On the algebra of regular sets (Q1926594)

Stone proved that every Boolean algebra has an isomorphic representation as the open sets of a topological space. Similarly, the regular open sets (and dually the regular closed sets) of a topological space also form Boolean algebras. It is only natural to ask what algebraic structures represent the regular sets of a topological space. This is still an open question. It is motivated by the theory of qualitative space, whose intended models contain only regular sets of a topological space. Generally speaking, Qualitative Spatial Reasoning (QSR) aims to capture relations in space qualitatively, e.g., without applying a specific metric. Topological and mereological relations are fundamental amongst such relations and are often captured in logical theories referred to as mereotopologies. For many practical representation and reasoning problems it is appropriate to assume that all regions are of the same dimension, we refer to such theories as equidimensional mereotopologies. Therein all regions must be regular, i.e., isolated points or other lower-dimensional artifacts cannot occur in order to ensure closure under sums and intersections amongst regions. This commonsensical notion of regularity corresponds to the definition of regularity known from topology. Several alternative axiomatizations of mereotopology have been proposed. One way to classify these axiomatizations of equidimensional mereotopology is by their intended topological models. Either it is assumed that all regions are regular closed (or dually regular open) or that all regions are regular, but not necessarily open or closed. The approach assuming that all regions are regular closed makes the ontological commitment that two regions with identical closures must be identical. Such an assumption is made for example by the most prominent mereotopology, the Region-Connection Calculus (RCC). Topological interpretations thereof include only regular closed sets and consequently each model forms a Boolean algebra. If we superimpose a contact relation on such a Boolean algebra we obtain so-called Boolean contact algebras. The alternative approach, first axiomatized as a first-order theory by Asher and Vieu, which we refer to as \(RT_0\), can distinguish regions with identical closures. For example, a region that is neither open nor closed is distinct from both its interior and its closure. This approach is maybe less popular, but the arising models are both algebraically and topologically interesting. Most importantly, the theory \(RT_0\) captures the same set of mereotopological relations as the RCC. Though regions with identical closures can be distinct, interiors and closures of all regions are still required to exist. Moreover, all regions must again be regular which leads to definitions of regular union and intersection that differ from standard set-theoretic union and intersection. Hence not every set of subsets of a topological space can be extended to a model of \(RT_0\). For studying the mereotopology \(RT_0\), it is convenient to study a weaker theory thereof which is referred to as \(RT^{-}\). The models of \(RT^{-}\) are structurally equivalent to the models of \(RT_0\). However, `trivial' models of the \(RT^{-}\), in which the extensions of external contact (EC) or weak contact (WCont) are empty, are not models of \(RT_0\). Recall that two regions are externally connected if they share a point, i.e., are in contact, but do not share an equidimensional part, i.e., do not overlap. Two regions are in weak contact if their closure are not in contact but any open region containing one of them is in contact to the other. Stonian \(p\)-ortholattices were introduced to characterize the models of \(RT^{-}\) algebraically. Moreover, it has been shown that the models of \(RT_0\) are several special kinds of Stonian \(p\)-ortholattices. Subsequent work demonstrated that the skeleton, i.e., the set of all closed sets, of a Stonian \(p\)-ortholattice is a model of RCC and every model of RCC can be extended to a Stonian \(p\)-ortholattice while preserving the contact relation. This verified the close relationship between the RCC and \(RT^{-}\). However, one important question has been left open: the topological, i.e., the point-set, representability of Stonian \(p\)-ortholattices. If every model of the theory \(RT_0\) would indeed capture the regular sets of some topological space, we could represent all Stonian \(p\)-ortholattices topologically by the regular sets of some topological space. An algebraic representation of the regular closed sets of a topological space would immediately follow. In this paper, the authors have proven that this is not the case: not all Stonian \(p\)-ortholattices can be represented as regular point sets of a topological space. Therefore, the completeness theorem for \(RT_0\) with respect to the intended models of regular subsets of a topological space is incorrect. In this work, significant progress is made towards a full topological representation of Stonian \(p\)-ortholattices in two ways. Examples of Stonian \(p\)-ortholattices that are not (topologically) representable are given and five necessary properties for Stonian \(p\)-ortholattices to be representable are identified. These properties, amongst them a conditional form of distributivity, are topologically motivated and can be used to eliminate unintended models of the theories \(RT^{-}\) and \(RT_0\).