This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Morse shellings on products

Jean-Yves Welschinger
Abstract

We recently defined a property of Morse shellability (and tileability) of finite simplicial complexes which extends the classical one and its relations with discrete Morse theory. We now prove that the product of two Morse tileable or shellable simplicial complexes carries Morse tileable or shellable triangulations under some tameness condition, and that any tiling or shelling becomes tame after one barycentric subdivision. We deduce that any finite product of closed manifolds of dimensions less than four carries Morse shellable triangulations whose critical and hh-vectors are palindromic. We also prove that the hh-vector of a Morse tiling is always palindromic in dimension less than four or in the case of an hh-tiling, provided its critical vector is palindromic.

Keywords : simplicial complex, shellable complex, hh-vector, tilings, discrete Morse theory, triangulation.

Mathematics subject classification 2020: 57Q70, 55U10, 52C22, 05E45.

1 Introduction

Recall that the face vector of a finite nn-dimensional simplicial complex KK encodes the number of simplices it contains in each dimension, that is its number of vertices, edges and so on, see §2.1. When KK is the boundary of a convex polytope for example, it has been (strikingly) understood by L.J. Billera, C.W. Lee and R. P. Stanley what this face vector can be, confirming an earlier conjecture of P. McMullen, see [1, 22, 8, 24]. The answer is expressed in terms of its hh-vector h(K)=(h0(K),,hn+1(K))h(K)=(h_{0}(K),\dots,h_{n+1}(K)), which is a linear recombination of its face vector, see §2.1, and turns out to coincide with the list of Betti numbers of the toric variety X(K)X(K) associated to the convex polytope. In particular, Poincaré duality in X(K)X(K) implies that hj(K)=hn+1j(K)h_{j}(K)=h_{n+1-j}(K) for every j{0,,n+1}j\in\{0,\dots,n+1\}, a result which also follows directly from the Dehn-Sommerville relations, see [15, 14, 11, 24] or also Theorem 1.11.1 of [18]. We will declare such a vector to be palindromic, see Definition 3.1. In general, it is unclear how to understand hh-vectors, except at least for shellable complexes, see §8.38.3 of [24]. We introduced in [17] a notion of tiling of simplicial complexes and when such a tiling τ\tau exists, e.g. for shellable complexes, defined its hh-vector h(τ)=(h0(τ),,hn+1(τ))h(\tau)=(h_{0}(\tau),\dots,h_{n+1}(\tau)) to be the number of tiles of each type used by τ\tau, see §2.2. A tile here, or basic tile, is just a simplex deprived of several of its codimension one faces, whose number is called its order, so that closed and open simplices are particular ones, of minimal and maximal order respectively, see Definition 2.2. By Theorem 4.94.9 and Corollary 4.104.10 of [17], two hh-tilings τ\tau and τ\tau^{\prime} of a simplicial complex KK have same hh-vector provided h0(τ)=h0(τ)h_{0}(\tau)=h_{0}(\tau^{\prime}) and when moreover the latter equals one, this hh-vector coincides with the hh-vector of KK. These results thus sometimes provide a geometric meaning of hh-vectors but also provide a larger class of vectors of interest, for h0(τ)h_{0}(\tau) need not be one. If among the closed manifolds, only the spheres carry shellable triangulations [12, 23], we do not know which ones carry hh-tileable triangulations. We however prove the following, see Corollary 3.13.

Theorem 1.1.

The product of a sphere and a torus of any dimensions carries hh-tileable triangulations.

Definitely, a closed manifold of even dimension and negative Euler characteristic has no hh-tileable triangulations though, see Lemma 2.9.

Influenced by the discrete Morse theory of R. Forman [6, 7], we enlarged in [19] the collection of tiles in each dimension by allowing a unique face of higher codimension to be removed from a simplex, thus introducing Morse tiles, see Definition 2.4. These include a collection of critical tiles of any index and led to properties of Morse tileable and shellable complexes, see §2.2. Among basic tiles, closed and open simplices are the only critical ones, of minimal and maximal indices respectively. We proved that any triangulation on a closed surface is Morse shellable and that any closed three-manifold carries Morse shellable triangulations, see Theorems 1.31.3 and 1.41.4 of [19]. Moreover, Morse tilings carry compatible discrete vector fields and in the case of Morse shellings, these are gradient vector fields of discrete Morse functions whose critical points are in one-to-one correspondence with the critical tiles of the tiling, preserving the index, see Theorem 1.21.2 of [19]. We now prove, see Corollary 3.9.

Theorem 1.2.

Any finite product of closed manifolds of dimensions less than four carries Morse shellable triangulations.

Recall that H. Bruggesser and P. Mani proved that the boundary of every convex polytope is shellable, while some triangulations on spheres are not, see [3, 13, 24]. We do not know any closed triangulated manifold which is not Morse tileable or shellable. We encode the number of critical tiles of each index used by an nn-dimensional Morse tiling τ\tau in a critical vector c(τ)=(c0(τ),,cn(τ))c(\tau)=(c_{0}(\tau),\dots,c_{n}(\tau)) and likewise, the number of tiles of each order it uses in an hh-vector h(τ)=(h0(τ),,hn+1(τ))h(\tau)=(h_{0}(\tau),\dots,h_{n+1}(\tau)), as in the case of hh-tilings, see §2.2. We then prove, see Corollary 3.3 and Theorem 3.5.

Theorem 1.3.

The hh-vector of an hh-tiling on a closed triangulated manifold is palindromic iff its cc-vector is. Likewise, the hh-vector of a Morse tiling on a closed triangulated manifold of dimension less than four is palindromic iff its cc-vector is.

The tilings given by Theorems 1.1 and 1.2 can be chosen to have palindromic hh-vectors as well. Does there exist a Morse tiling on a closed triangulated manifold which has palindromic cc-vector but non palindromic hh-vector? It would also be of interest to get Theorem 1.2 in any dimension. In dimension three, it has been obtained in [19] by successive attachments of triangulated handles equipped with Morse shellings. We now prove the existence of such shellings on every handle, see Corollary 3.17.

Theorem 1.4.

For every 0kn0\leq k\leq n, the handle Δk×Δnk\stackrel{{\scriptstyle\circ}}{{\Delta}}_{k}\times\Delta_{n-k} carries Morse shellable triangulations using a unique critical tile, of index kk.

The core of the paper actually aims at proving that the product of two Morse tileable or shellable simplicial complexes carries Morse tileable or shellable triangulations. We first prove this result for single Morse tiles, see Theorem 3.14 of which Theorem 1.4 is a special case, and then observe a duality phenomenon which makes it possible to get the palindromic property, see Theorem 3.16. We then prove the result in general under some tameness condition on the tilings, see §2.3, to get.

Theorem 1.5.

Let K1K_{1} and K2K_{2} be finite simplicial complexes equipped with tame Morse tilings (resp. shellings) τ1\tau_{1} and τ2\tau_{2}. Then, K1×K2K_{1}\times K_{2} carries tame Morse tileable (resp. shellable) primitive triangulations. Moreover, if τ1\tau_{1}, τ2\tau_{2} are pure dimensional, these Morse tilings have palindromic hh-vectors provided h(τ1)h(\tau_{1}) and h(τ2)h(\tau_{2}) are palindromic.

The critical vector of such tilings on K1×K2K_{1}\times K_{2} is a product of the ones of τ1\tau_{1} and τ2\tau_{2} while τ1\tau_{1}, τ2\tau_{2} are always pure dimensional in the case of triangulated manifolds, see Lemma 2.6 and Theorem 3.8. Theorem 1.5 suffices to deduce Theorems 1.1 and 1.2, for the tilings or shellings on each factor can be chosen to be tame and it has a counterpart which produces hh-tilings as well, see Theorem 3.10. In fact, any Morse tiling or shelling becomes tame after a single barycentric subdivision, see Proposition 2.14. We finally provide many examples of Morse shellings throughout the paper, see in particular §5.4.

We recall in section 2 the classical notions of face and hh-vectors of simplicial complexes, the notions of tilings and shellings defined in [17, 19] and we introduce the tameness condition needed to get Theorem 1.5. We then formulate our main results in section 3, devoting §3.1 to the palindromic property and Theorem 1.3, §3.2 to Theorem 1.5 and §3.3 to the special case of single tiles and Theorem 1.4. We study in §4 the cartesian products of two simplices together with the shellings of its staircase triangulations. This makes it possible to prove the main results in §§5 and 6.

Acknowledgement: This work was partially supported by the ANR project MICROLOCAL (ANR-15CE40-0007-01).

2 Preliminaries

2.1 Simplicial complexes

Let nn be a non-negative integer. An nn-simplex is the convex hull of n+1n+1 points affinely independent in some real affine space. A face of a simplex is the convex hull of a subset of its vertices and we call it a facet when it has codimension one in the simplex. The standard nn-simplex is the convex hull of the standard basis of n+1\mathbb{R}^{n+1}. It will be denoted by Δ[n]\Delta_{[n]}, or sometimes just by Δn\Delta_{n}, fixing an identification between its vertices and the set of integers [n]={0,,n}[n]=\{0,\dots,n\}. Likewise, for every subset JJ of {0,,n}\{0,\dots,n\}, we will denote by ΔJ\Delta_{J} the face of Δ[n]\Delta_{[n]} whose vertices belong to JJ. A total order on the vertices of any simplex prescribes then an affine isomorphism with the standard simplex of the corresponding dimension.

A finite simplicial complex KK is a finite collection of simplices which contains all faces of its simplices and such that the intersection of any two simplices in this collection is a face of each of them, see [16, 5]. The dimension of such a complex is the maximal dimension of its simplices and it is said to be pure nn-dimensional if all the simplices that are maximal with respect to the inclusion are of dimension nn. Such a simplicial complex KK inherits a topology and the underlying topological space is usually denoted by |K||K|, see [16, 5]. When it gets homeomorphic to some manifold, any such homeomorphism is called a triangulation of the manifold.

The face vector or ff-vector of an nn-dimensional finite simplicial complex KK is the vector f(K)=(f1(K),f0(K),,fn(K))f(K)=(f_{-1}(K),f_{0}(K),\dots,f_{n}(K)), where for every j{0,,n}j\in\{0,\dots,n\}, fj(K)f_{j}(K) denotes the number of jj-simplices of KK while f1(K)=1f_{-1}(K)=1 counts the empty set. Likewise, the hh-vector h(K)=(h0(K),,hn+1(K))h(K)=(h_{0}(K),\dots,h_{n+1}(K)) of KK is defined by the relation i=0n+1hi(K)Xn+1i=i=0n+1fi1(K)(X1)n+1i\sum_{i=0}^{n+1}h_{i}(K)X^{n+1-i}=\sum_{i=0}^{n+1}f_{i-1}(K)(X-1)^{n+1-i}, see [15, 21, 8, 24].

Example 2.1.

The boundary of a simplex is homeorphic to a sphere. Its hh-vector equals (1,,1)(1,\dots,1).

Let us finally recall that a finite simplicial complex is said to be shellable iff there exists an order σ1,,σN\sigma_{1},\dots,\sigma_{N} of its maximal simplices such that for every i{2,,N}i\in\{2,\dots,N\}, σi(j=1N1σj)\sigma_{i}\cap\big{(}\cup_{j=1}^{N-1}\sigma_{j}\big{)} is non-empty of pure dimension dimσi1\dim\sigma_{i}-1, see [12, 24] for instance. This means that the simplices σ1,,σN\sigma_{1},\dots,\sigma_{N} are not proper faces of any other simplex in KK and that any simplex in σi(j=1N1σj)\sigma_{i}\cap\big{(}\cup_{j=1}^{N-1}\sigma_{j}\big{)} is a face of a (dimσi1)(\dim\sigma_{i}-1)-dimensional one in this intersection. It is convenient for us to allow this intersection for being empty, so that a shelling for us need not be connected, see Remark 2.162.16 of [19] and §2.2.

2.2 Morse shellings

We now recall the notions of tilings and shellings introduced in [17, 19].

Definition 2.2.

A basic tile of dimension nn and order k{0,,n+1}k\in\{0,\dots,n+1\} is an nn-simplex deprived of kk of its facets.

Two basic tiles of same dimension and order are isomorphic to each other via some affine isomorphism. We denote by Tkn=Δ[n]j=0k1Δ[n]{j}T^{n}_{k}=\Delta_{[n]}\setminus\cup_{j=0}^{k-1}\Delta_{[n]\setminus\{j\}} the standard basic tile of dimension nn and order kk, compare [17].

Example 2.3.

1) The open (resp. closed) nn-simplex is the basic tile of dimension nn and order n+1n+1 (resp. 0).

2) Figure 1 depicts the four isomorphism classes of basic tiles in dimension two.

Refer to caption
Figure 1: The basic tiles in dimension two.
Definition 2.4 (Definition 2.42.4 of [19]).

A Morse tile of dimension nn and order k{0,,n+1}k\in\{0,\dots,n+1\} is an nn-simplex σ\sigma deprived of kk of its facets together with, if k1k\geq 1, a possibly empty face μ\mu of higher codimension. The simplex σ\sigma (resp. σ\sigma deprived of the kk facets) is called the underlying simplex (resp. basic tile), while μ\mu is called its Morse face.

When k1k\geq 1, the dimension of μ\mu ranges between k1k-1 and n2n-2 and the underlying basic tile has a unique face of dimension k1k-1, see [19]. Any basic tile is Morse and a Morse tile is said to be critical of index kk iff it is of order kk and its Morse face has minimal dimension k1k-1, while a closed simplex is critical of index zero. The other Morse tiles are said to be regular. We sometimes denote by CknC^{n}_{k} (resp. Tkn,lT^{n,l}_{k}) a nn-dimensional critical tile of index kk (resp. a nn-dimensional Morse tile of order kk with ll-dimensional Morse face), so that Ckn=Tkn,k1C^{n}_{k}=T^{n,k-1}_{k}. They are all isomorphic to each other via some affine isomorphism.

Definition 2.5 (Definition 2.82.8 of [19]).

A subset SS of the underlying topological space |K||K| of a finite simplicial complex KK is said to be Morse tileable iff it admits a partition by Morse tiles such that for every j0j\geq 0, the union of tiles of dimension j\geq j is closed in SS. Such a partition τ\tau is called a Morse tiling and the closure of SS in KK is called the underlying simplicial complex.

When the tiling uses only basic tiles, it is called an hh-tiling, see [17, 19]. Of special interest is the case S=S¯S=\overline{S} where a finite simplicial complex is Morse tiled, but Definition 2.5 is more general and includes sets such as the triangulated handles Δk×Δnk\stackrel{{\scriptstyle\circ}}{{\Delta}}_{k}\times\Delta_{n-k} of Theorem 1.4. The dimension of a tileable subset is the dimension of the underlying simplicial complex, that is the maximal dimension of the tiles in any Morse tiling. When all tiles have same dimension, the tiling is said to be pure dimensional. This is always the case on compact triangulated manifolds. Indeed,

Lemma 2.6.

Any Morse tiling on a compact connected triangulated manifold is pure dimensional.

Proof.

Let nn be the dimension of the triangulated manifold KK and let τ\tau be any Morse tiling on KK. Then, the nn-dimensional tiles of τ\tau cover all open nn-dimensional simplices of KK and by Definition 2.5, their union is closed in KK, so that it contains all closed nn-dimensional simplices of KK as well. Since KK is a compact connected triangulated manifold, the latter is KK itself. ∎

Definition 2.7.

The hh-vector (resp. cc-vector) of a nn-dimensional Morse tiling τ\tau is the vector h(τ)=(h0(τ),,hn+1(τ))h(\tau)=(h_{0}(\tau),\dots,h_{n+1}(\tau)) (resp. c(τ)=(c0(τ),,cn(τ))c(\tau)=(c_{0}(\tau),\dots,c_{n}(\tau))) whose entries hk(τ)h_{k}(\tau), k{0,,n+1}k\in\{0,\dots,n+1\} (resp. ck(τ)c_{k}(\tau), k{0,,n}k\in\{0,\dots,n\}), are the number of tiles of order kk (resp. critical tiles of index kk) used by τ\tau.

In particular, h0(τ)=c0(τ)h_{0}(\tau)=c_{0}(\tau) and hn+1(τ)=cn(τ)h_{n+1}(\tau)=c_{n}(\tau). Recall that the hh-vector h(τK)h(\tau_{K}) of any hh-tiling τK\tau_{K} of a finite simplicial complex KK coincides with the hh-vector of KK as soon as h0(τK)=1h_{0}(\tau_{K})=1 and in any cases, two hh-tilings τK\tau_{K} and τK\tau^{\prime}_{K} of KK have same hh-vector as soon as h0(τK)=h0(τK)h_{0}(\tau_{K})=h_{0}(\tau^{\prime}_{K}), by Theorem 4.94.9 and Corollary 4.104.10 of [17].

Example 2.8.

The boundary of an nn-simplex admits hh-tilings using exactly one (n1)(n-1)-dimensional tile of each order.

The cc-vector, or critical vector, of a Morse tiling encodes the Euler characteristic of the tiled simplicial complex. Indeed,

Lemma 2.9.

Let KK be an nn-dimensional finite simplicial complex equipped with a Morse tiling τK\tau_{K}. Then, its Euler characteristic satisfies χ(K)=k=0n(1)kck(τK)\chi(K)=\sum_{k=0}^{n}(-1)^{k}c_{k}(\tau_{K}).

Proof.

Let us equip the underlying topological space |K||K| with its structure of cellular complex given by open simplices and compute χ(K)\chi(K) as the alternate sum of the dimensions of its cellular chain complexes. By Lemma 2.52.5 of [19], the contribution of each regular Morse tile to this count vanishes while a critical tile of index kk contributes as (1)k(-1)^{k}. Hence the result. ∎

We finally recall the definition of Morse shellability given in [19].

Definition 2.10 (Definition 2.142.14 of [19]).

A subset SS of the underlying topological space |K||K| of a finite simplicial complex KK is said to be Morse shellable iff it admits a Morse tiling together with a filtration S1SN=S\emptyset\subset S_{1}\subset\dots\subset S_{N}=S of Morse tiled subsets such that for every i{2,,N}i\in\{2,\dots,N\}, SiSi1S_{i}\setminus S_{i-1} is a single tile of the tiling.

A Morse tiled subset of SS is a union of tiles which is closed in SS, see Definition 99 of [19]. When the tiling uses only basic tiles, this notion of Morse shelling coincides with the classical notion of shelling, without the non-emptyness assumption though, see §2.1, Theorem 2.152.15 and Remark 2.162.16 of [19]. A finite simplicial complex, when equipped with a Morse tiling, carries discrete vector fields which are compatible with the tiling and in the case of a Morse shelling, any of these is the gradient vector field of a discrete Morse function in the sense of R. Forman [6], whose critical points are in one-to-one correspondence, preserving the index, with the critical tiles of the tiling, see Theorem 1.21.2 of [19]. The Betti numbers of a Morse shelled finite simplicial complex thus get bounded from above by the number of critical tiles of the corresponding index of the shelling, see Corollary 1.51.5 of [19].

2.3 Tame Morse shellings

In order to get a Morse shelling on the product of two Morse shelled complexes, we need the shellings to satisfy some tameness condition which we now introduce.

Proposition 2.11.

Let KK be a finite simplicial complex whose edges are oriented. Then, the following properties are equivalent:

  1. 1.

    There is no triangle in KK whose boundary is an oriented one-cycle.

  2. 2.

    For every simplex of KK, the relation ”xyx\leq y iff x=yx=y or the edge between xx and yy is oriented from xx to yy” defines a total order on its vertices.

Moreover, under these conditions, the inclusion of faces define increasing maps, that is they preserve the order on the vertices.

Proof.

The second condition implies the first one by transitivity. Indeed, if x,y,zx,y,z denote the three vertices of a triangle θ\theta and if the edges are oriented from xx to yy and from yy to zz, then by transitivity of the order, xzx\leq z, so that the edge between xx and zz cannot be oriented from zz to xx, it would imply x=zx=z by antisymmetry and this order wouldn’t be total. Conversely, let σ\sigma be any simplex of KK, the relation defined in the second property is reflexive by definition and antisymmetric since two different vertices x,yx,y are joined by a unique edge so that xyx\leq y and yxy\leq x cannot happen unless x=yx=y. Now the transitivity follows from the first property. Indeed, if x,yx,y and zz are three different vertices of σ\sigma, then we may assume that the edge between xx and yy is oriented from xx to yy and that the edge between yy and zz is oriented from yy to zz. Let θ\theta be the face with vertices x,y,zx,y,z. By the first property, the edge between xx and zz has to be oriented from xx to zz. Hence the transitivity. This order relation is then total since any two vertices of σ\sigma are connected by an edge. ∎

Example 2.12.

1) If the vertices of a finite simplicial complex are totally ordered, then this order induces an orientation on every edge, from the minimal vertex to the maximal one, and the conditions of Proposition 2.11 get satisfied.

2) The boundary θ\partial\theta of a triangle θ\theta satisfies the properties of Proposition 2.11 whatever the orientations on its edges are, since it contains no two-simplex. However, θ\theta itself equipped with such orientations need not satisfy these properties.

The first part of Example 2.12 shows that it is always possible to orient the one-skeleton of a finite simplicial complex KK in order to define in a compatible way a total order on the vertices of each of its simplices, turning it into an ordered simplicial complex in the sense of Definition II.8.7II.8.7 of [5]. When KK is equipped with a Morse tiling, we would like in addition that for every Morse tile with underlying simplex σ\sigma and non-empty Morse face μ\mu, the vertices of μ\mu are the maximal ones among the ones of σ\sigma. Recall that the link Lkσ(μ)\textup{Lk}_{\sigma}(\mu) of μ\mu in σ\sigma is by definition the convex hull of the vertices of σμ\sigma\setminus\mu. We thus would like that the edges between Lkσ(μ)\textup{Lk}_{\sigma}(\mu) and μ\mu are oriented from Lkσ(μ)\textup{Lk}_{\sigma}(\mu) to μ\mu, see Figure 2.

Refer to caption
Figure 2: A Morse face μ\mu in a three-simplex.
Definition 2.13.

The tiling of a Morse tiled set SS is said to be tame iff there exists an orientation on the one-skeleton of the underlying simplicial complex KK which satisfies the following order and tameness conditions.

  1. 1.

    There is no triangle in KK whose boundary is an oriented one-cycle.

  2. 2.

    For every Morse tile with underlying simplex σ\sigma and non-empty Morse face μ\mu, the edges between Lkσ(μ)\textup{Lk}_{\sigma}(\mu) and μ\mu are oriented from Lkσ(μ)\textup{Lk}_{\sigma}(\mu) to μ\mu.

Every hh-tiling is tame by the first part of Example 2.12, since the second condition of Definition 2.13 is then empty and the first one satisfied. In fact, the order condition in Definition 2.13 provides a structure of ordered simplicial complex on KK given by Proposition 2.11, see Definition II.8.7II.8.7 of [5], while the tameness condition requires some compatibility between this structure and the tiling.

The tameness property gets satisfied by any Morse tiling after one barycentric subdivision for example.

Proposition 2.14.

The first barycentric subdivision of any Morse tiled (resp. shelled) set carries tame Morse tilings (resp. tame Morse shellings) containing the same number of critical tiles with the same indices.

Proof.

Let SS be a Morse tiled (resp. Morse shelled) set and let KK be its underlying simplicial complex. By Corollary 2.212.21 of [19], the first barycentric subdivision Sd(S)\textup{Sd}(S) of SS carries Morse tilings (resp. shellings) having the same number of critical tiles with the same indices. Its underlying simplicial complex is Sd(K)\textup{Sd}(K). Now, the one-skeleton of the latter is canonically oriented. Indeed, its vertices are by definition the barycenters σ^\hat{\sigma} of the simplices σ\sigma of KK while an edge connects two vertices σ^\hat{\sigma} and τ^\hat{\tau} iff σ\sigma is a face of τ\tau or vice-versa, see [16]. Let us orient such an edge from σ^\hat{\sigma} to τ^\hat{\tau} iff τ\tau is a face of σ\sigma. The order condition of Definition 2.13 gets satisfied by this order on the one-skeleton of Sd(K)\textup{Sd}(K). We have to prove that the tameness condition gets satisfied as well. Let TT^{\prime} be a Morse tile of Sd(S)\textup{Sd}(S) with underlying simplex σ\sigma^{\prime} and non-empty Morse face μ\mu^{\prime}. By construction, there exists a Morse tile TT of SS, with underlying simplex σ\sigma and non-empty Morse face μ\mu such that σSd(σ)\sigma^{\prime}\subset\textup{Sd}(\sigma) and μ=Sd(μ)σ\mu^{\prime}=\textup{Sd}(\mu)\cap\sigma^{\prime}, see [19]. There exists then a maximal flag σ0σ1σn=σ\sigma_{0}\subset\sigma_{1}\subset\dots\subset\sigma_{n}=\sigma such that the vertices of σ\sigma^{\prime} are the barycenters σ^i\hat{\sigma}_{i} of σi\sigma_{i}, i{0,,n}i\in\{0,\dots,n\}, where nn denotes the dimension of TT^{\prime}. By maximal flag we mean that for every 0in0\leq i\leq n, dimσi=i\dim\sigma_{i}=i and for every 0i<jn0\leq i<j\leq n, σi\sigma_{i} is a face of σj\sigma_{j}. Now such a vertex σ^j\hat{\sigma}_{j} belongs to μ\mu^{\prime} iff σj\sigma_{j} is a face of μ\mu and this then implies that σ^iμ\hat{\sigma}_{i}\in\mu^{\prime} for all iji\geq j, since then σiσj\sigma_{i}\subset\sigma_{j}. The vertices of σ\sigma^{\prime} that belong to the Morse face μ\mu^{\prime} are thus the maximal ones with respect to this canonical order. Hence the result. ∎

We may finally provide a criterium which ensures that a Morse tiling is tame.

Proposition 2.15.

A Morse tiling is tame if σμ=σμ\sigma\cap\mu^{\prime}=\sigma^{\prime}\cap\mu for every tiles T,TT,T^{\prime} with underlying simplices σ,σ\sigma,\sigma^{\prime} and non-empty Morse faces μ,μ\mu,\mu^{\prime}.

For example, the condition in Proposition 2.15 gets satisfied if all tiles with non-empty Morse faces have disjoint underlying simplices.

Proof.

Let KK be the underlying simplicial complex of such a Morse tiled set SS and let LL be the union of all Morse faces of its tiles. Let us fix a total order on the vertices of LL and a total order on the vertices of KLK\setminus L. Then, all edges of KK whose vertices are both in LL or both outside LL get oriented by these total orders from the minimal vertex to the maximal one. We finally orient the edges between KLK\setminus L and LL from KLK\setminus L to LL. These orientations satisfy the properties of Proposition 2.11, compare Example 2.12. Moreover, every tile TT with underlying simplex σ\sigma and non-empty Morse face μ\mu satisfies σL=μ\sigma\cap L=\mu by hypothesis and its edges between Lkσ(μ)\textup{Lk}_{\sigma}(\mu) and μ\mu are oriented from Lkσ(μ)\textup{Lk}_{\sigma}(\mu) to μ\mu by construction, so that the tiling is indeed tame by definition. ∎

3 Main results

3.1 Palindromic vectors

Let nn be a non-negative integer, the involution k{0,,n}nk{0,,n}k\in\{0,\dots,n\}\mapsto n-k\in\{0,\dots,n\} induces the automorphism v=(v0,,vn)nvˇ=(vn,,v0)nv=(v_{0},\dots,v_{n})\in\mathbb{R}^{n}\mapsto\check{v}=(v_{n},\dots,v_{0})\in\mathbb{R}^{n}.

Definition 3.1.

A vector vv of n\mathbb{R}^{n} is said to be palindromic iff vˇ=v\check{v}=v.

For example, the real Betti numbers of closed connected oriented manifolds define palindromic vectors by Poincaré duality, see [16, 2]. The hh-vectors of convex polytopes are palindromic as well, see [15, 22, 8, 24]. We are going to prove that hh-vectors of Morse tilings are likewise often palindromic.

Theorem 3.2.

Let KK be an nn-dimensional simplicial complex homeomorphic to a closed manifold and equipped with an hh-tiling τ\tau. Then,

  1. 1.

    If nn is odd, the hh-vector of τ\tau is palindromic.

  2. 2.

    If nn is even, for every i{0,,n+1}i\in\{0,\dots,n+1\},

    hi(τ)hn+1i(τ)=(1)i(n+1i)(h0(τ)hn+1(τ)).h_{i}(\tau)-h_{n+1-i}(\tau)=(-1)^{i}{n+1\choose i}(h_{0}(\tau)-h_{n+1}(\tau)).
Proof.

By Theorem 4.94.9 of [17], the hh-vector of τ\tau satisfies i=0n+1hi(τ)Xn+1i=i=0n+1fi1(K)(X1)n+1i\sum_{i=0}^{n+1}h_{i}(\tau)X^{n+1-i}=\sum_{i=0}^{n+1}f_{i-1}(K)(X-1)^{n+1-i} provided one sets f1(K)=h0(τ)f_{-1}(K)=h_{0}(\tau). By Theorem 2.12.1 of [14], the Dehn-Sommerville relations can be expressed by the relation RK(1X)=(1)n+1RK(X)R_{K}(-1-X)=(-1)^{n+1}R_{K}(X), where RK(X)=Xi=0nfi(X)Xiχ(K)XR_{K}(X)=X\sum_{i=0}^{n}f_{i}(X)X^{i}-\chi(K)X, see also Theorem 1.11.1 of [18]. Finally, we know from Lemma 2.9 that the Euler characteristic of KK satisfies χ(K)=h0(τ)+(1)nhn+1(τ)\chi(K)=h_{0}(\tau)+(-1)^{n}h_{n+1}(\tau), since the only critical tiles of an hh-tiling are the open and closed simplices. We deduce that RK(X)=i=0n+1hi(τ)Xi(X+1)n+1ih0(τ)χ(K)XR_{K}(X)=\sum_{i=0}^{n+1}h_{i}(\tau)X^{i}(X+1)^{n+1-i}-h_{0}(\tau)-\chi(K)X, so that Macdonald’s result [14] becomes

i=0n+1(hi(τ)hn+1i(τ))Xi(X+1)n+1i(h0(τ)hn+1(τ))=0 if n is even and\displaystyle\sum_{i=0}^{n+1}\big{(}h_{i}(\tau)-h_{n+1-i}(\tau)\big{)}X^{i}(X+1)^{n+1-i}-\big{(}h_{0}(\tau)-h_{n+1}(\tau)\big{)}=0\text{ if }n\text{ is even and}
i=0n+1(hi(τ)hn+1i(τ))Xi(X+1)n+1i(h0(τ)hn+1(τ))(1+2X)=0 if n is odd.\displaystyle\sum_{i=0}^{n+1}\big{(}h_{i}(\tau)-h_{n+1-i}(\tau)\big{)}X^{i}(X+1)^{n+1-i}-\big{(}h_{0}(\tau)-h_{n+1}(\tau)\big{)}(1+2X)=0\text{ if }n\text{ is odd.}

If nn is odd, the Euler characteristic of a closed nn-dimensional manifold vanishes from Poincaré duality, see [16] for example, so that h0(τ)=hn+1(τ)h_{0}(\tau)=h_{n+1}(\tau). We thus deduce in this case that h(τ)h(\tau) is palindromic. If nn is even, we observe that 1=i=0n+1(1)i(n+1i)Xi(X+1)n+1i1=\sum_{i=0}^{n+1}(-1)^{i}{n+1\choose i}X^{i}(X+1)^{n+1-i} and get the result, since the monomials (Xi(X+1)n+1i)i{0,,n+1}(X^{i}(X+1)^{n+1-i})_{i\in\{0,\dots,n+1\}} are linearly independant over \mathbb{R}. ∎

Corollary 3.3.

The hh-vector of any hh-tiling on a simplicial complex homeomorphic to a closed manifold is palindromic iff its cc-vector is palindromic.

Proof.

The hh-tiling τ\tau of a simplicial complex homeomorphic to a closed nn-dimensional manifold only contains nn-dimensional tiles by Lemma 2.6 and the singular ones are the open and closed simplices by definition. Thus, if h(τ)h(\tau) is palindromic, h0(τ)=hn+1(τ)h_{0}(\tau)=h_{n+1}(\tau), so that c0(τ)=cn(τ)c_{0}(\tau)=c_{n}(\tau) which means that c(τ)c(\tau) is palindromic as well. Converserly, if c(τ)c(\tau) is palindromic, then c0(τ)=cn(τ)c_{0}(\tau)=c_{n}(\tau), so that h0(τ)=hn+1(τ)h_{0}(\tau)=h_{n+1}(\tau) and the result follows from Theorem 3.2. ∎

Example 3.4.

1) The boundary of a (n+1)(n+1)-simplex is shellable and the associated hh-tiling uses one nn-dimensional basic tile of each order. Its hh-vector (1,,1)(1,\dots,1) is thus palindromic.

2) The boundary of a triangle is also tiled by three one-dimensional tile of order one, see Figure 3. The associated hh-vector (0,3,0)(0,3,0) is palindromic.

Refer to caption
Figure 3: The non-shellable tiling on Δ2\partial\Delta_{2}.

3) The cylinder Δ1×Δ2\Delta_{1}\times\partial\Delta_{2} has a triangulation tiled by six basic tiles of order one, obtained by gluing three copies of the square Δ1×T11\Delta_{1}\times T^{1}_{1} pictured in Figure 4. By caping this cylinder with two open triangles, we get an hh-tiled triangulation on the two-sphere for which neither the hh-vector (0,6,0,2)(0,6,0,2) nor the critical vector (0,0,2)(0,0,2) are palindromic.

Refer to caption
Figure 4: An hh-tiling on Δ1×T11\Delta_{1}\times T^{1}_{1}.

Example 3.4 exhibits in particular an hh-tiling with non-palindromic hh-vector on the triangulated two-sphere. Does there exist such hh-tilings on the other even-dimensional spheres?

In the case of Morse tilings, we observe.

Theorem 3.5.

Let KK be an nn-dimensional simplicial complex homeomorphic to a closed manifold of dimension at most three and equipped with a Morse tiling τ\tau. Then, the following three conditions are equivalent.

  1. 1.

    The tiling τ\tau uses as many open simplices as closed simplices.

  2. 2.

    The cc-vector of τ\tau is palindromic.

  3. 3.

    The hh-vector of τ\tau is palindromic.

Proof.

The implications 212\Rightarrow 1 and 313\Rightarrow 1 hold true in any dimension, while 121\Rightarrow 2 is obvious in dimension at most two and follows from Lemma 2.9 in dimension three, since the Euler characteristic of KK then vanishes from Poincaré duality, see [16]. Let us thus assume that 11 holds true and prove the implication 131\Rightarrow 3. By the simplest Dehn-Sommerville relation, every (n1)(n-1)-simplex σ\sigma of KK is the face of exactly two nn-simplices. The interiors of these two simplices are covered by two tiles of the tiling and the open face σ\stackrel{{\scriptstyle\circ}}{{\sigma}} is a facet of one of them and a missing facet of the other since the tiling defines a partition of KK. It follows that the total number of facets of the tiles of τ\tau coincides with the total number of missing facets of these tiles. By 11, the contributions to these totals of the tiles of order 0 and n+1n+1 coincide. If n=3n=3, the same holds true for tiles of order two since they have both two facets and two missing facets. In dimension two (resp. three), we deduce that τ\tau uses as many tiles of order one as tiles of order two (resp. three), so that h(τ)h(\tau) is palindromic. ∎

Example 3.6.

The octahedron carries a Morse tiling with non palindromic cc-vector and hh-vector. It is obtained by patching the two tiled squares pictured in Figure 5, so that its hh-vector (resp. cc-vector) equals (1,4,1,2)(1,4,1,2) (resp. (1,1,2)(1,1,2)).

Refer to caption
Figure 5: A Morse tiling on the octahedron.

The tiling given by Example 3.6 contains two critical tiles of indices one and two which could be replaced by two regular tiles of order two to produce a tiling with palindromic hh-vector. Such examples with non-palindromic hh-vectors and cc-vectors can be obtained in a similar way in higher dimensions. But we do not know any Morse tiled closed triangulated manifold with palindromic critical vector and non-palindromic hh-vector.

3.2 Tilings of products

Recall that the product of two simplicial complexes is not a simplicial complex, it is a polyhedral complex whose cells are products of two simplices. Such a product can nevertheless be triangulated in such a way that each product of two simplices becomes the union of simplices of the underlying affine space, see §II.8II.8 of [5], [9, 20] and §4.3. We are going to consider such triangulations, which are primitive in the sense of Definition 3.7 and in fact associated to staircases, see §4.1 and [9].

Definition 3.7.

A primitive triangulation of a polyhedral complex is a triangulation having the same set of vertices.

Our main result is the following Theorem 3.8, where we denote by uvuv the graded product of a vector u=(u0,,un)u=(u_{0},\dots,u_{n}) of n+1\mathbb{R}^{n+1} with a vector v=(v0,,vm)v=(v_{0},\dots,v_{m}) of m+1\mathbb{R}^{m+1}, that is the product of the corresponding polynomials, so that uv=(w0,,wn+m)uv=(w_{0},\dots,w_{n+m}) where for every k{0,,n+m}k\in\{0,\dots,n+m\}, wk=j=0kujvkjw_{k}=\sum_{j=0}^{k}u_{j}v_{k-j}.

Theorem 3.8.

Let S1S_{1} and S2S_{2} be two Morse tiled (resp. shelled) sets with tame tilings (resp. shellings) τ1\tau_{1} and τ2\tau_{2}. Then, S1×S2S_{1}\times S_{2} carries tame Morse tileable (resp. shellable) primitive triangulations with critical vector c(τ1)c(τ2)c(\tau_{1})c(\tau_{2}). Moreover, if τ1\tau_{1}, τ2\tau_{2} are pure dimensional, their hh-vector is palindromic provided h(τ1)h(\tau_{1}) and h(τ2)h(\tau_{2}) are.

By triangulation of S1×S2S_{1}\times S_{2} we mean triangulations on the product of the underlying simplicial complexes. We do not guarantee Morse tileability for all primitive triangulations on this product, the ones for which we do by Theorem 3.8 are given by the tameness of the tilings, see §6.1. Let us also recall that all hh-tilings are pure dimensional in the case of triangulated manifolds by Lemma 2.6. We deduce Theorem 1.2, namely.

Corollary 3.9.

Every finite product of closed manifolds of dimensions less than four carries triangulations which admit tame Morse shellings with palindromic cc-vectors and hh-vectors.

Proof.

By Theorem 1.41.4 of [19], every closed connected manifold of dimension less than four carries a Morse shellable triangulation which can moreover be chosen in such a way that the Morse shelling uses a unique critical tile of index 0 and a unique critical tile of maximal index, for there exists a Morse function on this manifold having a single minimum and a single maximum, see [4]. By Proposition 2.14, such a Morse shelling becomes tame after one barycentric subdivision and it keeps the property to use only one closed and one open simplex. By Theorem 3.5, its critical and h-vector are then palindromic. The result now follows by finite induction from Theorem 3.8. ∎

When one of the tilings τ1\tau_{1}, τ2\tau_{2} uses only regular tiles, the tilings given by Theorem 3.8 share the same property since c(τ1)c(τ2)c(\tau_{1})c(\tau_{2}) then vanishes. However, it is not supposed to be an hh-tiling even if τ1\tau_{1} and τ2\tau_{2} are, for c(τ1)c(τ2)c(\tau_{1})c(\tau_{2}) has more than two non-vanishing entries in general. The following variant of Theorem 3.8 fills this gap.

Theorem 3.10.

Let S1S_{1}, S2S_{2} be two hh-tiled sets and let the one-skeleton of their underlying simplicial complexes K1K_{1}, K2K_{2} be equipped with orientations given by Proposition 2.11. Then, the tiling of S1×S2S_{1}\times S_{2} given by Theorem 3.8 is an hh-tiling provided that if S1S_{1} (resp. S2S_{2}) contains a tile which has been deprived both of its facet not containing the biggest vertex and its facet not containing the least vertex, then every tile of S2S_{2} (resp. S1S_{1}) has been deprived either of its facet not containing the biggest vertex or of its facet not containing the least vertex.

Recall that the total orders on the vertices of the simplices of K1K_{1} and K2K_{2} are given by Proposition 2.11. If the hh-tiling of S1S_{1} (resp. S2S_{2}) contains an open simplex, then the hh-tiling of S2S_{2} (resp. of S1S_{1}) has in particular to be regular for Theorem 3.10 to apply. Now, if S2S_{2} is the non-shellable tiling of Δ2\partial\Delta_{2} given in the second part of Example 3.4, then Theorem 3.10 applies whatever S1S_{1} is. We may compute the hh-vector of S1×S2S_{1}\times S_{2} in this case.

Theorem 3.11.

Let SS be an hh-tiled set of pure dimension nn. Then, S×Δ2S\times\partial\Delta_{2} carries primitive triangulations which admit hh-tilings τ\tau such that h0(τ)=hn+2(τ)=0h_{0}(\tau)=h_{n+2}(\tau)=0 and for every j{1,,n+1}j\in\{1,\dots,n+1\}, hj(τ)=jhj(S)+(n+2j)hj1(S)h_{j}(\tau)=jh_{j}(S)+(n+2-j)h_{j-1}(S).

In Theorem 3.11 again, h(τ)h(\tau) is then palindromic as soon as h(S)h(S) is, or provided SS is homeomorphic to a closed manifold by Corollary 3.3, since c(τ)c(\tau) is palindromic. This result implies Theorem 1.1, showing that the product of a sphere and a torus of any dimensions carries hh-tileable triangulations. More precisely, we deduce the following Corollary 3.13.

Definition 3.12.

A walk of length mm from the integer aa to bb is a sequence wn,m=(wn,m(i))i{n,,n+m}w_{n,m}=\big{(}w_{n,m}(i)\big{)}_{i\in\{n,\dots,n+m\}} such that wn,m(n)=aw_{n,m}(n)=a, wn,m(n+m)=bw_{n,m}(n+m)=b and for every i{n,,n+m1}i\in\{n,\dots,n+m-1\}, either wn,m(i+1)=wn,m(i)w_{n,m}(i+1)=w_{n,m}(i) or wn,m(i+1)=wn,m(i)+1w_{n,m}(i+1)=w_{n,m}(i)+1. The weight of such a walk is the product p(wn,m)=Πi=nn+m1pi(wn,m)p(w_{n,m})=\Pi_{i=n}^{n+m-1}p_{i}(w_{n,m}), where pi(wn,m)=wn,m(i)p_{i}(w_{n,m})=w_{n,m}(i) if wn,m(i+1)=wn,m(i)w_{n,m}(i+1)=w_{n,m}(i) and pi(wn,m)=i+1wn,m(i)p_{i}(w_{n,m})=i+1-w_{n,m}(i) if wn,m(i+1)=wn,m(i)+1w_{n,m}(i+1)=w_{n,m}(i)+1.

Corollary 3.13.

For every n0n\geq 0 and m1m\geq 1, the product Δn+1×(Δ2)m\partial\Delta_{n+1}\times(\partial\Delta_{2})^{m} carries primitive triangulations which admit hh-tilings using no critical tile. Moreover, the hh-vector of such an hh-tiling τ\tau is palindromic and satisfies, for every k{0,,n+m+1}k\in\{0,\dots,n+m+1\}, hk(τ)=wn,mp(wn,m)h_{k}(\tau)=\sum_{w_{n,m}}p(w_{n,m}), where the sum is taken over all walks of length mm from an element of {0,,n+1}\{0,\dots,n+1\} to kk.

Proof.

Let us equip Δn+1\partial\Delta_{n+1} with the shelling given in the first part of Example 3.4, which uses one tile of each order k{0,,n+1}k\in\{0,\dots,n+1\}. The result follows by induction on mm, by successive applications of Theorem 3.11. Indeed, when m=1m=1, if k{0,n+2}k\in\{0,n+2\}, the walk is unique and its weight vanishes. If kk belongs to {1,,n+1}\{1,\dots,n+1\}, there are two walks leading to kk, namely (k,k)(k,k) and (k1,k)(k-1,k). The weight of the first one is kk by Definition 3.12 and the weight of the second one is n+2kn+2-k. By Theorem 3.11, hk(Δn+1×Δ2)=k+(n+2k)h_{k}(\partial\Delta_{n+1}\times\partial\Delta_{2})=k+(n+2-k) coincides with the sum of these walks. Let us now assume that the result is proven up to the rank mm and let us prove it for m+1m+1. Again, for k=0k=0 or n+m+2n+m+2, the walk leading to kk is unique and its weight vanishes. When kk belongs to {1,,n+m+1}\{1,\dots,n+m+1\}, a walk wn,m+1w_{n,m+1} leading to kk restricts either to a walk wn,mw_{n,m} leading to kk or to a walk wn,mw_{n,m} leading to k1k-1 while conversely, any such walk extends uniquely to a walk wn,m+1w_{n,m+1} leading to kk. Moreover, by Definition 3.12, in the first case, p(wn,m+1)=p(wn,m)kp(w_{n,m+1})=p(w_{n,m})k and in the second, p(wn,m+1)=p(wn,m)(n+m+2k)p(w_{n,m+1})=p(w_{n,m})(n+m+2-k), while by the induction hypothesis, wn,mp(wn,m)=hk(Δn+1×(Δ2)m)\sum_{w_{n,m}}p(w_{n,m})=h_{k}(\partial\Delta_{n+1}\times(\partial\Delta_{2})^{m}) (resp. wn,mp(wn,m)=hk1(Δn+1×(Δ2)m)\sum_{w_{n,m}}p(w_{n,m})=h_{k-1}(\partial\Delta_{n+1}\times(\partial\Delta_{2})^{m})) in the first case (resp. in the second case). The result now follows from Theorem 3.11 with S=Δn+1×(Δ2)mS=\partial\Delta_{n+1}\times(\partial\Delta_{2})^{m}, the palindromic property being ensured by Corollary 3.3. ∎

The hh-vector hn,mh^{n,m} of the hh-tilings given by Corollary 3.13 has its own interest. By Theorem 4.94.9 of [17], it does not depend on the choice of the tiling but differs from the hh-vector of the underlying primitive triangulation, see Corollary 4.104.10 of [17]. What is the asymptotic of hn,mh^{n,m} as mm grows to ++\infty? More precisely, what is the limit, as mm grows to ++\infty, of the probability measure

1k=0n+m+1hkn,mk=0n+m+1hkn,mδ2knm1n+m+1,\frac{1}{\sum_{k=0}^{n+m+1}h^{n,m}_{k}}\sum_{k=0}^{n+m+1}h^{n,m}_{k}\delta_{\frac{2k-n-m-1}{n+m+1}},

where δx\delta_{x} denotes the Dirac measure at xx\in\mathbb{R}?

We also do not know which closed manifolds carry hh-tileable triangulations. They have non-negative Euler characteristic by Lemma 2.9.

3.3 Tilings of handles and duality

The sets S1,S2S_{1},S_{2} in Theorems 3.8 and 3.10 may just consist of single tiles. In fact, these results follow from this special case to which we devote this section.

Theorem 3.14.

Let T1T_{1} and T2T_{2} be two basic tiles, one of which being regular. Then, T1×T2T_{1}\times T_{2} carries shellable primitive triangulations using only regular tiles in their shelling. If T1T_{1} and T2T_{2} are Morse tiles, then T1×T2T_{1}\times T_{2} carries Morse shellable primitive triangulations using a critical tile iff both T1T_{1} and T2T_{2} are critical and this critical tile is then unique of index the sum of the indices of T1T_{1} and T2T_{2}. Moreover, all these shellings are tame and pure dimensional.

These triangulations and shellings given by Theorem 3.14 are inherited from particular total orders on the vertices of the underlying simplices of T1T_{1} and T2T_{2}, see §5.2.4. They all have same hh-vector and satisfy the following duality property.

Definition 3.15.

Let σ\sigma be an nn-simplex and σ0,,σn\sigma_{0},\dots,\sigma_{n} be its facets. For every J{0,,n}J\subset\{0,\dots,n\}, the basic tiles T=σjJσjT=\sigma\setminus\bigcup_{j\in J}\sigma_{j} and T𝑤𝑖𝑑𝑒𝑐ℎ𝑒𝑐𝑘=σj{0,,n}Jσj\widecheck{T}=\sigma\setminus\bigcup_{j\in\{0,\dots,n\}\setminus J}\sigma_{j} are said to be dual to each other.

Theorem 3.16.

Let T1,T𝑤𝑖𝑑𝑒𝑐ℎ𝑒𝑐𝑘1T_{1},\widecheck{T}_{1} (resp. T2,T𝑤𝑖𝑑𝑒𝑐ℎ𝑒𝑐𝑘2T_{2},\widecheck{T}_{2}) be two Morse tiles whose underlying basic tiles are dual to each other. Then, the Morse shellings on T1×T2T_{1}\times T_{2} (resp. T𝑤𝑖𝑑𝑒𝑐ℎ𝑒𝑐𝑘1×T𝑤𝑖𝑑𝑒𝑐ℎ𝑒𝑐𝑘2\widecheck{T}_{1}\times\widecheck{T}_{2}) given by Theorem 3.14 all have same hh-vector and satisfy h(T𝑤𝑖𝑑𝑒𝑐ℎ𝑒𝑐𝑘1×T𝑤𝑖𝑑𝑒𝑐ℎ𝑒𝑐𝑘2)=h𝑤𝑖𝑑𝑒𝑐ℎ𝑒𝑐𝑘(T1×T2)h(\widecheck{T}_{1}\times\widecheck{T}_{2})=\widecheck{h}(T_{1}\times T_{2}).

In Theorem 3.16, hwidecheck(T1×T2)\widecheck{h}(T_{1}\times T_{2}) denotes the image of the hh-vector h(T1×T2)h(T_{1}\times T_{2}) under the palindromic automorphism defined in §3.1.

Theorem 3.14 provides in particular Morse shellings on every handle, whatever its index is, where by handle of index kk and dimension nn, we mean the product Δk×Δnk\stackrel{{\scriptstyle\circ}}{{\Delta}}_{k}\times\Delta_{n-k} as defined in [19]. Such a Morse shelling has already been obtained in index 11 and n1n-1, see Corollary 3.173.17 of [19].

Corollary 3.17.

For every 0kn0\leq k\leq n, the handle Δk×Δnk\stackrel{{\scriptstyle\circ}}{{\Delta}}_{k}\times\Delta_{n-k} carries Morse shellable primitive triangulations using a unique critical tile, of index kk. \square

Recall that Theorem 1.41.4 of [19], which provides Morse shelled triangulations on every closed three-manifold, has been obtained by successive attachments of such Morse shelled triangulated handles. We end this section by giving other remarkable shellings given by Theorem 3.14.

Corollary 3.18.

For every m,n>0m,n>0, Tnn×ΔmT_{n}^{n}\times\Delta_{m} (resp. T1n×ΔmT_{1}^{n}\times\stackrel{{\scriptstyle\circ}}{{\Delta}}_{m}) carries shellable primitive triangulations using basic tiles which are all isomorphic to each other, of order nn (resp. of order m+1m+1).

Recall that TnnT_{n}^{n} (resp. T1nT_{1}^{n}) denotes a basic tile of dimension nn and order nn (resp. order one), see §2.2.

Corollary 3.19.

Let T1T_{1} (resp T2T_{2}) be a Morse tile of odd dimension nn (resp. mm) and of order m+12\frac{m+1}{2} (resp. n+12\frac{n+1}{2}). Then, T1×T2T_{1}\times T_{2} carries Morse shellable primitive triangulations with palindromic hh-vector. Moreover, if the tiles are basic, these Morse shellings can be chosen to be shellings.

Proof.

Theorem 3.14 provides a Morse shelled triangulation on T1×T2T_{1}\times T_{2} and even a shelled triangulation if these tiles are basic, since they are regular. By hypothesis, the basic tile underlying T1T_{1} is isomorphic to its dual and likewise, the basic tile underlying T2T_{2} is isomorphic to its dual. The hh-vectors of T1×T2T_{1}\times T_{2} and Twidecheck1×Twidecheck2\widecheck{T}_{1}\times\widecheck{T}_{2} thus coincide, while they are dual to each other by Theorem 3.16. They must then be palindromic. ∎

4 Cartesian product of two simplices

The cartesian product of two simplices is a structure of simplicial complex on their product which is inherited from total orders on their vertices, see Definition II.8.8II.8.8 of [5]. We study these primitive triangulations in this section, whose simplices are associated to staircases, together with their shellings, see §4.3. The Cayley trick helps to visualize them, via the mixed decompositions induced on one of the simplices, see §4.2 and [20, 10].

4.1 Staircases

Let m,nm,n be two non-negative integers. We denote by C(n,m)C(n,m) the set of increasing -not strictly increasing- functions f:{0,,n}{0,,m}f:\{0,\dots,n\}\to\{0,\dots,m\} such that f(n)=mf(n)=m and by N(n,m)N(n,m) its cardinality. We recall that.

Lemma 4.1.

For every m,n0m,n\geq 0, N(n,m)=(n+mn).N(n,m)={n+m\choose n}.

Proof.

To every fC(n,m)f\in C(n,m) we may associate f~:k{0,,n}f(k)+k{0,,m+n}\tilde{f}:k\in\{0,\dots,n\}\mapsto f(k)+k\in\{0,\dots,m+n\}. This correspondence between C(n,m)C(n,m) and the set of strictly increasing functions {0,,n}{0,,m+n}\{0,\dots,n\}\to\{0,\dots,m+n\} such that f(n)=m+nf(n)=m+n is bijective. Moreover, the image of such a function is a subset of {0,,m+n}\{0,\dots,m+n\} containing m+nm+n and of cardinality n+1n+1 while every subset sharing these properties defines a strictly increasing map {0,,n}{0,,m+n}\{0,\dots,n\}\to\{0,\dots,m+n\} such that f(n)=m+nf(n)=m+n. The result follows. ∎

The space C(n,m)C(n,m) is equipped with the involution fC(n,m)fˇC(n,m)f\in C(n,m)\mapsto\check{f}\in C(n,m), where for every j{0,,n1}j\in\{0,\dots,n-1\}, fˇ(j)=mf(n1j)\check{f}(j)=m-f(n-1-j) and fˇ(n)=m\check{f}(n)=m. Also, the lexicographic order on the n-tuples of integers induces a total order on C(n,m)C(n,m), so that for every f,gC(n,m)f,g\in C(n,m), fgf\leq g iff (f(0),,f(n1))(g(0),,g(n1))(f(0),\dots,f(n-1))\leq(g(0),\dots,g(n-1)). The minimum of C(n,m)C(n,m) is thus a function which vanishes on {0,,n1}\{0,\dots,n-1\} while its maximum is the constant function equal to mm.

Likewise, we denote by I(n,m)I(n,m) the set of collections I=(Ij)j{0,,n}I=(I_{j})_{j\in\{0,\dots,n\}} of intervals Ij={i{0,,m}|bI(j)ieI(j)}I_{j}=\{i\in\{0,\dots,m\}\,|\,b_{I}(j)\leq i\leq e_{I}(j)\} which cover {0,,m}\{0,\dots,m\} and satisfy eI(j)=bI(j+1)e_{I}(j)=b_{I}(j+1) for every 0j<n0\leq j<n. In particular, bI(0)=0b_{I}(0)=0 and eI(n)=me_{I}(n)=m. This space of staircases, see §7.3.D7.3.D of [9], is equipped with the involution I=(Ij)j{0,,n}Iˇ=(Iˇj)j{0,,n}I=(I_{j})_{j\in\{0,\dots,n\}}\mapsto\check{I}=(\check{I}_{j})_{j\in\{0,\dots,n\}}, where for every j{0,,n}j\in\{0,\dots,n\}, Iˇj={meI(nj),,mbI(nj)}\check{I}_{j}=\{m-e_{I}(n-j),\dots,m-b_{I}(n-j)\}, so that bIˇ(j)=meI(nj)b_{\check{I}}(j)=m-e_{I}(n-j) and eIˇ(j)=mbI(nj)e_{\check{I}}(j)=m-b_{I}(n-j). This space also inherits a total order from the lexicographic order, so that for every I,JI(n,m)I,J\in I(n,m), IJI\leq J iff (eI(0),,eI(n1))(eJ(0),,eJ(n1))(e_{I}(0),\dots,e_{I}(n-1))\leq(e_{J}(0),\dots,e_{J}(n-1)).

These spaces of functions and staircases are in bijective correspondence. Namely, for every fC(n,m)f\in C(n,m), let us denote by IfI^{f} the element of I(n,m)I(n,m) such that eIf=fe_{I^{f}}=f.

Lemma 4.2.

The maps fC(n,m)IfI(n,m)f\in C(n,m)\mapsto I^{f}\in I(n,m) and II(n,m)eIC(n,m)I\in I(n,m)\mapsto e_{I}\in C(n,m) are bijective, /2\mathbb{Z}/2\mathbb{Z}-equivariant, order preserving and inverse one with respect to the other.

Proof.

The maps are order preserving and inverse one with respect to the other by definition. They are thus bijective as well. Now, let fC(n,m)f\in C(n,m), we have to check that Ifˇ=IfwidecheckI^{\check{f}}=\widecheck{I^{f}}. For every j{0,,n}j\in\{0,\dots,n\}, eIfˇ(j)=fˇ(j)=mf(n1j)e_{I^{\check{f}}}(j)=\check{f}(j)=m-f(n-1-j) while eIfwidecheck(j)=mbIf(nj)=meIf(n1j)=mf(n1j)e_{\widecheck{I^{f}}}(j)=m-b_{I^{f}}(n-j)=m-e_{I^{f}}(n-1-j)=m-f(n-1-j). Hence the result. ∎

Let us finally observe that exchanging the roles of nn and mm defines an involution I(n,m)I(m,n)I(n,m)\to I(m,n).

Lemma 4.3.

For every II(n,m)I\in I(n,m) and every i{0,,m}i\in\{0,\dots,m\}, set Ji={j{0,,n}|iIj}J_{i}=\{j\in\{0,\dots,n\}\,|\,i\in I_{j}\}. Then, J=(Ji)i{0,,m}J=(J_{i})_{i\in\{0,\dots,m\}} belongs to I(m,n)I(m,n) and the correspondence II(n,m)JI(m,n)I\in I(n,m)\mapsto J\in I(m,n) is bijective.

Proof.

Let II(n,m)I\in I(n,m) and i{0,,m}i\in\{0,\dots,m\}. We denote by bJ(i)b_{J}(i) (resp. eJ(i)e_{J}(i)) the least (resp. greatest) element of JiJ_{i}. If j{bJ(i),,eJ(i)}j\in\{b_{J}(i),\dots,e_{J}(i)\}, then bI(j)bI(eJ(i))ib_{I}(j)\leq b_{I}(e_{J}(i))\leq i and eI(j)eI(bJ(i))ie_{I}(j)\geq e_{I}(b_{J}(i))\geq i, so that bI(j)ieI(j)b_{I}(j)\leq i\leq e_{I}(j), that is iIji\in I_{j}. We deduce that JiJ_{i} is the interval {bJ(i),,eJ(i)}\{b_{J}(i),\dots,e_{J}(i)\}. Moreover, bJ(0)=0b_{J}(0)=0 and eJ(m)=ne_{J}(m)=n by definition. Finally, if jJij\in J_{i}, then either i=eI(j)i=e_{I}(j) and j+1=bI(j+1)Jij+1=b_{I}(j+1)\in J_{i} provided j<nj<n, or i<eI(j)i<e_{I}(j) and j=eJ(i)j=e_{J}(i). It follows that if i<mi<m, bJ(i+1)=eJ(i)b_{J}(i+1)=e_{J}(i) so that JJ(m,n)J\in J(m,n). Now, this correspondence II(n,m)JI(m,n)I\in I(n,m)\mapsto J\in I(m,n) is bijective. The preimage of an element J=(Ji)i{0,,m}J=(J_{i})_{i\in\{0,\dots,m\}} is the staircase (Ij)j{0,,n}(I_{j})_{j\in\{0,\dots,n\}} defined in a similar way, namely for every j{0,,n}j\in\{0,\dots,n\}, Ij={i{0,,m}|jJi}I_{j}=\{i\in\{0,\dots,m\}\,|\,j\in J_{i}\}. Indeed, we check likewise that II(n,m)I\in I(n,m) and for every (j,i){0,,n}×{0,,m}(j,i)\in\{0,\dots,n\}\times\{0,\dots,m\}, the conditions iIji\in I_{j} and jJij\in J_{i} are equivalent to each other, so that the maps are inverse one to another. ∎

4.2 Mixed decompositions of the simplex

Let us recall that Δ[m]\Delta_{[m]} denotes the standard mm-simplex whose vertices are labelled by the integers 0,,m0,\dots,m. Every mm-simplex whose vertices are totally ordered becomes canonically isomorphic to Δ[m]\Delta_{[m]}. Likewise, for every subset JJ of {0,,m}\{0,\dots,m\}, we denote by ΔJ\Delta_{J} the face of Δ[m]\Delta_{[m]} whose vertices belong to JJ.

Let then II(n,m)I\in I(n,m) and α=(α0,,αn)+n+1\alpha=(\alpha_{0},\dots,\alpha_{n})\in\mathbb{R}_{+}^{n+1} be such that α0++αn=1\alpha_{0}+\dots+\alpha_{n}=1. We set

ΔI,α={α0x0++αnxnΔ[m]|j{0,,n},xjΔIj}.\Delta_{I,\alpha}=\{\alpha_{0}x_{0}+\dots+\alpha_{n}x_{n}\in\Delta_{[m]}\,|\,\forall j\in\{0,\dots,n\},\,x_{j}\in\Delta_{I_{j}}\}.

When all the αj\alpha_{j}’s equal 1n+1\frac{1}{n+1}, this cell ΔI,α\Delta_{I,\alpha} is thus the rescaled Minkowski sum ΔI0++ΔIn\Delta_{I_{0}}+\dots+\Delta_{I_{n}}.

Likewise, for every j{0,,n}j\in\{0,\dots,n\}, we denote by TIjT_{I_{j}} the basic tile ΔIjΔIj{eI(j)}\Delta_{I_{j}}\setminus\Delta_{I_{j}\setminus\{e_{I}(j)\}} with the convention that Δ=\Delta_{\emptyset}=\emptyset and set

TI,α={α0x0++αnxnΔ[m]|j{0,,n1},xjTIj and xnΔIn}.T_{I,\alpha}=\{\alpha_{0}x_{0}+\dots+\alpha_{n}x_{n}\in\Delta_{[m]}\,|\,\forall j\in\{0,\dots,n-1\},\,x_{j}\in T_{I_{j}}\text{ and }x_{n}\in\Delta_{I_{n}}\}.
Example 4.4.

1) If m=n=2m=n=2 and α=(13,13,13)\alpha=(\frac{1}{3},\frac{1}{3},\frac{1}{3}), then I(2,2)I(2,2) consists of six staircases which, once labelled in the increasing order, are I1=({0},{0},{0,1,2})I^{1}=(\{0\},\{0\},\{0,1,2\}), I2=({0},{0,1},{1,2})I^{2}=(\{0\},\{0,1\},\{1,2\}), I3=({0},{0,1,2},{2})I^{3}=(\{0\},\{0,1,2\},\{2\}), I4=({0,1},{1},{1,2})I^{4}=(\{0,1\},\{1\},\{1,2\}), I5=({0,1},{1,2},{2})I^{5}=(\{0,1\},\{1,2\},\{2\}) and I6=({0,1,2},{2},{2})I^{6}=(\{0,1,2\},\{2\},\{2\}). The six cells (ΔIN,α)N{1,,6}\big{(}\Delta_{I^{N},\alpha}\big{)}_{N\in\{1,\dots,6\}} provide a mixed decomposition of the triangle Δ[2]\Delta_{[2]} and the family (TIN,α)N{1,,6}\big{(}T_{I^{N},\alpha}\big{)}_{N\in\{1,\dots,6\}} provides a partition of the latter, depicted in Figure 6.

Refer to caption
Figure 6: A mixed decomposition of the two-simplex.

2) If n=1n=1, m=3m=3 and α=(12,12)\alpha=(\frac{1}{2},\frac{1}{2}), then I(1,3)I(1,3) consists of four staircases which, once labelled in the increasing order, are I1=({0},{0,1,2,3})I^{1}=(\{0\},\{0,1,2,3\}), I2=({0,1},{1,2,3})I^{2}=(\{0,1\},\{1,2,3\}), I3=({0,1,2},{2,3})I^{3}=(\{0,1,2\},\{2,3\}) and I4=({0,1,2,3},{3})I^{4}=(\{0,1,2,3\},\{3\}). The corresponding cells (ΔIN,α)N{1,,4}\big{(}\Delta_{I^{N},\alpha}\big{)}_{N\in\{1,\dots,4\}} are depicted in Figure 7 and provide a mixed decomposition of the simplex Δ[3]\Delta_{[3]}, while the family (TIN,α)N{1,,4}\big{(}T_{I^{N},\alpha}\big{)}_{N\in\{1,\dots,4\}} provides a partition of the latter, depicted in Figure 8.

Refer to caption
Figure 7: The four cells when n=1n=1 and m=3m=3.
Refer to caption
Figure 8: A mixed decomposition of the three-simplex.

3) If n=2n=2, m=3m=3 and α=(13,13,13)\alpha=(\frac{1}{3},\frac{1}{3},\frac{1}{3}), then I(2,3)I(2,3) consists of ten staircases labelled in the increasing order by (IN)N{1,,10}(I^{N})_{N\in\{1,\dots,10\}}. The ten cells (ΔIN,α)N{1,,10}\big{(}\Delta_{I^{N},\alpha}\big{)}_{N\in\{1,\dots,10\}} provide a mixed decomposition of the simplex Δ[3]\Delta_{[3]} and the family (TIN,α)N{1,,10}\big{(}T_{I^{N},\alpha}\big{)}_{N\in\{1,\dots,10\}} provides a partition of the latter, depicted in Figure 9.

Refer to caption
Figure 9: Another mixed decomposition of the three-simplex.

The phenomenon observed in Example 4.4 is general, compare [20].

Theorem 4.5.

Let m,nm,n be two non-negative integers and let α=(α0,,αn)(+)n+1\alpha=(\alpha_{0},\dots,\alpha_{n})\in(\mathbb{R}_{+}^{*})^{n+1} be such that α0++αn=1\alpha_{0}+\dots+\alpha_{n}=1. Then, the Minkowski cells (ΔI,α)II(n,m)\big{(}\Delta_{I,\alpha}\big{)}_{I\in I(n,m)} provide a mixed decomposition of the simplex Δ[m]\Delta_{[m]} and the family (TI,α)II(n,m)\big{(}T_{I,\alpha}\big{)}_{I\in I(n,m)} provides a partition of the latter. Moreover, if we label the staircases of I(n,m)I(n,m) in increasing order by (IN)N{1,,N(n,m)}(I^{N})_{N\in\{1,\dots,N(n,m)\}}, then for every N{1,,N(n,m)}N\in\{1,\dots,N(n,m)\}, the unions k=1NΔIk,α\cup_{k=1}^{N}\Delta_{I^{k},\alpha} and k=1NTIk,α\cup_{k=1}^{N}T_{I^{k},\alpha} coincide and filtrate Δ[m]\Delta_{[m]}. Finally, the intersection of two cells ΔI,α\Delta_{I,\alpha} and ΔI,α\Delta_{I^{\prime},\alpha}, I,II(n,m)I,I^{\prime}\in I(n,m), is the face j=0nαjΔIjIj\sum_{j=0}^{n}\alpha_{j}\Delta_{I_{j}\cap I^{\prime}_{j}}.

Proof.

Let us identify each point of the simplex Δ[m]\Delta_{[m]} with its barycentric coordinates in the basis given by its vertices, so that Δ[m]={(λ0,,λm)+m+1|λ0++λm=1}\Delta_{[m]}=\{(\lambda_{0},\dots,\lambda_{m})\in\mathbb{R}_{+}^{m+1}\,|\,\lambda_{0}+\dots+\lambda_{m}=1\}. Then, for every I=(Ij)j{0,,n}I(n,m)I=(I_{j})_{j\in\{0,\dots,n\}}\in I(n,m),

ΔI,α={λ=(λi)Δ[m]|j{0,,n},i=0eI(j)1λil=0jαli=0eI(j)λi}\displaystyle\Delta_{I,\alpha}=\left\{\lambda=(\lambda_{i})\in\Delta_{[m]}\,\left|\,\forall j\in\{0,\dots,n\},\sum_{i=0}^{e_{I}(j)-1}\lambda_{i}\leq\sum_{l=0}^{j}\alpha_{l}\leq\sum_{i=0}^{e_{I}(j)}\lambda_{i}\right.\right\} (1)

Indeed, let xΔI,αx\in\Delta_{I,\alpha}, so that x=α0x0++αnxnx=\alpha_{0}x_{0}+\dots+\alpha_{n}x_{n} with xjΔIjx_{j}\in\Delta_{I_{j}}, j{0,,n}j\in\{0,\dots,n\}, and let us denote the barycentric coordinates of xjx_{j} by (λij)iIj(\lambda_{i}^{j})_{i\in I_{j}}. The barycentric coordinates (λi)i{0,,m}(\lambda_{i})_{i\in\{0,\dots,m\}} of xx then satisfy, for every i{0,,m}i\in\{0,\dots,m\}, λi=l=0nαlλil\lambda_{i}=\sum_{l=0}^{n}\alpha_{l}\lambda_{i}^{l}. Let j{0,,n}j\in\{0,\dots,n\}, we deduce that

i=0eI(j)λi\displaystyle\sum_{i=0}^{e_{I}(j)}\lambda_{i} =\displaystyle= i=0eI(j)l=0nαlλill=0jαli=0eI(j)λil=l=0jαl,\displaystyle\sum_{i=0}^{e_{I}(j)}\sum_{l=0}^{n}\alpha_{l}\lambda_{i}^{l}\geq\sum_{l=0}^{j}\alpha_{l}\sum_{i=0}^{e_{I}(j)}\lambda_{i}^{l}=\sum_{l=0}^{j}\alpha_{l},

since {0,,eI(j)}\{0,\dots,e_{I}(j)\} contains IlI_{l} if ljl\leq j. Likewise,

l=0jαl=l=0jαl(i=0eI(j)λil)=i=0eI(j)(l=0jαlλil)i=0eI(j)1(l=0jαlλil)=i=0eI(j)1λi,\sum_{l=0}^{j}\alpha_{l}=\sum_{l=0}^{j}\alpha_{l}\big{(}\sum_{i=0}^{e_{I}(j)}\lambda_{i}^{l}\big{)}=\sum_{i=0}^{e_{I}(j)}\big{(}\sum_{l=0}^{j}\alpha_{l}\lambda_{i}^{l}\big{)}\geq\sum_{i=0}^{e_{I}(j)-1}\big{(}\sum_{l=0}^{j}\alpha_{l}\lambda_{i}^{l}\big{)}=\sum_{i=0}^{e_{I}(j)-1}\lambda_{i},

since Il{0,,eI(j)1}=I_{l}\cap\{0,\dots,e_{I}(j)-1\}=\emptyset if l>jl>j.

Conversely, if λ=(λi)i{0,,m}\lambda=(\lambda_{i})_{i\in\{0,\dots,m\}} satisfies, for every j{0,,n}j\in\{0,\dots,n\}, i=0eI(j)1λil=0jαli=0eI(j)λi\sum_{i=0}^{e_{I}(j)-1}\lambda_{i}\leq\sum_{l=0}^{j}\alpha_{l}\leq\sum_{i=0}^{e_{I}(j)}\lambda_{i}, we set λi0=λiα0\lambda_{i}^{0}=\frac{\lambda_{i}}{\alpha_{0}} if i<eI(0)i<e_{I}(0), λi0=0\lambda_{i}^{0}=0 if i>eI(0)i>e_{I}(0) and λeI(0)0=1i=0eI(0)1λi0\lambda_{e_{I}(0)}^{0}=1-\sum_{i=0}^{e_{I}(0)-1}\lambda^{0}_{i}. Then, by induction on jj, we set

λij=λil=0j1αlλilαj if i<eI(j),λij=0 if i>eI(j) and  λeI(j)j=1i=0eI(j)1λij.\displaystyle\lambda_{i}^{j}=\frac{\lambda_{i}-\sum_{l=0}^{j-1}\alpha_{l}\lambda_{i}^{l}}{\alpha_{j}}\text{ if }i<e_{I}(j),\,\lambda_{i}^{j}=0\text{ if }i>e_{I}(j)\text{ and } \lambda_{e_{I}(j)}^{j}=1-\sum_{i=0}^{e_{I}(j)-1}\lambda^{j}_{i}. (2)

These coefficients are all non-negative and if we denote by xjx_{j} the point with barycentric coordinates (λij)iIj(\lambda_{i}^{j})_{i\in I_{j}}, we get x=α0x0++αnxnx=\alpha_{0}x_{0}+\dots+\alpha_{n}x_{n} by construction. The equality (1) is proved.

Now, for every N{1,,N(n,m)}N\in\{1,\dots,N(n,m)\}, let us denote by LNL_{N} the union k=1NΔIk,α\cup_{k=1}^{N}\Delta_{I^{k},\alpha}. Let λ=(λi)i{0,,m}Δ[m]\lambda=(\lambda_{i})_{i\in\{0,\dots,m\}}\in\Delta_{[m]} and for every j{0,,n1}j\in\{0,\dots,n-1\}, let f(j)f(j) be the least integer ee such that i=0eλil=0jαl\sum_{i=0}^{e}\lambda_{i}\geq\sum_{l=0}^{j}\alpha_{l}, so that l=0jαl>i=0f(j)1λi\sum_{l=0}^{j}\alpha_{l}>\sum_{i=0}^{f(j)-1}\lambda_{i}. We set f(n)=mf(n)=m. Then, λΔIf,α\lambda\in\Delta_{I^{f},\alpha} by (1) while if I<IfI<I^{f}, λΔI,α\lambda\notin\Delta_{I,\alpha} by definition of the lexicographic order. In particular, the cells (ΔI,α)II(n,m)(\Delta_{I,\alpha})_{I\in I(n,m)} cover Δ[m]\Delta_{[m]}. Let N{1,,N(n,m)}N\in\{1,\dots,N(n,m)\} be such that If=INI^{f}=I^{N}, we deduce that ΔIN,αLN1={λ=(λi)i{0,,m}ΔIN,α|j{0,,n1},l=0jαl=i=0eIN(j)1λi}\Delta_{I^{N},\alpha}\cap L_{N-1}=\{\lambda=(\lambda_{i})_{i\in\{0,\dots,m\}}\in\Delta_{I^{N},\alpha}\,|\,\exists j\in\{0,\dots,n-1\},\,\sum_{l=0}^{j}\alpha_{l}=\sum_{i=0}^{e_{I^{N}}(j)-1}\lambda_{i}\}. Let then λ=(λi)i{0,,m}ΔIN,αLN1\lambda=(\lambda_{i})_{i\in\{0,\dots,m\}}\in\Delta_{I^{N},\alpha}\cap L_{N-1} and j{0,,n1}j\in\{0,\dots,n-1\} be such that l=0jαl=i=0eIN(j)1λi\sum_{l=0}^{j}\alpha_{l}=\sum_{i=0}^{e_{I^{N}}(j)-1}\lambda_{i}. Since αj0\alpha_{j}\neq 0, this forces eIN(j1)<eIN(j)e_{I^{N}}(j-1)<e_{I^{N}}(j). Then, denoting by xx the point of Δ[m]\Delta_{[m]} with barycentric coordinates λ\lambda and writing it as before x=α0x0++αnxnx=\alpha_{0}x_{0}+\dots+\alpha_{n}x_{n}, where xlx_{l} has barycentric coordinates (λil)iIlN(\lambda^{l}_{i})_{i\in I^{N}_{l}}, l{0,,n}l\in\{0,\dots,n\}, this forces λeIN(j)j=0\lambda^{j}_{e_{I^{N}}(j)}=0. Indeed, by (2), i=0eIN(j)1λij=1αj(i=0eIN(j)1λil=0j1αl(i=0eIN(j)1λil))=1αj(i=0eIN(j)1λil=0j1αl)=αjαj=1\sum_{i=0}^{e_{I^{N}}(j)-1}\lambda_{i}^{j}=\frac{1}{\alpha_{j}}\big{(}\sum_{i=0}^{e_{I^{N}}(j)-1}\lambda_{i}-\sum_{l=0}^{j-1}\alpha_{l}(\sum_{i=0}^{e_{I^{N}}(j)-1}\lambda^{l}_{i})\big{)}=\frac{1}{\alpha_{j}}\big{(}\sum_{i=0}^{e_{I^{N}}(j)-1}\lambda_{i}-\sum_{l=0}^{j-1}\alpha_{l}\big{)}=\frac{\alpha_{j}}{\alpha_{j}}=1. We conclude that xjx_{j} belongs to the facet ΔIjN{eIN(j)}\Delta_{I^{N}_{j}\setminus\{e_{I^{N}}(j)\}} of ΔIjN\Delta_{I^{N}_{j}}. Conversely, if xjx_{j} belongs to the facet ΔIjN{eIN(j)}\Delta_{I^{N}_{j}\setminus\{e_{I^{N}}(j)\}} of ΔIjN\Delta_{I^{N}_{j}}, then eIN(j1)<eIN(j)e_{I^{N}}(j-1)<e_{I^{N}}(j) and the preceding computation shows that i=0eIN(j)1λi=l=0jαl\sum_{i=0}^{e_{I^{N}}(j)-1}\lambda_{i}=\sum_{l=0}^{j}\alpha_{l} since i=0eIN(j)1λij=1\sum_{i=0}^{e_{I^{N}}(j)-1}\lambda_{i}^{j}=1. We deduce that for every N{2,,N(n,m)}N\in\{2,\dots,N(n,m)\}, LNLN1=TIN,αL_{N}\setminus L_{N-1}=T_{I^{N},\alpha}.

Let us finally prove that the intersection of two cells is a common face of them. Let then I=(Ij)j{0,,n}I=(I_{j})_{j\in\{0,\dots,n\}} and I=(Ij)j{0,,n}I^{\prime}=(I^{\prime}_{j})_{j\in\{0,\dots,n\}} be two staircases of I(n,m)I(n,m). For every j{0,,n}j\in\{0,\dots,n\}, let xjΔIjΔIjx_{j}\in\Delta_{I_{j}}\cap\Delta_{I^{\prime}_{j}}. Then, j=0nαjxjΔI,αΔI,α\sum_{j=0}^{n}\alpha_{j}x_{j}\in\Delta_{I,\alpha}\cap\Delta_{I^{\prime},\alpha} and we have to prove that the intersection is reduced to this face. Let xx be a point in this intersection and let λ=(λi)i{0,,m}\lambda=(\lambda_{i})_{i\in\{0,\dots,m\}} be its barycentric coordinates. For every j{0,,n}j\in\{0,\dots,n\}, i=0eI(j)1λil=0jαli=0eI(j)λi\sum_{i=0}^{e_{I}(j)-1}\lambda_{i}\leq\sum_{l=0}^{j}\alpha_{l}\leq\sum_{i=0}^{e_{I}(j)}\lambda_{i} and i=0eI(j)1λil=0jαli=0eI(j)λi\sum_{i=0}^{e_{I^{\prime}}(j)-1}\lambda_{i}\leq\sum_{l=0}^{j}\alpha_{l}\leq\sum_{i=0}^{e_{I^{\prime}}(j)}\lambda_{i} by (1). We may write x=α0x0++αnxnx=\alpha_{0}x_{0}+\dots+\alpha_{n}x_{n} and x=α0x0++αnxnx=\alpha_{0}x^{\prime}_{0}+\dots+\alpha_{n}x^{\prime}_{n}, where for every j{0,,n}j\in\{0,\dots,n\}, xjΔIjx_{j}\in\Delta_{I_{j}} and xjΔIjx^{\prime}_{j}\in\Delta_{I^{\prime}_{j}} have barycentric coordinates given by (2). Let j{0,,n}j\in\{0,\dots,n\}. If eI(j)<eI(j)e_{I}(j)<e_{I^{\prime}}(j) and vice versa if eI(j)<eI(j)e_{I^{\prime}}(j)<e_{I}(j), we deduce that l=0jαl=i=0eI(j)λi=i=0eI(j)1λi\sum_{l=0}^{j}\alpha_{l}=\sum_{i=0}^{e_{I}(j)}\lambda_{i}=\sum_{i=0}^{e_{I^{\prime}}(j)-1}\lambda_{i}, so that λi=0\lambda_{i}=0 if eI(j)<i<eI(j)e_{I}(j)<i<e_{I^{\prime}}(j) and xjx_{j}^{\prime} belongs to the facet ΔIj{eI(j)}\Delta_{I^{\prime}_{j}\setminus\{e_{I^{\prime}}(j)\}} of ΔIj\Delta_{I^{\prime}_{j}} by what we proved above. By induction on jj, the formula (2) defining the barycentric coordinates of xjx_{j} and xjx^{\prime}_{j} give then the same numbers, so that xj=xjΔIjΔIjx_{j}=x^{\prime}_{j}\in\Delta_{I_{j}}\cap\Delta_{I^{\prime}_{j}} for every j{0,,n}j\in\{0,\dots,n\}. Hence the result. ∎

4.3 Staircases triangulations

The mixed decompositions given by Theorem 4.5 provide in fact primitive triangulations of the product of two simplices via the Cayley trick, see [9, 10, 20], inducing the cartesian product structure of [5]. Indeed, the Cayley trick makes it possible to switch from triangulations to mixed decompositions by associating to every maximal simplex of a primitively triangulated product of simplices Δ×Δ\Delta\times\Delta^{\prime} its intersection with the fiber {b}×Δ\{b\}\times\Delta^{\prime}, where bb denotes the barycenter of Δ\Delta. We are going to use this correspondence.

For every staircase II(n,m)I\in I(n,m), let ΔI\Delta_{I} be the convex hull in Δ[n]×Δ[m]\Delta_{[n]}\times\Delta_{[m]} of the faces ({j}×ΔIj)j{0,,n}(\{j\}\times\Delta_{I_{j}})_{j\in\{0,\dots,n\}} and let TIT_{I} be the convex hull of the tiles ({j}×TIj)j{0,,n1}(\{j\}\times T_{I_{j}})_{j\in\{0,\dots,n-1\}} and {n}×ΔIn\{n\}\times\Delta_{I_{n}}.

Corollary 4.6.

For every non-negative integers m,nm,n, the ordered collection of (m+n)(m+n)-simplices (ΔI)II(n,m)(\Delta_{I})_{I\in I(n,m)} provides a shelled primitive triangulation of the product Δ[n]×Δ[m]\Delta_{[n]}\times\Delta_{[m]}. The ordered collection of tiles (TI)II(n,m)(T_{I})_{I\in I(n,m)} provides the associated hh-tiling.

Corollary 4.6 corresponds to the case in Theorem 3.14 where both tiles are critical of vanishing index.

Proof.

We prove the result by induction on nn. If n=0n=0, there is nothing to prove, the set I(0,m)I(0,m) consists of a single staircase I0I_{0} and ΔI0=TI0\Delta_{I_{0}}=T_{I_{0}} coincides with the simplex {0}×Δ[m]\{0\}\times\Delta_{[m]}. Let us assume the result proven up to the rank n1n-1 and let us prove it for nn. For every j{0,,n}j\in\{0,\dots,n\}, the vertices of the facet Δ[n]{j}\Delta_{[n]\setminus\{j\}} inherit a total order and by the induction hypothesis, the product Δ[n]{j}×Δ[m]\Delta_{[n]\setminus\{j\}}\times\Delta_{[m]} inherits a triangulation with maximal simplices (ΔI~)I~I(n1,m)(\Delta_{\tilde{I}})_{\tilde{I}\in I(n-1,m)}. If II(n,m)I\in I(n,m), the intersection of ΔI\Delta_{I} with Δ[n]{j}×Δ[m]\Delta_{[n]\setminus\{j\}}\times\Delta_{[m]} is a face of codimension #Ij1\#I_{j}-1 in this triangulation by definition, since it is included in a simplex ΔI~\Delta_{\tilde{I}} for some I~I(n1,m)\tilde{I}\in I(n-1,m) and even coincides with this simplex if #Ij=1\#I_{j}=1. Likewise, for every jk{0,,n}j\neq k\in\{0,\dots,n\}, the vertices of the face Δ[n]{j,k}\Delta_{[n]\setminus\{j,k\}} inherit a total order and the product Δ[n]{j,k}×Δ[m]\Delta_{[n]\setminus\{j,k\}}\times\Delta_{[m]} inherits by the induction hypothesis a triangulation which coincides from what we just saw with the trace of the triangulations of Δ[n]{j}×Δ[m]\Delta_{[n]\setminus\{j\}}\times\Delta_{[m]} and Δ[n]{k}×Δ[m]\Delta_{[n]\setminus\{k\}}\times\Delta_{[m]}. Hence, we get from the induction hypothesis a triangulation on Δ[n]×Δ[m]\partial\Delta_{[n]}\times\Delta_{[m]}. Now, the interior points to Δ[n]\Delta_{[n]} are determined by their barycentric coordinates α=(α0,,αn)(+)n+1\alpha=(\alpha_{0},\dots,\alpha_{n})\in(\mathbb{R}_{+}^{*})^{n+1}, α0++αn=1\alpha_{0}+\dots+\alpha_{n}=1, in the affine basis given by its vertices. For every staircase II(n,m)I\in I(n,m), the intersection ΔI({α}×Δ[m])\Delta_{I}\cap(\{\alpha\}\times\Delta_{[m]}) coincides with ΔI,α\Delta_{I,\alpha} by definition. From Theorem 3.14 follows thus by induction that the collection of simplices (ΔI)II(n,m)(\Delta_{I})_{I\in I(n,m)} defines a primitive triangulation of the product Δ[n]×Δ[m]\Delta_{[n]}\times\Delta_{[m]} and that the intersection of two such simplices ΔIΔJ\Delta_{I}\cap\Delta_{J} is the convex hull of the vertices of ({j}×ΔIjJj)j{0,,n}(\{j\}\times\Delta_{I_{j}\cap J_{j}})_{j\in\{0,\dots,n\}} they have in common.

Let us now label the straicases in the increasing order by (IN)N{1,N(n,m)}(I^{N})_{N\in\{1\dots,N(n,m)\}} so that the unions k=1NΔIk\cup_{k=1}^{N}\Delta_{I^{k}}, N{1,N(n,m)}N\in\{1\dots,N(n,m)\}, filtrate Δ[n]×Δ[m]\Delta_{[n]}\times\Delta_{[m]}. Let N{1,N(n,m)}N\in\{1\dots,N(n,m)\}, we are going to prove that k=1NΔIkk=1N1ΔIk\cup_{k=1}^{N}\Delta_{I^{k}}\setminus\cup_{k=1}^{N-1}\Delta_{I^{k}} is the tile TINT_{I^{N}}. Let II(n,m)I\in I(n,m) be such that I<INI<I^{N}. By definition of the lexicographic order, there exists j{0,,n1}j\in\{0,\dots,n-1\} such that #IjN>1\#I_{j}^{N}>1 and eI(j)<eIN(j)e_{I}(j)<e_{I^{N}}(j), since eIe_{I} is increasing. If eI(j)=eIN(j)1e_{I}(j)=e_{I^{N}}(j)-1 and eI(l)=eIN(l)e_{I}(l)=e_{I^{N}}(l) for ljl\neq j, then ΔI\Delta_{I} contains all vertices of ΔIN\Delta_{I^{N}} but (j,eIN(j))(j,e_{I^{N}}(j)) so that ΔIΔIN\Delta_{I}\cap\Delta_{I^{N}} is the facet of ΔIN\Delta_{I^{N}} not containing (j,eIN(j))(j,e_{I^{N}}(j)). Otherwise, ΔI\Delta_{I} contains a subset of these vertices of ΔIN\Delta_{I^{N}}. Since TINT_{I^{N}} is the tile ΔIN\Delta_{I^{N}} deprived precisely of all those facets not containing (j,eIN(j))(j,e_{I^{N}}(j)) for all j{0,,n1}j\in\{0,\dots,n-1\} such that #Ij>1\#I_{j}>1, we deduce the result. ∎

Let us finally observe that the lexicographic order on the pairs (j,i){0,,n}×{0,,m}(j,i)\in\{0,\dots,n\}\times\{0,\dots,m\} induces a total order on the vertices of ΔI\Delta_{I} for all II(n,m)I\in I(n,m). If (j1,i1)(j_{1},i_{1}) and (j2,i2)(j_{2},i_{2}) are two vertices of ΔI\Delta_{I}, then, by definition of staircases, (j1,i1)(j2,i2)(j_{1},i_{1})\leq(j_{2},i_{2}) with respect to this order iff j1j2j_{1}\leq j_{2} and i1i2i_{1}\leq i_{2}. The product Δ[n]×Δ[m]\Delta_{[n]}\times\Delta_{[m]} equipped with the triangulation (ΔI)II(n,m)(\Delta_{I})_{I\in I(n,m)} given by Corollary 4.6 is thus the cartesian product of Δ[n]\Delta_{[n]} and Δ[m]\Delta_{[m]} in the sense of Definition II.8.8II.8.8 of [5].

4.4 The palindromic automorphism

Let 𝒫{\cal P} be the automorphism of Δ[n]×Δ[m]\Delta_{[n]}\times\Delta_{[m]} induced by reversing the total orders of the vertices of both simplices, so that for every (j,i){0,,n}×{0,,m}(j,i)\in\{0,\dots,n\}\times\{0,\dots,m\}, 𝒫(j,i)=(nj,mi){\cal P}(j,i)=(n-j,m-i).

Lemma 4.7.

For every non-negative integers m,nm,n, the automorphism 𝒫{\cal P} preserves the triangulation (ΔI)II(n,m)(\Delta_{I})_{I\in I(n,m)} of Δ[n]×Δ[m]\Delta_{[n]}\times\Delta_{[m]} and satisfies, for every II(n,m)I\in I(n,m), 𝒫(ΔI)=ΔIˇ{\cal P}(\Delta_{I})=\Delta_{\check{I}}.

Proof.

Let II(n,m)I\in I(n,m), it is enough to prove that 𝒫{\cal P} maps the vertices of ΔI\Delta_{I} on those of ΔIˇ\Delta_{\check{I}}. Now, if j{0,,n}j\in\{0,\dots,n\} and if iIji\in I_{j}, 𝒫(j,i)=(nj,mi){\cal P}(j,i)=(n-j,m-i) by definition while Iˇnj={meI(j),,mbI(j)}\check{I}_{n-j}=\{m-e_{I}(j),\dots,m-b_{I}(j)\} by definition, so that miIˇnjm-i\in\check{I}_{n-j} if iIji\in I_{j}. ∎

However, the automorphism 𝒫{\cal P} does not preserve the hh-tiling (TI)II(n,m)(T_{I})_{I\in I(n,m)} of Δ[n]×Δ[m]\Delta_{[n]}\times\Delta_{[m]} in general, so that (𝒫(TI))II(n,m)({\cal P}(T_{I}))_{I\in I(n,m)} provides another shelling of the triangulation (ΔI)II(n,m)(\Delta_{I})_{I\in I(n,m)}. We may also exchange the factors Δ[n]\Delta_{[n]} and Δ[m]\Delta_{[m]} to get from Corollary 4.6 a triangulation (ΔJ)JI(m,n)(\Delta_{J})_{J\in I(m,n)} on the product Δ[m]×Δ[n]\Delta_{[m]}\times\Delta_{[n]} together with a pair of shellings (TJ)JI(m,n)(T_{J})_{J\in I(m,n)} and (𝒫(TJ))JI(m,n)({\cal P}(T_{J}))_{J\in I(m,n)}.

Lemma 4.8.

The involution E:(x,y)Δ[n]×Δ[m](y,x)Δ[m]×Δ[n]E:(x,y)\in\Delta_{[n]}\times\Delta_{[m]}\mapsto(y,x)\in\Delta_{[m]}\times\Delta_{[n]} commutes with the action of 𝒫{\cal P} and defines an isomorphism between the simplicial complexes (ΔI)II(n,m)(\Delta_{I})_{I\in I(n,m)} and (ΔJ)JI(m,n)(\Delta_{J})_{J\in I(m,n)} which maps the shelling (TI)II(n,m)(T_{I})_{I\in I(n,m)} onto the shelling (𝒫(TJ))JI(m,n)({\cal P}(T_{J}))_{J\in I(m,n)}. Moreover, it preserves the order on the vertices of all simplices ΔI\Delta_{I}, II(n,m)I\in I(n,m).

Proof.

Let us identify xΔ[n]x\in\Delta_{[n]} (resp. yΔ[m]y\in\Delta_{[m]}) with its barycentric coordinates (αj)j{0,,n}(\alpha_{j})_{j\in\{0,\dots,n\}} (resp. (λi)i{0,,m}(\lambda_{i})_{i\in\{0,\dots,m\}}) in the affine basis given by the vertices of Δ[n]\Delta_{[n]} (resp. Δ[m]\Delta_{[m]}), so that αj0\alpha_{j}\geq 0 (resp. λi0\lambda_{i}\geq 0) and j=0nαj=1\sum_{j=0}^{n}\alpha_{j}=1 (resp. i=0mλi=1\sum_{i=0}^{m}\lambda_{i}=1 ). Then, 𝒫E((αj)j{0,,n},(λi)i{0,,m})=((λmi)i{0,,m},(αnj)j{0,,n})=E𝒫((αj)j{0,,n},(λi)i{0,,m}){\cal P}\circ E\big{(}(\alpha_{j})_{j\in\{0,\dots,n\}},(\lambda_{i})_{i\in\{0,\dots,m\}}\big{)}=\big{(}(\lambda_{m-i})_{i\in\{0,\dots,m\}},(\alpha_{n-j})_{j\in\{0,\dots,n\}}\big{)}=E\circ{\cal P}\big{(}(\alpha_{j})_{j\in\{0,\dots,n\}},(\lambda_{i})_{i\in\{0,\dots,m\}}\big{)}, hence the first part of the lemma.

Let now II(n,m)I\in I(n,m) and ΔI\Delta_{I} be the associated simplex in the triangulated Δ[m]×Δ[n]\Delta_{[m]}\times\Delta_{[n]}. Its vertices are the pairs (j,i){0,,n}×{0,,m}(j,i)\in\{0,\dots,n\}\times\{0,\dots,m\} such that iIji\in I_{j}. The image E(ΔI)E(\Delta_{I}) is thus the simplex of Δ[m]×Δ[n]\Delta_{[m]}\times\Delta_{[n]} with vertices (i,j){0,,m}×{0,,n}(i,j)\in\{0,\dots,m\}\times\{0,\dots,n\} such that iIji\in I_{j}. The conditions iIji\in I_{j} and jJij\in J_{i} are equivalent to each other, where JJ is the image of II under the involution II(n,m)JJ(m,n)I\in I(n,m)\mapsto J\in J(m,n) given by Lemma 4.3. We deduce that E(ΔI)=ΔJE(\Delta_{I})=\Delta_{J}, so that EE defines a /2\mathbb{Z}/2\mathbb{Z}-equivariant simplicial isomorphism between the complexes (ΔI)II(n,m)(\Delta_{I})_{I\in I(n,m)} and (ΔJ)JI(m,n)(\Delta_{J})_{J\in I(m,n)}. Moreover, if (j1,i1)(j2,i2)(j_{1},i_{1})\leq(j_{2},i_{2}) are vertices of ΔI\Delta_{I}, then j1j2j_{1}\leq j_{2} and i1i2i_{1}\leq i_{2} by definition of staircases, so that (i1,j1)(i2,j2)(i_{1},j_{1})\leq(i_{2},j_{2}) and E|ΔI:ΔIΔJE_{|\Delta_{I}}:\Delta_{I}\to\Delta_{J} preserves the order on the vertices.

Let us finally prove that 𝒫E{\cal P}\circ E maps the tiling (TI)II(n,m)(T_{I})_{I\in I(n,m)} onto (TJ)JI(m,n)(T_{J})_{J\in I(m,n)}. Let II(n,m)I\in I(n,m). By definition, TIT_{I} is the simplex ΔI\Delta_{I} with vertices {(j,i){0,,n}×{0,,m}|iIj}\{(j,i)\in\{0,\dots,n\}\times\{0,\dots,m\}\,|\,i\in I_{j}\} deprived, for every j<nj<n such that bI(j)eI(j)b_{I}(j)\neq e_{I}(j), of the facet not containing the vertex (j,eI(j))(j,e_{I}(j)). From the preceding part, the involution EE maps ΔI\Delta_{I} onto the simplex ΔJ\Delta_{J}, JJ(m,n)J\in J(m,n), and the vertices {(j,eI(j)){0,,n1}×{1,,m}|bI(j)<eI(j)}\{(j,e_{I}(j))\in\{0,\dots,n-1\}\times\{1,\dots,m\}\,|\,b_{I}(j)<e_{I}(j)\} onto the vertices {(i,bJ(i)){1,,m}×{0,,n1}|bJ(i)<eJ(i)}\{(i,b_{J}(i))\in\{1,\dots,m\}\times\{0,\dots,n-1\}\,|\,b_{J}(i)<e_{J}(i)\}. The involution 𝒫E{\cal P}\circ E thus maps ΔI\Delta_{I} onto ΔJwidecheck\Delta_{\widecheck{J}} and the vertices {(j,eI(j)){0,,n1}×{1,,m}|bI(j)<eI(j)}\{(j,e_{I}(j))\in\{0,\dots,n-1\}\times\{1,\dots,m\}\,|\,b_{I}(j)<e_{I}(j)\} onto the vertices {(i,eJ(i)){0,,m1}×{1,,n}|bJ(i)<eJ(i)}\{(i,e_{J}(i))\in\{0,\dots,m-1\}\times\{1,\dots,n\}\,|\,b_{J}(i)<e_{J}(i)\}, so that 𝒫E(TI)=TJwidecheck{\cal P}\circ E(T_{I})=T_{\widecheck{J}}. Hence the result. ∎

5 Shellings on products of two tiles

5.1 Preliminaries

The proofs of Theorems 3.14 and 3.16 are based on the following Propositions 5.1 and 5.3.

Proposition 5.1.

Let m,n0m,n\geq 0, II(n,m)I\in I(n,m) and J={b(J),,e(J)}J=\{b(J),\dots,e(J)\} be an interval of {0,,m}\{0,\dots,m\}. Then, the intersection of the tile TIT_{I} with Δ[n]×Δ[m]J\Delta_{[n]}\times\Delta_{[m]\setminus J} is:

  1. 1.

    empty if there exists j{0,,n1}j\in\{0,\dots,n-1\} such that bI(j)eI(j)b_{I}(j)\neq e_{I}(j) and eI(j)Je_{I}(j)\in J.

  2. 2.

    the convex hull of the tiles ({l}×TIl)l{0,,n1}{j}(\{l\}\times T_{I_{l}})_{l\in\{0,\dots,n-1\}\setminus\{j\}}, {n}×ΔIn\{n\}\times\Delta_{I_{n}} and {j}×TIjJ\{j\}\times T_{I_{j}\setminus J} if there exists j{0,,n1}j\in\{0,\dots,n-1\} such that JIj{bI(j),eI(j)}J\subset I_{j}\setminus\{b_{I}(j),e_{I}(j)\} and the convex hull of the tiles ({j}×TIj)j{0,,n1}(\{j\}\times T_{I_{j}})_{j\in\{0,\dots,n-1\}} and {n}×ΔInJ\{n\}\times\Delta_{I_{n}\setminus J} if JIn{bI(n)}J\subset I_{n}\setminus\{b_{I}(n)\}.

  3. 3.

    the convex hull of the tiles ({l}×TIl)l{j+1,,n1}(\{l\}\times T_{I_{l}})_{l\in\{j+1,\dots,n-1\}}, {n}×ΔIn\{n\}\times\Delta_{I_{n}} and {j}×TIjJ\{j\}\times T_{I_{j}\setminus J} if b(J)=0b(J)=0 and there exists j{0,,n1}j\in\{0,\dots,n-1\} such that JIj{eI(j)}J\subset I_{j}\setminus\{e_{I}(j)\} and restricts to {n}×ΔInJ\{n\}\times\Delta_{I_{n}\setminus J} if b(J)=0b(J)=0 and JInJ\subset I_{n}.

Remark 5.2.

In particular, the face TI(Δ[n]×Δ[m]J)T_{I}\cap(\Delta_{[n]}\times\Delta_{[m]\setminus J}) is empty in the case 1. of Proposition 5.1, of codimension #J\#J in the case 2. and of codimension #J+j1\#J+j-1 in the case 3.

Proof.

In the case 1., the intersection of Δ[n]×Δ[m]J\Delta_{[n]}\times\Delta_{[m]\setminus J} with the simplex ΔI\Delta_{I} is included in the facet of ΔI\Delta_{I} which does not contain the vertex (j,eI(j))(j,e_{I}(j)). Since j<nj<n, TIT_{I} is already deprived of this facet by definition and we get 1. In the case 2., Δ[n]×Δ[m]J\Delta_{[n]}\times\Delta_{[m]\setminus J} contains all the vertices of ΔI\Delta_{I} except those with coordinates (j,i)(j,i) with iJi\in J. This product intersects thus ΔI\Delta_{I} along a face of codimension #J\#J, convex hull of {j}×ΔIjJ\{j\}\times\Delta_{I_{j}\setminus J} and the simplices ({l}×ΔIl)lj(\{l\}\times\Delta_{I_{l}})_{l\neq j}. We deduce part 2. after intersecting TIT_{I} with this face. Finally, if JIjJ\subset I_{j} and b(J)=0b(J)=0, then Il={0}I_{l}=\{0\} if l<jl<j by definition. It follows that the intersection of Δ[n]×Δ[m]J\Delta_{[n]}\times\Delta_{[m]\setminus J} with the simplex ΔI\Delta_{I} is the convex hull of {j}×ΔIjJ\{j\}\times\Delta_{I_{j}\setminus J} and the faces ({l}×ΔIl)l>j(\{l\}\times\Delta_{I_{l}})_{l>j}. We deduce 3. after intersecting TIT_{I} with this face. ∎

Proposition 5.3.

Let m,n0m,n\geq 0, II(n,m)I\in I(n,m) and JJ be a subset of {0,,n}\{0,\dots,n\}. Then, the intersection of the tile TIT_{I} with Δ[n]J×Δ[m]\Delta_{[n]\setminus J}\times\Delta_{[m]} is:

  1. 1.

    empty if there exists jJ{n}j\in J\setminus\{n\} such that bI(j)eI(j)b_{I}(j)\neq e_{I}(j).

  2. 2.

    the convex hull of the tiles ({j}×TIj)j{0,,n1}J(\{j\}\times T_{I_{j}})_{j\in\{0,\dots,n-1\}\setminus J} together with {n}×ΔIn\{n\}\times\Delta_{I_{n}} if nJn\notin J otherwise.

Remark 5.4.

In particular, the face TI(Δ[n]J×Δ[m])T_{I}\cap(\Delta_{[n]\setminus J}\times\Delta_{[m]}) is of codimension at least #J\#J when nonempty.

Proof.

In the case 1.1., the intersection of Δ[n]J×Δ[m]\Delta_{[n]\setminus J}\times\Delta_{[m]} with the simplex ΔI\Delta_{I} is included in the facet of ΔI\Delta_{I} which does not contain the vertex (j,eI(j))(j,e_{I}(j)). Since j<nj<n, TIT_{I} has been deprived of this facet by definition and we get 1. The case 2. follows from the definition of TIT_{I}. ∎

5.2 Proof of Theorem 3.14

Let us denote by σ1,σ2\sigma_{1},\sigma_{2} the underlying simplices of T1T_{1} and T2T_{2} and by nn, mm their respective dimensions. Let us also choose total orders on the vertices of σ1\sigma_{1} and σ2\sigma_{2} in such a way that if T1T_{1} (resp. T2T_{2}) has been deprived of a Morse face, then the vertices of this face are the greatest of σ1\sigma_{1} (resp. σ2\sigma_{2}). These orders induce isomorphisms between σ1\sigma_{1} (resp. σ2\sigma_{2}) and Δ[n]\Delta_{[n]} (resp. Δ[m]\Delta_{[m]}), so that σ1×σ2\sigma_{1}\times\sigma_{2} inherits the triangulation (ΔI)II(n,m)(\Delta_{I})_{I\in I(n,m)} given by Corollary 4.6 together with the shelling (TI)II(n,m)(T_{I})_{I\in I(n,m)}. The palindromic automorphism 𝒫{\cal P} given by Lemma 4.7 induces then an isomorphism of the simplicial complex σ1×σ2\sigma_{1}\times\sigma_{2}. We are going to prove the following alternative. Either, for every II(n,m)I\in I(n,m), TI(T1×T2)T_{I}\cap(T_{1}\times T_{2}) is a Morse tile and the collection (TI(T1×T2))II(n,m)\big{(}T_{I}\cap(T_{1}\times T_{2})\big{)}_{I\in I(n,m)} provides a Morse shelling of T1×T2T_{1}\times T_{2} as claimed by Theorem 3.14 ; or, for every II(n,m)I\in I(n,m), 𝒫(TI)(T1×T2){\cal P}(T_{I})\cap(T_{1}\times T_{2}) is a Morse tile and the collection (𝒫(TI)(T1×T2))II(n,m)\big{(}{\cal P}(T_{I})\cap(T_{1}\times T_{2})\big{)}_{I\in I(n,m)} provides the Morse shelling claimed by Theorem 3.14. Hence, the trace with T1×T2T_{1}\times T_{2} of one of the two hh-tilings (TI)II(n,m)(T_{I})_{I\in I(n,m)} and (𝒫(TI))II(n,m)({\cal P}(T_{I}))_{I\in I(n,m)} of σ1×σ2\sigma_{1}\times\sigma_{2} provides the Morse shelling we are looking for. However, the total orders chosen on the vertices of σ1\sigma_{1} and σ2\sigma_{2} have to satisfy the previous condition for this result to hold true. To get the first part of Theorem 3.14, where T1T_{1} and T2T_{2} are basic, one of them being regular, and where we want a true shelling of T1×T2T_{1}\times T_{2}, that is using only basic tiles, we need an additional condition on these orders to be satisfied, namely Condition hh of §5.2.4. In what follows, we are going to consider separately the case where T1T_{1} and T2T_{2} are both basic, the case where they are both Morse non-basic and the mixed case, one being basic, the other one being Morse non-basic. Let us recall that the lexicographic order on the pairs (j,i){0,,n}×{0,,m}(j,i)\in\{0,\dots,n\}\times\{0,\dots,m\} induces a total order on the vertices of ΔI\Delta_{I} for every II(n,m)I\in I(n,m), so that the total orders on the vertices of σ1\sigma_{1} and σ2\sigma_{2} induce total orders on the vertices of all simplices of the complex σ1×σ2\sigma_{1}\times\sigma_{2}.

5.2.1 The case of two basic tiles

If both tiles are basic, the chosen order on the vertices of σ1\sigma_{1} (resp. σ2\sigma_{2}) induces an isomorphism between T1T_{1} (resp. T2T_{2}) and Δ[n]jJ1Δ[n]{j}\Delta_{[n]}\setminus\cup_{j\in J_{1}}\Delta_{[n]\setminus\{j\}} (resp. Δ[m]iJ2Δ[m]{i}\Delta_{[m]}\setminus\cup_{i\in J_{2}}\Delta_{[m]\setminus\{i\}}), where #J1\#J_{1} (resp. #J2\#J_{2}) is the order of T1T_{1} (resp. T2T_{2}). If one of these tiles is in addition regular, in order to get a shelling of T1×T2T_{1}\times T_{2} using only regular basic tiles, we need to assume that the total orders have been chosen in such a way that if {0,n}J1\{0,n\}\subset J_{1}, then {0,m}J2\{0,m\}\cap J_{2}\neq\emptyset and vice versa, that if {0,m}J2\{0,m\}\subset J_{2}, then {0,n}J1\{0,n\}\cap J_{1}\neq\emptyset, see Condition hh of §5.2.4. In this case, applying the involution T1×T2T2×T1T_{1}\times T_{2}\to T_{2}\times T_{1} which exchanges the roles of T1T_{1} and T2T_{2}, we may assume that if nJ1n\in J_{1}, then mJ2m\in J_{2} and that if 0J20\in J_{2}, then 0J10\in J_{1}. If we do not assume this additional Condition hh on the total orders, then, even with this possibility to apply the involution T1×T2T2×T1T_{1}\times T_{2}\to T_{2}\times T_{1}, we can only assume that one of these two properties holds true, either that if nJ1n\in J_{1}, then mJ2m\in J_{2}, or that if 0J20\in J_{2}, then 0J10\in J_{1}, but not both. The following proof then provides a Morse shelling on T1×T2T_{1}\times T_{2}, but not a true shelling in general, so that we need the additional Condition hh to get the first part of Theorem 3.14.

From Proposition 5.1 we know that for every II(n,m)I\in I(n,m) and every iJ2{0}i\in J_{2}\setminus\{0\}, TI(Δ[n]×Δ[m]{i})T_{I}\cap\big{(}\Delta_{[n]}\times\Delta_{[m]\setminus\{i\}}\big{)} is either empty, or of codimension one in TIT_{I}, see Remark 5.2. If 0J20\in J_{2}, then either #I0>1\#I_{0}>1 and TI(Δ[n]×Δ{1,,m})T_{I}\cap\big{(}\Delta_{[n]}\times\Delta_{\{1,\dots,m\}}\big{)} is of codimension one in TIT_{I}, or #I0=1\#I_{0}=1 and this intersection is included in the facet TI(Δ{1,,n}×Δ[m])T_{I}\cap\big{(}\Delta_{\{1,\dots,n\}}\times\Delta_{[m]}\big{)}, so that TI(Δ[n]×T2)T_{I}\cap\big{(}\Delta_{[n]}\times T_{2}\big{)} is a basic tile since we assumed that 0J20\in J_{2} implies 0J10\in J_{1}. Likewise, from Proposition 5.3 we know that for every II(n,m)I\in I(n,m) and every jJ1j\in J_{1}, TI(Δ[n]{j}×Δ[m])T_{I}\cap\big{(}\Delta_{[n]\setminus\{j\}}\times\Delta_{[m]}\big{)} is either empty, or of codimension one in TIT_{I}, with the exception of j=nj=n if #In>1\#I_{n}>1, but then the intersection is included in the facet TI(Δ[n]×Δ{0,,m1})T_{I}\cap\big{(}\Delta_{[n]}\times\Delta_{\{0,\dots,m-1\}}\big{)} since we also assumed that nJ1n\in J_{1} implies mJ2m\in J_{2}. We then deduce that for every II(n,m)I\in I(n,m), TI(T1×T2)T_{I}\cap(T_{1}\times T_{2}) is a basic tile. Moreover, by definition, the order of TIT_{I} equals n#{j{0,,n1}|#Ij=1}n-\#\{j\in\{0,\dots,n-1\}\,|\,\#I_{j}=1\}, so that the order of TI(T1×T2)T_{I}\cap(T_{1}\times T_{2}) equals

n#{j{0,,n1}J1|#Ij=1}+#{iJ2|ibI({1,,n})}\displaystyle n-\#\{j\in\{0,\dots,n-1\}\setminus J_{1}\,|\,\#I_{j}=1\}+\#\{i\in J_{2}\,|\,i\notin b_{I}(\{1,\dots,n\})\} (3)

plus one in case nJ1n\in J_{1} and #In=1\#I_{n}=1. Since one of the tiles T1,T2T_{1},T_{2} is regular, we deduce from Propositions 5.1 and 5.3 the upper and lower bounds 0<Ord(TI(T1×T2))m+n0<\textup{Ord}(T_{I}\cap(T_{1}\times T_{2}))\leq m+n, for the term (n#{j{0,,n1}J1|#Ij=1})(n-\#\{j\in\{0,\dots,n-1\}\setminus J_{1}\,|\,\#I_{j}=1\}) vanishes only if In={0,,m}I_{n}=\{0,\dots,m\} and in this case the second term in (3) does not vanish by hypothesis. Moreover, the last two terms equal m+1m+1 only if I0={0,,m}I_{0}=\{0,\dots,m\} and J2={0,,m}J_{2}=\{0,\dots,m\} but in this case the first term in (3) is less than nn by hypothesis, for J1J_{1} cannot be {0,,n}\{0,\dots,n\}. Hence, the shelled triangulation of Δ[n]×Δ[m]\Delta_{[n]}\times\Delta_{[m]} given by Corollary 4.6 induces a shelled primitive triangulation of T1×T2T_{1}\times T_{2} using only regular basic tiles. The first part of Theorem 3.14 is proven. When Condition hh of §5.2.4 is not satisfied, we may still assume, applying the involution E:T1×T2T2×T1E:T_{1}\times T_{2}\to T_{2}\times T_{1} if necessary, that if nJ1n\in J_{1}, then mJ2m\in J_{2}. If 0J20\in J_{2} but 0J10\notin J_{1}, then for every II(n,m)I\in I(n,m), ΔI(Δ[n]×Δ{1,,m})\Delta_{I}\cap\big{(}\Delta_{[n]}\times\Delta_{\{1,\dots,m\}}\big{)} is a face of codimension greater than one as soon as #I0=1\#I_{0}=1 and its vertices are the greatest of ΔI\Delta_{I}. The tile TI(T1×T2)T_{I}\cap(T_{1}\times T_{2}) is then Morse and we get a Morse shelling (TI(T1×T2))II(n,m)\big{(}T_{I}\cap(T_{1}\times T_{2})\big{)}_{I\in I(n,m)} which is tame. If nJ1n\in J_{1} and mJ2m\notin J_{2}, we apply the involution EE to the tame Morse shelling (TJ(T2×T1))JI(m,n)\big{(}T_{J}\cap(T_{2}\times T_{1})\big{)}_{J\in I(m,n)} we just obtained to deduce from Lemma 4.8 that the collection (𝒫(TI)(T1×T2))II(n,m)\big{(}{\cal P}(T_{I})\cap(T_{1}\times T_{2})\big{)}_{I\in I(n,m)} defines a tame Morse shelling on T1×T2T_{1}\times T_{2}.

If T1T_{1} and T2T_{2} are open simplices, so that J1={0,,n}J_{1}=\{0,\dots,n\} and J2={0,,m}J_{2}=\{0,\dots,m\}, then again, by Propositions 5.1 and 5.3, for every II(n,m)I\in I(n,m), every iJ2{0}i\in J_{2}\setminus\{0\} and every jJ1{n}j\in J_{1}\setminus\{n\}, TI(Δ[n]×Δ[m]{i})T_{I}\cap\big{(}\Delta_{[n]}\times\Delta_{[m]\setminus\{i\}}\big{)} and TI(Δ[n]{j}×Δ[m])T_{I}\cap\big{(}\Delta_{[n]\setminus\{j\}}\times\Delta_{[m]}\big{)} are of codimension one in TIT_{I} when non-empty. Moreover, TI(Δ[n]×Δ{1,,m})T_{I}\cap\big{(}\Delta_{[n]}\times\Delta_{\{1,\dots,m\}}\big{)} (resp. TI(Δ{0,,n1}×Δ[m])T_{I}\cap\big{(}\Delta_{\{0,\dots,n-1\}}\times\Delta_{[m]}\big{)}) is either of codimension one in TIT_{I} if #I0>1\#I_{0}>1 (resp. #In=1\#I_{n}=1), or included in the facet TI(Δ{1,,n}×Δ[m])T_{I}\cap\big{(}\Delta_{\{1,\dots,n\}}\times\Delta_{[m]}\big{)} (resp. TI(Δ[n]×Δ{0,,m1})T_{I}\cap\big{(}\Delta_{[n]}\times\Delta_{\{0,\dots,m-1\}}\big{)}) which is removed from TIT_{I}. Thus, the shelled triangulation of Δ[n]×Δ[m]\Delta_{[n]}\times\Delta_{[m]} given by Corollary 4.6 induces a shelled primitive triangulation of T1×T2T_{1}\times T_{2} also in this case and (3) remains valid. It remains to check that this hh-tiling uses a unique critical tile, of index m+nm+n. But as before, Ord(TI(T1×T2))>0\textup{Ord}(T_{I}\cap(T_{1}\times T_{2}))>0 for every II(n,m)I\in I(n,m) while Ord(TI(T1×T2))=n+m+1\textup{Ord}(T_{I}\cap(T_{1}\times T_{2}))=n+m+1 forces I0={0,,m}I_{0}=\{0,\dots,m\}. However now, J1={0,,n}J_{1}=\{0,\dots,n\} and J2={0,,m}J_{2}=\{0,\dots,m\}, so that this staircase provides a critical tile of index m+nm+n, namely an open simplex.

If on the contrary T1T_{1} is a closed simplex and T2T_{2} an open one, then for every II(n,m)I\in I(n,m), TIi=1m(Δ[n]×Δ[m]{i})T_{I}\setminus\cup_{i=1}^{m}\big{(}\Delta_{[n]}\times\Delta_{[m]\setminus\{i\}}\big{)} is a basic tile of order mm by Proposition 5.1 and TI(T1×T2)T_{I}\cap(T_{1}\times T_{2}) is a basic tile of order m+1m+1 if #I01\#I_{0}\neq 1 and a Morse tile of order mm otherwise, the Morse face being of codimension #bI1(0)\#b_{I}^{-1}(0). The only critical tile of the Morse tiling of T1×T2T_{1}\times T_{2} is thus the tile TI(T1×T2)T_{I}\cap(T_{1}\times T_{2}) with In={0,,m}I_{n}=\{0,\dots,m\} and Ij={0}I_{j}=\{0\} if j<nj<n. Its index equals mm. Moreover, for all non-basic tile TI(T1×T2)T_{I}\cap(T_{1}\times T_{2}) of the tiling, II(n,m)I\in I(n,m), the vertices of the Morse face TI(Δ[n]×Δ{1,,m})T_{I}\cap\big{(}\Delta_{[n]}\times\Delta_{\{1,\dots,m\}}\big{)} are the greatest of ΔI\Delta_{I} with respect to the lexicographic order on {0,,n}×{0,,m}\{0,\dots,n\}\times\{0,\dots,m\}, so that the Morse shelling is tame. If T2T_{2} is a closed simplex and T1T_{1} an open one, we apply the exchange involution E:T2×T1T1×T2E:T_{2}\times T_{1}\to T_{1}\times T_{2} to the Morse shelling (TJ(T2×T1))JI(m,n)\big{(}T_{J}\cap(T_{2}\times T_{1})\big{)}_{J\in I(m,n)} we just obtained, to deduce from Lemma 4.8 that the collection (𝒫(TI)(T1×T2))II(n,m)\big{(}{\cal P}(T_{I})\cap(T_{1}\times T_{2})\big{)}_{I\in I(n,m)} defines a tame Morse shelling on T1×T2T_{1}\times T_{2}.

5.2.2 The case of two non-basic tiles

If both T1T_{1} and T2T_{2} are Morse and not basic, then the chosen orders on the vertices of σ1\sigma_{1} and σ2\sigma_{2} induce isomorphisms between them and Δ[n](jJ1Δ[n]{j}Δ{k1,,n})\Delta_{[n]}\setminus\big{(}\cup_{j\in J_{1}}\Delta_{[n]\setminus\{j\}}\cup\Delta_{\{k_{1},\dots,n\}}\big{)} and Δ[m](iJ2Δ[m]{i}Δ{k2,,m})\Delta_{[m]}\setminus\big{(}\cup_{i\in J_{2}}\Delta_{[m]\setminus\{i\}}\cup\Delta_{\{k_{2},\dots,m\}}\big{)} respectively, where 1<k1n1<k_{1}\leq n, 1<k2m1<k_{2}\leq m, J1{k1, ,n}J_{1}\subset\{k_{1}, \dots,n\} and J2{k2, ,m}J_{2}\subset\{k_{2}, \dots,m\}. Applying the involution E:T1×T2T2×T1E:T_{1}\times T_{2}\to T_{2}\times T_{1} which exchanges the roles of T1T_{1} and T2T_{2} if necessary, we may assume that if nJ1n\in J_{1}, then mJ2m\in J_{2}, see Lemma 4.8. In this case, we deduce from §5.2.1 that for every II(n,m)I\in I(n,m), TI(jJ1(Δ[n]{j}×Δ[m])iJ2(Δ[n]×Δ[m]{i}))T_{I}\setminus\big{(}\cup_{j\in J_{1}}(\Delta_{[n]\setminus\{j\}}\times\Delta_{[m]})\cup_{i\in J_{2}}(\Delta_{[n]}\times\Delta_{[m]\setminus\{i\}})\big{)} is a basic tile of order n#{j{0,,n1}J1|#Ij=1}+#{iJ2|ibI({1,,n})}n-\#\{j\in\{0,\dots,n-1\}\setminus J_{1}\,|\,\#I_{j}=1\}+\#\{i\in J_{2}\,|\,i\notin b_{I}(\{1,\dots,n\})\} plus one in case nJ1n\in J_{1} and #In=1\#I_{n}=1, see (3). But from Proposition 5.3 we know that TI(Δ{k1,,n}×Δ[m])T_{I}\cap\big{(}\Delta_{\{k_{1},\dots,n\}}\times\Delta_{[m]}\big{)} is empty if there exists j{0,,k11}j\in\{0,\dots,k_{1}-1\} such that #Ij1\#I_{j}\neq 1 while in the opposite case, this intersection contains TI(Δ[n]×Δ{k2,,m})T_{I}\cap\big{(}\Delta_{[n]}\times\Delta_{\{k_{2},\dots,m\}}\big{)} by Proposition 5.1 and is of codimension k1k_{1} in TIT_{I}. In this second case, TI(T1×T2)T_{I}\cap(T_{1}\times T_{2}) is Morse and regular, since its order is not more than m+nk1k2+2m+n-k_{1}-k_{2}+2. Moreover, the vertices of the Morse face are the greatest of ΔI\Delta_{I}. In the first case, we know from Proposition 5.1 that TI(Δ[n]×Δ{k2,,m})T_{I}\cap\big{(}\Delta_{[n]}\times\Delta_{\{k_{2},\dots,m\}}\big{)} is empty if there exists j{0,,n}j\in\{0,\dots,n\} such that #Ij1\#I_{j}\neq 1 and eI(j)<k2e_{I}(j)<k_{2} and of codimension k+k2k+k_{2} otherwise, where kk denotes the least integer such that #Ik1\#I_{k}\neq 1. The tile TI(T1×T2)T_{I}\cap(T_{1}\times T_{2}) is then again Morse and it is singular iff k=k11k=k_{1}-1, Ik={0,,m}I_{k}=\{0,\dots,m\}, J1={k1, ,n}J_{1}=\{k_{1}, \dots,n\} and J2={k2, ,m}J_{2}=\{k_{2}, \dots,m\}, since its order is bounded from above by m+nk1k2+2m+n-k_{1}-k_{2}+2 with equality, when k=k11k=k_{1}-1, only in this case. Moreover, the vertices of the Morse face are the greatest of ΔI\Delta_{I}. Hence, the product T1×T2T_{1}\times T_{2} inherits the Morse shelled primitive triangulation (TI(T1×T2))II(n,m)\big{(}T_{I}\cap(T_{1}\times T_{2})\big{)}_{I\in I(n,m)} which uses a unique critical tile, of index m+nk1k2+2m+n-k_{1}-k_{2}+2, iff T1T_{1} and T2T_{2} are critical of indices nk1+1n-k_{1}+1 and mk2+1m-k_{2}+1 respectively. In the case that nJ1n\in J_{1}, but mJ2m\notin J_{2}, we apply the involution EE to the tame Morse shelling (TJ(T2×T1))JI(m,n)\big{(}T_{J}\cap(T_{2}\times T_{1})\big{)}_{J\in I(m,n)} we just obtained and deduce from Lemma 4.8 that the collection (𝒫(TI)(T1×T2))II(n,m)\big{(}{\cal P}(T_{I})\cap(T_{1}\times T_{2})\big{)}_{I\in I(n,m)} defines a tame Morse shelling on T1×T2T_{1}\times T_{2}. Theorem 3.14 is thus proven in the case of non basic tiles.

5.2.3 The mixed case of one basic and one non basic tiles

In the mixed case, we may assume that T1T_{1} is Morse and T2T_{2} basic, applying the involution E:T1×T2T2×T1E:T_{1}\times T_{2}\to T_{2}\times T_{1} if necessary. Then the chosen orders on the vertices of the underlying simplices σ1\sigma_{1} and σ2\sigma_{2} induce isomorphisms between them and Δ[n](jJ1Δ[n]{j}Δ{k1,,n})\Delta_{[n]}\setminus\big{(}\cup_{j\in J_{1}}\Delta_{[n]\setminus\{j\}}\cup\Delta_{\{k_{1},\dots,n\}}\big{)} and Δ[m]iJ2Δ[m]{i}\Delta_{[m]}\setminus\cup_{i\in J_{2}}\Delta_{[m]\setminus\{i\}} respectively, where 1<k1n1<k_{1}\leq n, J1{k1, ,n}J_{1}\subset\{k_{1}, \dots,n\} and J2{0, ,m}J_{2}\subset\{0, \dots,m\}. We first assume that if nJ1n\in J_{1}, then mJ2m\in J_{2}. In this case, we deduce from §5.2.1 that for every II(n,m)I\in I(n,m), TI(jJ1(Δ[n]{j}×Δ[m])iJ2{0}(Δ[n]×Δ[m]{i}))T_{I}\setminus\big{(}\cup_{j\in J_{1}}(\Delta_{[n]\setminus\{j\}}\times\Delta_{[m]})\cup_{i\in J_{2}\setminus\{0\}}(\Delta_{[n]}\times\Delta_{[m]\setminus\{i\}})\big{)} is a basic tile of order n#{j{0,,n1}J1|#Ij=1}+#{iJ2{0}|ibI({1,,n})}n-\#\{j\in\{0,\dots,n-1\}\setminus J_{1}\,|\,\#I_{j}=1\}+\#\{i\in J_{2}\setminus\{0\}\,|\,i\notin b_{I}(\{1,\dots,n\})\} plus one in case nJ1n\in J_{1} and #In=1\#I_{n}=1, see (3). But from Proposition 5.3 we know that TI(Δ{k1,,n}×Δ[m])T_{I}\cap\big{(}\Delta_{\{k_{1},\dots,n\}}\times\Delta_{[m]}\big{)} is empty if there exists j{0,,k11}j\in\{0,\dots,k_{1}-1\} such that #Ij1\#I_{j}\neq 1 and in the opposite case, this intersection contains TI(Δ[n]×Δ{1,,m})T_{I}\cap\big{(}\Delta_{[n]}\times\Delta_{\{1,\dots,m\}}\big{)} by Proposition 5.1 and is of codimension k1k_{1} in TIT_{I}. In this second case, TI(T1×T2)T_{I}\cap(T_{1}\times T_{2}) is Morse, since if nJ1n\in J_{1} and #In1\#I_{n}\neq 1, then TI(Δ{0,,n1}×Δ[m])T_{I}\cap\big{(}\Delta_{\{0,\dots,n-1\}}\times\Delta_{[m]}\big{)} is contained in the facet TI(Δ[n]×Δ{0,,m1})T_{I}\cap\big{(}\Delta_{[n]}\times\Delta_{\{0,\dots,m-1\}}\big{)} which is removed by hypothesis. This tile is moreover regular, since its order is not more than m+nk1m+n-k_{1}, and the vertices of the Morse face are the greatest of ΔI\Delta_{I}. In the first case, we know from Proposition 5.1 that TI(Δ[n]×Δ{1,,m})T_{I}\cap\big{(}\Delta_{[n]}\times\Delta_{\{1,\dots,m\}}\big{)} is of codimension k+1k+1, kk being the least integer less than k1k_{1} such that #Ik1\#I_{k}\neq 1. The tile TI(T1×T2)T_{I}\cap(T_{1}\times T_{2}) is then again Morse and it is singular iff k=k11k=k_{1}-1, Ik={0,,m}I_{k}=\{0,\dots,m\}, J1={k1, ,n}J_{1}=\{k_{1}, \dots,n\} and J2={0, ,m}J_{2}=\{0, \dots,m\}, since its order is bounded from above by m+nkm+n-k with equality only in this case. Moreover, the vertices of the Morse face are the greatest of ΔI\Delta_{I}. Hence, the product T1×T2T_{1}\times T_{2} inherits the tame Morse shelled primitive triangulation (TI(T1×T2))II(n,m)\big{(}T_{I}\cap(T_{1}\times T_{2})\big{)}_{I\in I(n,m)} which uses a unique critical tile, of index m+nk1+1m+n-k_{1}+1, iff T1T_{1} and T2T_{2} are critical, of indices nk1+1n-k_{1}+1 and mm respectively. If now nJ1n\in J_{1}, but mJ2m\notin J_{2}, we apply the exchange involution EE. We deduce from (3) that for every II(m,n)I\in I(m,n), TI(iJ2(Δ[m]{i}×Δ[n])jJ1(Δ[m]×Δ[n]{j}))T_{I}\setminus\big{(}\cup_{i\in J_{2}}(\Delta_{[m]\setminus\{i\}}\times\Delta_{[n]})\cup_{j\in J_{1}}(\Delta_{[m]}\times\Delta_{[n]\setminus\{j\}})\big{)} is a basic tile of order m#{i{0,,m1}J2|#Ii=1}+#{jJ1|jbI({1,,m})}m-\#\{i\in\{0,\dots,m-1\}\setminus J_{2}\,|\,\#I_{i}=1\}+\#\{j\in J_{1}\,|\,j\notin b_{I}(\{1,\dots,m\})\}. From Proposition 5.1 we know that TI(Δ[m]×Δ{k1,,n})T_{I}\cap(\Delta_{[m]}\times\Delta_{\{k_{1},\dots,n\}}) is empty if there exists i{0,,m1}i\in\{0,\dots,m-1\} such that bI(i)eI(i)<k1b_{I}(i)\neq e_{I}(i)<k_{1} and otherwise, it is of codimension k+k1k+k_{1} in TIT_{I}, where kk denotes the least integer such that #Ik1\#I_{k}\neq 1. Moreover, the vertices of the Morse face are the greatest of ΔI\Delta_{I}. We deduce that TI(T2×T1)T_{I}\cap(T_{2}\times T_{1}) is Morse and its order is bounded from above by m+n+1kk1m+n+1-k-k_{1}, with equality only if J1={k1, ,n}J_{1}=\{k_{1}, \dots,n\}, J2=J_{2}=\emptyset and Im={0,,n}I_{m}=\{0,\dots,n\}. The collection (TI(T2×T1))II(m,n)\big{(}T_{I}\cap(T_{2}\times T_{1})\big{)}_{I\in I(m,n)} thus defines a tame Morse shelling of T2×T1T_{2}\times T_{1} which contains a unique critical tile, of index n+1k1n+1-k_{1} iff T1T_{1} and T2T_{2} are both critical, of indices n+1k1n+1-k_{1} and 0 respectively. By Lemma 4.8, the collection (𝒫(TI)(T1×T2))II(n,m)\big{(}{\cal P}(T_{I})\cap(T_{1}\times T_{2})\big{)}_{I\in I(n,m)} then defines the tame Morse shelling on T1×T2T_{1}\times T_{2} we are looking for. Hence the result.

5.2.4 Remarks on the proof of Theorem 3.14

1) The shelling of T1×T2T_{1}\times T_{2} is inherited from the shelling of σ1×σ2\sigma_{1}\times\sigma_{2} given by Corollary 4.6, via the choice of total orders on the vertices of the underlying simplices σ1\sigma_{1} and σ2\sigma_{2}. When T1T_{1} and T2T_{2} are both basic, these orders fix isomorphisms between T1T_{1}, T2T_{2} and Δ[n]jJ1Δ[n]{j}\Delta_{[n]}\setminus\cup_{j\in J_{1}}\Delta_{[n]\setminus\{j\}}, Δ[m]iJ2Δ[m]{i}\Delta_{[m]}\setminus\cup_{i\in J_{2}}\Delta_{[m]\setminus\{i\}} respectively, where n=dimσ1n=\dim\sigma_{1}, m=dimσ2m=\dim\sigma_{2}, J1{0, ,n}J_{1}\subset\{0, \dots,n\} and J2{0, ,m}J_{2}\subset\{0, \dots,m\}. In order to get hh-tilings on T1×T2T_{1}\times T_{2}, we need to assume the

Condition hh : If {0,n}J1\{0,n\}\subset J_{1}, then {0,m}J2\{0,m\}\cap J_{2}\neq\emptyset and vice versa, if {0,m}J2\{0,m\}\subset J_{2}, then {0,n}J1\{0,n\}\cap J_{1}\neq\emptyset.

Indeed, for example, neither the shelling (TI)II(n,m)(T_{I})_{I\in I(n,m)}, nor the shelling (𝒫(TI))II(n,m)\big{(}{\cal P}(T_{I})\big{)}_{I\in I(n,m)} induces an hh-tiling on Δ[n]((Δ{0,,n1}×Δ[m])(Δ{1,,n}×Δ[m]))\Delta_{[n]}\setminus\big{(}(\Delta_{\{0,\dots,n-1\}}\times\Delta_{[m]})\cup(\Delta_{\{1,\dots,n\}}\times\Delta_{[m]})\big{)} in general.

2) Likewise, when T1T_{1} or T2T_{2} is not basic, we need to assume the

Condition MM : If T1T_{1} (resp. T2T_{2}) is a Morse tile which is not basic, then the vertices of its Morse face are the greatest among those of σ1\sigma_{1} (resp. of σ2\sigma_{2}).

Indeed, for example, neither the shelling (TI)II(n,m)(T_{I})_{I\in I(n,m)}, nor the shelling (𝒫(TI))II(n,m)\big{(}{\cal P}(T_{I})\big{)}_{I\in I(n,m)} induces a Morse tiling on (Δ[n](Δ{0,,n1}Δ{n1,n}))×(Δ[m](Δ{1,,m}Δ{0,1}))\big{(}\Delta_{[n]}\setminus(\Delta_{\{0,\dots,n-1\}}\cup\Delta_{\{n-1,n\}})\big{)}\times\big{(}\Delta_{[m]}\setminus(\Delta_{\{1,\dots,m\}}\cup\Delta_{\{0,1\}})\big{)} in general.

3) Finally, even if these Conditions hh or MM are satisfied, the shelling (TI)II(n,m)(T_{I})_{I\in I(n,m)} of Δ[n]×Δ[m]\Delta_{[n]}\times\Delta_{[m]} given by Corollary 4.6 need not induce a shelling on T1×T2T_{1}\times T_{2}, it has been sometimes necessary to apply the palindromic isomorphism 𝒫{\cal P} to (TI)II(n,m)(T_{I})_{I\in I(n,m)}, which amounts to reverse the total orders of the vertices of σ1\sigma_{1} and σ2\sigma_{2}. The latter does not affect the triangulation (ΔI)II(n,m)(\Delta_{I})_{I\in I(n,m)} given by Corollary 4.6.

5.3 Proof of Theorem 3.16

Let us first prove that we may assume the tiles to be basic.

Proposition 5.5.

Let T1,T2T_{1},T_{2} be two Morse tiles with underlying basic tiles T1,T2T^{\prime}_{1},T^{\prime}_{2}. Then, the Morse shellings of T1×T2T_{1}\times T_{2} and T1×T2T^{\prime}_{1}\times T^{\prime}_{2} given by Theorem 3.14 have same hh-vector.

Proof.

If T1T_{1} (resp. T2T_{2}) is a Morse tile which is not basic, we denote by σ1\sigma_{1} (resp. σ2\sigma_{2}) its underlying simplex and by τ1\tau_{1} (resp. τ2\tau_{2}) its Morse face. The Morse shelling given by Theorem 3.14 is inherited from a total order on the vertices of σ1\sigma_{1} (resp. σ2\sigma_{2}) such that the ones of τ1\tau_{1} (resp. τ2\tau_{2}) are the greatest, see Condition MM in §5.2.4. The product σ1×σ2\sigma_{1}\times\sigma_{2} inherits then a triangulation (ΔI)II(n,m)(\Delta_{I})_{I\in I(n,m)} and a shelling (TI)II(n,m)(T_{I})_{I\in I(n,m)} given by Corollary 4.6. We know from Proposition 5.3 (resp. Proposition 5.1) that for every II(n,m)I\in I(n,m), the intersection of TIT_{I} with τ1×σ2\tau_{1}\times\sigma_{2} (resp. σ1×τ2\sigma_{1}\times\tau_{2}) is either empty, or of codimension at least equal to the one of τ1\tau_{1} in σ1\sigma_{1} (resp. τ2\tau_{2} in σ2\sigma_{2}). This intersection thus does not contribute to the order of TI(T1×T2)T_{I}\cap(T_{1}\times T_{2}). Thus, if the tame Morse shellling of T1×T2T_{1}\times T_{2} given by Theorem 3.14 is (TI(T1×T2))II(n,m)\big{(}T_{I}\cap(T_{1}\times T_{2})\big{)}_{I\in I(n,m)}, then the Morse shelling of T1×T2T^{\prime}_{1}\times T^{\prime}_{2} given by Theorem 3.14 is (TI(T1×T2))II(n,m)\big{(}T_{I}\cap(T^{\prime}_{1}\times T^{\prime}_{2})\big{)}_{I\in I(n,m)} and has the same hh-vector. Otherwise, Theorem 3.14 provides the Morse shellling (𝒫(TI)(T1×T2))II(n,m)\big{(}{\cal P}(T_{I})\cap(T_{1}\times T_{2})\big{)}_{I\in I(n,m)}, which is the image of (TJ(T2×T1))JI(m,n)\big{(}T_{J}\cap(T_{2}\times T_{1})\big{)}_{J\in I(m,n)}, under the exchange involution by Lemma 4.8, so that it provides the Morse shellling (𝒫(TI)(T1×T2))II(n,m)\big{(}{\cal P}(T_{I})\cap(T^{\prime}_{1}\times T^{\prime}_{2})\big{)}_{I\in I(n,m)} on T1×T2T^{\prime}_{1}\times T^{\prime}_{2} as well, which has same hh-vector as (𝒫(TI)(T1×T2))II(n,m)\big{(}{\cal P}(T_{I})\cap(T_{1}\times T_{2})\big{)}_{I\in I(n,m)}. Hence the result. ∎

Let us now prove that all Morse tilings of a product T1×T2T_{1}\times T_{2} given by Theorem 3.14 have same hh-vector. By Proposition 5.5 we may assume the tiles to be basic.

Proposition 5.6.

Let T1,T2T_{1},T_{2} be two basic tiles. Then, all Morse shellings of T1×T2T_{1}\times T_{2} given by Theorem 3.14 have same hh-vector.

Proof.

We may assume that T1=Δ[n]jJ1Δ[n]{j}T_{1}=\Delta_{[n]}\setminus\cup_{j\in J_{1}}\Delta_{[n]\setminus\{j\}} and T2=Δ[m]iJ2Δ[m]{i}T_{2}=\Delta_{[m]}\setminus\cup_{i\in J_{2}}\Delta_{[m]\setminus\{i\}}, where J1{0,,n}J_{1}\subset\{0,\dots,n\} and J2{0,,m}J_{2}\subset\{0,\dots,m\}. We have to assume in addition that if nJ1n\in J_{1}, then mJ2m\in J_{2}. This can be done after possibly applying the involution E:T1×T2T2×T1E:T_{1}\times T_{2}\to T_{2}\times T_{1} which exchanges the roles of T1T_{1} and T2T_{2}. The tiles of the Morse tiling given by Theorem 3.14 are then the (TI(T1×T2))II(n,m)\big{(}T_{I}\cap(T_{1}\times T_{2})\big{)}_{I\in I(n,m)} and we are going to prove that for every k{0,,m+n+1}k\in\{0,\dots,m+n+1\}, the number of tiles of order kk in this collection only depends on #J1\#J_{1} and #J2\#J_{2}. This implies the result, for applying the exchange involution EE or not does not affect the hh-vector as well. We proceed by induction on the dimensions n,mn,m. For every j1J1{0,n}j_{1}\in J_{1}\setminus\{0,n\} (resp. i2J2{0,m}i_{2}\in J_{2}\setminus\{0,m\}), let us compare the hh-vectors of the tilings of T1×T2T_{1}\times T_{2} and T1×T2T^{\prime}_{1}\times T_{2} (resp. T1×T2T_{1}\times T^{\prime}_{2}), where T1=Δ[n]jJ1{j1}Δ[n]{j}T^{\prime}_{1}=\Delta_{[n]}\setminus\cup_{j\in J_{1}\setminus\{j_{1}\}}\Delta_{[n]\setminus\{j\}} (resp. T2=Δ[m]iJ2{i2}Δ[m]{i}T^{\prime}_{2}=\Delta_{[m]}\setminus\cup_{i\in J_{2}\setminus\{i_{2}\}}\Delta_{[m]\setminus\{i\}}). Let II(n,m)I\in I(n,m). From Proposition 5.3 (resp. Proposition 5.1) we know that the intersection of Δ[n]{j1}×Δ[m]\Delta_{[n]\setminus\{j_{1}\}}\times\Delta_{[m]} (resp. Δ[n]×Δ[m]{i2}\Delta_{[n]}\times\Delta_{[m]\setminus\{i_{2}\}}) with TIT_{I} is empty if #Ij1>1\#I_{j_{1}}>1 (resp. if i2bI({1,,n})i_{2}\in b_{I}(\{1,\dots,n\})) and is of codimension one otherwise. In the first case, j1j_{1} (resp. i2i_{2}) does not contribute to the order of TI(T1×T2)T_{I}\cap(T_{1}\times T_{2}), so that this order coincides with the one of TI(T1×T2)T_{I}\cap(T^{\prime}_{1}\times T_{2}) (resp. TI(T1×T2)T_{I}\cap(T_{1}\times T^{\prime}_{2})) and in the second case, its contribution equals one, so that Ord(TI(T1×T2))=Ord(TI(T1×T2))+1\textup{Ord}(T_{I}\cap(T_{1}\times T_{2}))=\textup{Ord}(T_{I}\cap(T^{\prime}_{1}\times T_{2}))+1 (resp. Ord(TI(T1×T2))=Ord(TI(T1×T2))+1\textup{Ord}(T_{I}\cap(T_{1}\times T_{2}))=\textup{Ord}(T_{I}\cap(T_{1}\times T^{\prime}_{2}))+1). The staircases II(n,m)I\in I(n,m) for which #Ij1=1\#I_{j_{1}}=1 (resp. i2bI({1,,n})i_{2}\notin b_{I}(\{1,\dots,n\})) are in bijective correspondence with the staircases I~I(n1,m)\tilde{I}\in I(n-1,m) (resp. I~I(n,m1)\tilde{I}\in I(n,m-1)), this correspondence Forj1\textup{For}_{j_{1}} (resp. Fori2\textup{For}_{i_{2}}) being induced by the inclusion {0,,n}{j1}{0,,n}\{0,\dots,n\}\setminus\{j_{1}\}\to\{0,\dots,n\} (resp. {0,,m}{i2}{0,,m}\{0,\dots,m\}\setminus\{i_{2}\}\to\{0,\dots,m\}). But if I~=Forj1(I)\tilde{I}=\textup{For}_{j_{1}}(I) (resp. I~=Fori2(I)\tilde{I}=\textup{For}_{i_{2}}(I)) with #Ij1=1\#I_{j_{1}}=1 (resp. i2bI({1,,n})i_{2}\notin b_{I}(\{1,\dots,n\})) and if T~1=Δ[n]{j1}jJ1{j1}Δ[n]{j,j1}\widetilde{T}_{1}=\Delta_{[n]\setminus\{j_{1}\}}\setminus\cup_{j\in J_{1}\setminus\{j_{1}\}}\Delta_{[n]\setminus\{j,j_{1}\}} (resp. T~2=Δ[m]{i2}iJ2{i2}Δ[m]{i,i2}\widetilde{T}_{2}=\Delta_{[m]\setminus\{i_{2}\}}\setminus\cup_{i\in J_{2}\setminus\{i_{2}\}}\Delta_{[m]\setminus\{i,i_{2}\}}), then Ord(TI(T1×T2))=Ord(TI~(T~1×T2))\textup{Ord}(T_{I}\cap(T^{\prime}_{1}\times T_{2}))=\textup{Ord}(T_{\tilde{I}}\cap(\widetilde{T}_{1}\times T_{2})) (resp. Ord(TI(T1×T2))=Ord(TI~(T1×T~2))\textup{Ord}(T_{I}\cap(T_{1}\times T^{\prime}_{2}))=\textup{Ord}(T_{\tilde{I}}\cap(T_{1}\times\widetilde{T}_{2}))) by Proposition 5.3 (resp. Proposition 5.1). We deduce the relation

h(T1×T2)=h(T1×T2)h(T~1×T2)[0]+h(T~1×T2)[1],\displaystyle h(T_{1}\times T_{2})=h(T^{\prime}_{1}\times T_{2})-h(\widetilde{T}_{1}\times T_{2})^{[0]}+h(\widetilde{T}_{1}\times T_{2})^{[1]}, (4)

where for every v=(v0,,vm+n)m+n+1v=(v_{0},\dots,v_{m+n})\in\mathbb{Z}^{m+n+1}, v[0]=(v0,,vm+n,0)m+n+2v^{[0]}=(v_{0},\dots,v_{m+n},0)\in\mathbb{Z}^{m+n+2} and v[1]=(0,v0,,vm+n)m+n+2v^{[1]}=(0,v_{0},\dots,v_{m+n})\in\mathbb{Z}^{m+n+2}. Likewise,

h(T1×T2)=h(T1×T2)h(T1×T~2)[0]+h(T1×T~2)[1].\displaystyle h(T_{1}\times T_{2})=h(T_{1}\times T^{\prime}_{2})-h(T_{1}\times\widetilde{T}_{2})^{[0]}+h(T_{1}\times\widetilde{T}_{2})^{[1]}. (5)

By deleting one after the other the elements of J1{0,n}J_{1}\setminus\{0,n\} and J2{0,m}J_{2}\setminus\{0,m\}, we then express the hh-vector of T1×T2T_{1}\times T_{2} in terms of the hh-vector of product of tiles of dimensions n\leq n and m\leq m having lower order, and this expression does not depend on the specific position of the elements in J1{0,n}J_{1}\setminus\{0,n\} and J2{0,m}J_{2}\setminus\{0,m\}, it only depends on the number of such elements. We can likewise delete nJ1n\in J_{1} if mJ2m\in J_{2} and 0J20\in J_{2} if 0J10\in J_{1}, the formula (4) being valid in this case and we may also delete 0J10\in J_{1} if 0J20\notin J_{2} and mJ2m\in J_{2} if nJ1n\notin J_{1}, in the same way as before. The only delicate case is the case where 0J20\in J_{2} but 0J10\notin J_{1} since we have applied the exchange involution T1×T2T2×T1T_{1}\times T_{2}\to T_{2}\times T_{1} in order to make sure that mJ2m\in J_{2} if nJ1n\in J_{1}. In such a case, the tiling of T1×T2T_{1}\times T_{2} is Morse and not an hh-tiling. Let II(n,m)I\in I(n,m). The intersection of Δ[n]×Δ{1,,m}\Delta_{[n]}\times\Delta_{\{1,\dots,m\}} with TIT_{I} is no longer empty if #I0=1\#I_{0}=1, but it is of codimension greater than one in TIT_{I}, so that it is a Morse face which does not contribute to the order of TI(T1×T2)T_{I}\cap(T_{1}\times T_{2}). If #I0>1\#I_{0}>1, it is of codimension one and contributes as one to the order of TI(T1×T2)T_{I}\cap(T_{1}\times T_{2}). If we delete 0J20\in J_{2}, we thus again get, in the same way, the formula (5), even if the natures of the tilings of T1×T2T_{1}\times T_{2}, T1×T2T_{1}\times T^{\prime}_{2} and T1×T~2T_{1}\times\widetilde{T}_{2} now differ. We thus deduce by applying inductively finitely many times (4), (5) an expression of h(T1×T2)h(T_{1}\times T_{2}) in terms of the hh-vectors of the product of closed simplices of dimensions n\leq n and m\leq m and this expression only depends on the cardinalities of J1J_{1} and J2J_{2}. Moreover, the hh-vectors of the tilings (TI)II(n,m)(T_{I})_{I\in I(n,m)} or (𝒫(TI))II(n,m)({\cal P}(T_{I}))_{I\in I(n,m)} given by Corollary 4.6 coincide and only depend on the dimension n,mn,m of the closed simplices. The result follows. ∎

Remark 5.7.

The formulas (4), (5) make it possible to compute by induction the hh-vector given by Theorem 3.14 in terms of the hh-vectors given by Corollary 4.6. Moreover, the latter can be computed using a similar induction or by computing the face vector of the cartesian product of two simplices, but we do not detail these computations here.

It remains to prove the formula given in Theorem 3.16 and thanks to Proposition 5.5, it is enough to prove it for basic tiles.

Proposition 5.8.

Let T1,T2T_{1},T_{2} be two basic tiles and T𝑤𝑖𝑑𝑒𝑐ℎ𝑒𝑐𝑘1,T𝑤𝑖𝑑𝑒𝑐ℎ𝑒𝑐𝑘2\widecheck{T}_{1},\widecheck{T}_{2} be their dual ones. Then, all Morse shellings of T1×T2T_{1}\times T_{2} and T𝑤𝑖𝑑𝑒𝑐ℎ𝑒𝑐𝑘1×T𝑤𝑖𝑑𝑒𝑐ℎ𝑒𝑐𝑘2\widecheck{T}_{1}\times\widecheck{T}_{2} given by Theorem 3.14 satisfy h(T𝑤𝑖𝑑𝑒𝑐ℎ𝑒𝑐𝑘1×T𝑤𝑖𝑑𝑒𝑐ℎ𝑒𝑐𝑘2)=h𝑤𝑖𝑑𝑒𝑐ℎ𝑒𝑐𝑘(T1×T2)h(\widecheck{T}_{1}\times\widecheck{T}_{2})=\widecheck{h}(T_{1}\times T_{2}).

Proof.

Let n,mn,m (resp. k,lk,l) be the dimensions (resp. orders) of T1T_{1} and T2T_{2}. Let us choose total orders on the vertices of T1,T2T_{1},T_{2} (resp. Twidecheck1,Twidecheck2\widecheck{T}_{1},\widecheck{T}_{2}) in order to obtain isomorphisms with Δ[n]j=0k1Δ[n]{j}\Delta_{[n]}\setminus\cup_{j=0}^{k-1}\Delta_{[n]\setminus\{j\}} and Δ[m]i=0l1Δ[m]{i}\Delta_{[m]}\setminus\cup_{i=0}^{l-1}\Delta_{[m]\setminus\{i\}} (resp. Δ[n]j=0nkΔ[n]{j}\Delta_{[n]}\setminus\cup_{j=0}^{n-k}\Delta_{[n]\setminus\{j\}} and Δ[m]i=0mlΔ[m]{i}\Delta_{[m]}\setminus\cup_{i=0}^{m-l}\Delta_{[m]\setminus\{i\}}). From (3), we know that for every II(n,m)I\in I(n,m),

Ord(TI(T1×T2))=n#{j{k,,n1}|#Ij=1}+#{i{0,,l1}|ibI({1,,n})}\textup{Ord}\big{(}T_{I}\cap(T_{1}\times T_{2})\big{)}=n-\#\{j\in\{k,\dots,n-1\}\,|\,\#I_{j}=1\}+\#\{i\in\{0,\dots,l-1\}\,|\,i\notin b_{I}(\{1,\dots,n\})\}

plus one in case k=n+1k=n+1 and #In=1\#I_{n}=1. Likewise, by definition of Iˇ\check{I} and (3),

Ord(TIˇ(Twidecheck1×Twidecheck2))=n#{j{1,,k1}|#Ij=1}+#{i{l,,m}|ieI({0,,n1})}\textup{Ord}\big{(}T_{\check{I}}\cap(\widecheck{T}_{1}\times\widecheck{T}_{2})\big{)}=n-\#\{j\in\{1,\dots,k-1\}\,|\,\#I_{j}=1\}+\#\{i\in\{l,\dots,m\}\,|\,i\notin e_{I}(\{0,\dots,n-1\})\}

plus one in case k=0k=0 and #I0=1\#I_{0}=1. By adding these quantities, whatever the value of kk is, we get

Ord(TI(T1×T2))+Ord(TIˇ(Twidecheck1×Twidecheck2))=2n+m+1#{j{1,,n1}|#Ij=1}#eI({0,,n1}),\begin{array}[]{l}\textup{Ord}\big{(}T_{I}\cap(T_{1}\times T_{2})\big{)}+\textup{Ord}\big{(}T_{\check{I}}\cap(\widecheck{T}_{1}\times\widecheck{T}_{2})\big{)}\\ =2n+m+1-\#\{j\in\{1,\dots,n-1\}\,|\,\#I_{j}=1\}-\#e_{I}(\{0,\dots,n-1\}),\\ \end{array}

since eI(j)=bI(j+1)e_{I}(j)=b_{I}(j+1) for every j{0,,n1}j\in\{0,\dots,n-1\} by definition, see §4.1. Moreover, since eIe_{I} is increasing, #eI({0,,n1})=n#{j{1,,n1}|#Ij=1}\#e_{I}(\{0,\dots,n-1\})=n-\#\{j\in\{1,\dots,n-1\}\,|\,\#I_{j}=1\}, so that Ord(TI(T1×T2))+Ord(TIˇ(Twidecheck1×Twidecheck2))=m+n+1\textup{Ord}\big{(}T_{I}\cap(T_{1}\times T_{2})\big{)}+\textup{Ord}\big{(}T_{\check{I}}\cap(\widecheck{T}_{1}\times\widecheck{T}_{2})\big{)}=m+n+1. ∎

Theorem 3.14 now follows from Propositions 5.5, 5.6 and 5.8.

5.4 Proof of Corollary 3.18 and further examples

1) Corollary 3.17 is a special case of Theorem 3.14 which produces Morse shellings on handles of any dimension and index. Figure 10 provides some examples of such shellings, depicted using the associated mixed decompositions of the simplex Δ[m]\Delta_{[m]}, see §4.2.

Refer to caption
Figure 10: Morse shellings on handles.

In general, we may check that the Morse shelling of the handle Δn×Δm\stackrel{{\scriptstyle\circ}}{{\Delta}}_{n}\times\Delta_{m} given by Theorem 3.14 uses one critical tile of index nn, (m+n1n1){m+n-1\choose n-1} basic tiles of order n+1n+1 and for every l{2,,m}l\in\{2,\dots,m\}, (m+nln1){m+n-l\choose n-1} Morse tiles isomorphic to Tnm+n,m+nlT_{n}^{m+n,m+n-l}. Likewise, the handle Δn×Δm\Delta_{n}\times\stackrel{{\scriptstyle\circ}}{{\Delta}}_{m} is tiled by one critical tile of index mm, (m+n1n){m+n-1\choose n} basic tiles of order m+1m+1 and for every l{m,,m+n2}l\in\{m,\dots,m+n-2\}, (ll+1m){l\choose l+1-m} Morse tiles isomorphic to Tmm+n,lT_{m}^{m+n,l}. We do not detail these computations here.

2) Corollary 3.18 provides other special cases of shellings given by Theorem 3.14, namely hh-tilings whose tiles are all isomorphic to each other. Let us prove now this corollary.

Proof of Corollary 3.18.

Let T1=Δ[n]j=0n1Δ[n]{j}T_{1}=\Delta_{[n]}\setminus\cup_{j=0}^{n-1}\Delta_{[n]\setminus\{j\}} and T2=Δ[m]T_{2}=\Delta_{[m]}. The shelled triangulation of T1×T2T_{1}\times T_{2} given by Theorem 3.14 uses the tiles (TI(T1×T2))II(n,m)\big{(}T_{I}\cap(T_{1}\times T_{2})\big{)}_{I\in I(n,m)}. From (3), we know that for every II(n,m)I\in I(n,m), Ord(TI(T1×T2))=n\textup{Ord}\big{(}T_{I}\cap(T_{1}\times T_{2})\big{)}=n. Moreover, these tiles are all basic, so that the first part of Corollary 3.18 is proved. Likewise, let T1=Δ[n]Δ{1,,n}T^{\prime}_{1}=\Delta_{[n]}\setminus\Delta_{\{1,\dots,n\}} and T2=Δ[m]i=0mΔ[m]{i}T^{\prime}_{2}=\Delta_{[m]}\setminus\cup_{i=0}^{m}\Delta_{[m]\setminus\{i\}}. From (3), we know that for every II(n,m)I\in I(n,m),

Ord(TI(T1×T2))\displaystyle\textup{Ord}\big{(}T_{I}\cap(T^{\prime}_{1}\times T^{\prime}_{2})\big{)} =\displaystyle= n#{j{1,,n1}|#Ij=1}+#{i[m]|ibI({1,,n})}\displaystyle n-\#\{j\in\{1,\dots,n-1\}\,|\,\#I_{j}=1\}+\#\{i\in[m]\,|\,i\notin b_{I}(\{1,\dots,n\})\}
=\displaystyle= #eI({0,,n1})+m+1#eI({0,,n1})\displaystyle\#e_{I}(\{0,\dots,n-1\})+m+1-\#e_{I}(\{0,\dots,n-1\})
=\displaystyle= m+1.\displaystyle m+1.

Again, all these tiles are basic, hence the result. ∎

In the same way, Δ[n]×Tmm\Delta_{[n]}\times T_{m}^{m} (resp. Δn×T1m\stackrel{{\scriptstyle\circ}}{{\Delta}}_{n}\times T_{1}^{m}) inherits a shelled triangulation whose tiles are all isomorphic to each other, of order mm (resp. of order n+1n+1).

3) We have observed in §5.3 that the symmetry given by Theorem 3.16 is induced by the involution II(n,m)IˇI(n,m)I\in I(n,m)\mapsto\check{I}\in I(n,m). This symmetry appears on the examples given by Figures 11, 12, 13 and 14.

Refer to caption
Figure 11: The symmetry observed on some tilings, with n=4n=4.
Refer to caption
Figure 12: The symmetry observed on more tilings, with n=4n=4.
Refer to caption
Figure 13: The symmetry observed on more tilings, with n=4n=4.
Refer to caption
Figure 14: The symmetry observed on more tilings, with n=4n=4.

6 Tilings on products of two complexes

6.1 Proof of Theorems 3.8 and 3.10

Let KK and LL be the simplicial complexes underlying S1S_{1} and S2S_{2}. Let us equip their edges with orientations given by Definition 2.13, the tilings of S1S_{1} and S2S_{2} being tame by hypothesis. Then, the vertices of every simplex σ\sigma of KK and θ\theta of LL inherit a total order given by Proposition 2.11, so that σ×θ\sigma\times\theta gets equipped with a staircase triangulation given by Corollary 4.6. Moreover, the face inclusions preserve these orders by Proposition 2.11, so that the staircase triangulations on these products glue together to define a primitive triangulation on K×LK\times L, compare Lemma II.8.9II.8.9 of [5]. If τ1\tau_{1} and τ2\tau_{2} are shelled, the tiles of τ1\tau_{1} get labelled T1,,TN1T_{1},\dots,T_{N_{1}} and the tiles of τ2\tau_{2} labelled T1,,TN2T^{\prime}_{1},\dots,T^{\prime}_{N_{2}}. Let us label the underlying simplices σ1,,σN1\sigma_{1},\dots,\sigma_{N_{1}} and θ1,,θN2\theta_{1},\dots,\theta_{N_{2}}, they shell KK and LL respectively. The products σk×θl\sigma_{k}\times\theta_{l} get then ordered by the lexicographic order on pairs (k,l){1,,N1}×{1,,N2}(k,l)\in\{1,\dots,N_{1}\}\times\{1,\dots,N_{2}\}. By Corollary 4.6, the triangulation of each product σk×θl\sigma_{k}\times\theta_{l} is itself shelled, providing an order on its maximal simplices (ΔI(k,l))II(n,m)(\Delta_{I}^{(k,l)})_{I\in I(n,m)}, where nn denotes the dimension of σk\sigma_{k} and mm the dimension of θl\theta_{l}. The lexicographic order on triplets (k,l,I){1,,N1}×{1,,N2}×I(n,m)(k,l,I)\in\{1,\dots,N_{1}\}\times\{1,\dots,N_{2}\}\times I(n,m) induces then a shelling on the triangulated product K×LK\times L. We have here used a slight abuse of notation since the dimension nn (resp. mm) depends on kk (resp. ll) in general. Now, S1×S2S_{1}\times S_{2} is partitionned by the products Tk×TlT_{k}\times T^{\prime}_{l}, (k,l){1,,N1}×{1,,N2}(k,l)\in\{1,\dots,N_{1}\}\times\{1,\dots,N_{2}\}, and by Theorem 3.14, these products, equipped with the preceding triangulation, are Morse shellable. Indeed, the tilings of S1S_{1} and S2S_{2} being tame, we know from Definition 2.13 that Condition MM of §5.2.4 is satisfied and this guaranties the Morse shellability of Tk×TlT_{k}\times T^{\prime}_{l}. Again, the lexicographic order on triplets (k,l,I){1,,N1}×{1,,N2}×I(n,m)(k,l,I)\in\{1,\dots,N_{1}\}\times\{1,\dots,N_{2}\}\times I(n,m) induces a Morse shelling on the triangulated product S1×S2S_{1}\times S_{2}. Moreover, this shelling is tame since the ones of Tk×TlT_{k}\times T^{\prime}_{l} are by Theorem 3.14.

By Theorem 3.14, the critical tiles of S1×S2S_{1}\times S_{2} are then in bijective correspondence with the products Tk×TlT_{k}\times T^{\prime}_{l} of a critical tile TkT_{k} of S1S_{1} and a critical tile TlT^{\prime}_{l} of S2S_{2}, their indices being the sum of the index of TkT_{k} with the index of TlT^{\prime}_{l}. The cc-vector of S1×S2S_{1}\times S_{2} is thus the product c(S1)c(S2)c(S_{1})c(S_{2}) by definition. Finally, if the tiles of S1S_{1} (resp. of S2S_{2}) all have same dimension nn (resp. mm) and if h(S1)h(S_{1}) and h(S2)h(S_{2}) are palindromic, then, we may group the tiles of S1S_{1} (resp. of S2S_{2}) by pairs of tiles of order jj and n+1jn+1-j (resp. ii and m+1im+1-i), 0j<n+120\leq j<\frac{n+1}{2} (resp. 0i<m+120\leq i<\frac{m+1}{2}), leaving alone the tiles of order n+12\frac{n+1}{2} (resp. m+12\frac{m+1}{2}) in case nn (resp. mm) is odd. The products Tk×TlT_{k}\times T^{\prime}_{l}, (k,l){1,,N1}×{1,,N2}(k,l)\in\{1,\dots,N_{1}\}\times\{1,\dots,N_{2}\}, are then grouped by quadruples, pairs or left alone depending on the cases, but Theorem 3.16 ensures that the contribution of each group to the hh-vector of S1×S2S_{1}\times S_{2} is palindromic. Adding together the contributions of all these groups, we deduce Theorem 3.8. Now, under the hypothesis of Theorem 3.10, for every (k,l){1,,N1}×{1,,N2}(k,l)\in\{1,\dots,N_{1}\}\times\{1,\dots,N_{2}\}, Condition hh of §5.2.4 gets satisfied by the total orders on the vertices of TkT_{k} and TlT^{\prime}_{l}, so that the shelling (TI(Tk×Tl))II(n,m)\big{(}T_{I}\cap(T_{k}\times T^{\prime}_{l})\big{)}_{I\in I(n,m)} given by Theorem 3.14 uses only basic tiles, which proves Theorem 3.10. \square

6.2 Proof of Theorem 3.11

The simplicial complex Δ2\partial\Delta_{2} can be equipped with the hh-tiling using three tiles of dimension and order one, see Example 3.4. We may orient each edge in such a way that it goes towards the remaining vertex of each tile. Let KK be the simplicial complex underlying SS. We choose a total order on its vertices. The conditions of Theorem 3.10 are then satisfied so that S×Δ2S\times\partial\Delta_{2} inherits an hh-tiled primitive triangulation whose hh-vector is moreover palindromic provided h(S)h(S) is. For every tile T1T_{1} of SS and T2T_{2} of Δ2\partial\Delta_{2}, the orders chosen on vertices provide an isomorphism between T1×T2T_{1}\times T_{2} and (Δ[n]jJ1Δ[n]{j})×(Δ{0,1}Δ{0})(\Delta_{[n]}\setminus\cup_{j\in J_{1}}\Delta_{[n]\setminus\{j\}})\times(\Delta_{\{0,1\}}\setminus\Delta_{\{0\}}), where nn is the dimension of T1T_{1}. For every II(n,1)I\in I(n,1), only one interval Ij0I_{j_{0}} is not a singleton, j0{0,,n}j_{0}\in\{0,\dots,n\}, and the tile TI(Δ[n]×T2)T_{I}\cap(\Delta_{[n]}\times T_{2}) is of order one. Thus, the tile TI(T1×T2)T_{I}\cap(T_{1}\times T_{2}) is of order #J1\#J_{1} if j0J1j_{0}\in J_{1} and of order #J1+1\#J_{1}+1 otherwise, see (3). The tiling (TI(T1×T2))II(n,1)\big{(}T_{I}\cap(T_{1}\times T_{2})\big{)}_{I\in I(n,1)} uses then Ord(T1)\textup{Ord}(T_{1}) basic tiles of order Ord(T1)\textup{Ord}(T_{1}) and dim(T1)+1Ord(T1)\dim(T_{1})+1-\textup{Ord}(T_{1}) basic tiles of order Ord(T1)+1\textup{Ord}(T_{1})+1. Theorem 3.11 follows.

References

  • [1] L. J. Billera and C. W. Lee. A proof of the sufficiency of McMullen’s conditions for ff-vectors of simplicial convex polytopes. J. Combin. Theory Ser. A, 31(3):237–255, 1981.
  • [2] R. Bott and L. W. Tu. Differential forms in algebraic topology, volume 82 of Graduate Texts in Mathematics. Springer-Verlag, New York-Berlin, 1982.
  • [3] H. Bruggesser and P. Mani. Shellable decompositions of cells and spheres. Math. Scand., 29:197–205 (1972), 1971.
  • [4] J. Cerf and A. Gramain. Le théorème du hh-cobordisme (Smale). École Normale Supérieure, 1968.
  • [5] S. Eilenberg and N. Steenrod. Foundations of algebraic topology. Princeton University Press, Princeton, New Jersey, 1952.
  • [6] R. Forman. Morse theory for cell complexes. Adv. Math., 134(1):90–145, 1998.
  • [7] R. Forman. A user’s guide to discrete Morse theory. Sém. Lothar. Combin., 48:Art. B48c, 35, 2002.
  • [8] W. Fulton. Introduction to toric varieties, volume 131 of Annals of Mathematics Studies. Princeton University Press, Princeton, NJ, 1993. The William H. Roever Lectures in Geometry.
  • [9] I. M. Gelfand, M. M. Kapranov, and A. V. Zelevinsky. Discriminants, resultants, and multidimensional determinants. Mathematics: Theory & Applications. Birkhäuser Boston, Inc., Boston, MA, 1994.
  • [10] B. Huber, J. Rambau, and F. Santos. The Cayley trick, lifting subdivisions and the Bohne-Dress theorem on zonotopal tilings. J. Eur. Math. Soc. (JEMS), 2(2):179–198, 2000.
  • [11] S. Klain. Dehn-sommerville relations for triangulated manifolds. unpublished manuscript available at http://faculty.uml.edu/dklain/ds.pdf.
  • [12] D. Kozlov. Combinatorial algebraic topology, volume 21 of Algorithms and Computation in Mathematics. Springer, Berlin, 2008.
  • [13] W. B. R. Lickorish. Unshellable triangulations of spheres. European J. Combin., 12(6):527–530, 1991.
  • [14] I. G. Macdonald. Polynomials associated with finite cell-complexes. J. London Math. Soc. (2), 4:181–192, 1971.
  • [15] P. McMullen. The maximum numbers of faces of a convex polytope. Mathematika, 17:179–184, 1970.
  • [16] J. R. Munkres. Elements of algebraic topology. Addison-Wesley Publishing Company, Menlo Park, CA, 1984.
  • [17] N. Salepci and J.-Y. Welschinger. Tilings, packings and expected Betti numbers in simplicial complexes. arXiv:1806.05084, 2018.
  • [18] N. Salepci and J.-Y. Welschinger. Asymptotic measures and links in simplicial complexes. Discrete Comput. Geom., 62(1):164–179, 2019.
  • [19] N. Salepci and J.-Y. Welschinger. Morse shellability, tilings and triangulations. arXiv:1910.13241, 2019.
  • [20] F. Santos. The Cayley trick and triangulations of products of simplices. In Integer points in polyhedra—geometry, number theory, algebra, optimization, volume 374 of Contemp. Math., pages 151–177. Amer. Math. Soc., Providence, RI, 2005.
  • [21] R. P. Stanley. The upper bound conjecture and Cohen-Macaulay rings. Studies in Appl. Math., 54(2):135–142, 1975.
  • [22] R. P. Stanley. The number of faces of a simplicial convex polytope. Adv. in Math., 35(3):236–238, 1980.
  • [23] J. H. C. Whitehead. Simplicial Spaces, Nuclei and m-Groups. Proc. London Math. Soc. (2), 45(4):243–327, 1939.
  • [24] G. M. Ziegler. Lectures on polytopes, volume 152 of Graduate Texts in Mathematics. Springer-Verlag, New York, 1995.

Univ Lyon, Université Claude Bernard Lyon 1, CNRS UMR 5208, Institut Camille Jordan, 43 blvd. du 11 novembre 1918, F-69622 Villeurbanne cedex, France