In general, let and be two families of objects of some category . We say that is a refinement of if there are a function of indices and a morphism for each .
Very often we do this in the slice category for some object . If you spell this out, then you have families and of morphisms to ; is a refinement of if there are a function and a commutative diagram
for each .
More specifically, apply this to the poset of subobjects of . Then you have families and of subobjects of ; is a refinement of if there are a function and a commutative diagram (1) for each .
Yet more specifically, apply this to the lattice of subsets of some set . Then you have families and of subsets of ; is a refinement of if there is a function such that each is contained in .
Yet more specifically, let the families of subsets be indexed by themselves. Then have collections and of subsets of ; is a refinement of if for each there is an such that is contained in .
Actually, this definition is slightly weaker than the previous one in the absence of the axiom of choice. Perhaps in that case the general definition should say that for each there is an and a morphism .
Special cases of this last example include refinement of filters and refinement of open covers.
On the other hand, you might want to generalise the case of open covers to covers or covering sieves on a site. In that case, the general definition still applies; you have covering families and of some object ; is a refinement of if there are a map and a commutative diagram (1) for each .