Bertrand Guenin - the set covering problem1. Perfect and Ideal cluttersA family of sets over a finite groundset where no set is properly contained in another is a clutter. Given a clutter we can define the set packing polytope,
The clutter is said to be perfect if the associated set packing polytope is integral. Padberg, Chvátal, and Lovász proved that a clutter is perfect if and only if its members are the maximal cliques in a perfect graph. Thus the study of perfect clutters reduces to the study of perfect graphs. The Strong Perfect Graph theorem of Chudnowsky, Robertson, Seymour and Thomas, then easily leads to an excluded minor characterization for perfect clutters. Given a clutter we can define the set covering polyhedron,
The clutter is said to be ideal if the associated set covering polyhedron is integral. Ideal clutters, in contrast to perfect clutters, are not associated with a single class of combinatorial objects. Examples of ideal clutters,
Not all examples of ideal clutters arise from graphs, there are some purely geometric constructions as well. Thus ideal clutter form a very rich and complex class of objects. Problem
Characterize ideal clutters. A solution to this problem seems to be out of range at this juncture, so we will consider special cases. Here we shall focus on excluded minor characterizations. Let us formalize this. Let be a clutter and let be the polyhedron obtained from by setting variables to zero and projecting variables. Then is of the form for some clutter We then say that is a minor of As projection and faces of integral polyhedra are integral, idealness is a minor closed property. A clutter is said to be minimally non-ideal (mni) if it is not ideal, but every proper minor is. Lehman proved that minimally non-ideal clutters have a unique fractional extreme point, moreover, that fractional point satisfies some very strong regularity properties. However, this is only a partial characterization, and (in contrast to minimally non perfect clutters) there are complicated infinite classes, and lots of sporadic examples of mni clutters. An accessible proof of Lehman's proof is available in the next reference. Results about general mni clutters can be found in the second reference.
To make further headway we need to restrict ourselves to special classes of clutters. Binary cluttersA signed matroid is a pair where is a binary matroid and A cycle of is even (resp. odd) if is even (resp. odd). Clearly, the set of inclusion-wise minimal odd cycles of is a clutter. Clutters that can be obtained by this construction are said to be binary. If characterizing all mni clutters is difficult, what about binary mni clutters? At least we have the following beautiful 1977 conjecture by Seymour, The Flowing Conjecture
The only minimally non-ideal binary clutters are the clutters of
For (1) the groundset of the clutter are the elements of the Fano matroids, and the members the lines. We illustrate the corresponding clutter in the following picture,
For (2) and (3) the groundset are the edges of the complete graph on vertices. Note for (2) the members are the triangles and the pentagons and for (3) the members are triangles together with an independent edge and complete graphs on four vertices. The conjecture remains open but there are lots of known special cases as we will discuss next. Known cases of the Flowing ConjectureConsider a binary clutter and it associated signed matroid , recall that is a binary matroid. When is co-graphic the members of are precisely the -cuts of some graft As none of outcomes (1)-(3) of the Flowing Conjecture can be expressed as -cuts, the conjecture implies that clutters of -cuts are ideals. This is indeed the case and was proved by Edmonds and Johnson. When is graphic the members of are precisely the odd circuits of some signed-graph It is easy to verify that (1) and (3) are not clutters of odd circuits of any signed-graph. Thus the Flowing Conjecture predicts that (2) is the only obstruction to idealness in that case. Indeed this is the case, Weakly Bipartite Theorem
Clutters of odd circuits of signed-graphs are ideal if and only if it they do not contain as a minor the clutter of odd circuits of
An immediate consequence of this result are examples (1) and (2). This has implications for multi-commodity flow problems. For a survey on flows in graphs and matroids see,
Combining the Weakly Bipartite Theorem, the Theorem of Edmonds and Johnson on -cuts and Seymour's decomposition of regular matroids, we can obtain sufficient conditions, in term of excluded minors, for a binary clutter to be ideal.
An extension to the Weakly Bipartite Theorem and an extension to the Edmonds and Johnson -cuts Theorem is given in
Furthermore, we develop an algorithmic version of Lehman's theorem with applications, including the following, Theorem
For any graph with non-negative weight edges, we can in polynomial time do one of the following, (a) find a maximum weight cut, or (b) find a set of edges and a cut disjoint from such that the graph obtained from by deleting and contracting all edges of is
The Weak Flowing ConjectureConsider a clutter , the blocker of is the clutter with the same groundset as where the members are the inclusion-wise minimal sets that intersect every member of The blocking operation is an involution and Lehman proved that idealness is preserved under taking the blocker. This implies in particular, that if a clutter is mni then so is its blocker. Observe that the blocker of the Lines of the Fano is itself, and that the blocker of the complements of cuts of is the clutter of odd circuits of As the blocker of a binary clutter is also binary, the Flowing Conjecture can be restated as follows, The Flowing Conjecture - alternate statement
Let be a minimally non-ideal binary clutter. Then or is the clutter of the Lines of the Fano or the clutter of the odd circuits of Observe that both the clutter of the Lines of the Fano and the clutter of odd circuits of have a member of cardinality three. Hence, a weakening of the Flowing Conjecture is the following, The Weak Flowing Conjecture
Let be a minimally non-ideal binary clutter. Then or has a member of cardinality three. We proved that in fact the Weak Flowing Conjecture is equivalent to the Flowing Conjecture.
2. The Cycling ConjectureA family of sets over a finite groundset where no set is properly contained in another is a clutter. A cover of a clutter is a subset of the groundset that intersects every member of the clutter. The maximum number of pairwise disjoint sets in a clutter is the packing number and the cardinality of the smallest cover is the covering number. Clearly, the packing number is at most the covering number and a clutter packs if both numbers coincide. Not every clutter packs, consider for instance,
then the packing number is but the covering number is . Examples of clutters that pack,
We will describe (4) in more details later. Examples (3) and (4) both involves a parity condition. For (3) all circuits have even length and for (4) all edge cuts have an even number of edges. The Cycling conjecture will provide sufficient conditions for a clutter, satisfying a particular parity condition, to pack. The conjecture,Consider a clutter with an element of the groundset. The deletion minor is defined as , and the contraction minor is defined as the inclusion-wise minimal sets in . A minor of is a clutter obtained by a sequence of contractions and deletions. A signed matroid is a pair where is a binary matroid and . A cycle of is even (resp. odd) if is even (resp. odd). Clearly, the set of inclusion-wise minimal odd cycles of is a clutter. Clutters that can be obtained by this construction are said to be binary. A clutter is Eulerian if the parity of the cardinality of all inclusion-wise minimal covers is the same. At first glance this condition may appear non intuitive, so let us analyze what it says for a clutter of -cuts in a graph with terminals . The inclusion-wise minimal covers of are -joins, and the parity of all -joins is the same exactly when the symmetric difference of any two -joins is a set with even cardinality. As the difference of two -joins is an Eulerian subgraph, the condition translates to the requirement that be bipartite. We have the following conjecture that is essentially due to Seymour, The Cycling Conjecture
An Eulerian binary clutter packs if it does not contain any of the following clutters as a minor,
For (1) the groundset of the clutter are the elements of the Fano matroids, and the members the lines. For (2) and (3) the groundset are the edges of the complete graph on vertices. Note for (2) the members are the triangles and the pentagons and for (3) the members are triangles together with an independent edge and complete graphs on four vertices. For (4) the groundset are the edges of the Petersen graph and the member subgraphs where every vertex has odd degree. This conjecture is hard as we will see that it implies the Four Colour Theorem. A special case of the conjecture,A signed-graph is a pair where is a graph and is a subset of the edges. We say that is even (resp. odd) if is even (resp. odd). We call a subset of the edges of where an odd -walk if it is either an odd -path, or it is the union of an even -path and an odd circuit where and share at most one vertex. We consider the empty set to be an (even) -path for the case where . Thus when , odd -walks are odd circuits. We proved that the Cycling conjecture holds for the clutter of odd -walks, more precisely Theorem 1
An Eulerian clutter of odd -walks packs if it does not contain any of the following clutters as a minor,
The proof and applications can be found here.
Suppose is a signed-graph with vertices and let be the clutter of odd -walks of . Then is Eulerian if either and the degree of every vertex is even, or and and the degree of every vertex in are even. The theorem states that in that case either: the maximum number of pairwise (edge) disjoint odd -walks is equal to the minimum number of edges needed to intersect all odd -walk, or must have as a minor the clutter of the lines of the Fano or the clutter of odd circuits of . When , the condition that is Eulerian says that all vertices of have even degree. Moreover, in that case odd -walks are odd circuits. Hence, Theorem 1 becomes, Theorem 2
Let be a graph that does not contain as an odd minor and where every vertex has even degree. Then the minimum number of edges needed to intersect all odd circuits is equal to the maximum number of pairwise disjoint odd circuits.
A blocking vertex (resp. blocking pair) is a vertex (resp. pair of vertices) that intersects every odd circuit. Next we indicate a number of classes of signed-graphs for which the clutter of odd -walks does not have as a minor the lines of the Fano or the clutter of odd circuits of .
Thus for each of the classes of signed-graphs (1)-(6), as long as the Eulerian condition holds, the minimum number of edges needed to intersect all odd -walk is equal to the maximum number of disjoint odd -walks. Applying Theorem 1 to class (1) we can obtain, Corollary
Let be a graft with . Suppose that every vertex of not in has even degree and that all the vertices in have degrees of the same parity. Then the maximum number of pairwise disjoint -joins is equal to the minimum size of a -cut. In fact this result holds as long as . Applying Theorem 1 to class (2) we can derive the two-commodity flow of Hu, Rothschild and Whinston, namely, Corollary
Let be a graph with vertices where , , all of have the same parity,and all the other vertices have even degree. Then the maximum number of pairwise disjoint paths that are between and for some , is equal to the minimum size of an edge subset whose deletion removes all - and -paths. Consider obtained as follows: () start from a plane graph with exactly two faces of odd length and distinct vertices and and identify and . Applying Theorem 1 to class (3) we can obtain, Corollary
Let be a graph as in () and suppose that the length of the shortest odd circuit is . Then there exists cuts such that every edge is in at least of Suppose that is as in () and is loopless. Then by Corollary 6, there exists cuts such that every edge is in . It follows that for all where , is a stable set. Hence, is -colourable. The following conjecture would generalize the previous result as well as the 4-colour theorem, Conjecture
Let be a graph that does not contain as an odd minor and suppose that the length of the shortest odd circuit is . Then there exists cuts such that every edge is in at least of . It can be shown that this conjecture is another special case of the Cycling Conjecture. It is open for planar graphs.
|