On pseudomodular matroids and adjoints (Q1894347)

Let \(L\) be a geometric lattice with rank function \(r\). A geometric lattice \(L^\Delta\) with \(r(L^\Delta)= r(L)\) is said to be an adjoint of \(L\) if there is a map \(\phi: L\to L^\Delta\) taking coatoms to points such that if \(a\) covers \(b\) in \(L\), then \(\phi(b)\) covers \(\phi(a)\) in \(L^\Delta\), and such that \(\phi(a\vee b)= \phi(a)\wedge \phi(b)\). The construction, where possible, of an adjoint may involve the addition to \(\phi(L)\) of certain other elements, to provide closure under intersection. A necessary local condition for the possibility of such point extensions to a matroid is the generalized Euclidean intersection property, which requires that for any non-modular pair in the matroid, there be a proper point extension in which the closures of the pairs intersect. The weaker Euclidean intersection property makes the same assertion only for pairs consisting of a non-intersecting hyperplane and line. Finally, a matroid is pseudomodular if a certain rather complicated identity involving the rank function of its geometric lattice implies another one; this condition was introduced by Björner and Lovász in 1987. In this paper, the authors show that pseudomodular matroids are Euclidean, and that any pseudomodular matroid of rank 4 has an adjoint. To show that these results are tight, they present an example of a pseudomodular matroid of rank 5 which does not have the generalized Euclidean intersection property, and hence has no adjoint. They also give an example of a nonlinear matroid of rank 4 which is pseudomodular (and thus has an adjoint), disproving a conjecture of Mnëv.