Why the "curious" omission of this topos from most discussions of combinatorial topology ? The introduction of ordered simplices by Eilenberg 60 years ago is usually explained in subjective terms like the annoying extra degeneracies that had to be eliminated in the previous theory . However there is a objective requirement clearly pointed out by Gabriel & Zisman. (They spoke of Kelley spaces but that was a mistake due to Kelley's excellent exposition of the k-spaces of Hurewicz). The requirement is that geometric realization r (left adjoint to "singular" s) be left exact, so that products and equations on fleshed out spaces can be the reflection simply of the same operations on the combinatorial models. If we construe spaces as Johnstone did (or in many other possible ways) as forming themselves a topos, the the above requirement is simply that r/s constitute a geometric morphism of toposes. To understand the qualitative distinction between the possible codomain combinatorial toposes, it is very helpful to note their role as CLASSIFYING toposes. That role is not always easy to grasp in the specific if one starts with primitives and axioms for a first order theory, tries to present the corresponding Lindenbaum category, then takes sheaves on that, etc. Fortunately in some cases one can bypass that presentation process because the resulting small category is already well known. Preheaves on the category of non-empty finite posets clearly classify arbitrary non-trivial distributive lattices in any Grothendieck topos. In other words an s/r teory can be based on an "interval" object in spaces that has a DL structure. If we want that to factor through the subtopos of simplicial sets, we note that the latter is the classifier for those special DLs that are totally ordered, which translates geometrically to a condition on an interval that the square is a union of two triangles; since "union" depends on the topology of space, that condition is often not true for DLs that at first glance look like intervals. There are other relevant subtoposes (ie positive classes of DLs) for example those for which the canonical map from the generic one I to the truth-object is actually a homomorphism with respect to both lattice operations. But the presheaves on non-empty finite sets is the subtopos that classifies Boolean algebras. The generic BA is the obvious one. It does not really describe well its relation to the total orderings to call it the "synmmetric version". As the one contains all small categories, this one analogously contains all groupoids. It can still receive an r/s pair as required if only a space with a BA structure is used as the "interval". The natural choice for that is the in finite dimensional sphere, which indeed has a continuous BA structure that contains the usual interval as a subDL. If only this had been known 60 years ago, we could have done without the simplicial sets, for this singular theory reads to the same homotopy category. Note that any Grothendieck topos (including ss!) has a canonical BA object, hence enjoys a canonical r/s theory valued here. Quoting Marco Grandis <grandis@dima.unige.it>:
Vaughan Pratt asked about:
undirected graphs ... as presheaves on the full subcategory 1 and 2 of Set?
It is the 2-truncation of "symmetric simplicial sets" as presheaves on finite cardinals, cf (*).
Curiously, symmetric simplicial sets have been rarely considered. Even if simplicial complexes (well-known!) are a symmetric notion and have a natural embedding in symmetric simplicial sets. While simplicial sets are a directed notion, used as an undirected one in classical Algebraic Topology.
(*) M. Grandis, Finite sets and symmetric simplicial sets, Theory Appl. Categ. 8 (2001), No. 8, 244-252.
Marco Grandis