nLab pullback-stable colimit



(,1)(\infty,1)-Category theory

(,1)(\infty,1)-Topos Theory

(∞,1)-topos theory

structures in a cohesive (∞,1)-topos



Let CC be a category with pullbacks and colimits of some shape DD. We say that colimits of shape DD are stable by base change, or stable under pullback, or that these colimits are universal if for every functor G:DCG : D \to C and for all pullback diagrams of the form

(colim DG)× ZY colim DG Y f Z \array{ (colim_D G) \times_Z Y &\to& colim_D G \\ \downarrow && \downarrow \\ Y &\underset{f}{\to} & Z }

the canonical morphism

(1)colimdD(G(d)× ZY)(colimdDG(d))× ZY \underset{d \in D}{colim} (G(d) \times_Z Y) \stackrel{\simeq}{\to} (\underset{d \in D}{colim} G(d)) \times_Z Y

is an isomorphism. This is equivalent to saying that every pullback functor f *:C/ZC/Yf^*: C/Z \to C/Y preserves DD-colimits. Similar definitions can be given for higher categories.

To see that this definition is equivalent to the universality condition given at van Kampen colimit, observe that, given a natural transformation α:FG\alpha' : F' \Rightarrow G', the diagram

F(*) α * colim DG id id F(*) α * colim DG \array{ F'(*) &\overset{\alpha'_*}{\to}& colim_D G \\ ^\id\downarrow && \downarrow^\id \\ F'(*) &\underset{\alpha'_*}{\to} & colim_D G }

is a degenerate pullback square, hence there is a canonical isomorphism

colim dD(G(d)× colim DGF(*))F(*). colim_{d \in D}(G(d) \times_{colim_D G} F'(*)) \simeq F'(*).

But if α\alpha' is equifibered, we have G(d)× colim DGF(*)F(d)G(d) \times_{colim_D G} F'(*) \simeq F(d), hence we get the desired isomorphism F(*)colim DFF'(*) \simeq colim_D F.

Conversely, given a pullback diagram as above, let F=f *GF' = f^* \circ G' (viewing GG' as a functor DC/ZD \to C/Z and remembering that colimits in C/ZC/Z are computed as colimits in CC) and α:FG\alpha' : F' \Rightarrow G' the natural transformation induced by the pullback projections, which is equifibered as a consequence of the pasting law for pullbacks. Then f *Gf^* \circ G' is a colimiting cocone, which is to say that f *f^* preserves colim DGcolim_D G.



The stability of all colimits is one of Giraud's axioms that characterize Grothendieck toposes in the 1-categorical context and Grothendieck-Rezk-Lurie (∞,1)-toposes in the higher categorical context. The fact that colimits are stable in toposes can be seen from the characterization of toposes as left-exact reflective subcategories of presheaf categories as follows:

  • First observe that colimits are stable in C=C = Set.
  • Now colimits are stable for C=C = a presheaf category [S op,Set][S^{op},Set], since colimits in such CC are computed objectwise in SetSet. (See limits and colimits by example.)
  • Finally, stability of colimits is preserved in reflective subcategories, since the reflector preserves both colimits and pullbacks.

For (∞,1)-toposes, this is HTT, theorem (3) ii)


More generally, colimits are stable in any locally cartesian closed category, since in that case the pullback functors f *f^* all have right adjoints. Conversely, if CC is cocomplete with all stable colimits, and the adjoint functor theorem applies to all its slice categories, then it is locally cartesian closed.

Colimits are also stable in any regular infinitary extensive category.

But colimits are not stable in, for instance, C=C = Ab.

Relation to commutativity and distributivity

Although stability of colimits appears as a sort of “commutativity” between colimits and pullbacks, it is not literally an instance of commutativity of limits and colimits. It is an example of the latter if the colimit over DD of the diagram constant on a single object (such as YY) is that single object. For ordinary colimits in category theory this is a mild condition, requiring DD to be a connected category; but in higher category theory this becomes an ever stronger condition; for colimits in an (infinity,1)-category it means that the infinity-groupoid generated by DD is contractible homotopy type (see this corollary).

It is generally true that (1) is an example of distributivity of limits over colimits; see there.


Last revised on April 16, 2023 at 17:28:07. See the history of this page for a list of all contributions to it.