Independent scholarship · Open accessThe monthly digest

Full manuscript · browser edition

Actions of Toposes on Toposes

Continuous, single-column reading. Equations, tables, proofs, references and appendices stay in document order.

Contents · 20 sections

Actions of Toposes on Toposes

Geometric actions, internal symmetry, and quotient constructions

Anonymous November 2026

Abstract. A monoidal Grothendieck topos can act on a topos by a geometric morphism out of their bicategorical product. We compare this notion with discrete groups acting by geometric autoequivalences, internal groups acting on objects, and groupoid actions in an ∞\infty-topos. We prove how a discrete parameter topos recovers a weak group action and explain why the equivariant topos is a quotient in the direction of geometric morphisms. Classifying topoi, localic actions, and a point-free example show what is retained by generalized points, torsors, and descent. We review the canonical isotropy action on topoi relative to a base. We also distinguish the action of a discrete group from the multiplication of its classifying topos and give the higher bar construction with its slice universal property.

Keywords. Grothendieck topoi, geometric morphisms, actions, classifying topoi, torsors, groupoid objects, descent.

The geometric meaning of an action

An action of a group on a space consists of a parameter object, an evaluation, and composition laws. The same description makes sense in the bicategory of Grothendieck topoi and geometric morphisms. There is also a distinct internal description in which the group and the acted-on space are objects of a fixed topos. The two descriptions can interact through classifying topoi and descent. The parameter object, the direction of inverse image, and the kind of quotient determine the relationship in a given example.

Use Top\mathbf{Top} for the bicategory of Grothendieck topoi, with arrows directed as geometric morphisms f:X→Yf:\mathcal X\to\mathcal Y. Thus f∗:Y→Xf^*:\mathcal Y\to\mathcal X is left exact and left adjoint to f∗f_*. By a geometric transformation we mean a natural transformation between inverse images in the displayed order. Invertible geometric transformations are the coherence cells in our actions. We fix universes large enough for the sites occurring below. Products mean bicategorical products in this geometric direction. The terminal topos is Set\mathrm{Set}. For the product-of-sites description see [8, 10].

Definition 1 (A monoidal topos and its action). A monoidal topos in this sense is a pseudomonoid T\mathcal T for the product of Top\mathbf{Top}. It has geometric morphisms m:T×T→T,e:Set→Tm:\mathcal T\times\mathcal T\to\mathcal T,\qquad e:\mathrm{Set}\to\mathcal T with invertible associativity and unit cells satisfying the pentagon and triangle. A left action on X\mathcal X is a geometric morphism a:T×X→Xa:\mathcal T\times\mathcal X\to\mathcal X with invertible cells a(m×1)≃a(1×a)a(m\times1)\simeq a(1\times a) and a(e×1)≃1Xa(e\times1)\simeq1_{\mathcal X} satisfying the action coherences. Product comparisons are understood.

This is the usual action definition in a cartesian bicategory. In inverse-image notation the evaluation is a∗:X→T×Xa^*:\mathcal X\to\mathcal T\times\mathcal X, with the associativity comparison between the two composites out of X\mathcal X. It is important here to use the geometric product before taking inverse image. The cartesian product of the underlying categories of sheaves typically gives a different topos in Top\mathbf{Top}.

For an exponentiable X\mathcal X in the bicategorical sense, an action can be curried to a morphism T→[X,X]\mathcal T\to[\mathcal X,\mathcal X]. The endomorphism topos then has a multiplication induced by composition, and the action cells say that this map preserves it coherently. The evaluation definition works without this exponential hypothesis. Exponentiability of Grothendieck topoi is a separate property studied by Johnstone and Joyal [5].

Generalized points determine the action

Write Geom⁡(B,X)\operatorname{Geom}(\mathcal B,\mathcal X) for the category of geometric morphisms and transformations. An action evaluates on each test topos B\mathcal B by the functor Geom⁡(B,T)×Geom⁡(B,X)⟶Geom⁡(B,X),(t,x)⟼a⟨t,x⟩.\begin{aligned} \operatorname{Geom}(\mathcal B,\mathcal T)\times \operatorname{Geom}(\mathcal B,\mathcal X) &\longrightarrow \operatorname{Geom}(\mathcal B,\mathcal X),\\ (t,x)&\longmapsto a\langle t,x\rangle . \end{aligned} The action cells induce the corresponding cells on all these functors. In particular, t:B→Tt:\mathcal B\to\mathcal T acts on B×X\mathcal B\times\mathcal X over B\mathcal B by the pair ⟨πB,a⟨tπB,πX⟩⟩\langle\pi_{\mathcal B},a\langle t\pi_{\mathcal B},\pi_{\mathcal X}\rangle\rangle. If T\mathcal T has an inverse morphism with the usual group identities up to coherent invertible cells, this endomorphism over B\mathcal B is an equivalence. The inverse is induced by the inverse generalized point, using the unit and multiplication cells.

Proposition 2 (Test-topos description). The evaluation aa and its coherence cells are equivalent to the operation on geometric morphisms from every test topos, with its coherence cells and pseudonaturality under precomposition.

Proof. A product gives a pseudonatural equivalence between the pair of input hom-categories and Geom⁡(B,T×X)\operatorname{Geom}(\mathcal B,\mathcal T\times\mathcal X). At B=T×X\mathcal B=\mathcal T\times\mathcal X, apply the operation to its two projections. The resulting geometric morphism is aa. For a pair (t,x)(t,x) with test topos B\mathcal B, pseudonaturality under ⟨t,x⟩\langle t,x\rangle identifies its evaluation with a⟨t,x⟩a\langle t,x\rangle. Evaluating transformations at the same projections recovers the transformation between action morphisms. Applying this argument with the products of two and three parameter factors recovers the coherence cells and their equations. This is the bicategorical Yoneda argument. ◻

Testing only at Set\mathrm{Set} can fail entirely. Let LL be the locale of the complete Boolean algebra of regular open subsets of R\mathbb R. Its Boolean algebra is atomless since any nonempty regular open contains a nonempty smaller interval with closure inside it. A locale point h:O(L)→2h:\mathcal O(L)\to 2 preserves arbitrary joins and finite meets. Set u=⋁{v:h(v)=0}u=\bigvee\{v:h(v)=0\}, so h(u)=0h(u)=0. Every nonzero element below ¬u\neg u has image 11. Two disjoint nonzero elements below ¬u\neg u would violate preservation of meets. Thus ¬u\neg u would be an atom. There is no such point.

The topos X=Sh⁡(L⨿L)\mathcal X=\operatorname{Sh}(L\amalg L) has no ordinary points. The identity and the swap of its two summands give two different actions of the two-element discrete group (realized geometrically as in the next subsection). On ordinary points the actions return the same empty description. With test topos Sh⁡(L)\operatorname{Sh}(L), the inclusion of the first summand is sent to itself or to the second summand. Pullback of the subterminal selecting the first summand distinguishes those two morphisms. For the locale and sheaf correspondence see [4, 8].

Discrete parameters and composition

For a set II write DI=SetID_I=\mathrm{Set}^I, the sheaf topos of the discrete locale on II. Its underlying category is a product of copies of Set\mathrm{Set}. In Top\mathbf{Top}, however, it is their coproduct. More generally a product of categories ∏iXi\prod_i\mathcal X_i is a coproduct of topoi in the geometric direction: a left exact left adjoint into it is a family of left exact left adjoints into the individual factors. Given their right adjoints, the right adjoint of the family is the product of their values in the common source category.

Proposition 3 (Discrete parameter topos). There is an equivalence in Top\mathbf{Top} DI×X≃∐i∈IX.D_I\times\mathcal X\simeq\coprod_{i\in I}\mathcal X . For a discrete monoid MM, the multiplication on MM induces one on DMD_M. Actions of DMD_M on X\mathcal X are weak monoid actions by geometric endomorphisms. For a group GG these are weak group actions by geometric autoequivalences.

Proof. A morphism B→DI\mathcal B\to D_I is classified by a partition of its terminal object, 1B=∐i∈IUi1_{\mathcal B}=\coprod_{i\in I}U_i, into disjoint subterminal objects. To see this, apply inverse image to the subterminal supported at each coordinate. In the other direction the inverse image sends (Si)(S_i) to ∐iSi⋅Ui\coprod_i S_i\cdot U_i, where Si⋅UiS_i\cdot U_i denotes the coproduct of SiS_i copies of UiU_i. It has right adjoint A↦(Hom⁡B(Ui,A))iA\mapsto(\operatorname{Hom}_{\mathcal B}(U_i,A))_i. The disjointness and pullback stability of coproducts give left exactness. A natural transformation gives inequalities Ui≤ViU_i\leq V_i between the two partitions. Each ViV_i is the coproduct of its intersections with all the UjU_j, of which only UiU_i remains. Hence the partitions coincide, and on copowers the transformation is fixed by its restriction to the UiU_i.

Disjoint stable coproducts give an equivalence B≃∏iB/Ui\mathcal B\simeq\prod_i\mathcal B/U_i. A morphism B→X\mathcal B\to\mathcal X therefore restricts to a family of morphisms B/Ui→X\mathcal B/U_i\to\mathcal X, and the restrictions determine it. A map B→∐iX\mathcal B\to\coprod_i\mathcal X has exactly the same data: pull back the subterminals selecting its summands and take the induced maps on the pieces. These descriptions are natural in B\mathcal B and also identify transformations. This proves the product formula.

In particular DI×J≃DI×DJD_{I\times J}\simeq D_I\times D_J. The action evaluation for DMD_M is therefore specified by its restrictions ama_m to the summands, and the cells restrict to amn≃amana_{mn}\simeq a_m a_n and the identity at the unit. Conversely these data assemble by the coproduct property. For a group the component at g−1g^{-1} supplies the inverse to the component at gg. ◻

In particular DGD_G and the classifying topos of GG-sets serve different purposes. The former parametrizes group elements. The latter classifies torsors and retains their automorphisms. For trivial action on Set\mathrm{Set}, a geometric quotient is the classifying topos, as Section 2 explains.

The quotient convention

For a geometric action, the action nerve has terms Tn×X\mathcal T^n\times\mathcal X. The first face at degree one is the action, and the other is projection. The higher faces compose parameters or act with one of them. The simplicial identities for a weak action hold up to coherent isomorphism. One can specify its geometric quotient by a bicategorical codescent property. A geometric morphism from the quotient to Y\mathcal Y amounts to f:X→Yf:\mathcal X\to\mathcal Y equipped with an invertible comparison fa≃fpr⁡Xf a\simeq f\operatorname{pr}_{\mathcal X}, subject to unit and composition. This specifies the quotient when the indicated universal object exists. The invertibility convention matters for monoids. For the trivial action of the additive monoid N\mathbb N on Set\mathrm{Set}, invertible comparisons are one automorphism of the selected point of Y\mathcal Y, with its powers. They are equivalently Z\mathbb Z-actions. The quotient for this convention is the classifying topos of Z\mathbb Z-sets by the trivial-action case of the next section. Allowing noninvertible comparisons instead gives actions of a single endomorphism. The classifying topos for these data is the category of sets with an endomorphism: an inverse image into it consists of a left exact cocontinuous functor to sets and a natural endomorphism, since limits and colimits of such diagrams are computed pointwise. The forgetful point of this topos has the natural endomorphism supplied by the structure map of each set. Its value on the set of natural numbers with successor is not invertible, so this point with its comparison gives only a lax cocone.

External actions by geometric automorphisms

Let GG be a small discrete group and E\mathcal E a Grothendieck 11-topos. An external action of GG is a pseudofunctor from the one-object groupoid BGBG to geometric morphisms, with value E\mathcal E at the object. Write ρg\rho_g for the resulting geometric autoequivalence. Composition satisfies ρgρh≃ρgh\rho_g\rho_h\simeq\rho_{gh}, with the usual associativity and unit coherences. In the direction of inverse images it is convenient to index by Tg=ρg−1∗T_g=\rho_{g^{-1}}^*. Then there are coherent isomorphisms μg,h:TgTh→Tgh\mu_{g,h}:T_gT_h\to T_{gh} and ϵ:id→T1\epsilon:\mathrm{id}\to T_1.

The equivariant category EG\mathcal E^G consists of pairs (X,θ)(X,\theta) where θg:TgX→X\theta_g:T_gX\to X are isomorphisms and θghμg,h,X=θgTg(θh),θ1ϵX=idX. \theta_{gh}\mu_{g,h,X}=\theta_gT_g(\theta_h), \qquad \theta_1\epsilon_X=\mathrm{id}_X . Morphisms commute with all θg\theta_g. Thus the underlying object need only be identified with each translate, and the identifications themselves are part of the object. A strict action is the case in which the composition and unit comparisons are identities.

Proposition 4 (Quotient and fixed-point variance). The category EG\mathcal E^G is a Grothendieck topos. Its forgetful functor U:EG→EU:\mathcal E^G\to\mathcal E is the inverse image of a geometric morphism q:E→EGq:\mathcal E\to\mathcal E^G. With the canonical isomorphisms qρg≃qq\rho_g\simeq q, this is a pseudocolimit of the action in geometric direction. In inverse-image direction it is the pseudolimit of the TgT_g in the 22-category of topoi and left exact left adjoints.

Proof. Each TgT_g is an equivalence and so preserves all limits and colimits. The underlying limit or colimit of an equivariant diagram inherits its equivariance by applying TgT_g to the defining cones. Coherence follows by the universal property, and UU creates these constructions and reflects isomorphisms. Disjointness of coproducts, pullback stability of colimits, and effectiveness of equivalence relations can therefore be checked in E\mathcal E. For a separating set of objects AA of E\mathcal E, use the equivariant objects L(A)=∐g∈GTgA.L(A)=\coprod_{g\in G} T_g A . The structure maps move summands by the μh,g\mu_{h,g}. A map L(A)→XL(A)\to X is equivariant precisely when it is determined by its component A→UXA\to UX at the identity summand, using ϵA\epsilon_A. Its other components are TgA→Tg(f)TgUX→θgUXT_gA\xrightarrow{T_g(f)}T_gUX\xrightarrow{\theta_g}UX. Since ordinary maps out of the chosen AA’s separate maps in E\mathcal E, the L(A)L(A)’s separate equivariant maps. Giraud’s characterization gives the topos claim [8, Ch. VII]. The object ∏gTgA\prod_g T_gA, with factors permuted using the same coherences, is right adjoint to UU. Indeed an equivariant map to that product is likewise determined by its identity component f:UX→Af:UX\to A, with gg-component Tg(f)θg−1T_g(f)\theta_g^{-1}. Thus UU is a left exact left adjoint.

For a test topos F\mathcal F, an inverse image into EG\mathcal E^G is a functor S:F→EGS:\mathcal F\to\mathcal E^G preserving finite limits and small colimits. Composing with UU gives F:F→EF:\mathcal F\to\mathcal E, together with natural isomorphisms TgF≃FT_gF\simeq F obeying (1). Conversely these data lift FF to EG\mathcal E^G, and preservation is checked by the created limits and colimits. A colimit-preserving functor between Grothendieck topoi has a right adjoint by the adjoint functor theorem for locally presentable categories. Compatible natural transformations are exactly transformations of the lifts. This gives the pseudolimit property. Reversing geometric arrows exchanges the cone with a cocone and gives the claimed pseudocolimit. ◻

The adjunction L⊣UL\dashv U also gives an algebraic description. The monad on E\mathcal E sends AA to ∐gTgA\coprod_{g}T_gA, with unit at the identity summand and multiplication carrying (g,h)(g,h) to ghgh using μg,h\mu_{g,h}. An algebra is a family of maps βg:TgX→X\beta_g:T_gX\to X satisfying (1). Even if invertibility is omitted from this definition, the equation for h=g−1h=g^{-1} gives a right inverse to βg\beta_g after identifying TgTg−1XT_gT_{g^{-1}}X with XX. The equation with gg and g−1g^{-1} reversed gives a left inverse by whiskering with TgT_g and using the coherences. Thus the monad’s algebras are exactly equivariant objects. For the trivial action the monad is the copower by the set GG, and its algebras are actions of the constant internal group. For a nontrivial action, its summands transport the object before mapping it back.

Similarly an equivariant geometric morphism f:A→Ef:\mathcal A\to\mathcal E has an inverse-image functor with coherent identifications Sgf∗≃f∗TgS_gf^*\simeq f^*T_g, where SgS_g acts on A\mathcal A. Applying f∗f^* to the underlying object and using these identifications defines EG→AG\mathcal E^G\to\mathcal A^G. It preserves finite limits and colimits under the forgetful functors, hence induces a geometric morphism AG→EG\mathcal A^G\to\mathcal E^G.

A site calculation and its examples

Suppose the action is realized on a small site (C,J)(C,J) by strict functors ag:C→Ca_g:C\to C preserving the topology. Then TgT_g acts by precomposition with ag−1opa_{g^{-1}}^{\mathrm{op}}. Form the Grothendieck construction C⋊GC\rtimes G, with objects those of CC and arrows c→dc\to d pairs (g,u:agc→d)(g,u:a_gc\to d). Their composite with (h,v):d→e(h,v):d\to e is (hg,v∘ah(u))(hg,v\circ a_h(u)). Give this category the topology generated by JJ-covers in the identity copy of CC.

A presheaf on C⋊GC\rtimes G is a presheaf on CC with the equivariance in (1). To see this, factor (g,u)(g,u) through the canonical isomorphism c→agcc\to a_gc followed by the identity-indexed arrow uu. Restriction along these canonical isomorphisms encodes the θg\theta_g’s, and factorization and the cocycle law recover all restrictions. A covering sieve SS on dd in CC extends to the sieve of all (g,u)(g,u) with u∈Su\in S in the larger category. The extended covering sieves form a topology since pullback along each canonical isomorphism is transported by a site automorphism, and pullback and transitivity for the identity-indexed arrows use JJ. In particular, to test transitivity for all arrows of an extended sieve it suffices to test the identity-indexed arrows, since each other arrow factors through one of those by an isomorphism. A matching family for such a sieve is determined by its entries along the identity-indexed arrows. The entry along (g,u)(g,u) must be obtained from the entry along (1,u)(1,u) by restriction along c→agcc\to a_gc, and this assignment satisfies the remaining compatibilities by the same factorization. Therefore the sheaf condition is exactly the JJ-sheaf condition on CC, and this site presents EG\mathcal E^G.

For the trivial action this gives EG≃Fun⁡(BG,E)\mathcal E^G\simeq\operatorname{Fun}(BG,\mathcal E). It is the geometric product E×BG\mathcal E\times\mathbf B G, where BG\mathbf B G denotes the topos of GG-sets. Indeed in this case C⋊G=C×BGC\rtimes G=C\times BG and the topology is the product-site topology with the trivial topology on BGBG [10, Ch. II]. In particular the trivial action on Set\mathbf{Set} has quotient BG\mathbf B G. A limit in geometric direction of the constant action on the terminal topos Set\mathbf{Set} would instead be Set\mathbf{Set}. The quotient includes a GG-action on every object.

For a discrete group acting on a space XX by homeomorphisms, apply the site construction to opens with their usual covering families. Equivariant objects are sheaves on XX with lifts of the action, or equivalently equivariant local homeomorphisms to XX. This is the classifying topos of the action groupoid G×X⇉XG\times X\rightrightarrows X [9]. For the transitive discrete space X=G/HX=G/H, taking the fiber over the identity coset recovers an HH-set AA. Its inverse construction has total space G×HAG\times_H A over G/HG/H. Hence the quotient topos is BH\mathbf B H although the orbit set has one point. In this example the remaining isotropy determines the entire quotient.

Internal actions and relative classifying topoi

Let E\mathcal E be a Grothendieck topos. An internal group GG is a group object in E\mathcal E. Write BEGB_{\mathcal E}G for the topos of left GG-objects in E\mathcal E. Its projection π:BEG→E\pi:B_{\mathcal E}G\to\mathcal E has inverse image the trivial-action functor and direct image the invariants functor. The latter assigns to a GG-object the subobject on which its action agrees with the projection. In this setting a right GG-torsor is a right GG-object TT such that T→1T\to1 is an epimorphism and T×G⟶T×T,(t,g)⟼(t,tg)T\times G\longrightarrow T\times T,\qquad (t,g)\longmapsto(t,tg) is an isomorphism. The epimorphism requires local inhabitation. Together the two conditions give a local identification with the regular action.

The universal torsor and change of base

The regular left GG-object GG has a commuting right action of the trivial GG-object π∗G\pi^*G. With this right action it is a π∗G\pi^*G-torsor, denoted UGU_G. For a geometric morphism f:F→Ef:\mathcal F\to\mathcal E, pullback of UGU_G gives an equivalence of groupoids Geom⁡/E≃(F,BEG)≃Tors⁡F(f∗G).\operatorname{Geom}_{/\mathcal E}^\simeq (\mathcal F,B_{\mathcal E}G) \simeq \operatorname{Tors}_{\mathcal F}(f^*G). The domain takes the isomorphism groupoid of geometric morphisms over E\mathcal E, with transformations written in the direction of inverse images. The comparison over E\mathcal E is a specified isomorphism. This is the relative classifying property [1, Thm. 2.3].

To see its construction, let TT be a right f∗Gf^*G-torsor. On a left GG-object AA, the associated inverse image is T×f∗Gf∗AT\times^{f^*G}f^*A, the quotient identifying (tg,a)(tg,a) with (t,ga)(t,ga). After pullback along T↠1T\twoheadrightarrow 1 the torsor has a section, so the contracted product becomes the underlying f∗Af^*A in that slice. Finite limits are therefore preserved, since this pullback is conservative. Colimit preservation follows from the contracted-product coequalizer and the preservation of colimits by products in a topos. Thus the functor has a right adjoint by the adjoint functor theorem for presentable categories. On trivial actions it agrees with f∗f^*, and on the regular left object its value is TT. For the other composite, apply an inverse image over E\mathcal E to the natural presentation of AA as UG×π∗Gπ∗A0U_G\times^{\pi^*G}\pi^*A_0, where π∗A0\pi^*A_0 bears the pullback of the original action as an internal action of π∗G\pi^*G. It gives the contracted product with the pulled-back torsor. This presentation also identifies natural isomorphisms over the base with torsor isomorphisms.

For any ff as above there is a pullback of topoi in the bicategorical sense: F×EBEG≃BF(f∗G).\mathcal F\times_{\mathcal E} B_{\mathcal E}G \simeq B_{\mathcal F}(f^*G). Indeed, a test topos over F\mathcal F sees the same pulled-back group and the same torsor groupoid on either side. Relative transformations between the corresponding inverse images reduce to equivariant torsor maps, which are isomorphisms because they are locally maps of regular torsors. See [1, Cor. 2.4].

Groupoid actions and descent

For an internal groupoid R⇉UR\rightrightarrows U with source ss and target tt, an action is an object A→UA\to U together with transport along arrows. Equivalently it is an isomorphism s∗A≃t∗As^*A\simeq t^*A over RR that is the identity on units and satisfies the composition law over composable pairs. Write BERB_{\mathcal E}R for the category of these objects. Limits and colimits are computed on the anchored objects in E/U\mathcal E/U, because s∗,t∗s^*,t^* preserve them. Transport then inherits the laws from the diagram. The slice’s universal colimits and effective equivalence relations give the same properties for action objects. To see the generator condition, use the free action on A→UA\to U, R×s,UA→UR\times_{s,U}A\to U, with anchor given by tt. Free actions on slice generators detect maps by this adjunction and the faithful forgetful functor. Giraud’s criterion makes BERB_{\mathcal E}R a Grothendieck topos. Its projection to E\mathcal E has inverse image sending CC to U×CU\times C with transport trivial in the second factor.

One can use the same contracted-product argument to classify right f∗Rf^*R-torsors. Here TT has an anchor to f∗Uf^*U, is locally inhabited, and the action induces an isomorphism T×f∗Uf∗R≃T×TT\times_{f^*U} f^*R\simeq T\times T, using the target of the arrow in the fiber product. The universal torsor is the arrow object with left transport by composition and right multiplication along its source anchor. After locally choosing a section of TT with anchor uu, its contracted product with a left action is the pullback of that action’s underlying object along uu. This proves the finite-limit condition in the classification as before. For the colimit condition, first form the fiber product over f∗Uf^*U in the slice and then take the quotient in F\mathcal F. Pullbacks and the functor composing structure maps preserve colimits. Reconstruction uses the natural isomorphism R×RA≃AR\times^R A\simeq A by evaluation. The identity at the anchor of aa represents its inverse on the class of aa.

For example, for the action groupoid G⋉XG\ltimes X of a left GG-object XX, an anchored object transports exactly when it is an equivariant map to XX. Hence BE(G⋉X)≃(BEG)/X.B_{\mathcal E}(G\ltimes X) \simeq (B_{\mathcal E}G)/X. A map from F\mathcal F over E\mathcal E into this quotient topos is specified by a right f∗Gf^*G-torsor TT and an equivariant map T→f∗XT\to f^*X, with a(tg)=g−1a(t)a(tg)=g^{-1}a(t). This follows because a map to the slice also supplies a section of the associated object T×f∗Gf∗XT\times^{f^*G}f^*X. When E=Set⁡\mathcal E=\operatorname{Set} and X=G/HX=G/H for a subgroup HH, taking the fiber above the identity coset yields an equivalence of this slice with the topos of HH-sets. Induction G×H(−)G\times^H(-) gives an inverse, since its fibers are transported by left translations and the ambiguities in the transport are exactly the HH-action.

There is also an explicit atlas for BERB_{\mathcal E}R. Regard RR as a left action via its target map. The slice (BER)/R(B_{\mathcal E}R)/R is equivalent to E/U\mathcal E/U: take the fiber of an equivariant map over the identity arrows, and reconstruct its fiber over rr by transport along rr from the identity of s(r)s(r). The projection of this slice is an étale surjection since R→1R\to 1 is epic in the action topos. Its double overlap is the slice over R×RR\times R in BERB_{\mathcal E}R. Under the first projection and the identity-fiber equivalence, it becomes E/R\mathcal E/R: the pair (1u,r)(1_u,r) is specified by an arrow with target uu. The overlap maps are given by the two anchors of rr, and the cocycle on triple overlaps composes arrows. Effective descent along this atlas recovers the transport description of BERB_{\mathcal E}R.

For an epimorphism U→VU\to V the Čech groupoid U×VU⇉UU\times_V U\rightrightarrows U gives BE(U×VU⇉U)≃E/VB_{\mathcal E}(U\times_V U\rightrightarrows U) \simeq\mathcal E/V. Indeed, its arrows identify the objects over UU by exactly the descent cocycle. Quotient the resulting equivalence relation on an action object to descend it to VV. Any local representative of a quotient element can be transported to a specified UU-coordinate, and that representative is then unique. Pullback to UU therefore recovers the original object’s action. In particular refining the object space of an internal groupoid along an epimorphism preserves its classifying topos by this descent argument. This is a basic Morita invariance mechanism. For continuous and localic groupoids, object and arrow spaces have topology and the actions used in the classifying topos are on equivariant sheaves. The geometric hypotheses and the descent results for those settings are developed in [9, 10, 6].

Example 5 (A sheaf of groups with trivial global group). Over S1=R/ZS^1=\mathbb R/\mathbb Z, let GG be the locally constant sheaf with fiber Z/3\mathbb Z/3 and monodromy a↦−aa\mapsto-a. It is represented as an étale group bundle by identifying (t,a)(t,a) with (t+1,−a)(t+1,-a) in R×Z/3\mathbb R\times\mathbb Z/3. A global section is fixed by inversion, so the group of global sections is trivial. Pullback of BSh⁡(S1)GB_{\operatorname{Sh}(S^1)}G over any point gives the topos of Z/3\mathbb Z/3-sets by base change. This topos has a nontrivial regular action that is noninitial in the action topos and has no equivariant global element. In Set⁡\operatorname{Set} every noninitial object has a global element. The global section group has classifying topos Set⁡\operatorname{Set}. The relative classifying topos thus retains the local group and its transport.

Concrete monoidal topoi and actions

An ordinary group GG produces two different topoi. The topos Sh⁡(Gdisc)\operatorname{Sh}(G_{\mathrm{disc}}) is the sheaf topos of the discrete locale with set of points GG. Multiplication of the discrete locale gives a monoidal multiplication of this topos, with inverse induced by g↦g−1g\mapsto g^{-1}. By contrast, write BG\mathbf B G for the topos of left GG-sets. It classifies right GG-torsors, and the ordinary group law on GG does not in general induce a multiplication of BG\mathbf B G. The distinction already appears over Set\mathbf{Set}: the discrete locale has one geometric point for every element of GG, whereas BG\mathbf B G has one point up to isomorphism, with automorphism group GG.

From locales to monoidal topoi

The construction Sh⁡\operatorname{Sh} preserves products of locales as products of topoi. One explanation uses the localic reflection: for a Grothendieck topos U\mathcal U, maps U→Sh⁡(L)\mathcal U\to\operatorname{Sh}(L) classify frame maps O(L)⟶Sub⁡U(1).\mathcal O(L)\longrightarrow \operatorname{Sub}_{\mathcal U}(1). Here a frame map preserves arbitrary joins and finite meets. The frame of opens of a locale product is the coproduct of the two frames. Hence giving the classifying data for a map into the sheaves on a product is equivalent to giving them for each factor. The same statement holds for the comparisons of maps: they are determined by the order on the images of opens, which can be checked on the two factor frames [8, Ch. IX, §5, Prop. 3]. Thus if a group locale KK acts on a locale LL, the multiplication and action induce Sh⁡(K)×Sh⁡(L)⟶Sh⁡(L)\operatorname{Sh}(K)\times\operatorname{Sh}(L) \longrightarrow\operatorname{Sh}(L) with the required associativity and unit. The construction is pointfree and applies to localic group actions used in topos descent [6, Chs. V–VIII].

For example, ordinary real addition and real addition with the discrete topology give different group topoi. Both have (R,+)(\mathbb R,+) as their group of geometric points. Their subterminal frames distinguish them. For the discrete topology the frame is a Boolean power set. For the usual topology, a proper nonempty open interval has no open complement in the connected real line. The two topoi therefore cannot be equivalent. Both act on the sheaf topos of the usual real line by translations. The two actions agree on individual ordinary points, while their parameter topoi encode different families of translations.

A simple monoid that is not a group is the Sierpiński locale Σ\Sigma, whose frame is free on one generator. A geometric map U→Sh⁡(Σ)\mathcal U\to\operatorname{Sh}(\Sigma) classifies a subobject of 11 in U\mathcal U [12]. The product Sh⁡(Σ)2\operatorname{Sh}(\Sigma)^2 classifies two such subobjects. The universal operation (U,V)↦U∩V(U,V)\mapsto U\cap V, together with the unit 1↪11\hookrightarrow 1, defines a monoidal structure on Sh⁡(Σ)\operatorname{Sh}(\Sigma). The operation commutes with inverse image functors because they preserve finite limits. On the product topos Sh⁡(Σ)r\operatorname{Sh}(\Sigma)^r it defines an action by (U,V1,…,Vr)⟼(U∩V1,…,U∩Vr).(U,V_1,\ldots,V_r) \longmapsto (U\cap V_1,\ldots,U\cap V_r). Applying first UU and then WW gives the same family as applying U∩WU\cap W. The point classifying 0↪10\hookrightarrow1 sends every tuple to the one consisting of zero subobjects. For r≥1r\geq1 it cannot act by an equivalence since this map identifies distinct tuples already over Set\mathbf{Set}. This example has a noninvertible action in geometric morphisms, expressed entirely by geometric logical operations.

When a torsor classifier is monoidal

The classifying property of BG\mathbf B G identifies geometric maps U→BG\mathcal U\to \mathbf B G with right GG-torsors in U\mathcal U; GG is interpreted as the constant group [8, Ch. VIII, Thm. 2.7]. For an ordinary group HH, a right GG-torsor in BH\mathbf B H is a right GG-torsor of sets equipped with a commuting left HH-action. Choose an element pp of the torsor and write h⋅p=pϕ(h)h\cdot p=p\phi(h). Commutation of the actions gives ϕ(hh′)=ϕ(h)ϕ(h′)\phi(hh')=\phi(h)\phi(h'). Replacing pp by pgpg conjugates ϕ\phi by g−1g^{-1}. Conversely, any homomorphism ϕ:H→G\phi:H\to G gives a torsor by taking the regular right GG-set with left HH-action by ϕ\phi. Thus maps BH→BG\mathbf B H\to \mathbf B G are classified up to isomorphism by homomorphisms H→GH\to G up to conjugation. A map induced by ϕ\phi has inverse image given by restriction of action along ϕ\phi. Composition of the geometric maps therefore agrees with composition of the inducing homomorphisms.

There is a product equivalence BG×BH≃B(G×H)\mathbf B G\times \mathbf B H\simeq \mathbf B(G\times H). Indeed, the product of a right GG-torsor and a right HH-torsor is a right (G×H)(G\times H)-torsor. Conversely, a (G×H)(G\times H)-torsor PP recovers the two torsors by quotienting by the HH- and GG-actions respectively. The reconstructed product identifies with PP after a covering over which the torsor is trivial, hence globally by descent. The two constructions are natural in the test topos, so represent the same product.

Proposition 6. The topos BG\mathbf B G admits a monoidal structure for the Cartesian product of Grothendieck topoi if and only if GG is abelian. For abelian GG the standard multiplication makes BG\mathbf B G a group object, with all laws interpreted up to the canonical equivalences of products.

Proof. A multiplication is classified by a homomorphism μ:G×G→G\mu:G\times G\to G. The unit map Set→BG\mathbf{Set}\to \mathbf B G is unique up to isomorphism, since right GG-torsors in sets are isomorphic to GG. The two unit comparisons imply that each restriction of μ\mu to a factor is conjugate to the identity. Both restrictions are therefore surjective. Their images commute elementwise, since the factors of G×GG\times G do. Thus any two elements of GG commute.

When GG is abelian, (g,h)↦gh(g,h)\mapsto gh is a homomorphism. Its associativity and unit equalities induce the required coherences by restriction of actions. The inverse homomorphism g↦g−1g\mapsto g^{-1} similarly supplies inversion. ◻

There are actions of such group topoi that are visible on automorphisms of points. Let AA be abelian and let λ:A→Z(H)\lambda:A\to Z(H) be a homomorphism into the center of an arbitrary group HH. The map A×H⟶H,(a,h)⟼λ(a)hA\times H\longrightarrow H,\qquad (a,h)\longmapsto\lambda(a)h is a homomorphism. It gives a left action BA×BH→BH\mathbf B A\times \mathbf B H\to \mathbf B H because the unit and associativity equalities hold on these homomorphisms. Conversely, an action morphism with invertible unit comparison is represented by a homomorphism ϕ:A×H→H\phi:A\times H\to H whose restriction to HH is conjugate to the identity. After conjugating the whole homomorphism this restriction is the identity, so ϕ(a,1)\phi(a,1) must lie in Z(H)Z(H). This recovers λ\lambda at the level of action morphisms. Equivalences that also keep track of all chosen coherence data require their own comparison.

Taking geometric points of this action, each of the three classifying topoi involved has just one isomorphism class of point. But after choosing the regular torsors as representatives, the action on their automorphisms is the displayed map (a,h)↦λ(a)h(a,h)\mapsto\lambda(a)h. In particular, λ\lambda can be read off by setting h=1h=1. The groupoid of points thus retains information lost on passing to mere isomorphism classes, even for these atomic topoi.

Internal actions in an ∞\infty-topos

An internal group can act on an object of an ∞\infty-topos. In this setting the quotient is again an object, and the category of equivariant objects is itself a slice ∞\infty-topos. The construction uses effective groupoids and descent for slices [7, §6.1.3]. The resulting interpretation of actions and torsors is developed in [11, §§3–4.1].

Action groupoids and effective descent

Let X\mathcal X be an ∞\infty-topos. Present a group object GG by a reduced simplicial groupoid G∙G_\bullet in X\mathcal X. Thus G0=1G_0=1 and the Segal maps identify GnG_n with GnG^n, where G=G1G=G_1. Write BG=∣G∙∣BG=|G_\bullet| for the geometric realization. Groupoid effectivity identifies G∙G_\bullet with the Čech nerve of the effective epimorphism 1→BG1\to BG. In particular, G≃1×BG1G\simeq 1\times_{BG}1 with its loop group structure. The equivalence between group objects and pointed connected objects is given by this delooping and loop construction [7, Thm. 6.1.0.6 and Lemma 7.2.2.11].

One definition of an action on an object XX is a cartesian transformation U∙→G∙U_\bullet\to G_\bullet with U0=XU_0=X. Here cartesian means that the naturality square for each simplicial operator is a homotopy pullback. In particular Un≃X×GnU_n\simeq X\times G^n by using the initial vertex in [n][n]. The two vertex maps U1⇉XU_1\rightrightarrows X are then the projection and the action. For an ordinary right action, for example, these take (x,g)(x,g) to xx and xgxg. Composition and the higher compatibility of the action are encoded by the simplicial maps. In this description the Segal and invertibility conditions for U∙U_\bullet are obtained by pulling back those for G∙G_\bullet: a string of arrows over (g1,…,gn)(g_1,\ldots,g_n) is determined by a starting object. Thus U∙U_\bullet is an action groupoid. The category Act⁡G(X)\operatorname{Act}_G(\mathcal X) uses all natural transformations over G∙G_\bullet as its equivariant maps.

Proposition 7 (The slice of actions). The functor that sends an action groupoid to its realized map over BGBG gives an equivalence Act⁡G(X)≃X/BG.\operatorname{Act}_G(\mathcal X) \simeq \mathcal X_{/BG}. The image of an action on XX has source the homotopy quotient X//G=∣[n]↦X×Gn∣X//G=|[n]\mapsto X\times G^n| with the action bar structure. Its fiber over 1→BG1\to BG is XX.

Proof. For each morphism Y→BGY\to BG, pull back G∙→BGG_\bullet\to BG along it. The result is a cartesian transformation to G∙G_\bullet with degree-zero object Y×BG1Y\times_{BG}1. Its realization is YY: colimits in an ∞\infty-topos are stable under pullback. Conversely let U∙→G∙U_\bullet\to G_\bullet be a cartesian transformation and realize both diagrams. The cartesian descent criterion for colimit diagrams makes the augmented transformation U∙→G∙U_\bullet\to G_\bullet cartesian as well [7, Thm. 6.1.3.9(4) and Prop. 6.1.3.10]. In particular U0⟶∣U∙∣↓↓1⟶BG\begin{array}{ccc} U_0 & \longrightarrow & |U_\bullet|\\ \downarrow && \downarrow\\ 1 & \longrightarrow & BG \end{array} is a homotopy pullback, and the same holds with UnU_n and GnG_n in place of U0U_0 and 11. These comparisons identify the original diagram with the one pulled back from its realization. They are natural in transformations over G∙G_\bullet, so the equivalence covers maps and their higher coherences. ◻

By taking the slice version of descent one can also express the result as X/BG≃lim⁡[n]∈ΔX/Gn.\mathcal X_{/BG}\simeq \lim_{[n]\in\Delta}\mathcal X_{/G_n}. An object of this limit is an object over each stage together with coherent identifications along the pullback functors. This describes the action without requiring a chosen model of the quotient. A geometric inverse image commutes with these particular action constructions: it preserves finite limits and geometric realizations, so it carries the action groupoid, the homotopy quotient, and the fiber square to their counterparts for the pulled-back group and object.

What a family of quotient points specifies

For a parameter object TT, a map T→BGT\to BG pulls back the universal effective epimorphism to a GG-torsor P→TP\to T. A torsor here is an effective epimorphism equipped with a right action such that (p,g)↦(p,pg)(p,g)\mapsto(p,pg) identifies P×GP\times G with P×TPP\times_T P. Its action groupoid is the Čech nerve of P→TP\to T, so realization recovers TT and supplies a map to BGBG. Proposition 7 makes these constructions inverse. More generally, a map T→X//GT\to X//G specifies the torsor induced by its composite with BGBG together with an equivariant map P→XP\to X. Indeed a lift of T→BGT\to BG to X//GX//G is a map in the slice, which under the proposition is the asserted equivariant map. Thus the fibers of the mapping-space map to Map⁡X(T,BG)\operatorname{Map}_{\mathcal X}(T,BG) record equivariant maps for each torsor. This is the family-level refinement of an orbit space.

There is also an object of homotopy fixed points. Given Y=X//G→BGY=X//G\to BG, the pullback along q:BG→1q:BG\to 1 has a right adjoint Πq\Pi_q since X\mathcal X is locally cartesian closed. Its value Πq(Y)\Pi_q(Y) represents sections of YY parametrized by an arbitrary object: Map⁡X(T,ΠqY)≃Map⁡X/BG(T×BG,Y).\operatorname{Map}_{\mathcal X}(T,\Pi_q Y) \simeq \operatorname{Map}_{\mathcal X_{/BG}} (T\times BG,Y). Under Proposition 7, T×BGT\times BG corresponds to the trivial action on TT, because the quotient of the trivial bar construction is its product with BGBG. Consequently Πq(Y)\Pi_q(Y) represents equivariant maps out of objects carrying the trivial action. It is denoted XhGX^{hG}. The homotopy orbit object X//GX//G and the homotopy fixed-point object XhGX^{hG} therefore come from the left and right adjoints to the trivial-action functor. Indeed, the left adjoint to q∗q^* takes a map over BGBG to its domain, forgetting the structure map. In spaces these two objects specialize to the colimit and limit of the coherent GG-diagram, respectively.

Example 8 (Reflection of a circle). In spaces let C2C_2 act on S1S^1 by complex conjugation. The coarse orbit space is an interval. The homotopy quotient has a fibration sequence S1→S1//C2→BC2S^1\to S^1//C_2\to BC_2. It admits a section because 1∈S11\in S^1 is a fixed point. The fundamental group sequence is a split extension of C2C_2 by Z\mathbb Z, using π2(BC2)=0\pi_2(BC_2)=0. The nonidentity element acts on Z\mathbb Z by inversion since conjugation reverses the orientation of S1S^1. It follows that S1//C2≃B(Z⋊C2).S^1//C_2\simeq B(\mathbb Z\rtimes C_2). Here both the base and the fiber are aspherical, so the long exact homotopy sequence makes the total space aspherical as well. The quotient has infinite dihedral fundamental group, a distinction invisible to the interval of orbits.

Comparison with external symmetry

An external action on an ∞\infty-topos assigns geometric autoequivalences to elements of a group and coherence to their composition. The associated category of equivariant objects is the limit of the resulting diagram of categories of inverse images. The action groupoid above uses objects and arrows internal to a single X\mathcal X. The two constructions meet for a trivial external action of a discrete group HH. Use the constant internal group H‾=∐h∈H1\underline H=\coprod_{h\in H}1. By disjoint universal coproducts, X/H‾n≃∏HnX\mathcal X_{/\underline H^n}\simeq \prod_{H^n}\mathcal X. Slice descent then gives X/BH‾≃Fun⁡(BH,X)\mathcal X_{/B\underline H}\simeq \operatorname{Fun}(BH,\mathcal X): the cosimplicial limit of these products imposes the coherent action along the nerve of HH. This identification supplies an underlying internal object group for that particular category of equivariant objects.

Isotropy actions and crossed topoi

A topos carries an intrinsic source of symmetry. For a Grothendieck topos E\mathcal E and an object UU, write pU:E/U→Ep_U:\mathcal E/U\to\mathcal E for the slice projection, with inverse image A↦A×U→UA\mapsto A\times U\to U. Funk, Hofstra, and Steinberg prove that the presheaf of groups U⟼Aut⁡(pU∗)U\longmapsto \operatorname{Aut}(p_U^*) is represented by a group object ZEZ_{\mathcal E} [2, §4.1]. The restrictions of this presheaf pull natural automorphisms back along maps of slices. The identity of ZEZ_{\mathcal E} therefore gives a natural automorphism of pZE∗p_{Z_{\mathcal E}}^*. Reading its component at AA over ZEZ_{\mathcal E} gives the canonical action ZE×A→AZ_{\mathcal E}\times A\to A, with the order of factors chosen for left actions and multiplication given by composition. Every morphism of E\mathcal E intertwines these actions by naturality. The action on the group object itself is conjugation.

On a fixed base E\mathcal E, a crossed topos of Funk, Hofstra, and Steinberg is described by an internal crossed module ∂:K→ZE\partial:K\to Z_{\mathcal E} using the canonical ZEZ_{\mathcal E}-action on KK. Equivariance of ∂\partial follows from naturality, and the Peiffer identity says that ∂(k)\partial(k) acts on KK by conjugation by kk. These are the internal conditions for algebras of their isotropy monad on grouped topoi [2, §2 and Cor. 5.10]. The action induced on each AA restricts the canonical one along ∂\partial. Objects on which this induced action is trivial form a topos whose inclusion in E\mathcal E is the inverse image of a connected atomic isotropy quotient [2, Prop. 4.17].

Example 9 (Isotropy of the topos of GG-sets). For a discrete group GG and E=BG\mathcal E=\mathbf B G, the group ZEZ_{\mathcal E} is GG with its conjugation action. Indeed, writing a component of an automorphism of pU∗p_U^* as (x,u)↦(αX(x,u),u)(x,u)\mapsto(\alpha_X(x,u),u), put zu=αG(1,u)z_u=\alpha_G(1,u) on the left regular GG-set. Naturality along G→X, g↦gxG\to X,\ g\mapsto gx forces αX(x,u)=zux\alpha_X(x,u)=z_u x. Equivariance of the components gives zhu=hzuh−1z_{hu}=hz_uh^{-1}. Conversely this condition makes the displayed formula an equivariant natural automorphism. Global sections of the internal isotropy group form the centre of GG, which can be trivial even for a nontrivial GG. The quotient by the full isotropy action is Set\mathbf{Set}, since its objects are the sets with trivial GG-action.

Henry’s localic isotropy IE→EI_{\mathcal E}\to\mathcal E extends the construction to all geometric probes f:F→Ef:\mathcal F\to\mathcal E. Its group of sections after base change along ff is Aut⁡(f∗)\operatorname{Aut}(f^*). As a topos over E\mathcal E, IEI_{\mathcal E} is the pseudopullback of the diagonal of E\mathcal E along itself. A test morphism into this pullback gives two morphisms to E\mathcal E and two invertible comparisons between them. Composing one comparison with the inverse of the other gives a morphism and its automorphism. Composition gives the group law over E\mathcal E, and ZEZ_{\mathcal E} is its internal group of points [3, Props. 2.2–2.3 and 6.1]. This group acts on every topos Y→EY\to\mathcal E in the pseudo-slice. On test morphisms it changes the specified comparison between a map through YY and the map to E\mathcal E by composing with the isotropy automorphism [3, Prop. 2.5].

For a locally positive (overt) internal localic group KK and an isotropy action K→IEK\to I_{\mathcal E}, the quotient E→EK\mathcal E\to\mathcal E_K has inverse image the inclusion of objects with trivial induced action and is connected atomic. Conversely every connected atomic morphism f:E→Ff:\mathcal E\to\mathcal F is such a quotient by the relative isotropy IE/FI_{\mathcal E/\mathcal F}, which consists of automorphisms becoming trivial over F\mathcal F [3, Props. 5.4 and 5.9]. Local positivity is a hypothesis of this atomicity assertion. Arbitrary localic isotropy quotients give hyperconnected morphisms [3, §4].

Comparing ordinary and higher quotients

The central action of Section 4 admits a direct quotient calculation at both levels. Let AA be abelian, HH a discrete group, and χ:A→Z(H)\chi:A\to Z(H) a homomorphism into its center. Write BH\mathbf B H for the topos of left HH-sets, and BHBH for the pointed connected classifying space. At the topos level the action of BA\mathbf B A on BH\mathbf B H is induced by the group homomorphism (a,h)↦χ(a)h(a,h)\mapsto\chi(a)h. At the space level the same homomorphism gives an action of the group object BABA on BHBH. One can take the usual nerves of the one-object groupoids, in which products and the action are given levelwise by the group laws.

Proposition 10 (A central-action quotient). With invertible comparison cells, the geometric quotient of BH\mathbf B H by BA\mathbf B A is B(H/χ(A))\mathbf B(H/\chi(A)). The quotient space BH//BABH//BA is connected and has homotopy groups π1(BH//BA)≃H/χ(A),π2(BH//BA)≃ker⁡χ.\pi_1(BH//BA)\simeq H/\chi(A),\qquad \pi_2(BH//BA)\simeq\ker\chi . Its homotopy groups in degrees greater than two vanish.

Proof. Put K=H/χ(A)K=H/\chi(A). A geometric morphism f:BH→Yf:\mathbf B H\to\mathcal Y has inverse image F:Y→H-SetF:\mathcal Y\to H\text{-}\mathrm{Set}. A comparison between the action and projection on this functor is an isomorphism between two functors into (A×H)-Set(A\times H)\text{-}\mathrm{Set}, given by restricting the HH-actions along (a,h)↦χ(a)h(a,h)\mapsto\chi(a)h and (a,h)↦h(a,h)\mapsto h respectively. On the second functor AA acts trivially. An equivariant isomorphism can exist only if χ(A)\chi(A) acts trivially on each F(Y)F(Y). The unit condition forces the comparison restricted to HH to be the identity. That restriction is faithful on morphisms, so the only possible comparison is the identity. It satisfies the composition condition whenever χ(A)\chi(A) acts trivially.

Thus a cocone is precisely a left exact cocontinuous FF taking values in the full subcategory of HH-sets on which χ(A)\chi(A) acts trivially. This subcategory is K-SetK\text{-}\mathrm{Set}, and its inclusion creates finite limits and small colimits. A transformation of cocones is exactly a transformation of the resulting functors into it. This gives the geometric universal property of BK\mathbf B K.

For the higher calculation, the bar description gives a fiber sequence BH→BH//BA→B2ABH\to BH//BA\to B^2A. The boundary A=π2(B2A)→π1(BH)=HA=\pi_2(B^2A)\to\pi_1(BH)=H in its long exact sequence is χ\chi up to the choice of boundary sign. Indeed, the orbit map BA→BHBA\to BH of the base point is BχB\chi and is equivariant for translation acting on BABA. It induces a map over B2AB^2A from BA//BA≃∗BA//BA\simeq * to BH//BABH//BA. The connecting homomorphism for the first fiber sequence identifies π2(B2A)\pi_2(B^2A) with π1(BA)\pi_1(BA). Naturality gives the stated boundary in the second. Exactness yields the displayed groups, since BHBH is a 11-type and B2AB^2A is a 22-type. ◻

If χ\chi is trivial, the geometric quotient is BH\mathbf B H. The space construction gives BH×B2ABH\times B^2A. In particular, for HH trivial it still retains AA in degree two. This example compares a specific codescent construction in Grothendieck topoi with a homotopy quotient in spaces. Each universal property uses its own category of test objects.

Which description to use

A monoidal topos acting by geometric morphisms can be evaluated on all test topoi. Discrete parameter topoi specialize to families of geometric automorphisms, and the topos of equivariant objects realizes their geometric quotient. Internal actions have a fixed base topos and are represented by relative classifying topoi or their slices. They retain torsors and local transport. Isotropy supplies a canonical symmetry attached to a topos itself. Passing to higher topoi replaces ordinary descent isomorphisms by coherent groupoid objects and permits higher homotopy in the quotient. The descriptions agree where their hypotheses identify the same descent data, as with ordinary discrete groups, and the central-action calculation specifies one place where their invariants differ.

References

  1. P. Cassou-Noguès, T. Chinburg, B. Morin, and M. J. Taylor. The classifying topos of a group scheme and invariants of symmetric bundles. Proc. London Math. Soc. 109 (2014), 1093–1136. DOI.

  2. J. Funk, P. Hofstra, and B. Steinberg. Isotropy and crossed toposes. Theory Appl. Categ. 26(24) (2012), 660–709. Journal article.

  3. S. Henry. The localic isotropy group of a topos. Theory Appl. Categ. 33(41) (2018), 1318–1345. Journal article.

  4. P. T. Johnstone. The point of pointless topology. Bull. Amer. Math. Soc. 8 (1983), 41–53. DOI.

  5. P. T. Johnstone and A. Joyal. Continuous categories and exponentiable toposes. J. Pure Appl. Algebra 25 (1982), 255–296. DOI.

  6. A. Joyal and M. Tierney. An Extension of the Galois Theory of Grothendieck. Memoirs Amer. Math. Soc. 51(309), 1984. DOI.

  7. J. Lurie. Higher Topos Theory. Annals of Mathematics Studies 170. Princeton University Press, 2009. Author’s copy.

  8. S. Mac Lane and I. Moerdijk. Sheaves in Geometry and Logic: A First Introduction to Topos Theory. Springer, 1992. DOI.

  9. I. Moerdijk. The classifying topos of a continuous groupoid. I. Trans. Amer. Math. Soc. 310 (1988), 629–668. DOI.

  10. I. Moerdijk. Classifying Spaces and Classifying Topoi. Lecture Notes in Mathematics 1616. Springer, 1995. DOI.

  11. T. Nikolaus, U. Schreiber, and D. Stevenson. Principal ∞\infty-bundles: general theory. J. Homotopy Relat. Struct. 10 (2015), 749–801. DOI.

  12. S. Vickers. The double powerlocale and exponentiation: A case study in geometric logic. Theory Appl. Categ. 12(13) (2004), 372–422. Journal article.

Complete flattened LaTeX source
\documentclass[10pt,letterpaper,twocolumn]{article}
\usepackage[T1]{fontenc}

\usepackage{lmodern}

\usepackage[
  letterpaper,left=.72in,right=.72in,top=.75in,bottom=.73in,
  headheight=14pt,headsep=14pt,footskip=26pt
]{geometry}

\usepackage{microtype}

\usepackage{mathtools,amssymb,amsthm}

\usepackage{xcolor}

\usepackage{tikz-cd}

\usepackage{titlesec}

\usepackage{fancyhdr}

\usepackage{caption}

\usepackage{booktabs,balance}

\usepackage{xurl}

\usepackage{hyperref}

\usepackage{bookmark}


\definecolor{paperink}{HTML}{202824}
\definecolor{paperaccent}{HTML}{145A51}
\definecolor{papermuted}{HTML}{51655F}
\definecolor{paperrule}{HTML}{B6CBC4}
\hypersetup{
  pdftitle={Actions of Toposes on Toposes},
  pdfauthor={Anonymous},
  pdfsubject={Geometric actions, internal symmetry, and quotient constructions},
  colorlinks=true,linkcolor=paperaccent,citecolor=paperaccent,urlcolor=paperaccent
}
\AtBeginDocument{\color{paperink}}
\setlength{\columnsep}{.29in}
\setlength{\parindent}{1em}
\setlength{\parskip}{1pt plus .5pt}
\setlength{\emergencystretch}{1.7em}
\linespread{1.035}
\allowdisplaybreaks[1]
\raggedbottom
\clubpenalty=1000
\widowpenalty=1000

\titleformat{\section}{\large\bfseries\raggedright}
  {\color{paperaccent}\thesection}{.55em}{}
\titlespacing*{\section}{0pt}{14pt plus 3pt minus 2pt}{5pt}
\titleformat{\subsection}{\normalsize\bfseries\raggedright}
  {\color{paperaccent}\thesubsection}{.55em}{}
\titlespacing*{\subsection}{0pt}{10pt plus 2pt minus 1pt}{4pt}
\captionsetup{font=small,labelfont={bf,color=paperaccent}}

\pagestyle{fancy}
\fancyhf{}
\fancyhead[L]{\footnotesize\sffamily\color{papermuted}Actions of toposes on toposes}
\fancyfoot[R]{\footnotesize\sffamily\color{papermuted}\thepage}
\renewcommand{\headrulewidth}{0pt}
\fancypagestyle{plain}{  \fancyhf{}  \fancyfoot[R]{\footnotesize\sffamily\color{papermuted}\thepage}  \renewcommand{\headrulewidth}{0pt}}

\theoremstyle{plain}
\newtheorem{theorem}{Theorem}[section]
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{corollary}[theorem]{Corollary}
\theoremstyle{definition}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{example}[theorem]{Example}
\theoremstyle{remark}
\newtheorem{remark}[theorem]{Remark}
\DeclareMathOperator{\Map}{Map}
\DeclareMathOperator{\Tot}{Tot}
\DeclareMathOperator{\Fun}{Fun}
\DeclareMathOperator{\Hom}{Hom}
\DeclareMathOperator{\grad}{grad}
\DeclareMathOperator{\Ext}{Ext}
\DeclareMathOperator{\Res}{Res}
\DeclareMathOperator{\Kl}{Kl}
\DeclareMathOperator{\id}{id}
\DeclareMathOperator{\rank}{rank}
\newcommand{\E}{\mathcal E}
\newcommand{\Spc}{\mathcal S}
\newcommand{\R}{\mathbb R}
\newcommand{\C}{\mathbb C}
\newcommand{\Z}{\mathbb Z}
\newcommand{\overB}{\E_{/B}}
\tikzcdset{every label/.append style={font=\small}}
\begin{document}
\twocolumn[
\begin{@twocolumnfalse}
\noindent\textcolor{paperaccent}{\rule{\textwidth}{1pt}}\par
\vspace{11pt}
{\noindent\fontsize{29}{33}\selectfont\bfseries
Actions of Toposes on Toposes\par}
\vspace{8pt}
{\noindent\large\color{papermuted}
Geometric actions, internal symmetry, and quotient constructions\par}
\vspace{10pt}
{\noindent\small Anonymous \quad\textbullet\quad November 2026\par}
\vspace{9pt}
\noindent\begin{minipage}{\textwidth}\small
\textbf{Abstract.}
A monoidal Grothendieck topos can act on a topos by a geometric
morphism out of their bicategorical product. We compare this notion
with discrete groups acting by geometric autoequivalences, internal
groups acting on objects, and groupoid actions in an $\infty$-topos.
We prove how a discrete parameter topos recovers a weak group action
and explain why the equivariant topos is a quotient in the direction
of geometric morphisms. Classifying topoi, localic actions, and a
point-free example show what is retained by generalized points,
torsors, and descent. We review the canonical isotropy action on topoi
relative to a base. We also distinguish the action of a discrete
group from the multiplication of its classifying topos and give the
higher bar construction with its slice universal property.
\par\medskip
\noindent\textbf{Keywords.}\enspace
Grothendieck topoi, geometric morphisms, actions, classifying topoi,
torsors, groupoid objects, descent.
\end{minipage}
\vspace{17pt}
\end{@twocolumnfalse}
]
\thispagestyle{plain}

\section{The geometric meaning of an action}
\label{sec:geometry}

An action of a group on a space consists of a parameter
object, an evaluation, and composition laws. The same
description makes sense in the bicategory of Grothendieck
topoi and geometric morphisms. There is also a distinct
internal description in which the group and the acted-on
space are objects of a fixed topos. The two descriptions
can interact through classifying topoi and descent. The parameter object, the direction of inverse image,
and the kind of quotient determine the relationship in
a given example.

Use $\mathbf{Top}$ for the bicategory of Grothendieck
topoi, with arrows directed as geometric morphisms
$f:\mathcal X\to\mathcal Y$. Thus
$f^*:\mathcal Y\to\mathcal X$ is left exact and left
adjoint to $f_*$. By a geometric transformation we
mean a natural transformation between inverse images
in the displayed order. Invertible geometric
transformations are the coherence cells in our actions.
We fix universes large enough for the sites occurring
below. Products mean bicategorical products in this
geometric direction. The terminal topos is $\mathrm{Set}$.
For the product-of-sites description see
\cite{maclane-moerdijk,moerdijk-classifying}.

\begin{definition}[A monoidal topos and its action]
A monoidal topos in this sense is a pseudomonoid
$\mathcal T$ for the product of $\mathbf{Top}$. It
has geometric morphisms
\[
 m:\mathcal T\times\mathcal T\to\mathcal T,\qquad
 e:\mathrm{Set}\to\mathcal T
\]
with invertible associativity and unit cells satisfying the
pentagon and triangle. A left action on $\mathcal X$ is
a geometric morphism $a:\mathcal T\times\mathcal X\to\mathcal X$
with invertible cells
$a(m\times1)\simeq a(1\times a)$ and
$a(e\times1)\simeq1_{\mathcal X}$ satisfying the action
coherences. Product comparisons are understood.
\end{definition}

This is the usual action definition in a cartesian
bicategory. In inverse-image notation the evaluation is
$a^*:\mathcal X\to\mathcal T\times\mathcal X$, with the
associativity comparison between the two composites out
of $\mathcal X$. It is important here to use the geometric
product before taking inverse image. The cartesian product
of the underlying categories of sheaves typically gives a
different topos in $\mathbf{Top}$.

For an exponentiable $\mathcal X$ in the bicategorical
sense, an action can be curried to a morphism
$\mathcal T\to[\mathcal X,\mathcal X]$. The endomorphism
topos then has a multiplication induced by composition,
and the action cells say that this map preserves it
coherently. The evaluation definition works without this
exponential hypothesis. Exponentiability of Grothendieck
topoi is a separate property studied by Johnstone and
Joyal \cite{johnstone-joyal-exponential}.

\subsection{Generalized points determine the action}

Write $\operatorname{Geom}(\mathcal B,\mathcal X)$ for
the category of geometric morphisms and transformations.
An action evaluates on each test topos $\mathcal B$ by
the functor
\[
 \begin{aligned}
 \operatorname{Geom}(\mathcal B,\mathcal T)\times
 \operatorname{Geom}(\mathcal B,\mathcal X)
 &\longrightarrow \operatorname{Geom}(\mathcal B,\mathcal X),\\
 (t,x)&\longmapsto a\langle t,x\rangle .
 \end{aligned}
\]
The action cells induce the corresponding cells on all
these functors. In particular, $t:\mathcal B\to\mathcal T$
acts on $\mathcal B\times\mathcal X$ over $\mathcal B$
by the pair
$\langle\pi_{\mathcal B},a\langle t\pi_{\mathcal B},\pi_{\mathcal X}\rangle\rangle$.
If $\mathcal T$ has an inverse morphism with the usual
group identities up to coherent invertible cells, this
endomorphism over $\mathcal B$ is an equivalence.
The inverse is induced by the inverse generalized point,
using the unit and multiplication cells.

\begin{proposition}[Test-topos description]
The evaluation $a$ and its coherence cells are equivalent
to the operation on geometric morphisms from every test
topos, with its coherence cells and pseudonaturality under
precomposition.
\end{proposition}

\begin{proof}
A product gives a pseudonatural equivalence between the
pair of input hom-categories and
$\operatorname{Geom}(\mathcal B,\mathcal T\times\mathcal X)$.
At $\mathcal B=\mathcal T\times\mathcal X$, apply the
operation to its two projections. The resulting
geometric morphism is $a$. For a pair $(t,x)$ with test
topos $\mathcal B$, pseudonaturality under
$\langle t,x\rangle$ identifies its evaluation with
$a\langle t,x\rangle$. Evaluating transformations at
the same projections recovers the transformation
between action morphisms. Applying this argument with
the products of two and three parameter factors recovers
the coherence cells and their equations. This is the
bicategorical Yoneda argument.
\end{proof}

Testing only at $\mathrm{Set}$ can fail entirely.
Let $L$ be the locale of the complete Boolean algebra
of regular open subsets of $\mathbb R$. Its Boolean
algebra is atomless since any nonempty regular open
contains a nonempty smaller interval with closure
inside it. A locale point $h:\mathcal O(L)\to 2$
preserves arbitrary joins and finite meets. Set
$u=\bigvee\{v:h(v)=0\}$, so $h(u)=0$.
Every nonzero element below $\neg u$ has image $1$.
Two disjoint nonzero elements below $\neg u$ would
violate preservation of meets. Thus $\neg u$ would
be an atom. There is no such point.

The topos $\mathcal X=\operatorname{Sh}(L\amalg L)$
has no ordinary points. The identity and the swap of
its two summands give two different actions of the
two-element discrete group (realized geometrically
as in the next subsection). On ordinary points the
actions return the same empty description. With test
topos $\operatorname{Sh}(L)$, the inclusion of the
first summand is sent to itself or to the second
summand. Pullback of the subterminal selecting the
first summand distinguishes those two morphisms.
For the locale and sheaf correspondence see
\cite{johnstone-pointless,maclane-moerdijk}.

\subsection{Discrete parameters and composition}

For a set $I$ write $D_I=\mathrm{Set}^I$, the sheaf
topos of the discrete locale on $I$. Its underlying
category is a product of copies of $\mathrm{Set}$.
In $\mathbf{Top}$, however, it is their coproduct.
More generally a product of categories
$\prod_i\mathcal X_i$ is a coproduct of topoi in the
geometric direction: a left exact left adjoint into
it is a family of left exact left adjoints into the
individual factors. Given their right adjoints, the
right adjoint of the family is the product of their
values in the common source category.

\begin{proposition}[Discrete parameter topos]
\label{prop:discrete-param}
There is an equivalence in $\mathbf{Top}$
\[
 D_I\times\mathcal X\simeq\coprod_{i\in I}\mathcal X .
\]
For a discrete monoid $M$, the multiplication on $M$
induces one on $D_M$. Actions of $D_M$ on $\mathcal X$
are weak monoid actions by geometric endomorphisms.
For a group $G$ these are weak group actions by geometric
autoequivalences.
\end{proposition}

\begin{proof}
A morphism $\mathcal B\to D_I$ is classified by a
partition of its terminal object,
$1_{\mathcal B}=\coprod_{i\in I}U_i$, into disjoint
subterminal objects. To see this, apply inverse image to
the subterminal supported at each coordinate. In the
other direction the inverse image sends $(S_i)$ to
$\coprod_i S_i\cdot U_i$, where $S_i\cdot U_i$ denotes
the coproduct of $S_i$ copies of $U_i$. It has right
adjoint $A\mapsto(\operatorname{Hom}_{\mathcal B}(U_i,A))_i$.
The disjointness and pullback stability of coproducts
give left exactness. A natural transformation gives inequalities $U_i\leq V_i$
between the two partitions. Each $V_i$ is the coproduct of its
intersections with all the $U_j$, of which only $U_i$ remains.
Hence the partitions coincide, and on copowers the transformation
is fixed by its restriction to the $U_i$.

Disjoint stable coproducts give an equivalence
$\mathcal B\simeq\prod_i\mathcal B/U_i$.
A morphism $\mathcal B\to\mathcal X$ therefore
restricts to a family of morphisms
$\mathcal B/U_i\to\mathcal X$, and the restrictions
determine it. A map
$\mathcal B\to\coprod_i\mathcal X$ has exactly the same
data: pull back the subterminals selecting its summands
and take the induced maps on the pieces.
These descriptions are natural in $\mathcal B$ and
also identify transformations. This proves the product
formula.

In particular $D_{I\times J}\simeq D_I\times D_J$.
The action evaluation for $D_M$ is therefore specified
by its restrictions $a_m$ to the summands, and the cells
restrict to $a_{mn}\simeq a_m a_n$ and the identity at
the unit. Conversely these data assemble by the coproduct
property. For a group the component at $g^{-1}$ supplies
the inverse to the component at $g$.
\end{proof}

In particular $D_G$ and the classifying topos of $G$-sets
serve different purposes. The former parametrizes group
elements. The latter classifies torsors and retains their
automorphisms. For trivial action on $\mathrm{Set}$, a
geometric quotient is the classifying topos, as
Section~\ref{sec:external} explains.

\subsection{The quotient convention}

For a geometric action, the action nerve has terms $\mathcal T^n\times\mathcal X$. The first
face at degree one is the action, and the other is
projection. The higher faces compose parameters or
act with one of them. The simplicial identities for a weak
action hold up to coherent isomorphism. One can specify its geometric
quotient by a bicategorical codescent property.
A geometric morphism from the quotient to $\mathcal Y$
amounts to $f:\mathcal X\to\mathcal Y$ equipped
with an \emph{invertible} comparison
$f a\simeq f\operatorname{pr}_{\mathcal X}$, subject to
unit and composition. This specifies the quotient
when the indicated universal object exists.
The invertibility convention matters for monoids.
For the trivial action of the additive monoid
$\mathbb N$ on $\mathrm{Set}$, invertible comparisons are
one automorphism of the selected point of $\mathcal Y$,
with its powers. They are equivalently $\mathbb Z$-actions.
The quotient for this convention is the classifying
topos of $\mathbb Z$-sets by the trivial-action case
of the next section. Allowing noninvertible comparisons
instead gives actions of a single endomorphism. The classifying
topos for these data is the category of sets with an endomorphism:
an inverse image into it consists of a left exact cocontinuous
functor to sets and a natural endomorphism, since limits and
colimits of such diagrams are computed pointwise. The forgetful
point of this topos has the natural endomorphism supplied by the
structure map of each set. Its value on the set of natural numbers
with successor is not invertible, so this point with its comparison
gives only a lax cocone.
\par
\section{External actions by geometric automorphisms}
\label{sec:external}

Let \(G\) be a small discrete group and \(\mathcal E\) a
Grothendieck \(1\)-topos. An external action of \(G\) is a
pseudofunctor from the one-object groupoid \(BG\) to geometric
morphisms, with value \(\mathcal E\) at the object. Write
\(\rho_g\) for the resulting geometric autoequivalence.
Composition satisfies \(\rho_g\rho_h\simeq\rho_{gh}\), with
the usual associativity and unit coherences. In the direction of
inverse images it is convenient to index by
\(T_g=\rho_{g^{-1}}^*\). Then there are coherent isomorphisms
\(\mu_{g,h}:T_gT_h\to T_{gh}\) and
\(\epsilon:\mathrm{id}\to T_1\).

The equivariant category \(\mathcal E^G\) consists of pairs
\((X,\theta)\) where \(\theta_g:T_gX\to X\) are isomorphisms and
\begin{equation}
 \label{eq:ext-cocycle}
 \theta_{gh}\mu_{g,h,X}=\theta_gT_g(\theta_h),
 \qquad \theta_1\epsilon_X=\mathrm{id}_X .
\end{equation}
Morphisms commute with all \(\theta_g\). Thus the underlying
object need only be identified with each translate, and the
identifications themselves are part of the object. A strict
action is the case in which the composition and unit
comparisons are identities.

\begin{proposition}[Quotient and fixed-point variance]
\label{prop:ext-quotient}
The category \(\mathcal E^G\) is a Grothendieck topos. Its
forgetful functor \(U:\mathcal E^G\to\mathcal E\) is the
inverse image of a geometric morphism
\(q:\mathcal E\to\mathcal E^G\). With the canonical
isomorphisms \(q\rho_g\simeq q\), this is a pseudocolimit
of the action in geometric direction. In inverse-image
direction it is the pseudolimit of the \(T_g\) in the
\(2\)-category of topoi and left exact left adjoints.
\end{proposition}

\begin{proof}
Each \(T_g\) is an equivalence and so preserves all limits
and colimits. The underlying limit or colimit of an
equivariant diagram inherits its equivariance by applying
\(T_g\) to the defining cones. Coherence follows by the
universal property, and \(U\) creates these constructions
and reflects isomorphisms. Disjointness of coproducts,
pullback stability of colimits, and effectiveness of
equivalence relations can therefore be checked in \(\mathcal E\).
For a separating set of objects \(A\) of \(\mathcal E\), use
the equivariant objects
\[
 L(A)=\coprod_{g\in G} T_g A .
\]
The structure maps move summands by the \(\mu_{h,g}\).
A map \(L(A)\to X\) is equivariant precisely when it is
determined by its component \(A\to UX\) at the identity
summand, using \(\epsilon_A\). Its other components are
\(T_gA\xrightarrow{T_g(f)}T_gUX\xrightarrow{\theta_g}UX\).
Since ordinary maps out of the chosen \(A\)'s separate maps
in \(\mathcal E\), the \(L(A)\)'s separate equivariant maps.
Giraud's characterization gives the topos claim
\cite[Ch.~VII]{maclane-moerdijk}. The object
\(\prod_g T_gA\), with factors permuted using the same
coherences, is right adjoint to \(U\). Indeed an equivariant
map to that product is likewise determined by its identity
component \(f:UX\to A\), with \(g\)-component
\(T_g(f)\theta_g^{-1}\). Thus \(U\) is a left exact left adjoint.

For a test topos \(\mathcal F\), an inverse image into
\(\mathcal E^G\) is a functor \(S:\mathcal F\to\mathcal E^G\)
preserving finite limits and small colimits. Composing
with \(U\) gives \(F:\mathcal F\to\mathcal E\), together
with natural isomorphisms \(T_gF\simeq F\) obeying
\eqref{eq:ext-cocycle}. Conversely these data lift \(F\)
to \(\mathcal E^G\), and preservation is checked by the
created limits and colimits. A colimit-preserving functor
between Grothendieck topoi has a right adjoint by the
adjoint functor theorem for locally presentable categories.
Compatible natural transformations are exactly
transformations of the lifts. This gives the pseudolimit
property. Reversing geometric arrows exchanges the
cone with a cocone and gives the claimed pseudocolimit.
\end{proof}

The adjunction \(L\dashv U\) also gives an algebraic
description. The monad on \(\mathcal E\) sends \(A\) to
\(\coprod_{g}T_gA\), with unit at the identity summand and
multiplication carrying \((g,h)\) to \(gh\) using
\(\mu_{g,h}\). An algebra is a family of maps
\(\beta_g:T_gX\to X\) satisfying \eqref{eq:ext-cocycle}.
Even if invertibility is omitted from this definition,
the equation for \(h=g^{-1}\) gives a right inverse to
\(\beta_g\) after identifying \(T_gT_{g^{-1}}X\) with \(X\).
The equation with \(g\) and \(g^{-1}\) reversed gives a
left inverse by whiskering with \(T_g\) and using the
coherences. Thus the monad's algebras are exactly
equivariant objects. For the trivial action the monad
is the copower by the set \(G\), and its algebras are
actions of the constant internal group. For a
nontrivial action, its summands transport the object
before mapping it back.

Similarly an equivariant geometric morphism
\(f:\mathcal A\to\mathcal E\) has an inverse-image functor
with coherent identifications \(S_gf^*\simeq f^*T_g\),
where \(S_g\) acts on \(\mathcal A\). Applying \(f^*\)
to the underlying object and using these identifications
defines \( \mathcal E^G\to\mathcal A^G\). It
preserves finite limits and colimits under the forgetful
functors, hence induces a geometric morphism
\(\mathcal A^G\to\mathcal E^G\).

\subsection{A site calculation and its examples}

Suppose the action is realized on a small site \((C,J)\)
by strict functors \(a_g:C\to C\) preserving the topology.
Then \(T_g\) acts by precomposition with
\(a_{g^{-1}}^{\mathrm{op}}\). Form the Grothendieck
construction \(C\rtimes G\), with objects those of \(C\)
and arrows \(c\to d\) pairs \((g,u:a_gc\to d)\).
Their composite with \((h,v):d\to e\) is
\((hg,v\circ a_h(u))\). Give this category the topology
generated by \(J\)-covers in the identity copy of \(C\).

A presheaf on \(C\rtimes G\) is a presheaf on \(C\) with
the equivariance in \eqref{eq:ext-cocycle}. To see this,
factor \((g,u)\) through the canonical isomorphism
\(c\to a_gc\) followed by the identity-indexed arrow \(u\).
Restriction along these canonical isomorphisms encodes the
\(\theta_g\)'s, and factorization and the cocycle law recover
all restrictions. A covering sieve \(S\) on \(d\) in \(C\)
extends to the sieve of all \((g,u)\) with \(u\in S\) in the
larger category. The extended covering sieves form a
topology since pullback along each canonical isomorphism
is transported by a site automorphism, and pullback and
transitivity for the identity-indexed arrows use \(J\).
In particular, to test transitivity for all arrows of an
extended sieve it suffices to test the identity-indexed
arrows, since each other arrow factors through one of those
by an isomorphism. A matching family for such a sieve is
determined by its entries along the identity-indexed
arrows. The entry along \((g,u)\) must be obtained from
the entry along \((1,u)\) by restriction along
\(c\to a_gc\), and this assignment satisfies the remaining
compatibilities by the same factorization. Therefore
the sheaf condition is exactly the \(J\)-sheaf condition
on \(C\), and this site presents \(\mathcal E^G\).

For the trivial action this gives
\(\mathcal E^G\simeq\operatorname{Fun}(BG,\mathcal E)\).
It is the geometric product \(\mathcal E\times\mathbf B G\),
where \(\mathbf B G\) denotes the topos of \(G\)-sets.
Indeed in this case \(C\rtimes G=C\times BG\) and the
topology is the product-site topology with the trivial
topology on \(BG\) \cite[Ch.~II]{moerdijk-classifying}.
In particular the trivial action on \(\mathbf{Set}\)
has quotient \(\mathbf B G\). A limit in geometric
direction of the constant action on the terminal topos
\(\mathbf{Set}\) would instead be \(\mathbf{Set}\).
The quotient includes a \(G\)-action on every object.

For a discrete group acting on a space \(X\) by
homeomorphisms, apply the site construction to opens with
their usual covering families. Equivariant objects are sheaves
on \(X\) with lifts of the action, or equivalently
equivariant local homeomorphisms to \(X\). This is the
classifying topos of the action groupoid
\(G\times X\rightrightarrows X\)
\cite{moerdijk-groupoid}. For the transitive discrete
space \(X=G/H\), taking the fiber over the identity coset
recovers an \(H\)-set \(A\). Its inverse construction has
total space \(G\times_H A\) over \(G/H\). Hence the
quotient topos is \(\mathbf B H\) although the orbit set
has one point. In this example the remaining isotropy
determines the entire quotient.
\par
\section{Internal actions and relative classifying topoi}
\label{sec:internal}

Let \(\mathcal E\) be a Grothendieck topos. An internal
group \(G\) is a group object in \(\mathcal E\).
Write \(B_{\mathcal E}G\) for the topos of left \(G\)-objects
in \(\mathcal E\). Its projection
\(\pi:B_{\mathcal E}G\to\mathcal E\) has inverse image the
trivial-action functor and direct image the invariants
functor. The latter assigns to a \(G\)-object the subobject
on which its action agrees with the projection. In this
setting a right \(G\)-torsor is a right \(G\)-object \(T\)
such that \(T\to1\) is an epimorphism and
\[
 T\times G\longrightarrow T\times T,\qquad
 (t,g)\longmapsto(t,tg)
\]
is an isomorphism. The epimorphism requires local
inhabitation. Together the two conditions give a
local identification with the regular action.

\subsection{The universal torsor and change of base}

The regular left \(G\)-object \(G\) has a commuting right
action of the trivial \(G\)-object \(\pi^*G\). With this
right action it is a \(\pi^*G\)-torsor, denoted \(U_G\).
For a geometric morphism \(f:\mathcal F\to\mathcal E\),
pullback of \(U_G\) gives an equivalence of groupoids
\begin{equation}
 \operatorname{Geom}_{/\mathcal E}^\simeq
   (\mathcal F,B_{\mathcal E}G)
 \simeq \operatorname{Tors}_{\mathcal F}(f^*G).
 \label{eq:torsor-classification}
\end{equation}
The domain takes the isomorphism groupoid of geometric
morphisms over \(\mathcal E\), with transformations
written in the direction of inverse images. The
comparison over \(\mathcal E\) is a specified isomorphism.
This is the relative classifying property
\cite[Thm.~2.3]{cassou-torsors}.

To see its construction, let \(T\) be a right \(f^*G\)-torsor.
On a left \(G\)-object \(A\), the associated inverse
image is \(T\times^{f^*G}f^*A\), the quotient identifying
\((tg,a)\) with \((t,ga)\).
After pullback along \(T\twoheadrightarrow 1\) the
torsor has a section, so the contracted product becomes
the underlying \(f^*A\) in that slice. Finite limits are
therefore preserved, since this pullback is conservative.
Colimit preservation follows from the contracted-product
coequalizer and the preservation of colimits by products
in a topos. Thus the functor has a right adjoint by the
adjoint functor theorem for presentable categories. On
trivial actions it agrees with \(f^*\), and on the regular
left object its value is \(T\). For the other composite,
apply an inverse image over \(\mathcal E\) to the
natural presentation of \(A\) as
\(U_G\times^{\pi^*G}\pi^*A_0\), where \(\pi^*A_0\) bears the
pullback of the original action as an internal action of
\(\pi^*G\). It gives the contracted product with the pulled-back torsor. This presentation
also identifies natural isomorphisms over the base with
torsor isomorphisms.

For any \(f\) as above there is a pullback of topoi in the
bicategorical sense:
\[
 \mathcal F\times_{\mathcal E} B_{\mathcal E}G
   \simeq B_{\mathcal F}(f^*G).
\]
Indeed, a test topos over \(\mathcal F\) sees the same
pulled-back group and the same torsor groupoid on either
side. Relative transformations between the corresponding
inverse images reduce to equivariant torsor maps, which
are isomorphisms because they are locally maps of regular
torsors. See \cite[Cor.~2.4]{cassou-torsors}.

\subsection{Groupoid actions and descent}

For an internal groupoid \(R\rightrightarrows U\) with
source \(s\) and target \(t\), an action is an object
\(A\to U\) together with transport along arrows.
Equivalently it is an isomorphism \(s^*A\simeq t^*A\)
over \(R\) that is the identity on units and satisfies
the composition law over composable pairs. Write
\(B_{\mathcal E}R\) for the category of these objects.
Limits and colimits are computed on the anchored objects
in \(\mathcal E/U\), because \(s^*,t^*\) preserve them.
Transport then inherits the laws from the diagram.
The slice's universal colimits and effective equivalence
relations give the same properties for action objects.
To see the generator condition, use the free action on
\(A\to U\), \(R\times_{s,U}A\to U\), with anchor given by
\(t\). Free actions on slice generators detect maps by
this adjunction and the faithful forgetful functor.
Giraud's criterion makes \(B_{\mathcal E}R\) a Grothendieck
topos. Its projection to \(\mathcal E\) has inverse image
sending \(C\) to \(U\times C\) with transport trivial in
the second factor.

One can use the same contracted-product argument to
classify right \(f^*R\)-torsors. Here \(T\) has an anchor
to \(f^*U\), is locally inhabited, and the action induces
an isomorphism \(T\times_{f^*U} f^*R\simeq T\times T\),
using the target of the arrow in the fiber product.
The universal torsor is the arrow object with left
transport by composition and right multiplication along
its source anchor. After locally choosing a section of
\(T\) with anchor \(u\), its contracted product with a
left action is the pullback of that action's underlying
object along \(u\). This proves the finite-limit condition
in the classification as before. For the colimit condition,
first form the fiber product over \(f^*U\) in the slice
and then take the quotient in \(\mathcal F\).
Pullbacks and the functor composing structure maps
preserve colimits. Reconstruction uses the natural
isomorphism \(R\times^R A\simeq A\) by evaluation.
The identity at the anchor of \(a\) represents its
inverse on the class of \(a\).

For example, for the action groupoid \(G\ltimes X\) of a
left \(G\)-object \(X\), an anchored object transports
exactly when it is an equivariant map to \(X\). Hence
\[
 B_{\mathcal E}(G\ltimes X)
   \simeq (B_{\mathcal E}G)/X.
\]
A map from \(\mathcal F\) over \(\mathcal E\) into this
quotient topos is specified by a right \(f^*G\)-torsor
\(T\) and an equivariant map \(T\to f^*X\), with
\(a(tg)=g^{-1}a(t)\). This follows because a map to the
slice also supplies a section of the associated object
\(T\times^{f^*G}f^*X\).
When \(\mathcal E=\operatorname{Set}\) and \(X=G/H\)
for a subgroup \(H\), taking the fiber above the
identity coset yields an equivalence of this slice
with the topos of \(H\)-sets. Induction \(G\times^H(-)\)
gives an inverse, since its fibers are transported by
left translations and the ambiguities in the transport
are exactly the \(H\)-action.

There is also an explicit atlas for \(B_{\mathcal E}R\).
Regard \(R\) as a left action via its target map. The slice
\((B_{\mathcal E}R)/R\) is equivalent to \(\mathcal E/U\):
take the fiber of an equivariant map over the identity
arrows, and reconstruct its fiber over \(r\) by transport
along \(r\) from the identity of \(s(r)\). The projection
of this slice is an \'{e}tale surjection since \(R\to 1\)
is epic in the action topos. Its double overlap is the
slice over \(R\times R\) in \(B_{\mathcal E}R\).
Under the first projection and the identity-fiber
equivalence, it becomes \(\mathcal E/R\): the pair
\((1_u,r)\) is specified by an arrow with target \(u\).
The overlap maps are given by the two anchors of \(r\),
and the cocycle on triple overlaps composes arrows.
Effective descent along this atlas recovers the
transport description of \(B_{\mathcal E}R\).

For an epimorphism \(U\to V\) the \v{C}ech groupoid
\(U\times_V U\rightrightarrows U\) gives
\(B_{\mathcal E}(U\times_V U\rightrightarrows U)
\simeq\mathcal E/V\). Indeed, its arrows identify the
objects over \(U\) by exactly the descent cocycle.
Quotient the resulting equivalence relation on an
action object to descend it to \(V\). Any local
representative of a quotient element can be transported
to a specified \(U\)-coordinate, and that representative
is then unique. Pullback to \(U\) therefore recovers
the original object's action. In particular refining
the object space of an internal groupoid along an
epimorphism preserves its classifying topos by this
descent argument. This is a basic Morita invariance
mechanism. For continuous and localic groupoids, object
and arrow spaces have topology and the actions used in
the classifying topos are on equivariant sheaves.
The geometric hypotheses and the descent results for
those settings are developed in
\cite{moerdijk-groupoid,moerdijk-classifying,joyal-tierney}.

\begin{example}[A sheaf of groups with trivial global group]
Over \(S^1=\mathbb R/\mathbb Z\), let \(G\) be the locally
constant sheaf with fiber \(\mathbb Z/3\) and monodromy
\(a\mapsto-a\). It is represented as an \'{e}tale group
bundle by identifying \((t,a)\) with \((t+1,-a)\) in
\(\mathbb R\times\mathbb Z/3\). A global section is fixed
by inversion, so the group of global sections is
trivial. Pullback of \(B_{\operatorname{Sh}(S^1)}G\)
over any point gives the topos of \(\mathbb Z/3\)-sets
by base change. This topos has a nontrivial regular
action that is noninitial in the action topos and has no
equivariant global element. In \(\operatorname{Set}\) every noninitial
object has a global element. The global section group
has classifying topos \(\operatorname{Set}\). The relative
classifying topos thus retains the local group and its
transport.
\end{example}
\par
\section{Concrete monoidal topoi and actions}
\label{sec:module-examples}

An ordinary group $G$ produces two different topoi. The
topos $\operatorname{Sh}(G_{\mathrm{disc}})$ is the sheaf topos
of the discrete locale with set of points $G$. Multiplication
of the discrete locale gives a monoidal multiplication of this
topos, with inverse induced by $g\mapsto g^{-1}$. By contrast,
write $\mathbf B G$ for the topos of left $G$-sets. It classifies right
$G$-torsors, and the ordinary group law on $G$ does not in
general induce a multiplication of $\mathbf B G$. The distinction
already appears over $\mathbf{Set}$: the discrete locale has
one geometric point for every element of $G$, whereas $\mathbf B G$
has one point up to isomorphism, with automorphism group $G$.

\subsection{From locales to monoidal topoi}

The construction $\operatorname{Sh}$ preserves products of
locales as products of topoi. One explanation uses the
localic reflection: for a Grothendieck topos $\mathcal U$,
maps $\mathcal U\to\operatorname{Sh}(L)$ classify frame maps
\[
  \mathcal O(L)\longrightarrow
       \operatorname{Sub}_{\mathcal U}(1).
\]
Here a frame map preserves arbitrary joins and finite meets.
The frame of opens of a locale product is the coproduct of
the two frames. Hence giving the classifying data for a map
into the sheaves on a product is equivalent to giving them
for each factor. The same statement holds for the comparisons
of maps: they are determined by the order on the images of
opens, which can be checked on the two factor frames
\cite[Ch.~IX, \S5, Prop.~3]{maclane-moerdijk}. Thus if a group locale $K$
acts on a locale $L$, the multiplication and action induce
\[
 \operatorname{Sh}(K)\times\operatorname{Sh}(L)
     \longrightarrow\operatorname{Sh}(L)
\]
with the required associativity and unit. The construction is
pointfree and applies to localic group actions used in
topos descent \cite[Chs.~V--VIII]{joyal-tierney}.

For example, ordinary real addition and real addition with
the discrete topology give different group topoi. Both
have $(\mathbb R,+)$ as their group of geometric points.
Their subterminal frames distinguish them. For the discrete
topology the frame is a Boolean power set. For the usual
topology, a proper nonempty open interval has no open
complement in the connected real line. The two topoi
therefore cannot be equivalent. Both act on the sheaf topos
of the usual real line by translations. The two actions
agree on individual ordinary points, while their parameter
topoi encode different families of translations.

A simple monoid that is not a group is the Sierpi\'nski locale
$\Sigma$, whose frame is free on one generator. A geometric map
$\mathcal U\to\operatorname{Sh}(\Sigma)$ classifies a subobject of
$1$ in $\mathcal U$ \cite{vickers-double}. The product
$\operatorname{Sh}(\Sigma)^2$ classifies two such subobjects.
The universal operation $(U,V)\mapsto U\cap V$, together with
the unit $1\hookrightarrow 1$, defines a monoidal structure on
$\operatorname{Sh}(\Sigma)$. The operation commutes with inverse
image functors because they preserve finite limits. On the product
topos $\operatorname{Sh}(\Sigma)^r$ it defines an action by
\[
   (U,V_1,\ldots,V_r)
      \longmapsto (U\cap V_1,\ldots,U\cap V_r).
\]
Applying first $U$ and then $W$ gives the same family as applying
$U\cap W$. The point classifying $0\hookrightarrow1$ sends every
tuple to the one consisting of zero subobjects. For $r\geq1$ it
cannot act by an equivalence since this map identifies distinct
tuples already over $\mathbf{Set}$. This example has a
noninvertible action in geometric morphisms, expressed
entirely by geometric logical operations.

\subsection{When a torsor classifier is monoidal}

The classifying property of $\mathbf B G$ identifies geometric maps
$\mathcal U\to \mathbf B G$ with right $G$-torsors in $\mathcal U$;
$G$ is interpreted as the constant group
\cite[Ch.~VIII, Thm.~2.7]{maclane-moerdijk}. For an ordinary group
$H$, a right $G$-torsor in $\mathbf B H$ is a right $G$-torsor of sets
equipped with a commuting left $H$-action. Choose an element $p$
of the torsor and write $h\cdot p=p\phi(h)$. Commutation of the
actions gives $\phi(hh')=\phi(h)\phi(h')$. Replacing $p$ by
$pg$ conjugates $\phi$ by $g^{-1}$. Conversely, any
homomorphism $\phi:H\to G$ gives a torsor by taking the
regular right $G$-set with left $H$-action by $\phi$.
Thus maps $\mathbf B H\to \mathbf B G$ are classified up to isomorphism
by homomorphisms $H\to G$ up to conjugation. A map induced
by $\phi$ has inverse image given by restriction of action
along $\phi$. Composition of the geometric maps therefore
agrees with composition of the inducing homomorphisms.

There is a product equivalence $\mathbf B G\times \mathbf B H\simeq \mathbf B(G\times H)$.
Indeed, the product of a right $G$-torsor and a right
$H$-torsor is a right $(G\times H)$-torsor. Conversely,
a $(G\times H)$-torsor $P$ recovers the two torsors by
quotienting by the $H$- and $G$-actions respectively.
The reconstructed product identifies with $P$ after a covering
over which the torsor is trivial, hence globally by descent.
The two constructions are natural in the test topos,
so represent the same product.

\begin{proposition}
\label{prop:torsor-classifier-monoidal}
The topos $\mathbf B G$ admits a monoidal structure for the Cartesian
product of Grothendieck topoi if and only if $G$ is abelian.
For abelian $G$ the standard multiplication makes $\mathbf B G$
a group object, with all laws interpreted up to the
canonical equivalences of products.
\end{proposition}

\begin{proof}
A multiplication is classified by a homomorphism
$\mu:G\times G\to G$. The unit map $\mathbf{Set}\to \mathbf B G$ is
unique up to isomorphism, since right $G$-torsors in sets
are isomorphic to $G$. The two unit comparisons imply
that each restriction of $\mu$ to a factor is conjugate
to the identity. Both restrictions are therefore surjective.
Their images commute elementwise, since the factors of
$G\times G$ do. Thus any two elements of $G$ commute.

When $G$ is abelian, $(g,h)\mapsto gh$ is a homomorphism.
Its associativity and unit equalities induce the required
coherences by restriction of actions. The inverse
homomorphism $g\mapsto g^{-1}$ similarly supplies inversion.
\end{proof}

There are actions of such group topoi that are visible on
automorphisms of points. Let $A$ be abelian and let
$\lambda:A\to Z(H)$ be a homomorphism into the center of
an arbitrary group $H$. The map
\[
 A\times H\longrightarrow H,\qquad
 (a,h)\longmapsto\lambda(a)h
\]
is a homomorphism. It gives a left action $\mathbf B A\times \mathbf B H\to \mathbf B H$
because the unit and associativity equalities hold on these
homomorphisms. Conversely, an action morphism with invertible
unit comparison is represented by a homomorphism
$\phi:A\times H\to H$ whose restriction to $H$ is conjugate
to the identity. After conjugating the whole homomorphism
this restriction is the identity, so $\phi(a,1)$ must lie
in $Z(H)$. This recovers $\lambda$ at the level of action
morphisms. Equivalences that also keep track of all chosen
coherence data require their own comparison.

Taking geometric points of this action, each of the three
classifying topoi involved has just one isomorphism class
of point. But after choosing the regular torsors as
representatives, the action on their automorphisms is the
displayed map $(a,h)\mapsto\lambda(a)h$. In particular, $\lambda$
can be read off by setting $h=1$. The groupoid of points thus
retains information lost on passing to mere isomorphism
classes, even for these atomic topoi.
\par
\section{Internal actions in an \texorpdfstring{$\infty$}{infinity}-topos}
\label{sec:higher}

An internal group can act on an object of an
$\infty$-topos. In this setting the quotient is again
an object, and the category of equivariant objects
is itself a slice $\infty$-topos. The construction
uses effective groupoids and descent for slices
\cite[{\S}6.1.3]{htt}. The resulting interpretation
of actions and torsors is developed in
\cite[{\S}{\S}3--4.1]{nss}.

\subsection{Action groupoids and effective descent}

Let $\mathcal X$ be an $\infty$-topos. Present a
group object $G$ by a reduced simplicial groupoid
$G_\bullet$ in $\mathcal X$. Thus $G_0=1$ and the
Segal maps identify $G_n$ with $G^n$, where $G=G_1$.
Write $BG=|G_\bullet|$ for the geometric realization.
Groupoid effectivity identifies $G_\bullet$ with
the \v{C}ech nerve of the effective epimorphism
$1\to BG$. In particular,
$G\simeq 1\times_{BG}1$ with its loop group structure.
The equivalence between group objects and pointed
connected objects is given by this delooping and
loop construction
\cite[Thm.~6.1.0.6 and Lemma~7.2.2.11]{htt}.

One definition of an action on an object $X$ is
a cartesian transformation $U_\bullet\to G_\bullet$
with $U_0=X$. Here cartesian means that the naturality
square for each simplicial operator is a homotopy
pullback. In particular $U_n\simeq X\times G^n$ by
using the initial vertex in $[n]$. The two vertex
maps $U_1\rightrightarrows X$ are then the projection
and the action. For an ordinary right action, for
example, these take $(x,g)$ to $x$ and $xg$.
Composition and the higher compatibility of the
action are encoded by the simplicial maps. In this
description the Segal and invertibility conditions
for $U_\bullet$ are obtained by pulling back those
for $G_\bullet$: a string of arrows over
$(g_1,\ldots,g_n)$ is determined by a starting
object. Thus $U_\bullet$ is an action groupoid.
The category $\operatorname{Act}_G(\mathcal X)$
uses all natural transformations over $G_\bullet$
as its equivariant maps.

\begin{proposition}[The slice of actions]
\label{prop:higher-slice}
The functor that sends an action groupoid to
its realized map over $BG$ gives an equivalence
\[
 \operatorname{Act}_G(\mathcal X)
     \simeq \mathcal X_{/BG}.
\]
The image of an action on $X$ has source the
homotopy quotient
$X//G=|[n]\mapsto X\times G^n|$ with the action
bar structure. Its fiber over $1\to BG$ is $X$.
\end{proposition}

\begin{proof}
For each morphism $Y\to BG$, pull back
$G_\bullet\to BG$ along it. The result is
a cartesian transformation to $G_\bullet$
with degree-zero object $Y\times_{BG}1$.
Its realization is $Y$: colimits in an
$\infty$-topos are stable under pullback.
Conversely let $U_\bullet\to G_\bullet$
be a cartesian transformation and realize
both diagrams. The cartesian descent
criterion for colimit diagrams makes the
augmented transformation
$U_\bullet\to G_\bullet$ cartesian as well
\cite[Thm.~6.1.3.9(4) and Prop.~6.1.3.10]{htt}.
In particular
\[
\begin{array}{ccc}
 U_0 & \longrightarrow & |U_\bullet|\\
 \downarrow && \downarrow\\
 1   & \longrightarrow & BG
\end{array}
\]
is a homotopy pullback, and the same holds
with $U_n$ and $G_n$ in place of $U_0$ and $1$.
These comparisons identify the original diagram
with the one pulled back from its realization.
They are natural in transformations over
$G_\bullet$, so the equivalence covers maps and
their higher coherences.
\end{proof}

By taking the slice version of descent one can
also express the result as
\[
 \mathcal X_{/BG}\simeq
 \lim_{[n]\in\Delta}\mathcal X_{/G_n}.
\]
An object of this limit is an object over each
stage together with coherent identifications along
the pullback functors. This describes the action
without requiring a chosen model of the quotient.
A geometric inverse image commutes with these
particular action constructions: it preserves
finite limits and geometric realizations, so it
carries the action groupoid, the homotopy
quotient, and the fiber square to their counterparts
for the pulled-back group and object.

\subsection{What a family of quotient points specifies}

For a parameter object $T$, a map $T\to BG$ pulls
back the universal effective epimorphism to a
$G$-torsor $P\to T$. A torsor here is an effective
epimorphism equipped with a right action such
that $(p,g)\mapsto(p,pg)$ identifies $P\times G$
with $P\times_T P$. Its action groupoid is the
\v{C}ech nerve of $P\to T$, so realization
recovers $T$ and supplies a map to $BG$.
Proposition~\ref{prop:higher-slice} makes these
constructions inverse. More generally, a map
$T\to X//G$ specifies the torsor induced by its
composite with $BG$ together with an equivariant
map $P\to X$. Indeed a lift of $T\to BG$ to
$X//G$ is a map in the slice, which under the
proposition is the asserted equivariant map.
Thus the fibers of the mapping-space map to
$\operatorname{Map}_{\mathcal X}(T,BG)$ record
equivariant maps for each torsor. This is the
family-level refinement of an orbit space.

There is also an object of homotopy fixed points.
Given $Y=X//G\to BG$, the pullback along
$q:BG\to 1$ has a right adjoint $\Pi_q$ since
$\mathcal X$ is locally cartesian closed. Its
value $\Pi_q(Y)$ represents sections of $Y$
parametrized by an arbitrary object:
\[
 \operatorname{Map}_{\mathcal X}(T,\Pi_q Y)
 \simeq \operatorname{Map}_{\mathcal X_{/BG}}
             (T\times BG,Y).
\]
Under Proposition~\ref{prop:higher-slice},
$T\times BG$ corresponds to the trivial action on
$T$, because the quotient of the trivial bar
construction is its product with $BG$.
Consequently $\Pi_q(Y)$ represents equivariant maps
out of objects carrying the trivial action. It is
denoted $X^{hG}$. The homotopy orbit object $X//G$
and the homotopy fixed-point object $X^{hG}$
therefore come from the left and right adjoints to
the trivial-action functor. Indeed, the left
adjoint to $q^*$ takes a map over $BG$ to its
domain, forgetting the structure map. In spaces
these two objects specialize to the colimit and
limit of the coherent $G$-diagram, respectively.

\begin{example}[Reflection of a circle]
In spaces let $C_2$ act on $S^1$ by complex
conjugation. The coarse orbit space is an
interval. The homotopy quotient has a
fibration sequence
$S^1\to S^1//C_2\to BC_2$. It admits a section
because $1\in S^1$ is a fixed point.
The fundamental group sequence is a split
extension of $C_2$ by $\mathbb Z$, using
$\pi_2(BC_2)=0$. The nonidentity element
acts on $\mathbb Z$ by inversion since
conjugation reverses the orientation of
$S^1$. It follows that
\[
 S^1//C_2\simeq B(\mathbb Z\rtimes C_2).
\]
Here both the base and the fiber are
aspherical, so the long exact homotopy
sequence makes the total space aspherical
as well. The quotient has infinite
dihedral fundamental group, a distinction
invisible to the interval of orbits.
\end{example}

\subsection{Comparison with external symmetry}

An external action on an $\infty$-topos assigns
geometric autoequivalences to elements of a group
and coherence to their composition. The associated
category of equivariant objects is the limit of
the resulting diagram of categories of inverse
images. The action groupoid above
uses objects and arrows internal to a single
$\mathcal X$. The two constructions meet for a
trivial external action of a discrete group $H$.
Use the constant internal group
$\underline H=\coprod_{h\in H}1$. By disjoint
universal coproducts,
$\mathcal X_{/\underline H^n}\simeq
\prod_{H^n}\mathcal X$. Slice descent then gives
$\mathcal X_{/B\underline H}\simeq
\operatorname{Fun}(BH,\mathcal X)$:
the cosimplicial limit of these products imposes
the coherent action along the nerve of $H$.
This identification supplies an underlying
internal object group for that particular
category of equivariant objects.
\par
\section{Isotropy actions and crossed topoi}
\label{sec:isotropy}

A topos carries an intrinsic source of symmetry.
For a Grothendieck topos \(\mathcal E\) and an object
\(U\), write \(p_U:\mathcal E/U\to\mathcal E\) for
the slice projection, with inverse image
\(A\mapsto A\times U\to U\). Funk, Hofstra, and
Steinberg prove that the presheaf of groups
\[
 U\longmapsto \operatorname{Aut}(p_U^*)
\]
is represented by a group object \(Z_{\mathcal E}\)
\cite[{\S}4.1]{fhs-isotropy}. The restrictions of
this presheaf pull natural automorphisms back along
maps of slices. The identity of \(Z_{\mathcal E}\)
therefore gives a natural automorphism of
\(p_{Z_{\mathcal E}}^*\). Reading its component at \(A\)
over \(Z_{\mathcal E}\) gives the canonical action
\(Z_{\mathcal E}\times A\to A\), with the order of
factors chosen for left actions and multiplication
given by composition. Every morphism of
\(\mathcal E\) intertwines these actions by
naturality. The action on the group object itself
is conjugation.

On a fixed base \(\mathcal E\), a \emph{crossed topos}
of Funk, Hofstra, and Steinberg is described by an
internal crossed module \(\partial:K\to Z_{\mathcal E}\)
using the canonical \(Z_{\mathcal E}\)-action on \(K\).
Equivariance of \(\partial\) follows from naturality,
and the Peiffer identity says that \(\partial(k)\) acts
on \(K\) by conjugation by \(k\). These are the internal
conditions for algebras of their isotropy monad on
grouped topoi \cite[{\S}2 and Cor.~5.10]{fhs-isotropy}.
The action induced on each \(A\) restricts the canonical
one along \(\partial\). Objects on which this induced
action is trivial form a topos whose inclusion in
\(\mathcal E\) is the inverse image of a connected atomic
isotropy quotient \cite[Prop.~4.17]{fhs-isotropy}.

\begin{example}[Isotropy of the topos of \(G\)-sets]
For a discrete group \(G\) and \(\mathcal E=\mathbf B G\),
the group \(Z_{\mathcal E}\) is \(G\) with its conjugation
action. Indeed, writing a component of an automorphism
of \(p_U^*\) as \((x,u)\mapsto(\alpha_X(x,u),u)\),
put \(z_u=\alpha_G(1,u)\) on the left regular \(G\)-set.
Naturality along \(G\to X,\ g\mapsto gx\) forces
\(\alpha_X(x,u)=z_u x\).
Equivariance of the components gives
\(z_{hu}=hz_uh^{-1}\).
Conversely this condition makes the displayed
formula an equivariant natural automorphism. Global
sections of the internal isotropy group form the centre
of \(G\), which can be trivial even for a nontrivial
\(G\). The quotient by the full isotropy action is
\(\mathbf{Set}\), since its objects are the sets with
trivial \(G\)-action.
\end{example}

Henry's localic isotropy \(I_{\mathcal E}\to\mathcal E\)
extends the construction to all geometric probes
\(f:\mathcal F\to\mathcal E\). Its group of sections after
base change along \(f\) is \(\operatorname{Aut}(f^*)\).
As a topos over \(\mathcal E\), \(I_{\mathcal E}\)
is the pseudopullback of the diagonal of \(\mathcal E\)
along itself. A test morphism into this pullback gives
two morphisms to \(\mathcal E\) and two invertible
comparisons between them. Composing one comparison
with the inverse of the other gives a morphism and
its automorphism. Composition gives the group law
over \(\mathcal E\), and \(Z_{\mathcal E}\) is its
internal group of points
\cite[Props.~2.2--2.3 and 6.1]{henry-isotropy}.
This group acts on every topos \(Y\to\mathcal E\) in the
pseudo-slice. On test morphisms it changes the specified
comparison between a map through \(Y\) and the map to
\(\mathcal E\) by composing with the isotropy
automorphism \cite[Prop.~2.5]{henry-isotropy}.

For a locally positive (overt) internal localic group
\(K\) and an isotropy action \(K\to I_{\mathcal E}\), the
quotient \(\mathcal E\to\mathcal E_K\) has inverse image
the inclusion of objects with trivial induced action
and is connected atomic. Conversely every connected
atomic morphism \(f:\mathcal E\to\mathcal F\) is such a
quotient by the relative isotropy \(I_{\mathcal E/\mathcal F}\),
which consists of automorphisms becoming trivial over
\(\mathcal F\) \cite[Props.~5.4 and 5.9]{henry-isotropy}.
Local positivity is a hypothesis of this atomicity
assertion. Arbitrary localic isotropy quotients give
hyperconnected morphisms
\cite[{\S}4]{henry-isotropy}.
\par
\section{Comparing ordinary and higher quotients}
\label{sec:comparison}

The central action of Section~\ref{sec:module-examples}
admits a direct quotient calculation at both levels. Let
$A$ be abelian, $H$ a discrete group, and
$\chi:A\to Z(H)$ a homomorphism into its center.
Write $\mathbf B H$ for the topos of left $H$-sets,
and $BH$ for the pointed connected classifying space.
At the topos level the action of $\mathbf B A$
on $\mathbf B H$ is induced by the group homomorphism
$(a,h)\mapsto\chi(a)h$. At the space level the same
homomorphism gives an action of the group object
$BA$ on $BH$. One can take the usual nerves of the
one-object groupoids, in which products and the action
are given levelwise by the group laws.

\begin{proposition}[A central-action quotient]
\label{prop:central-codescent}
With invertible comparison cells, the geometric quotient
of $\mathbf B H$ by $\mathbf B A$ is
$\mathbf B(H/\chi(A))$. The quotient space $BH//BA$
is connected and has homotopy groups
\[
 \pi_1(BH//BA)\simeq H/\chi(A),\qquad
 \pi_2(BH//BA)\simeq\ker\chi .
\]
Its homotopy groups in degrees greater than two vanish.
\end{proposition}

\begin{proof}
Put $K=H/\chi(A)$. A geometric morphism
$f:\mathbf B H\to\mathcal Y$ has inverse image
$F:\mathcal Y\to H\text{-}\mathrm{Set}$. A comparison
between the action and projection on this functor
is an isomorphism between two functors into
$(A\times H)\text{-}\mathrm{Set}$, given by restricting
the $H$-actions along $(a,h)\mapsto\chi(a)h$ and
$(a,h)\mapsto h$ respectively. On the second functor
$A$ acts trivially. An equivariant isomorphism can
exist only if $\chi(A)$ acts trivially on each $F(Y)$.
The unit condition forces the comparison restricted
to $H$ to be the identity. That restriction is
faithful on morphisms, so the only possible
comparison is the identity. It satisfies the
composition condition whenever $\chi(A)$ acts
trivially.

Thus a cocone is precisely a left exact cocontinuous
$F$ taking values in the full subcategory of
$H$-sets on which $\chi(A)$ acts trivially.
This subcategory is $K\text{-}\mathrm{Set}$, and its
inclusion creates finite limits and small colimits.
A transformation of cocones is exactly a
transformation of the resulting functors into it.
This gives the geometric universal property
of $\mathbf B K$.

For the higher calculation, the bar description gives
a fiber sequence $BH\to BH//BA\to B^2A$.
The boundary $A=\pi_2(B^2A)\to\pi_1(BH)=H$ in its
long exact sequence is $\chi$ up to the choice of
boundary sign. Indeed, the orbit map $BA\to BH$
of the base point is $B\chi$ and is equivariant for
translation acting on $BA$. It induces a map over
$B^2A$ from $BA//BA\simeq *$ to $BH//BA$.
The connecting homomorphism for the first fiber
sequence identifies $\pi_2(B^2A)$ with $\pi_1(BA)$.
Naturality gives the stated boundary in the second.
Exactness yields the displayed groups, since $BH$
is a $1$-type and $B^2A$ is a $2$-type.
\end{proof}
\balance

If $\chi$ is trivial, the geometric quotient is
$\mathbf B H$. The space construction gives
$BH\times B^2A$. In particular, for $H$ trivial it
still retains $A$ in degree two. This example
compares a specific codescent construction in
Grothendieck topoi with a homotopy quotient in
spaces. Each universal property uses its own
category of test objects.

\section{Which description to use}

A monoidal topos acting by geometric morphisms can
be evaluated on all test topoi. Discrete parameter
topoi specialize to families of geometric
automorphisms, and the topos of equivariant
objects realizes their geometric quotient.
Internal actions have a fixed base topos and are
represented by relative classifying topoi or
their slices. They retain torsors and local
transport. Isotropy supplies a canonical symmetry
attached to a topos itself. Passing to higher
topoi replaces ordinary descent isomorphisms
by coherent groupoid objects and permits higher
homotopy in the quotient. The descriptions agree
where their hypotheses identify the same descent
data, as with ordinary discrete groups, and the
central-action calculation specifies one place
where their invariants differ.
\par
\begin{thebibliography}{99}
\small\raggedright

\bibitem{cassou-torsors}
P. Cassou-Nogu\`es, T. Chinburg, B. Morin, and M. J. Taylor.
The classifying topos of a group scheme and invariants of symmetric bundles.
\emph{Proc. London Math. Soc.} 109 (2014), 1093--1136.
\href{https://doi.org/10.1112/plms/pdu017}{DOI}.

\bibitem{fhs-isotropy}
J. Funk, P. Hofstra, and B. Steinberg.
Isotropy and crossed toposes.
\emph{Theory Appl. Categ.} 26(24) (2012), 660--709.
\href{https://www.tac.mta.ca/tac/volumes/26/24/26-24abs.html}{Journal article}.

\bibitem{henry-isotropy}
S. Henry. The localic isotropy group of a topos.
\emph{Theory Appl. Categ.} 33(41) (2018), 1318--1345.
\href{https://www.tac.mta.ca/tac/volumes/33/41/33-41abs.html}{Journal article}.

\bibitem{johnstone-pointless}
P. T. Johnstone. The point of pointless topology.
\emph{Bull. Amer. Math. Soc.} 8 (1983), 41--53.
\href{https://doi.org/10.1090/S0273-0979-1983-15080-2}{DOI}.

\bibitem{johnstone-joyal-exponential}
P. T. Johnstone and A. Joyal.
Continuous categories and exponentiable toposes.
\emph{J. Pure Appl. Algebra} 25 (1982), 255--296.
\href{https://doi.org/10.1016/0022-4049(82)90083-4}{DOI}.

\bibitem{joyal-tierney}
A. Joyal and M. Tierney.
\emph{An Extension of the Galois Theory of Grothendieck}.
Memoirs Amer. Math. Soc. 51(309), 1984.
\href{https://doi.org/10.1090/memo/0309}{DOI}.

\bibitem{htt}
J. Lurie. \emph{Higher Topos Theory}.
Annals of Mathematics Studies 170. Princeton University Press, 2009.
\href{https://www.math.ias.edu/~lurie/papers/HTT.pdf}{Author's copy}.

\bibitem{maclane-moerdijk}
S. Mac Lane and I. Moerdijk.
\emph{Sheaves in Geometry and Logic: A First Introduction to Topos Theory}.
Springer, 1992.
\href{https://doi.org/10.1007/978-1-4612-0927-0}{DOI}.

\bibitem{moerdijk-groupoid}
I. Moerdijk. The classifying topos of a continuous groupoid. I.
\emph{Trans. Amer. Math. Soc.} 310 (1988), 629--668.
\href{https://doi.org/10.1090/S0002-9947-1988-0973173-9}{DOI}.

\bibitem{moerdijk-classifying}
I. Moerdijk. \emph{Classifying Spaces and Classifying Topoi}.
Lecture Notes in Mathematics 1616. Springer, 1995.
\href{https://doi.org/10.1007/BFb0094441}{DOI}.

\bibitem{nss}
T. Nikolaus, U. Schreiber, and D. Stevenson.
Principal $\infty$-bundles: general theory.
\emph{J. Homotopy Relat. Struct.} 10 (2015), 749--801.
\href{https://doi.org/10.1007/s40062-014-0083-6}{DOI}.

\bibitem{vickers-double}
S. Vickers.
The double powerlocale and exponentiation: A case study in geometric logic.
\emph{Theory Appl. Categ.} 12(13) (2004), 372--422.
\href{https://www.tac.mta.ca/tac/volumes/12/13/12-13abs.html}{Journal article}.

\end{thebibliography}
\end{document}

End of manuscript · Back to top ↑