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

\marginsize

2.5cm3cm2cm2cm

Edge connectivity of simplicial polytopes

Guillermo Pineda-Villavicencio & Julien Ugon Federation University Australia School of Information Technology, Deakin University work@guillermo.com.au julien.ugon@deakin.edu.au
Abstract.

A simplicial polytope is a polytope with all its facets being combinatorially equivalent to simplices. We deal with the edge connectivity of the graphs of simplicial polytopes. We first establish that, for any d3d\geq 3, for any d3d\geq 3, every minimum edge cut of cardinality at most 4d74d-7 in such a graph is trivial, namely it consists of all the edges incident with some vertex. A consequence of this is that, for d3d\geq 3, the graph of a simplicial dd-polytope with minimum degree δ\delta is min{δ,4d6}\min\{\delta,4d-6\}-edge-connected. In the particular case of d=3d=3, we have that every minimum edge cut in a plane triangulation is trivial; this may be of interest to researchers in graph theory.

Second, for every d4d\geq 4 we construct a simplicial dd-polytope whose graph has a nontrivial minimum edge cut of cardinality (d2+d)/2(d^{2}+d)/2. This gives a simplicial 4-polytope with a nontrivial minimum edge cut that has ten edges. Thus, the aforementioned result is best possible for simplicial 44-polytopes.

Key words and phrases:
stacked polytope, simplex, simplicial polytope, edge connectivity, edge cut
2010 Mathematics Subject Classification:
Primary 52B05; Secondary 52B12
Julien Ugon’s research was partially supported by ARC discovery project DP180100602.

1. Introduction

A (convex) polytope is the convex hull of a finite set XX of points in d\mathbb{R}^{d}; the convex hull of XX is the smallest convex set containing XX. The dimension of a polytope in d\mathbb{R}^{d} is one less than the maximum number of affinely independent points in the polytope; a set of points p1,,pkp_{1},\ldots,p_{k} in d\mathbb{R}^{d} is affinely independent if the k1k-1 vectors p1pk,,pk1pkp_{1}-p_{k},\ldots,p_{k-1}-p_{k} are linearly independent. A polytope of dimension dd is referred to as a dd-polytope.

A polytope is structured around other polytopes, its faces. A face of a polytope PP in d\mathbb{R}^{d} is PP itself, or the intersection of PP with a hyperplane in d\mathbb{R}^{d} that contains PP in one of its closed halfspaces. A face of dimension 0, 1, and d1d-1 in a dd-polytope is a vertex, an edge, and a facet, respectively. The sets of vertices and edges of a polytope or a graph are denoted by 𝒱\mathcal{V} and \mathcal{E}, respectively. The graph G(P)G(P) of a polytope PP is the graph with vertex set 𝒱(P)\mathcal{V}(P) and edge set (P)\mathcal{E}(P).

Unless otherwise stated, the graph theoretical notation and terminology follow from [2] and the polytope theoretical notation and terminology from [14]. Moreover, when referring to graph-theoretical properties of a polytope such as minimum degree, connectivity, and edge connectivity, we mean properties of its graph.

This paper studies the edge connectivity of a simplicial polytope, namely the edge connectivity of the graph of the polytope. A simplicial polytope is a polytope with all its facets being simplices. By a simplex we mean any polytope that is combinatorially isomorphic to a simplex. Two polytopes PP and PP^{\prime} are combinatorially isomorphic if their face lattices are isomorphic.

For a graph GG, we say that a set Z𝒱(G)(G)Z\subseteq\mathcal{V}(G)\cup\mathcal{E}(G) separates two distinct vertices x,yx,y, if every path in GG from xx to yy contains an element of ZZ. A set ZZ separates GG if it separates two vertices of GG. A graph is rr-edge-connected, for r0r\geq 0, if no two vertices are separated by fewer than rr edges. And a graph with at least r+1r+1 vertices is rr-connected if removing any r1r-1 vertices leaves a connected subgraph. Balinski [1] showed the following, which implies that the graph of a dd-polytope is dd-edge-connected.

Theorem 1 (Balinski [1]).

For d2d\geq 2, the graph of a dd-polytope is dd-connected.

We have recently studied other notions of connectivity in graphs of polytopes. The paper [4] studied the vertex connectivity of cubical polytopes and showed a cubical dd-polytope with minimum degree δ\delta is min{δ,2d2}\min\{\delta,2d-2\}-connected, which implies that min{δ,2d2}\min\{\delta,2d-2\}-edge-connected. In addition, the paper [5] analysed the linkedness of cubical polytopes, a stronger notion of connectivity.

A separating set DD of edges is an edge cut in GG if there exists a nonempty proper subset X𝒱(G)X\subseteq\mathcal{V}(G) such that D=(X,𝒱(G)X)D=\mathcal{E}(X,\mathcal{V}(G)\setminus X); here (X,𝒱(G)X)\mathcal{E}(X,\mathcal{V}(G)\setminus X) denotes the set of all edges from a vertex xXx\in X to a vertex y𝒱(G)Xy\in\mathcal{V}(G)\setminus X. Henceforth, we write (X,X¯)\mathcal{E}(X,\overline{X}) instead of (X,𝒱(G)X)\mathcal{E}(X,\mathcal{V}(G)\setminus X). Every minimal separating set of edges in a connected graph GG is an edge cut [2, Sec. 2.5]. In addition, a trivial edge cut is one whose edges are the ones incident with a single vertex; otherwise the edge cut is nontrivial. We prove the following.

Theorem (Edge connectivity theorem). For every d3d\geq 3, in a simplicial dd-polytope every minimum edge cut of cardinality at most 4d74d-7 is trivial.

Let δ(G)\delta(G) or δ(P)\delta(P) be the minimum degree of a vertex in a graph GG or a polytope PP. A corollary trivially ensues.

Corollary. For every d3d\geq 3, a simplicial dd-polytope with minimum degree δ\delta is min{δ,4d6}\min\{\delta,4d-6\}-edge-connected.

The graph of a 3-polytope is 3-connected and planar graph by Steinitz’s theorem [13], and a planar embedding of the graph of a simplicial 3-polytope is a plane triangulation GG; therefore, abusing terminology slightly, we will use interchangeably the terms plane triangulation and simplicial 3-polytope. In this case, Euler’s formula [6, 7] implies that minimum degree of GG is at most five. Hence, a minimum separating set of edges in GG would have at most five edges. Thus, another corollary is the following.

Corollary. Every minimum edge cut in a plane triangulation is trivial.

For the interest of graph theorists, we provide a short, graph-theoretical proof of this corollary in Section 2; surprisingly, this result seems to be new.

Finally, for every d4d\geq 4, we construct a simplicial dd-polytope with a minimum edge cut that contains (d2/2+d/2)(d^{2}/2+d/2) edges and does not consist of the edges incident with a vertex. This shows that the aforementioned theorem is best possible for simplicial 4-polytopes. We suspect the following is true.

Conjecture 2.

For every d3d\geq 3, there is a function f(d)f(d) quadratic in dd such that, in a simplicial dd-polytope, every minimum edge cut of cardinality at most f(d)f(d) is trivial.

The simplicial dd-polytopes we constructed contain an empty (d1)(d-1)-simplex, a set of dd vertices that does not form a face of the polytope but every proper subset does, and their graphs contain large complete graphs. So perhaps our theorem holds for all minimum edge cuts in certain classes of simplicial polytopes.

Problem 3.

For d3d\geq 3, is every minimum edge cut of a flag dd-polytope or a balanced dd-polytope trivial?

A flag polytope is a simplicial polytope whose proper faces are the complete graphs of the graph, and so it contains no empty simplices. And a balanced dd-polytope is a simplicial dd-polytope whose vertices can be coloured with dd colours such that adjacent vertices receive different colours, which implies that its graph does not contain complete graphs with more than dd vertices.

2. Plane triangulations

Our first theorem (Theorem 5) is about plane triangulations and may be of independent interest to graph theorists, and so we provide an independent, graph-theoretical proof.

For two vertices x,yx,y of a graph GG, we say that a path L:=x1xnL:=x_{1}\ldots x_{n} is an xyx-y path in GG if 𝒱(L){x}={x1}\mathcal{V}(L)\cap\{x\}=\{x_{1}\} and 𝒱(L){y}={xn}\mathcal{V}(L)\cap\{y\}=\{x_{n}\}. For an edge cut DD, let 𝒱(D)\mathcal{V}(D) denote the vertices in the edges of DD.

If GG is a 3-connected plane graph, for a vertex xx let x\mathcal{F}_{x} be the faces of GG that contain xx. The link of xx in GG, denoted link(x)\operatorname{link}(x), is the subgraph of GG induced by the vertices and edges in x\mathcal{F}_{x} that are disjoint from xx. It is a standard result that the link of a vertex in a 3-connected plane graph is a cycle that contains the neighbours of the vertex; see, for instance, [2, Cor. 10.8]. Since a plane triangulation can be considered as a simplicial 3-polytope, we could have used the standard definition of a link in Section 3 and Proposition 7 to justify these assertions, but we want this section to be devoid of polytope theory.

Lemma 4.

Let GG be a dd-connected graph, and let D:=(X,X¯)D:=\mathcal{E}(X,\overline{X}) be a nontrivial minimum edge cut of GG for some X𝒱(G)X\subseteq\mathcal{V}(G). Then #(X𝒱(D))d\#(X\cap\mathcal{V}(D))\geq d and #(X¯𝒱(D))d\#(\overline{X}\cap\mathcal{V}(D))\geq d.

Proof.

Let nX:=#(X𝒱(D))n_{X}:=\#(X\cap\mathcal{V}(D)) and let δ:=δ(G)\delta:=\delta(G). If nXd1n_{X}\leq d-1, then removing these nXn_{X} vertices does not disconnect the graph. Thus X=X𝒱(D)X=X\cap\mathcal{V}(D). For any xXx\in X, xx is incident with at least δnX+1\delta-n_{X}+1 edges in DD (as deg(x)δ\deg(x)\geq\delta). Since #Dδ\#D\leq\delta, we have nX(δnX+1)δn_{X}(\delta-n_{X}+1)\leq\delta, and so δnXd1\delta\leq n_{X}\leq d-1, a contradiction to the fact that δd\delta\geq d. Hence nXdn_{X}\geq d. The same reasoning yields that #(X¯𝒱(D))d\#(\overline{X}\cap\mathcal{V}(D))\geq d. ∎

Theorem 5 (Edge connectivity of plane triangulations).

Every minimum edge cut in a plane triangulation is trivial.

Proof.

Let GG be a plane triangulation, and let D:=(X,X¯)D:=\mathcal{E}(X,\overline{X}) be a minimum edge cut for some X𝒱(G)X\subseteq\mathcal{V}(G). By Euler’s formula, the minimum degree of a plane graph is at most five. Thus #D5\#D\leq 5. Suppose, by way of contradiction, that DD is nontrivial. It is a standard result of graph theory that a plane triangulation is 3-connected; this also follows from Balinski’s theorem (1). Let nX:=#(X𝒱(D))n_{X}:=\#(X\cap\mathcal{V}(D)) and nX¯:=#(X¯𝒱(D))n_{\overline{X}}:=\#(\overline{X}\cap\mathcal{V}(D)). Then, Lemma 4 ensures that nX3n_{X}\geq 3 and nX¯3n_{\overline{X}}\geq 3.

There is a vertex uX𝒱(D)u\in{X}\cap\mathcal{V}(D) with a unique neighbour vv in X¯𝒱(D)\overline{X}\cap\mathcal{V}(D); if every vertex in X𝒱(D){X}\cap\mathcal{V}(D) was incident with at least two edges from DD, then #D6\#D\geq 6, a contradiction. Moreover, since DD is nontrivial, the vertex uu has at least one neighbour in XX, say u1u_{1}. It follows that every u1vu_{1}-v path in GG passes through an edge of DD. Thus, the edge cut DD must separate u1u_{1} from vv in the link of uu. Let Du:=D(link(u))D_{u}:=D\cap\mathcal{E}(\operatorname{link}(u)). The set DuD_{u} contains precisely the edges incident with vv in (link(u))\mathcal{E}(\operatorname{link}(u)); otherwise either DuD_{u} consists of the edges incident with u1u_{1} or there would be at least two nonadjacent edges in DuD_{u}, both cases implying the existence of at least two neighbours of uu in X¯\overline{X}. Because link(u)\operatorname{link}(u) is a cycle (see also Proposition 7), we conclude that vv is incident with at least three edges in DD, including uvuv.

Again, since #D5\#D\leq 5 and vv is incident with at least three edges in DD, there is a vertex vX¯𝒱(D)v^{\prime}\in\overline{X}\cap\mathcal{V}(D) with a unique neighbour uu^{\prime} in X𝒱(D){X}\cap\mathcal{V}(D). Following the same line of reasoning as before, the vertex uu^{\prime} is incident with uvu^{\prime}v^{\prime} and exactly two edges in Dlink(v)D\cap\operatorname{link}(v^{\prime}), one of these two edges is uvu^{\prime}v. By the 3-connectivity of GG, the vertices u,vu^{\prime},v cannot disconnect the graph. Thus, there is at least one vertex in (X𝒱(D)){u,u}(X\cap\mathcal{V}(D))\setminus\{u,u^{\prime}\} adjacent to a vertex in X¯{v,v}\overline{X}\setminus\{v^{\prime},v\}. Since this means that #D6\#D\geq 6, we arrive at a contradiction. The proof of the theorem is complete. ∎

As a corollary we get the following.

Corollary 6.

A plane triangulation with minimum degree δ\delta is δ\delta-edge-connected.

3. Simplicial polytopes

The boundary complex of a polytope PP is the set of faces of PP other than PP itself. And the link of a vertex xx in PP, denoted link(x)\operatorname{link}(x), is the set of faces of PP that do not contain xx but lie in a facet of PP that contains xx. We require a result from [14], which we proved in [3, Prop. 12].

Proposition 7 ([14, Ex. 8.6]).

Let PP be a dd-polytope. Then the link of a vertex in PP is combinatorially isomorphic to the boundary complex of a (d1)(d-1)-polytope. In particular, for each d3d\geq 3, the graph of the link of a vertex is isomorphic to the graph of a (d1)(d-1)-polytope.

Let GG be a graph and let X𝒱(G)X\subseteq\mathcal{V}(G), then G[X]G[X] denotes the subgraph of GG induced by the set XX and GXG-X denotes the subgraph G[𝒱(G)X]G[\mathcal{V}(G)\setminus X] of GG. We also require Menger’s theorem [10].

Theorem 8 (Menger, [10]).

Let GG be a graph, and let xx and yy be two nonadjacent of its vertices. Then the minimum number of vertices separating xx from yy in GG equals the maximum number of pairwise internally disjoint xyx-y paths in GG.

An extension of the reasoning in the proof of Theorem 5 yields the main result of the paper.

A vertex adjacent to a vertex xx in a graph GG is a neighbour of xx. We denote by 𝒩G(x)\mathcal{N}_{G}(x) the set of neighbours of xx in GG. We extend this notation to neighbours that belong to a subgraph or a subset of vertices of GG; for instance, if Y𝒱(G)Y\subseteq\mathcal{V}(G) then 𝒩Y(x)\mathcal{N}_{Y}(x) denote the neighbours of xx in YY.

Theorem 9.

For each d3d\geq 3, in a simplicial dd-polytope every minimum edge cut of cardinality at most 4d74d-7 is trivial.

Proof.

Let PP be a simplicial dd-polytope, let GG be its graph, let δ\delta be the minimum degree of GG, and let D:=(X,X¯)D:=\mathcal{E}(X,\overline{X}) be a minimum edge cut of GG for some X𝒱(G)X\subseteq\mathcal{V}(G). By our hypothesis, #D4d7\#D\leq 4d-7. The theorem holds for d=3d=3 by Theorem 5, and so assume that d4d\geq 4. Let nX:=#(X𝒱(D))n_{X}:=\#(X\cap\mathcal{V}(D)) and nX¯:=#(X¯𝒱(D))n_{\overline{X}}:=\#(\overline{X}\cap\mathcal{V}(D)). Suppose, by way of contradiction, that DD is nontrivial. By Balinski’s theorem (1), GG is dd-connected, in which case Lemma 4 ensures that nXdn_{X}\geq d and nX¯dn_{\overline{X}}\geq d. We need two simple claims.

Claim 1.

There is a vertex wX𝒱(D)w\in X\cap\mathcal{V}(D) such that 1#𝒩X¯(w)d21\leq\#\mathcal{N}_{\overline{X}}(w)\leq d-2. Similarly, there is a vertex zX¯𝒱(D)z\in\overline{X}\cap\mathcal{V}(D) such that 1#𝒩X(z)d21\leq\#\mathcal{N}_{{X}}(z)\leq d-2.

Proof. If, for every vertex xx in X𝒱(D)X\cap\mathcal{V}(D), we have that #𝒩X¯(x)d1\#\mathcal{N}_{\overline{X}}(x)\geq d-1, then #Dd(d1)>4d7\#D\geq d(d-1)>4d-7 for d4d\geq 4 (as nXdn_{X}\geq d), a contradiction. The other statement is proved analogously. \square

Claim 2.

For every vertex wX𝒱(D)w\in X\cap\mathcal{V}(D) such that 1#𝒩X¯(w)d21\leq\#\mathcal{N}_{\overline{X}}(w)\leq d-2, then every vertex in 𝒩X¯(w)\mathcal{N}_{\overline{X}}(w) is incident with at least d+1#𝒩X¯(w)d+1-\#\mathcal{N}_{\overline{X}}(w) edges from DD. Similarly, for every vertex zX¯𝒱(D)z\in\overline{X}\cap\mathcal{V}(D) such that 1#𝒩X(z)d21\leq\#\mathcal{N}_{{X}}(z)\leq d-2, then every vertex in 𝒩X(z)\mathcal{N}_{{X}}(z) is incident with at least d+1#𝒩X(z)d+1-\#\mathcal{N}_{{X}}(z) edges from DD.

Proof. Since DD is nontrivial, there exists a vertex w1𝒩X(w)w_{1}\in\mathcal{N}_{X}(w). Because #𝒩G(w)d\#\mathcal{N}_{G}(w)\geq d and #𝒩X¯(w)d2\#\mathcal{N}_{\overline{X}}(w)\leq d-2, the vertex ww has neighbours in both XX and X¯\overline{X}, and so there must exist a separating set of edges in D(link(w))D\cap\mathcal{E}(\operatorname{link}(w)).

Consider a vertex v𝒩X¯(w)v\in\mathcal{N}_{\overline{X}}(w). As the graph of link(w)\operatorname{link}(w) is isomorphic to the graph of a simplicial (d1)(d-1)-polytope (Proposition 7), the vertex vv has degree at least d1d-1 in link(w)\operatorname{link}(w). Of the neighbours of vv in link(w)\operatorname{link}(w), at most #𝒩X¯(w)1\#\mathcal{N}_{\overline{X}}(w)-1 of them are in link(w)X¯\operatorname{link}(w)\cap\overline{X}, since 𝒱(link(w))=𝒩X(w)𝒩X¯(w)\mathcal{V}(\operatorname{link}(w))=\mathcal{N}_{X}(w)\cup\mathcal{N}_{\overline{X}}(w). It follows that, of the neighbours of vv in link(w)\operatorname{link}(w), at least d1(#𝒩X¯(w)1)d-1-(\#\mathcal{N}_{\overline{X}}(w)-1) are in XX; in other words, vv is incident with at least d𝒩X¯(w)d-\mathcal{N}_{\overline{X}}(w) edges in D(link(w))D\cap\mathcal{E}(\operatorname{link}(w)). If we also count the edge wv(link(kw))wv\notin\mathcal{E}(\operatorname{link}(kw)), then we get the desired number of edges incident with vv. The statement about the vertex zX¯𝒱(D)z\in\overline{X}\cap\mathcal{V}(D) is proved analogously. \square

Let k:=min({#𝒩X¯(x):xX𝒱(D)}{#𝒩X(x):xX¯𝒱(D)})k:=\min(\{\#\mathcal{N}_{\overline{X}}(x):x\in X\cap\mathcal{V}(D)\}\cup\{\#\mathcal{N}_{{X}}(x):x\in\overline{X}\cap\mathcal{V}(D)\}), and let u𝒱(D)u\in\mathcal{V}(D) be a vertex such that #𝒩X¯(u)=k\#\mathcal{N}_{\overline{X}}(u)=k. Without loss of generality, assume that uX𝒱(D)u\in X\cap\mathcal{V}(D). Then, by Claim 1 we have that #𝒩X¯(u)d2\#\mathcal{N}_{\overline{X}}(u)\leq d-2, and by Claim 2 every vertex vv in 𝒩X¯(u)\mathcal{N}_{\overline{X}}(u) is incident with d+1kd+1-k edges of DD. Every vertex in 𝒩X¯(u)\mathcal{N}_{\overline{X}}(u) is in 𝒱(D)X¯\mathcal{V}(D)\cap\overline{X}, and so there are at least dkd-k vertices in (𝒱(D)X¯)link(u)(\mathcal{V}(D)\cap\overline{X})\setminus\operatorname{link}(u) (as nX¯dn_{\overline{X}}\geq d). Because every vertex in 𝒱(D)\mathcal{V}(D) is incident with at least kk edges from DD, for k2k\geq 2 we get that

#Dk(d+1k)+k(dk)=2k(dk)+k4d6,\#D\geq k(d+1-k)+k(d-k)=2k(d-k)+k\geq 4d-6,

a contradiction. Therefore k=1k=1. Denote by vv the unique vertex in 𝒩X¯(u)\mathcal{N}_{\overline{X}}(u). By Claim 2 we have that #𝒩X(v)d\#\mathcal{N}_{X}(v)\geq d.

If every vertex in X¯𝒱(D)\overline{X}\cap\mathcal{V}(D) other than vv was incident with at least three edges in DD, then we would have #Dd+3(d1)=4d3\#D\geq d+3(d-1)=4d-3, and so there exists a vertex vX¯𝒱(D)v^{\prime}\in\overline{X}\cap\mathcal{V}(D) such that #𝒩X(v)2\#\mathcal{N}_{X}(v^{\prime})\leq 2. We consider two cases according to the cardinality of 𝒩X(v)\mathcal{N}_{X}(v^{\prime}).

First suppose that #𝒩X(v)=1\#\mathcal{N}_{X}(v^{\prime})=1. Then the unique vertex u𝒩X(v)u^{\prime}\in\mathcal{N}_{X}(v^{\prime}) is incident with at least dd edges from DD by Claim 2. Since GG is dd-connected by Balinski’s theorem (1), GG would remain (d2)(d-2)-connected after removing uu^{\prime} and vv. By Menger’s theorem (8), there are (d2)(d-2) pairwise internally disjoint uvu-v^{\prime} paths in G{u,v}G-\{u^{\prime},v\}, and so DD contains d2d-2 pairwise disjoint edges not containing uu^{\prime} or vv, say u1v1,,ud2vd2u_{1}v_{1},\ldots,u_{d-2}v_{d-2} with uiXu_{i}\in X and viX¯v_{i}\in\overline{X}. It is not possible that each uiu_{i} is incident with at least three edges in DD, as otherwise Claim 2 would ensure that every vertex viv_{i} (i=1,,d2i=1,\ldots,d-2) would be incident with at least d2d-2 edges in DD. Thus, counting the edges in DD incident with vv, vv^{\prime}, and v1,,vd2v_{1},\ldots,v_{d-2} we get that #Dd+1+3(d2)=4d5\#D\geq d+1+3(d-2)=4d-5. So there is a vertex uiXu_{i}\in X adjacent to at most two vertices in X¯\overline{X}, one of which is viv_{i}. By Claim 2, this implies that #𝒩X(vi)d+12\#\mathcal{N}_{X}(v_{i})\geq d+1-2. Of the edges in DD not incident with any vertex in {v,v,vi}\{v,v^{\prime},v_{i}\}, d3d-3 are edges ujvju_{j}v_{j} for jij\neq i, and another d3d-3 of them are edges incident with uu^{\prime}. Therefore,

#D#𝒩X(v)+#𝒩X(vi)+#𝒩X(v)+2(d3)d+d1+1+2(d3)=4d6,\#D\geq\#\mathcal{N}_{X}(v)+\#\mathcal{N}_{X}(v_{i})+\#\mathcal{N}_{X}(v^{\prime})+2(d-3)\geq d+d-1+1+2(d-3)=4d-6,

another contradiction. Thus, #𝒩X(v)=2\#\mathcal{N}_{X}(v^{\prime})=2.

The proof of the case #𝒩X(v)=2\#\mathcal{N}_{X}(v^{\prime})=2 is analogous to the proof of the case #𝒩X(v)=1\#\mathcal{N}_{X}(v^{\prime})=1, but we provide the details for the sake of completeness. Let {u,u′′}:=𝒩X(v)\{u^{\prime},u^{\prime\prime}\}:=\mathcal{N}_{X}(v^{\prime}). Then Claim 2 yields that both uu^{\prime} and u′′u^{\prime\prime} are incident with d1d-1 edges from DD. Removing uu^{\prime}, u′′u^{\prime\prime}, and vv would make GG (d3)(d-3)-edge-connected by Balinski’s theorem (1). So DD contains d3d-3 pairwise disjoint edges not containing uu^{\prime}, u′′u^{\prime\prime}, or vv, say u1v1,,ud3vd3u_{1}v_{1},\ldots,u_{d-3}v_{d-3} with uiXu_{i}\in X and viX¯v_{i}\in\overline{X}. It is not possible that each uiu_{i} is incident with at least three edges in DD; otherwise Claim 2 would ensure that every vertex vi,i=1,,d3v_{i},i=1,\ldots,d-3 would be incident with at least d2d-2 edges in DD, which yields that #Dd+2+3(d2)=4d4>4d7\#D\geq d+2+3(d-2)=4d-4>4d-7 by counting the edges of DD incident with vv, vv^{\prime}, and v1,,vd3v_{1},\ldots,v_{d-3}. So there is a vertex uiXu_{i}\in X adjacent to at most two vertices in X¯\overline{X}, one of which is viv_{i}. By Claim 2, this implies that #𝒩X(vi)d1\#\mathcal{N}_{X}(v_{i})\geq d-1. Of the edges in DD not incident with any vertex in {v,v,vi}\{v,v^{\prime},v_{i}\}, d4d-4 are edges ujvju_{j}v_{j} for jij\neq i, and another 2d262d-2-6 of them are edges incident with uu^{\prime} or u′′u^{\prime\prime}. Therefore, for d5d\geq 5 we have that

#D#𝒩X(v)+#𝒩X(vi)+#𝒩X(v)+d4+2d8d+d1+2+3d12=5d11>4d7,\#D\geq\#\mathcal{N}_{X}(v)+\#\mathcal{N}_{X}(v_{i})+\#\mathcal{N}_{X}(v^{\prime})+d-4+2d-8\geq d+d-1+2+3d-12=5d-11>4d-7,

another contradiction. In the case d=4d=4 we get #D=5d11\#D=5d-11. This means that #𝒩X(v)=4\#\mathcal{N}_{X}(v)=4, #𝒩X(vi)=3\#\mathcal{N}_{X}(v_{i})=3, and #𝒩X(v)=2\#\mathcal{N}_{X}(v^{\prime})=2. But, since nX¯dn_{\overline{X}}\geq d, there should exist a vertex v′′v^{\prime\prime} in X¯𝒱(D)\overline{X}\cap\mathcal{V}(D) other than {v,vi,v}\{v,v_{i},v^{\prime}\}; this gives an edge that has not been accounted for. This final contradiction completes the proof of the theorem. ∎

A simple corollary of Theorem 9 is the following.

Corollary 10.

For d3d\geq 3, a simplicial dd-polytope with minimum degree δ\delta is min{δ,4d6}\min\{\delta,4d-6\}-edge-connected.

3.1. A construction of nontrivial minimum edge cuts

We construct a simplicial dd-polytope that shows that the bound of 4d74d-7 in Theorem 9 is best possible for d=4d=4.

Let PP and PP^{\prime} be two dd-polytopes with a facet FF of PP projectively isomorphic to a facet FF^{\prime} of PP^{\prime}. The connected sum of PP and PP^{\prime} is obtained by “gluing” PP and PP^{\prime} along FF and FF^{\prime}. Projective transformations on the polytopes PP and PP^{\prime} may be required for the connected sum to be convex. The connected sum of two polytopes is depicted in Fig. 1. Our construction is based on performing connected sums along simplex facets. This is always possible because every polytope combinatorially isomorphic to a simplex is projectively isomorphic to the simplex [9].

Refer to caption
Figure 1. Connected sum of two polytopes.

We define stacked polytopes recursively. A dd-simplex is stacked, and each stacked dd-polytope with nd+2n\geq d+2 vertices is obtained as the connected sum of a stacked dd-polytope with n1n-1 vertices and a dd-simplex along a simplex facet.

Proposition 11.

For each d4d\geq 4, there is a simplicial dd-polytope with minimum degree at least (d2+d)/2{(d^{2}+d)}/{2} and a nontrivial minimum edge cut with (d2+d)/2{(d^{2}+d)}/{2} edges.

Proof.

Consider a family of stacked dd-polytopes PjP_{j} with j=d+1,,2dj=d+1,\ldots,2d vertices constructed as follows:

  1. (i)

    Pd+1P_{d+1} is a dd-simplex, with vertices labelled x1,,xd+1x_{1},\ldots,x_{d+1}.

  2. (ii)

    For j=d+1,,2d1j=d+1,\ldots,2d-1, the polytope Pj+1P_{j+1} is the connected sum of the dd-simplex and the polytope PjP_{j}, along the simplex facet of PjP_{j} containing the vertices xj+1d,xj+1d+1,,xjx_{j+1-d},x_{j+1-d+1},\ldots,x_{j}. Label the last vertex added xj+1x_{j+1}. The vertices xj+2d,xj+2d+1,,xj+1x_{j+2-d},x_{j+2-d+1},\ldots,x_{j+1} form a facet of Pj+1P_{j+1}, since they form a facet of the simplex we just added.

The polytope P2dP_{2d} contains two disjoint simplex facets F0F_{0} containing the vertices x1,,xdx_{1},\ldots,x_{d} and F1F_{1} containing the vertices xd+1,,x2dx_{d+1},\ldots,x_{2d}. Let DD be the set of the edges in P2dP_{2d} from a vertex in F0F_{0} to a vertex in F1F_{1}. Then DD is a nontrivial edge cut of P2dP_{2d} with (3d2d)/2d(d1)=(d2+d)/2{(3d^{2}-d)}/{2}-d(d-1)={(d^{2}+d)}/{2} edges.

Let CC be a cyclic 44-polytope with degree at least (d2+d)/2{(d^{2}+d)}/{2}, and let PP be the connected sum C#F0P2d#F1CC\#_{F_{0}}P_{2d}\#_{F_{1}}C. The polytope PP is the desired simplicial dd-polytope. It has minimum degree at least (d2+d)/2{(d^{2}+d)}/{2}, and the set DD is a nontrivial edge cut of PP. Say that D=(X,X¯)D=\mathcal{E}(X,\overline{X}) for some X𝒱(G)X\subseteq\mathcal{V}(G). Then the edge cut DD is minimal in PP, as both subgraphs G[X]G[X] and G[X¯]G[\overline{X}] are connected.

It remains to prove that DD is a minimum edge cut of G(P)G(P). Let C0C_{0} and C1C_{1} be the two copies of CC in PP. Then (P)=(C0)(C1)D\mathcal{E}(P)=\mathcal{E}(C_{0})\cup\mathcal{E}(C_{1})\cup D. Consider a minimum edge cut FF in G(P)G(P). It follows that #F#D=(d2+d)/2\#F\leq\#D={(d^{2}+d)}/{2}. We can assume that F=FF′′F=F^{\prime}\cup F^{\prime\prime} where FDF^{\prime}\subseteq D and F′′D=F^{\prime\prime}\cap D=\emptyset. Because F′′(C0)(C1)F^{\prime\prime}\subseteq\mathcal{E}(C_{0})\cup\mathcal{E}(C_{1}) and both C0C_{0} and C1C_{1} are (d2+d)/2{(d^{2}+d)}/{2}-edge-connected, removing the edges of F′′F^{\prime\prime} does not disconnect the subgraph G(C0)G(C_{0}) or G(C1)G(C_{1}). In addition, if FF^{\prime} is a proper subset of DD, then removing FF^{\prime} does not disconnect G(P2d)G(P_{2d}), as DD is minimal. It is now plain that F=DF=D, implying that DD is a minimum edge cut.

When d=4d=4 this construction is best possible by Theorem 9. ∎

References

  • [1] Balinski, M. L. (1961). On the graph structure of convex polyhedra in nn-space. Pacific J. Math., 11, 431–434.
  • [2] J. A. Bondy and U. S. R. Murty, Graph Theory, Graduate Texts in Mathematics, vol. 244, Springer, New York, 2008.
  • [3] H. T. Bui, G. Pineda-Villavicencio, and J. Ugon, The linkedness of cubical polytopes, arXiv:1802.09230, 2019.
  • [4] H. T. Bui, G. Pineda-Villavicencio, and J. Ugon, Connectivity of cubical polytopes, J. Combin. Theory Ser. A 169 (2020), 105–126.
  • [5] H. T. Bui, G. Pineda-Villavicencio, and J. Ugon, The linkedness of cubical polytopes: The cube, Electron. J. Comb. 28 (2021), P3.45.
  • [6] L. Euler, Elementa doctrinae solidorum, Novi Comm. Acad. Sci. Imp. Petropol. (1758), 109–140.
  • [7] L. Euler, Demonstratio nonnullarum insignium proprietatum quibas solida hedris planis inclusa sunt praedita, Novi Comm. Acad. Sci. Imp. Petropol. 4 (1758), 140–160.
  • [8] J. E. Goodman, J. O’Rourke, and C. D. Tóth (eds.), Handbook of discrete and computational geometry, 3rd ed., Chapman & Hall/CRC, Boca Raton, FL, 2017.
  • [9] P. McMullen, Constructions for projectively unique polytopes, Discrete Math. 14 (1976), no. 4, 347–358.
  • [10] K. Menger, Zur allgemeinen kurventheorie, Fundam. Math. 10 (1927), no. 1, 96–115 (ger).
  • [11] G. Pineda-Villavicencio, A new proof of Balinski’s theorem on the connectivity of polytopes, Discrete Math. 344 (2021), Article 112408.
  • [12] J. Richter-Gebert, Realization spaces of polytopes, Springer, Berlin, 2006.
  • [13] E. Steinitz, Polyeder und raumeinteilungen, Encyk der Math Wiss 3 (1922), 1–139.
  • [14] G. M. Ziegler, Lectures on polytopes, Graduate Texts in Mathematics, vol. 152, Springer, New York, 1995.