The purpose of this section is to prove Theorem 1.1.
3.1. Reduction to almost braid closures
Given a braid word for a braid , we can get a 1-1-tangle
diagram by taking the braid closure of the rightmost strands. We say that such 1-1-tangle
diagrams are in almost braid closure form. From a 1-1-tangle diagram , one can
obtain link diagrams and by taking either the left- or right-handed closure of the single
open strand. These diagrams are illustrated at the top left and top right of (3.1)
respectively.
We note the following straightforward extension of the Alexander theorem.
Proposition 3.2.If the sweep-around map is homotopic to the identity for 1-1-tangles in almost braid
closure form, then the same is true for all 1-1-tangle diagrams.
Proof.Consider an isotopy that brings the tangle diagram into almost braid closure form
and denote its image under the Khovanov invariant as . Furthermore, let the maps
associated to the sweep-around for and be denoted by and
respectively. Now, note that because the
underlying link cobordisms are isotopic in .
By assumption and thus also .
∎
3.2. The game plan
Fix an almost closure of a braid word for . We call the
right-hand closure and the left-hand closure . We consider the following movies of
intermediate diagrams and their associated chain maps between Khovanov–Rozansky complexes.
(3.1)
In the first row, the signs indicate the two versions of this movie, in
which the horizontal strand passes in front of () or behind () . We
denote the composition along the top by and the composition along the bottom
by . In either case we first see a Reidemeister I move (denoted by ),
then a composite of Reidemeister II moves (denoted by ), a number of
Reidemeister III moves (each denoted by ), a composite of inverse
Reidemeister II moves (), and finally an inverse Reidemeister I
move (). Our goal is to show that, after making careful use of
the freedom, described later, to choose up-to-homotopy representatives of the
chain maps for Reidemeister III moves, we have the following:
Theorem 3.3.For every almost braid closure diagram , the front sweep and the back sweep chain
maps constructed above are identical (not just merely homotopic).
Together with Proposition 3.2, this will imply Theorem 1.1.
The proof of Theorem 3.3 will occupy the rest of this section.
We distinguish two types of crossings in the intermediate diagrams . The crossings of
the moving, horizontal, strand with everything else will be called external. The remaining
crossings were already present in and will be called internal.
Definition 3.5. The homological grading on splits into the sum of the internal and
external homological gradings, contributed by resolutions of internal and external crossings
respectively. The internal and external homological degrees of a web appearing in
will be denoted by and respectively.
The braid word determines an ordering of the crossings in , , and , namely
from top to bottom. This ordering also induces an ordering of the internal crossings in all
other diagrams in (3.1). The diagrams and have one additional
external crossing. The diagrams for all have external
crossings, which are ordered from right to left. We will classify webs in each of these
complexes according to the resolutions that appear at the crossings. For the following, let
denote the number of crossings in , , and .
Definition 3.6. The type of a web in any of the complexes in (3.1) is the element that records in the -th coordinate whether the -th internal crossing in the
respective link diagram is resolved in a parallel way (p), or using the thick edge (t).
The offset of a web in any of the complexes is the element that records the resolution of the leftmost external crossing.
The state of a web in any of the complexes for is
the element , which records the resolutions of the rightmost external
crossings (that is, all except the leftmost external crossing). Such a web is said to be
palindromic if is a palindrome.
Remark.
The webs in the complexes and are indexed by their types . The webs
in , and are indexed by the pairs . The
webs in the complexes for are indexed by the triples
.
Definition 3.7. If , , , and
, we will use or to denote the
web in with indexing data or , as appropriate.
Analogously, we write and for -indexed webs in and
respectively. If the indexing data is fixed, we will sometimes omit it from the notation (e.g.
and ) and say that the webs and
correspond to each other.
If is a chain map and and are webs in the source and target complexes, then we write
for the component of from to .
Remark.
Suppose , and . For we have
as webs, and for we have
as webs. Moreover, .
3.3. Reidemeister I moves
The Reidemeister I chain maps are the following.
dcapcupcapdcupdcapcupcapdcup
Here and simply denote the cap and cup foams,
while and denote decorated cap and cup foams.
The decoration is by the polynomial where denotes
the dot on the strand and the dot on the circle; see (2.2).
We have only assigned notation and to those
Reidemeister I chain maps that are relevant for the sweep-around move.
Lemma 3.8.The Reidemeister I chain maps and preserve the internal and external homological degrees individually.
Moreover, their only non-zero components are in external homological grading zero.
Proof.The chain maps and each consist of identity foams decorated by the polynomial .
In , the dots and are placed next to the Reidemeister I crossing, as shown in the
first picture on the right. These dots are spatially separated from the region in which the
and moves are taking place, so we can slide them spatially lower in the diagram, and
timewise past all the and moves. At that point, shown in the second diagram on the
right, the dots are in exactly the positions to give .
∎
3.4. Reidemeister II moves
We will use Elias–Khovanov’s Soergel calculus [EK10a] to describe the
chain maps associated to Reidemeister II and III moves. The Soergel calculus of
type is a graphical incarnation of the 2-category of Soergel
bimodules, which categorifies the Hecke algebra for . For any , it
admits a 2-functor to the monoidal subcategory of of webs and foams with
boundary components with suitable orientations, see e.g. [MV10].
Instead of describing these -functors formally, we will just use the Soergel
calculus as shorthand notation for foams using the following dictionary:
•
In the calculus, we have only a blue object, which we will interpret as the two strand web
•
In the calculus, we have red and blue objects, interpreted as three strand webs
•
Start dots and end dots (in any color)
correspond to zip and unzip foams.
•
The trivalent vertices and correspond to
digon creation and annihilation foams respectively.
We also use cups
and caps .
•
The 6-valent vertex corresponds to the foam shown in Figure 1.
The Reidemeister II chain maps are the following.
(3.2)
In both cases we have chosen to order the crossings from the top to the bottom.
Now we can record two observations concerning the composite (inverse)
Reidemeister II chain maps and .
Lemma 3.10.The chain maps and preserve the internal and external homological gradings
individually and their only non-zero components involve palindromic resolutions.
Lemma 3.11.Let and be pairs of corresponding webs
in and respectively. Further, let be a
palindrome in which appears times, and consider and
in and respectively.
Then
Proof.In a single Reidemeister II move, the identity resolution is always sent to the identity resolution
via the identity. The maps involving the resolution with two thick edges are negatives of each
other, when comparing the two types of Reidemeister II moves with fixed order of crossings as in
(3.2).
∎
3.5. Reidemeister III moves
In (3.1) we encounter four types of Reidemeister III moves. Namely, the moving
strand can pass in front of or behind a positive or a negative crossing. In the following we show
the front and back versions alongside each other. In every case, the moving strand is the one
connecting the bottom left and top right boundary points.
In each variant of Reidemeister III, we order the crossings in each tangle from top to bottom. The
parts of the complexes with internal homological degree zero—where the internal crossing is
resolved in the parallel fashion—are highlighted in blue. The parts with internal homological
degree are highlighted in yellow.
There is a 2-dimensional space of chain maps between the two sides of each Reidemeister
III move [EK10b]. There is a 1-dimensional affine subspace of these chain maps which, given the previous
choices for Reidemeister I and II maps, provides a functorial link invariant, by Theorem
2.4. (Note that their proof does not rely on any particular choice of chain
maps from this subspace; any will do!) This subspace is characterised by the condition that the
component of the chain map between parallel resolutions is the identity (this condition corresponds
to the appearance of a blue highlighted in each chain map below). In the diagrams below, we
parametrise this subspace by a variable ; shortly we shall specialize to .
All choices of chain map in this affine subspace are homotopic, so for many purposes this
structure can be ignored. For the present proof, however, it is quite important that we make
the most convenient choice of up-to-homotopy representative.
When the moving strand passes a positive crossing we have:
(3.3)
0
Next, we consider the two ways in which the moving strand may pass a negative crossing:
(3.4)
For the remainder of this paper we specialise to the choice . (Note in particular that the
statements immediately below are not true for other choices!)
Proof.Since chain maps are of homological degree zero, the statement is equivalent to saying that the
Reidemeister III chain maps in (3.1) never increase the internal
homological grading. This can be verified by inspecting (3.3) and (3.4). For
the reader’s convenience we have highlighted the components of negative internal homological degree
in green. All other non-zero components are highlighted blue or yellow and have internal homological
degree zero because they map between the yellow and blue layers of the relevant complexes. Thus we
only need to worry about components of the chain map which are not highlighted in the
diagrams above. With , these components all vanish.
∎
In other words, the Reidemeister III maps are filtered with respect to the filtration determined by
the internal homological degree, which we shall call the internal filtration.
Proof.By inspecting (3.3) and (3.4) —
for each of the 1+9+9+1 components of the chain map, check that the corresponding component of the chain map is the same (recalling ).
∎
Corollary 3.14.The filtration-preserving component of the chain maps
agree. More precisely, we have
for pairs of corresponding webs in
and in with
.
Remark.
The Reidemeister III chain maps shown in (3.3) and (3.4), their inverses,
and four additional variations were studied by Elias–Krasner [EK10b]. Note, however, the
following differences in conventions. Their positive crossings are our negative crossings and the
crossings in their braids are ordered from bottom to top, while we order them from top to bottom.
Finally, they read Soergel diagrams from left to right, while we read them from right to left.
Proof of Theorem 3.3.We need to show that the two chain maps and from (3.1) are equal. For
this, let and be webs in and respectively. We shall compare the
components of and between and .
By Proposition 3.13, the maps do not decrease the external homological degree,
but by Lemmas 3.8 and 3.10, the and maps
preserve the external homological degree. Since , the increasing components of
do not contribute to or . Now suppose that are corresponding
webs in and are corresponding webs in with
. Then, by Corollary 3.14,
Let us also record that if
has a non-zero component between two webs and , then first digits of
agree with the first digits of . (Recall that the first digits describe the
rightmost crossings, which are spatially separated from the region in which Reidemeister III
moves occur.)
Next we consider the pair of corresponding webs in
, which appear in the image of under , and the pair
of corresponding webs in ,
which have as image under . The components of
between
these webs are sums over components through many possible intermediate webs
and . By the previous argument, the Reidemeister III
portions of the - and the -version of the map agree. By
Lemma 3.11, the Reidemeister II portions could at most cause a
sign-discrepancy. However, since the first digits of all agree, and since Reidemeister II
chain maps are zero on non-palindromic webs by Lemma 3.10, there is no
sign-discrepancy. Thus, we record: