Geometric combinatorial algebras: cyclohedron and simplex
aa r X i v : . [ m a t h . C O ] D ec GEOMETRIC COMBINATORIAL ALGEBRAS: CYCLOHEDRON ANDSIMPLEX
STEFAN FORCEY AND DERRIELL SPRINGFIELD
Abstract.
In this paper we report on results of our investigation into the algebraic struc-ture supported by the combinatorial geometry of the cyclohedron. Our new graded algebrastructures lie between two well known Hopf algebras: the Malvenuto-Reutenauer algebraof permutations and the Loday-Ronco algebra of binary trees. Connecting algebra mapsarise from a new generalization of the Tonks projection from the permutohedron to theassociahedron, which we discover via the viewpoint of the graph associahedra of Carr andDevadoss. At the same time, that viewpoint allows exciting geometrical insights into themultiplicative structure of the algebras involved. Extending the Tonks projection also re-veals a new graded algebra structure on the simplices. Finally this latter is extended to anew graded Hopf algebra (one-sided) with basis all the faces of the simplices. Introduction
Background: Polytope algebras.
In 1998 Loday and Ronco found an intriguingHopf algebra of planar binary trees lying between the Malvenuto-Reutenauer Hopf algebraof permutations [13] and the Solomon descent algebra of Boolean subsets [11]. They alsodescribed natural Hopf algebra maps which neatly factor the descent map from permutationsto Boolean subsets. Their first factor turns out to be the restriction (to vertices) of the Tonksprojection from the permutohedron to the associahedron. Chapoton made sense of this latterfact when he found larger Hopf algebras based on the faces of the respective polytopes [5].Here we study several new algebraic structures based on polytope sequences, including thecyclohedra, W n , and the simplices, ∆ n . In Figure 1 we show the central polytopes, in threedimensions. Figure 1.
The main characters, left to right: P , W , K , and ∆ . We will be referring to the algebras and maps discussed in [2] and [1]. In these two papers,Aguiar and Sottile make powerful use of the weak order on the symmetric groups and the
Mathematics Subject Classification.
Primary 52B11.
Key words and phrases.
Hopf algebra, graph associahedron, cyclohedron, graded algebra.
Tamari order on binary trees. By leveraging the M¨obius function of these two lattices theyprovide clear descriptions of the antipodes and of the geometric underpinnings of the Hopfalgebras. The also demonstrate cofreeness, characterize primitives of the coalgebras, and, inthe case of binary trees, demonstrate equivalence to the non-commutative Connes-KreimerHopf algebra from renormalization theory.Here we generalize the well known algebras based on associahedra and permutohedrato new ones on cyclohedra and simplices. The cyclohedra underlie graded algebras, andthe simplices underlie a new (one-sided) Hopf algebra. We leave for future investigationmany potential algebras and coalgebras based on novel sequences of graph associahedra.The phenomenon of polytopes underlying Hopf structure may be rare, but algebras andcoalgebras based on polytope faces are beginning to seem ubiquitous. There does exist alarger family of Hopf algebras to be uncovered in the structure of the polytope sequencesderived from trees, including the multiplihedra, their quotients and their covers. These arestudied in [9].1.2.
Background: Cyclohedra.
The cyclohedron W n of dimension n − n a positiveinteger was originally developed by Bott and Taubes [3], and received its name from Stasheff.The name points out the close connection to Stasheff’s associahedra, which we denote K n ofdimension n − . The former authors described the facets of W n as being indexed by subsetsof [ n ] = 1 , , . . . , n of cardinality ≥ n ( n −
1) facets. Allthe faces can be indexed by cyclic bracketings of the string 123 . . . n , where the facets haveexactly one pair of brackets. The vertices are complete bracketings, enumerated by (cid:0) n − n − (cid:1) . (123)(12)3 1(23)(231)(31)2(312)((31)2) (3(12)) ((12)3) (1(23)) (2(31)) ((23)1)123 Figure 2.
The cyclohedron W with various indexing.The space W n × S , seen as the compactification of the configuration space of n distinctpoints in R which are constrained to lie upon a given knot, is used to define new invariantswhich reflect the self linking of knots [3]. Since their inception the cyclohedra have provento be useful in many other arenas. They provide an excellent example of a right operadmodule (over the operad of associahedra) as shown in [14]. Devadoss discovered a tiling ofthe ( n − n − W n in [7]. Recently the cyclohedra have been usedto look for the statistical signature of periodically expressed genes in the study of biologicalclocks [15].The faces of the cyclohedra may also be indexed by the centrally symmetric subdivisionsof a convex polygon of 2 n sides, as discovered by Simion [19]. In this indexing the ver-tices are centrally symmetric triangulations, which allowed Hohlweg and Lange to developgeometric realizations of the cyclohedra as convex hulls [10]. This picture is related to the EOMETRIC COMBINATORIAL ALGEBRAS: CYCLOHEDRON AND SIMPLEX 3 work of Fomin, Reading and Zelevinsky, who see the cyclohedra as a generalization of theassociahedra corresponding to the B n Coxeter diagrams [18]. From their perspective the facestructure of the cyclohedron is determined by the sub-cluster structure of the generators ofa finite cluster algebra.In contrast, for Devadoss the cyclohedra arise from truncating simplex faces correspondingto sub-diagrams of the ˜ A n Coxeter diagram, or cycle graph [4]. In this paper we will workfrom the point of view taken by Devadoss and consider the faces as indexed by tubings of thecycle graph on n vertices. Given a graph G , the graph associahedron K G is a convex polytopegeneralizing the associahedron, with a face poset based on the full connected subgraphs, ortubes of G . For instance, when G is a path, a cycle, or a complete graph, K G resultsin the associahedron, cyclohedron, and permutohedron, respectively. In [8], a geometricrealization of K G is given, constructing this polytope from truncations of the simplex. In [4]the motivation for the development of K G is that it appears in tilings of minimal blow-upsof certain Coxeter complexes, which themselves are natural generalizations of the modulispaces M ,n ( R ).The value of the graph associahedron to algebraists, as we hope to demonstrate here,is twofold. First, a unified description of so many combinatorial polytopes allows usefulgeneralizations of the known algebraic structures on familiar polytope sequences. Second, therecursive structure described by Carr and Devadoss in a general way for graph associahedraturns out to lend new geometrical meaning to the graded algebra structures of both theMalvenuto-Reutenauer and Loday-Ronco algebras. The product of two vertices, from terms P i and P j of a given sequence of polytopes { P n } ∞ n =0 , is described as a sum of vertices ofthe term P i + j to which the operands are mapped. The summed vertices in the productare the images of classical inclusion maps composed with our new extensions of the Tonksprojection.1.2.1. Acknowledgements.
We would like to thank the referees for helping us make connec-tions to other work, and for making some excellent suggestions about presentation of themain ideas. We also thank the following for tutorials and helpful conversations on the subjectmatter: Satyan Devadoss, Aaron Lauve, Maria Ronco and Frank Sottile.1.3.
Summary.
Notation.
The Hopf algebras of permutations, binary trees, and Boolean subsets aredenoted respectively S Sym, Y Sym and Q Sym, as in [2] and [1]. Note that some of oursources, including Loday and Ronco’s original treatment of binary trees, actually deal withthe dual graded algebras. The larger algebras of faces of permutohedra, associahedra andcubes are denoted ˜ S Sym, ˜ Y Sym and ˜ Q Sym.
The new algebras of cyclohedra vertices andfaces are denoted W Sym and ˜ W Sym.
The new algebra of vertices of the simplices is denoted∆
Sym.
Finally the new one-sided Hopf algebra of faces of the simplices is denoted ˜∆
Sym.
Throughout there are three important maps we will use. First we see them as polytopemaps: ˆ ρ is the inclusion of facets defined by Devadoss as a generalization of the recursivedefinitions of associahedra, simplices and permutohedra; η is the projection from the polytopeof a disconnected graph to its components; and Θ is the generalization of Tonks’s projectionfrom the permutohedron to the associahedron. This last gives rise to maps ˆΘ which arealgebra homomorphisms. STEFAN FORCEY AND DERRIELL SPRINGFIELD
Main Results.
In Theorem 5.3 we demonstrate that W Sym is an associative gradedalgebra. In Theorem 6.4 we extend this structure to the full poset of faces to describe theassociative graded algebra ˜ W Sym.
In Theorem 7.3 we demonstrate an associative gradedalgebra structure on ∆
Sym.
In Theorem 7.11 we show how to extend this structure to becomea new graded (one-sided) Hopf algebra ˜∆
Sym, based upon all the faces of the simplices. Thusits graded dimension is 2 n .Theorems 4.1, 4.3 and 6.1 and Remark 5.5 point out that the multiplications in all thealgebras studied here can be understood in a unified way via the recursive structure of theface posets of associated polytopes. To summarize, the products can be seen as a process ofprojecting and including of faces, via Θ , ˆ ρ and η . The algebras based on the permutohedrondon’t use Θ or η, and the algebras based on the simplices don’t use Θ , but otherwise thedefinitions will seem quite repetitive. We treat each definition separately to highlight thosedifferences, but often the proofs of an early version will be referenced rather than repeatedverbatim. The coproducts of Y Sym and S Sym are also understandable as projections ofthe polytope geometry, as mentioned in Remarks 4.2 and 4.4.Before discussing algebraic structures, however, we build a geometric combinatorial foun-dation. We show precisely how the graph associahedra are all cellular quotients of thepermutohedra (Lemma 3.2), and how the associahedron is a quotient of any given connectedgraph associahedron (Theorem 3.5). Results similar to these latter statements are impliedby the work of Postnikov in [17], and were also reportedly known to Tonks in the casesinvolving the cyclohedron [4].The various maps arising from factors of the Tonks projection are shown to be algebra ho-momorphisms in Theorems 5.6, 5.7 and Lemma 7.5. Corollary 5.9 points out the implicationof module structures on W Sym over S Sym ; and on Y Sym over W Sym.
Overview of subsequent sections.
Section 2 describes the posets of connected sub-graphs which are realized as the graph associahedra polytopes. We also review the cartesianproduct structure of their facets. Section 3 shows how the Tonks cellular projection from thepermutohedron to the associahedron can be factored in multiple ways so that the interme-diate polytopes are graph associahedra for connected graphs. We also show that the Tonksprojection can be extended to cellular projections to the simplices, again in such a way thatthe intermediate polytopes are graph associahedra. In particular, we focus on a factoriza-tion of the Tonks projection which has the cyclohedron (cycle graph associahedra) as anintermediate quotient polytope between permutohedron and associahedron. In Section 4 weuse the viewpoint of graph associahedra to redescribe the products in S Sym and Y Sym ,and to point out their geometric interpretation. The latter is based upon our new cellularprojections as well as classical inclusions of (cartesian products of) low dimensional associ-ahedra and permutohedra as faces in the higher dimensional polytopes of their respectivesequences. In Section 5 we begin our exploration of new graded algebras with the vertices ofthe cyclohedron. Then we show that the linear projections following from the factored Tonksprojection (restricted to vertices) are algebra maps. We extend these findings in Section 6to the full algebras based on all the faces of the polytopes involved. Finally in Section 7 wegeneralize our discoveries to the sequence of edgeless graph associahedra. This allows us tobuild a graded algebra based upon the vertices of the simplices. By carefully extending thatstructure to the faces of the simplices we find a new graded (one-sided) Hopf algebra, withgraded dimension 2 n . EOMETRIC COMBINATORIAL ALGEBRAS: CYCLOHEDRON AND SIMPLEX 5 Review of some geometric combinatorics
We begin with definitions of graph associahedra; the reader is encouraged to see [4, Section1] and [6] for further details.
Definition 2.1.
Let G be a finite connected simple graph. A tube is a set of nodes of G whose induced graph is a connected subgraph of G . Two tubes u and v may interact on thegraph as follows:(1) Tubes are nested if u ⊂ v .(2) Tubes are far apart if u ∪ v is not a tube in G, that is, the induced subgraph of theunion is not connected, or none of the nodes of u are adjacent to a node of v .Tubes are compatible if they are either nested or far apart. We call G itself the universaltube . A tubing U of G is a set of tubes of G such that every pair of tubes in U is compatible;moreover, we force every tubing of G to contain (by default) its universal tube. By the term k - tubing we refer to a tubing made up of k tubes, for k ∈ { , . . . , n } . When G is a disconnected graph with connected components G , . . . , G k , an additionalcondition is needed: If u i is the tube of G whose induced graph is G i , then any tubing of G cannot contain all of the tubes { u , . . . , u k } . However, the universal tube is still includeddespite being disconnected. Parts (a)-(c) of Figure 3 from [6] show examples of allowabletubings, whereas (d)-(f) depict the forbidden ones. ( a ) ( b ) ( c ) ( d ) ( e ) ( f ) Figure 3. (a)-(c) Allowable tubings and (d)-(f) forbidden tubings, figure from [6].
Theorem 2.2. [4, Section 3]
For a graph G with n nodes, the graph associahedron K G isa simple, convex polytope of dimension n − whose face poset is isomorphic to the set oftubings of G , ordered by the relationship U ≺ U ′ if U is obtained from U ′ by adding tubes. The vertices of the graph associahedron are the n -tubings of G. Faces of dimension k areindexed by ( n − k )-tubings of G. In fact, the barycentric subdivision of K G is precisely thegeometric realization of the described poset of tubings. Many of the face vectors of graphassociahedra for path-like graphs have been found, as shown in [16]. This source also containsthe face vectors for the cyclohedra. There are many open questions regarding formulas forthe face vectors of graph associahedra for specific types of graphs. Example 2.3.
Figure 4, partly from [6], shows two examples of graph associahedra. Thesehave underlying graphs a path and a disconnected graph, respectively, with three nodes each.These turn out to be the 2 dimensional associahedron and a square. The case of a threenode complete graph, which is both the cyclohedron and the permutohedron, is shown inFigure 2.To describe the face structure of the graph associahedra we need a definition from [4,Section 2].
Definition 2.4.
For graph G and a collection of nodes t , construct a new graph G ∗ ( t ) calledthe reconnected complement : If V is the set of nodes of G , then V − t is the set of nodes of STEFAN FORCEY AND DERRIELL SPRINGFIELD
Figure 4.
Graph associahedra of a path and a disconnected graph. The 3-cube is found as the graph-associahedron of two disjoint edges on four nodes,but no simple graph yields the 4-cube. G ∗ ( t ). There is an edge between nodes a and b in G ∗ ( t ) if { a, b } ∪ t ′ is connected in G forsome t ′ ⊆ t . Example 2.5.
Figure 5 illustrates some examples of graphs along with their reconnectedcomplements.
Figure 5.
Examples of tubes and their reconnected complements.For a given tube t and a graph G , let G ( t ) denote the induced subgraph on the graph G .By abuse of notation, we sometimes refer to G ( t ) as a tube. Theorem 2.6. [4, Theorem 2.9]
Let V be a facet of K G, that is, a face of dimension n − of K G , where G has n nodes. Let t be the single, non-universal, tube of V . The face posetof V is isomorphic to K G ( t ) × K G ∗ ( t ) . A pair of examples is shown in Figure 6. The isomorphism described in [4, Theorem 2.9] iscalled ˆ ρ. Since we will be using this isomorphism more than once as an embedding of faces inpolytopes, then we will specify it, according to the tube involved, as ˆ ρ t : K G ( t ) × K G ∗ ( t ) ֒ →K G. Given a pair of tubings from T ∈ K G t and T ′ ∈ K G ∗ ( t ) the image ˆ ρ ( T, T ′ ) consists ofall of T and an expanded version of T ′ . In the latter each tube of T ′ is expanded by takingits union with t if that union is itself a tube. A specific example of the action of ˆ ρ t is inFigure 7. In fact, often there will be more than one tube involved, as we now indicate: Corollary 2.7.
Let { t , . . . , t k , G } be an explicit tubing of G , such that each pair of non-universal tubes in the list is far apart. Then the face of K G associated to that tubing isisomorphic to K G ( t ) × · · · × K G ( t n ) × K G ∗ ( t ∪ · · · ∪ t k ) . EOMETRIC COMBINATORIAL ALGEBRAS: CYCLOHEDRON AND SIMPLEX 7
We will denote the embedding by:ˆ ρ t ...t k : K G ( t ) × · · · × K G ( t n ) × K G ∗ ( t ∪ · · · ∪ t k ) ֒ → K G. Proof.
This follows directly from Theorem 2.6. Note that the reconnected complement withrespect to the union of several tubes t , . . . , t k is the same as taking successive iteratedreconnected complements with respect to each tube in the list. That is, G ∗ ( t ∪ · · · ∪ t k ) = ( . . . (( G ∗ ( t )) ∗ ( t )) ∗ . . . ) ∗ ( t k ) . (cid:3) K W K W Figure 6.
Two facets of the cyclohedron W . Remark 2.8.
As shown in [8], the graph associahedron of the graph consisting of n nodesand no edges is the ( n − ( n − . Thus the graph associahedron of a graph withconnected components G , G , . . . , G k is actually equivalent to the polytope K G × · · · ×K G k × ∆ ( k − . This equivalence implies that:
Lemma 2.9.
For a disconnected graph G with multiple connected components G , G , . . . , G k ,there is always a cellular surjection η : K G → K G × · · · × K G k . An example is in Figure 7. , ( ( η ρ Figure 7.
Example of the action of η and ˆ ρ t , where t is the cycle sub-graphof the final graph. STEFAN FORCEY AND DERRIELL SPRINGFIELD Factoring and extending the Tonks projection.
Loday and Ronco’s Hopf algebra map.
The two most important existing math-ematical structures we will use in this paper are the graded Hopf algebra of permutations, S Sym , and the graded Hopf algebra of binary trees, Y Sym . The n th component of S Sym has basis the symmetric group S n , with number of elements counted by n !. The n th compo-nent of Y Sym has basis the collection of binary trees with n interior nodes, and thus n + 1leaves, denoted Y n . These are counted by the Catalan numbers.The connection discovered by Loday and Ronco between the two algebras is due to thefact that a permutation on n elements can be pictured as a binary tree with n interior nodes,drawn so that the interior nodes are at n different vertical heights from the base of the tree.This is called an ordered binary tree . The interior nodes are numbered left to right. Wenumber the leaves 0 , , . . . , n − , , . . . , n − . The i th node is “between”leaf i − i where “between” might be described to mean that a rain drop fallingbetween those leaves would be caught at that node. Distinct vertical levels of the nodes arenumbered top to bottom. Then for a permutation σ ∈ S n the corresponding tree has node i at level σ ( i ) . The map from permutations to binary trees is achieved by forgetting the levels,and just retaining the tree. This classical surjection is denoted τ : S n → Y n . An example isin Figure 8.By a cellular surjection of polytopes P and Q we refer to a map f from the face posetof P to that of Q which is onto and which preserves the poset structure. That is, if x is asub-face of y in P then f ( x ) is a sub-face of or equal to f ( y ) . A cellular projection is a cellularsurjection which also has the property that the dimension of f ( x ) is less than or equal to thedimension of x. In [20] Tonks extended τ to a cellular projection from the permutohedronto the associahedron: Θ : P n → K n . In the cellular projection a face of the permutohedron,which is leveled tree, is taken to its underlying tree, which is a face of the associahedron.Figure 9 shows an example. The new revelation of Loday and Ronco is that the map τ givesrise to a Hopf algebraic projection τ : S Sym → Y
Sym , so that the algebra of binary treesis seen to be embedded in the algebra of permutations.
12 34 1 2 34
Figure 8.
The permutation σ = (2431) ∈ S pictured as an ordered treeand as a tubing of the complete graph; the unordered binary tree, and itscorresponding tubing.3.2. Tubings, permutations, and trees.
Our new approach to the Tonks projection ismade possible by the recent discovery of Devadoss in [8] that the graph-associahedron of thecomplete graph on n vertices is precisely the n th permutohedron P n . Each of its verticescorresponds to a permutation of n elements. Its faces in general correspond to ordered EOMETRIC COMBINATORIAL ALGEBRAS: CYCLOHEDRON AND SIMPLEX 9 partitions of [ n ] . Keep in mind that for a permutation σ ∈ S n , the corresponding orderedpartition of [ n ] is ( { σ − (1) } , { σ − (2) } , . . . { σ − ( n ) } ) . Here is how to describe a bijection from the n -tubings on the complete graph to thepermutations, as found by Devadoss in [8]. First a numbering of the n nodes must bechosen. Given an n -tubing, since the tubes are all nested we can number them starting withthe innermost tube. Then the permutation σ ∈ S n pictured by our n -tubing is such thatnode i is within tube σ ( i ) but not within any tube nested inside of tube σ ( i ) . Figure 8 showsan example.It is easy to extend this bijection between n -tubings and permutations to all tubings ofthe complete graph and ordered partitions of [ n ]. Given a k -tubing of the complete graph,each tube contains some numbered nodes which are not contained in any other tube. Thesesubsets of [ n ] , one for each tube, make up the k -partition, and the ordering of the partitionis from innermost to outermost. Recall that an ordered k -partition of [ n ] corresponds to aleveled tree with n + 1 leaves and k levels, numbered top to bottom, at which lie the internalnodes. Numbering the n spaces between leaves from left to right (imagine a raindrop fallinginto each space), we see that the raindrops collecting at the internal nodes at level i representthe i th subset in the partition. We denote our bijection by: f : { leveled trees with n leaves } → K (complete graph on n − . Figure 9 shows an example.
Figure 9.
The ordered partition ( { , , } , { } ) pictured as a leveled tree andas a tubing of the complete graph; the underlying tree, and its correspondingtubing.The binary trees with n + 1 leaves (and n internal nodes) correspond to the vertices ofthe ( n − K n . In the world of graph-associahedra, these vertices correspond to the n -tubings of the path graph on n nodes. Carrand Devadoss realized that in fact the path graph associahedron is precisely the Stasheffassociahedron [4]. Thus for any tree with n + 1 leaves we can bijectively determine a tubingon the path graph. This correspondence can be described by creating a tube of the pathgraph for each internal node of the tree. We number the leaves from left to right 0 , , . . . , n and the nodes of the path from left to right 1 , . . . , n. The tube we create contains the samenumbered nodes of the path graph as all but the leftmost leaf of the subtree determined bythe internal node. This bijection we denote by: g : { trees with n leaves } → K (path on n − . Figures 8 and 9 show examples.
Generalizing the Tonks projection.
The fact that every graph of n nodes is asubgraph of the complete graph leads us to a grand factorization of the Tonks cellularprojection through all connected graph associahedra. Incomplete graphs are formed simplyby removing edges from the complete graph. As a single edge is deleted, the tubing ispreserved up to connection. That is, if the nodes of a tube are no longer connected, itbecomes two tubes made up of the two connected subgraphs spanned by its original set ofnodes. Definition 3.1.
Let G be a graph on n nodes, and let e be an edge of G . Let G − e denotethe graph achieved be the deletion of e, while retaining all n nodes. We define a cellularprojection Θ e : K G ։ K ( G − e ) . First, allowing an abuse of notation, we define Θ e onindividual tubes. For t a tube of G such that t is not a tube of G − e, then let t ′ , t ′′ be thetubes of G − e such that t ′ ∪ t ′′ = t. Then:Θ e ( t ) = ( { t } , t a tube of G − e { t ′ , t ′′ } , otherwise. Now given a tubing T on G we define its image as follows:Θ e ( T ) = [ t ∈ T Θ e ( t ) . See Figures 10, 11 and 12 for examples.
123 4 1 2 3 4123 4 123 4
Figure 10.
The Tonks projection performed on (1243), factored by graphs.
Lemma 3.2.
For a graph G with an edge e , Θ e is a cellular surjection of polytopes K G ։ K ( G − e ) . Proof.
By Theorem 2.2 we have that the face posets of the polytopes are isomorphic to theposets of tubings. The map takes a tubing on G to a tubing on G − e with a greater or equalnumber of tubes. This establishes its projective property. Also, for two tubings U ≺ U ′ of G we see that either Θ e ( U ) = Θ e ( U ′ ) or Θ e ( U ) ≺ Θ e ( U ′ ) . Thus the poset structure is preservedby the projection.Finally, the map is surjective, since given any tubing T on G − e we can find a (maximal)preimage T ′ as follows: First consider all the tubes of T as a candidate tubing of G. If it is avalid tubing, we have our T ′ . If not, then there must be a pair of tubes t ′ , t ′′ ∈ T which areadjacent via the edge e, and for which there are no tubes containing one of t ′ , t ′′ but not theother. Then let T be the result of replacing that pair in T with the single tube t = t ′ ⊔ t ′′ . If T is a valid tubing of G then let T ′ = T . If not, continue inductively. (cid:3)
Composition of these cellular projections is commutative.
Lemma 3.3.
Let e, e ′ be edges of G. Then Θ e ◦ Θ e ′ = Θ e ′ ◦ Θ e . EOMETRIC COMBINATORIAL ALGEBRAS: CYCLOHEDRON AND SIMPLEX 11
Proof.
Consider the image of a tubing of G under either composition. A tube of G that is atube of both G − e and G − e ′ will persist in the image. Otherwise it will be broken, perhapstwice. The same smaller tubes will result regardless of the order of the breaking. (cid:3) By Lemma 3.3 we can unambiguously use the following notation:
Definition 3.4.
For any collection E of edges of G we denote by Θ E : K G ։ K ( G − E ) thecomposition of projections { Θ e | e ∈ E } . Now the Tonks projection from leveled trees to trees can be described in terms of thetubings. Beginning with a tubing on the complete graph with numbered nodes, we achieveone on the path graph by deleting all the edges of the complete graph except for thoseconnecting the nodes in consecutive order from 1 to n . See Figures 8, 9 and 10 for a pictureto accompany the following: Theorem 3.5.
Let e i,k be the edge between the nodes i, i + k of the complete graph G on n nodes. Let P = { e i,k | i ∈ { , . . . , n − } and k ∈ { , . . . , n − i }} . These are all but the edgesof the path graph. Then the following composition gives the Tonks map: g − ◦ Θ P ◦ f = Θ . Proof.
We begin with an ordered j -partition of n drawn as a leveled tree t with n + 1 leavesnumbered left to right, and j levels numbered top to bottom. The bijection f tells us howto draw t as a tubing of the complete graph K n with numbered nodes.Consider the internal nodes of t at level i . In f ( t ) there corresponds a tube u i containingthe nodes of K n with the same numbers as all the spaces between leaves of the subtreesrooted at the internal nodes at level i. The set in the partition which is represented by theinternal nodes at level i is the precise set of nodes of u i which are not contained by any othertube in f ( t ) . The relative position of this subset of [ n ] in our ordered partition is reflectedby the relative nesting of the tube u i .The map Θ P has the following action on tubings. Let u be a tube of f ( t ) . We partition u into subsets of consecutively numbered nodes such that no union of two subsets is consecu-tively numbered. Then the tubing Θ P ( f ( t )) contains the tubes given by these subsets.Now we claim that the tree g − (Θ P ( f ( t ))) has the same branching structure as t itself.First, for any interior node of t there is a tube of consecutively numbered nodes of the pathgraph, arising from the original tube of f ( t ). This then becomes a corresponding interiornode of our tree, under the action of g − . Secondly, if any interior node of t lies between theroot of t and a second interior node, then the same relation holds in the tree between thecorresponding interior nodes. This follows from the fact that the action of any of our mapsΘ e i,k will preserve relative nesting. That is, for two tubes u ⊂ v we have that in the imageof Θ e i,k any tube resulting from u must lie within a tube resulting from v. (cid:3) To sum up, there is a factorization of the Tonks cellular projection through various graph-associahedra. An example on an n -tubing is shown in Figure 10, and another possiblefactorization of the projection in dimension 3 is demonstrated in Figure 11.3.4. Disconnected graph associahedra.
The special case of extending the Tonks pro-jection to graphs with multiple connected components will be useful. Consider a partition S ⊔ · · · ⊔ S k of the n nodes of a connected graph G , chosen such that we have connectedinduced subgraphs G ( S i ) . Let E S be the set of edges of G not included in any G ( S i ) . Thusthe graph G − E S formed by deleting these edges will have the k connected components Θ Θ Θ
Figure 11.
A factorization of the Tonks projection through 3 dimensionalgraph associahedra. The shaded facets correspond to the shown tubings, andare collapsed as indicated to respective edges. The first, third and fourthpictured polytopes are above views of P , W and K respectively. G ( S i ) . In this situation Θ E S will be a generalization of the Tonks projection to the graphassociahedron of a disconnected graph.In Figure 12 we show the extended Tonks projections in dimension 2. Θ Θ Θ
Figure 12.
Extending the Tonks projection to the 2-simplex. The high-lighted edges are collapsed to the respective vertices.4.
Geometrical view of S Sym and Y Sym
Before proving the graded algebra structures on graph associahedra which our title promised,we motivate our point of view by showing how it will fit with the well known graded algebrastructures on permutations and binary trees.4.1.
Review of S Sym . Let S Sym be the graded vector space over Q with the n th com-ponent of its basis given by the permutations S n . An element σ ∈ S n is given by its image( σ (1) , . . . , σ ( n )), often without commas. We follow [1] and [2] and write F u for the basiselement corresponding to u ∈ S n and 1 for the basis element of degree 0. A graded Hopfalgebra structure on S Sym was discovered by Malvenuto and Reutenauer in [13]. First wereview the product and coproduct and then show a new way to picture those operations.Recall that a permutation σ is said to have a descent at location p if σ ( p ) > σ ( p + 1) . The ( p, q )-shuffles of S p + q are the ( p + q ) permutations with at most one descent, at position p. We denote this set as S ( p,q ) . The product in S Sym of two basis elements F u and F v for u ∈ S p and v ∈ S q is found by summing a term for each shuffle, created by composing the EOMETRIC COMBINATORIAL ALGEBRAS: CYCLOHEDRON AND SIMPLEX 13 juxtaposition of u and v with the inverse shuffle: F u · F v = X ι ∈ S ( p,q ) F ( u × v ) · ι − . Here u × v is the permutation ( u (1) , . . . , u ( p ) , v (1) + p, . . . , v ( q ) + p ). Geometry of S Sym . The algebraic structure of S Sym can be linked explicitly to therecursive geometric structure of the permutohedra. In S Sym we may view our operands (apair of permutations) as a vertex of the cartesian product of permutohedra P p × P q . Thentheir product is the formal sum of images of that vertex under the collection of inclusions of P p × P q as a facet of P p + q . An example is in Figure 13, where the product is shown alongwith corresponding pictures of the tubings on the complete graphs. To make this geometricclaim precise we use the facet isomorphism ˆ ρ t which exists by Theorem 2.6. Theorem 4.1.
The product in S Sym of basis elements F u and F v for u ∈ S p and v ∈ S q may be written: F u · F v = X ι ∈ S ( p,q ) F ˆ ρ ι ( u,v ) where ˆ ρ ι is shorthand for ˆ ρ ι ([ p ]) . (2 1)(1 2) (1 2 4 3) (1 4 2 3) (1 4 3 2) (4 1 2 3) (4 1 3 2) (4 3 1 2) (1 2) (1 2)(2 1) (2 1) P = P P P P
12 34
12 34 =
12 34 12 34 + ++
12 34 12 34 + + ρ Figure 13.
The product in S Sym . Here (and similarly in all our pictorialexamples) we let the picture of the graph tubing u stand in for the basiselement F u . In the picture of P the circled vertices from bottom to top arethe permutations in the product, as listed from left to right above. Proof.
From Theorem 2.6 we have that for each tube t of the graph, the corresponding facetof K G is isomorphic to K G ∗ ( t ) × K G ( t ) . In the case of the complete graph G = K p + q , forany tube t of p nodes, we have the facet inclusion ˆ ρ t : P q × P p → P p + q . We just need to review the definition of the isomorphism ˆ ρ from the proof of [4, Theorem2.9], and point out that the permutation associated to the tubing ˆ ρ ι ( u, v ) is indeed ( u × v ) · ι − . Given a shuffle ι , the ( p + q )-tubing ˆ ρ ι ( u, v ) on K p + q is given by including for each tube t of u the tube with nodes the image ι ( t ) . Denote the resulting set of tubes by ι ( u ) . For eachtube s of v we include the tube formed by ι ([ p ]) ∪ ˆ ι ( s ) , where ˆ ι ( i ) = ι ( i + p ) . We denote theresulting set of tubes as ˆ ι ( v ) . Now the tubing ˆ ρ ι ( u, v ) is defined to be ι ( u ) ∪ ˆ ι ( v ) . This tubing is precisely the completetubing which represents the permutation ( u × v ) · ι − . (cid:3) To sum up, in our view of S Sym each permutation is pictured as a tubing of a completegraph with numbered nodes. Since any subset of the nodes of a complete graph spans asmaller complete graph, we can draw the terms of the product in S Sym directly. Choosinga ( p, q )-shuffle is accomplished by choosing p nodes of the ( p + q )-node complete graph K p + q . First the permutation u (as a p -tubing) is drawn upon the induced p -node completesubgraph, according to the ascending order of the chosen nodes. Then the permutation v isdrawn upon the subgraph induced by the remaining q nodes–with a caveat. Each of the tubesof v is expanded to also include the p nodes that were originally chosen. This perspectivewill generalize nicely to the other graphs and their graph associahedra.In [1] and [2] the authors give a related geometric interpretation of the products of S Sym and Y Sym as expressed in the M¨obius basis. An interesting project for the future would beto apply that point of view to W Sym.
Remark 4.2.
The coproduct of S Sym can also be described geometrically in terms of thegraph tubings. The coproduct is usually described as a sum of all the ways of splitting apermutation u ∈ S n into two permutations u i ∈ S i and u n − i ∈ S n − i : ∆( F u ) = n X i =0 F u i ⊗ F u n − i where u i = ( u (1) . . . u ( i )) and u n − i = ( u ( i + 1) − i . . . u ( n ) − i ) . Given an n -tubing u of the complete graph on n vertices we can find u i and u n − i just byrestricting to the sub graphs (also complete) induced by the nodes 1 , . . . , i and i + 1 , . . . , n respectively. For each tube t ∈ u the two intersections of t with the respective subgraphs areincluded in the respective tubings u i and u n − i . An example is in Figure 14.Notice that this restriction of the tubings to subgraphs is the same as the result of per-forming the Tonks projection. Technically, ( u i , u n − i ) = η (Θ E i ( u )) , where E i = { e an edge of G | e connects a node j ≤ i to a node j ′ > i } . and η is from Lemma 2.9. ∆ = + + + + Figure 14.
The coproduct in S Sym. ∆ F (1342) = 1 ⊗ F (1342) + F (1) ⊗ F (231) + F (12) ⊗ F (21) + F (123) ⊗ F (1) + F (1342) ⊗ . EOMETRIC COMBINATORIAL ALGEBRAS: CYCLOHEDRON AND SIMPLEX 15
Review of Y Sym . The product and coproduct of Y Sym are described by Aguiar andSottile in terms of splitting and grafting binary trees [2]. We can vertically split a tree intosmaller trees at each leaf from top to bottom–as if a lightning strike hits a leaf and splits thetree along the path to the root. We graft trees by attaching roots to leaves (without creatinga new interior node at the graft.) The product of two trees with n and m interior nodes(in that order) is a sum of (cid:0) n + mn (cid:1) terms, each with n + m interior nodes. Each is achievedby vertically splitting the first tree into m + 1 smaller trees and then grafting them to thesecond tree, left to right. A picture is in Figure 15.4.4. Geometry of Y Sym . Since Loday and Ronco demonstrated that the Tonks projection,restricted to vertices, gives rise to an algebra homomorphism τ : S Sym → Y
Sym, itis no surprise that the processes of splitting and grafting have geometric interpretations.Grafting corresponds to certain face inclusions of associahedra, and splitting correspondsto the extension of Tonks’s projection to disconnected graphs. To see the latter, note thatsplitting at leaf i is the same as deleting the edge from node i to node i + 1 . Thus the productin Y Sym can be described with the language of path graph tubings, using a combination offacet inclusion and the extended Tonks projection. An example is shown in Figure 15.Let U be the path graph with p -tubing u and V the path graph with q -tubing v. Givena shuffle ι ∈ S ( p,q ) we can partition the nodes of U into the preimages s , . . . , s k of theconnected components ι ( s ) , . . . , ι ( s k ) of the possibly disconnected subgraph induced by thenodes ι ([ p ]) on the ( p + q )-path. Let E ι be the edges of U not included in the subgraphs U ( s i )induced by our partition. For short we denote the extended Tonks projection Θ E ι as simplyΘ ι . Now Θ ι ( u ) is the projection of u onto the possibly disconnected graph U − E ι . Recall fromLemma 2.9 that a vertex of K ( U − E ι ) may be mapped to a vertex of K U ( s ) × · · · × K U ( s k ) . We call this map η ι . Theorem 4.3.
The product in Y Sym can be written: F u · F v = X ι ∈ S ( p,q ) F ˆ ρ ι ( η ι (Θ ι ( u )) ,v ) . Where ˆ ρ ι is shorthand notation for the isomorphism ˆ ρ ι ( s ) ...ι ( s k ) from Corollary 2.7.Proof. We will explain how the splitting and grafting of trees in a term of the product may beput into the language of tubings, and then argue that the term thus described is indeed theimage of projections and inclusions as claimed. The product in Y Sym of two basis elements F u and F v for u ∈ Y p and v ∈ Y q is found by summing a term for each shuffle ι ∈ S ( p,q ) . Wedraw u and v in the form of tubings on path graphs of p and q nodes, respectively.Here is the non-technical description: first the p -tubing u is drawn upon the inducedsubgraph of the nodes ι ([ p ]) according to the ascending order of the chosen nodes. However,each tube may need to be first broken into several sub-tubes. Then the q -tubing v is drawnupon the subgraph induced by the remaining q nodes. In this last step, each of the tubes of v is expanded to also include any of the previously drawn tubes that its nodes are adjacentto.To be precise, we first choose a shuffle ι ∈ S ( p,q ) . Let ˆ ι ( i ) = ι ( i + p ) . Our term of F u · F v isthe ( p + q )-tubing on the ( p + q )-path given by the following:First for each tube t ∈ u we include in our new tubing the tubes which are the connectedcomponents of the subgraph induced by the nodes ι ( t ) . Let ι ( u ) denote the tubing constructedthus far. After this step in terms of trees, we have performed the splitting and chosen where K = K K K K
12 34
12 34 =
12 34 12 34 + ++
12 34 12 34 + + K K K K K K = η Θ ρ Figure 15.
The product in Y Sym . The circled vertices of K which are atthe upper end of highlighted edges are the fifth, third and second terms of theproduct, in that order respectively from bottom to top in the picture.to graft the (non-trivial) subtrees. In terms of trees the splitting occurs at leaves labeled byˆ ι ( i ) − i for i ∈ [ q ] . Second, for each tube s ∈ v we include in our term of the product the tube formed byˆ ι ( s ) ∪ { j ∈ t ′ ∈ ι ( u ) | t ′ is adjacent to ˆ ι ( s ) } . Let ˆ ι ( v ) denote the tubes added in this second step. Now we have completed the graftingoperation.Now we just point out that ˆ ρ ι ( η ι (Θ ι ( u )) , v ) is precisely the same as ι ( u ) ∪ ˆ ι ( v ) . Thesplitting of tubes of u is accomplished by Θ ι ; and η ι simply recasts the result as an elementof the appropriate cartesian product. Then ˆ ρ ι , as defined in [4], performs the inclusion ofthat element paired together with v. (cid:3) To summarize, the product in Y Sym can be seen as a process of splitting and grafting, orequivalently of projecting and including. From the latter viewpoint, we see the product oftwo associahedra vertices being achieved by projecting the first onto a cartesian product ofsmaller associahedra, and then mapping that result paired with the second operand into alarge associahedron via face inclusion. Notice that the reason the second path graph tubing v can be input here is that any reconnected complement of a path graph is another pathgraph. Remark 4.4.
The coproduct of Y Sym can also be described geometrically in terms of thegraph tubings. The coproduct is usually described as a sum of all the ways of splitting a
EOMETRIC COMBINATORIAL ALGEBRAS: CYCLOHEDRON AND SIMPLEX 17 binary tree u ∈ Y n along leaf i into two trees: u i ∈ Y i and u n − i ∈ Y n − i : ∆( F u ) = n X i =0 F u i ⊗ F u n − i . Given an n -tubing u of the path graph on n vertices we can find u i and u n − i just byrestricting to the sub graphs (also paths) induced by the nodes 1 , . . . , i and i + 1 , . . . , n respectively. For each tube t ∈ u the two intersections of t with the respective subgraphs areincluded in the respective tubings u i and u n − i . An example is in Figure 16.Notice that this restriction of the tubings to subgraphs is the same as the result of per-forming the extended Tonks projection, just as described in Remark 4.2. ∆ = + + + + Figure 16.
The coproduct in Y Sym. The algebra of the vertices of cyclohedra
Recall that the ( n − W n has vertices which are indexed by n -tubings on the cycle graph of n -nodes. We will define a graded algebra with a basis whichcorresponds to the vertices of the cyclohedra, and whose grading respects the dimension(plus one) of the cyclohedra. Definition 5.1.
Let W Sym be the graded vector space over Q with the n th component ofits basis given by the n -tubings on the cycle graph of n cyclically numbered nodes. By W n we denote the set of n -tubings on the cycle graph C n . We write F u for the basis elementcorresponding to u ∈ W n and 1 for the basis element of degree 0.5.1. Graded algebra structure.
Now we demonstrate a product which respects the grad-ing on W Sym by following the example described above for S Sym.
The product in W Sym of two basis elements F u and F v for u ∈ W p and v ∈ W q is found by summing a term foreach shuffle ι ∈ S ( p,q ) . First the p -tubing u is drawn upon the induced subgraph of the nodes ι ([ p ]) according to the ascending order of the chosen nodes. However, each tube may need tobe first broken into several sub-tubes, since the induced graph on the nodes ι ([ p ]) may haveconnected components ι ( s ) , . . . , ι ( s k ) (as described in Section 4.4). Then the q -tubing v isdrawn upon the subgraph induced by the remaining q nodes. However, each of the tubes of v is expanded to also include any of the previously drawn tubes that its nodes are adjacentto. Definition 5.2. F u · F v = X ι ∈ S ( p,q ) F ˆ ρ ι ( η ι (Θ ι ( u )) ,v ) . Where ˆ ρ ι is shorthand notation for the isomorphism ˆ ρ ι ( s ) ...ι ( s k ) from Corollary 2.7. Also 1is a two-sided unit. An example is shown in Figure 17. For an example of finding a term in the product givena specific shuffle ι, see the second part of Figure 19. Theorem 5.3.
The product we have just described makes W Sym into an associative gradedalgebra.Proof.
Given a shuffle ι we will use the fact thatˆ ρ ι ( η ι (Θ ι ( u )) , v ) = ι ( u ) ∪ ˆ ι ( v )as defined in the proof of Theorem 4.3. First we must check that the result of the productis indeed a sum of valid ( p + q )-tubings of the cycle graph C p + q . We claim that in each termthe new tubes we have created are pairwise compatible. This being shown, we will be ableto deduce that since p + q tubes were used in the construction, then the resulting term willnecessarily have p + q tubes as well. To check our claim we compare pairs of tubes in one orboth of ι ( u ) and ˆ ι ( v ) . There are six cases:(1) By our method of construction, any tube of ι ( u ) is either nested within some of orfar apart from all of the tubes of ˆ ι ( v ) . (2) If two tubes of ι ( u ) are made up of nodes in ι ( t ) and ι ( t ′ ) respectively for nestedtubes t and t ′ of u then they will be similarly nested.(3) Two tubes from ι ( u ) might both be made up of nodes in ι ( t ) for a single tube t of u. In that case they are guaranteed to be far apart, since their respective nodes togethercannot be a consecutive string (mod p + q ).(4) If two tubes of ι ( u ) are made up of nodes in ι ( t ) and ι ( t ′ ) respectively for far aparttubes t and t ′ of u, then we claim they will be far apart. This is true since if twonodes a and b are nonadjacent in the cycle graph C p then the two nodes ι ( a ) and ι ( b )will be nonadjacent in C p + q . (5) If two tubes of ˆ ι ( v ) contain some nodes in ˆ ι ( t ) and ˆ ι ( t ′ ) respectively for nested tubes t and t ′ of u then they will be similarly nested. This follows from the fact that ι ( t )will only be adjacent to nodes that ι ( t ′ ) is also adjacent to.(6) Finally, if two tubes of ˆ ι ( v ) contain some nodes in ˆ ι ( t ) and ˆ ι ( t ′ ) respectively for farapart tubes t and t ′ of u, then we claim they will be far apart. This final case dependson a special property which the cycle graphs exemplify. Given any subset of k of thenodes of C n , the reconnected complement of that subset is the cycle graph C n − k . Specifically the reconnected complement of C p + q with respect to the nodes ι ( p ) is thegraph C q . Thus even the expanded tubes of ˆ ι ( v ) remain far apart as long as theircomponents from ˆ ι ( q ) were far apart; and this last property is guaranteed since ι preserves the cyclic order.Thus we have shown that the result of multiplying two basis elements is again a basis element,and that this multiplication respects the grading. That this multiplication is associative is acorollary of the following result regarding how the multiplication is preserved under a mapfrom S Sym, specifically a corollary of Theorem 5.6. (cid:3)
Remark 5.4.
Note that the cases (1) and (2) are explainable simply by the fact that westart with two valid tubings and multiply them in the given order. Cases (3)-(6) however canbe jointly explained based upon the fact that given any subset of k of the nodes of C n , thereconnected complement of that subset is the cycle graph C n − k . Cases (3)-(5) specifically relyon the fact that the reconnected complement of C p + q with respect to the nodes ˆ ι ( q ) is thegraph C p . This property is true of many other graph sequences, including the complete graphs
EOMETRIC COMBINATORIAL ALGEBRAS: CYCLOHEDRON AND SIMPLEX 19 = + + . . . + Figure 17.
A product of cycle graph tubings (10 terms total).and the path graphs. The property is also more simply stated as follows: the reconnectedcomplement of G i with respect to any single node is G i − . Remark 5.5.
Once again we can interpret the product geometrically. The entire contents ofthe proof of Theorem 4.3 apply here, with the term “path” everywhere replaced with the term“cycle.” Thus a term in the product can be seen as first projecting the cyclohedron vertex u onto a collection of sub-path-graphs of the cycle. We then map the vertex of this cartesianproduct of associahedra, paired with the second vertex of the cyclohedron represented by v, into the large cyclohedron via the indicated face inclusion. The usual picture is in Figure 18. W
12 34
12 34 =
12 34 12 34 + ++
12 34 12 34 + + W = W W W W K K K K W = η Θ ρ Figure 18.
The product in W Sym . The second and fifth terms of the prod-uct are the ones that use a Tonks projection; they are found as the two verticesat the tips of included (highlighted) edges.5.2.
Algebra homomorphisms.
Next we consider the map from the permutohedra to thecyclohedra which is described via the deletion of edges–from the complete graphs to the cyclegraphs–and point out that this is an algebra homomorphism. Recall from Theorem 3.5 thatΘ P is the Tonks projection viewed from the graph associahedra point of view; via deletionof the edges of the complete graph except for the path connecting the numbered nodes inorder. Let Θ c be the map defined just as Θ P but without deleting the edge from node n to node 1 . Thus we will be deleting all the edges except those making up the cycle of numberednodes in cyclic order. Define a map from S Sym to W Sym on basis elements by:ˆΘ c ( F u ) = F Θ c ( u ) . Theorem 5.6.
The map ˆΘ c is an algebra homomorphism from S Sym onto W Sym.
Proof.
For u ∈ S p and v ∈ S q we compare ˆΘ c ( F u · F v ) with ˆΘ c ( F u ) · ˆΘ c ( F v ) . Each of themultiplications results in a sum of (cid:0) p + qp (cid:1) terms. It turns out that comparing the results ofthe two operations can be done piecewise. Thus we check that for a given shuffle ι ∈ S ( p,q ) the respective terms of our two operations agree: we claim thatΘ c ( ˆ ρ ι ( η ι (Θ ι ( u )) , v )) = ˆ ρ ι ( η ι (Θ ι (Θ c ( u ))) , Θ c ( v ))or equivalently: Θ c ( ι ( u ) ∪ ˆ ι ( v )) = ι (Θ c ( u )) ∪ ˆ ι (Θ c ( v )) . Here ι ( u ) and ˆ ι ( v ) are as described in the proof of Theorem 4.3, and the righthand side ofthe equation is using the notation of Definition 5.2. The justification is a straightforwardcomparison of the indicated operations on individual tubes. There are two cases:(1) For t a tube of u , the right-hand side first breaks t into several smaller tubes bydeleting certain edges of the p -node complete graph, then takes each of these to theirimage under ι, breaking them again whenever they are no longer connected. Theleft-hand side takes t to ι ( t ) , a tube of the complete graph on p + q nodes, andthen breaks ι ( t ) into the tubes that result from deleting the specified edges of the( p + q )-node complete graph. In this last step, by Lemma 3.3, we can delete firstthose edges that also happen to lie in the complete subgraph induced by ι ([ p ]) , andthen the remaining specified edges. Thus we have duplicated the left-hand side andget the same final set of tubes on either side of the equation.(2) For s a tube of v , the right-hand side first breaks s into several smaller tubes bydeleting certain edges of the q -node complete graph, then takes each of these to theirimage under ˆ ι, expanding them to include tubes of ι (Θ c ( u )). The left-hand side takes s to ˆ ι ( s ) ∪ ι ([ p ]) , and then breaks the result into the tubes that result from deletingthe specified edges of the ( p + q )-node complete graph. Again we can delete firstthose edges that also happen to lie in the complete subgraph induced by ι ([ p ]) , andthen the remaining specified edges, duplicating the left-hand process.An illustration of the two sides of the equation is in Figure 19. The surjectivity of ˆΘ c followsfrom the surjectivity of the generalized Tonks projection, as shown in Lemma 3.2. (cid:3) Let C n be the cycle graph on n numbered nodes and let w be the edge from node 1 tonode n. We can define a map from W Sym to Y Sym by:ˆΘ w ( F u ) = F Θ w ( u ) , for u ∈ W n . Theorem 5.7. ˆΘ w is a surjective homomorphism of graded algebras W Sym → Y
Sym .Proof.
In [11] it is shown that the map we call τ is an algebra homomorphism from S Sym to Y Sym.
In the previous theorem we demonstrated that the map ˆΘ c is a surjective algebrahomomorphism from S Sym to W Sym.
Now, since τ is the same map on vertices as our Θ P (from Theorem 3.5), then the relationship of these three is: τ = ˆΘ w ◦ ˆΘ c . EOMETRIC COMBINATORIAL ALGEBRAS: CYCLOHEDRON AND SIMPLEX 21
12 34
12 3
123 4 5 67
12 34
12 3
12 3 4 5 67 Θ c u v Figure 19.
An example of the equation Θ c ( ι ( u ) ∪ ˆ ι ( v )) = ι (Θ c ( u )) ∪ ˆ ι (Θ c ( v )) . where u = (2314) , v = (312) and ι ([4]) = { , , , } . Thus ˆΘ w is an algebra homomorphism from S Sym onto W Sym. (cid:3)
Remark 5.8.
The existence of a surjective algebra homomorphism from S Sym not onlyallows us a shortcut to demonstrating associativity, but leads to an alternate description ofthe product in the range of that homomorphism. This may be achieved in three steps:(1) Lifting our p and q -tubings in W Sym to any preimages of the generalized Tonksprojection Θ c on the complete graphs on p and q nodes.(2) Performing the product of these complete graph tubings in S Sym ,(3) and finding the image of the resulting terms under the homomorphism ˆΘ c .This description is independent of our choices of preimages in S Sym due to the homomor-phism property.The following corollary follows directly from the properties of the surjective algebra ho-momorphisms of Theorems 5.6 and 5.7.
Corollary 5.9. W Sym is a left S Sym -module under the definition F u · F Θ c ( v ) = ˆΘ c ( F u · F v ) and a right S Sym -module under the definition F Θ c ( u ) · F v = ˆΘ c ( F u · F v ) . Y Sym is a left W Sym -module under the definition F u · F Θ w ( v ) = ˆΘ w ( F u · F v ) and a right W Sym -moduleunder the definition F Θ w ( u ) · F v = ˆΘ w ( F u · F v ) . Extension to the faces of the polytopes
Chapoton was the first to point out the fact that in studying the Hopf algebras basedon vertices of permutohedra, associahedra and cubes, one need not restrict their attentionto just the zero-dimensional faces of the polytopes. He has shown that the Loday RoncoHopf algebra Y Sym, the Hopf algebra of permutations S Sym, and the Hopf algebra ofquasisymmetric functions Q Sym are each subalgebras of algebras based on the the trees,the ordered partitions, and faces of the hypercubes respectively [5]. Furthermore, he hasdemonstrated that these larger algebras of faces are bi-graded and possess a differential. Here we point out that the cyclohedra based algebra W Sym can be extended to a larger algebrabased on all the faces of the cyclohedra as well, and conjecture the additional properties.Chapoton’s product structure on the permutohedra faces is given in [5] in terms of orderedpartitions. Let ˜ S Sym be the graded vector space over Q with the n th component of its basisgiven by the ordered partitions of [ n ] . We write F u for the basis element corresponding tothe m -partition u : [ n ] → [ m ] , for 0 ≤ m ≤ n, and 1 for the basis element of degree 0. Theproduct in ˜ S Sym of two basis elements F u and F v for u : [ p ] → [ k ] and v : [ q ] → [ l ] is foundby summing a term for each shuffle, created by composing the juxtaposition of u and v withthe inverse shuffle: F u · F v = X σ ∈ S ( p,q ) F ( u × v ) · σ − . Here u × v is the ordered ( k + l )-partition of [ p + q ] given by:( u × v )( i ) = ( u ( i ) , i ∈ [ p ] v ( i − p ) + k, i ∈ { p + 1 , . . . , p + q } . The bijection between tubings of complete graphs and ordered partitions allows us to writethis product geometrically.
Theorem 6.1.
The product in ˜ S Sym may be written as: F u · F v = X ι ∈ S ( p,q ) F ˆ ρ ι ( u,v ) which is just Definition 4.1 extended to all pairs of faces ( u, v ) . Proof.
Recall that we found the bijection between tubings of complete graphs and orderedpartitions by noting that each tube contains some numbered nodes which are not containedin any other tube. These subsets of [ n ] , one for each tube, make up the partition, andthe ordering of the partition is from innermost to outermost tube. Now the Carr-Devadossisomorphism ˆ ρ is a bijection of face posets. With this in mind, the same argument appliesas in the proof of Theorem 4.1. (cid:3) In other words, we view our operands (a pair of ordered partitions) as a face of the cartesianproduct of permutohedra P p × P q . Then the product is the formal sum of images of that faceunder the collection of inclusions of P p × P q as a facet of P p + q . An example is in Figure 20.We leave to the reader the by now straightforward tasks of finding the geometric inter-pretations of the coproduct on ˜ S Sym and of the Hopf algebra structure on the faces of theassociahedra, ˜ Y Sym.
Each one can be done simply by repeating earlier definitions usingtubings, but with all sizes of k -tubings as operands.Recall that the faces of the cyclohedra correspond to tubings of the cycle graph. We willdefine a graded algebra with a basis which corresponds to the faces of the cyclohedra, andwhose grading respects the dimensions (plus one) of the cyclohedra. Definition 6.2.
Let ˜ W Sym be the graded vector space over Q with the n th component ofits basis given by all the tubings on the cycle graph of n numbered nodes. By W n = K C n we denote the poset of tubings on the cycle graph C n with a cyclic numbering of nodes. Wewrite F u for the basis element corresponding to u ∈ W n and 1 for the basis element of degree0. EOMETRIC COMBINATORIAL ALGEBRAS: CYCLOHEDRON AND SIMPLEX 23 ({1 2})({1 2}) ({1 2},{3 4}) ({1 3},{2 4}) ({1 4},{2 3}) ({2 3},{1 4}) ({2 4},{1 3}) ({3 4},{1 2}) P = P P P P
12 34
12 34 =
12 34 12 34 + ++
12 34 12 34 + + {1 2} {1 2} ρ Figure 20.
The product in ˜ S Sym . As an exercise the reader can deleteedges in the graphs to illustrate the products in ˜ W Sym and ˜ Y Sym.
The product in ˜ W Sym of two basis elements F u and F v for u ∈ W p and v ∈ W q is foundby summing a term for each shuffle ι ∈ S ( p,q ) . First the l -tubing u is drawn upon the inducedsubgraph of the nodes ι ([ p ]) according to the ascending order of the chosen nodes. However,each tube may need to be first broken into several sub-tubes. Then the m -tubing v is drawnupon the subgraph induced by the remaining q nodes. However, each of the tubes of v isexpanded to also include any of the previously drawn tubes that its nodes are adjacent to. Definition 6.3. F u · F v = X ι ∈ S ( p,q ) F ˆ ρ ι ( η ι (Θ ι ( u )) ,v ) . where the facet inclusion ˆ ρ ι is the same as in the two previous definitions using this template:Definitions 7.2 and 5.2. Theorem 6.4.
The product we have just defined makes ˜ W Sym into an associative gradedalgebra.Proof.
The proof is almost precisely the same as for the algebra of vertices of the cyclohedra, W Sym.
The only difference is that the tubings do not always have the maximum number oftubes. First we must check that the result of the product is indeed a sum of valid tubings ofthe cycle graph C p + q . We claim that in each term the new tubes we have created are pairwisecompatible. The cases to be checked and the reasoning for each are exactly as shown in theproof of Theorem 5.3. Associativity is shown by lifting the tubings to be multiplied totubings on the complete graphs which are preimages of the extended Tonks projection, andperforming the multiplication in Chapoton’s algebra. The fact that the extended Tonksprojection does preserve the structure here is again an easy check of corresponding terms,following the same pattern as for the n -tubings in Theorem 5.6. (cid:3) Remark 6.5.
Chapoton has shown the existence of a differential graded structure on thealgebras of faces of the permutohedra, associahedra, and cubes [5]. The basic idea is simple,to define the differential as a signed sum of the bounding sub-faces of a given face. Here we leave for future investigation the possibility of extending this differential to algebras ofgraph associahedra. 7.
Algebras of simplices
We introduce a curious new graded algebra whose n th component has dimension n. Wedenote it ∆
Sym.
In fact ∆
Sym may be thought of as a graded algebra whose basis is madeup of all the standard bases for Euclidean spaces.The graph associahedron of the edgeless graph on n vertices is the ( n − n − ofdimension n − . Thus the final range of our extension of the Tonks projection is the ( n − Definition 7.1.
Let ∆
Sym be the graded vector space over Q with the n th component of itsbasis given by the n -tubings on the edgeless graph of n numbered nodes. By D n we denotethe set of n -tubings on the edgeless graph. We write F u for the basis element correspondingto u ∈ D n and 1 for the basis element of degree 0.7.1. Graded algebra structure on vertices.
Now we demonstrate a product which re-spects the grading on ∆
Sym by following the example described above for S Sym.
Theproduct in ∆
Sym of two basis elements F u and F v for u ∈ D p and v ∈ D q is found bysumming a term for each shuffle ι ∈ S ( p,q ) . Our term of F u · F v will be a p + q tubing of theedgeless graph on p + q nodes. Its tubes will include all the nodes numbered by ι ([ p ]) . Wedenote these by ι ( u ) . In addition we include all the tubes ˆ ι ( t ) for non-universal t ∈ v, andthe universal tube. We denote these by ˆ ι ( v ) . By inspection we see that in this case the union ι ( u ) ∪ ˆ ι ( v ) = ˆ ρ ι ( η ( u ) , v ) . Thus we write:
Definition 7.2. F u · F v = X ι ∈ S ( p,q ) F ˆ ρ ι ( η ( u ) ,v ) . Also 1 is a two-sided unit.An example is shown in Figure 21. = + + + ++ + + ++
Figure 21.
The product in ∆
Sym.
Theorem 7.3.
The product we have just described makes ∆ Sym into an associative gradedalgebra.
EOMETRIC COMBINATORIAL ALGEBRAS: CYCLOHEDRON AND SIMPLEX 25
Proof.
First we must check that the result of the product is indeed a sum of valid ( p + q )-tubings of the edgeless graph on p + q nodes. This is clearly true, since there will always beonly one node that is not a tube in any given term of the product. Associativity is shownby the existence of an algebra homomorphism S Sym → ∆ Sym which we will see next inLemma 7.5. (cid:3)
Definition 7.4.
Let E n = the set of edges of the complete graph on n nodes. Deleting thesegives a projection Θ n from the permutohedra to the simplices, as in Definition 3.4. Then wedefine ˆΘ δ : S Sym → ∆ Sym by ˆΘ δ ( F u ) = F Θ n ( u ) for u an n -tubing of K n . Lemma 7.5.
The map ˆΘ δ is an algebra homomorphism.Proof. The proof follows precisely the same arguments as the proof for Theorem 5.6, exceptthat the cases are simpler. We consider ˆΘ δ ( F u · F v ) for u ∈ S p and v ∈ S q . We compare theresult with ˆΘ δ ( F u ) · ˆΘ E ( F v ) . Each of the multiplications results in a sum of (cid:0) p + qp (cid:1) terms. Itturns out that comparing the results of the two operations can be done piecewise. Thus wecheck that for a given shuffle ι the respective terms of our two operations agree: we claimthat Θ p + q ( ι ( u ) ∪ ˆ ι ( v )) = ι (Θ p ( u )) ∪ ˆ ι (Θ q ( v )) . On the left-hand side, Θ p + q forgets all the information in the tubing it is applied to; exceptfor the following data: which of the nodes is the only node in the universal tube and not inany other tube. This particular node is actually the node numbered by j = ˆ ι ( v − ( q )) . Theeffect of Θ p + q is to create the n -tubing of the edgeless graph with node j as the only nodethat is not a tube.On the right hand side Θ p and Θ q forget all but the value of u − ( p ) and v − ( q ) respectively.Then we create the tubing of the edgeless graph by first including all the nodes numberedby ι ([ p ]) and then all the nodes except node j = ˆ ι ( v − ( q )) . (cid:3) Remark 7.6.
There are clearly algebra homomorphisms Y Sym → ∆ Sym and W Sym → ∆ Sym described by the extended Tonks projections from associahedra and cyclohedra tothe simplices.7.2.
Formula for the product structure.
There is a simple bijection from n -tubings ofthe edgeless graph on n nodes to standard basis elements of Q n . Let e nm be the column vectorof Q n with all zero entries except for a 1 in the m th position. Associate e nj = e u with the n -tubing u whose nodes are all tubes except for the j th node. Then use the product of ∆ Sym to define a product of two standard basis vectors of varying dimension: e u · e v is the sum ofall e w for F w a term in the product F u · F v . Then:
Theorem 7.7. e pj · e ql = p + l X i = l (cid:18) i − l − (cid:19)(cid:18) p + q − iq − l (cid:19) e p + qi Proof.
Let e ql = e v for the associated tubing v. The only node not a tube of v is node l. Weneed only keep track of where l lands under ˆ ι : [ q ] → [ p + q ] , where ˆ ι is as in Definition 7.2.The only possible images of l are from l to p + l, thus the limits of the summation. Whenˆ ι ( l ) = i there are several ways this could have occurred. ˆ ι must have mapped [ l −
1] to [ i − and the set { l + 1 , . . . , q } to { i + 1 , . . . , p + q } . The ways this can be done are enumeratedby the combinations in the sum. (cid:3)
Example 7.8.
Consider the example product performed in Figure 21. Here the formulagives the observed quantities: e · e = 3 e + 4 e + 3 e . Faces of the simplex.
Note that a product of tubings on the edgeless graph (withany number of tubes) is easily defined by direct analogy. Now we present a Hopf algebrawith one-sided unit based upon the face posets of simplices. Tubings on the edgeless graphslabel all the faces of the simplices. The number of faces of the n -simplex, including the nullface and the n -dimensional face, is 2 n . By adjoining the null face here we thus have a gradedbialgebra with n th component of dimension 2 n . It would be of interest to compare this withother algebras of similar dimension, such as Q Sym . Definition 7.9.
Let ˜∆
Sym be the graded vector space over Q with the n th component of itsbasis given by the tubings on the edgeless graph of n numbered nodes, with one extra basiselement included in each component: corresponding to the null facet ∅ n is the collectionconsisting of all n of the singleton tubes and the universal tube. By D n we denote the setof n -tubings on the edgeless graph, together with ∅ n . We write F u for the basis elementcorresponding to u ∈ D n and 1 for the basis element of degree 0.Now we define the product and coproduct, with careful description of the units.Let u ∈ D p and v ∈ D q . For a given shuffle ι our term of F u · F v will be indexed by anelement of D p + q which will include all the nodes numbered by ι ([ p ]) . In addition we includeall the tubes ˆ ι ( t ) for non-universal t ∈ v, and the universal tube. Thus the product is anextension of Definition 7.2, with a redefined right multiplication by the unit: Definition 7.10. F u · F v = X ι ∈ S ( p,q ) F ˆ ρ ι ( η ( u ) ,v ) ; 1 · F u = F u ; F u · F ∅ p . An example is shown in Figure 22. = + + + ++ + + ++
Figure 22.
The product in ˜∆
Sym.
The coproduct is defined simply by restricting an element of D n to its subgraphs inducedby the nodes 1 , . . . , i and i + 1 , . . . n. Given a tubing u of the edgeless graph on n verticeswe can find tubings u i and u n − i as follows: for each tube t ∈ u we find the intersectionsof t with the two sub-graphs (also edgeless) induced by the nodes 1 , . . . , i and i + 1 , . . . , n respectively. ∆( F u ) = n X i =0 F u i ⊗ F u n − i . An example is shown in Figure 23.
EOMETRIC COMBINATORIAL ALGEBRAS: CYCLOHEDRON AND SIMPLEX 27 = + ++ + ∆ + Figure 23.
The coproduct in ˜∆
Sym.
Theorem 7.11.
The product and coproduct just defined form a graded bialgebra structureon ˜∆ Sym , and therefore a (one-sided) Hopf algebra.Proof.
Associativity of the product is due to the observation that the result of multiplying aseries of basis elements only depends on the final operand. Coassociativity of the coproductfollows from the fact that terms of both (1 ⊗ ∆)∆ F u and (∆ ⊗ F u involve simply splittingthe edgeless graph into three consecutive pieces, and restricting u to those pieces. Now wedemonstrate that the product and coproduct commute; that is, ∆( F u · F v ) = (∆ F u ) · (∆ F v ) . We describe a bijection between the terms on the left-hand and right-hand sides, which turnsout to be the identity map. Let u ∈ D p and v ∈ D q . A term of the left-hand side dependsfirst upon a ( p, q )-shuffle ι to choose p of the nodes of the ( p + q )-node edgeless graph. Thenafter forming the product, a choice of i ∈ , . . . , p + q determines where to split the resultinto the pieces of the final tensor product. Let m = |{ x ∈ ι ([ p ]) : x ≤ i }| . Thus m countsthe number of nodes in the image of ι which lie before the split.To create a term of the right hand side, we first split both u and v, then interchange andmultiply the resulting four terms as prescribed in the definition of the tensor product ofalgebras. Our matching term on the right-hand side is the one formed by first splitting u after node m, and v after node i − m. Then in the first multiplication choose the ( m, i − m )-shuffle σ ( x ) = ι ( x ), and in the second use the ( q − m, q − i + m )-shuffle σ ′ ( x ) = ι ( x + m ) − i. These choices define an identity map from the set of terms of the left hand side to those onthe right. (cid:3)
Remark 7.12. ˜∆ Sym is closely related to the free associative trialgebra on one variabledescribed by Loday and Ronco in [12]. In [12, Proposition 1.9] the authors describe theproducts for that trialgebra. Axiomatically, the first two products automatically form a dialgebra . The sum of these two products appears as two of the terms of our shuffle-basedproduct! We leave it to an interested reader to uncover the precise relationship, perhapsduality, between the two structures.
References [1] Marcelo Aguiar and Frank Sottile,
Structure of the Malvenuto-Reutenauer Hopf algebra of permutations ,Adv. Math. (2005), no. 2, 225–275. MR MR2103213 (2005m:05226)[2] ,
Structure of the Loday-Ronco Hopf algebra of trees , J. Algebra (2006), no. 2, 473–511.MR MR2194965 (2006k:16078)[3] Raoul Bott and Clifford Taubes,
On the self-linking of knots , J. Math. Phys. (1994), no. 10, 5247–5287, Topology and physics. MR MR1295465 (95g:57008)[4] Michael P. Carr and Satyan L. Devadoss, Coxeter complexes and graph-associahedra , Topology Appl. (2006), no. 12, 2155–2168. MR MR2239078 (2007c:52012)[5] Fr´ed´eric Chapoton,
Big`ebres diff´erentielles gradu´ees associ´ees aux permuto`edres, associa`edres et hyper-cubes , Ann. Inst. Fourier (Grenoble) (2000), no. 4, 1127–1153. MR MR1799740 (2002f:16081)[6] Satyan Devadoss and Stefan Forcey, Marked tubes and the graph multiplihedron , Algebr. Geom. Topol. (2008), no. 4, 2081–2108. MR MR2460880[7] Satyan L. Devadoss, A space of cyclohedra , Discrete Comput. Geom. (2003), no. 1, 61–75.MR MR1946794 (2003j:57027)[8] , A realization of graph associahedra , Discrete Math. (2009), no. 1, 271–276. MR MR2479448 [9] Stefan Forcey, Aaron Lauve, and Frank Sottile,
Constructing cofree combinatorial hopf algebras , inpreparation.[10] Christophe Hohlweg and Carsten E. M. C. Lange,
Realizations of the associahedron and cyclohedron ,Discrete Comput. Geom. (2007), no. 4, 517–543. MR MR2321739 (2008g:52021)[11] Jean-Louis Loday and Mar´ıa O. Ronco, Hopf algebra of the planar binary trees , Adv. Math. (1998),no. 2, 293–309. MR MR1654173 (99m:16063)[12] ,
Trialgebras and families of polytopes , Homotopy theory: Relations with Algebraic Geometry,Group Cohomology and Algebraic K -Theory, Contemp. Math., vol. 346, Amer. Math. Soc., Providence,RI, 2004, pp. 369–398. MR MR2066507 (2006e:18016)[13] Clauda Malvenuto and Christophe Reutenauer, Duality between quasi-symmetric functions and theSolomon descent algebra , J. Algebra (1995), no. 3, 967–982. MR MR1358493 (97d:05277)[14] Martin Markl,
Simplex, associahedron, and cyclohedron , Higher homotopy structures in topology andmathematical physics (Poughkeepsie, NY, 1996), Contemp. Math., vol. 227, Amer. Math. Soc., Provi-dence, RI, 1999, pp. 235–265. MR MR1665469 (99m:57020)[15] Jason Morton, Anne Shiu, Lior Pachter, and Bernd Sturmfels,
The cyclohedron test for finding peri-odic genes in time course expression studies , Stat. Appl. Genet. Mol. Biol. (2007), Art. 21, 25 pp.(electronic). MR MR2349914[16] Alex Postnikov, Victor Reiner, and Lauren Williams, Faces of generalized permutohedra , Doc. Math. (2008), 207–273. MR MR2520477[17] Alexander Postnikov, Permutohedra, associahedra, and beyond , Int. Math. Res. Not. IMRN (2009),no. 6, 1026–1106. MR MR2487491[18] Nathan Reading,
Cambrian lattices , Adv. Math. (2006), no. 2, 313–353. MR MR2258260(2007g:05195)[19] Rodica Simion,
A type-B associahedron , Adv. in Appl. Math. (2003), no. 1-2, 2–25, Formal powerseries and algebraic combinatorics (Scottsdale, AZ, 2001). MR MR1979780 (2004h:52013)[20] Andy Tonks, Relating the associahedron and the permutohedron , Operads: Proceedings of RenaissanceConferences (Hartford, CT/Luminy, 1995), Contemp. Math., vol. 202, Amer. Math. Soc., Providence,RI, 1997, pp. 33–36. MR MR1436915 (98c:52015)
S. Forcey: Tennessee State University, Nashville, TN 37209
E-mail address ::