Formal Concepts and Residuation on Multilattices

TL;DR

This paper explores the structure of multilattices with residuation, introduces minimal examples, and applies these concepts to formal concept analysis, generalizing existing lattice results.

Contribution

It demonstrates the smallest pure multilattice, constructs new residuated multilattices, and extends formal concept analysis to multilattice-based truth-values.

Findings

The smallest pure multilattice is identified and characterized.

Residuated multilattices with fewer than seven elements do not exist.

The set of formal concepts forms a complete residuated multilattice.

Abstract

Multilattices are generalisations of lattices introduced by Mihail Benado. He replaced the existence of unique lower (resp. upper) bound by the existence of maximal lower (resp. minimal upper) bound(s). A multilattice will be called pure if it is not a lattice. Multilattices could be endowed with a residuation, and therefore used as set of truth-values to evaluate elements in fuzzy setting. In this paper we exhibit the smallest pure multilattice and show that it is a sub-multilattice of any pure multilattice. We also prove that any bounded residuated multilattice that is not a residuated lattice has at least seven elements. We apply the ordinal sum construction to get more examples of residuated multilattices that are not residuated lattices. We then use these residuated multilattices to evaluate objects and attributes in formal concept analysis setting, and describe the structure of the set of corresponding formal concepts. More precisely, if $\mathcal{A}_i: =(A_i,\le_i,\top_i,\odot_i,\to_i,\bot_i)$, $i=1,2$ are two complete residuated multilattices, $G$ and $M$ two nonempty sets and $(\varphi, \psi)$ a Galois connection between $A_1^G$ and $A_2^M$ that is compatible with the residuation, then we show that \[\mathcal{C}: =\{(h,f)\in A_1^G\times A_2^M; \varphi(h)=f \text{ and } \psi(f)=h \}\] can be endowed with a complete residuated multilattice structure. This is a generalization of a result by Ruiz-Calvi{\~n}o and Medina saying that if the (reduct of the) algebras $\mathcal{A}_i$, $i=1,2$ are complete multilattices, then $\mathcal{C}$ is a complete multilattice.