Diagrammatics for Coxeter groups and their braid groups
aa r X i v : . [ m a t h . R T ] J a n DIAGRAMMATICS FOR COXETER GROUPS AND THEIR BRAID GROUPS
BEN ELIAS AND GEORDIE WILLIAMSONA
BSTRACT . We give a monoidal presentation of Coxeter and braid 2-groups, in termsof decorated planar graphs. This presentation extends the Coxeter presentation. Wededuce a simple criterion for a Coxeter group or braid group to act on a category.
1. I
NTRODUCTION AND PRELIMINARIES
Strictifying group actions.
A group acting on a category is, roughly speaking,an assignment of a functor for each element of the group. This can be thought ofas a categorification of the usual notion of a group representation. The recent para-digm in categorification dictates that studying the (isomorphism classes of) functorswhich appear is not sufficient; one should study the natural transformations betweenthem. This philosophy can be found, for instance, in the seminal paper of Chuangand Rouquier [7], which deduced strong structural results for categorified sl repre-sentations given the existence of a certain algebra of natural transformations.The nuance in the work of Chuang and Rouquier was specifying an interestingalgebra of natural transformations between functors. For groups, the nuance comesfrom the opposite goal: showing that the algebra of natural transformations betweenfunctors corresponding to the same element of the group can be trivialized. Thedesired structure is a strict action of a group on a category, where each element ofthe group is (compatibly) assigned a canonical functor (see Definition 1.3). This isto be contrasted with a weak action, where each element of the group is assignedan isomorphism class of functor. Given two words in the group which multiply tothe same element, a weak action guarantees that the corresponding compositions offunctors are isomorphic, while a strict action fixes a natural transformation whichrealizes this isomorphism.Here we pause to distinguish between the two most common descriptions of groupsand their representations, which we call the holistic and the combinatorial. In theholistic approach, the action of each element is given in a general way. An exam-ple is the standard representation of GL ( n ) (or its exterior and symmetric powers),where each matrix g ∈ GL ( n ) acts via a general formula. Another example is theaction of an automorphism group of a variety acting on the cohomology ring, bypullback. In the combinatorial approach, one describes a group combinatorially bygenerators and relations. A representation can be defined by giving an endomor-phism for each generator, and checking the relations. This can save a great deal oflabor, replacing the computation of the entire multiplication table with a manageable amount of data. Representations of Coxeter groups and their Artin braid groups areoften defined in this fashion.These two approaches are also common when defining actions of groups on cat-egories. The holistic approach lends itself easily to the notion of a strict action. Forexample, the automorphism group of a variety acts on the derived category of coher-ent sheaves by pullback, and the composition of pullbacks is naturally isomorphic tothe pullback of the composition. On the other hand, given a group presentation, onecould define a functor for each generator, and check an isomorphism of functors foreach relation, but this would only define a weak action. If one works directly fromthe definition of a strict group action, the additional data required to make this actionstrict is not made any simpler by the presentation: one needs a natural transforma-tion for each entry of the multiplication table, satisfying a host of compatibilities.This “strictification” data is prohibitive to provide in practice, and is not in keepingwith the labor-saving combinatorial nature of the presentation.1.2. Braid groups and Coxeter groups.
The primary goal of this paper is to give anexplicit and efficient criterion for establishing a strict action of a Coxeter group orits braid group on a category, extending the Coxeter presentation of said group. Forinstance, to make an action of the type A braid group strict, one need only checka single equality: the so-called Zamolodzhikov relation . This improves slightly upona similar result of Deligne [9] and Digne-Michel [10] for braid groups. The litera-ture does not seem to contain any previous results on strictifying actions of Coxetergroups.Many topics in category theory can be more intuitively phrased using the lan-guage of topology, and strictifying a group action is a fine example. The equiva-lence between group presentations and 2-dimensional cell complexes (with a single0-cell) is well-known. Finding strictification data for this presentation is equivalentto finding a collection of 3-cells which kill π of this complex. Essentially, one issearching for a combinatorially-defined 3-skeletal approximation for the classifyingspace BG of the group. An equivalent question is to find an appropriate 3-skeletalapproximation for its universal cover EG . In this paper we discuss two separate cellcomplexes attached to a Coxeter system, one for the Coxeter group and the other forits associated braid group. Though the complexes are different, the correspondingstrictification criteria are closely related.To a Coxeter group W one can associate a real hyperplane arrangement, and canconsider the complement of these hyperplanes in the complexification Y W . The K ( π, -conjecture , originally due to Arnold, Brieskorn, Pham and Thom states that Y W should be a classifying space for the pure braid group, and thus a natural quo-tient Y W /W = Z W should be a classifying space for the braid group. This wasproven for finite Coxeter groups by Deligne [9]. The K ( π, -conjecture has also been Z W is denoted X W in [9] and [30]. We have reserved X W for another purpose. IAGRAMMATICS FOR COXETER GROUPS AND THEIR BRAID GROUPS 3 proven for a broad class of infinite Coxeter groups. We give some further introduc-tion in § Z W as a cell complex. In hisproof, Deligne introduces a realization involving reduced expressions for elementsin the Coxeter group. Correspondingly, in [8], Deligne provides an analogous crite-rion for a strict action of the positive braid monoid of a finite Coxeter group. Thiscriterion extends the positive lift presentation of the braid group, where the generatorsare positive lifts of each element in the Coxeter group. This result has been gen-eralized to arbitrary Coxeter groups by Digne and Michel [10], using the Garsidestructure on the braid group.In [29], Salvetti proved that an arbitrary hyperplane complement has the same ho-motopy type as a combinatorially-defined cell complex. In [30], Salvetti providedan analogous cell complex realization of the quotient Z W , which he used to reprovethe K ( π, -conjecture for finite Coxeter groups. Both cell complexes are called Sal-vetti complexes in the literature; in this paper, we reserve the term for the realization of Z W . Independently, Paris [26] also used Salvetti’s complexes for hyperplane comple-ments to reprove this result, introducing along the way a combinatorial constructionfor the (conjectural) universal cover of the Salvetti complex.The Salvetti complex differs from Deligne’s complex, in that the 2-skeleton corre-sponds to the Artin presentation of the braid group rather than the positive lift pre-sentation. Regardless of the validity of the K ( π, -conjecture, the results of Digneand Michel mentioned above imply that π ( Z W ) = 0 , and therefore the Salvetti com-plex gives a valid 3-skeletal approximation of the classifying space. Our strictifica-tion data for the Artin presentation is the extrapolation of the 3-cells in the Salvetticomplex.In similar fashion, assuming the K ( π, -conjecture, presumably one can use the k -skeleton of the Salvetti complex to concoct a strictification procedure for actions ofbraid groups on ( k − -categories. One can see this as a higher categorical gener-alization of the Coxeter presentation of a braid group; on the k -th categorical level,there is but a single relation for each finite rank k (standard) parabolic subgroup, a“higher Zamolodzhikov relation.” We do not pursue this any further in this paper.We are certainly not the first to observe this type of phenomenon. Strictificationdata for braid group has been studied by several authors, and the Zamolodzhikovrelation goes back at least to Deligne [8]. One can find the Zamolodzhikov relation intype A described using Igusa pictures in Loday [25, Figure 19]. A similar approachfor general braid groups using the language of coherent presentations was taken in[15].In contrast to the situation for braid groups, there do not seem to be cell complexrealizations of the classifying spaces of Coxeter groups in the literature. However,strictification data for Coxeter groups is essential in the authors’ work on Soergelbimodules [13]. We construct a new and somewhat unfamiliar cell complex as our BEN ELIAS AND GEORDIE WILLIAMSON EW . We study this model by relating it tothe dual Coxeter complex, in a way to be described in § π of a cell complex that we found in a book byFenn [14] (see Remark 2.1). Applying these techniques to the Salvetti complex and toour complex for EW , one obtains a depiction of elements of π as decorated planargraphs. This diagrammatic calculus is new for both braid groups (outside of type A )and Coxeter groups, and could potentially lead to new, diagrammatic proofs of ourmain result.In the remainder of this introductory chapter, we spell out the connection withtopology in more detail, in order to describe our results and discuss some of thetechniques we use. Then we state our results in § § Strict actions and 3-presentations.Definition 1.1.
For a group G , let Ω G be the monoidal category defined as follows.The objects consist of the set G , and the only morphisms are identity maps id g foreach g ∈ G . The monoidal structure on objects is given by the group structure on G ,and the monoidal structure on morphisms is uniquely determined. Definition 1.2.
Given a category C , let Aut( C ) denote the monoidal category whoseobjects are autoequivalences of C , and whose morphisms are invertible natural trans-formations. The monoidal structure is given by composition of functors. Definition 1.3. A strict group action of G on a category C is a monoidal functor Ω G → Aut( C ) . Remark . The usual definition of a strict group action involves providing a functor F g for each g ∈ G , isomorphisms a g,h : F g ◦ F h → F gh , and an isomorphism ǫ : F e → C for the identity element e ∈ G , satisfying some natural compatibilities, including anassociativity compatibility. The isomorphisms a g,h are the image of the unique mor-phism g ⊗ h → gh in Ω G , and the isomorphism ǫ is the fixed isomorphism betweenmonoidal identities given as part of the data of a monoidal functor.Suppose that G has a presentation P = ( S , R ) with generators S and relations R .As discussed above, it is common in the literature to define a weak group action bygiving an invertible functor F s for each s ∈ S , and checking an isomorphism foreach r ∈ R . Doing this is implicitly defining a monoidal functor from the monoidalcategory Ω P defined below. The inverse functor F − s is not usually given explicitly,but one can choose F − s to be any inverse to F s . Definition 1.5.
For a presentation P = ( S , R ) , let Ω P be the monoidal category de-fined as follows. Its objects are words in the letters S ∪ S − , with monoidal structuregiven by concatenation (we let denote the monoidal identity, the empty word). Itsmorphisms are monoidally generated by the following maps: IAGRAMMATICS FOR COXETER GROUPS AND THEIR BRAID GROUPS 5 • (Cups and Caps) Inverse isomorphisms ss − ⇄ for each s ∈ S , as well asinverse isomorphisms s − s ⇄ . • (Relation symbols) Inverse isomorphisms r ⇄ for each word r ∈ R .One imposes the following relations: • The generating “inverse isomorphisms” are actually inverse isomorphisms. • Cups and caps form the units and counits of adjunction between the biadjointfunctors s ⊗ ( · ) and s − ⊗ ( · ) . • The relation symbols are cyclic with respect to these biadjunction structures.
Remark . This remark is for the reader unfamiliar with biadjunction and cyclicity.For a broader introduction to biadjunction, cyclicity, and the associated diagram-matic notation, see [23, 24].Observe that any invertible functor F ∈ Aut( C ) is both left and right adjoint toits inverse functor F − . Adjunction is a structure, not a property, and a biadjunction is a choice of both a left and right adjunction between F and F − . Having choseninverse isomorphisms F s F − s ⇄ , there is a unique choice for the isomorphisms F − s F s ⇄ such that the isomorphisms also provide (the units and counits of) abiadjunction. Biadjunctions occur more generally than between inverse functors,but this will suffice for our purposes. Cyclicity is a property of a general morphism (i.e. natural transformation) betweencompositions of functors equipped with biadjunctions, stating that this morphism issomehow compatible with the right versus the left adjunction. An explicit statementof this compatibility can be found (in diagrammatic language) later in this paper.Identity morphisms are axiomatically cyclic, but general morphisms need not becyclic.In a monoidal category (like
Aut( C ) ), the biadjoint of an object, if it exists, is well-defined up to unique isomorphism, but nonetheless this assignment of a biadjoint toeach object need not be functorial. In a pivotal category, there exists a duality func-tor D sending each object to a biadjoint, and equipped with a natural isomorphism φ : → D . One can always adjust the functor D up to isomorphism to guaranteethat D and agree on objects, but it need not be the case that φ is the identity map.If φ is the identity map, the category is strictly pivotal . Equivalently, every morphismis cyclic, so the category is also called cyclic biadjoint . This happens frequently ingeometric and algebraic examples of group actions. It also happens in fundamental2-groupoids, the topological framework of this paper. Remark . Note that a functor Ω P →
Aut( C ) is not quite as general as a weak groupaction defined by generators and relations, because there need not exist relation sym-bols which are cyclic. Any weak action which can be extended to a strict action willcertainly satisfy cyclicity. Remark . What is the difference between a categorical action of a monoid wherethe generators happen to act by invertible functors, and a categorical action of its
BEN ELIAS AND GEORDIE WILLIAMSON associated group? From the definition of a weak categorical action, there is no dif-ference. Philosophically, however, one might desire some new condition which con-nects the new inversion structure with the existing relation isomorphisms. Said an-other way, one now has a host of new relations obtained by conjugating existingrelations, and one might expect these to be somehow mutually compatible. We be-lieve that cyclicity is precisely the correct structure one should impose.Every morphism in Ω P is an isomorphism. The isomorphism classes of objects in Ω P can be identified with G , and there is a monoidal functor Ω P → Ω G . However,endomorphism spaces in Ω P can be quite large, so this functor is not faithful. Be-cause of biadjunction, every endomorphism space is a principal space for the group End( ) . Definition 1.9.
Let Z be a chosen subset of End( ) within the category Ω( S , R ) , forsome group presentation ( S , R ) of G . We call P = ( S , R , Z ) a of G , andwe simply write ( S , R ) when Z is empty. We let Ω P denote the quotient of Ω( S , R ) by the relations z = id for each z ∈ Z .For any 3-presentation P of G , there is still a monoidal functor Ω P → Ω G . When Z generates the group End( ) ⊂ Ω( S , R ) , then morphism spaces in Ω P are trivial,consisting only of identity maps, and the functor to Ω G is an equivalence. We callsuch a 3-presentation acyclic . In the acyclic case, giving a monoidal functor Ω P →
Aut( C ) is equivalent to giving a strict action of G , but it has a different recipe: providea functor F s and its biadjoint inverse for each s ∈ S , provide an isomorphism and itsinverse for each r ∈ R , and check a relation for each z ∈ Z .This recipe need not be interesting or useful. We do not expect there to be ageneral method to extend an arbitrary -presentation of a group into an acyclic -presentation in a useful way. Example 1.10.
Every group has a universal presentation where S = G and R consistsof relations stating that g · h = ( gh ) . The corresponding monoidal category Ω( S , R ) has isomorphisms a g,h as in Remark 1.4, but no compatibility requirements. Letting Z be the set of associativity requirements (one for each triple g, h, k ∈ G ), one hasthat ( S , R , Z ) is acyclic. This is the universal 3-presentation of a group, and onecould say that it is the only uninteresting 3-presentation of the group, since it doesnot reduce the labor required to construct a strict action. Example 1.11.
Given any presentation ( S , R ) , the 3-presentation ( S , R , End( )) isits universal acyclic extension. This example is also not very interesting or useful,because computing End( ) and checking a relation for each element of End( ) canbe prohibitive.In this paper we will give interesting examples of acyclic 3-presentations, extend-ing the usual presentations of Coxeter groups and their braid groups. Said anotherway, we find an interesting presentation of the uninteresting monoidal category Ω G for these groups. IAGRAMMATICS FOR COXETER GROUPS AND THEIR BRAID GROUPS 7
Diagrammatics and Topology.
To any topological space X one may associateits fundamental 2-groupoid π ( X ) ≤ (see e.g. [2, 8.2]) . In this 2-category, the objectsare the points of X , the 1-morphisms from x to y are given by paths from x to y ,and the 2-morphisms are given by “paths of paths” up to homotopy. Our notationis intended to suggest that π ( X ) ≤ is a truncation of the fundamental ∞ -groupoid π ( X ) , which encodes the homotopy type of X . When X is a cell complex, π ( X ) ≤ only depends on the 3-skeleton X ⊂ X .There is an explicit diagrammatic interpretation of π ( X ) ≤ for a cell complex X ,known in the literature as Igusa’s pictures [19] (see Remark 2.1). After fixing someadditional data, one can encode the structure of the cell complex combinatorially.A sufficiently nice map D → X is depicted as a decorated oriented planar graph;we call such a map pictorial . Any map D → X is homotopic to a pictorial map.Moreover, there is a list of relations which account for all homotopies between picto-rial maps, essentially arising from Morse theory. This diagrammatic calculus can beviewed as a tool that takes a 3-skeleton of a cell complex X and returns a combinatorially-defined 2-category Π( X ) ≤ , which is equivalent to π ( X ) ≤ . The objects of Π( X ) ≤ arethe 0-cells of X ; the generating 1-morphisms are the 1-cells and their formal (biad-joint) inverses; the generating 2-morphisms are the 2-cells and their formal inverses,along with units and counits of biadjunction; and the relations between 2-morphismsare given by the 3-cells, along with some general relations (inverses are inverses,other 2-morphisms are cyclic). A 2-morphism in Π( X ) ≤ will be represented by adecorated planar graph, whose regions are labelled by 0-cells, whose edges are la-belled by 1-cells with an orientation, and whose vertices are labelled by 2-cells withsome additional data.Let P = ( S , R ) be a presentation of G . In a standard way, this is also a recipefor a 2-complex X P with a single 0-cell, for which π ( X P ) ∼ = G . The correspondingmonoidal category Π( X P ) ≤ is the category Ω P defined above. Similarly, for a 3-presentation P = ( S , R , Z ) there is a 3-complex X P , and Π( X P ) ≤ equals Ω P . The3-presentation is acyclic if and only if π ( X P ) is trivial, in which case Π( X P ) ≤ ∼ = Ω G ,and X P is the 3-skeleton of some realization of the classifying space BG = K ( G, .The presentation of a Coxeter group has a number of symmetries (though the pre-sentation of its braid group does not). Exploiting these symmetries, we can modifythis construction of Π( X W ) ≤ to produce a simpler diagrammatic calculus. For ex-ample, when constructing a 2-morphism as a decorated planar graph, one need notspecify the orientation on the edges; think of this as using the relation s = 1 tocanonically identify s and s − .1.5. The K ( π, conjecturette. Let ( W, S ) be a Coxeter system and B W be the corre-sponding Artin braid group.In chapter § P B W of B W . The corresponding 3-complex X P BW is the 3-skeleton of a cell complex we shall call X B W , which is the Salvetti Technically, this is called the “homotopy bigroupoid” in [2].
BEN ELIAS AND GEORDIE WILLIAMSON complex. Further discussion of this complex can be found in § X B W is the subject of a famous conjecture. Conjecture 1.12. (The K ( π, -conjecture) X B W is the classifying space of B W . However, the fact that P B W is acyclic is equivalent to a much weaker condition,which we call the K ( π, -conjecturette . Proposition 1.13. (The K ( π, -conjecturette) π ( X B W ) is trivial. Via the work of Salvetti [29], this becomes a question about hyperplane comple-ments, and Digne-Michel’s generalization [10, §
6] of Deligne’s finite-type argumentsgives a proof of the K ( π, -conjecturette for all Coxeter groups. We will quote thisresult henceforth. However, we believe our diagrammatic tools should allow for anelementary and direct proof, which unfortunately has not yet materialized.1.6. Modified Coxeter complexes and half-skeletons. In § P W of W . The corresponding 3-complex X W , which we call (the 3-skeleton of) the modified Coxeter complex , is not a familiar topological space. However, a natural W -fold cover e X W of X W has an equivalent fundamental -groupoid to the 3-skeletonof the completed dual Coxeter complex g Cox W . The completed dual Coxeter com-plex will be discussed further in §
4. Thankfully, g Cox W is known to be contractible,therefore giving the proof that π ( e X W ) ∼ = π ( X W ) is trivial.The classifying space BW (for which X W is supposed to be a model) is a quotientof its universal cover EW (for which e X W is supposed to be a model). The space EW must satisfy two conditions: it must be contractible, and it must admit a free W -action. When trying to build EW as a cell complex, one is torn between these twogoals. Perhaps the neatest approach is to alternate between them, first constructing acontractible space, then extending it until it admits a free W -action, then extending itto make it contractible again, and so forth. This leads to the concept of half-skeletons ,which we use to prove the result about e X W and g Cox W . Half-skeletons are not in-tended to be a complete theory, just a heuristic organizational tool.Let us illustrate the approach using the simplest example of a Coxeter group, W = Z / Z , an example which is treated in more detail within the body of the paper. Theclassifying space of W is RP ∞ , with universal cover S ∞ . The standard cell complexconstruction of S ∞ has two k -cells for each k ≥ , such that the k -skeleton is S k .Note that S k admits a free W -action compatible with the cell decomposition, but isnot contractible.Now take the k -skeleton S k and attach just one of the two ( k + 1) -cells; one obtainsthe disk D k +1 , which is contractible, but does not admit a free W -action. This is the ( k + 0 . -skeleton of EW . To get from the ( k + 0 . -skeleton to the ( k + 1) -skeleton,one attaches an additional ( k + 1) -cell, but along an attaching map which is nulho-motopic in the ( k + 0 . -skeleton (unsurprisingly). Topologically, this operation isjust wedging with S k +1 , and hence does not change π l of the space for any l ≤ k . IAGRAMMATICS FOR COXETER GROUPS AND THEIR BRAID GROUPS 9
Let X k or X k +0 . denote such a skeleton, for k ∈ N . To reiterate, this setup isdesigned so that: • X k admits a free cellular W -action. • X k +0 . is contractible (or at least has trivial fundamental k -groupoid). • To get from X k +0 . to X k +1 , one attaches a set of ( k + 1) -cells (along attachingmaps which are necessarily nulhomotopic).In particular, this guarantees that X k +0 . and X k +1 have equivalent fundamental k -groupoids, so that π l ( X k +1 ) = 0 for l ≤ k .In §
6, for a general Coxeter group W , we construct a . -skeleton and a -skeletonfor EW . The . -skeleton will naturally deformation retract to the completed dualCoxeter complex g Cox W , and is therefore contractible. The -skeleton is exactly e X P W for our chosen 3-presentation. This explains why e X P W and g Cox W have equivalentfundamental -groupoids π ≤ .We do not propose a combinatorial method to construct higher skeletons and half-skeletons for EW , largely because the “diagrammatic” technology for understand-ing higher fundamental groupoids is undeveloped.1.7. Results.
Let ( W, S ) be a Coxeter system and let B W be the corresponding Artinbraid group. We now describe our presentations of Ω B W and Ω W without any ref-erence to topology. Definition 1.14.
Let B diag be the monoidal category with(1) objects – words in S ∪ S − , or equivalently, sequences of oriented dots on aline colored by S . s s stt u (2) morphisms – planar strip diagrams, generated by oriented cups, caps andtwo-colored m -valent vertices. (In the example below, m st = 3 and m su =2 . We will continue to use these to exemplify the general case. There is nogenerator when m st = ∞ .) s s st us ss s s s ts st s su Morphisms are taken modulo the relations below. Each relation holds for any valid“coloring,” i.e. any valid labeling of the edges by elements of S .(3) The standard relations :(1.1a) = (1.1b) = = (1.1c) = = (4) The isotopy relations :(1.2a) = = == (1.2b) = (This picture illustrates the case m = 3 . We require a similar relation for any m -valent vertex.)(5) The generalized Zamolodzhikov relations , one for each finite (standard) parabolicsubgroup of rank 3:(1.3a) Type A × I ( m ) : = (1.3b) Type A : = (1.3c) Type B : = IAGRAMMATICS FOR COXETER GROUPS AND THEIR BRAID GROUPS 11 (1.3d) Type H : = This ends the definition.
Remark . The isotopy relations are equivalent to the statement that one can con-sider these embedded planar graphs up to isotopy.
Remark . In type A , the relation (1.3b) is related to the Zamolodzhikov relation.For this reason, we call the relations in (1.3) generalized Zamolodzhikov relations. Theorem 1.17.
The obvious monoidal functor B diag → Ω B W is an equivalence of categories. Corollary 1.18.
To define a strict action of B W on a category C is equivalent to giving thefollowing data:(1) Functors F s and F − s for each s ∈ S .(2) Natural transformations F s F − s ⇄ and F − s F s ⇄ for each s ∈ S .(3) For each s, t ∈ S with m st finite, natural transformations F s F t F s . . . ⇄ F t F s F t . . . ;here each expression has length m st .This data is subject to the following conditions:(4) The pairs of natural transformations above are inverse isomorphisms.(5) The natural transformations identifying F − s as the inverse of F s form a biadjointstructure.(6) The natural transformations F s F t F s . . . ⇄ F t F s F t . . . are cyclic with respect to thisbiadjoint structure.(7) The generalized Zamolodzhikov relations hold.These conditions correspond to (1.1) , (1.2a) , (1.2b) , and (1.3) respectively. Note that the generalized Zamolodzhikov relations do not involve the functors F − s at all, and can be checked without needing to fix these inverse functors andthe cups and caps. Given a collection of invertible functors F s acting on a suit-ably nice category (i.e. one with functorial biadjoints) and satisfying the generalizedZamolodzhikov relations, one can cook up the rest of the data. Definition 1.19.
Let W diag be the category defined as in Definition 1.14, except with-out any orientations. In other words, objects are words in S , or equivalently, se-quences of (unoriented) colored dots on a line. Morphisms are diagrams up to iso-topy, generated by (unoriented) cups, caps, and m -valent vertices, modulo the un-oriented versions of all the relations above. Theorem 1.20.
The obvious monoidal functor W diag → Ω W is an equivalence of categories. There is a monoidal functor B diag → W diag , which on objects sends both s + and s − to s , and on morphisms sends an oriented diagram to its unoriented version. Thiscategorifies the quotient map B W → W .1.8. Organization of the paper.
We have divided the paper into two parts: thepurely topological, and the Coxeter-theoretic.The first half of this paper will give an exposition of the diagrammatic approachto π ≤ ( § § § §
4) by providing background on Coxetergroups, Coxeter complexes, and Salvetti complexes. In § B diag above, and whose 3-complexagrees with the 3-skeleton of a quotient of the Salvetti complex. The K ( π, -conjecturettestates that π of the Salvetti complex is trivial, which implies Theorem 1.17. Thischapter only requires §
2, not needing the modified diagrams from §
3. In § W diag above.We show that its 3-complex has a 2.5-skeleton which deformation retracts to the dualCoxeter complex, as discussed in § § § §
6, and can ig-nore any mention of half-skeletons. The reader interested only in the Coxeter groupcan safely skip § Applications and further directions.
The authors came to this topic in theirstudy of Soergel bimodules [32], which provide a categorical action of the Hecke al-gebra of W . Certain complexes of Soergel bimodules (known as Rouquier complexes ,see [28]) give braid group actions that, after localization, become Coxeter group ac-tions. The description of strict Coxeter group actions given in this paper also givesa presentation of the monoidal category of localized Soergel bimodules, which isessential to our description of the category of Soergel bimodules [13].Braid group actions on categories appear to be ubiquitous in modern geometricrepresentation theory. These braid group actions are defined using the Coxeter pre-sentation, and so the authors have typically (understandably) neglected to makethese actions strict. However, many of these examples have since been proven tobe strict. Examples of such braid group actions include: Bondal-Kapranov’s con-struction of mutations on triangulated categories [4]; Brou´e-Michel’s constructionin Deligne-Lusztig theory [5]; Khovanov’s homology of tangles [21]; Seidel-Thomastwists around (-2)-curves on derived categories of coherent sheaves [31] and gener-alizations [3]; and braid group actions via shuffling functors in highest weight repre-sentation theory. Deligne had the Bondal-Kapranov and Brou´e-Michel constructions
IAGRAMMATICS FOR COXETER GROUPS AND THEIR BRAID GROUPS 13 in mind as applications when he gave his criterion for a braid group action [8]. In-deed, Deligne’s criterion does seem to be sufficient for many constructions.Other categorical actions of the braid group, such as those arising in categoricalactions of Kac-Moody algebras, have not yet been proven to be strict.Occasionally one can show that a certain space of natural transformations is onlyone-dimensional, and can conclude that strictification data exists without being forcedto provide it explicitly. This is the approach taken by Rouquier [28] and Khovanov-Thomas [22]. In these cases, it is now trivial to make the strictification data explicit.For Rouquier complexes in type A , an explicit approach was taken in [12].Many categorical actions of the braid group descend to actions of the Hecke alge-bra on the Grothendieck group, such as those arising in highest weight representa-tion theory. One expects these to arise from a categorical action of the Hecke algebra,which would then imply the strictness of these actions. Spherical twists such as thosein [31] can be investigated using technology developed by Joseph Grant [17, 16]. Inupcoming work of the first author and Grant, we will demonstrate the connectionbetween certain actions by spherical twist and categorical Hecke algebra actions.In type A , words in the braid group are themselves topological objects, and onehas the notion of braid cobordisms between such words. Braid cobordisms also havea description by generators and relations due to Carter-Saito [6]: the morphisms arecalled movies , and the relations movie moves . Not all movies are invertible, however. Corollary 1.21.
To give a strict braid group action in type A is the same data as an action ofthe invertible braid cobordism category.Proof. In fact, our description of B diag by generators and relations agrees with that ofCarter-Saito for the invertible part of their braid cobordism category. Movie moves3, 5, 6, and 7 correspond to the standard relations (1.1); movie moves 1, 2, and 8correspond to the isotopy relations (1.2); and movie moves 4, 9, and 10 correspondto the Zamolodzhikov relations (1.3). (cid:3) In [12], the first author and Daniel Krasner proved that the entire braid cobor-dism category acts on Rouquier complexes, not just the invertible part. Which otheractions admit such an extension, and what corresponds to the non-invertible braidcobordisms in other types, are both interesting questions.Finally, one should note that Theorems 1.17 and 1.20, which are stated in purelydiagrammatic language, could admit purely diagrammatic proofs. This can be ac-complished in a variety of special cases (e.g. Coxeter groups in type A , by an ar-gument similar to the one used in [11]). However, no general diagrammatic proofcurrently exists.Affine Weyl groups are Coxeter groups, but also admit another presentation, the loop presentation . Often, weak categorical actions of affine Weyl groups are givenusing the loop presentation rather than the Coxeter presentation, and thus differentstrictification data are required. We do not consider this (interesting) question in thispaper. Acknowledgements:
We would like to thank Ruth Charney and Jean Michel for use-ful correspondence. The first author was supported by NSF Postdoctoral FellowshipDMS-1103862.
Part Topology and diagrammatics
2. I
GUSA D IAGRAMS
The goal of this chapter will be to take a cell complex X and construct, by gen-erators and relations, a cyclic biadjoint -category Π( X ) ≤ which is equivalent tothe fundamental 2-groupoid π ( X ) ≤ of X . We assume the reader is familiar withdiagrammatic interpretations of cyclic biadjoint -categories, for which an excellentintroduction can be found in [23, 24].The -morphisms in Π( X ) ≤ should be combinatorial encodings of maps D → X .Note that Π( X ) ≤ only depends on the 3-skeleton of X , so we may assume that X is a3-complex. We follow the procedure described by Roger Fenn in his book [14]. First,enrich the notion of a 3-complex to make it more combinatorial, by adding a smallamount of data for each cell and placing minor restrictions on attaching maps, noneof which is significant up to homotopy equivalence. Given an enriched 2-complex,certain planar diagrams, Igusa diagrams , can be used to encode (nice) maps D → X ,such as the attaching maps of the 3-cells. Finally, one lists the relations betweendiagrams which correspond to homotopy in an enriched 3-complex.To any -presentation P = ( S , R , Z ) of a group G , one can associate a 3-complex X P for which π ( X P ) = G . One can also construct the universal cover e X P → X P insuch a way that the action of G is inherent from the cell complex structure on e X P . Wediscuss the diagrammatics for the corresponding fundamental 2-groupoids below.Fenn’s exposition is highly recommended. We give a quick summary, followingsections 1.2, 2.3 and 2.4 of [14]. Fenn’s discussion requires that X have a unique 0-cell, but it is straightforward to generalize to a cell complex with multiple 0-cells, aswe do below. It is also straightforward to reorganize everything into a 2-category,with one object for each 0-cell. Remark . This diagrammatic interpretation of π ( X ) ≤ for a cell complex X is cred-ited to Whitehead by Igusa [19, Remark following Proposition 7.4]. Subsequent pa-pers (e.g. [25, 33, 20]) call these diagrams “Igusa pictures.” It seems likely that Fennindependently discovered this diagrammatic description [14].2.1. Cell complexes and pictorial maps.Definition 2.2. An (enriched) 3-complex will be the following data. • A set of 0-cells O . • A set of 1-cells S , viewed as oriented edges D between 0-cells. For each s ∈ S , we fix a point b s ∈ Int( s ) , and let b S = { b s } s ∈S . • A set of 2-cells R , viewed as oriented disks D attached along their boundaryto the above oriented graph. We assume each attaching map is pictorial , in a IAGRAMMATICS FOR COXETER GROUPS AND THEIR BRAID GROUPS 15 sense to be defined shortly. For each r ∈ R , we fix a point b r ∈ Int( r ) and apoint pt r ∈ ∂ ( r ) , and let b R = { b r } r ∈R . • A set of 3-cells Z , viewed as oriented balls D attached along their boundaryto the above 2-skeleton. We assume each attaching map is pictorial , in a senseto be defined shortly. For each z ∈ Z we fix a point pt z ∈ ∂ ( z ) .We now define diagrammatic, combinatorial ways to encode maps from D → X and D → X . Definition 2.3. A line diagram is an interval D decorated as follows. A finite numberof points in Int( D ) are labeled with an element of S and an orientation ± , i.e. withan element of S ∪ S − . In this paper we associate a color to each s ∈ S , and we referto this labeling as a “coloring.” The regions between those points are labeled withelements of O . The region to the left of a point labelled s + (resp. s − ) must be thesource (resp. target) of the oriented edge s , and the region to the right must be thetarget (resp. source).To a line diagram f we have a word w ( f ) in the letters S ∪ S − , which determinesthe line diagram uniquely. We also have a word o ( f ) in the letters O . Clearly w ( f ) determines o ( f ) , while o ( f ) determines w ( f ) so long as X has no loops or doubleedges. Example 2.4.
This is an example where the -cells are labelled { a , b , c } and the -cellsare labelled { r, g, b } for r ed, g reen and b lue. We will continue this example below. Wehave drawn a line diagram whose word w is brg − brr − . a b c b c baa b c Suppose that f : D → X is a map. It is represented by a given line diagram if • For each s ∈ S , f − ( b s ) is the collection of points colored s . • Each point colored s in D has a neighborhood which maps homeomorphi-cally to a neighborhood of b s . The sign on that point is + if the homeomor-phism preserves orientation, and − if it reverses it.Note that each connected component of X \ b S is star-shaped and deformation re-tracts to a single 0-cell o ∈ O . The conditions above imply that the entire regionlabelled o in D will map to the corresponding connected component. Also note thatthe endpoints ∂ ( D ) can not map to b S . Definition 2.5.
A map D → X is pictorial if it is represented by some line diagram.A map ( S , pt) → X is pictorial if the corresponding map D → X is pictorial, givenby identifying ∂ ( D ) with pt ∈ S .It is easy to modify the notion of a line diagram to obtain that of a circle diagram ,representing a map ( S , pt) → X . We keep track of the marked point with a tag. One can flip a line or circle diagram, which will invert all the orientations, and willcorrespond to the obvious precomposition with the flip map D → D or S → S . Example 2.6.
This is a loop with word g − brr − r based at c , and its flip r − rr − b − g . b cb c a b cb c a Any line diagram is clearly realized by some map D → X . Any two pictorialmaps D → X with isotopic line diagrams are clearly homotopic (via a homotopysending ∂ ( D ) → X \ b S ). In the definition of an enriched 3-complex, the attachingmap of a 2-cell r is assumed to be pictorial, and thus has a circle diagram; the markedpoint pt r ∈ ∂ ( r ) corresponds to the tag. Definition 2.7. A disk diagram is a particular kind of oriented planar graph in thedisk D . Each edge of the graph is colored with some s ∈ S , and each region islabeled with some o ∈ O , compatible with the orientations on edges via a “left-handed rule.” Edges may run to the boundary ∂ ( D ) , yielding a circle diagram onthe boundary (see example for orientation rules). Edges need not meet any vertices,forming circles, or arcs at the boundary. Each vertex is labelled with an element of R and an orientation ± , i.e. with an element of R ∪ R − . A small circle around a vertexlabelled r + (resp. r − ) must yield the circle diagram of r (resp. the flip of the circlediagram of r ). A disk diagram with marked points on the boundary is exactly that, withthe additional assumption that the marked points do not meet the edges. Example 2.8.
To the -skeleton of the previous examples we have glued a -cell w along b − brr − (based at the -cell b ) and another v along gr − b − (based at a ). Nowwe have constructed a map from the disk which uses w and v − . b cb ca w v a Suppose that one takes a disk diagram and excises a neighborhood of each vertex.What remains is a colored, oriented 1-manifold embedded in the punctured disk.Suppose that f : D → X is a map. It is represented by a given disk diagram if • For each r ∈ R , f − ( b r ) is the collection of vertices labeled r . • Each vertex labeled r has a neighborhood which maps homeomorphically tothe 2-cell r . The sign on the vertex is + if the homeomorphism preservesorientation, and − if it reverses it. IAGRAMMATICS FOR COXETER GROUPS AND THEIR BRAID GROUPS 17 • Let Y denote the disk with those neighborhoods excised. Then Y maps to X .The remainder of these criteria address the restricted map f Y : Y → X . • For each s ∈ S , f − Y ( b s ) is the 1-manifold colored s . • Each connected 1-manifold colored s in Y has a tubular neighborhood map-ping by projection to a neighborhood of b s . The orientation of the manifoldobeys the obvious rule.Once again, these conditions imply that the remainder of the disk (i.e. Y minus thesetubular neighborhoods) is sent to X \ b S , and maps to the connected componentcorresponding to the label on each region. Definition 2.9.
A map D → X is pictorial if it is represented by some disk diagram.A map ( S , pt) → X is pictorial if the corresponding map D → X is pictorial, givenby collapsing the boundary to pt . This implies that the corresponding disk diagramis closed , i.e. its boundary has the empty word.Any disk diagram is clearly realized by some map D → X . Any two picto-rial maps D → X which agree on the boundary and have isotopic disk diagramsare clearly homotopic relative to the boundary. In the definition of an enriched 3-complex, the attaching map of a 3-cell z is assumed to be pictorial, and thus has aclosed disk diagram. One can also define the flip operation on disk diagrams, whichinverts all the orientations.Every map D → X or D → X is homotopic to a pictorial map (and if theboundary is already nice enough, this homotopy can be performed relative to theboundary). Any 3-complex is homotopy equivalent to a 3-complex with pictorial at-taching maps. The choice of additional data needed to enrich a 3-complex is uniqueup to homotopy. Therefore, when studying arbitrary maps from D to arbitrary3-complexes up to homotopy, it is sufficient to study pictorial maps from D to en-riched 3-complexes.For more details, see Fenn [14].Henceforth, we will use the term Igusa diagram to refer to any diagram (on the line,circle, or disk) constructed above. We only consider Igusa diagrams up to isotopy.We also use the word symbol to refer to a vertex in a disk diagram.2.2.
Homotopy relations on diagrams.
Consider a disk diagram with a sub-disk-diagram containing no symbols. This subdiagram represents a map D → X .There are two local transformations of diagrams which result in homotopic maps ( D , ∂D ) → X . These are called bridging and removing circles , and they can be ap-plied to any s ∈ S (we have omitted the labeling of regions). We write the moves byplacing an equal sign between the two diagrams. The transformation (2.2) can alsobe applied with the other orientation.(2.1) = (2.2) = Claim 2.10.
Any two symbol-less disk diagrams which yield homotopic maps ( D , ∂D ) → X are related by a sequence of (2.1) and (2.2) . If we allow symbols, there is a new local transformation of diagrams which resultsin homotopic maps ( D , ∂D ) → X . It is called canceling pairs , and can be appliedto any r ∈ R . In this relation, the orientations must be opposite and the tags must liein the same region.(2.3) = Together, (2.3), (2.2) and (2.1) are called the standard relations . Exercise 2.11.
Use (2.2) and (2.1) to prove that (2.3) is equivalent to the local move(2.4) = =
Claim 2.12.
Two diagrams with the same boundary represent relatively homotopic maps ( D , ∂D ) → X if and only if they are related by the standard relations. The above claims are proven in [14, § Z of oriented balls D along maps ∂D ∼ = S → X . The effect of adding a 3-cell z ∈ Z to X is that itmakes the corresponding closed diagram ( ∂z, pt z ) nulhomotopic. The correspond-ing local move on disk diagrams would be to replace the diagram ( ∂z, pt z ) withthe empty diagram, or vice versa. Note that a disk diagram always represents a mapwhose image lies in X , but this local move corresponds to a homotopy which passesthrough X . We typically do not bother to draw the tag corresponding to pt z on sucha disk diagram, because its location on the empty boundary is irrelevant.Alternatively, one can also consider ∂z ∼ = S as a union of two copies of D alonga common boundary S (containing the marked point pt z ). The two hemisphereswould represent two (possibly non-closed) diagrams with the same boundary, andthe corresponding local move would be to replace one diagram with the other. Givena closed diagram, one can obtain the two hemispheres by slicing the disk in half toform two disks, and taking the flip of one. The relation which replaces ∂ z with an IAGRAMMATICS FOR COXETER GROUPS AND THEIR BRAID GROUPS 19 empty diagram and the relation which replaces one hemisphere with another areequivalent modulo the standard relations.
Example 2.13.
The following two relations, which could arise from the gluing of a3-cell, are equivalent. The attaching maps w and v come from the previous examples. w vv = w = vv Theorem 2.14.
Two diagrams with the same boundary represent relatively homotopic maps ( D , ∂D ) → X if and only if they are related by the standard relations and the new rela-tions imposed by Z .Remark . If one glues in a new 3-cell along an attaching map ( ∂z, pt z ) which is al-ready nulhomotopic, then the new relation is clearly redundant. In other words, anyhomotopy of maps D → X which passes through z could have instead avoided z (though there may be no homotopy of homotopies). If two diagrams are homotopic,it is easy to deduce from the standard relations that their flips are also homotopic.Therefore, after gluing in z , gluing in a 3-cell z along the flipped attaching map willnot affect the diagrammatic calculus.2.3. -categorical language. We can also draw Igusa diagrams in the planar strip R × [0 , rather than the planar disk, and they will be called strip diagrams . They rep-resent (pictorial) maps ( D , pt , pt) → X with two marked points on the boundary.The same local moves as above will describe homotopy classes of such diagrams. Definition 2.16.
Let X be an (enriched) 3-complex, with 0-cells O , 1-cells S , 2-cells R , and 3-cells Z . We define a -category Π( X ) ≤ as follows. The objects will be O . The -morphisms will be generated by s : o → o and s − : o → o , where o (resp. o ) is the 0-cell at the source (resp. target) of the oriented edge s . Thus anarbitrary -morphism is a compatible word in S ∪ S − . The 2-morphisms w → w between compatible words will be the set of strip diagrams constructed with thesymbols r, r − for r ∈ R , modulo isotopy, the standard relations, and a relation foreach z ∈ Z . Composition of 2-morphisms is given by vertical concatenation. Remark . One can also phrase this definition in terms of generators and relations.The 2-morphisms are generated by oriented cups and caps for each s ∈ S , and by symbols r and r − for each r ∈ R . In addition to the standard relations and Z , oneimposes certain “isotopy relations.” See Lauda [23] for more details.Note that oriented cups and caps give 2-morphisms ss − → , etc. Relations (2.1)and (2.2) prove that cups and caps form inverse isomorphisms. Similarly, the symbol r gives a map from w ( r ) → , and r − gives a map → w ( r ) . Relations (2.4) and(2.3) prove that these are inverse isomorphisms.This combinatorially-defined -category encodes everything one needs to knowabout π ≤ ( X ) . In particular, the previous results immediately imply this corollary. Corollary 2.18.
There is an obvious -functor Π( X ) ≤ → π ( X ) ≤ , sending each object o ∈ O to the corresponding point in X . This is a -categorical equivalence. Group presentations.
Let P = ( S , R ) be a 2-presentation of a group G . Thecorresponding 2-complex is the Cayley complex X P , and is constructed in the familiarway. It has a single 0-cell, a 1-cell for each s ∈ S , and a 2-cell for each r ∈ R ,glued in the obvious fashion along its corresponding word (see also [14, § G ∼ = π ( X P ) , although the higher homotopy groups depend on the presentationchosen. Recall that a P = ( S , R , Z ) is a 2-presentation of G with acollection Z of 3-cells; we also denote the corresponding 3-complex X P . We call theelements of Z . Example 2.19.
Suppose G = { e } and P = ( ∅ , R ) , where each element of R is theempty word. Then X G will be a rosette of 2-spheres, one for each element of R .When R is a singleton so that X G ∼ = S , the commutative group π is isomorphic to Z , based on a signed count of appearances of the relation symbol. = Exercise 2.20.
Suppose G = { e } and P = ( { a } , { a } ) . Then X G ∼ = D . Show explicitlythat any two disk diagrams with the same boundary are equivalent.We can also construct a G -fold (universal) cover of X P , which we will denote e X P .The 0-cells will be e O = G , the 1-cells will be e S = S × G , the 2-cells will be e R = R × G , and so forth. Each 1-cell ( s, g ) will go from g to gs ; by convention, edgescorrespond to right multiplication. Each 2-cell ( r, g ) will be attached along the edgescorresponding to the word of r , beginning at the base point g ∈ e O . The 3-cell ( z, g ) isglued into the closed diagram corresponding to z , with the outer region labelled g ,and the other regions labeled in the only consistent way. Clearly e X P comes equippedwith a free action of G by left multiplication on cell names, and e X P /G ∼ = X P . Claim 2.21. π ( e X G ) ∼ = . Moreover, π n ( e X G ) ∼ = π n ( X G ) for all n ≥ .Proof. This is immediate from the long exact sequence associated to the coveringmap. (cid:3)
IAGRAMMATICS FOR COXETER GROUPS AND THEIR BRAID GROUPS 21
Remark . It is somewhat presumptuous to assume that, given P , one knows what G is, or even how big G is. While X P can be constructed explicitly without the set ofelements of G , e X P can not be. When G is infinite so is e X P , but it is locally finite sothat usual topological intuition applies.We use the following conventions for Igusa diagrams of group presentations. Wedo not bother to label the regions of a diagram for X P with the unique element of O .Given a disk diagram for e X P , the label g ∈ G = e O of a single region will determinethe label of every other region. Moreover, having fixed a label on a single region, thecolor ( s, g ) ∈ e S of any edge is determined only by the color s ∈ S , and similarly forsymbols ( r, g ) ∈ e R . We omit the redundant data, coloring edges only by s ∈ S andnaming symbols by r ∈ R . Thus a disk diagram for e X P is the same data as a diskdiagram for X P with an arbitrary choice of label g ∈ G for a single chosen region.Postcomposing a map D → e X P with the quotient map e X P → X P corresponds toforgetting the label on that region.The following example will be crucial in chapter 3. Example 2.23.
Let P = ( { s } , { s } , { z } ) be a presentation for the group G = Z / Z .The 3-cell z is glued in along the picture:Then the 2-skeleton of X P will be RP , and X P will be RP . In particular, π ( X P ) istrivial.To construct e X P , we take two points, add two edges to get S , add two disks toget S , and add two 3-cells to get S . This is the 3-skeleton of S ∞ ∼ = EG in its usualconstruction.Let Y denote the 2-skeleton of e X P with only a single 3-cell added, so that Y ∼ = D .We think of Y as the “2.5-skeleton” of EG . To obtain e X P from Y , one attaches a 3-cell which is redundant in the sense of Remark 2.15. Therefore, Y and e X P have thesame category Π ≤ . Of course, Y does not admit a free action of G , but it has otheradvantages. For instance, Y deformation retracts to a pole between the two 0-cells,which is the (completed) dual Coxeter complex of G .3. M ODIFIED F ENN D IAGRAMS
The Coxeter presentation has a number of natural symmetries, and we wish to ex-ploit them in order to simplify our diagrammatic description of Ω W . In this chapterwe develop some general machinery which yields simpler diagrammatics for specialkinds of 3-presentations.Suppose that ( S , R ) is a group presentation, where s ∈ R for some s ∈ S . Thereis a particular 3-cell one can glue in, which will cause s and s − to be canonicallyisomorphic, and this allows us to ignore the orientations on the strands colored s in diagrams for e X P . Heuristically, these s -unoriented diagrams depict Π ≤ for somedeformation retract of a “2.5-skeleton” of e X P , as in Example 2.23. In similar fashion,we describe a modification adapted to rotational and flip symmetries in relations,such as in the braid relation.3.1. Modified diagrammatics for involutions.
First let us consider diagrams for P = ( { s } , { s } ) , so that X P ∼ = RP and e X P ∼ = S . The relation s allows us to drawbivalent vertices which look like this.The sign on the symbol is determined by the orientations of the strands, but thelocation of the tag is not. Therefore, the bivalent vertex gives two natural maps s → s − , depending on the placement of the tag, and two natural maps the otherdirection.Using (2.3) and (2.4) we have(3.1) = (3.2) = = We now add a -cell z s to obtain the higher presentation P = ( { s } , { s } , { z s } ) , andtemporarily write e X = e X P and X = X P . The new -cell is meant to kill π ( RP ) , andin e X to kill π ( S ) . If we attach two bivalent vertices together so that the tags do notcancel, this represents the map that z s is glued into. Thus we have a new relation:(3.3) = Splitting a new -cell into hemispheres, we obtain the equivalent relations:(3.4) = = We may introduce a new symbol: a bivalent vertex without a tag. This symbol isset equal to the bivalent vertex with either placement of the tag. Now (3.2) becomes
IAGRAMMATICS FOR COXETER GROUPS AND THEIR BRAID GROUPS 23 (3.5) = Thus the bivalent vertex gives a (canonical) isomorphism between s and s − . Exercise 3.1.
Any closed diagram for X P is equal to the empty diagram. In otherwords, π ( X P ) = 1 . The topological statement is obvious: find a diagrammatic proof.Now let P = ( S , R , Z ) be any 3-presentation, and let s ∈ S be such that s ∈ R and z s ∈ Z for z s as above. The same arguments as above show that we may ignorethe tag on the bivalent vertex associated to s . Moreover, the bivalent vertices ofeither sign form inverse isomorphisms between s and s − , and we wish to use themto canonically identify the two objects. Given any Fenn diagram we may forget theorientation data associated to s to get an s -unoriented Fenn diagram . In particular, eachrelation has an s -unoriented symbol. Let s denote a point s without an orientation. Definition 3.2.
Suppose that P = ( S , R , Z ) is a 3-presentation containing ( s, s , z ) .Let Π( X P , s ) ≤ be the 2-category with a single object, defined as follows. The 1-morphisms are generated by S ′ ∪ ( S ′ ) − ∪ { s } , where S ′ = S \ { s } . The 2-morphismsare generated by the s -unoriented symbols of R ′ ∪ ( R ′ ) − , for R ′ = R \ { s } , and thuscorrespond to s -unoriented disk diagrams. The relations are generated by Z \{ z s } , aswell as the usual Fenn relations for oriented parts of the diagram and the unorientedFenn relations for s :(3.6) = (3.7) = (3.8) = There is a natural 2-functor Π( X P ) ≤ → Π( X P , s ) ≤ . It sends both s and s − to s . Itsends the bivalent vertex corresponding to s to the identity map of s . To every diskdiagram without bivalent vertices, it forgets the orientation data associated to s . Itis easy to show that this 2-functor is an equivalence. Given any s -unoriented diskdiagram, and any choice of orientations of s on the boundary, one may choose a diskdiagram by placing orientations on s -strands willy-nilly, and adding bivalent ver-tices whenever necessary for consistency. While there are multiple such diagrams,they are all equal in Π( X P ) ≤ . If P contains ( s, s , z s ) for multiple distinct involutions in S , there is no obstructionto forgetting the orientations on multiple colors at once. We write Π( X P ) un-or ≤ for the2-category where every such orientation is ignored.The case of e X P can be treated in the same way. One must glue in a copy of z s forevery possible region labeling. As before, diagrams for e X P will be s -unoriented diskdiagrams with a label in a single region. Remark . Here is a heuristic topological understanding of unoriented diagrams, atleast for e X P . As in Example 2.23, let Y ∼ = D be the 2.5-skeleton of S ∞ , which has two0-cells and s , two 1-cells ( s, and ( s, s ) , two 2-cells ( s , and ( s , s ) , and a single3-cell ( z s , . One can construct a new “unoriented” cell complex Y un-or , consistingof two 0-cells and s , and a single unoriented 1-cell s between them. We think of s as a pole inside Y ∼ = D . Clearly Y un-or ⊂ Y is a deformation retract, under a retractsending both edges ( s, and ( s, s ) to s .Similarly, suppose that ( s, s , z s ) ⊂ P for a general 3-presentation. After construct-ing e X P , one can repeat the above construction for each coset { x, xs } ∈ G to obtaina 3-complex Y which deformation retracts to a 1-complex Y un-or . In Y un-or , x and xs are connected by a single edge ( s, x ) . The attaching maps of other 2-cells in R can bedeformed to lie on Y un-or , and similarly for the other 3-cells in Z , yielding a defor-mation retract Y un-or of a 2.5-skeleton Y of e X P . We think of unoriented diagrams asdescribing maps to Y un-or (even though Fenn diagrams for Y un-or are actually quitedifferent). There is no reasonable Z / Z action on Y or Y un-or whose quotient has π = G , so we do not use this heuristic when thinking about X P , only e X P .3.2. Rotational invariance and flip invariance.
When a relation does not have ro-tational invariance, there is no need to keep track of the tag on the correspondingsymbol in a Fenn diagram. The location of the tag can be deduced from the edgecoloring. When a relation does have rotational invariance, the tag is not redundant.However, if an appropriate 3-cell is glued in, all possible locations of the tag will beset equal, and the tag will become redundant. An exactly analogous procedure willwork to make the sign on a symbol redundant when a relation has flip invariance.We will use specific examples to illustrate general principles, because it is hard todraw a general example.First consider P = ( { r, g, b } , { w = rgbrgbrgb } ) . The symbol for w can be rotated by120 degrees and 240 degrees to give a morphism with the same boundary. This is adifferent morphism because the tag is in the wrong place.If we set two of these to be equal by gluing in a 3-cell, then the third will be equalas well. In general, if w is invariant under rotation by θ then setting w equal to θ ( w ) IAGRAMMATICS FOR COXETER GROUPS AND THEIR BRAID GROUPS 25 will also set it equal to nθ ( w ) for any n ∈ Z . The 3-cell z w would be glued in alongthe following closed diagram, which is a “mismatched pair.”Once this 3-cell is glued in, one need not draw the tag on this symbol any longer.There are only three valid locations for the tag (it must be before r and after b ), andthey all give equal 2-morphisms.In the previous section, we had to construct a new 2-category which equated twocanonically isomorphic objects. In this section, we are not changing the category, butare merely using a notational convenience, using one symbol to represent severaldistinct symbols which happen to be equal.The case of e X P can be treated in the same way. One must glue in a 3-cell as abovefor every possible region labeling. Remark . As in Remark 3.3, there is a topological heuristic for the new diagram-matic calculus. Suppose that P contains ( { r, g, b } , { w = rgbrgbrgb } , z w ) as above. Forany x ∈ G there are three different -cells being glued to the same S ⊂ e X P : ( w, x ) , ( w, xrgb ) and ( w, xrgbrgb ) . Fixing the same base point in S for all three, they are ( w, x ) , ( θ ( w ) , x ) and ( θ ( w ) , x ) . One can visualize this part of the 2-skeleton as a stackof pancakes, glued together along their rim. The 3-cell ( z w , x ) fills in the gap betweenthe first two pancakes, while the 3-cell ( z w , xrgb ) fills in the gap between the secondand third pancakes. With these two 3-cells glued in, the result is a copy of D . There-fore the last 3-cell ( z w , xrgbrgb ) would be redundant, and we need not glue it in.Ignoring this 3-cell (for each x ) one obtains the “2.66-skeleton” of e X P (we continueto call it the 2.5-skeleton), and it deformation retracts to a central D pancake-shapedslice. This central slice is what the tagless symbol is meant to represent.The reader can deduce the rest of the analogy. Unlike Remark 3.3, replacing e X P with the deformation retract of its 2.5-skeleton does not change the 1-skeleton, whichis why one need not change the objects in the category.Now consider G = ( { r, g } , { w = rgg − r − } ) . It lacks any rotational symmetry, butit does have a flip symmetry: the symbol for w and some rotation of the oppositeorientation of w have the same boundary. We can glue in a 3-cell z w along a “mis-matched pair.” Once this 3-cell is glued in, one need not keep track of the sign on the symbol anylonger.Of course, the relation w ∈ π ( X P ) is already nulhomotopic even in X P , as anyrelation with flip symmetry will be! This restricts the notion of flip symmetry tounusual presentations.Flip symmetry becomes more interesting for unoriented Fenn diagrams. Supposethat P = ( { r, g, b } , { r , g , b , rgbrbg = w } , Z ) and that Z contains the 3-cells whichallow for unoriented diagrams as in the previous section. The unoriented symbol for w has no rotational symmetry, but it does have flip symmetry.Without this 3-cell added as a relation, a diagram with boundary rgbrbg could beeither w + or a rotated w − . This 3-cell would set them equal. For this example it isnot terribly meaningful to say that we can remove the ± decoration on w , becauseone must keep the tag, and the sign can be deduced from the tag. The followingexample combines all three modifications, and gives a situation where removing thesign does have a noticeable effect. Example 3.5.
Consider the Coxeter presentation ( { s, t } , { s , t , stst − s − t − } ) . Now,glue in the 3-cells for each generating involution, so we may work with unorienteddiagrams. Then glue in a 3-cell for rotational invariance. At this point, the followingtwo diagrams do not represent the same 2-morphism (tags included for clarity).Gluing in one more 3-cell for flip symmetry, we can ignore the sign on the symbol,and draw the 2-morphism unambiguously as a 6-valent vertex. IAGRAMMATICS FOR COXETER GROUPS AND THEIR BRAID GROUPS 27
Example 3.6.
Let P be the 3-presentation of the previous example. What is e X P ? Itis more complicated than it looks, because each cell appears 6 times, once for eachelement of W .We begin with six 0-cells. Instead of labeling them by elements of W , let us men-tally arrange them as a hexagon and label them by numbers modulo 6, as on a 6-hour clock. We then glue in twelve 1-cells (six for s and six for t ). These connect eachneighboring pair of 0-cells with two edges, yielding six copies of S welded into ahexagonal loop. Next, the relations in Q attach twelve 2-cells (six of each), two gluedinto each copy of S . This yields six copies of S welded into a loop. Then we glue inthe “orientation 3-cells”, twelve of them, two glued into each copy of S . This yieldssix copies of S welded into a loop. However, six of the twelve 3-cells are redundant:after a 3-cell has turned S into D , the other 3-cell will have a nulhomotopic attach-ing map. Ignoring these six redundant 3-cells, we have six copies of D welded intoa loop. This space deformation retracts to the 1-skeleton of g Cox (to be defined in § g Cox, a solid hexagon.
Part Coxeter groups and braid groups
4. C
OXETER GROUPS AND TOPOLOGY
In this chapter we give some background information on Coxeter groups, theirArtin braid groups, and some associated topological spaces.4.1.
Coxeter groups.
Fix a set S , and for each pair s = t ∈ S fix an element m st ∈ Z ≥ ∪ {∞} . The Coxeter group W is defined by its Coxeter presentation ( S , Q ∪ B ) , where the quadratic relations are Q = { s } s ∈S and the braid relations are B = { b s,t } s = t ∈S for b s,t = sts . . . | {z } m st . . . t − s − t − | {z } m st . There is no braid relation when m st = ∞ . There is only one braid relation for eachpair s, t ∈ S ; we will not redundantly use both b s,t and b t,s . The corresponding Artinbraid group B W has presentation ( S , B ) . We let B + W ⊂ B W denote the monoid of positive braids , which is the monoid with the same presentation ( S , B ) .We assume that S is finite, though this is not strictly necessary for our arguments.We let r = |S| be the rank of W . Let ℓ denote the length function.For a subset I ⊂ S , there is a parabolic subgroup W I ⊂ W generated by s ∈ I . Itis also a Coxeter group, with presentation ( I, Q I ∪ B I ) . When there exists a partition S = I ` I such that m st = 2 for all s ∈ I and t ∈ I , then W ∼ = W I × W I , andwe say that W is reducible . When W I is finite, we say that I is finitary , and we let w I denote the longest element of W I .A Coxeter group of rank 2 is determined by m = m st , and is said to be of type I ( m ) . It is finite unless m = ∞ . The group I (2) is the reducible group A × A .There is a classification of all finite Coxeter groups. The finite Coxeter groups ofrank 3 are types A , B , H and the reducible types A × I ( m ) for m < ∞ .For an element w ∈ W , we will use an underline w = s s · · · s d to indicate anexpression for w in terms of S . If we need to differentiate between two expressionsfor w we will write w and w . We say that w is reduced if d = ℓ ( w ) . Given w ∈ W , achoice of reduced expression w will also yield an element e w of B + W , independent ofthe reduced expression chosen. We call this the positive lift of w to B W .See Humphreys [18] for more details.4.2. The Coxeter complex.
To a Coxeter system ( W, S ) one may associate a simpli-cial complex, the Coxeter complex | ( W, S ) | , as follows:(1) Choose an arbitrary total order on S .(2) Color the r faces of the ( r − -simplex by S , matching the lexicographic orderon faces to the total order on S ; call the resulting simplex ∆ .(3) Take one copy ∆ w of ∆ for each w ∈ W .(4) Glue ∆ w to ∆ ws along the face colored by s , for all w ∈ W and s ∈ S . There isonly one possible gluing which preserves the orientation.The result is a connected ( r − -dimensional simplicial complex with simplices ofmaximal dimension labelled by W and codimension one simplices (or walls ) coloredby S . Moreover, W acts on | ( W, S ) | by automorphisms preserving the coloring ofwalls.If one chooses a different total order on S , one obtains the same complex withdifferent orientations. We do not care about the simplicial orientations in the Coxeter IAGRAMMATICS FOR COXETER GROUPS AND THEIR BRAID GROUPS 29 complex (or in any of the other complexes we construct in this chapter; the avidreader may fill in the details). In the examples below we will draw orientations onthe walls; these orientations have nothing to do with the simplicial orientations, butinstead record the Bruhat order on W . Example 4.1.
Let W be a finite subgroup of the orthogonal group O ( V ) of a Euclideanvector space V of dimension r , and assume W is generated by reflections and actsirreducibly. Let T denote the subset of W of elements which act as reflections on W .Consider the space U := V − [ t ∈ T V t obtained from V by deleting all reflecting hyperplanes. Then W acts simply transi-tively on the connected components of U . If one fixes a connected component C of U ,then C is a simplicial cone, and ( W, S ) is a Coxeter system of rank r , where S denotesthe set of reflections in the walls of C . If one intersects C with the unit sphere in V then one obtains a closed subset ∆ homeomorphic to an ( r − -simplex, whose facesare colored by S . The W -translates of { w ∆ | w ∈ W } give a triangulation of the unitsphere, giving a realization of the Coxeter complex. In fact, all Coxeter complexesassociated to finite Coxeter systems can be realized in this way. Example 4.2.
As for any finite rank 3 Coxeter system, the Coxeter complex for A × A × A is a triangulation of the sphere. The triangles are labeled by w ∈ W . Thetriangle closest to the reader is labeled with the identity, and the triangle furthest isthe longest element. We place orientations on edges such that going from the leftside of an edge to the right side will increase the length of w ∈ W by . > < < > < > > > > > < < Given x, y ∈ W , a gallery from x to y in the Coxeter complex | ( W, S ) | is a pathbetween the simplices corresponding to x and y , which does not meet any simplexof codimension ≥ . We regard two galleries as equivalent if they visit the samesimplices in the same order. A gallery from x to y is minimal if it crosses the leastnumber of walls amongst all galleries from x to y . Giving a gallery from x to y isthe same thing as giving an expression for x − y . Indeed, a gallery is determineduniquely by the ordered list of walls crossed in the path from x to y . A gallery from x to y corresponding to an expression st · · · u for x − y is minimal if and only if st · · · u is reduced. The Dual Coxeter complex.
For our purposes it will be more convenient to usethe dual Coxeter complex , which is the CW-complex | ( W, S ) | ∨ dual to | ( W, S ) | . It has a0-cell for each w ∈ W , and a gallery in the Coxeter complex corresponds to a path inthe 1-skeleton of the dual Coxeter complex.Let C be any face of codimension k < r in | ( W, S ) | . One can label C by the rank k subset I ⊂ S , consisting of the colors on the walls which contain that face. Thenthere is a face labeled by I if and only if I is finitary. Moreover, the ( r − -simplicescontaining such a face C are labeled by elements of W forming a coset in W/W I .Hence one can construct | ( W, S ) | ∨ as follows:(1) Take a 0-cell for each w ∈ W .(2) Attach a 1-cell from x to xs , when xs > x .(3) Attach a 2-cell between the two minimal galleries from x to xw s,t , when m s,t is finite and x is a minimal length coset representative.(4) . . . Let us elaborate upon the inductive step. Fix any coset C in W/W I for I finitaryof rank k . Consider the cells whose closure only contains 0-cells corresponding toelements in C . After the k − -st step, the union of these cells will be homeomorphic S k − . The k -th step is to glue in a k -cell and obtain D k instead. As | ( W, S ) | is ( r − -dimensional, this process ends after ( r − steps.One can also form the completed dual Coxeter complex , which includes the r -th stepabove. We denote it by g Cox. It differs from | ( W, S ) | ∨ in a single r -cell when W isfinite, and does not differ otherwise. In the finite case, g Cox gives a CW-complexstructure for the unit ball in Euclidean space, rather than the unit sphere.
Example 4.3.
When W is a finite dihedral group of size m , g Cox is the solid m -gon. Exercise 4.4.
Suppose that W = W × W is a product of two other Coxeter groups.Show that g Cox W ∼ = g Cox W × g Cox W , compatibly with the CW structure. Proposition 4.5.
The completed dual Coxeter complex g Cox is contractible.Proof.
By the exercise above, we may assume that W is irreducible. When W is in-finite, the result follows from the contractibility of the Coxeter complex (see e.g. [1,Theorem 4.127]). When W is finite, the completed dual Coxeter complex is a unitball. (cid:3) The (completed) dual Coxeter complex g Cox does have an action of W , which actsby left multiplication on 0-cells. However, this action is not free, and the quotientdoes not inherit a nice CW-complex structure. The dihedral group acting on theregular m -gon provides a familiar example. Thus g Cox does not provide a good CW-complex model for EG . Instead, the (3-dimensional) model we construct in chapter6 will contain (the 3-skeleton of) g Cox as a deformation retract.
IAGRAMMATICS FOR COXETER GROUPS AND THEIR BRAID GROUPS 31
The Salvetti complex.
The completed dual Coxeter complex has one k -cell foreach pair ( I, C ) , where I is finitary of rank k , and C is a coset of W/W I . Suppose weplace an equivalence relation on g Cox, identifying any two k -cells ( I, C ) and ( I, C ′ ) .The quotient is still a CW-complex, having a single k -cell for each finitary I ⊂ S .For instance, there is a single 0-cell, a 1-cell for each s ∈ S , and a 2-cell for each pair s = t ∈ S with m s,t < ∞ . We call this CW-complex | B W | .Similarly, one can construct a W -fold cover of this CW-complex, called the Salvetticomplex
Sal. It has one k -cell for each pair ( I, w ) , with w ∈ W and I ⊂ S finitary ofrank k . The k -cell ( I, w ) is glued in such a way that it contains 0-cells labeled by wu for u ∈ W I .Note that the Salvetti complex is different from g Cox, despite having the same 0-cells, and | B W | is different from the quotient of g Cox by the action of W describedabove. Example 4.6.
Consider type A . Then g Cox ∼ = D is an interval connecting two 0-cells and s . The complex Sal ∼ = S has two -cells connecting the 0-cells and s . Thequotient g Cox /W is also an interval, folded in half. Meanwhile | B W | ∼ = S identifiesthe endpoints of the interval, or wraps the Salvetti complex in half.We have already discussed the K ( π, -conjecture in some detail in the introduc-tion § π ( | B W | ) ∼ = B W . The K ( π, -conjecture states that all higherhomotopy groups vanish; the K ( π, -conjecturette states that π ( B W ) = 0 . The K ( π, -conjecturette is known for all Coxeter groups W , thanks to work of Digne-Michel [10].For more information on the K ( π, -conjecture and a list of cases where it isknown, see the survey paper [27].5. D IAGRAMMATICS FOR B RAID GROUPS In § | B W | . We now seek to describe π ( | B W | ) ≤ diagrammatically. Definition 5.1.
Let B diag denote Π( X P ) for the 3-presentation P = ( S , B , Z ) below.The presentation ( S , B ) agrees with the presentation of the braid group given in § Π( X P ) is a word in the letters S ∪S − . The morphisms are gen-erated by oriented cups and caps, as well as m st -valent vertices as pictured below,whenever m st < ∞ for the two colors present.These morphisms satisfy the Fenn relations:(5.1) = (5.2) = = (5.3) = = Remark . There is only one 2-cell for each pair s = t ∈ S with m s,t < ∞ . The twodifferent kinds of m -valent vertices are the two orientations of the correspondingsymbol. Both the tag and the orientation on the symbol can be determined from thecoloring and orientation on the strands, so we do not draw them in our diagramshenceforth.In addition, for any three colors forming a finite parabolic subgroup, there is asingle 3-cell in Z . The corresponding relation is the generalized Zamolodzhikovequation, given in (1.3).By now, it is clear that Theorem 1.17 is equivalent to the K ( π, -conjecturette, andis thus proven. 6. D IAGRAMMATICS FOR C OXETER GROUPS
Let ( W, S ) be a Coxeter group, with the usual presentation ( S , Q ∪ B ) . Let g Cox beits completed dual Coxeter complex.
Definition 6.1. A standard diagram for W , will be a diagram with unlabeled regions,unoriented edges colored by s ∈ S , and (untagged, unoriented) m -valent verticeswhich alternate between edges colored s and t for which m st = m < ∞ . A labeledstandard diagram is a standard diagram with a single region labeled by an element of W .As noted previously, it is equivalent to give a label in W for a single region, and toconsistently label each region by an element of W , such that two regions separatedby an edge s differ by that element in W . Definition 6.2.
Let W diag denote the monoidal category whose objects are generatedby s ∈ S , and whose morphisms are given by standard diagrams modulo isotopyand the following relations (the Fenn relations and the Zamolodzhikov relations).(6.1) = (6.2) = IAGRAMMATICS FOR COXETER GROUPS AND THEIR BRAID GROUPS 33 (6.3) = (6.4) = (6.5) = (6.6) = (6.7) = Let f W diag denote the 2-category whose objects are elements of W , and whose 2-morphisms are labeled standard diagrams, modulo the relations above.In § Definition 6.3.
Let P = ( S , Q ∪ B , Z ∪ M ) be the following 3-presentation, ex-tending the usual presentation of W , which we call the Coxeter 3-presentation . TheZamolodzhikov 3-cells Z are the same as in the previous chapter. The “diagram-simplifying” 3-cells M consist of: • one 3-cell z s for each generating involution s ∈ S , as in (3.3); • one 3-cell for each braid relation ( st ) m = 1 accounting for rotational symme-try; • one 3-cell for each braid relation ( st ) m = ( ts ) m = 1 , accounting for flip sym-metry.The 3-cells accounting for rotational and flip symmetry were described in § Remark . As discussed in § Proposition 6.5. f W diag is isomorphic (not just equivalent) to Π( e X P ) un-or ≤ as -categories. Now we restate and prove one of our main theorems from the introduction.
Theorem 6.6.
The obvious functor W diag → Ω W is an equivalence of categories.Proof. It is enough to prove that π ( e X P ) = 0 . This follows from the lemma below. (cid:3) Lemma 6.7.
By removing redundant 3-cells from e X P , one obtains a space which will de-formation retract to the 3-skeleton of g Cox. In other words, e X P is homotopy equivalent to g Cox ∨ S ∨ · · · ∨ S . Example 3.6 illustrates the basic idea of this proof.
Proof.
We prove this lemma in steps. At the k -th step, we construct a sub-complex e X ( k ) of e X P by choosing certain cells to include. The sub-complex e X ( k ) is not the k -skeleton, though it will contain all k -cells of e X P when k < . We show that e X ( k ) deformation retracts to g Cox k . In particular, up to homotopy equivalence, we canconstruct e X ( k +1) by gluing higher cells to g Cox k instead of e X ( k ) . For k = 3 , the differ-ence between e X (3) and e X P will consist entirely of redundant 3-cells. Both g Cox and e X P have the same 0-skeleton, so we begin with e X (0) = e X P .Now consider a single s ∈ S , and its parabolic subgroup W s ⊂ W . By gluingin the 1-cells, 2-cells, and 3-cells corresponding to the sub-presentation ( s, s , z s ) ,one obtains a copy of S for each coset of W s in W (see § D for each coset. Each D willdeformation retract to a single edge between the two 0-cells, which can be thought ofas an s -colored edge in g Cox . Thus if we take e X (0) and add both 1-cells, both 2-cells,and one 3-cell of ( s, s , z s ) for each s ∈ S , we obtain a space e X (1) which deformationretracts to g Cox .Now consider a single pair s, t ∈ S with m = m s,t < ∞ , and its parabolic subgroup W s,t ⊂ W . Let b s,t denote the braid relation inside B , r s,t denote the rotation 3-cell in-side M , and f s,t denote the flip 3-cell inside M . Each coset of W s,t in W correspondsto a hollow m -gon in g Cox (or something which deformation retracts to a hollow m -gon in e X (1) ). Gluing in the 2-cells corresponding to b s,t , each coset will look like m disks, each glued along their boundary to a common S . One can visualize thisas an amalgamation of m − copies of S , where the southern hemisphere of the i -th copy is identified with the northern hemisphere of the i + 1 -st copy. There area total of m r s,t and f s,t ( m of each), each of which gives IAGRAMMATICS FOR COXETER GROUPS AND THEIR BRAID GROUPS 35 a cobordism between two different disks. One can choose m − such 3-cells to fillin the m − copies of S , yielding a cell complex structure on D . The remaining (2 m + 1) e X (2) . This copyof D for each coset will deformation retract to a single solid m -gon, which is a 2-cell in g Cox corresponding to W s,t . Thus if we take e X (1) and add all m m − ( b s,t , { r s,t , f s,t } ) for each s, t ∈ S with m s,t < ∞ , we obtain a space e X (2) which deformation retracts to g Cox .Now consider a single triple s, t, u ∈ S whose parabolic subgroup W s,t,u has finitesize n . Let Z s,t,u denote the Zamolodzhikov 3-cell in Z . Each coset of W s,t,u gives asubspace of g Cox which is a particular cell structure for S . In e X P , Z s,t,u correspondsto n S into D . Clearly only one such3-cell is necessary, after which the remaining ones are redundant. This single 3-cellcorresponds precisely to the 3-cell in g Cox for that coset. Thus if we take e X (2) andadd a single 3-cell of the form Z s,t,u for each coset of W s,t,u , we obtain the desiredspace e X (3) which deformation retracts to g Cox . (cid:3) Remark . It is not unreasonable to expect a purely diagrammatic proof of Theorem6.6, and certainly this can be achieved in special cases. However, the difficulty infinding this proof was what led the authors to this topological detour.R
EFERENCES [1] Peter Abramenko and Kenneth S. Brown.
Buildings , volume 248 of
Graduate Texts in Mathematics .Springer, New York, 2008. Theory and applications.[2] John C. Baez and Aaron D. Lauda. Higher-dimensional algebra. V. 2-groups.
Theory Appl. Categ. ,12:423–491, 2004.[3] Roman Bezrukavnikov and Simon Riche. Affine braid group actions on derived categories ofSpringer resolutions.
Ann. Sci. ´Ec. Norm. Sup´er. (4) , 45(4):535–599 (2013), 2012.[4] Alexei I Bondal and Mikhail M Kapranov. Representable functors, Serre functors, and mutations.
Mathematics of the USSR-Izvestiya , 35(3):519, 1990.[5] Michel Brou´e and Jean Michel. Sur certains ´el´ements r´eguliers des groupes de Weyl et les vari´et´esde Deligne-Lusztig associ´ees. In
Finite Reductive Groups: Related Structures and Representations ,pages 73–139. Springer, 1997.[6] J. Scott Carter and Masahico Saito.
Knotted surfaces and their diagrams , volume 55 of
MathematicalSurveys and Monographs . American Mathematical Society, Providence, RI, 1998.[7] Joseph Chuang and Rapha¨el Rouquier. Derived equivalences for symmetric groups and-categorification.
Annals of Mathematics , pages 245–298, 2008.[8] P. Deligne. Action du groupe des tresses sur une cat´egorie.
Invent. Math. , 128(1):159–175, 1997.[9] Pierre Deligne. Les immeubles des groupes de tresses g´en´eralis´es.
Inventiones mathematicae ,17(4):273–302, 1972.[10] F. Digne and J. Michel. Parabolic Deligne–Lusztig varieties.
Adv. Math. , 257:136–218, 2014.[11] Ben Elias and Mikhail Khovanov. Diagrammatics for Soergel categories.
Int. J. Math. Math. Sci. ,pages Art. ID 978635, 58, 2010.[12] Ben Elias and Dan Krasner. Rouquier complexes are functorial over braid cobordisms.
Homology,Homotopy Appl. , 12(2):109–146, 2010. [13] Ben Elias and Geordie Williamson. Soergel calculus.
Represent. Theory , 20:295–374, 2016.arXiv:1309.0865.[14] Roger A. Fenn.
Techniques of geometric topology , volume 57 of
London Mathematical Society LectureNote Series . Cambridge University Press, Cambridge, 1983.[15] St´ephane Gaussent, Yves Guiraud, and Philippe Malbos. Coherent presentations of Artin groups. arXiv preprint arXiv:1203.5358 , 2012.[16] Joseph Grant. Lifts of longest elements to braid groups acting on derived categories. arXivpreprint arXiv:1207.2758 , 2012.[17] Joseph Grant. Derived autoequivalences from periodic algebras.
Proc. Lond. Math. Soc. (3) ,106(2):375–409, 2013.[18] James E. Humphreys.
Reflection groups and Coxeter groups , volume 29 of
Cambridge Studies in Ad-vanced Mathematics . Cambridge University Press, Cambridge, 1990.[19] Kiyoshi Igusa. The Borel regulator map on pictures. I. A dilogarithm formula. K -Theory ,7(3):201–224, 1993.[20] Mikhail M. Kapranov and Masahico Saito. Hidden Stasheff polytopes in algebraic K -theory andin the space of Morse functions. In Higher homotopy structures in topology and mathematical physics(Poughkeepsie, NY, 1996) , volume 227 of
Contemp. Math. , pages 191–225. Amer. Math. Soc., Provi-dence, RI, 1999.[21] Mikhail Khovanov. A functor-valued invariant of tangles.
Algebr. Geom. Topol. , 2:665–741, 2002.[22] Mikhail Khovanov and Richard Thomas. Braid cobordisms, triangulated categories, and flagvarieties.
Homology Homotopy Appl. , 9(2):19–94, 2007.[23] Aaron D. Lauda. A categorification of quantum sl(2) . Adv. Math. , 225(6):3327–3424, 2010.[24] Aaron D. Lauda. An introduction to diagrammatic algebra and categorified quantum sl . Bull.Inst. Math. Acad. Sin. (N.S.) , 7(2):165–270, 2012.[25] Jean-Louis Loday. Homotopical syzygies. In
Une d´egustation topologique [Topological morsels]: ho-motopy theory in the Swiss Alps (Arolla, 1999) , volume 265 of
Contemp. Math. , pages 99–127. Amer.Math. Soc., Providence, RI, 2000.[26] Luis Paris. Universal cover of Salvetti’s complex and topology of simplicial arrangements ofhyperplanes.
Transactions of the American Mathematical Society , 340(1):149–178, 1993.[27] Luis Paris. K ( π, conjecture for Artin groups. Ann. Fac. Sci. Toulouse Math. (6) , 23(2):361–415,2014.[28] Raphael Rouquier. Categorification of the braid groups. arXiv preprint math/0409593 , 2004.[29] Mario Salvetti. Topology of the complement of real hyperplanes in C n . Inventiones mathematicae ,88(3):603–618, 1987.[30] Mario Salvetti. The homotopy type of Artin groups.
Math. Res. Lett , 1(5):565–577, 1994.[31] Paul Seidel, Richard Thomas, et al. Braid group actions on derived categories of coherentsheaves.
Duke Mathematical Journal , 108(1):37–108, 2001.[32] Wolfgang Soergel. Kazhdan-Lusztig-Polynome und unzerlegbare Bimoduln ¨uber Polynomrin-gen.
J. Inst. Math. Jussieu , 6(3):501–525, 2007.[33] J. B. Wagoner. A picture description of the boundary map in algebraic K -theory. In Algebraic K -theory, Part I (Oberwolfach, 1980) , volume 966 of Lecture Notes in Math. , pages 362–389. Springer,Berlin-New York, 1982.M
ASSACHUSETTS I NSTITUTE OF T ECHNOLOGY , B
OSTON M AX -P LANCK -I NSTITUT F ¨ UR M ATHEMATIK , B, B