nLab atomic site



Topos Theory

topos theory



Internal Logic

Topos morphisms

Extra stuff, structure, properties

Cohomology and homotopy

In higher category theory




Atomic sites are sites (𝒞,J at)(\mathcal{C}, J_{at}) equipped with the atomic topology J atJ_{at}. The corresponding sheaf toposes Sh(𝒞,J at)Sh(\mathcal{C}, J_{at}) are precisely the atomic Grothendieck toposes.



A site (𝒞,J at)(\mathcal{C}, J_{at}) is called atomic if the covering sieves SS of J atJ_{at} are exactly the inhabited sieves SS\neq\emptyset. A Grothendieck topology J atJ_{at} of this form is called atomic.


Let FinSet mono opFinSet^{op}_{mono} be the opposite of the category FinSet monoFinSet_{mono} with objects finite sets and monomorphisms. Then (FinSet mono op,J at)(FinSet^{op}_{mono}, J_{at}) is an atomic site and the corresponding sheaf topos Sh(FinSet mono op,J at)Sh(FinSet^{op}_{mono}, J_{at}) is the Schanuel topos. That J atJ_{at} is indeed a Grothendieck topology is ensured by prop. .



Let 𝒞\mathcal{C} be a category. Then 𝒞\mathcal{C} can be made into an atomic site if and only if for any diagram

A B C \array{ & & A \\ & & \downarrow\\ B & \to & C }

there is an object DD and arrows DA,BD \to A, B such that the following diagram commutes:

D A B C \array{ D & \to & A \\ \downarrow & & \downarrow\\ B & \to & C }

This is exactly what is needed for the pullback stability axiom to hold, and the other axioms are immediate.

The condition occurring in the proposition is called the (right) Ore condition. It is a result by P. T. Johnstone (1979) that Set 𝒞 opSet^{\mathcal{C}^{op}} is a De Morgan topos precisely if 𝒞\mathcal{C} satisfies the Ore condition. Whence we see that every atomic Grothendieck toposes is a (Boolean) subtopos of a De Morgan presheaf topos.

Recall that the dense topology J dJ_d on a category 𝒞\mathcal{C} consists of all sieves SJ d(C)S\in J_d(C) with the property that given f:DCf:D\to C there exists g:EDg:E\to D such that fgSf\cdot g\in S. The atomic topology is a special case of this:


Let 𝒞\mathcal{C} be a category satisfying the Ore condition. Then the atomic topology J atJ_{at} coincides with the dense topology J dJ_d.


For 𝒞=\mathcal{C}=\emptyset the claim is trivial. So let C𝒞C\in\mathcal{C} be an object and SS a sieve on CC.

Assume SJ d(C)S\in J_d(C), then for id:CCid\colon C\to C there exists g:ECg:E\to C with idgSid\cdot g\in S whence SJ at(C)S\in J_{at}(C).

Conversely, assume SJ at(C)S\in J_{at}(C) and let f:DCf:D\to C be a morphism. Then there exists gSg\in S by assumption and the diagram DfCgED\overset{f}{\rightarrow} C \overset{g}{\leftarrow} E can be completed to a commutative square fi=ghf\cdot i = g\cdot h but ghSg\cdot h\in S since gSg\in S and SS is a sieve. Whence fiSf \cdot i\in S and, accordingly, SJ d(C)S\in J_d(C).

In other words, the atomic topology is just the dense topology on categories satisfying the Ore condition. Since the corresponding sheaf toposes of the dense topology are just the double negation subtoposes of the corresponding presheaf topos we finally get:


Atomic Grothendieck toposes i.e. toposes (equivalent to) Sh(𝒞,J at)Sh(\mathcal{C}, J_{at}) for (𝒞,J at)(\mathcal{C}, J_{at}) an atomic site are precisely (the toposes equivalent to) the double negation subtoposes Sh ¬¬(Set 𝒞 op)Sh_{\neg\neg}(Set^{\mathcal{C}^{op}}) for a De Morgan presheaf topos Set 𝒞 opSet^{\mathcal{C}^{op}}. \qed

The sheaves of atomic sheaf toposes Sh(𝒞,J at)Sh(\mathcal{C}, J_{at}) are easy to describe:


Let (𝒞,J at)(\mathcal{C}, J_{at}) be an atomic site. A presheaf PSet 𝒞 opP\in Set^{\mathcal{C}^{op}} is a sheaf for J atJ_{at} iff for any morphism f:DCf:D\to C and any yP(D)y\in P(D) , if P(g)(y)=P(h)(y)P(g)(y)=P(h)(y) for all diagrams

EhgDfC E\overset{g}{\underset{h}{\rightrightarrows}} D\overset{f}{\to} C

with fg=fhf\cdot g=f\cdot h , then y=P(f)(x)y=P(f)(x) for a unique xP(C)x\in P(C).

For the proof see Mac Lane-Moerdijk (1994, pp.126f).

The more general definition

Thus far we have presented the classical approach as presented in Mac Lane-Moerdijk (1994) going back to Barr-Diaconescu (1980) but it was observed by O. Caramello (2012) that the atomic topology can in fact be defined on arbitrary categories not only on those satisfying the Ore condition.


Let 𝒞\mathcal{C} be a category. The atomic topology J atJ_{at} on 𝒞\mathcal{C} is the smallest Grothendieck topology containing all the nonempty sieves. A site of the form (𝒞,J at)(\mathcal{C}, J_{at}) is called atomic.

Note that J atJ_{at} is well defined as the intersection of all Grothendieck topologies with the property that all nonempty sieves cover. The following proposition justifies the terminology:


Let (𝒞,J at)(\mathcal{C}, J_{at}) be an atomic site. Then Sh(𝒞,J at)Sh(\mathcal{C}, J_{at}) is an atomic Grothendieck topos.


The main idea is to consider the full subcategory 𝒞\mathcal{C}' on those objects UU with J at(U)\emptyset\notin J_{at}(U) together with the induced topology J at=J at| 𝒞J'_{at}=J_{at}|_{\mathcal{C}'}. Then one shows that 𝒞\mathcal{C}' satisfies the Ore condition and concludes by the comparison lemma that Sh(𝒞,J at)Sh(𝒞,J at)Sh(\mathcal{C}', J'_{at})\simeq Sh(\mathcal{C}, J_{at}). For the details see Caramello (2012, prop.1.4).


Consider the category 𝒞\mathcal{C} on the ‘walking co-span’ AfCgBA\overset{f}{\rightarrow} C\overset{g}{\leftarrow} B. 𝒞\mathcal{C} does not satisfy the Ore condition. The atomic topology J atJ_{at} is given by

J at(A)={{id A},}J at(B)={{id B},} J_{at}(A)=\{\{id_A\},\emptyset\} \qquad J_{at}(B)=\{\{id_B\},\emptyset\}
J at(C)={{id C,f,g},{f},{g},{f,g},}. J_{at}(C)=\{\{ id_C, f,g\},\{f \}, \{ g \},\{f,g\},\emptyset\} \quad .

Here J at(A)\emptyset\in J_{at}(A) , respectively J at(B)\emptyset\in J_{at}(B), due to the stability axiom applied to f *({g})=f^\ast(\{g\})=\emptyset , respectively to g *({f})=g^\ast(\{f\})=\emptyset . Whereas J at(C)\emptyset\in J_{at}({C}) by the transitivity axiom applied to {f}J at(C)\{f\}\in J_{at}(C) and the sieve \emptyset since f *()=J at(A)f^{\ast}(\emptyset)=\emptyset\in J_{at}(A).

Accordingly the subcategory 𝒞\mathcal{C}' is empty and Sh(𝒞,J at)1Sh(\mathcal{C},J_{at})\simeq 1 is degenerate. In particular, Sh(𝒞,J at)Sh(\mathcal{C},J_{at}) is not equivalent to Sh(𝒞,J d)Sh ¬¬(Set 𝒞 op)Set×SetSh(\mathcal{C},J_d)\simeq Sh_{\neg\neg}(Set^{\mathcal{C}^{op}})\simeq Set\times Set. So we see that the atomic topology on 𝒞\mathcal{C} is distinct from the dense topology. For completeness we describe the latter:

J d(A)={{id A}}J d(B)={{id B}} J_{d}(A)=\{\{id_A\}\} \qquad J_{d}(B)=\{\{id_B\}\}
J d(C)={{id C,f,g},{f,g}}. J_{d}(C)=\{\{id_C, f,g\},\{f,g\}\} \quad .

Further details on Set 𝒞 opSet^{\mathcal{C}^{op}}, the topos of hypergraphs, may be found at hypergraph.

In the example, we observed that dense and the atomic topology need not coincide for categories not satisfying the Ore condition. In fact more can be said here:


Let 𝒞\mathcal{C} be a category. Then J dJ atJ_{d}\subseteq J_{at} in general, but J d=J atJ_d=J_{at} precisely if 𝒞\mathcal{C} satisfies the Ore condition.


The proof of prop. already showed that the sieves of the dense topology J dJ_d are never empty regardless of the Ore condition. From prop. follows that the atomic topology J atJ_{at} will additionally contain empty sieves precisely if 𝒞\mathcal{C} does not satisfy the Ore condition.

In particular, Sh(𝒞,J at)Sh(𝒞,J d)Sh ¬¬(Set 𝒞 op)Sh(\mathcal{C},J_{at})\subseteq Sh(\mathcal{C},J_{d})\simeq Sh_{\neg\neg}(Set^{\mathcal{C}^{op}}) .


Last revised on May 31, 2022 at 18:21:00. See the history of this page for a list of all contributions to it.