
= {( ), ( ), ( ), ( ), ( ), ( ), ( ),
( ), ( ), ( ), ( ), ( ), ( ), ( )}.
3. Partial Ordering on CnFct(Bn )
In this section we will discuss a natural partial ordering on the set CnFct(Bn )
arising from the diagrams.
Definition 3.1. The set of convex hulls of a canonical factor A, denoted

by cvh(A), is the union of the diagram A and the region(s) enclosed by the
edges.
Example 3.2. The set of convex hulls of a triangle cvh( ) consists of the
2-dimensional triangle and the area inside the triangle enclosed by the three
edges. The set of convex hulls cvh( ) only includes the two disconnected
edges. We do not include the area between the two edges, as they do not
enclose a connected region.
Similar to the partial ordering ≤ on Bn (Definition 2.8), we define a partial

ordering ≺ on the set CnFct(Bn ).
Definition 3.3. For canonical factors A, B ∈ CnFct(Bn ), we say A ≺ B if
AQ = B for some canonical factor Q ∈ CnFct(Bn ).
We note that A ≺ B implies A ≤ B.

The next lemma gives a diagrammatic interpretation of the partial order-
ing.
Theorem 3.4. Let A, B ∈ CnFct(Bn ) be canonical factors. We have cvh(A) ⊂
cvh(B) if and only if A ≺ B.
The theorem will be proved in Propositions 3.18 and 5.4.

Using Theorem 3.4, we can easily obtain the Hasse diagram (Figure 2)
for the partial ordering ≺ on the set CnFct(B4 ). Moreover we notice that ≺
and ≤ are equivalent partial orderings on CnFct(B4 ).
3.1. Operations and ∗ on CnFct(Bn ). To prove Theorem 3.4 we will

introduce operations t, and ∗ on CnFct(Bn ). We use the convention that a
canonical factor is denoted with a roman font (eg. A) and its corresponding
diagram is denoted by the same alphabet in calligraphy style font (eg. A).
Definition 3.5. Let A and B be canonical factors whose diagrams do not
intersect; A ∩ B = ∅. The product AB in the braid group Bn is denoted by
A t B and the corresponding diagram is denoted by A t B. Since AB = BA
we have A t B = B t A.
Definition 3.6. (See Figure 3) Let A ∈ CnFct(Bn ) be a canonical factor
whose diagram A is disjoint union of polygon components. We say that
components X and Y of A are facing to each other if there exist unique
edges x0 ⊂ X and y0 ⊂ Y such that no other components of A lie between
x0 and y0 . The pair of edges x0 and y0 is called a facing pair.
The definition of the -operation is divided into 4 steps.

Figure 2. Hasse Diagram for the partially ordered set

(CnFct(B4 ), ≺)
Figure 3. On the left, A = X t Y and X and Y are facing

each other. On the right, A = X t Y t Z. There exists a
3-gon Z in-between X and Y, so X and Y are not facing
each other.
Definition 3.7. (Step 1. See Figure 4) Let A, B ∈ CnFct(Bn ) be canonical

factors such that
(1) both of their diagrams A and B are connected,
(2) diagrams A and B intersect at a single vertex, say P , and
(3) B lies on the left side of A near the vertex P (if you stand on p and
see the interior of the disk in front of you).
Let a0 and b0 be the unique edges of A and B respectively such that a0 ∩b0 =
P and A∪B lies outside the fan-shape region between a0 and b0 . Assume the
polygon A (resp. B) has k (resp. l) sides. Label the edges of A starting from
a0 clockwise, a1 , . . . , ak−1 . Label the edges of B starting from b0 clockwise,

b1 , . . . , bl−1 . Abusing the notation, each ai and bj represents a positive band
generator.
When A and B satisfy the above conditions (and if we emphasize this
fact), the usual product AB in the braid group Bn is denoted by A B.
A B := AB = a1 · · · ak−1 b1 · · · bl−1 .
Figure 4. An example of Definition 3.7, A B. The left

image shows two polygons that intersect at the point P .
Here, A is a 4-gon, B is a 3-gon, and A B is a 6-gon.
We list properties of A B:
• The diagram of A B, which is denoted by A B, is a single polygon
with k + l − 1 sides.
• By Theorem 2.9 the product A B is a canonical factor.
• cvh(A) ∪ cvh(B) ⊂ cvh(A B). More precisely, A B is the minimal
polygon whose convex hull contains A and B.
By the condition (3), BA does not make sense, i.e., is a non-commutative
operation.
A simple example of the operation is:
( )( )=( )
We will extend the operation to less restrictive pairs of canonical factors.

Definition 3.8. (Step 2. See Figure 5) Let A = A1 t · · · t Ak and B ∈
CnFct(Bn ) be canonical factors such that
• each of the diagrams A1 , . . . , Ak and B is a connected component.
• for each i = 1, . . . , k the pair (Ai , B) satisfy the (Step 1) condition;
namely, the intersection Ai ∩ B is a single vertex, say PAi , of the disk
Dn , and
• in a small neighborhood of PAi , the diagram B lies on the left of Ai .
We extend the operation to the pair A and B by

A B := A1 (A2 (· · · (Ak B)))
where the in the right hand side is defined in (Step 2) in Definition 3.7.
Figure 5. An example of Definition 3.8. We have

A = A1 t A2 and a 3-gon B. We apply Definition 3.8 to
obtain A B = A1 (A2 B).
We remark that is well defined since for every permutation σ ∈ Sk of k

elements,
A1 (A2 (· · · (Ak B))) = Aσ(1) (Aσ(2) (· · · (Aσ(k) B))).
Proposition 3.9. We list properties of the operation below:
(a) As braids, A B = AB.
(b) The corresponding diagram of AB, denoted by AB, is the minimal
disjoint union of polygons whose convex hull contains cvh(A) and
cvh(B).
(c) In particular, cvh(A) ∪ cvh(B) ⊂ cvh(A B).
(d) A B ∈ CnFct(Bn ) by Theorem 2.9.
For Step 2, with regard to (b) the diagram A B is a single polygon.

Definition 3.10. (Step 3. See Figure 6) Let A = A1 t · · · t Ak t A0 and
B ∈ CnFct(Bn ) be canonical factors such that
• the pair (A1 t · · · t Ak , B) satisfy the (Step 2) condition for Defini-
tion 3.8,
• and A0 ∩ B = ∅.
We define
A B := A0 t ((A1 t · · · t Ak ) B)
where on the right hand side is in the sense of (Step 2). The disjoint
operation in the right hand side between A0 and (A1 t · · · t Ak ) B is
justified by the above property (b) and the disjointness A0 ∩ B = ∅. The
diagram A B is the disjoint union of A0 and the polygon (A1 t · · · t Ak ) B.
Therefore, above properties (a), (b), (c), and (d) also hold.
Figure 6. An example of Definition 3.10. We have

A = A1 t A2 t A0 and a 3-gon B. We apply Definition 3.10
to obtain A B = A0 t ((A1 t A2 ) B).
Definition 3.11. (Step 4) Let A, B ∈ CnFct(Bn ) such that

• A ∩ B = cvh(A) ∩ cvh(B) ⊂ {vertices of Dn } and
• if P ∈ A ∩ B, near P the diagram B lies on the left of A.
Therefore, A and B do not share any edges or points away from the vertices.
It is possible that A ∩ B = ∅.
The factor may admit a decomposition B = B1 t· · ·tBl tB 0 ∈ CnFct(Bn )
such that for each i = 1, . . . , l, the pair (A, Bi ) satisfy (Step 3) condition in
Definition 3.10 and A ∩ B 0 = ∅.
We define
A B := B 0 t ((((A B1 ) B2 ) · · · ) Bl )
where the in the right hand side is in the sense of Definition 3.10. The op-
eration is independent of the ordering of factors Bi s. The above properties
(a), (c) and (d) are also satisfied.
Next we define the operation ∗.

Definition 3.12. (See Figure 7) Let A, B ∈ CnFct(Bn ) be canonical factors
such that
• both of their diagrams A and B are connected, and
• their diagrams do not intersect A ∩ B = ∅ (i.e., AB = BA in Bn ).
Suppose that the polygon A (resp. B) has k (resp. l) sides. Let a0 and
b0 be edges of A and B that form the facing pair. Edges of A and B are
named a0 , . . . , ak−1 and b0 , . . . , bl−1 labeled clockwise. Thus A = a1 · · · ak−1
and B = b1 · · · bl−1 . Denote by PA the vertex of A where the edges a0 and
ak−1 meet. Similarly, denote by PB the vertex of B where the edges b0 and
bl−1 meet. Let C denote the edge joining the vertices PA and PB . We call C
the joining edge of A and B. The joining edge C corresponds to a positive
band generator, denoted by C ∈ BGen(Bn ) and can be thought as a bigon

(cf. Remark 2.3.) Define an operation ∗ by
A ∗ B := A (B C) = B (A C) = (A t B) C.
As braids, A ∗ B = ABC = BAC in the braid group Bn .
Figure 7. An example of the operation A ∗ B. The left

image shows two polygons that do not intersect. A is a
4-gon, B is a 3-gon, and C is the joining edge. The middle
image shows the computation of B C. The right image
shows the computation of A ∗ B.
Remark 3.13. For later use, we note that in neighborhood of PA the joining
edge C lies on left of cvh(A). Likewise, near PB the edge C lies on left of
cvh(B).
We list properties of the operation ∗ immediately follow from the defini-

tion.
Proposition 3.14. The operation ∗ is commutative: A ∗ B = B ∗ A. The
diagram of A ∗ B, denoted by A ∗ B, satisfies
A ∗ B = A (B C) = B (A C) = (A t B) C
and it is a single polygon with k + l sides. Thus, by Theorem 2.9, we see
that A ∗ B ∈ CnFct(Bn ). The definition also implies
cvh(A) t cvh(B) ⊂ cvh(A ∗ B).
3.2. Proof of Theorem 3.4.

Lemma 3.15. (See Figure 8) Let A, B ∈ CnFct(Bn ) be canonical factors
such that
• the diagram B is connected,
• the diagram A has k(> 1) connected components, and
• cvh(A) ⊂ cvh(B).
Then there exists a band generator C ∈ BGen(Bn ) such that such that A0 :=
AC satisfies
• A0 = A C,
• A0 is a canonical factor,
• its diagram A0 has k − 1 connected components, and
• cvh(A) ∪ C ⊂ cvh(A0 ) ⊂ cvh(B).
Proof. The diagram A contains a disjoint pair of connected components

facing to each other. Call them X and Y and their corresponding canonical
factors are denoted by X and Y respectively. Therefore, A = X t Y t Z
for some Z ∈ CnFct(Bn ). Denote the joining edge of X and Y by C. Let
C ∈ BGen(Bn ) denote the corresponding positive band generator for C. Note
that the pair (A, C) satisfy the (Step 3) condition of . We may define
A0 := A C = (X ∗ Y ) t Z
Its diagram A0 is exactly A except for the two components X and Y are
replaced by the single polygon X ∗Y = X (YC). Therefore, A0 is a canonical
factor by Theorem 2.9 and A0 consists of n − 1 components. By the nature
of the operation ∗ we have the inclusions cvh(A) ⊂ cvh(A0 ) ⊂ cvh(B).
Figure 8. The above figure is an example of Lemma ??.

The left image contains a 10-gon B and a 3-component
diagram A = X t Y t Z in red. On the right image, we
compute A0 = (X ∗ Y) t Z, which is a 2-component diagram
in red.
Corollary 3.16. (See Figures 9 & 10) Let A, B ∈ CnFct(Bn ) be canonical

factors such that
• the diagram A = A1 t · · · t Ak where k > 1 and Ai is connected, and
Then there exists a canonical factor C ∈ CnFct(Bn ) such that A0 := AC
satisfies
• A0 = A C,
• A0 is a canonical factor,
• its diagram A0 is a single polygon, and

• cvh(A) ⊂ cvh(A0 ) ⊂ cvh(B).
Moreover, up to reordering of the disjoint components,
A0 = (((A1 ∗ A2 ) ∗ A3 ) ∗ · · · Ak−1 ) ∗ Ak .
Figure 9. An example of Corollary 3.16. The left image

contains a 10-gon B, a 3-component diagram
A = X t Y t Z in red, and C = C1 t C2 . On the right image,
we compute A0 , which is a single polygon in red.
Figure 10. Another example of Corollary 3.16. The left

image contains a 3-component diagram A = X t Y t Z. On
the right image, we compute A0 = A C, which is a single
polygon.
Proof. Applying Lemma 3.15 k − 1 times, we inductively obtain band gen-

erators C1 , · · · , Ck−1 ∈ BGen(Bn ) such that
A0 := (· · · ((A C1 ) C2 ) · · · ) Ck−1 (3.1)
= AC1 C2 · · · Ck−1
is a canonical factor, the diagram A0 is connected, and cvh(A) ⊂ cvh(A0 ) ⊂
cvh(B). Therefore, it is enough to show the product C := C1 · · · Ck−1 is a
canonical factor.
By the construction in Lemma 3.15 we see

C1 ⊂ cvh(A C1 ) and cvh(A C1 ) ∩ int(C2 ) = ∅.
Thus interiors of the joining edges C1 and C2 are disjoint. Likewise for every
i = 1, . . . , k − 2 we have
C1 , · · · , Ci ⊂ cvh(AC1 · · · Ci )
and
cvh(AC1 · · · Ci ) ∩ int(Ci+1 ) = ∅.
This means interiors of the joining edges C1 , . . . , Ck are pairwise disjoint.
If Ci and Cj intersect, they only intersect at a single vertex, say P ∈ Dn .
Moreover, Remark 3.13 about joining edges implies that: i < j if and only if
in small neighborhood of P the edge Cj lies on the left of Ci . In other words,
near P , joining edges meet in the counterclockwise manner. The interior
disjointness property and the counterclockwise property guarantee that
C = C1 · · · Ck−1 = (((C1 C2 ) C3 ) · · · ) Ck−1
and its diagram is disjoint union of polygons, that is, by Theorem 2.9, C is
a canonical factor. Since the pair (A, C) satisfy the (Step 4) condition of
in Definition 3.11 we have A0 = A C.
We note that the interior disjointness property and the counterclockwise

property have already appeared in the setting of Definition 3.7.
Lemma 3.17. (See Figure 11) Let A, B ∈ CnFct(Bn ) be canonical factors
such that
• the diagram A is connected or empty, and
Then there exists a canonical factor C ∈ CnFct(Bn ) such that
B = A C = AC.
Proof. If A is empty, then A = e and we set C := B, done. In the following

we assume A is non-empty and connected.
The set B \ (A ∩ B) consists of disjoint arcs. Let k denote the number
of these disjoint arcs. If k = 0 then there is nothing to argue since A = B.
Assume that k > 0. Pick one of the arcs. Let m be the number of edges
to form the arc. Orient the polygon B clockwise. Following the orientation,
denote the m edges of the arc by b1 , . . . , bm . The product C1 := b1 · · · bm−1
is a canonical factor whose diagram C1 is an m-gon. We note that the pair
(A, C1 ) satisfy the (Step 1) condition for the operation . By Definition 3.7,
we see that
• the product AC1 = A C1 is a canonical factor
Figure 11. The above figure is an example of Lemma

3.17. The top left image contains a 10-gon B and a 3-gon A
in red. On the top right image, we draw C1 in blue. On the
bottom left image, we compute A C1 . We also draw C2 in
blue. On the bottom right, we compute B = (A C1 ) C2
.
• its diagram A C1 is connected,

• the number of arc components of B \ ((A C1 ) ∩ B) is k − 1, and
• cvh(A) ⊂ cvh(AC1 ) ⊂ cvh(B).
Thus, we can repeat the procedure for AC1 and B to obtain a canonical
factor AC1 C2 such that
cvh(AC1 ) ⊂ cvh(AC1 C2 ) ⊂ cvh(B).
We continue the procedure until we exhaust all the original k arcs of

B \ (A ∩ B). At the end, we have canonical factors C1 , · · · , Ck ∈ CnFct(Bn )
such that
AC1 · · · Ck = (((A C1 ) C2 ) · · · ) Ck = B.
Since their diagrams C1 , · · · , Ck are pairwise disjoint, by Theorem 2.9, the
product C := C1 · · · Ck = C1 t · · · t Ck is also a canonical factor and we
have B = AC = A C.
We prove the only-if part of Theorem 3.4. The if part of Theorem 3.4 will
be proved in Proposition 5.4.
Proposition 3.18. Let A, B ∈ CnFct(Bn ). If cvh(A) ⊂ cvh(B) then there

exists a Q ∈ CnFct(Bn ) such that B = AQ; thus, A ≺ B.
Proof. Suppose that cvh(A) ⊂ cvh(B) and the diagram B has k connected
components B1 , . . . , Bk . Suppose that A admits a decomposition A = A1 t
· · · t Ak such that cvh(Ai ) ⊂ cvh(Bi ) or Ai = e. The diagram Ai has
possibly multiple components.
If Ai 6= e applying Corollary 3.16 to every pair (Ai , Bi ), we can find a
canonical factor Ci ∈ CnFct(Bn ) such that
• A0i := Ai Ci = Ai Ci ∈ CnFct(Bn ),
• the new diagram A0i is connected and
• cvh(Ai ) ⊂ cvh(A0i ) ⊂ cvh(Bi ).
• cvh(Ci ) ⊂ cvh(A0i ) ⊂ cvh(Bi ).
If Ai = e then we may set Ci = e.

Next we apply Lemma 3.17 to the pair (A0i , Bi ) for each i = 1, . . . , k. Then
there exists a canonical factor Di ∈ CnFct(Bn ) such that Bi = A0i Di .
Since components of B are pairwise disjoint, diagrams Ci are pairwise
disjoint, so the product C := C1 t · · · t Ck is a canonical factor. Similarly,
the product D := D1 t · · · t Dk is a canonical factor. With the C and D we
have B = (A C) D.
It remains to prove that the product Q := CD satisfies CD = C D;
thus, Q is a canonical factor and B = A Q.
Note that
cvh(C) ⊂ cvh(A C) ⊂ cvh(B), (3.2)

cvh(D) ⊂ cvh(B), (3.3)
int(cvh(D)) ∩ cvh(A C) = ∅.
This means if a component of C and a component of D intersect, then they

must intersect at a vertex of Dn . Moreover, near that vertex, the component
of D lies on the left of the component of C. Thus the (Step 4) condition of in
Definition 3.7 is satisfied which yields that the product Q = CD = CD.
Corollary 3.19. For every canonical factor A ∈ CnFct(Bn ) there exists a

B ∈ CnFct(Bn ) such that A B = AB = δ.
4. Partially Ordered Set (Wds(β), ⇒)
Given β ∈ Bn , we define a set:

r
W is a word representing β,
Wds(β) := W = δ A1 · · · Ak

r ∈ Z, Ai ∈ CnFct(Bn )
In this section, extending the partial ordering ≺ on the set CnFct(Bn ), we

introduce a partial ordering ⇒ on the set Wds(β) of braid words representing
β ∈ Bn .
We start with defining a relation ⇒ between words.
Definition 4.1. Let A, B, A0 , B 0 ∈ CnFct(Bn ). We denote
AB ⇒ A0 B 0
if the following are satisfied:
a) AB = A0 B 0 as braid elements in Bn .
b) A ≺ A0 (equivalently, cvh(A) ⊂ cvh(A0 )).
In other words, the word A0 B 0 is more left-weighted than AB. If no A0 , B 0 ∈

CnFct(Bn ) satisfy AB ⇒ A0 B 0 then we say that the word AB is maximally
left weighted. In case of B 0 = e, B 0 may be omitted.
Before giving examples of AB ⇒ A0 B 0 we recall the right complementary

set R(A) and the starting set S(B). The reason we do so here is that: The
word AB admits a more left-weighted word A0 B” if and only if R(A) ∩
S(B) 6= ∅ (see Remark 4.4).
Definition 4.2. Let A ∈ CnFct(Bn ). Define the right complementary set

R(A) and the starting set S(A) as follows:
R(A) = {c ∈ BGen(Bn ) | Ac ∈ CnFct(Bn )}

S(A) = {c ∈ BGen(Bn ) | A = cA0 for some A0 ∈ CnFct(Bn )}.
In terms of the diagram, R(A) is the set of edges C which are
• disjoint from the diagram A of A; i.e., A ∩ C = ∅, or

• satisfying the two conditions in Definition 3.11; i.e., cvh(A) ∩ C =
A ∩ C ⊂ {vertices of Dn } and in small neighborhood of each P ∈
cvh(A) ∩ C the edge C lies on the left of A.
It is easy to see that S(A) is the set of all edges of the diagram A.
Example 4.3. For each canonical factor A ∈ CnFct(B4 ) of B4 , the set R(A)
is described as follows (up to rotation):
R( ) = {( ), ( ), ( )} = ( ) R( ) = {( ), ( )} = ( )
R( ) = {( )} = ( ) R( ) = {( )} = ( )
R( ) = {( )} = ( )
The thick highlighted edges represent elements of R(A), while the black
edges represent A. See Example 4.6 below for explanation of R( ).
The set S(A) is described as follows (up to rotation):
S( ) = {( )} S( )={ }
S( ) = {( ), ( ), ( )} S( ) = {( ), ( )}
S( ) = {( ), ( ), ( ), ( ), ( ), ( )}
The last equation comes from (2.2).
The maximally left weighted condition (Definition 4.1) of word AB can

be interpreted in terms of the sets R(A) and S(B).
Remark 4.4. The algebraic condition R(A) ∩ S(B) = ∅ is equivalent to the
pictorial condition that the diagram B does not intersect the highlighted
edges of A (cf. Example 4.3). Therefore, R(A) ∩ S(B) = ∅ if and only if
no factors A, B ∈ CnFct(B4 ) can increase the partial order AB ⇒ A0 B 0 ,
equivalently the AB is maximally left weighted.
Next, we will show two examples of AB ⇒ A0 B 0 using R(A) and S(B).

Example 4.5. This example will show that
AB := ( )( )⇒( )( )⇒( )( )=( ).
We will start by highlighting A as follows:
( )( ).
Note that B is contained in the highlighted edges of A. Move one of these

lines from B to A to make it more left weighted. It does not matter which
line you choose to move. Then, we obtain
AB ⇒ A0 B 0 := ( )( ).
Next, we repeat this process until R(A0 ) ∩ S(B 0 ) = ∅. If we highlight A0 , we

have
( )( ).
Note that the edge of B 0 is contained in the highlighted edge of A0 . Move

this line from B 0 to A0 to make it more left weighted. Then, we obtain
A0 B 0 ⇒ A00 B 00 := ( )( ).
Example 4.6. This example will show
( )( )⇒( ).
Note that the left factor ( ) consists of two lines. Whenever the left factor
consists of disjoint two lines, we must take an additional step before we
highlight it. First, move one of the lines from ( ) to ( ) to decrease the
partial ordering and obtain the word ( )( ). We have
AB := ( )( )⇒( )( ).
Now we may highlight A:

( )( ).
Note that the edges of B are contained in the highlighted edges of A. Move
one of the edges from B to A to increase the partial ordering. We chose to
move the leftmost edge and obtain:
AB ⇒ A0 B 0 := ( )( ).
Keep in mind that triangles are oriented clockwise, so we get B 0 =( ).
Writing B 0 =( ) does not follow the orientation, so this is not allowed.

Then, we rehighlight A0 :
( )( ).
Again, we see that the edges of B 0 are contained in the highlighted edges of
A0 . Move the remaining edge from B 0 to A0 to have more left weighted word
A0 B 0 ⇒ A00 B 00 := ( )( ).
This concludes: ( )( )⇒ ( ).
Pairs (A, B) of canonical factors of 4-braids with R(A) ∩ S(B) 6= ∅ (resp.

= ∅) are listed in Table 1 (resp. Table 2) below.
Example 4.7. In Table 1, we have listed all possible pairs (Ai , Ai+1 ) of
canonical factors up to rotation whose partial ordering can be increased.
Namely, there are A0i , A0i+1 ∈ CnFct(B4 ) such that Ai Ai+1 ⇒ A0i A0i+1 .
Order Increasable Multiplications for CnFct(B4 ) up to rotation

Ai+1
Single Edge Disjoint Edges Triangle
Ai
( )( )⇒( ) ( )( )⇒ ( )∗∗ ( )( )⇒( )( )
( )( )⇒( ) ( )( )⇒( )( ) ( )( )⇒( )( )
Single ( )( )⇒( ) ( )( )⇒( )( ) ( )( )⇒( )( )
Edge ( )( )⇒( ) ( )( )⇒( )
( )( )⇒( )( )
( )( )⇒( )( )
Disjoint ( )( )⇒ ( )∗ ( )( )⇒( )( )∗
Edges
( )( )⇒( ) ( )( )⇒( )( ) ( )( )⇒( )( )
Triangle ( )( )⇒( )( )
Table 1. Order increasable multiplications for CnFct(B4 )

up to rotation.
* see Example 2.16.
** see Example 2.15.
Example 4.8. In addition to the list in Table 1, if Ai ∈ CnFct(B4 ) and

Ai+1 = ( ), then
Ai ( )⇒( ) τ (Ai )
where τ : B4 → B4 is the inner automorphism defined by τ (ai ) = δ −1 ai δ =

ai+1 and τ (bi ) = δ −1 bi δ = bi+1 . Thus, τ rotates a canonical factor diagram
90◦ counterclockwise.
Example 4.9. In Table 2, we list all possible pairs (Ai , Ai+1 ) up to rotation
whose partial order cannot be increased. Equivalently, R(Ai ) ∩ S(Ai+1 ) = ∅.
We omitted pairs of (Ai , Ai+1 ) where Ai = Ai+1 .
Finally, we extend the relation ⇒ (Definition 4.1) to the set Wds(β).

Order Non-Increasing Multiplications for B4 up to rotation

Ai+1
Single Edge Disjoint Edges Triangle
Ai
( )( ) ( )( )
Single ( )( )
Edge ( )( )
( )( )
Disjoint ( )( ) ( )( ) ( )( )
Edges ( )( )
( )( )
( )( ) ( )( ) ( )( )
( )( )
Triangle ( )( )
( )( )
( )( )
Table 2. Maximally left weighted words Ai Ai+1 up to

rotation. The cases where Ai = Ai+1 or Ai = δ are omitted.
Definition 4.10. For braid words δ ` A1 · · · Ak and δ m A01 · · · A0n in Wds(β),

we define
δ ` A1 · · · Ak ⇒ δ m A01 · · · A0n
if one of the following is true:
a) ` < m
b) ` = m, k = n + 1, and there exists some i ∈ {1, . . . , k} such that
Ai = ( ), Aj = A0j for all j < i and Aj = A0j−1 for all j > i.
c) ` = m, k = n, and there exists some i ∈ {1, . . . , k − 1} such that
Aj = A0j for all j 6= i, i + 1 and Ai Ai+1 ⇒ A0i A0i+1 .
With the relation, (Wds(β), ⇒) becomes a partially ordered set.

5. The Left Canonical Form
5.1. Definition of the Left Canonical Form. In this section we intro-

duce the left canonical form LCF(β) of a braid β ∈ Bn as the maximal
element in the partially ordered set (Wds(β), ⇒).
The word problem can be solved using the left canonical form by Xu for
3-braids [25], Kang, Ko and Lee for 4-braids [15], and Birman, Ko, Lee for
general n-braids [3]. Namely, β = β 0 if and only if LCF(β) = LCF(β 0 ).
We first introduce inf(β), sup(β), `(β) since the left canonical form LCF(β)
will be defined to be the unique element of Wds(β) which
(1) achieves the inf(β) and sup(β) simultaneously, and
(2) is maximal in the partially ordered set (Wds(β), ⇒).
Definition 5.1. For β ∈ Bn , define
inf(β) := max{r ∈ Z | δ r ≤ β},
sup(β) := min{s ∈ Z | β ≤ δ s },
`(β) := sup(β) − inf(β).
Here, `(β) is called the canonical length which is different from ||β||, the
usual minimal word length in band generators. For example, `( ) = 1 and
||( )|| = ||a2 a1 || = 2.
Using these definitions, we can introduce the left canonical form.

Theorem 5.2. [25, 15, 3] For any n-braid β, there exist unique r ∈ Z,
unique k ∈ Z≥0 and unique canonical factors A1 , . . . , Ak ∈ CnFct(Bn )\{e, δ}
such that
• β = δ r A1 A2 · · · Ak as braid elements in Bn and
• any consecutive pairs A1 A2 , . . . , Ak−1 Ak are maximally left weighted.
Moreover r and k satisfy
inf(β) = r, sup(β) = r + k, and `(β) = k.
The unique factorization δ r A1 A2 · · · Ak of β is called the left canonical form
of β and denoted by LCF(β).
Remark 5.3. The left canonical form is not a conjugacy invariant. See
Example 5.6.
We end the section by proving the “if part” of Theorem 3.4.

Proposition 5.4. Let A, B ∈ CnFct(Bn ). If there exists Q ∈ CnFct(Bn )
such that AQ = B then cvh(A) ⊂ cvh(B).
Proof. Assume that canonical factors A, B, Q satisfy AQ = B. If the pair

(A, Q) satisfy all the conditions of in Definition 3.11, so that AQ = A Q,
then by the property (c) above Definition 3.10, it follows that cvh(A) ⊂
cvh(B).
If (A, Q) does not satisfy the conditions of in Definition 3.11, then
LCF(AQ) has canonical length `(AQ) = 2. However since LCF(AQ) =
LCF(B) = B, the canonical length `(AQ) = `(B) = 1, which is a con-
tradiction.
5.2. Left Canonical Form Algorithm. In this section, we will review

Kang, Ko, and Lee’s algorithm for left canonical form [15] for 4-braids. It
can be extended to general n-braids [3]. This is a polynomial time algorithm
which gives a unique left canonical form for a braid word β.
Let β ∈ B4 be a braid element. Assume that β is represented by a
positive braid word in band generators; β = A1 A2 · · · Ak ∈ B4+ . Note that
Ai ∈ BGen(B4 ) ⊂ CnFct(B4 ) so Ai ∈ CnFct(B4 ). Change the composition
of β by the following rules.
a) If Ai = ( ) for some i, then remove Ai from the word A1 · · · Ak .

b) Replace Ai Ai+1 with A0i A0i+1 so that Ai Ai+1 ⇒ A0i A0i+1 (as listed in
Table 1).
Apply (a) and (b) until the partial order cannot be raised anymore. The
resulting word is the left canonical form, LCF(β), as defined in Theorem 5.2.
We remark that the resulting word LCF(β) does not depend on the way we
apply (a) and (b). One may start with increasing the partial order of any
pair (Ai , Ai+1 ) in the braid word.
If β is not positive (β ∈/ B4+ ), we get rid of the negative exponent terms.
2
Suppose β = A11 A2 · · · Akk , where Ai ∈ CnFct(B4 ) and i = ±1. Suppose
that i = −1 for some i. Then, replace A−1 i with A0i δ −1 for some canonical
0
factor Ai ∈ CnFct(B4 ). Up to rotation, we have:
( )−1 = ( )δ −1
( )−1 = ( )δ −1
Applying the rule A0i δ −1 = δ −1 τ −1 (A0i ) a number of times will shift δ −1 to

the beginning of the word. Repeat this procedure until we exhaust all the
factors with negative exponents. It will give us a word δ −r P for some r ∈ N
and positive word P ∈ B4+ in band generators. Finally we apply the above
algorithm to the positive word P to obtain LCF(β).
Example 5.5. We will use the Diagrammatic Left Canonical Form Algo-
rithm to find LCF(β), where
β = b2 a1 b1 a4 a2
=( )( )( )( )( )
= A1 A2 A3 A4 A5
in band generators. We will look at each pair of (Ai , Ai+1 ) to determine if
we can increase the partial order using part (b) of the algorithm.
β=( )( )( )( )( ) Highlight A4
=( )( )( )( ) Increase partial order
=( )( )( ) Increase partial order (See Ex. 2.16)
=( )( )( ) Apply Ai ( )⇒( )τ (Ai )
=( )( )( ) Highlight A02
=( )( ) Increase partial order
Thus,
LCF(β) = ( )( ).
Note that it does not matter which pair (Ai , Ai+1 ) is considered first.
Example 5.6. Consider the knot 72 represented by the 4-braid
β = a1 a1 a1 a2 a−1 −1
1 a2 a3 a2 a3 .
Using the LCF Algorithm, we find

LCF(β) = δ −1 a4 a4 a4 a1 b2 a2 a3 a3 .
The knot 72 is strongly quasi-positive, as it can be represented by another
braid β 0 = a1 a1 b2 b1 a3 which is strongly quasi-positive. We can observe
LCF(β 0 ) = β 0 . For this example, β and β 0 are conjugate but LCF(β) 6=
LCF(β 0 ).
5.3. Super Summit Set and The Left Canonical Form. In this section,
we review the definition for the super summit set, which is a conjugacy class
invariant for a braid. We also give important properties of the super summit
set in Proposition 5.9 and Theorem 5.10. The conjugacy problem has been
solved using band generators for B3 by Xu [25], for B4 by Kang, Ko, and
Lee [15], and for Bn by Birman, Ko, and Lee [3].
Definition 5.7. For the conjugacy class [β] of β ∈ Bn we define:

inf[β] := max{inf(β 0 ) | β 0 is conjugate to β} (5.1)
sup[β] := min{sup(β 0 ) | β 0 is conjugate to β}
Definition 5.8. The super summit set of a braid β, denoted SSS(β), is the
set of conjugates β 0 of β such that the canonical length `(β 0 ) is minimal
among all the conjugates of β.
In [15, Corollary 4.6] it is proved that the inf[β] and sup[β] can be achieved
simultaneously by the same element. Since the canonical length `(β) =
sup(β) − inf(β) we have the following:
Proposition 5.9. Any super summit element β 0 ∈ SSS(β) realizes inf[β]
and sup[β]. Namely inf[β] = inf(β 0 ) and sup[β] = sup(β 0 ).
An important fact proved by Elrifai and Morton is that: two braids are
conjugate if and only if their super summit sets are identical if and only if
their super summit sets intersect [7, 9].
Let W := LCF(β) = δ r A1 A2 · · · Ak be the left canonical form of β ∈ Bn .
Let τ be the inner automorphism of Bn defined by
τ (β) = δ −1 βδ.
In terms of the non-crossing partition diagram, τ rotates the canonical factor
diagram 2π/n counterclockwise. Define the cycling c(W ) and the decycling
d(W ) of the word W as follows:
c(W ) := δ r A2 · · · Ak τ −r (A1 )
d(W ) := δ r τ r (Ak )A1 · · · Ak−1
In general, β ∈ Bn is not necessarily an element of SSS(β). The next

theorem characterizes elements of SSS(β).
Theorem 5.10 (Kang, Ko, Lee [15] and Birman, Ko, Lee [3]). Let β ∈ Bn
be an n-braid with canonical length `(β) ≥ 3. Let W = LCF(β). If `(W ) =
`(c(W )) = `(d(W )), then β is an element of the super summit set SSS(β).
This is a key theorem to solving the conjugacy problem. For the solution
to the conjugacy problem, readers may refer to [3, Section 5].
6. Detection of SQP and ASQP braids
In this section, we explore how the left canonical form looks like for (al-
most) strongly quasipositive braids.
Definition 6.1. A braid β is a strongly quasipositive braid (SQP) if it can
be represented by a word W written as a product of positive powers of some
of the band generators. A link K is a strongly quasipositive link if K can be
represented by a strongly quasipositive braid.
Definition 6.2. A braid β is an almost strongly quasipositive braid (ASQP)

if it can be represented by a word W written as a product of band generators
and inverses such that the number of inverses used is at most one. A link K
is an almost strongly quasipositive link if K can be represented by an almost
strongly quasipositive braid.
Additionally, a braid β is a strictly almost strongly quasipositive braid if
β is almost strongly quasipositive but not strongly quasipositive. A link K
is strictly almost strongly quasipositive if K is almost strongly quasipositive
but not strongly quasipositive.
We have four results on SQP braids and ASQP braids.

Theorem 6.3. Let n ≥ 3. An n-braid β ∈ Bn is strongly quasipositive if
and only if inf(β) ≥ 0.
Proof. (⇒) Let β ∈ Bn be strongly quasipositive. By definition, β can be

represented by a positive word, W , in band generators of Bn . Apply the
algorithm for left canonical form (Section 5.2) to W we obtain inf(β) ≥ 0.
(⇐) Suppose β has inf(β) ≥ 0. Then,
LCF(β) = δ r A1 A2 · · · Ak
where r ≥ 0, Ai ∈ CnFct(Bn ), and R(Ai )∩S(Ai+1 ) = ∅. Since CnFct(Bn ) ⊂
Bn+ , each Ai is a positive word in band generators. Therefore, β ∈ Bn+ , in
particular β is strongly quasipositive.
The corollary below follows from Proposition 5.9 on super summit set.
Corollary 6.4. A braid β ∈ Bn is conjugate to a strongly quasipositive
braid if and only if every element β 0 ∈ SSS(β) has inf(β 0 ) ≥ 0.
In Example 5.6, we give a braid representative β ∈ B4 of the knot that

is not strongly quasipositive but is conjugate to a strongly quasipositive
braid. The next theorem is the analog of Theorem 6.3 in the almost strongly
quasipositive case.
Theorem 6.5. Let β ∈ Bn with n ≥ 3. If β is almost strongly quasipositive,
then inf(β) ≥ −1. Further, if β is strictly almost strongly quasipositive, then
inf(β) = −1.
Proof. Suppose β ∈ Bn is almost strongly quasipositive. If β is strongly

quasipositive, we can apply Theorem 6.3. Therefore, we will assume that β
is strictly almost strongly quasipositive. Suppose that β can be represented
by a word, W , with exactly one negative band. In the algorithm for left
canonical form, a negative band contributes δ −1 . This implies that inf(W ) ≥
−1. On the other hand, since β is not strongly quasipositive, we have
inf(β) < 0 by Theorem 6.3. Therefore, we obtain inf(W ) = −1. Since
LCF(β) = LCF(W ), we get inf(β) = −1.
Theorems 6.3 and 6.5, and Corollary 6.4 give the following:
Corollary 6.6. If β ∈ Bn with n ≥ 3 is conjugate to an ASQP (resp.
strictly ASQP) braid, then every element β 0 ∈ SSS(β) has inf(β 0 ) ≥ −1
(resp. inf(β 0 ) = −1).
For both Theorem 6.5 and Corollary 6.6, the converse direction does not
hold in general. See Example 7.4 below. However, if we add an additional
condition, the converse of Corollary 6.6 holds. See Theorem 9.4 below.
7. The Bennequin Inequality and LCF(β)
We define three negative band numbers nb(β), nb[β], nb(K) as follows:

Definition 7.1. For a braid β ∈ Bn the minimal number of negative bands,
nb(β), is defined as follows:

β is represented by a word in band
nb(β) = min k
generators containing k negative bands
Similarly for the conjugacy class [β] of β we define:
nb[β] = min{nb(β 0 ) | β 0 is conjugate to β }
For a knot or link K ⊂ S 3 we define:
nb(K) = min {nb(β) | β is a braid representative of K }
For a braid representative β of a knot K,

0 ≤ nb(K) ≤ nb[β] ≤ nb(β).
We can characterize SQP braids and ASQP braids in terms of the negative
band number.
• The braid β is SQP if and only if nb(β) = 0.
• β is ASQP if and only if nb(β) ≤ 1.
• β is strictly ASQP if and only if nb(β) = 1.
The invariant nb(K) is closely related to the Bennequin type inequalities.
Bennequin [2] showed the maximal self-linking number SL(K) of a given
knot type K is bounded above by −χ(K) = 2g(K) − 1 where g(K) is the
genus of the knot.
SL(K) ≤ 2g(K) − 1
Bennequin’s inequality has been extended to concordance invariants includ-
ing the slice genus g4 (K), Ozsváth-Szabó’s tau-invariant τ (K) [21] and Rus-
mussen’s s-invariant s(K) [22] as follows.
SL(K) ≤ 2τ (K) − 1 ≤ 2g4 (K) − 1 ≤ 2g(K) − 1
SL(K) ≤ s(K) − 1 ≤ 2g4 (K) − 1 ≤ 2g(K) − 1
Ito and Kawamuro [13] defined the defect D(K) of the (original) Bennequin
inequality
1
D(K) = (2g(K) − 1 − SL(K)).
2
It is shown that D(K) ∈ Z and D(K) ≤ nb(K). In fact, it is conjectured
that D(K) = nb(K).
We will study the negative band numbers nb(β), nb[β], and nb(K) in
terms of the left canonical form.
Lemma 7.2. We have nb(β) ≥ 1 if and only if inf(β) < 0. (Equivalently,
nb(β) = 0 if and only if inf(β) ≥ 0.)
Proof. Let β ∈ Bn with n ≥ 3 and suppose that nb(β) ≥ 1. Notice nb(β) ≥ 1

if and only if β is not strongly quasipositive. Thus, nb(β) ≥ 1 is equivalent
to inf(β) < 0 by Theorem 6.3.
Theorem 7.3. Let β ∈ Bn with n ≥ 3 and nb(β) ≥ 1.
0 < − inf(β) = | inf(β)| ≤ nb(β).
Proof. The first inequality follows from Lemma 7.2.

Suppose that β is represented by a word W in band generators with
nb(β)(≥ 1) negative bands. Apply the algorithm for left canonical form
(Section 5.2) to W . The algorithm first replaces each negative band with
an Aδ −1 for some canonical factor A ∈ CnFct(Bn ), then shifting δ −1 to the
left. During the shifting process it is possible that some δ −1 vanishes if
cancellation occurs. Therefore, we obtain −nb(β) ≤ inf(β).
We study when the inequality −nb(β) ≤ inf(β) in Theorem 7.3 becomes

a strict inequality in Proposition 8.6.
Example 7.4. We give an example of Theorem 7.3 where strict inequality
−nb(β) < inf(β) holds. Consider a 4-braid β := a3 a−1 −1
1 a2 b2 b1 a1 b2 b1 a3 .
Running the LCF algorithm (Section 5.2) we obtain
LCF(β) = δ −1 a2 a3 b2 b1 a1 b2 b1 a3 ,
i.e., inf(β) = −1. Clearly, nb(β) ≤ 2. In the following, we will show nb(β) =
2.
Figure 12. The braid β := a3 a−1 −1

1 a2 b2 b1 a1 b2 b1 a3 .
To this end, we first show that the topological knot type of β has braid
index 4. To do this, we will use the Morton-Franks-Williams Inequality
[20, 8]. Let K be the topological type of the braid closure β̂. Let i be
the braid index of K. We will show i(K) = 4, i.e., β realizes the braid
index of K. Let E be the largest power and e the smallest power of a
in the HOMFLY-PT polynomial of K. Then, the Morton-Franks-Williams
inequality states
1
i ≥ (E − e) + 1.
2
Mathematica’s Knot Theory package computes the HOMFLY-PT polyno-
mial for K as
−a−6 +a−4 +a−2 +z 2 a−8 −3z 2 a−6 +4z 2 a−4 −2z 4 a−6 +3z 4 a−4 −z 4 a−2 +z 5 a−4 .
In the above polynomial, we have E = −2 and e = −8. Therefore, the
Morton-Franks-Williams inequality tells us that
i ≥ 4.
Therefore, the braid index for K is 4. Now we can proceed to our claim.
Claim: nb(β) ≥ 2.
Proof. First, we will show that g(K) ≤ 3, where g(K) is the genus of K.
Consider a Bennequin surface, ΣW , of K coming from the braid β repre-
sented by the word W = a3 a−1 −1
1 a2 b2 b1 a1 b2 b1 a3 made of four disks joined
together with nine twisted bands.
Figure 13. Bennequin Surface ΣW
Therefore,
χ(K) ≥ χ(ΣW ) = 4 − 9 = −5.
Using the fact that χ(K) = 1−2g(K) for any knot K, we find that g(K) ≤ 3.
Next, we will show that g(K) ≥ 3. Recall that
degree(∆K (t)) ≤ 2g(K)
where ∆K (t) is the Alexander polynomial of the knot [23]. Using Math-
ematica’s Knot Theory Package, the Alexander polynomial of K is given
by
1 − 6t + 17t2 − 23t3 + 17t4 − 6t5 + t6 .
The degree of this polynomial is 6. Therefore, g(K) ≥ 3.
We have shown that g(K) = 3. Hence, χ(K) = −5. Recall the Bennequin
Inequality [2]
SL(K) ≤ −χ(K)
where SL(K) is the maximal self linking number of the link type K. By the
truth of the Generalized Jones Conjecture [6, 16, 18], SL(K) = −4 + p − n.
Here, p is the number of positive bands and n is the number of negative
bands in a braid representative which realizes the minimal braid index 4. In
our example SL(K) = −4 + 7 − 2 = 1. Observe the following inequality:
SL(K) ≤ −χ(K) ≤ −4 + p + n
Subtracting SL(K) = −4 + p − n from each side, we get
0 ≤ −χ(K) − SL(K) ≤ 2n
We have χ(K) = −5 and SL(K) = 1. Therefore,
0 ≤ 5 − 1 ≤ 2n
Therefore, 2 ≤ n. This shows that β must contain at least 2 negative bands,
i.e., nb(β) ≥ 2.
We may conclude nb(β) = 2.

We note that nb(β) = 2 implies β is not ASQP. Further, after computing
cyclings and decyclings of LCF(β), we can apply Theorem 2.27 to verify that
β ∈ SSS(β). We also have that inf(β) = −1, which means the converse
direction of Corollary 6.6 does not hold in general.
8. The Negative Band Number and Reduction of LCF(β)
The reduction operation for braids has been introduced by Kang, Ko

and Lee [15] in order to solve the shortest word problem for 4-braids (see
Theorem 9.1). We first review the reduction operation then apply it to
investigate the relation between the negative band number nb(β) and the
left canonical form LCF(β).
8.1. Review of Reduction Operation.

Definition 8.1. Given a word W = δ r W1 W2 · · · Ws with δ −1 < Wi < δ and
Wi 6= e, define red(W ) as:
(1) if r ≥ 0 or Wi < e for all i = 1, . . . , s, then red(W ) = W ;

(2) otherwise (i.e., r < 0 and there exists some i with Wi ∈ CnFct(B4 )),
choose a word Wk ∈ CnFct(B4 ) whose word length ||Wk || is maximal
among all W1 , . . . , Wk .
By Corollary 3.19 there exists Vk ∈ CnFct(B4 ) such that Wk Vk =
Wk Vk = δ. Put Wk0 = (Vk )−1 . We define
red(W ) = δ r+1 τ (W1 ) · · · τ (Wk−1 )Wk0 Wk+1 · · · Ws ,
Define Red(W ) := red|r| (W ), i.e., repeat the above algorithm until the ex-
ponent of δ is non-negative.
Example 8.2. This example will show how to compute Red(W ), where
W = δ −2 (a3 a2 )(a4 a3 )a4 b1 b2
=( )−2 ( )( )( )( )( )
= δ −2 W1 W2 W3 W4 W5
Here, r < 0, so we will proceed to option (2) in the algorithm. We will
choose the word W2 = (a4 a3 ) of maximal word length among all Wi . Then,
W20 = δ −1 W2
= δ −1 (a4 a3 )
= (a4 a3 a2 )−1 (a4 a3 )
= a−1 −1 −1
2 a3 a4 a4 a3
= a−1
2
Thus,
red(W ) = δ −1 τ (W1 )W20 W3 W4 W5
= δ −1 (a4 a3 )a−1
2 a4 b1 b2
We will repeat this process one more time. Again, we will choose the word
(a4 a3 ) of maximal length among all Wi . Then,
Red(W ) = red2 (W ) = a−1 −1
2 a2 a4 b1 b2 .
We note that both red(W ) and Red(W ) depend on choices. However, the
next lemma shows that the word-length ||Red(W ))|| is uniquely determined.
Lemma 8.3. (Kang, Ko, Lee [15, Lemma 5.1]) Let β ∈ Bn be an n-braid.
For any word representative W of β, every reduced word of the left canonical
form minimizes the word length; namely,
||Red(LCF(β))|| ≤ ||W ||.
Regarding the negative band number nb(β) and a shortest word we ob-
serve the following:
Lemma 8.4. Let W be a word representing β ∈ Bn . The negative band

number nb(β) is realized by W if and only if W gives a shortest word repre-
senting β.
Proof. Let W be a shortest word representing β and let p (resp. n) be the

number of positive (resp. negative) bands in the word W . Assume that a
braid word W 0 represents β and realizes nb(β). Let p0 (resp. n0 ) be the
number of positive (resp. negative) bands in the word W 0 . We get
n0 = nb(β) ≤ n.
Since W is a shortest word,

p0 + n0 = ||W 0 || ≥ ||W || = p + n.
Since W and W 0 belong to the same conjugacy class [β], they have the same
writhe.
p0 − n0 = writhe(W 0 ) = writhe(W ) = p − n
Therefore, n0 ≥ n.
Above two paragraphs give n = n0 = nb(β).
As a corollary of the above two lemmas, we observe the following:

Corollary 8.5. Every reduced word Red(LCF(β)) achieves nb(β).
8.2. Bounds of nb(β) in Terms of inf(β) and sup(β). With the above
preparation, we are now able to discuss the relation between nb(β) and
inf(β). We first discuss when the inequality −nb(β) ≤ inf(β) in Theorem 7.3
becomes a strict inequality.
Proposition 8.6. Let β ∈ Bn with n ≥ 3 and nb(β) ≥ 1. If there is a
shortest word representing β that contains A−1 for some canonical factor
A ∈ CnFct(Bn ) of word length ||A|| ≥ 2 then we get a strict inequality
−nb(β) < inf(β).
Proof. Let W be a shortest word representing β and containing A−1 for some
canonical factor A ∈ CnFct(Bn ) of word length ||A|| ≥ 2. By Lemma 8.4,
W realizes the negative band number nb(β), and the contribution of A−1 to
the numerical value nb(β) is ||A|| ≥ 2.
We note that W is a non-positive word. According to the LCF algorithm
(Section 5.2) for non-positive words, every canonical factor with negative
exponent contributes one δ −1 to the left canonical form. In particular, the
factor A−1 ∈ W contributes δ −1 (or δ 0 = e if some cancellation occurs)
to LCF(β). In other words, A−1 contributes −1 (or 0 if some cancellation
occurs) to inf(β).
Comparing the contribution of A−1 to nb(β) and inf(β), we obtain the
desired strict inequality.
In the next theorem, using the reduction operation we improve the in-
equality −nb(β) ≤ inf(β) in Theorem 7.3 to an equation for 3-braids:
Theorem 8.7. Let β be an n-braid. If inf(β) < 0 then
nb(β) ≤ (n − 2)| inf(β)| − min{0, sup(β)}.
Moreover, the equality holds when n = 3 and we have
nb(β) = | inf(β)| − min{0, sup(β)}.
Proof. Recall Definition 5.1 and put r := − inf(β) = | inf(β)| > 0 and
k := `(β) ≥ 0. We have
sup(β) = −r + k.
Theorem 5.2 implies that for some A1 , · · · , Ak ∈ CnFct(Bn ) \ {e, δ} we have
LCF(β) = δ −r A1 · · · Ak . By Corollary 3.19 for each i = 1, . . . , k there exists
a canonical factor A0i ∈ CnFct(Bn ) \ {e, δ} such that Ai A0i = δ.
Recall the inner automorphism τ : Bn → Bn defined by τ (β) = δ −1 βδ.
For a canonical factor A ∈ CnFct(Bn ), τ (A) is diagrammatically counter-
clockwise 2π/n rotation of A and τ (A−1 ) = (τ (A))−1 . Thus τ preserves the
word length;
||τ (A)|| = ||A|| = ||τ (A−1 )||.
Case 1: Suppose −r + k ≥ 0. Then min{0, sup(β 0 )} = 0. We apply

the reduction operation to LCF(β) and obtain a Red(LCF(β)). By Defini-
tion 8.1, among the canonical factors A1 , · · · , Ak in LCF(β) the ones with
r largest word length, say Ai1 , · · · , Air , are replaced by the negative words
(A0i1 )−1 , · · · , (A0ir )−1 up to rotation by τ . The rest of the k − r canonical
factors in LCF(β) are kept the same up to rotation by τ .
Since every canonical factor in CnFct(Bn ) \ {e, δ} has word length at
most n − 2 and τ preserves the word lengths of canonical factors (and their
inverses), the number of negative bands in Red(W ) is at most (n − 2)r. This
gives
nb(β) ≤ (n − 2)r = −(n − 2) inf(β).
When n = 3 we can say further. Every canonical factor in

CnFct(B3 ) \ {e, δ} = {a1 , a2 , a3 }
has word length exactly 1. This means each of (A0i1 )−1 , · · · , (A0ir )−1 corre-
sponds to exactly one negative band and we obtain
nb(β) ≤ r = − inf(β).
On the other hand, Theorem 7.3 gives −nb(β 0 ) ≤ inf(β 0 ). Therefore, for
3-braids we obtain
−nb(β) = inf(β).
Case 2: If −r + k < 0 we have min{0, sup(β)} = sup(β). Reduction

operation gives: up to rotation by τ
Red(LCF(β)) = δ −r+k (A01 )−1 · · · (A0k )−1 .
Recall Lemma 8.3 and Lemma 8.4, which state that Red(LCF(β)) realizes
the negative band number nb(β). Thus, we have
nb(β) = ||δ||(r − k) + ||A01 || + · · · + ||A0k ||.
Since the word length of δ is ||δ|| = n − 1 and ||A0i || ≤ n − 2, we get an

upper bound of nb(β)
nb(β) ≤ (n − 1)(r − k) + (n − 2)k (8.1)
= (n − 2)r + r − k (8.2)
= −(n − 2) inf(β) − sup(β).
For 3-braids, since ||δ|| = 2 and ||A0i || = 1, we get an explicit formula of
nb(β)
nb(β) = 2(r − k) + k = r − (−r + k) = − inf(β) − sup(β).

9. Relations Between inf(β) and nb(β) for β ∈ SSS(β)
The solution to the shortest word problem for 4-braids was given by Kang-
Ko-Lee [15]. The result is a generalization of Xu’s solution to the shortest
word problem for 3-braids [25].
Theorem 9.1. (Kang, Ko, Lee [15, Theorem 5.2]) Given a 4-braid β, let β 0
be an element in SSS(β). Then Red(LCF(β 0 )) gives a shortest word among
all the conjugates of β.
In this section, we explore the relation between inf(β) and nb(β) for a
super summit element β ∈ SSS(β).
Recall nb[β] = min{nb(β 0 ) | β 0 is conjugate to β} in Definition 7.1, the
minimal number of negative bands for the conjugacy class of β. In the next
lemma, we show that every shortest word realizes nb[β].
Lemma 9.2. Let β ∈ Bn . A word W is a shortest word among all the
conjugates of β if and only if the number of negative bands in W is nb[β].
Proof. Parallel argument in the proof of Lemma 8.4 applies.
The next lemma claims that super summit elements realize nb[β] for 3-
and 4-braids.
Lemma 9.3. Let β be an n-braid where n = 3 or 4. For every super

summit element β 0 ∈ SSS(β), we have nb[β] = nb(β 0 ). Moreover, a reduced
word Red(LCF(β 0 )) contains nb(β 0 ) negative bands.
Proof. Let β 0 be a super summit element in SSS(β). By Theorem 9.1 and

Lemma 9.2, a reduced word Red(LCF(β 0 )) contains nb[β] negative bands.
Since Red(LCF(β 0 )) = β 0 as braids, Corollary 8.5 implies that Red(LCF(β 0 ))
contains nb(β 0 ) negative bands.
Therefore, nb[β] = nb(β 0 ).
The next theorem shows that the converse direction holds for Corollary 6.6
in the strictly almost strongly quasipositive case if we add an additional
condition.
Theorem 9.4. A braid β ∈ Bn with n ≤ 4 (see Remark 9.5 below) is
conjugate to a strictly almost strongly quasipositive braid if and only if every
element β 0 ∈ SSS(β) has inf(β 0 ) = −1 and LCF(β 0 ) contains a canonical
factor of word length n − 2.
Remark 9.5. In the proof below, we note that the restriction on the braid
index n = 3, 4 is only required for the only-if (⇒) direction in which we use
Lemma 9.3 that is only proved for n = 3, 4 at this writing. The statement
of the if-direction (⇐) holds for general n.
Proof. (⇒) Suppose that β is conjugate to a strictly almost strongly quasi-

positive braid. By Corollary 6.6 every super summit element β 0 ∈ SSS(β)
has inf(β 0 ) = −1. Denote LCF(β 0 ) = δ −1 A1 · · · Ak . Suppose that Ai achieves
the maximal word length among A1 , · · · , Ak . Let Bi ∈ CnFct(B4 ) \ {e, δ} be
the canonical factor satisfying Ai Bi = Ai Bi = δ whose existence is guar-
anteed by Corollary 3.19. We apply the reduction operation (Definition 8.1)
with respect to the Ai . We get
Red(LCF(β 0 )) = δ −1+1 τ (A1 ) · · · τ (Ai−1 )(Bi )−1 Ai+1 · · · Ak (9.1)
−1
= τ (A1 ) · · · τ (Ai−1 )(Bi ) Ai+1 · · · Ak .
By Lemma 8.3, Red(LCF(β 0 ))gives a shortest word representing β 0 . By
Lemma 9.3, the word Red(LCF(β 0 )) realizes the number nb(β 0 ) = nb[β].
Since β is conjugate to a strictly ASQP braid, nb[β] = 1. The negative
exponent term (Bi )−1 in the reduced word Red(LCF(β 0 )) has word length
||(Bi )−1 || = ||Bi || = 1 and contributes the unique negative band to the
ASQP braid.
The diagram Bi of the band generator Bi is an edge connecting vertices of
Dn . If Bi joins consecutive vertices then the diagram Ai has one connected
component and Ai is an n − 1 gon. Otherwise, Ai has two components, a
k-gon and an n − k-gon for some k joined by the edge Bi . For both cases
the word length of Ai is ||Ai || = n − 2 (See Figure 14).
Figure 14. On the left figure, Ai = a3 a2 has one

connected component. On the right figure, Ai = a1 a3 has
two components. In both cases, the word length of Ai is
||Ai || = n − 2 = 2.
(⇐) Suppose that β ∈ Bn and that every element β 0 ∈ SSS(β) has

inf(β 0 ) = −1 and LCF(β 0 ) = δ −1 A1 · · · Ak contains some factor Ai ∈ CnFct(Bn )
of word length ||Ai || = n−2 for some i. By Corollary 3.19 there is a canonical
factor Bi ∈ CnFct(Bn ) \ {e, δ} such that Ai Bi = Ai Bi = δ.
Claim 9.6. We claim that ||Bi || = 1.
proof of the claim. (Case 1) If the diagram Ai consists of a single connected

component, then it is an (n − 1)-gon. Since Ai Bi = δ, the factor Bi is a
single band generator. Thus ||Bi || = 1.
(Case 2) If Ai has two connected components, say an x-gon X and a y-gon
Y (i.e., Ai = X t Y ), then
n − 2 = ||Ai || = ||X|| + ||Y || = (x − 1) + (y − 1).
We get x + y = n. By Proposition 3.14 we know that X ∗ Y is an (x + y)-gon.
On the other hand, δ is an n-gon. The only possibility is that the diagram
Bi is the facing edge of X and Y so that δ = Ai Bi = X (Y Bi ). Thus
||Bi || = 1.
(Case 3) Assume that the diagram Ai consists of three disjoint polygons
X , Y, Z. Suppose that X has x sides, Y has y sides and Z has z sides. Since
||Ai || = n − 2, by the disjointness of X , Y, Z, we have n − 2 = (x − 1) + (y −
1) + (z − 1). We obtain x + y + z = n + 1. We may assume that X and Y
are facing to each other (cf Definition 3.6) so that we can define X ∗ Y .
Apply Corollary 3.16 for the pair A = X t Y t Z and δ. Then we have
A ≺ (X ∗ Y ) ∗ Z ≺ δ. By Proposition 3.14 we know the diagram X ∗ Y is an
(x + y)-gon, and (X ∗ Y) ∗ Z is an (x + y + z)-gon. Since x + y + z = n + 1 and
δ is an n-gon, it is impossible that cvh((X ∗ Y) ∗ Z) ⊂ cvh(δ). Therefore,
Ai cannot have more than two components.
We continue the proof of the theorem. We have

Red(LCF(β 0 )) = τ (A1 ) · · · τ (Ai−1 )(Bi )−1 Ai+1 · · · Ak .
Since ||Bi || = 1 this shows that nb[β] ≤ 1. If nb[β] = 0 then β is conjugate
to a SQP braid. By Corollary 6.4 we obtain inf(β 0 ) = inf[β] ≥ 0. This
contradicts our assumption that inf(β 0 ) = −1. Thus, nb[β] = 1. Therefore,
β is conjugate to a strictly almost strongly quasipositive braid.
Recall the definition of inf[β] and sup[β] in Definition 5.7 and that they are
achieved by a super summit element simultaneously. The next two theorems
follow from Proposition 5.9, Theorem 8.7 and Lemma 9.3.
Theorem 9.7. Let β be a 3-braid with inf[β] < 0. Then
nb[β] = | inf[β]| − min{0, sup[β]}.
Theorem 9.8. Let K be a knot or link in S 3 of braid index n. Let β ∈ Bn
be a braid representative of K. The following holds:
• If inf[β] ≥ 0 then D(K) = nb(K) = 0.
• If inf[β] < 0 then
| inf[β]| ≤ nb[β] ≤ (n − 2)| inf[β]| − min{0, sup[β]}.
10. Fractional Dehn Twist Coefficient
In this section we apply the dual Garside structure to compute the frac-
tional Dehn twist coefficient (denoted FDTC) of a braid.
The FDTC is a Q-valued map c : MCG(S) → Q from the mapping class
group of a surface S. See Honda, Kazez and Matić’s paper [14] for the
definition. Intuitively, this invariant quantifies how much twisting a diffeo-
morphism φ : S → S possesses near a boundary component.
We will review some of the important properties of the FDTC.
Proposition 10.1. [14], [12] Let C be a boundary component of S and
φ ∈ Aut(S, ∂S). We have:
• c(φn , C) = n · c(φ, C).
• c(TC , C) = 1 and c(φ ◦ TC , C) = c(TC ◦ φ, C) = 1 + c(φ, C).
• c(φ, C) = c(ψ ◦ φ ◦ ψ −1 , C) for any ψ ∈ Aut(S, ∂S).
To formalize the concept of twisting with respect to a boundary com-

ponent, it is necessary to compare two arcs γ and η that share the same
starting point p ∈ C ⊂ ∂S with respect to the boundary component C. In-
formally, we want to define an order on arcs such that γ ≤ δ if δ veers to the
right of γ when the chosen representatives realize the geometric intersection
number.
Definition 10.2. [14] Let γ, η ⊂ S be two distinct oriented arcs that start
on the same point p ∈ C ⊂ ∂S. Isotope γ and η such that they minimally
intersect transversely. Consider the tangent vectors of the arcs γ̇(0) and
η̇(0). Then η is to the right of γ, denoted γ ≤ η, if the oriented basis
< η̇(0), γ̇(0) > agrees with the orientation on S. Equivalently, we pass to
the universal cover S̃. Since we have isotoped γ, η to minimally intersect in
the universal cover, these will only intersect at p̃. Then η is to the right of
γ if the interior of η̃ is in the region to the right of γ̃.
The following propositions are crucial when it comes to computing the

FDTC.
Lemma 10.3. [12] Let C ⊂ ∂S be a boundary component of S and φ ∈
Aut(S, ∂S). If there exists an essential arc γ ⊂ S that starts on C such that
T m (γ) ≤ φN (γ) ≤ T M (γ)
for some m, N, M ∈ Z then m/N ≤ c(φ, C) ≤ M/N.
Theorem 10.4. [12], [14] The FDTC of φ ∈ Aut(S, ∂S) defines a homoge-
nous quasi-morphism of defect 1. That is for all φ, η ∈ Aut(S, ∂S)
|c(φ ◦ η, C) − c(φ, C) − c(η, C)| ≤ 1
and
c(φn , C) = n · c(φ, C).
∼ Bn we can view a braid as a mapping class of the n-
Since MCG(Dn ) =
punctured disk Dn . Our goal is to find bounds on the fractional Dehn twist
of a braid β ∈ Bn using the LCF(β). Our surface is the n-punctured disk
Dn such that there is only one boundary component. As a result, we will
denote the fractional Dehn twist of a braid β by c(β).
Lemma 10.5. The FDTC satisfies the following:
(1) Let β ∈ Bn . If there exists an essential arc γ such that β(γ) and γ
are isotopic then c(β) = 0.
(2) c(A) = 0 for all A ∈ CnFct(Bn ) \ {δ = σn−1 · · · σ2 σ1 }.
(3) c(β1 β2 ) = c(β2 β1 ) for all β1 , β2 ∈ Bn .
(4) c(δ) = 1/n.
(5) c(( )( )) = 1/3.
Proof. (1) Notice that if φ(α) ∼

= α then α = TC0 (α) ≤ φ(α) ≤ TC0 (α) = α.
By Lemma 10.3 we see that
0 ≤ c(φ) ≤ 0
giving us the result.
(2) This follows as a direct application of part (1); for all A ∈ CnFct(Bn )\
{δ}, there exists an essential arc from the boundary to the boundary that
is preserved by the homeomorphism β ∈ CnFct(Bn ). In fact, A is reducible

type in Thurston’s classification.
(3) By the conjugacy invariant property (Proposition 10.1), we see that
for any β1 , β2 ∈ Bn
c(β1 β2 ) = c(β1−1 (β1 β2 )β1 ) = c(β2 β1 ).
(4) Notice that δ n = (σn−1 σn ...σ1 )n = ∆2 = TC . It follows that

1 1
c(δ) = c(TC ) = .
n n
(5) Using the property that A( )=( )τ (A), where τ (A) is the rota-
tion of A ∈ CnFct(B4 ) by π/2 as discussed in Example 4.8, we have
c((( )( ))3 ) = c(( )( )( )( )( )( )) (10.1)
= c(( )3 ( )( )( )) (10.2)
= c(( )4 ) = 1.
Therefore, c(( )( )) = 1/3.
Using Lemma 10.5 (alongside a train track idea for computing FDTC) we
can compute the FDTC of a braid β. As an instructional example, we will
compute the FDTC of a product of two canonical factors A, B ∈ CnFct(B4 ).
Proposition 10.6. c(( )( )) = 1/2
Proof. See Figure 15. Consider the arc α ⊂ D4 in the 4-punctured disk that
goes from the boundary back to the boundary while encapsulating one of
the punctures (the bottom left puncture). Similar to the concept of train
tracks, we will do a zip move isotoping the endpoints of the arc to one point
p ∈ ∂D2 as in the right of Figure 16.
After applying the braid β = ( )( ) twice, one sees that β 2 (α) wraps
around the bottom left puncture P1 three times whereas it wraps around the
bottom right puncture P2 four times. Applying a zip move on the bottom
face of the square made by the punctures, one gets a train-track of β 2 (α)
with labels 3 and 4 respectively. Further, note that β 2 (α) ≤ TC (α); it follows
that c(β) ≤ 1/2 by Lemma 10.3.
After applying ( ), the resulting train track will be as in Figure 16.

Notice that the resulting arc is to the right of TC (α), the boundary Dehn
twist on α.
Figure 15. This is the zip process the arcs α and β 2 (α)
into a train-track
Figure 16. On the top is the train track after applying

the canonical factor b1 to β 2 (α); on the bottom is the train
track for β 3 (α).
Similarly, applying ( ) will increase the weights while passing over the
second puncture after a first Dehn twist. It follows that TC (α) ≤ β 3 (α) and
hence 1/3 ≤ c(β). Using these train tracks, it is easy to see that for n ∈ N
TCn+1 (α) ≤ β 2n+3 (α)
n+1
⇒ c(( )( )) ≥ lim = 1/2.
n→∞ 2n + 3
Combining the two inequalities, we get c(( )( )) = 1/2.
The following table is a list of the FDTC of the product AB of canonical

factors A, B ∈ CnFct(B4 ).
Products FDTC Products FDTC Products FDTC

( )( ) 0 ( )( ) 0 ( )( ) 0
( )( ) 0 ( )( ) 0 ( )( ) 0
( )( ) 0 ( )( ) 1/4 ( )( ) 1/4
( )( ) 1/4 ( )( ) 1/4 ( )( ) 1/3
( )( ) 1/3 ( )( ) 1/3 ( )( ) 1/3
( )( ) 3/8 ( )( ) 3/8 ( )( ) 1/2
( )( ) 1/2 ( )( ) 1/2 ( )( ) 1/2
Table 3. The FDTC for AB where A, B ∈ CnFct(B4 ).
Suppose LCF(β) = δ r A1 · · · Ak . Recall the invariants inf(β) = r and

sup(β) = r + k.
Proposition 10.7. For a braid β we have
inf[β] sup[β]
≤ c(β) ≤ .
n n
Proof. Notice that c(δ) = 1/n by Lemma 10.5; it then follows by Lemma 10.3
that
inf(β) r r+k sup(β)
= = c(δ r ) ≤ c(δ r A1 · · · Ak ) ≤ c(δ r+k ) = = .
n n n n
Since c(β), inf[β], and sup[β] are conjugacy invariants (see Def 5.7) the
statement follows.
It is proved by Malyutin [19] for braids that the Dehornoy floor bβcD and
the FDTC are numerically close to each other; bβcD ≤ c(β) ≤ bβcD + 1.
Since sup(β) ≤ ||β||, the minimal word length in band generators, our bound
improves current bounds in the literature such as the following bound by
Ito.
Proposition 10.8. [11]. If an n-braid β is conjugate to a braid β 0 repre-
sented by a product of m band generators then
m+1
bβcD < .
n
References
1. E. Artin, Theorie der zopfe, Hamburg Abh. 4 (1925), 47–72.
2. D. Bennequin, Entrelacements et equations de Pfaff, Asterisque 107/108 (1983).
3. J. Birman, K. H. Ko, and S. J. Lee, A new approach to the word and conjugacy
problems in the braid groups, Adv. Math 139 (1998), 322–353.
4. M. Calvez and B. Wiest, A fast solution to the conjugacy problem in the four-strand
braid group, J. Group Theory 17 (2014), no. 5, 757–780.
5. P. Dehornoy, I. Dynnikov, D. Rolfsen, and B. Wiest, Ordering braids, vol. 148, AMS,
Providence, 2008.
6. I. A. Dynnikov and M. V. Prasolov, Bypasses for rectangular diagrams: A proof of the
jones conjecture and related questions, Trans. Moscow Math. Soc. 74 (2013), 97–144.
7. E. A. Elrifai and H. R. Morton, Algorithms for positive braids, Quart. J. Math. Oxford
45 (1994), no. 2, 479–497.
8. J. Franks and R.F. Williams, Braids and the jones polynomial, Trans. Amer. Math.
Soc. 303 (1987), 97–108.
9. F. A. Garside, The braid group and other groups, Quart. J. Math. Oxford 20 (1969),
no. 78, 235–254.
10. J. Hammer, T. Ito, and K. Kawamuro, Positivities of knots and links and the defect
of bennequin inequality, Exp. Math. (2022), no. 1, 199–255.
11. T. Ito, Braid ordering and the geometry of closed braids, Geometry and Topology 15
(2011), 473–498.
12. T. Ito and K. Kawamuro, Essential open book foliations and fractional dehn twist
coefficient, Geom. Dedicata 187 (2017), 17–67.
13. , The defect of the bennequin-eliashberg inequality and bennequin surfaces, In-
diana University Mathematics Journal 68 (2019), no. 3, 799–833.
14. W. Kazez K. Honda and G. Matic, Right-veering diffeomorphisms of compact surfaces
with boundary, Inventiones Mathematicae 169 (2007), 427–449.
15. E. S. Kang, K. H. Ko, and S. J. Lee, Band-generator presentation for the 4-braid
group, Topology Appl 78 (1997), 39–60.
16. K. Kawamuro, The algebraic crossing number and the braid index of knots and links,
Algebr. Geom. Topol. 6 (2006), 2313–2350.
17. K. H. Ko and S. J. Lee, Genera of some closed 4-braids, Topology Appl 78 (1997),
61–77.
18. D. LaFountain and W. Menasco, Embedded annuli and jones’ conjecture, Algebr.
Geom. Topol. 14 (2014), 3589–3601.
19. A.V. Malyutin, Writhe of (closed) braids (russian, with russian summary), Algebra i
Analiz 16 (2004), no. 5, 59–91, translation in St. Petersburg Math. J. 16(2005), no.5,
791-813.
20. H. R. Morton, Seifert circles and knot polynomials, Math. Proc. Cambridge Philos.
Soc. 99 (1986), 107–109.
21. P. Osváth and Z. Szabó, Knot floer homology and the four-ball genus, Geom. Topol.
7 (2003), 615–639.
22. J.A. Rasmussen, Khovanov homology and the slice genus, Invent. Math. 182(2)
(2010), 419–447.
23. D. Rolfsen, Knots and links, AMS Chelsea Publishing, Providence, 2003.
24. D. B. A. Epstein (with Cannon, Holt, Levey, Patterson, and Thurston), Word pro-
cessing in groups, Jones and Barlett, Boston, MA, 1992.
25. P. J. Xu, The genus of closed 3-braids, J. Knot Theory Ramifications 1 (1992), no. 3,
303–326.
Department of Mathematics, University of Iowa, Iowa City, IA 52242

Email address: michele-capovilla-searle@uiowa.edu
Email address: keiko-kawamuro@uiowa.edu
Email address: rebecca-sorsen@uiowa.edu

Michele Capovilla-Searle, Keiko Kawamuro, and Rebecca Sorsen

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Michele Capovilla-Searle, Keiko Kawamuro, and Rebecca Sorsen

Uploaded by

Copyright:

Available Formats

BIRMAN-KO-LEE LEFT CANONICAL FORM AND ITS

MICHELE CAPOVILLA-SEARLE, KEIKO KAWAMURO, AND REBECCA SORSEN

The braid groups Bn for n = 1, 2, 3, . . . were originally introduced by

In the standard presentation of Bn , Artin generators only swap adjacent

The infimum of a braid β, denoted by inf(β), is an invariant of the braid

In § 8, we review the reduction operation introduced by Kang, Ko, and

β 0 ∈ SSS(β) has inf(β 0 ) = −1 and LCF(β 0 ) contains a canonical factor of

Acknowledgements. The authors would like to thank Juan Gonzalez-

2. Canonical Factors via Diagram

2.1. Band Generators. In the traditional presentation of Bn , Artin gen-

The set of band generators of Bn is denoted by BGen(Bn ). Band generators

In B4 , following the notations in [15], the band generators are renamed

Figure 1. Band generators for B4

a1 , . . . , a4 , b1 , b2 are positive bands and their inverses a−1 −1 −1 −1

2.2. Non-Crossing Partition Diagrams. In this subsection, we review

Note that we placed the punctures in a counterclockwise direction, while in

Similarly, we define an n punctured disk Dn . Punctures are labeled

With the new notation we have:

Remark 2.3. In the following, we will define 3-gons, 4-gons, . . . so that an

Definition 2.5. (Square, 4 gon) Consider a square with vertices Pi , Pj , Pk , P` ∈

This square coincides with the fundamental element δ := a3 a2 a1 of B4 . It is

We can generalize the above construction to a k-gon.

Definition 2.6. (k-gon) Let 2 ≤ k ≤ n. A k-gon in Dn , whose vertices

By the disjointedness condition, each of these diagrams admits two factor-

2.3. Canonical Factors. In this section, we define canonical factors that

For example, there are 14 canonical factors in B4 .

3. Partial Ordering on CnFct(Bn )

Definition 3.1. The set of convex hulls of a canonical factor A, denoted

Similar to the partial ordering ≤ on Bn (Definition 2.8), we define a partial

We note that A ≺ B implies A ≤ B.

The theorem will be proved in Propositions 3.18 and 5.4.

3.1. Operations  and ∗ on CnFct(Bn ). To prove Theorem 3.4 we will

The definition of the -operation is divided into 4 steps.

Figure 2. Hasse Diagram for the partially ordered set

Figure 3. On the left, A = X t Y and X and Y are facing

Definition 3.7. (Step 1. See Figure 4) Let A, B ∈ CnFct(Bn ) be canonical

a0 clockwise, a1 , . . . , ak−1 . Label the edges of B starting from b0 clockwise,

Figure 4. An example of Definition 3.7, A  B. The left

We will extend the operation  to less restrictive pairs of canonical factors.

We extend the operation  to the pair A and B by

Figure 5. An example of Definition 3.8. We have

We remark that  is well defined since for every permutation σ ∈ Sk of k

For Step 2, with regard to (b) the diagram A  B is a single polygon.

Figure 6. An example of Definition 3.10. We have

Definition 3.11. (Step 4) Let A, B ∈ CnFct(Bn ) such that

Next we define the operation ∗.

band generator, denoted by C ∈ BGen(Bn ) and can be thought as a bigon

Figure 7. An example of the operation A ∗ B. The left

We list properties of the operation ∗ immediately follow from the defini-

3.2. Proof of Theorem 3.4.

Proof. The diagram A contains a disjoint pair of connected components

Figure 8. The above figure is an example of Lemma ??.

Corollary 3.16. (See Figures 9 & 10) Let A, B ∈ CnFct(Bn ) be canonical

• its diagram A0 is a single polygon, and

Figure 9. An example of Corollary 3.16. The left image

Figure 10. Another example of Corollary 3.16. The left

Proof. Applying Lemma 3.15 k − 1 times, we inductively obtain band gen-

By the construction in Lemma 3.15 we see

We note that the interior disjointness property and the counterclockwise

Proof. If A is empty, then A = e and we set C := B, done. In the following

Figure 11. The above figure is an example of Lemma

• its diagram A  C1 is connected,

3.1. Operations and ∗ on CnFct(Bn ). To prove Theorem 3.4 we will

The definition of the -operation is divided into 4 steps.

Figure 4. An example of Definition 3.7, A B. The left

We will extend the operation to less restrictive pairs of canonical factors.

We extend the operation to the pair A and B by

We remark that is well defined since for every permutation σ ∈ Sk of k

For Step 2, with regard to (b) the diagram A B is a single polygon.

• its diagram A C1 is connected,

cvh(C) ⊂ cvh(A C) ⊂ cvh(B), (3.2)