Research Article  Open Access
Francesco M. Malvestuto, Mauro Mezzini, Marina Moscarini, "Equivalence between Hypergraph Convexities", International Scholarly Research Notices, vol. 2011, Article ID 806193, 22 pages, 2011. https://doi.org/10.5402/2011/806193
Equivalence between Hypergraph Convexities
Abstract
Let be a connected graph on . A subset of is allpaths convex (or convex) if contains each vertex on every path joining two vertices in and is monophonically convex (or convex) if contains each vertex on every chordless path joining two vertices in . First of all, we prove that convexity and convexity coincide in if and only if is a tree. Next, in order to generalize this result to a connected hypergraph , in addition to the hypergraph versions of convexity and convexity, we consider canonical convexity (or convexity) and simplepath convexity (or convexity) for which it is well known that convexity is finer than both convexity and convexity and convexity is finer than convexity. After proving convexity is coarser than convexity, we characterize the hypergraphs in which each pair of the four convexities above is equivalent. As a result, we obtain a convexitytheoretic characterization of Bergeacyclic hypergraphs and of acyclic hypergraphs.
1. Introduction
Convexity is a fundamental concept occurring in geometry, topology, and functional analysis, and the problem of computing convex hulls is at the core of many computer engineering applications, for instance, in robotics, computer graphics, or optimization (see pageββ125 in [1]). For the theory of abstract convex structures see [2].
This paper is motivated by the paper by Farber and Jamison [3] where structural properties of different notions of convexity in graphs and hypergraphs are stated. We focus on the following four notions of convexities in hypergraphs: convexity (for monophonic convexity) [4], convexity (for canonical convexity) [4], convexity (for simplepath convexity) [3], and convexity (for allpaths convexity), which are known to be related to each other by the following implications: Note that hypergraph convexity generalizes graph convexity [3, 5], and both hypergraph convexity and convexity generalize graph convexity [6, 7]. After proving the implication convex ββconvex, we characterize the hypergraphs in which each pair of the four convexities above is equivalent.
As an application, suppose that for two hypergraph convexities and we have two algorithms and to compute convex hulls and convex hulls for hypergraphs and that the computational complexity of is less than . If and are equivalent in a class of hypergraphs then, for every hypergraph belonging to this class, we can compute convex hulls using the algorithm instead of .
As a consequence of our equivalence results, we obtain the following convexitytheoretic characterizations of the socalled βBergeacyclicβ hypergraphs and βacyclicβ hypergraphs [8]:(i)a hypergraph is Bergeacyclic if and only if convexity and convexity are equivalent in the hypergraph (see Theorem 6.9);(ii)a hypergraph is Bergeacyclic if and only if convexity and convexity are equivalent in the hypergraph (see Theorem 6.10);(iii)a hypergraph is acyclic if and only if convexity and convexity are equivalent in the hypergraph (see Theorem 5.5);(iv)a hypergraph is acyclic if and only if convexity and convexity are equivalent in the hypergraph (see Theorem 5.6).
It is worth noting that Bergeacyclic and acyclic hypergraphs belong to a family of hypergraphs (, and Bergeacyclic hypergraphs) which enjoy a number of theoretical and computational properties in database theory [8, 9], artificial intelligence [10, 11], and statistics [12, 13]. Moreover, convexity is a βconvex geometryβ in acyclic hypergraphs [4], and acyclicity characterizes those hypergraphs in which convexity yields a βconvex geometryβ [3].
To achieve the abovementioned results, we need several standard hypergraphtheoretic definitions (such as the four acyclicity types above) and additional notions which are introduced ad hoc. The outline of the paper is as follows. Section 2 contains basic definitions and results on graphs and hypergraphs. In Section 3 we review basic results on structural properties of convexity, convexity, convexity, and convexity in graphs and hypergraphs; moreover, we state some preliminary results. In Section 4 we characterize the hypergraphs in which convexity and convexity are equivalent. In Section 5 we prove that acyclicity characterizes those hypergraphs in which convexity (or convexity) and convexity are equivalent. In Section 6 we characterize the hypergraphs in which convexity and convexity are equivalent; moreover, we prove that Bergeacyclicity characterizes those hypergraphs in which convexity (or convexity) and convexity are equivalent. Section 7 contains an open problem for future research.
2. Terminology and Notation
2.1. Graphs
Henceforth, we only consider graphs [14] with no loops and no multiple edges, which henceforth will be referred simply to as graphs. Let be a graph. Two vertices of are adjacent if they are joined by some edge of . A nonempty subset of is a clique if every two distinct vertices in are adjacent. The subgraph of induced by a nonempty subset of is the graph, denoted by , with vertex set in which two distinct vertices are adjacent if and only if they are adjacent in . The notation is abridged into .
A path is a sequence , , of distinct vertices such that and are adjacent for . The path is said to join and (or, equivalently, to be an β path) and to have length ; moreover, if , is said to pass through each , , and two vertices and on are said to be consecutive if . By we denote the vertex set .
The distance between two vertices and is the length of any minimumlength β path. A graph is distance hereditary if, for every two vertices and of and for every connected, induced subgraph containing and , the distances between and in and are the same.
A cycle of length , , is a sequence where is a path, and the vertices and are adjacent. Two vertices and on the cycle are consecutive if either or . By we denote the set of vertices . A chord of is an edge joining two nonconsecutive vertices on .
A graph is chordal if every cycle of length at least 4 has a chord. A graph is strongly chordal if it is chordal and, for every cycle of even length, there are two nonconsecutive vertices at odd distance on that are adjacent. A graph is Ptolemaic if it is distance hereditary and chordal.
Let be a graph with at least two vertices. A vertex of is a cut vertex (or an βarticulation pointβ) of if the number of connected components of is greater than the number of connected components of or, equivalently, there exist two vertices and in the connected component of containing such that every β path passes through [14]. A block of is a maximal connected partial graph of containing no cut vertices. A block of is trivial if it consists of two vertices and nontrivial otherwise. Finally, is a block graph if the vertex set of every block of is a clique.
Proposition 2.1 (see [14]). Let be a nontrivial block. For every three distinct vertices , and of , there exists a β path that passes through .
Proposition 2.2 (see [14]). Let be a connected graph, and let be a block of . If is not a vertex of , then contains a cut vertex of such that, for every vertex of B, every β path passes through .
Let be a graph. Two connected vertices are separated by a subset of if they are in two distinct components of the induced graph . A nonempty subset of is nonseparable if is connected and no two vertices in are separated by a clique of . The prime components of are the subgraphs of induced by maximal nonseparable sets.
2.2. Hypergraphs
Generalizing notions and convexities from graphs to hypergraphs is not always straightforward, because there are often several nonequivalent ways to do this and different terminologies. This is true also for notions that hypergraph convexities are based on. For example, βsimple pathsβ in [3] are called βchordless chainsβ in [15], βsimple circuitsβ in [3] are called βchordless cyclesβ in [15], and βweak cyclesβ in [3] if they are of length at least 3, βnestβ vertices in [3] are called βsimpleβ vertices in [15].
The following basic definitions are taken from [16].
A (generic) hypergraph is a (possibly empty) set of nonempty sets; the elements of are the (hyper) edges of and their union is the vertex set of , denoted by .
A hypergraph is trivial if it has only one edge and nontrivial otherwise. A partial (sub) hypergraph of hypergraph is any subset of . A hypergraph is simple if no edge is contained in another edge. The reduction of a hypergraph is the partial hypergraph of whose edges are the maximal (with respect to setinclusion) edges of .
Let be a nonempty subset of . The subhypergraph of induced by is the hypergraph, denoted by , whose edges are exactly the maximal (with respect to setinclusion) edges of the hypergraph . The notation is abridged into .
A partial edge is a nonempty vertex set that is contained in some edge. Two vertices are adjacent if they belong together in some edge. A nonempty subset of is a clique if every two distinct vertices in are adjacent. A hypergraph is conformal if every clique is a partial edge.
A path is a sequence , , where the βs are pairwise distinct vertices, the βs are pairwise distinct edges, and for . The path is said to join and (or, equivalently, to be an β path), to have length and, if , to pass through each , . Moreover, two vertices and on the path are consecutive if . Finally, by we denote the vertex set , and by we denote the partial hypergraph of .
A hypergraph is connected if any two vertices are joined by a path. A nonempty subset of is connected if the induced subhypergraph is connected. Note that if is a graph, then the subgraph of induced by equals the reduction of .
The connected components of are the subhypergraphs of induced by maximal connected subsets of .
The following definitions of an βarticulation setβ and of a βblockβ in a hypergraph are the natural generalizations of the notions of a cut vertex and of a block in a graph [8, 17].
Let be a reduced hypergraph. A separator is a nonempty subset of such that there exist two connected vertices of that are in two distinct components of the induced subhypergraph . A separator of is an articulation set if it is the intersection of two edges of . A nonempty subset of is nonseparable if is connected and has no articulation set. A block of is the reduction of the subhypergraph of induced by a maximal nonseparable set.
Example 2.3. Let , where , , , , and . The hypergraph is shown in Figure 1. The set is the only articulation set of , and the blocks of are shown in Figure 2.
Finally, with a hypergraph we can associate two graphs: the β2sectionβ of and the βincidence graphβ of , which are defined as follows.
The 2section (also called βadjacency graphβ or βunderlying graphβ or βprimal graphβ, or βGaifman graphβ) of is the graph with vertex set in which two vertices are adjacent if and only if they are adjacent in . We denote the 2section of by .
The incidence graph of is the bipartite graph with bipartition , where and are joined by an edge if and only if . We denote the incidence graph of by , and the size of is the number of vertices and edges of [18]. Note that, if is connected, then the size of is where is the number of edges of .
2.3. Acyclicity
Fagin [8] introduced four notions of acyclicity for hypergraphs which are now recalled and, in the next sections, will be proven to be closely related to hypergraph convexities.
A cycle (also called a βcircuitβ [3]) is a sequence , , where is a path, and for . The cycle is said to have length ; moreover, two vertices and on are consecutive if either or . By we denote the set of vertices , and by we denote the partial hypergraph of .
A cycle is a cycle of length at least 3 such that at most one vertex in belongs to three or more edges of .
A cycle (a βweak cycleβ in [8]) is a cycle of length at least 3 such that every vertex in belongs to exactly two edges of .
A hypergraph is Bergeacyclic if it contains no cycles, acyclic if it contains no cycles, and acyclic if it contains no cycles (or, equivalently, if every partial hypergraph is acyclic).
A reduced hypergraph is acyclic if all its blocks are trivial hypergraphs. A hypergraph is acyclic precisely if its reduction is acyclic.
It is well known [8] that the following implications on hypergraphs hold: but none of their reverse implications holds in general.
Several characterizations of Bergeacyclicity, acyclicity, acyclicity, and acyclicity exist, and the following is based on the 2section of a hypergraph.
Proposition 2.4 (see [19]). A hypergraph is Bergeacyclic if and only if it is conformal and its 2section is a block graph. A hypergraph is acyclic if and only if it is conformal and its 2section is a Ptolemaic graph. A hypergraph is acyclic if and only if it is conformal and its 2section is a strongly chordal graph. A hypergraph is acyclic if and only if it is conformal and its 2section is a chordal graph.
By Proposition 2.4, acyclic hypergraphs are the same as βtotally balancedβ hypergraphs [3] and as βtotally decomposableβ hypergraphs in [12]. Finally, note that acyclic hypergraphs are called βacyclicβ hypergraphs in [4, 9, 11, 17, 18] and βdecomposableβ hypergraphs in [12, 13, 20, 21].
Before closing this subsection, we mention two acyclic hypergraphs which in some sense represent the βsuperstructureβ of a graph and of a hypergraph.
The prime hypergraph of graph is the (reduced) hypergraph whose edges are precisely the vertex sets of the prime components of .
Proposition 2.5 (see [21]). The prime hypergraph of a graph is a reduced, acyclic hypergraph.
A nonempty subset of is a compact set of if is connected and no partial edge of is a separator. Note that if is a compact set, then has no articulation set; but the reverse does not hold in general (see Example 2.3). A compact component of is the reduction of the subhypergraph of induced by a maximal compact set. The compact hypergraph of is the (reduced) hypergraph whose edges are precisely the vertex sets of the compact components of .
Proposition 2.6 (see [22, 23]). The compact hypergraph of a hypergraph is a reduced, acyclic hypergraph. Moreover, a hypergraph is acyclic if and only if its reduction equals its compact hypergraph.
Example 2.3 (continued). The compact components of are shown in Figure 3 and the compact hypergraph of is .
3. Convexities in Graphs and Hypergraphs
In this section, we recall the definitions and basic results on some convexities in graphs and hypergraphs. Moreover, we state some preliminary results.
Let be a connected hypergraph. A set of subsets of is a convexity space [24, 25] if(i)the empty set, the singletons, and belong to ,(ii) is closed under set intersection,(iii)every set in is connected.
The sets in a convexity space are called the convex sets of . For a subset of , the convex hull of is the minimal (with respect to set inclusion) convex set of that includes .
3.1. Graph Convexities
In this section, we recall the definitions and basic results on monophonic convexity (convexity) and allpaths convexity (convexity) on a connected graph [14, 17, 24β26]. Let be a connected graph. By and we denote the convexity space and the convexity space on , respectively.
A chord of a path is an edge of that joins two nonconsecutive vertices on . A path is chordless (or βinducedβ or βminimalβ) if it has no chords. A subset of is convex if, for every chordless path joining two vertices in , each vertex on belongs to .
The following result provides a known characterization of convex sets. Let be a subset of ; an β path is a path with , .
Theorem 3.1 (see [5]). Let be a connected graph. A subset of is convex if and only if, for every two distinct vertices and in , and are joined by an β path, then and are adjacent in .
Let be a connected graph with vertices and edges. Dourado et al. [25] gave an algorithm for computing the convex hull of a subset of which runs in time. A better algorithm was given by Kannan and Changat [27], which runs in time.
A subset of is if, for every path joining two vertices in , each vertex on belongs to . It is easily seen that if is a tree, then the convex hull of a subset of can be computed in time simply by deleting the leaves of that are not in .
We now state a characterization of those graphs on which = .
Theorem 3.2. Let be a connected graph.The equality holds if and only if is a tree.
Proof. Since every chordless path is trivially a path, the inclusion is obvious. If is a tree, then trivially one has . Assume that is not a tree. To prove that , consider any nontrivial block of , say . Let and be two adjacent vertices of . The path is the only chordless β path in and, hence, is convex. On the other hand, is not convex since, by Propositions 2.1 and 2.2, its convex hull is , which proves that .
By Theorem 3.2, if is a tree, then the convex hull of a subset of coincides with the convex hull of and, hence, can be computed in time.
3.2. Hypergraph Convexities
Let be a connected hypergraph. In this section we recall the definitions and basic results on monophonic convexity (convexity), canonical convexity (convexity), simplepath convexity (convexity), and allpaths convexity (convexity) on . By , , , and we denote the convexity space, the convexity space, the convexity space, and the convexity space on , respectively.
3.2.1. Convexity
Hypergraph convexity [4] was not defined in terms of paths but using the hypergraphtheoretic version of Theorem 3.1, that is, a subset of is convex if, for every two distinct vertices and in , and are joined by an β path, then and are adjacent in .
We now recall a useful characterization of convex sets. To this end, we need further definitions.
Let be a subset of ; two edges and of are connected outside , written , if(i) or(ii) or(iii)there exists an edge of such that and .
The edge relation is an equivalence relation; the classes of the resultant partition of will be referred to as the components of . The hypergraph is connected if has exactly one component. Note that is connected if the graph is connected and no edge of is completely contained in . For an component of , we call the set the boundary of with .
Example 3.3. Consider again the hypergraph of Example 2.3 (see Figure 1). For , the components of are and the boundary of with is , and the boundary of with is , and the boundary of with is .
Theorem 3.4 (see [4]). Let be a connected hypergraph. βA subset of is convex if and only if either or the boundary of with every component of is a clique of .
As we noted above, in contrast with convexity in graphs, the original definition of convexity in hypergraphs was given without having recourse to any path type. We now prove that convexity in hypergraphs can be related to the following generalization of the notion of a chordless path in a graph, which (is different from that given in [15] and) is defined as follows.
A chord of a path is a pair of nonconsecutive vertices on which are adjacent in . A path in is chordless if it has no chords.
Theorem 3.5. Let be a connected hypergraph. A subset of is convex if and only if, for every chordless path joining two vertices in , each vertex on belongs to .
Proof. (if) Assume that contains for every chordless path joining two vertices in . By Theorem 3.4, it is sufficient to prove that the boundary of with every component of is a clique of . Suppose, by contradiction, that there exists an component of such that the boundary of with contains two vertices and that are not adjacent in . Let be the boundary of with . Since and are not adjacent in and are not adjacent in and, since is connected, there exists a β path in of length at least 2 such that . Let be a β path of minimum length in . Of course, and is a chordless path; moreover, one has . Since and are not adjacent in must be of length at least 2 and, hence, there exists a vertex in that does not belong to . Since , does not belong to so that does not contain (contradiction).
(only if) Assume that is convex. By Theorem 3.4, the boundary of with every component of is a clique of . Suppose, by contradiction, that there exist two vertices and in and a chordless β path such that . Let and let . Of course, both and belong to the boundary of with the component of containing and, since the boundary of with every component of is a clique, and are adjacent. Since and are nonconsecutive on , the pair , is a chord of (contradiction).
Finally, let be a connected hypergraph with vertices and edges. An algorithm for computing convex hulls in a hypergraph is the βmonophonicclosure algorithmβ [4] which runs in time if the prime hypergraph of is given. Since the time needed to construct the prime hypergraph of is [21], where is the number of edges of , and since , the time complexity of the monophonicclosure algorithm is . On the other hand, it is easy to check that is a chordless path in if and only if is a chordless path in , which implies that so that, given , the convex hull of any vertex set in can be computed in time, that is, in time using the KannanChangat algorithm.
3.2.2. Convexity
A subset of is convex if the boundary of with every component of is a partial edge of . Let be a connected hypergraph with vertices and edges. It is proven in [4] that convex hulls can be computed using the MaierUllman algorithm [28], which runs in time. A more efficient algorithm is the βcanonicalclosure algorithmβ [4], which runs in time if the compact hypergraph of is given. Since the time needed to construct the compact hypergraph of is [23], the time complexity of the canonicalclosure algorithm is . Note that if is acyclic, then the time complexity of the algorithm reduces to ; however, we can do better using the βselectivereduction algorithmβ [18] which is linear in the size of .
3.2.3. Convexity
A path in is simple [3] if every two nonconsecutive vertices on are not adjacent in the partial hypergraph ; equivalently, a path in is simple if is a chordless path in . A subset of is convex if, for every simple path joining two vertices in , each vertex on belongs to . Let be a connected hypergraph with vertices and edges. An efficient algorithm for computing convex hulls was given in [29], which runs in , where is the number of edges of the incidence graph of . Since , the time complexity of the algorithm is . However, if is acyclic, then using the AnsteeFarber algorithm [15], the convex hull of vertex set can be computed in time simply by deleting the βnestβ vertices of that are not in (see [12]).
3.2.4. Convexity
Let be a connected hypergraph. The convexity space is defined in the same way as in Section 3.1, that is, a subset of is convex if, for every path joining two vertices in , each vertex on belongs to . Again one always has . In Section 6 we will give an efficient algorithm for computing convex hulls.
4. Convexity versus Convexity
Since every partial edge is a clique, one always has . In this section we characterize the class of hypergraphs for which . First of all, observe that if is conformal, then every clique is a partial edge so that, by Theorem 3.4, every convex set of is also convex so that . We will see that conformality is not a necessary condition for .
A clique of is a boundary clique if there exists an component of such that equals the boundary of with . A hypergraph is weakly conformal if every boundaryclique of is a partial edge of . Of course, every conformal hypergraph is weakly conformal. The following is an example of a weakly conformal hypergraph that is not conformal.
Example 4.1. Consider the (hyper)graph of Figure 4. The only cliques of that are not boundary cliques are the two cliques with cardinality 3, namely, the sets and . Since each clique of with cardinality less than 3 is a partial edge of , each boundary clique is a partial edge and, hence, is weakly conformal.
Theorem 4.2. Let be a connected hypergraph. The equality holds if and only if is weakly conformal.
Proof. (if) Let be any convex set. Let be any component of , and let be the boundary of with . By the very definition of convexity, is a clique; moreover, is also a component of and the boundary of with is itself. Therefore, is a boundary clique of . Since is weakly conformal, is a partial edge of . It follows that the boundary of with every component of is a partial edge of and, hence, is convex.
(only if) Let be any boundary clique of . Since is a clique, from the very definition of convexity it follows that is convex and, since by hypothesis, is convex. From the very definition of convexity, it follows that the boundary of with every component of is a partial edge. Finally, since is a boundary clique of , there exists an component of such that is the boundary of with . So, is a partial edge of . It follows that is weakly conformal.
5. Convexity versus Convexity and Convexity
In this section we characterize the class of hypergraphs for which and the class of hypergraphs for which .
5.1. Equivalence between Convexity and Convexity
Let be a connected hypergraph. We first prove that . To achieve this, we need the following two technical lemmas.
Lemma 5.1. If is a β path in and and are not adjacent in , then there exists in a simple β path of length at least 2 and with .
Proof. Let be a β path. Let max and are adjacent in and let be an edge of that contains both and . Since and are not adjacent in , one has . Consider the β path . Let and are adjacent in , and let be any edge of that contains both and . If , then the β path is simple since and . If then let and are adjacent in , and let be any edge of that contains both are . If , then the β path is simple since and , and so on.
The next lemma characterizes convex sets.
Lemma 5.2. A subset of is convex if and only if either or, for every two distinct vertices and in , there exists no β path joining and in the partial hypergraph of obtained by deleting the edges that contain both and .
Proof. (only if) Assume that is convex and . Suppose, by contradiction, that there exist two vertices and in and an β path joining and in . By construction of , the vertices and are not adjacent in so that by Lemma 5.1, there exists a simple β path of length at least 2 joining and in . Since is also a simple path in and is not contained in is not convex (contradiction).
(if) If then is trivially convex. Assume that and, for every two distinct vertices and in , there exists no β path joining and in . Suppose, by contradiction, that is not convex. Then, there exist two vertices and in and a simple β path in with . Let be a vertex in , and let be the subpath of such that is on and is an β path. Of course,is a simple path in and has length at least 2. Let and be the vertices (in ) that are joined by . Since is a simple path in , no edge of contains both and , so that is also a path in . To sum up, is an β path that in joins the vertices and in so that is not convex (contradiction).
Example 5.3. Let , where , , ,, , and . The hypergraph is shown in Figure 5. Let . Consider the three vertex pairs in . For the vertex pair , the partial hypergraph of is and, since 1 and 3 belong to distinct connected components of , there exists no β path joining 1 and 3 in . For the vertex pair , the partial hypergraph of is itself and, since every path joining 1 and 4 in passes through 3, there exists no β path joining 1 and 4 in . For the vertex pair , the partial hypergraph of is and, since 4 is not a vertex of , there exists no β path joining 3 and 4 in . By Lemma 5.2, the set is convex, which is confirmed by the fact that the only simple paths joining two vertices in are , , , , and .
We first prove the inclusion .
Theorem 5.4. Let be a connected hypergraph. Every convex set of is convex.
Proof. Suppose, by contradiction, that there exists an convex set of that is not convex. Then, there exists an component of such that the boundary of with (i.e., the set is not a partial edge of . Of course, and the boundary of with is not a partial edge of . Let be an edge of such that for every edge of , either or there exists . Since the boundary of with is not a partial edge of is a proper subset of ; therefore, there exist two vertices and such that and . Let be any edge of that contains . Since is an component of , and, hence, there exists an β path in of length at least 2 (i.e., ) with . If no edge of contains both and , then is not convex by Lemma 5.2 and a contradiction arises. Otherwise, let be the first edge on that contains both and . Since and , is not a subset of so that there exists . Consider the β path . If no edge of contains both and , then is not convex by Lemma 5.2 and a contradiction arises. Otherwise, let be the first edge on that contains both and . Since and is not a subset of so that there exists . Consider the β path . If no edge of contains both and , then is not convex by Lemma 5.2 and a contradiction arises, and so on. Thus, ultimately one obtains an β path of length at least 2 that joins two vertices and in and is such that no edge of contains both and . By Lemma 5.2, is not convex and a contradiction arises.
We now characterize the class of hypergraphs for which .
Theorem 5.5. Let be a connected hypergraph. The equality holds if and only if is acyclic.
Proof. (if) Assume that is acyclic and suppose, by contradiction, that . Let be a convex set that is not convex. Then, there exist two vertices and in and a simple β path (of length at least 2) such that . Let , let be such that , let and , and let and . Consider the subpath of . Since is a simple path, is a simple path of length at least 2, and by construction. Let be the component of such that . Then, both and belong to the boundary of with . Since is convex, the boundary of with is contained in an edge of , say . Since is a simple path of length at least 2, and contains both and , is not an edge of so that is a cycle; moreover, since is a path of length at least 2, has length at least 3. Distinguish two cases depending on whether or not .Case 1 (). Then, ) is a cycle of length 3 and, since only the vertex belongs to the three edges of , is a cycle (contradiction).Case 2 (). Then, there exists in a vertex for some , . Let and and and . Then is a cycle of length at least 3 and, since every vertex in belongs to exactly two edges of , is a cycle (contradiction).
(only if) Assume that every convex set of is also convex and suppose, by contradiction, that is not acyclic. Let , , be a cycle. Distinguish two cases depending on whether or not each vertex in belongs to exactly two edges of .Case 1. Each vertex in belongs to exactly two edges of . Thus, is a cycle. Let . Since , for each component of is a partial edge of and, hence, is convex. On the other hand, since each vertex in belongs to exactly two edges of , is simple path of length at least 2 and, since is not convex (contradiction).Case 2. There exists a vertex in that belongs to more than two edges of . Without loss of generality, let it be . Since is a cycle, each , , belongs to exactly two edges of . Let be any edge of , containing , and let , . Since , for each component of is a partial edge of and, hence, is convex. Let
It is easy to see that is a simple path of length at least 2 and, since is not convex (contradiction).
Let be a connected hypergraph with vertices and edges. If is acyclic, then is acyclic and, hence, convex hulls can be computed in using the AnsteeFarber algorithm. On the other hand, if is acyclic, then is acyclic and, hence, convex hulls can be computed in linear time using the TarjanYannakakis algorithm. By Theorem 5.5, convex hulls can be computed in linear time, that is, more efficiently than using the AnsteeFarber algorithm.
5.2. Equivalence between spConvexity and Convexity
Note that, by Theorem 3.5 and by the fact that every chordless path in is a simple path in , one always has . The following is another convexitytheoretic characterization of acyclic hypergraphs.
Theorem 5.6. A connected hypergraph is Ξ³acyclic if and only if .
Proof. (only if) By hypothesis, is acyclic so that by Theorem 5.5. Moreover, by Proposition 2.4, is conformal and, hence, weakly conformal, so that by Theorem 4.2. To sum up, .
(if) By hypothesis, . Since by Theorem 5.5 and , one has so that is acyclic again by Theorem 5.5.
By Theorem 5.6, convex hulls can be computed in linear time more efficiently than using the monophonicclosure algorithm.
6. Convexity versus Convexity, Convexity, and Convexity
Let be a connected hypergraph. In this section we characterize the three classes of hypergraphs for which , , and . To achieve this, we first give a polynomial algorithm for computing convex hulls.
6.1. Computing Convex Hulls
We represent by its incidence graph .
Remark 6.1. For every two vertices and of , every β path in is a β path in and vice versa; moreover, every cycle in is a cycle in and vice versa.
To avoid ambiguities, we call the vertices and edges of the nodes and arcs of , respectively. A node of is a vertexnode or an edgenode depending on whether or . Moreover, we call cutpoints the cut vertices of ; furthermore, a cutpoint of is a vertexcutpoint or an edgecutpoint depending on whether it is a vertexnode or an edgenode. Note that, if is a vertexcutpoint of , then either the induced subhypergraph is not connected (see the vertexnode 3 in Figure 6) or the singleton is an edge of (see the vertexnode 1 in Figure 6); moreover, if is an edgecutpoint of , then either the partial hypergraph is not connected (see the edgenode in Figure 6) or there exist one or more vertices of that belong to and to no other edge of (see the edgenode in Figure 6). Our algorithm works with the βblockcutpoint treeβ of , which is defined as follows. Let be the bipartite graph whose nodes are the cutpoints and blocks of and where is an arc if the cutpoint of is a node of the block of . A node of is a blocknode if it is a block of and a cutpointnode otherwise. It is well known [14] that is a tree, which is called the block cutvertex tree of . We also label each blocknode of by the vertex set .
Example 6.1. Consider again the hypergraph of Example 5.3 (see Figure 5). The incidence graph of is shown in Figure 6, in which the cutpoints of are circled.
Note that the induced subhypergraph is connected and the induced subhypergraph is not connected; moreover, the partial hypergraph is not connected and the partial hypergraph is connected.
The blocks of are reported in Figure 7.
The blockcutpoint tree of is shown in Figure 8.
The six blocknodes of are labeled as follows:
Note that each leaf of is a blocknode. Moreover, if is not a trivial hypergraph and is a onepoint tree, then the node of is a nontrivial block of . Finally, if is a trivial block of with and , then the blocknode of is not a leaf if and only if and are both cutpoints of .
Algorithm 1 constructs the convex hull of any subset of .

Example 6.1 (continued). When we apply Algorithm 1 with input , the tree resulting from the pruning of is shown in Figure 9. So, the output of Algorithm 1 is .
When we apply Algorithm 1 with input , the tree resulting from the pruning of is shown in Figure 10. So, the output of Algorithm 1 is .
Remark 6.2. Each time a blocknode is deleted during the pruning process, either or and the vertex in belongs to for some undeleted blocknode . Therefore, one has that is a subset of .
Fact 1. Each leaf of is a blocknode and, if is a leaf of , then . and is a onepoint tree, then the blocknode of is a nontrivial block of .
Fact 2. Let be a blocknode of that is a trivial block of with and . , then is not a onepoint tree and, furthermore,(i)if then is a node of adjacent to (see the cutpointnode and the block node in Figure 7) for, otherwise, the blocknode would be a leaf of and