Linear Hypergraph Edge Coloring-Generalizations of the EFL Conjecture

Motivated by the Erdos-Faber-Lovász (EFL) conjecture for hypergraphs, we consider the edge coloring of linear hypergraphs. We discuss several conjectures for coloring linear hypergraphs that generalize both EFL and Vizing's theorem for graphs. For example, we conjecture that in a linear hypergraph of rank 3, the edge chromatic number is at most 2 times the maximum degree unless the hypergraph is the Fano plane where the number is 7. We show that for sufficiently large fixed rank and sufficiently large degree, the conjectures are true. Introduction In 1972, a three week conference on hypergraphs was held at The Ohio State University. It was during this conference that the seeds of EFL were planted. In particular, in Problem 18 (see below and [1]) we asked for bounds on edge coloring. Later, we added the condition that the hypergraph be linear which modified the conjectured bounds in Problem 18 and when we couldn’t prove or disprove that we added the additional constraints that created EFL. In this paper, I go back to the precursor of EFL and show that some of the facts we know about EFL from [2] apply equally well to the precursor. Generalizations Preliminaries. Before we can discuss extensions to EFL, we need to give a short list of the concepts involved. Notation. Let ) , ( E V H  be a hypergraph (see, for example [3]): a set of subsets E of the set V . We call the elements of V the vertices and the elements of E the edges. We often write | |V n  and | | E m  . The degree of a vertex x is the number of edges ) (x d which include it. We let the minimum degree be  and the maximum degree be  . If all vertices have the same degree, we say the hypergraph is regular. The rank of an edge e is the cardinality ) (e r of e . We let the minimum rank be  and the maximum rank be  . If all edges have the same rank, we say the hypergraph is uniform. If 2 ) (  e r for every edge then H is a graph. If the intersection of any two edges has at most one vertex, we call the hypergraph linear. Incidence matrix formulation. An equivalent formulation for a hypergraph is to consider H to be the incidence matrix of the hypergaph. In this case, H is an m n matrix: a row of H is the transpose of the characteristic vector of a vertex and a column of H is the characteristic vector of a edge. We use these two formulations interchangeably. It is often easier to understand a fact in one formulation or the other. For example, a fundamental theorem for a hypergraph is that the sum of the ranks is equal to the sum of the degrees. This is trivial to see in the matrix formulation because both sides of the equality are clearly equal to the number of non-zero entries in the matrix H . In this formulation, an edge e is a column vector and a vertex x is a row vector. Two vertices x and y are independent if and only if they are orthogonal, that is, the inner product 0 ) , (      y x xy . Two edges are independent if and only if they are orthogonal, that is 0 ) , (    f e f e . Bulletin of Mathematical Sciences and Applications Submitted: 2016-05-18 ISSN: 2278-9634, Vol. 17, pp 1-9 Revised: 2016-08-24 doi:10.18052/www.scipress.com/BMSA.17.1 Accepted: 2016-08-26 2016 SciPress Ltd, Switzerland Online: 2016-11-01 SciPress applies the CC-BY 4.0 license to works we publish: https://creativecommons.org/licenses/by/4.0/ Duality. If H is a hypergraph, so is the transpose  H called the dual hypergraph. The theorems we need are often stated in a form that applies most naturally to  H and we have to translate them to H to apply them. Clearly the edges of  H are the vertices of H and vice versa, ranks swap with degrees, etc. It is easy to see that H is linear if and only if  H is linear. This is because H is linear if and only if every 2 2 minor has a zero entry. Clique and line graphs. The clique graph ) (H C has the same vertex set as H and an edge for every pair of vertices in some edge. Each edge in H then appears as a clique in ) (H C . The line graph of H is the clique graph of  H ; it has a vertex for each edge of H and an edge between two edges of H if they intersect. Note that in a linear hypergraph, the cliques in the clique graph which come from the edges in the hypergraph are edge disjoint; the clique graph is a set of edge disjoint complete subgraphs of the complete graph on n vertices. Coloring. A coloring of the edges of a hypergraph is a function  from the edges of a hypergraph into a set  , called colors, such that ) ( ) ( f e    only if e and f are orthogonal. We let ) (H q , called the chromatic index of H , be the cardinality of the smallest  for which there is a coloring. The chromatic index of H is the chromatic number of the line graph. If H is colored with q colors, the columns of the matrix H are the disjoint union of q sets of columns, each one of which consists of pairwise orthogonal columns. Conjectures. There is a long standing conjecture known now as the Erdos-Faber-Lovász conjecture which in its edge formulation says Conjecture C1 (EFL). Let H be a linear hypergraph with n vertices and no rank 1 edges. Then n H q  ) ( . In fact, there are stronger conjectures from which EFL stemmed that appear to have been forgotten. Definition. We define the clique degree of a vertex x by ) 1 ) ( ( ) (     x e e r x D . This is the degree of x in the clique graph. The dual concept is the clique rank of an edge ) 1 ) ( ( ) (     e x x d e R . Conjecture C2. Let H be a linear hypergraph with maximum rank  , maximum degree  and no rank 1 edges such that for every vertex k x D  ) ( . Then 1 ) (   k H q . Conjecture C3. Let H be a linear hypergraph with maximum rank  , maximum degree  and no rank 1 edges then 1 ) , max( ) (       H q . Discussion about designs. Most examples that we use come from designs. A ) 1 , , ( r v balanced incomplete block design ( ) 1 , , ( r v BIBD) is a linear hypergraph with v vertices, rank r and every pair of vertices is in exactly one edge. A projective ) 1 , , , (  r n design (see [4]) is a uniform regular hypergraph with n vertices, rank r and degree  and every two edges have a non-empty intersection. A projective design is not required to be linear. Note that the dual of a ) 1 , , ( k v BIBD is projective design with parameters k   , ) 1 /( ) 1 (    k v r , and ) 1 /( ) 1 )( / (    k v k v n . We know that if H is a projective design, then ) (H q is the number of edges but we often do not know what *) (H q is. A projective plane of order r is a projective ) 1 , 1 , 1 ( 2    r r r BIBD; its dual is also a projective design. 2 Volume 17


Introduction
In 1972, a three week conference on hypergraphs was held at The Ohio State University. It was during this conference that the seeds of EFL were planted. In particular, in Problem 18 (see below and [1]) we asked for bounds on edge coloring. Later, we added the condition that the hypergraph be linear which modified the conjectured bounds in Problem 18 and when we couldn't prove or disprove that we added the additional constraints that created EFL. In this paper, I go back to the precursor of EFL and show that some of the facts we know about EFL from [2] apply equally well to the precursor.

Generalizations
Preliminaries. Before we can discuss extensions to EFL, we need to give a short list of the concepts involved.
be a hypergraph (see, for example [3]): a set of subsets E of the set V . We call the elements of V the vertices and the elements of E the edges. We often write | |V n  and | | E m  . The degree of a vertex x is the number of edges ) (x d which include it. We let the minimum degree be  and the maximum degree be  . If all vertices have the same degree, we say the hypergraph is regular. The rank of an edge e is the cardinality ) (e r of e . We let the minimum rank be  and the maximum rank be  . If all edges have the same rank, we say the hypergraph is uniform. If 2 ) (  e r for every edge then H is a graph. If the intersection of any two edges has at most one vertex, we call the hypergraph linear.
Incidence matrix formulation. An equivalent formulation for a hypergraph is to consider H to be the incidence matrix of the hypergaph. In this case, H is an m n  matrix: a row of H is the transpose of the characteristic vector of a vertex and a column of H is the characteristic vector of a edge. We use these two formulations interchangeably. It is often easier to understand a fact in one formulation or the other. For example, a fundamental theorem for a hypergraph is that the sum of the ranks is equal to the sum of the degrees. This is trivial to see in the matrix formulation because both sides of the equality are clearly equal to the number of non-zero entries in the matrix H . In this formulation, an edge e is a column vector and a vertex x is a row vector. Two vertices x and y are independent if and only if they are orthogonal, that is, the inner product Two edges are independent if and only if they are orthogonal, that is 0 Conjectures. There is a long standing conjecture known now as the Erdos-Faber-Lovász conjecture which in its edge formulation says

Conjecture C1 (EFL). Let
H be a linear hypergraph with n vertices and no rank 1 edges. Then n H q  ) ( .
In fact, there are stronger conjectures from which EFL stemmed that appear to have been forgotten.
Definition. We define the clique degree of a vertex x by This is the degree of x in the clique graph. The dual concept is the clique rank of an edge

Conjecture C2. Let
H be a linear hypergraph with maximum rank  , maximum degree  and no rank 1 edges such that for every vertex

Conjecture C3. Let
H be a linear hypergraph with maximum rank  , maximum degree  and no rank 1 edges then

Discussion about designs.
Most examples that we use come from designs.
is a linear hypergraph with v vertices, rank r and every pair of vertices is in exactly one edge. A projective ) 1 , , , (  r n design (see [4]) is a uniform regular hypergraph with n vertices, rank r and degree  and every two edges have a non-empty intersection. A projective design is not required to be linear. Note that the dual of a ) 1 , H is a projective design, then ) (H q is the number of edges but we often do not know what *) (H q is. A projective plane of order r is a projective BIBD; its dual is also a projective design.

BMSA Volume 17
Conjecture C4. Let H be a linear hypergraph with maximum rank 3   , maximum degree  and no rank 1 edges then and H is a projective design in which Conjecture C5. Let H be a linear hypergraph with maximum rank 3   , maximum degree  and no rank 1 edges such that for every vertex Background Facts (Edge Formulations). In this section, we list some well-known theorems that we shall utilize to deduce relationships between the conjectures. In some cases, this requires giving the theorem in a form dual to its usual presentation.

Fact 1 (Brooks' Theorem -edge formulation). For any hypergraph if
unless the clique graph of the dual is either an odd cycle or a complete graph in which case the bound is Note that if the clique graph of the dual is an odd cycle, then the dual is an odd cycle also so 2 If the clique graph is a complete graph, then every pair of vertices in  H is in some edge so every pair of edges in H meets and 1 ) (

Fact 2 (Vizing's Theorem). For any graph
Note that Conjecture C2 is a natural extension of Vizing's theorem. If you think of H as an edge disjoint union of cliques, then the hypothesis of Conjecture C2 is that the sum total of all the edges of all the cliques that meet at a fixed vertex x is at most k . Vizing's theorem colors these edges so that they all have different colors. Conjecture C2 demands that all the edges that belong to the same clique have the same color.
Definition. Three edges ) , , ( . Then there exists a universal constant c such that Some Theorems. In this section we use the facts to prove theorems.
Thus degree ) (e R of a vertex e in the line graph L can be no more than In particular, if we let the line graph has maximum degree at most R . Furthermore, by Eq. 1 and Eq.2, , utilizing inequality Eq. 4 gives Thus we have shown that under the given hypothesis that But from Eq. 2 . If H is uniform,

Bulletin of Mathematical Sciences and Applications Vol. 17 5
and so Theorem 7 (see [6] . Similarly, by counting the number of vertices adjacent to any vertex x of degree ) (x d , Solve this quadratic to get , then any n edge coloring of 3 H can be extended to an n edge coloring of H .
Proof. The degree of a vertex x in the graph 2 H satisfies

Definition.
A critical hypergraph is one that satisfies the hypotheses of Theorem 12.
Hypergraphs that are not necessarily linear. We provide a partial solution to Problem 18 in Hypergraph Seminar.
Problem 18 (Faber-Lovász). Find the best possible function ) , . It is easy to show Discussion. There may be an omission in part (iii). It certainly does not seem easy in general. The construction that I recall only works for r a power of a prime plus 2. I summarize what I know about this problem in the following theorem. is covered in part (i) of the theorem so r r n   2 is the next smallest value.

Summary
We have discussed a number of ways to generalize the EFL conjecture to arbitrary linear hypergraphs and exhibited various relationships between these new conjectures. In particular, for large enough fixed rank we can generalize EFL to obtain a generalization of Vizing's theorem that we can show is true except for possibly a finite number of hypergraphs.