Closure operators and their middle-interchange law (Q645209): Difference between revisions

Let \(\mathcal{M}\) be a left-cancellable family of monomorphisms in a category \(\mathcal{K}\), containing isomorphisms and closed under composition with them. The paper first proposes a new definition of a closure operator on \(\mathcal{M}\) as a (well-) pointed endofunctor \(\Gamma : I \to C : \mathcal{M} \to \mathcal{M}\) on the corresponding full subcategory of \(\mathcal{K}^2\), with cod \(\Gamma = 1_{cod}\). Equivalently as a copointed functor \(\Delta : \widetilde{C} \to I : \mathcal{M} \to \mathcal{M}\) with dom \(\Delta = 1_{dom}\). This encompasses the definitions of Dikranjan-Giuli and of Bourn-Gran, avoiding the conditions of existence of direct or inverse images on the category \(\mathcal{\mathcal{K}}\). The two compositions of closure operators \(C\) and \(D\) on \(\mathcal{M}\) are now defined uniformly as \(D \cdot C = DC\) and \(\widetilde{D \ast C} = \widetilde{D} \widetilde{C}\). \(C\) is idempotent (resp. weakly hereditary) if \(\Gamma C\) (resp. \(\Delta\widetilde{C}\)) is an isomorphism. The author proves the lax middle-interchange law \[ (C \ast D) \cdot (E \ast F) \leq (C \cdot E) \ast (D \cdot F) \] (for closure operators \(C,D,E,F\) on \(\mathcal{M}\)), and that it holds strictly (i.e. the inequality is an equality) for all cases with \(E \leq C \leq D\) iff \(C\) is idempotent and weakly hereditary. A closure operator of \(\mathcal{M}\) in particular provides a factorization of the morphisms in \(\mathcal{M}\); the author intends to present a middle-interchange law for general types of (weak) factorization systems in a sequel of the paper.