Independent scholarship · Open accessThe monthly digest

Full manuscript · browser edition

Galois Theory in Higher Categories

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

Contents · 96 sections

RESEARCH MONOGRAPH

Galois Theory
in Higher Categories

Descent, monodromy, and higher topoi

Manuscript for Higher Categorical Semantics

November 2026

Preface

The ordinary expression L⊗KLL\otimes_K L in Galois theory serves two purposes. It describes the overlap of the extension with itself, and, when splitting holds, it encodes the symmetries by which copies of a descended object are identified. The higher analogue is the entire simplicial object of overlaps. Its realization recovers the base under effective descent. Its Segal operators describe multiplication in a groupoid, internal composition in a category, or a monad on a category of objects over a splitting base. These three conclusions require different hypotheses. Much of the work below is to identify them and prove the comparisons between their outputs.

The point of departure is Francis Borceux and George Janelidze’s Galois Theories [1]. Their sequence of generalizations passes from subgroup correspondences to split algebras, profinite symmetry, Pierce spectra, relative reflections, coverings, and generalized descent. This monograph follows those questions through the passage from ordinary categories to (∞,1)(\infty,1)-categories. The organizing results are homotopy-coherent descent in an ∞\infty-topos and the classification of families by transport. They are joined by versions in stable and directed settings where the relevant additional structure survives.

Reading the book

Part I develops the correspondence between families and actions and proves the restriction of effective descent to chosen coefficients. Part II treats continuity of monodromy, the shape of a higher topos, and obstructions to lifting actions through truncations. Part III compares Pierce descent, connective spectral étale descent, and Galois extensions in stable homotopy theory. Part IV studies admissible reflections in slices, factorization, and centrality. Part V turns from invertible transport to right and left fibrations, directed descent, and exodromy. Part VI returns to toposes themselves and compares internal, geometric, and external presentation theorems.

The correspondence with the original book is thematic. Its Chapters 1–2 are treated in Part I, its Chapter 3 in Part II, and its Chapter 4 and appendix in Part III. Its Chapter 5 chiefly informs Part IV. Its Chapter 6 is distributed over Parts I, II, and IV, where the notions of covering, connectedness, and local triviality can be stated independently. Its Chapter 7 motivates Parts V and VI. The exposition assumes familiarity with adjunctions, slices, and elementary homotopy theory. It gives universe and variance conventions before using any classification theorem. Deep inputs receive references at their points of use.

Abstract

We develop higher analogues of the categorical Galois theories of Borceux and Janelidze. Effective descent in an ∞\infty-topos yields actions of Čech groupoids and classifies torsors by delooping. Pro-spaces encode continuous monodromy under coefficient hypotheses, while Postnikov extensions give obstructions to lifting actions. We compare Pierce and spectral descent, admissible reflections and relative factorization, directed transport and exodromy, and groupoid presentations of topoi. The ordinary and higher statements are connected by specified truncations, duality, and base-change comparisons.

Manuscript disclosure

This manuscript is an independent mathematical treatment with the source book explicitly cited. Generative AI assisted with drafting, bibliography research, mathematical consistency checks, and typesetting. The journal submission metadata provides the responsible author or research group.

Torsors and descent

The language of coherent descent

In its simplest form a Galois correspondence compares subobjects of a given extension with subgroups of its symmetries. A stronger statement recovers the category of objects that split after a base change from trivial objects on that base and their gluing data. When gluing takes place in an ∞\infty-category, a gluing map comes with compatibilities among homotopies and coherences among those compatibilities. A simplicial overlap diagram organizes this information without requiring a separate definition at each level.

Conventions and the ambient category

Fix a universe of small spaces and write S\mathcal{S} for their ∞\infty-category. Functor categories, categories of large coefficients, and pro-categories are formed in larger universes when needed. A space means a homotopy type. An ∞\infty-category means an (∞,1)(\infty,1)-category, so the morphisms between two given objects form a space Map(−,−)\mathop{\mathrm{Map}}(-,-). For a topological space TT, we distinguish its topology, its singular homotopy type, and the ∞\infty-topos of sheaves on its open-set site. These inputs agree on certain classifications under the local hypotheses given below.

Limits, colimits, slices, and fibers are understood in the higher categorical sense. The space MapCat∞(C,D)=Fun(C,D)≃\mathop{\mathrm{Map}}_{\mathrm{Cat}_{\infty}}(\mathcal C,\mathcal D)= \mathop{\mathrm{Fun}}(\mathcal C,\mathcal D)^{\simeq} is the maximal subgroupoid of the functor category. The functor category itself uses all natural transformations, an important distinction when we study geometric morphisms in Chapter [ch:geometric].

Definition 1.1. An ∞\infty-topos is an accessible left exact localization of a presheaf ∞\infty-category Fun(Cop,S)\mathop{\mathrm{Fun}}(C^{\mathrm{op}},\mathcal{S}) for a small ∞\infty-category CC. A geometric morphism f:Y→Xf:\mathcal Y\to\mathcal X is specified by its inverse-image functor f∗:X→Yf^*:\mathcal X\to\mathcal Y, which preserves colimits and finite limits. Its right adjoint is denoted f∗f_*.

A geometric morphism points in the opposite direction from its inverse image. For a map p:U→Bp:U\to B inside X\mathcal{X}, the same notation p∗p^* means base change from X/B\mathcal{X}_{/B} to X/U\mathcal{X}_{/U}. The left adjoint p!p_! composes a map with pp. The right adjoint p∗p_* is dependent product. Its existence follows from local cartesian closure. If pp is a map of spaces and V→UV\to U is a family, the fiber of p!Vp_!V over bb is the total space over the homotopy fiber UbU_b. The fiber of p∗Vp_*V over bb is the space of sections over UbU_b.

Lemma 1.2 (Mapping from a section). Given q:E→Bq:E\to B and a:A→Ba:A\to B in X\mathcal{X}, one has MapX/B(A,E)≃fib⁡a(MapX(A,E)→q∘−MapX(A,B)).\mathop{\mathrm{Map}}_{\mathcal{X}_{/B}}(A,E) \simeq \operatorname{fib}_{a}\bigl(\mathop{\mathrm{Map}}_{\mathcal{X}}(A,E) \xrightarrow{q\circ-}\mathop{\mathrm{Map}}_{\mathcal{X}}(A,B)\bigr). This is also the space of sections of a∗E→Aa^*E\to A.

Proof. A point of the fiber consists of a map to EE and a homotopy identifying its composite with aa. These are the data in the slice mapping space, with the same assertion for families parametrized by a test space. The pullback universal property rewrites that homotopy fiber as maps from the terminal object of X/A\mathcal{X}_{/A} to a∗Ea^*E. ◻

If FF is a fixed fiber, its space of automorphisms Aut(F)\mathop{\mathrm{Aut}}(F) consists of the equivalence components of Map(F,F)\mathop{\mathrm{Map}}(F,F). Composition makes it a group object in spaces. Its group of components π0Aut(F)\pi_0\mathop{\mathrm{Aut}}(F) classifies automorphisms up to homotopy. A family with fiber FF can depend on homotopies within a single component of Aut(F)\mathop{\mathrm{Aut}}(F), even when the group of components of the automorphism object acts trivially.

Cartesianness and realizations

A cartesian family over a diagram U:I→XU:I\to\mathcal{X} is a natural transformation V→UV\to U whose square over every arrow is a pullback. Let B=colimIUB=\mathop{\mathrm{colim}}_I U. Pullback defines a comparison X/B⟶lim⁡i∈IopX/Ui. \mathcal{X}_{/B}\longrightarrow\lim_{i\in I^{\mathrm{op}}}\mathcal{X}_{/U_i}. The target consists of cartesian families. The identifications over arrows and their higher coherences are part of the data. A colimit for which this comparison is an equivalence is called van Kampen.

Proposition 1.3 (A test for descent of a diagram). Suppose the ambient category admits pullbacks and the required colimits, and these colimits are preserved by base change. The comparison (1) is fully faithful. It is an equivalence exactly when for each cartesian V→UV\to U, the squares image are pullbacks.

Proof. Realization takes VV to colimIV→B\mathop{\mathrm{colim}}_I V\to B. For any W→BW\to B, maps from the realization to WW are the limit of maps from Vi→BV_i\to B to WW, by the universal property of the colimit in the slice. Base change rewrites the latter mapping spaces as maps from Vi→UiV_i\to U_i to W×BUiW\times_B U_i. Their limit is the space of maps of cartesian families into the comparison image of WW. Realization is therefore left adjoint to the comparison. The counit colimI(W×BUi)→W\mathop{\mathrm{colim}}_I(W\times_B U_i)\to W is an equivalence by the base-change assumption, so the right adjoint is fully faithful. For each VV, the unit is an equivalence exactly when its components Vi→Ui×BcolimjVjV_i\to U_i\times_B\mathop{\mathrm{colim}}_j V_j are equivalences. ◻

A candidate over the base is determined by its restrictions as soon as the comparison is fully faithful. Existence of a candidate for every cartesian family requires the unit condition. Each covering theorem below uses a specific argument for existence.

Theorem 1.4 (Van Kampen property of higher topoi). Small colimits in an ∞\infty-topos are van Kampen.

Proof. In spaces, straightening over a space KK gives S/K≃Fun(K,S)\mathcal{S}_{/K}\simeq\mathop{\mathrm{Fun}}(K,\mathcal{S}), covariantly for a family viewed by transport. The inclusion of spaces into Cat∞\mathrm{Cat}_{\infty} preserves colimits because it is left adjoint to the maximal subgroupoid functor. The internal-hom universal property turns colimits of bases into limits of the functor categories. In a presheaf ∞\infty-category, the two criteria in Proposition 1.3 then hold objectwise.

Write L:P⇄X:iL:\mathcal P\rightleftarrows\mathcal{X}:i for a left exact localization. A diagram of local objects and its cartesian families can be viewed in P\mathcal P, since the inclusion preserves limits. Realizing there gives pullback squares as in Proposition 1.3. Applying LL preserves those pullbacks and gives their realizations in X\mathcal{X}. Universality of colimits in X\mathcal{X} follows by the same localization argument. If BPB_P is the colimit of the diagram in P\mathcal P, pull back that colimit along iW×iBBP→BPiW\times_{iB}B_P\to B_P and apply LL. This identifies WW as the colimit of the diagram base changed along W→BW\to B. The two criteria now apply in X\mathcal{X}. The straightening input is developed in [2, Ch. 2]. ◻

The overlap diagram

The Cˇech\text{\v{C}ech} nerve of p:U→Bp:U\to B has kk-simplices Uk=U×B⋯×BU⏟k+1.U_k=\underbrace{U\times_B\cdots\times_B U}_{k+1}. Its operators omit or repeat factors. It is a groupoid object because the edges encode pairs of local lifts and the inverse operation reverses a pair. The map pp is an effective epimorphism if the augmentation U∙→BU_\bullet\to B exhibits BB as its realization. In spaces this amounts to surjectivity on components. In a topos of sheaves it expresses local existence, so a section over the full base is unnecessary.

Corollary 1.5 (Effective descent over one cover). For an effective epimorphism p:U→Bp:U\to B in X\mathcal{X}, the natural comparison is an equivalence X/B≃Tot([k]⟼X/Uk).\mathcal{X}_{/B}\simeq\mathop{\mathrm{Tot}}\bigl([k]\longmapsto \mathcal{X}_{/U_k}\bigr). Moreover p∗p^* is conservative.

Proof. Apply Theorem 1.4 to the Cˇech\text{\v{C}ech} nerve. For a morphism in the limit whose degree-zero component is an equivalence, each degree-kk component is an equivalence by restriction along a vertex map [0]→[k][0]\to[k]. Equivalences in the limit are detected componentwise. ◻

A stable category or a category of algebras can require a different proof. Theorem 1.4 concerns slices of an ∞\infty-topos. A comparison for modules over a spectral ring computes a totalization of module categories using scalar extension. Its conservativity and continuity require hypotheses on that extension. Chapter [ch:spectral] gives such hypotheses.

Restricting coefficients

Let D∙D^\bullet be a cosimplicial diagram of ∞\infty-categories. Let EkE^k be replete full subcategories of DkD^k, preserved by all structure functors.

Lemma 1.6 (Restricting full coefficients). The functor TotE∙→TotD∙\mathop{\mathrm{Tot}}E^\bullet\to\mathop{\mathrm{Tot}}D^\bullet is fully faithful. An object of TotD∙\mathop{\mathrm{Tot}}D^\bullet lies in its essential image exactly when its degree-zero component lies in E0E^0.

Proof. For any pair of compatible families, the mapping space is the limit of the corresponding levelwise mapping-space diagram. Fullness makes the two diagrams equivalent in every degree. The essential image consists of objects whose components lie in the replete full subcategories. Each structure equivalence and coherence of the given diagram belongs to the full subcategory as soon as its vertices do. Applying a vertex map [0]→[k][0]\to[k] to the degree-zero component puts the degree-kk component in EkE^k. ◻

In applications one must establish the unrestricted comparison and then check stability of the coefficients or construct a comparison on which stability holds. In algebraic Galois theory the second step can follow from a relative left exactness condition. In stratified geometry the comparison must respect which strata are connected by directed arrows. An equivalence of cores classifies objects and their equivalences. We use a comparison of full categories when morphisms between the coefficients also matter.

Torsors, monodromy, and intermediate objects

The first two chapters of Borceux and Janelidze pass from intermediate fields to a category of split algebras. The Galois group acts on a set of embeddings. Inside the resulting action category, the quotients of the regular action recover the intermediate fields, with arrows reversed by the passage from algebras to actions [1, Chapters 1–2]. The argument suggests two distinct operations. One classifies objects trivialized by a cover. The other identifies which pointed intermediate objects are governed by subgroups. In an ∞\infty-category these operations have different answers unless a truncation condition is imposed.

Pointed connected objects

Write S\mathcal S for the ∞\infty-category of spaces and X\mathcal X for an ∞\infty-topos. Pullbacks, colimits, and fibers are taken in the ∞\infty-categorical sense. We write 11 for a terminal object and call a map 1→B1\to B a point of BB internal to X\mathcal X. This means a section of an object in the topos. A geometric morphism S→X\mathcal S\to\mathcal X is a different operation.

A simplicial object D∙D_\bullet is a groupoid object if its Segal maps are equivalences and the arrows are invertible in the internal sense. For example, the Čech nerve of a map q:U→Bq:U\to B is the groupoid object with Cˇ(q)n=U×B⋯×BU⏟n+1 factors.\check C(q)_n=\underbrace{U\times_B\cdots\times_B U}_{n+1\ \mathrm{factors}}. Its face maps delete vertices and its degeneracies repeat vertices. The map qq is an effective epimorphism if the augmentation ∣Cˇ(q)∙∣→B|\check C(q)_\bullet|\to B is an equivalence. Effectivity of groupoids in an ∞\infty-topos says that for every groupoid object D∙D_\bullet, the canonical map D0→∣D∙∣D_0\to |D_\bullet| is an effective epimorphism and its Čech nerve identifies with D∙D_\bullet [2, Section 6.1.3].

A group object is a groupoid object G∙G_\bullet with terminal object of objects. Its arrow object is denoted by G=G1G=G_1. In degree nn the Segal maps identify GnG_n with GnG^n. The two ways of composing three arrows agree through the simplicial structure, which also supplies all higher coherences. The delooping is BG=∣G∙∣BG=|G_\bullet|, pointed by the degree-zero inclusion. A pointed object b:1→Bb:1\to B is connected if bb is an effective epimorphism. In this situation, ΩbB=1×B1\Omega_b B=1\times_B 1 is the based loop group object, with simplicial description Cˇ(b)∙\check C(b)_\bullet. The definition of connectedness here is equivalent to τ≤0B≃1\tau_{\leq 0}B\simeq 1. In fact a subobject of BB is classified by a map to the subobject classifier, which is a 00-truncated object. If τ≤0B≃1\tau_{\leq 0}B\simeq 1, any subobject containing bb is therefore all of BB, so the image of bb is all of BB. Conversely, if bb is an effective epimorphism, a map from BB to a 00-truncated object is determined by its restriction along bb. Every map from 11 to that object extends along B→1B\to 1. The resulting universal property identifies τ≤0B\tau_{\leq 0}B with 11.

Proposition 1.1 (Loop and delooping). Taking loops and taking delooping define inverse equivalences between the ∞\infty-category of group objects in X\mathcal X and the ∞\infty-category of pointed connected objects of X\mathcal X.

Proof. A map of group objects induces a pointed map on geometric realizations. A pointed map induces a map on Čech nerves of its points. For a pointed connected object the augmentation of its Čech nerve realizes to the object. For a group object its simplicial groupoid is the Čech nerve of the point of its realization by effectivity. Both identifications are canonical maps of diagrams. They give unit and counit equivalences of the two functors, including on mapping spaces. ◻

These operations also work in every slice X/B\mathcal X_{/B}. The terminal object in that slice is the identity of BB. A point is a section, and the corresponding notion of connectedness says that the section is an effective epimorphism. Slicing is necessary when a group varies with parameters. A single external group of global sections need not recover an internal group object. The family of mapping spaces Map⁡X(T,G)\operatorname{Map}_{\mathcal X}(T,G), with its action by change of TT, does recover it by the ∞\infty-categorical Yoneda lemma.

Actions and classifying maps

A left GG-object can be presented as a simplicial object E∙E_\bullet over G∙G_\bullet such that restriction to the initial vertex induces En→ ∼ E0×Gn.E_n\xrightarrow{\ \sim\ }E_0\times G^n . The map to the final vertex at n=1n=1 supplies the action. With this orientation, successive arrows g,hg,h act by h⋅(g⋅x)=(hg)⋅xh\cdot(g\cdot x)=(hg)\cdot x. Here and below formulas in elements specify maps of objects and their coherent simplicial identities. The category of actions is the full subcategory of the category of simplicial objects over G∙G_\bullet specified by the displayed pullback property. In particular its morphisms are coherent equivariant morphisms. The quotient E0//GE_0//G means ∣E∙∣|E_\bullet|.

Theorem 1.2 (Actions as objects over a delooping). The functor sending E∙→G∙E_\bullet\to G_\bullet to ∣E∙∣→BG|E_\bullet|\to BG is an equivalence Act⁡G(X)≃X/BG.\operatorname{Act}_{G}(\mathcal X)\simeq \mathcal X_{/BG}. Its inverse sends Y→BGY\to BG to the action diagram G∙×BGYG_\bullet\times_{BG}Y.

Proof. For a groupoid, cartesianness over the initial vertex implies cartesianness over every vertex. In degree one, transport to the other vertex has inverse induced by inversion. In higher degrees the Segal maps express transport along a string of arrows as successive transport in degree one. The resulting squares are cartesian for all simplex operators. The Segal pullbacks for G∙G_\bullet, together with the cartesian vertex squares, then give the Segal pullbacks for E∙E_\bullet. Invertibility is inherited from the base groupoid by transport. Thus E∙E_\bullet is a groupoid object. Put Y=∣E∙∣Y=|E_\bullet|. Effectivity gives E1≃E0×YE0E_1\simeq E_0\times_Y E_0 and G1≃1×BG1G_1\simeq 1\times_{BG}1. The square E0⟶1↓↓Y⟶BG\begin{array}{ccc} E_0&\longrightarrow&1\\ \downarrow&&\downarrow\\ Y&\longrightarrow&BG \end{array} is cartesian. To see this, test the comparison E0→Y×BG1E_0\to Y\times_{BG}1 after pullback along the effective epimorphism E0→YE_0\to Y. The resulting map is E0×YE0⟶E0×BG1.E_0\times_Y E_0\longrightarrow E_0\times_{BG}1. Under the identifications above it is exactly the degree-one vertex equivalence E1≃E0×G1E_1\simeq E_0\times G_1, using the first projection to E0E_0. Equivalences can be tested after an effective epimorphism, as follows by reconstructing both source and target by the base changes of its Čech nerve and realizing. The vertex equivalence now gives E∙≃G∙×BGYE_\bullet\simeq G_\bullet\times_{BG}Y in all degrees. In the other direction, universality of colimits implies ∣G∙×BGY∣≃Y|G_\bullet\times_{BG}Y|\simeq Y. All comparisons are natural in the diagrams, so they give inverse equivalences of ∞\infty-categories. ◻

A GG-torsor over TT is an action in X/T\mathcal X_{/T} on an object P→TP\to T such that P→TP\to T is an effective epimorphism and the transport map P×G⟶P×TP,(x,g)⟼(x,g⋅x)P\times G\longrightarrow P\times_T P, \qquad (x,g)\longmapsto(x,g\cdot x) is an equivalence over TT. The universal torsor has underlying object 1→BG1\to BG. Its action is specified by the identification of its Čech groupoid with the bar groupoid of GG. Conventions using right actions give the same classifier via inversion.

Corollary 1.3 (Torsor classification). The space of GG-torsors over TT, with equivariant equivalences as arrows, is naturally equivalent to Map⁡X(T,BG)\operatorname{Map}_{\mathcal X}(T,BG). Pullback in TT corresponds to precomposition with the classifying map.

Proof. The classifying equivalence of Theorem 1.2 in X/T\mathcal X_{/T} sends an action on PP to an object P//G→T×BGP//G\to T\times BG over TT. The comparison from its action groupoid to the Čech nerve of P→TP\to T is an equivalence in degree one by the torsor condition. The two Segal properties make it an equivalence in every degree. Since P→TP\to T is effective, this means P//G≃TP//G\simeq T over TT. Conversely, if P//G≃TP//G\simeq T, effectivity identifies the action groupoid with the Čech nerve of P→TP\to T, giving both torsor conditions. The core of objects Y→T×BGY\to T\times BG over TT with Y→TY\to T an equivalence is Map⁡X(T,BG)\operatorname{Map}_{\mathcal X}(T,BG), by replacing the source by TT. This identification and the resulting pullback of the universal torsor commute with base change. ◻

This formulation retains automorphisms of a torsor and their higher homotopies. At a classifying map c:T→BGc:T\to BG, the loop space of the mapping space in Corollary 1.3 is the space of equivariant automorphisms of the classified torsor. Choosing a section of the torsor identifies it with the trivial torsor through the transport equivalence. Changing the section can change that identification. For an ordinary nonabelian group the induced change is conjugation. For an internal abelian group AA with Eilenberg–Mac Lane deloopings, the same corollary applied to G=K(A,m)G=K(A,m) identifies torsors with maps to K(A,m+1)K(A,m+1). With the convention Hi(X,A)=π0Map⁡X(1,K(A,i))H^i(\mathcal X,A)=\pi_0\operatorname{Map}_{\mathcal X}(1,K(A,i)), global isomorphism classes form Hm+1(X,A)H^{m+1}(\mathcal X,A). The based homotopy groups of the torsor space at the trivial class recover lower degrees, because looping mapping spaces commutes with looping the target.

Transport over a space

For a space BB there is a canonical equivalence S/B≃Fun⁡(B,S). \mathcal S_{/B}\simeq \operatorname{Fun}(B,\mathcal S). The functor associated to Y→BY\to B sends bb to Yb={b}×BYY_b=\{b\}\times_B Y. Its maps are transport along paths, and homotopies of paths act through the higher coherences of transport. One construction uses straightening of left fibrations [2, Section 2.2.1]. Over a groupoidal base, the Grothendieck construction of a space-valued functor has only invertible arrows. Its total space is the homotopy sum colim⁡b∈BF(b)\operatorname{colim}_{b\in B}F(b) over colim⁡b∈B∗≃B\operatorname{colim}_{b\in B}*\simeq B. In the other direction an arbitrary map of spaces can be represented by a Kan fibration, which in particular is a left fibration. Thus the straightening correspondence restricts to the displayed equivalence.

For connected BB the point bb is an effective epimorphism in S\mathcal S, and the space BB is the delooping of ΩbB\Omega_bB. Equation (2) and Theorem 1.2 identify evaluation at bb with the forgetful functor from ΩbB\Omega_bB-actions. The point b:∗→Bb: *\to B has fiber ΩbB\Omega_bB. Under the equivalence it is the regular action, also characterized as the free action on the point. Indeed maps ∗→Y*\to Y over BB are points of YbY_b, and the Yoneda lemma in Fun⁡(B,S)\operatorname{Fun}(B,\mathcal S) characterizes the free action by the representable functor Map⁡B(b,−)\operatorname{Map}_B(b,-).

Pullback along a map f:C→Bf:C\to B corresponds to restriction of representations. Both adjoints to restriction exist and are computed by Kan extension. At bb, left Kan extension of F:C→SF:C\to\mathcal S takes the homotopy sum of FF over the groupoid of pairs (c,f(c)→b)(c,f(c)\to b). Right Kan extension takes the limit over pairs (c,b→f(c))(c,b\to f(c)). Inversion identifies the two indexing groupoids. The left adjoint is composition of total-space maps over the base, since giving a map from that composite is equivalent to giving a map to the pullback by the universal property of a fiber product. In the right adjoint the fiber over bb is the space of sections of the restricted family over the fiber of ff. In particular the two adjoints along B→∗B\to * are homotopy orbits and homotopy fixed points.

If a connected space BB has a fiber FF at bb, then after choosing an identification with a specified space KK, the action can be expressed by a based map B⟶BAut⁡(K).B\longrightarrow B\operatorname{Aut}(K). Here Aut⁡(K)\operatorname{Aut}(K) is the space of equivalences K→KK\to K with composition. To verify the classification, take the core S≃\mathcal S^\simeq in a universe containing the given fibers, choose the component containing KK, and use Proposition 1.1 to identify the component with BAut⁡(K)B\operatorname{Aut}(K). Every functor from a space lands in the core, since all its arrows are invertible, and connectedness forces the selected component. Taking the homotopy fiber of evaluation at bb gives the based mapping space and records the chosen identification. Forgetting the identification gives the unpointed mapping space. Taking π0\pi_0 of Aut⁡(K)\operatorname{Aut}(K) loses the higher transition data. For instance, let K=S1K=S^1. The identity component of Aut⁡(S1)\operatorname{Aut}(S^1) is equivalent to S1S^1, as follows by evaluating a degree-one map at a point and noting that the degree-one component of the based mapping space is contractible. Consequently π2(BAut⁡(S1))≅Z.\pi_2\bigl(B\operatorname{Aut}(S^1)\bigr)\cong \mathbb Z. There are based families with fiber S1S^1 over S2S^2 whose classifying maps represent a nonzero integer, even though π1(S2)\pi_1(S^2) is trivial.

The subgroup correspondence and its higher replacement

Fix a pointed connected space (B,b)(B,b). A pointed intermediate object is a factorization ∗→yY→qB*\xrightarrow{y}Y\xrightarrow{q}B of bb, including an identification of the composite with bb. Its morphisms are maps of such factorizations, with coherent triangles. By (2) the category of factorizations is the category of actions equipped with a map from the regular action, or equivalently the category of actions with a specified point of the fiber over bb. This statement is true whether or not YY is connected. It follows because taking the undercategory of a given object preserves equivalences of ∞\infty-categories.

Proposition 1.4 (Connected intermediates). The full subcategory of pointed intermediate objects with connected YY is equivalent to the slice of group objects in spaces over G=ΩbBG=\Omega_b B. Under the equivalence Y→BY\to B corresponds to H=ΩyY→GH=\Omega_yY\to G. Its fiber over bb, with the distinguished point, is the homotopy quotient G//HG//H of the right HH-action on GG through H→GH\to G.

Proof. Slicing the equivalence of Proposition 1.1 gives the categorical claim. For the fiber, model the map as BH→BGBH\to BG. Pull back the family ∗→BG*\to BG to BHBH. Its fiber at the point of BHBH is GG. Choose the path coordinate of a point in this family to run from the image of the moving point of BHBH to the fixed point of BGBG. If ϕ:H→G\phi:H\to G is the homomorphism, transport in this coordinate along hh sends gg to gϕ(h)−1g\phi(h)^{-1}. This is the covariant presentation of the right action g⋅h=gϕ(h)g\cdot h=g\phi(h). The total space is its homotopy quotient G//HG//H, and the chosen point comes from the identity in GG. Equivalently, this quotient is the realization obtained by pulling the augmented bar diagram of HH back along ∗→BG*\to BG, so its description fixes the same map and basepoint without choosing path coordinates. ◻

The connectedness condition on the total space is the homotopy-theoretic transitivity condition. In a pointed factorization, the map from the regular action to the fiber is the orbit map induced by the distinguished point. It is an effective epimorphism exactly when YY is connected. In one direction the orbit map is the base change of ∗→Y*\to Y along Yb→YY_b\to Y, and in the other direction effective epimorphisms descend along the effective epimorphism Yb→YY_b\to Y. The latter map is a base change of ∗→B*\to B. To justify the descent assertion, factor a map into its effective epimorphism and monomorphism. The subobject given by the second factor is terminal after base change precisely if it was terminal before base change, since base change along an effective epimorphism detects equivalences. For spaces this criterion says that the regular orbit is surjective on components and places no discreteness condition on the fiber.

Theorem 1.5 (Subgroups at the covering level). The category of pointed factorizations over (B,b)(B,b) with YY connected and q:Y→Bq:Y\to B 00-truncated is equivalent to the poset of subgroups of π1(B,b)\pi_1(B,b), ordered by inclusion.

Proof. Under (2), the condition on qq says every value of the functor is discrete. Such functors factor through the ordinary fundamental groupoid of BB. This follows from the universal property of the homotopy category, since the category of sets is ordinary and a functor on a groupoid takes its arrows to bijections. Choosing the object bb of the connected fundamental groupoid gives a π1(B,b)\pi_1(B,b)-action on the fiber. The distinguished point xx and connectedness make the action transitive, by the orbit criterion above. The stabilizer KK of xx is a subgroup, and the pointed action is isomorphic to the left action on π1(B,b)/K\pi_1(B,b)/K via gK↦g⋅xgK\mapsto g\cdot x. A pointed equivariant map exists from the quotient by KK to the quotient by LL precisely when K⊆LK\subseteq L, and is then uniquely determined by the displayed formula. Mapping spaces between the pointed actions are discrete because maps between discrete fibers form discrete spaces and naturality is imposed by a limit of such spaces. The result is the stated equivalence of categories. ◻

This is the regular-quotient argument of [1, Proposition 2.4.1] at the covering level. The opposite category reverses its containment order, as for intermediate fields. Subgroups up to conjugacy classify connected covers up to equivalence when the point in each intermediate cover is forgotten. For example, the equivariant endomorphisms of the transitive set π1(B,b)/K\pi_1(B,b)/K correspond to the cosets fixed under KK. The group of invertible endomorphisms is Nπ1(B,b)(K)/KN_{\pi_1(B,b)}(K)/K, up to inversion in the convention for right translation.

The loop homomorphism for a connected covering is the inclusion of a union of components of GG. Indeed its fiberwise description as the stabilizer is the pullback of the orbit map G→YbG\to Y_b along the point xx, and YbY_b is discrete. For a general intermediate object the same pullback is a homotopy fiber with a chosen nullhomotopy of motion of xx. Such a fiber need not be a union of components. Take B=∗B=* and Y=K(Z,2)Y=K(\mathbb Z,2). The map of loop groups is K(Z,1)→∗K(\mathbb Z,1)\to *, whose only homotopy fiber is neither empty nor contractible. Hence it fails to be a monomorphism in spaces. Proposition 1.4 accommodates this data by allowing arbitrary homomorphisms into GG.

Corollary 1.6 (Relative truncation of the stabilizer). For n≥0n\geq 0, a map BH→BGBH\to BG of pointed connected spaces is nn-truncated if and only if H→GH\to G is (n−1)(n-1)-truncated.

Proof. Set F=BH×BG∗F=BH\times_{BG}*, pointed by the chosen points. The homotopy fiber of H→GH\to G over the identity is ΩF\Omega F, since based loops preserve limits. Every fiber of H→GH\to G over a point in its image on components is equivalent to the identity fiber by translation, and all other fibers are empty. The degree-one orbit argument also shows that G→FG\to F meets every component of FF, and the left GG-action on FF gives equivalences carrying the distinguished point into each component. Thus FF is nn-truncated if and only if its loop space at that point is (n−1)(n-1)-truncated. The fibers of BH→BGBH\to BG are all equivalent to FF since BGBG is connected. Combining these observations proves the claim. ◻

For n=0n=0, this identifies the covering condition with monomorphisms of loop groups. For n=1n=1, the stabilizer map can have nontrivial discrete homotopy fibers, as for Y=BZY=B\mathbb Z over B=∗B=*. A classification by subgroup inclusions already fails at this level. The full slice of group objects records how the stabilizer maps into the ambient group, including the coherent data through which it acts.

Regular intermediates and delooped quotients

For a pointed connected cover classified by a subgroup K⊆π1(B,b)K\subseteq\pi_1(B,b), the deck transformations act transitively on the fiber precisely when KK is normal. Indeed the orbit of the distinguished coset under deck transformations is indexed by Nπ1(B,b)(K)/KN_{\pi_1(B,b)}(K)/K, so it is all of π1(B,b)/K\pi_1(B,b)/K precisely when the normalizer is all of π1(B,b)\pi_1(B,b). In that case the quotient is itself a group and the cover is the pullback of ∗⟶B(π1(B,b)/K)*\longrightarrow B\bigl(\pi_1(B,b)/K\bigr) along the map B→B(π1(B,b)/K)B\to B(\pi_1(B,b)/K) induced by the quotient homomorphism. The path space construction gives left transport by the quotient group, and the right deck transformations commute with it.

At higher truncation levels the homotopy quotient G//HG//H is always defined. It can have higher homotopy groups even when GG is discrete. The existence of a delooped quotient is a separate requirement and can be expressed in terms of the pointed action already obtained from the factorization.

Proposition 1.7 (Criterion for a delooped quotient). Let i:H→Gi:H\to G be a map of group objects in spaces and let QQ be another group object. For a homomorphism ϕ:G→Q\phi:G\to Q, there is a pointed equivalence over BGBG BH≃BG×BQ∗BH\simeq BG\times_{BQ}* if and only if G//HG//H, as a pointed left GG-space, is equivalent to the regular left QQ-space restricted along ϕ\phi. In the pointed actions the specified points are those induced by the identities.

Proof. The object ∗→BQ*\to BQ corresponds to the regular action by (2). Pullback along BϕB\phi restricts the action. Proposition 1.4 computes the fiber of BH→BGBH\to BG as G//HG//H. Transport from the fixed base along loops of BGBG acts on this fiber by left multiplication. This action commutes coherently with the right HH-action used to form the quotient. Specifying a point in the fiber is the same as giving a pointed map from the regular action of GG. Hence (2), applied also to the undercategory of the point, gives the stated equivalence of conditions. ◻

For a solution ϕ\phi, the action on components of QQ is transitive because BHBH is connected. Thus π0G→π0Q\pi_0 G\to \pi_0 Q is surjective. This condition is automatic for the ordinary quotient by a normal subgroup, but the full equivalence of pointed actions in Proposition 1.7 probes every homotopy group of the fiber and its transport.

Effective descent and Čech actions

The split-algebra equivalence in [1, Chapter 2] classifies objects by transport among their trivializations. Later chapters of that book make the descent morphism explicit and replace a single group by an internal groupoid. In a higher topos this construction starts with the entire Čech nerve of an effective epimorphism. Each composable string of changes of trivialization comes with its own coherence, and these coherences are part of the gluing datum.

Descent data and their realization

For a map q:U→Bq:U\to B, put Un=U×B(n+1)U_n=U^{\times_B(n+1)}. The cosimplicial diagram of slices [n]⟼X/Un[n]\longmapsto \mathcal X_{/U_n} acts on a simplex operator by pullback. Define Desc⁡(q)\operatorname{Desc}(q) to be its limit in ∞\infty-categories. This limit can be described as the category of simplicial objects V∙→U∙V_\bullet\to U_\bullet such that for every map [m]→[n][m]\to[n] in the simplex category the square Vn⟶Vm↓↓Un⟶Um\begin{array}{ccc} V_n&\longrightarrow&V_m\\ \downarrow&&\downarrow\\ U_n&\longrightarrow&U_m \end{array} is cartesian. Indeed the straightening of a diagram taking an object AA to X/A\mathcal X_{/A} is the target projection of the arrow category with its cartesian morphisms the pullback squares. A limit of such a diagram classifies its cartesian sections. In this description the two vertex identifications in degree one give an equivalence of the two pullbacks of V0V_0 to U1U_1. The higher simplices specify coherent composition of those equivalences.

Theorem 1.1 (Effective descent). For an effective epimorphism q:U→Bq:U\to B in an ∞\infty-topos, the comparison X/B⟶Desc⁡(q),E⟼(E×BUn)n≥0,\mathcal X_{/B}\longrightarrow \operatorname{Desc}(q), \qquad E\longmapsto (E\times_B U_n)_{n\geq 0}, is an equivalence. Its inverse is geometric realization over BB.

Proof. The realization of the displayed diagram for EE is EE, by universality of colimits and ∣U∙∣≃B|U_\bullet|\simeq B. For the other composite let V∙V_\bullet be a descent datum and put Y=∣V∙∣Y=|V_\bullet|. First, V∙V_\bullet is a groupoid object. The Segal squares can be checked using a vertex at the beginning of a string together with cartesian transport to each next vertex. A composable string of arrows from V0V_0 then consists of its initial point and the underlying composable string in U∙U_\bullet. Transport along the inverse of a base arrow gives the inverse arrow of V∙V_\bullet, again by the cartesian vertex condition.

By effectivity V0→YV_0\to Y is an effective epimorphism and V1≃V0×YV0V_1\simeq V_0\times_Y V_0. There is a canonical comparison V0→Y×BUV_0\to Y\times_B U over YY. After pullback along V0→YV_0\to Y, this comparison becomes V0×YV0⟶V0×BU.V_0\times_Y V_0\longrightarrow V_0\times_B U. The source is V1V_1, and the target is V0×UU1V_0\times_U U_1 where U1→UU_1\to U uses the initial vertex. The comparison is therefore the cartesian vertex equivalence of the datum. To descend that equivalence, pull it and both its domain and codomain to the entire Čech nerve of V0→YV_0\to Y. The comparison is an equivalence in every such degree. Each original object is recovered by realizing its pullback diagram, by universality of colimits. Hence V0≃Y×BUV_0\simeq Y\times_B U. The vertex comparison in each degree identifies Vn≃Y×BUnV_n\simeq Y\times_B U_n, compatibly with all simplex operators. This gives the natural equivalences of the two composites. ◻

This gives the underlying mechanism of the action equivalence of Theorem [thm:actions-delooping]. When U=1U=1 and B=BGB=BG, the Čech nerve is the bar groupoid of GG. At the other extreme, there may be no distinguished global point of UU or BB. The internal groupoid then describes transitions between varying objects of objects. Replacing it by the loops at a chosen object would require extra hypotheses that are absent for most covers in a topos.

Corollary 1.2 (What a general map sees). For any q:U→Bq:U\to B, let i:I↪Bi:I\hookrightarrow B denote its image and p:U→Ip:U\to I the associated effective epimorphism. There is a canonical equivalence Desc⁡(q)≃X/I\operatorname{Desc}(q)\simeq\mathcal X_{/I}, under which comparison from X/B\mathcal X_{/B} is restriction along ii. Comparison is an equivalence if and only if qq is an effective epimorphism.

Proof. The
vCech diagrams for pp and qq coincide because ii is a monomorphism. More explicitly I×BI≃II\times_B I\simeq I, so the comparison U×IU→U×BUU\times_I U\to U\times_B U is an equivalence, and the same holds for every iterated pullback. Theorem 1.1 applied to pp gives the conclusion about restriction. If i∗i^* is an equivalence, it reflects equivalences and carries (I→B)→(B→B)(I\to B)\to(B\to B) to an equivalence. Thus ii is an equivalence. ◻

Mapping spaces and properties local on the base

Descent reconstructs morphisms as well as objects. For E,F→BE,F\to B and qq effective, it identifies Map⁡X/B(E,F)≃lim⁡[n]∈ΔMap⁡X/Un(E×BUn,F×BUn). \operatorname{Map}_{\mathcal X_{/B}}(E,F) \simeq \lim_{[n]\in\Delta} \operatorname{Map}_{\mathcal X_{/U_n}} (E\times_B U_n,F\times_B U_n). To check this directly, realize the augmented diagram E×BU∙→EE\times_B U_\bullet\to E in the slice over BB. Mapping into FF turns its colimit into a limit of mapping spaces. The adjunction between composition and pullback along Un→BU_n\to B identifies each term of that limit with the corresponding term on the right of (3). An element of the right-hand side specifies a local map together with compatibility homotopies over intersections and higher coherent comparisons on all iterated intersections.

Let D(B)⊆X/B\mathcal D(B)\subseteq \mathcal X_{/B} be a full subcategory closed under equivalences, for each BB, and assume pullback takes objects of D\mathcal D to objects of D\mathcal D. Call membership local on the base if E→BE\to B belongs to D(B)\mathcal D(B) whenever its pullback belongs to D(U)\mathcal D(U) along an effective epimorphism U→BU\to B.

Proposition 1.3 (Descent for full subcategories). If membership in D\mathcal D is local on the base, then for every effective epimorphism q:U→Bq:U\to B, D(B)≃lim⁡[n]∈ΔD(Un).\mathcal D(B)\simeq \lim_{[n]\in\Delta}\mathcal D(U_n). Conversely, if this comparison is an equivalence for all effective epimorphisms, membership is local on the base.

Proof. The proposed target is a full subcategory of Desc⁡(q)\operatorname{Desc}(q), since limits preserve full inclusions of ∞\infty-categories. A cartesian section lies in it exactly when its degree-zero object is in D(U)\mathcal D(U), since the other entries are equivalent to pullbacks along a vertex. Theorem 1.1 identifies such sections with the objects over BB whose pullback along qq belongs to D(U)\mathcal D(U). This proves both directions. ◻

For example, the class of nn-truncated maps is local on the base for each integer n≥−2n\geq -2. The case n=−2n=-2 is the conservativity of pullback along an effective epimorphism. For the induction step, a map E→BE\to B is nn-truncated precisely if its diagonal E→E×BEE\to E\times_B E is (n−1)(n-1)-truncated. Forming the diagonal commutes with base change. The cover on its codomain after pullback along U→BU\to B is effective, so the induction hypothesis applies. In particular this yields descent for monomorphisms and for covering objects with discrete fibers. Effective epimorphisms themselves are local, by applying the equivalence test to the image inclusion as in the orbit argument of Chapter [ch:torsors]. The locality proof uses finite-level properties and is valid in a nonhypercomplete topos.

The descent monad

Effective descent also has a monadic formulation. The functor q∗:X/B→X/Uq^*:\mathcal X_{/B}\to\mathcal X_{/U} has left adjoint q!q_! given by composition with qq. It preserves colimits by universality and detects equivalences by the Čech argument. The ∞\infty-categorical Barr–Beck theorem then gives X/B≃Alg⁡T(X/U),T=q∗q!.\mathcal X_{/B}\simeq \operatorname{Alg}_{T}(\mathcal X_{/U}), \qquad T=q^*q_!. In the language of the Barr–Beck hypotheses, preserving all colimits implies preservation of realizations of those simplicial objects that become split after q∗q^* [6, Theorem 4.7.3.5]. This proves monadicity without assuming that the underlying morphism qq has a section.

The monad can be expressed in terms of the two projections s,t:U1⇉Us,t:U_1\rightrightarrows U, with ss the source and tt the target. For A→UA\to U, T(A)≃t!s∗A≃U×BA.T(A)\simeq t_!s^* A \simeq U\times_B A . In the last expression the first coordinate is the target in UU. The unit is induced by the diagonal of UU over BB. The multiplication takes (u2,u1,a)(u_2,u_1,a) to (u2,a)(u_2,a) with the prescribed composition of paths in BB. Its simplicial powers are indexed by strings in the Čech groupoid. For A=E×BUA=E\times_B U, the monad action identifies the point ee over the preceding vertex with the same point over the new vertex. Thus the map T(A)→AT(A)\to A is the transport correspondence from the canonical descent datum. The monadic comparison and the Čech comparison agree on every pullback from BB, and both comparisons are equivalences. This relates their coherent algebra and coherent transport descriptions without an extra hypothesis on qq.

As with the classical treatment in [1, Section 5.1], the monad lives on a slice over the splitting object. If a geometric problem begins with algebras and replaces them by opposite affine objects, geometric pullback usually becomes extension of scalars. Passing to opposites converts the preceding monadic expression into a comonadic one with the corresponding variance of the adjunction. In a general algebraic category, extension of scalars still must be proved to satisfy a descent theorem. The argument in this chapter uses groupoid effectivity and universal colimits in a specific ambient topos, which are the hypotheses making arbitrary effective epimorphisms work here.

Refinements and changes of groupoid presentation

Every effective internal groupoid presents a base. Let D∙D_\bullet be an internal groupoid, let B=∣D∙∣B=|D_\bullet|, and apply Theorem 1.1 to D0→BD_0\to B. This identifies cartesian objects over D∙D_\bullet with X/B\mathcal X_{/B}. A map of groupoids induces a pullback functor between their cartesian action categories, and under this equivalence it is pullback along the map of realizations. In fact, pulling back E×∣D∙∣D∙E\times_{|D_\bullet|}D_\bullet along the map of groupoids gives its pullback from EE degree by degree. Refinement of an effective cover is a particular case, in which the map of realizations is the identity. Restricting a datum to a refinement changes the representatives for its local objects and their transition maps while preserving its realized object.

The following criterion gives a specific test for changing presentations. For a map F∙→D∙F_\bullet\to D_\bullet of internal groupoids, call it fully faithful if the comparison F1⟶(F0×F0)×D0×D0D1F_1\longrightarrow (F_0\times F_0)\times_{D_0\times D_0}D_1 is an equivalence. Call it essentially surjective if the map F0×D0,sD1→ t D0F_0\times_{D_0,s}D_1\xrightarrow{\,t\,}D_0 is an effective epimorphism.

Proposition 1.4 (Change of presentation). If a map of internal groupoids is fully faithful and essentially surjective, it induces an equivalence of realizations and an equivalence of cartesian action categories. Conversely, a map inducing an equivalence of realizations is fully faithful and essentially surjective.

Proof. Write A=∣F∙∣A=|F_\bullet| and B=∣D∙∣B=|D_\bullet|. Effectivity gives (F0×F0)×D0×D0D1≃F0×BF0.(F_0\times F_0)\times_{D_0\times D_0}D_1\simeq F_0\times_B F_0. Full faithfulness thus identifies the degree-one terms of the Čech nerves of F0→AF_0\to A and F0→BF_0\to B, compatibly with both vertices. By the Segal identities their entire nerves agree. Their realizations identify AA with the image of F0→BF_0\to B. On the other hand F0×D0,sD1≃F0×BD0F_0\times_{D_0,s}D_1\simeq F_0\times_B D_0, so essential surjectivity is equivalent to F0→BF_0\to B being effective after base change along D0→BD_0\to B. Effectivity descends along that cover. The image is therefore BB. If A→BA\to B is an equivalence, the same formulas give full faithfulness, and essential surjectivity follows by base change from the effective map F0→AF_0\to A. The result for actions follows from the descent comparisons. ◻

An equivalence here allows the objects of objects to change. In later applications one can replace a groupoid associated to a cover by a more convenient presentation as long as the change of realization, or the two conditions in the proposition, can be verified.

Fixed split type and the torsor of frames

Fix K∈XK\in\mathcal X. An object E→BE\to B is a form of KK if there exists an effective epimorphism U→BU\to B and an equivalence E×BU≃U×KE\times_B U\simeq U\times K over UU. For a specified effective epimorphism qq, the objects it splits with this type correspond in Desc⁡(q)\operatorname{Desc}(q) to those data with degree-zero entry equivalent to U×KU\times K. The category of these objects, with all maps over BB, is full by Theorem 1.1. To compute the space of forms we restrict to equivalences of total objects.

The group object G=Aut⁡‾(K)G=\underline{\operatorname{Aut}}(K) is the object of self-equivalences with its coherent composition. Concretely, a section of GG over WW is an equivalence W×K→W×KW\times K\to W\times K over WW, and these spaces of sections are represented by the equivalence part of the internal hom. For a form EE, its frame object over BB represents equivalences W×K→E×BWW\times K\to E\times_B W over WW, for test objects W→BW\to B. Local cartesian closedness gives the internal hom over BB, and the equivalence locus can be represented by requiring a two-sided inverse. More precisely, the object of maps with left and right inverses and witnesses for each gives this locus. At a map that is an equivalence the space of choices of each witness together with its inverse is contractible, by cancellation along an equivalence. At any other map at least one of the two spaces of choices is empty.

Proposition 1.5 (Classification of forms). The frame construction identifies the space of forms of KK over BB with the space of right GG-torsors over BB, hence with Map⁡X(B,BG)\operatorname{Map}_{\mathcal X}(B,BG).

Proof. Precomposition equips the frame object Fr⁡(E)\operatorname{Fr}(E) with a right GG-action. A choice of local equivalence with U×KU\times K gives a section over UU and identifies the action there with the regular right action. The torsor comparison Fr⁡(E)×G⟶Fr⁡(E)×BFr⁡(E)\operatorname{Fr}(E)\times G\longrightarrow \operatorname{Fr}(E)\times_B\operatorname{Fr}(E) is an equivalence after pulling back along U→BU\to B, and is therefore an equivalence globally. The frame map to BB is effective since it has a section over UU. Conversely, for a right GG-torsor PP form P×GKP\times^G K by realizing the diagonal action groupoid, where gg acts by (p,x)↦(pg−1,gx)(p,x)\mapsto(p g^{-1},g x). Over the effective epimorphism P→BP\to B the pulled-back torsor has its tautological section. The associated object then identifies with the product with KK. The map from PP to the frames of this associated object sends a local point pp to x↦[p,x]x\mapsto[p,x]. It is equivariant by the relation [pg,x]=[p,gx][p g,x]=[p,g x], and becomes an equivalence where PP is trivialized, so it is an equivalence. The evaluation of a frame on KK gives Fr⁡(E)×GK→E\operatorname{Fr}(E)\times^G K\to E, which becomes the product equivalence on the same trivializing cover and is thus an equivalence. Both maps are natural, giving inverse equivalences on cores. Corollary [cor:torsor-classification] gives the mapping-space description. ◻

If a finite-level truncation of KK is imposed, its automorphism object also has bounded truncation degree, because mapping into an nn-truncated object gives nn-truncated mapping spaces. The classification then uses a correspondingly truncated target BGBG. For general KK, it involves coherence at every degree. A section of the frame torsor is a chosen trivialization, and two sections may differ by a nontrivial automorphism. This distinguishes a property, being split, from the moduli of its witnesses.

The boundary between Čech descent and hyperdescent

A hypercover V∙→BV_\bullet\to B is an augmented simplicial object whose relative matching maps are effective epimorphisms, including V0→BV_0\to B. The Čech nerve of an effective epimorphism is a hypercover, with the matching maps in positive degrees equivalences. A general hypercover may have additional local choices in higher degrees. It need not be a groupoid, so the effectivity argument for Theorem 1.1 does not prove descent along it.

For an object EE of X\mathcal X, hypercompleteness means that mapping into EE inverts all infinitely connective maps. The hypercover criterion identifies this with the requirement that the augmentation induce an equivalence Map⁡X(B,E)⟶lim⁡[n]∈ΔMap⁡X(Vn,E)\operatorname{Map}_{\mathcal X}(B,E) \longrightarrow \lim_{[n]\in\Delta}\operatorname{Map}_{\mathcal X}(V_n,E) for all hypercovers V∙→BV_\bullet\to B [2, Theorem 6.5.3.12]. This uses the fact that ∣V∙∣→B|V_\bullet|\to B is infinitely connective. In a hypercomplete topos it is an equivalence. A truncated target EE also satisfies the displayed condition in an arbitrary ∞\infty-topos, because maps into a target of fixed finite truncation cannot detect an infinitely connective discrepancy. The distinction matters when local trivializations or coefficients are given as unbounded towers. Categorical descent for one effective epimorphism here avoids any completeness hypothesis because the augmentation of that particular nerve is already an equivalence.

Monodromy and shape

Pro-spaces and continuous monodromy

The passage from a single splitting field to a directed family of splitting fields in [1, Ch. 3] changes the meaning of an action. At each finite stage the Galois group acts discretely. An orbit in the limit need only be controlled by a quotient large enough to see that orbit. A continuous action on an entire space can require a more restrictive or a more flexible definition, depending on whether one demands a single trivializing stage for the object or permits a filtered assembly of objects with that property. This chapter distinguishes those constructions before passing to the shape of a topos.

Pro-objects and the role of the test space

Write S\mathcal S for the ∞\infty-category of spaces. A pro-space PP is represented by a small cofiltered diagram i↦Pii\mapsto P_i. For a second pro-space Q={Qj}j∈JQ=\{Q_j\}_{j\in J}, the mapping formula is Map⁡Pro⁡(S)(P,Q)≃lim⁡j∈Jcolim⁡i∈IopMap⁡S(Pi,Qj). \operatorname{Map}_{\operatorname{Pro}(\mathcal S)}(P,Q) \simeq \lim_{j\in J}\operatorname*{colim}_{i\in I^{\mathrm{op}}} \operatorname{Map}_{\mathcal S}(P_i,Q_j). We place colim⁡i\operatorname{colim}_{i} over the filtered direction throughout this chapter. Formula (4) comes from the embedding of Pro⁡(S)op\operatorname{Pro}(\mathcal S)^{\mathrm{op}} in Fun⁡(S,S)\operatorname{Fun}(\mathcal S,\mathcal S) taking PP to FP(T)=colim⁡iMap⁡(Pi,T).F_P(T)=\operatorname*{colim}_{i}\operatorname{Map}(P_i,T). Indeed the representable Map⁡(Qj,−)\operatorname{Map}(Q_j,-) is a term in the filtered presentation of FQF_Q. A natural transformation from FQF_Q to FPF_P is a compatible family of transformations from those representables. The Yoneda lemma identifies the family with the right side of (4). In particular, an equivalence of pro-spaces can be checked against every constant test TT. Here and below an ambient size enlargement may be used for the categories of test functors.

The realized inverse limit lim⁡iPi\lim_i P_i loses this test property. Take Pm={m,m+1,… }P_m=\{m,m+1,\dots\} with the inclusions as transition maps. As discrete spaces their inverse limit is empty. Yet FP(∅)F_P(\varnothing) is empty because every PmP_m is nonempty. The constant empty pro-space has F∅(∅)≃∗F_{\varnothing}(\varnothing)\simeq *. Thus even this tower of discrete spaces cannot be replaced by its realized inverse limit for a classification problem.

Definition 1.1. A space is π\pi-finite if it is truncated, has finitely many components, and has finite positive homotopy groups at every basepoint. Let Sπ\mathcal S_\pi be the full subcategory on these spaces, and let S<∞\mathcal S_{<\infty} be the full subcategory of all truncated spaces. The corresponding completions of PP are written Pπ∧∈Pro⁡(Sπ)P^{\wedge}_\pi\in\operatorname{Pro}(\mathcal S_\pi) and P<∞∈Pro⁡(S<∞)P_{<\infty}\in\operatorname{Pro}(\mathcal S_{<\infty}).

These completions are characterized by retaining exactly the tests of the indicated kind. In the functor presentation, restrict FPF_P to the indicated full subcategory. Each subcategory admits finite limits. In the π\pi-finite case, this follows from the homotopy exact sequence for a pullback and the finiteness of the component sets. Both restrictions preserve finite limits because filtered colimits of spaces do so. The pro-object characterization by accessible finite-limit-preserving functors then gives the completions. For truncated spaces the accessibility issue can be handled by the cofiltered diagram {τ≤dPi}i,d\{\tau_{\leq d}P_i\}_{i,d}, since maps to any fixed truncated test stabilize in the Postnikov coordinate. A particular consequence that will be used repeatedly is Map⁡(Pπ∧,T)≃Map⁡(P,T)(T∈Sπ). \operatorname{Map}(P^{\wedge}_\pi,T) \simeq\operatorname{Map}(P,T) \qquad (T\in\mathcal S_\pi). There is an analogous formula for P<∞P_{<\infty} with T∈S<∞T\in\mathcal S_{<\infty}. Either construction is left adjoint to the inclusion of the corresponding pro-category. This follows for a general target by taking the limit of the equivalences for its constant stages. The same argument in degree 00 recovers the equivalence between pro-finite sets and Stone spaces used in [1, Ch. 3]. The higher completion tests Eilenberg–Mac Lane spaces, classifying spaces, and their finite Postnikov combinations.

A representation category at stages

For a coefficient ∞\infty-category D\mathcal D, set Rep⁡D(P)=colim⁡iFun⁡(Pi,D), \operatorname{Rep}_{\mathcal D}(P)= \operatorname*{colim}_{i}\operatorname{Fun}(P_i,\mathcal D), where the colimit is in Cat⁡∞\operatorname{Cat}_{\infty}, and the arrows are given by restriction. A representative of an object is a family and all its transport coherence at a single stage. A representative of a morphism is a natural transformation after passing to a common stage. More precisely, if FF and GG are already represented at a common stage ii, their mapping space is Map⁡Rep⁡D(P)(F,G)≃colim⁡(k→i)Map⁡Fun⁡(Pk,D)(F∣Pk,G∣Pk). \operatorname{Map}_{\operatorname{Rep}_{\mathcal D}(P)}(F,G) \simeq \operatorname*{colim}_{(k\to i)} \operatorname{Map}_{\operatorname{Fun}(P_k,\mathcal D)} (F|_{P_k},G|_{P_k}). This follows by computing mapping spaces as fibers of the core of the arrow category over the core of pairs of objects. Filtered colimits of ∞\infty-categories commute with forming these cores and arrow categories, and filtered colimits of spaces commute with the fibers. The reason for the first commutation can also be seen in simplicial models: every prescribed simplex and every witness that an edge is invertible involves a finite diagram, so it is already present at a single refinement. Two stages that represent a prescribed datum become comparable after a further refinement. This proves invariance under cofiltered change of presentation and recovers the functorial construction of [3, §2].

The coefficient category may have infinitely many objects up to equivalence. A finite stage can still carry an arbitrary object of D\mathcal D. The crucial finiteness for completion is instead the size and truncation of the spaces classifying possible transport and coherent maps.

Lemma 1.2. If AA and BB are π\pi-finite, then Map⁡(A,B)\operatorname{Map}(A,B) is π\pi-finite. The component of AA in the core of spaces is BAut⁡(A)B\operatorname{Aut}(A) and is π\pi-finite.

Proof. A π\pi-finite space has a simplicial model with finitely many simplices in every degree. For a connected 11-type use the nerve of its finite fundamental group. At a higher Postnikov step, the fiber K(M,n)K(M,n) for a finite abelian group MM has a simplicial abelian model with finitely many simplices in each degree. Represent the step by its twisted kk-invariant into the K(M,n+1)K(M,n+1)-fibration over the nerve of the finite fundamental group. This target fibration can be modeled by the Borel construction for that group acting on K(M,n+1)K(M,n+1). Its total space and the fiberwise path fibration are finite in each simplicial degree. Pulling the path fibration back along the twisted invariant gives a degreewise finite model for the extension. This inductive construction and a finite disjoint union give the asserted model.

If BB is nn-truncated with n≥0n\geq 0, the inclusion of the (n+1)(n+1)-skeleton of this model into AA induces an equivalence on maps to BB, since the cells subsequently attached have dimension at least n+2n+2. The skeleton is a finite cell complex. Mapping from a cell attachment into BB forms a finite homotopy pullback of mapping spaces, starting with BB and a point. Closure under finite homotopy limits gives the claim. A (−1)(-1)-truncated target is either empty or contractible and the conclusion holds directly. The equivalences A→AA\to A form a union of components of Map⁡(A,A)\operatorname{Map}(A,A), hence a π\pi-finite group-like space. Its delooping is connected and shifts its homotopy groups by one degree, proving the second assertion. ◻

To specify the completion comparison for all morphisms, let Rep⁡Sπfr(P)\operatorname{Rep}_{\mathcal S_\pi}^{\mathrm{fr}}(P) consist of representations admitting a representative at a stage with a finite range of equivalence classes of fibers. Here fr\mathrm{fr} refers to the range of isomorphism types, and the superscript imposes a condition on objects only. Thus all maps in (7) between those objects are retained.

Proposition 1.3. Precomposition with P→Pπ∧P\to P^{\wedge}_\pi gives an equivalence Rep⁡Sπ(Pπ∧)≃Rep⁡Sπfr(P).\operatorname{Rep}_{\mathcal S_\pi}(P^{\wedge}_\pi) \simeq \operatorname{Rep}_{\mathcal S_\pi}^{\mathrm{fr}}(P). When PP admits a presentation by connected spaces, the finite-range condition on the right is automatic.

Proof. A representation over a π\pi-finite stage has finite range because the stage has finitely many components. Conversely, for a finite list DD of equivalence classes of fibers, let CDC_D be the union of those components in the core of Sπ\mathcal S_\pi. It is π\pi-finite by Lemma 1.2. The moduli space of stage objects with range in DD is Map⁡(P,CD)\operatorname{Map}(P,C_D), because a functor from a space factors through the core of its target. Thus every object lifts through the completion by (5).

For the map space between two such objects choose finite lists D0,D1D_0,D_1 for the source and target. Form the core of the arrow category and restrict its endpoints to CD0×CD1C_{D_0}\times C_{D_1}. The resulting space EE has a projection to that product with fiber Map⁡(A,B)\operatorname{Map}(A,B) at (A,B)(A,B). Each fiber and the base are π\pi-finite. A uniform truncation bound for the fibers exists because the lists are finite. The homotopy exact sequence shows that the total space is also π\pi-finite. The space of morphisms for stage representations with given endpoints is the fiber of Map⁡(P,E)⟶Map⁡(P,CD0×CD1).\operatorname{Map}(P,E)\longrightarrow \operatorname{Map}(P,C_{D_0}\times C_{D_1}). This fiber description follows at each stage and passes through the filtered colimit. Both its source and target are preserved under (5), which proves full faithfulness. A connected stage maps to one component of the core, proving the final assertion. ◻

For example, take the constant pro-space N\mathbb N, with N\mathbb N discrete. A family of finite discrete spaces with fiber cardinality m+1m+1 over m∈Nm\in\mathbb N belongs to Rep⁡Sπ(N)\operatorname{Rep}_{\mathcal S_\pi}(\mathbb N). It cannot descend from its profinite completion by Proposition 1.3. Each individual fiber satisfies the requisite finiteness condition. The failure comes from the infinitely many equivalence classes occurring in a single family.

From one-stage actions to continuous discrete actions

Let H=lim⁡iH/UiH=\lim_i H/U_i be a profinite group, where the UiU_i form the directed system of open normal subgroups, and set BproH={B(H/Ui)}i.B_{\mathrm{pro}}H=\{B(H/U_i)\}_{i}. A left action on a discrete set MM is continuous exactly when the stabilizer of each element is open. One direction follows by taking the preimage of the singleton under each orbit map. For the converse, the action map restricted to H×{m}H\times\{m\} is locally constant for each mm, and these pieces are open in H×MH\times M.

The stage category Rep⁡Set(BproH)\operatorname{Rep}_{\mathrm{Set}}(B_{\mathrm{pro}}H) realizes the full subcategory of discrete actions with a common open normal subgroup acting trivially. An action at one stage has that property. Conversely the common subgroup allows the action to be written over a quotient. If two actions come from H/UiH/U_i and H/UjH/U_j, then a function between them is HH-equivariant exactly when it is equivariant for H/(Ui∩Uj)H/(U_i\cap U_j), using surjectivity of the quotient map. Therefore passing to stages identifies their mapping sets. Every finite continuous HH-set belongs to this subcategory, as a finite intersection of open stabilizers is open.

Every continuous discrete HH-set MM is a filtered union of finite continuous HH-sets. Every orbit is finite by the compactness of HH and the openness of its stabilizer, and one can take the finite unions of orbits. A finite HH-set is compact among continuous discrete HH-sets. To see this, transport a map into a filtered colimit to some stage as a function on its finite underlying set. Choose an open normal subgroup fixing all elements of the source and their chosen representatives at that stage. Equivariance can now be checked against the finitely many elements of the quotient by that subgroup. The required equalities hold after some further refinement, by the description of filtered colimits of sets. The same argument with finitely many pointwise equalities proves uniqueness in the colimit of mapping sets. Thus the category of all continuous discrete actions is the ind-completion of its finite subcategory.

The difference between stage representations and this ind-completion already occurs for H=ZpH=\mathbb Z_p. On the disjoint union ∐m≥1Z/pmZ\coprod_{m\geq 1}\mathbb Z/p^m\mathbb Z, the translation action is continuous and requires arbitrarily deep quotients. Its kernel is trivial, which is a closed subgroup that is not open. This object is built from stage actions by taking a filtered union, but no single stage represents it.

A different construction first discards the topology and takes B(Hδ)B(H^{\delta}), the delooping of the underlying discrete group. For H=∏m≥1F2H=\prod_{m\geq 1}\mathbb F_2, there is an abstract homomorphism H→F2H\to\mathbb F_2 that annihilates ⨁m≥1F2\bigoplus_{m\geq 1}\mathbb F_2 and takes the all-ones sequence to 11. Indeed, that sequence defines a nonzero vector in the quotient by the direct sum. Extend this vector to an algebraic basis and project to its coordinate. The resulting homomorphism is discontinuous because its kernel is dense and proper. Its action by translations on F2\mathbb F_2 is an action of HδH^{\delta} with no continuous counterpart. Thus using the abstract classifying space instead of the pro-classifying space can change even the finite-coefficient theory.

Closed stabilizers and the infinite correspondence

The compact Hausdorff replacement of a finite discrete transitive action is a transitive action on a profinite space. For every closed subgroup L≤HL\leq H, there is a homeomorphism of HH-spaces H/L⟶lim⁡i(H/Ui)/(LUi/Ui). H/L\longrightarrow\lim_i (H/U_i)/(LU_i/U_i). Here H/LH/L denotes the left HH-space of cosets hLhL, with the quotient topology. To prove injectivity use L=⋂iLUiL=\bigcap_i LU_i. Indeed, if h∉Lh\notin L, some coset neighborhood hUihU_i is disjoint from LL, by closedness and the neighborhood basis at the identity. A compatible point of the inverse limit specifies a family of closed cosets of the groups LUiLU_i. Any finite subfamily intersects by refining the indices. Compactness gives a point in their intersection, proving surjectivity. A continuous bijection from a compact space to a Hausdorff space is a homeomorphism.

Conversely, given a transitive continuous action on a nonempty profinite space YY and a point y∈Yy\in Y, its stabilizer LL is closed as the inverse image of {y}\{y\} under the orbit map. The induced continuous bijection H/L→YH/L\to Y is again a homeomorphism. Between pointed homogeneous spaces (H/L,L)(H/L,L) and (H/M,M)(H/M,M) there is a unique pointed equivariant map precisely when L⊆ML\subseteq M. In fact its value at hLhL is forced to be hMhM, and the subgroup condition makes this formula well defined and continuous. An unpointed equivariant map can send LL to any coset hMhM with L⊆hMh−1L\subseteq hMh^{-1}; the map is then determined by its value there. In the discrete continuous theory, transitive objects instead correspond to open stabilizers.

For a Galois field extension K⊆LK\subseteq L, the ordinary subextension correspondence uses closed subgroups of H=Gal⁡(L/K)H=\operatorname{Gal}(L/K). The equality of a subgroup with the limit of its finite images in (8) shows why closedness is the condition visible to all finite subextensions. In greater algebraic generality the stages of the classical correspondence can likewise assemble into profinite HH-spaces, as in [1, §3.5]. Continuous actions on profinite spaces and continuous actions on discrete sets therefore enter at different points of infinitary Galois theory. Replacing both by one category of HδH^\delta-sets identifies constructions that have different stabilizers and different colimits.

Continuous cochains and twisted coefficients

For a discrete continuous HH-module MM, consider the cochain groups Cctsq(H,M)=Cont⁡(Hq,M)(q≥0)C^q_{\mathrm{cts}}(H,M)=\operatorname{Cont}(H^q,M) \qquad(q\geq0) with the group-cohomology differential. For a fixed cochain the image is finite, since a compact space has finite image in a discrete space. Moreover continuity implies that the cochain factors as a function through some (H/Ui)q(H/U_i)^q. Pulling back its finitely many fibers gives a clopen partition of HqH^q. Each point has an open product of cosets contained in its partition member. A finite subcover by such products exists, and refining all the coset neighborhoods to one UiU_i proves the factorization. After shrinking UiU_i further, it fixes every element of the image. Thus the cochain corresponds to an element of Cq(H/Ui,MUi)C^q(H/U_i,M^{U_i}). In degree 00 the identical conclusion holds by openness of the stabilizer of the single value. The differentials respect refinement of the quotient and inclusion of the fixed submodules, giving an isomorphism of complexes Ccts∙(H,M)≅colim⁡iC∙(H/Ui,MUi).C^\bullet_{\mathrm{cts}}(H,M) \cong \operatorname*{colim}_{i} C^\bullet(H/U_i,M^{U_i}). Filtered colimits are exact in abelian groups. Taking cohomology therefore gives Hctsq(H,M)≅colim⁡iHq(H/Ui,MUi). H^q_{\mathrm{cts}}(H,M) \cong \operatorname*{colim}_{i} H^q(H/U_i,M^{U_i}). If the action on MM factors through a fixed stage, the subgroups contained in its kernel form a cofinal subsystem and one can use MM throughout on the right.

Let MM be finite with continuous action, and take q≥1q\geq1. The stage module actions induce families of pointed spaces K(M,q)K(M,q) over B(H/Ui)B(H/U_i) for all sufficiently deep ii. The mapping space of sections at a stage is the homotopy fixed point space of K(M,q)K(M,q), modeled by totalizing the usual bar cosimplicial object. Its set of components is Hq(H/Ui,M)H^q(H/U_i,M). Taking components commutes with filtered colimits, so the stage mapping space has components Hctsq(H,M)H^q_{\mathrm{cts}}(H,M) by (9). This identifies pro-monodromy with twisted cohomology directly from the stage dynamics. An untwisted test of the form Map⁡(P,K(M,q))\operatorname{Map}(P,K(M,q)) only computes the version with trivial action. In particular, truncation of an underlying homotopy type does not justify erasing the transport of its coefficient groups.

The three continuous constructions above use the same inverse system in different ways. Discrete ind-objects permit each orbit to descend separately. Profinite spaces permit closed stabilizers and keep the successive quotients of an orbit. Stagewise higher representations retain the full space of local coefficient transports at every stage, with refinement identifying equivalent presentations. The next chapter explains how a single ∞\infty-topos supplies the pro-space that indexes such stages.

The shape of a topos and its local systems

Covering theory in [1, Ch. 6] extracts a fundamental group from a category of coverings and a fiber functor. A higher topos contains families with fibers of all homotopy types. The role of the fundamental invariant is then played by a space or pro-space, and it can be constructed before deciding which coefficients the classification ought to allow.

Shape as a functor of constant tests

Let X\mathcal X be an ∞\infty-topos with constant-object functor Δ ⁣:S→X\Delta\colon\mathcal S\to\mathcal X and global sections Γ\Gamma. For U∈XU\in\mathcal X define Π(U)\Pi(U) by Map⁡Pro⁡(S)(Π(U),T)≃Map⁡X(U,ΔT). \operatorname{Map}_{\operatorname{Pro}(\mathcal S)}(\Pi(U),T) \simeq \operatorname{Map}_{\mathcal X}(U,\Delta T). The right side preserves finite limits in TT. It is also accessible, since Δ\Delta preserves colimits and UU is compact for some regular cardinal in the presentable category X\mathcal X. The pro-representability criterion from Chapter [ch:profinite] applies. Equivalently Π(U)\Pi(U) is the shape of the slice X/U\mathcal X_{/U}, whose constant-object functor is ΔUT=(U×ΔT→U)\Delta_U T=(U\times\Delta T\to U). The shape of X\mathcal X is Π∞X=Π(1)\Pi_\infty\mathcal X=\Pi(1) and corepresents the functor T↦ΓΔTT\mapsto\Gamma\Delta T. These are the relative and absolute constructions of [2, §7.1.6] and [3, §2].

One presentation of Π(U)\Pi(U) uses the category of pairs (T,t)(T,t) with t ⁣:U→ΔTt\colon U\to\Delta T. The maps are maps of test spaces together with specified coherent compatibilities with tt. The category of pairs is cofiltered since the functor of tests preserves finite limits. Indeed, a finite diagram of such pairs determines a diagram of test spaces whose limit inherits a section by finite-limit preservation. The resulting pair supplies a cone. An essentially small cofiltered presentation exists by accessibility, and the projection to TT pro-represents (10). Using all pairs is convenient for comparisons even when the class of tests is large.

There is a relation to actions internal to X\mathcal X. For a point a∈Ta\in T put Pa=U×ΔTΔ{a}.P_a=U\times_{\Delta T}\Delta\{a\}. The pullbacks inherit transport along TT, producing a functor a↦Paa\mapsto P_a in X/U\mathcal X_{/U}. The colimit of this functor is the terminal object of the slice. Indeed the family of all points of TT, with all transport, has colimit the identity of TT in S/T\mathcal S_{/T}. Apply Δ\Delta and then pull back along tt, both of which preserve colimits. Conversely a functor T→X/UT\to\mathcal X_{/U} with colimit the terminal object is cartesian over the constant diagram of terminal objects. The correspondence between families over that constant diagram and objects over its colimit follows from universal descent in an ∞\infty-topos. It gives a section U→U×ΔTU\to U\times\Delta T in the slice. This is the sense in which the indexing pairs are principal TT-actions, even when TT has many components or has higher homotopy.

The functor U↦Π(U)U\mapsto\Pi(U) sends colimits to colimits. For U≃colim⁡dUdU\simeq\operatorname{colim}_{d} U_d, the right side of (10) is the limit of the corresponding right sides. Constant tests are jointly conservative, hence the comparison from the colimit of the Π(Ud)\Pi(U_d) is an equivalence. For a geometric morphism, pullback of constant objects along f ⁣:Y→Xf\colon\mathcal Y\to\mathcal X induces a map from Π∞Y\Pi_\infty\mathcal Y to Π∞X\Pi_\infty\mathcal X. It is an equivalence exactly when pullback gives equivalences on the spaces of all constant tests.

From stage representations to objects

For (T,t)(T,t) as above and a family E→TE\to T of spaces, put Et=U×ΔTΔE⟶U. E_t=U\times_{\Delta T}\Delta E\longrightarrow U. This operation respects both the reindexing of tests and natural transformations of families. Straightening over TT and then taking the filtered colimit of stage representations gives a functor RU ⁣:Rep⁡S(Π(U))⟶X/U.R_U\colon\operatorname{Rep}_{\mathcal S}(\Pi(U)) \longrightarrow\mathcal X_{/U}. Its values are locally constant. By definition this means that there is an effective epimorphism ∐αUα→U\coprod_\alpha U_\alpha\to U such that the restriction to every UαU_\alpha is equivalent in the slice to a constant object. In this case use the fibers Pa→UP_a\to U for one point aa of each component of TT. The sum of the selected points is an effective epimorphism onto TT, and Δ\Delta and pullback preserve effective epimorphisms. Over PaP_a, the object EtE_t is constant with fiber fib⁡a(E)\operatorname{fib}_a(E). The same argument applies with S≤n\mathcal S_{\leq n} or Sπ\mathcal S_\pi as the coefficient category.

A useful reduction in proving converse classification theorems is that stage representations satisfy descent on UU. To prove this, let D\mathcal D be any chosen coefficient category, and write Wm(D)W_m(\mathcal D) for the core of Fun⁡([m],D)\operatorname{Fun}([m],\mathcal D). The complete Segal nerve of Rep⁡D(P)\operatorname{Rep}_{\mathcal D}(P) has mm-th space Wm(Rep⁡D(P))≃Map⁡(P,Wm(D)). W_m(\operatorname{Rep}_{\mathcal D}(P)) \simeq \operatorname{Map}(P,W_m(\mathcal D)). Indeed, at a stage this follows by taking the core of Fun⁡(Pi,Fun⁡([m],D))\operatorname{Fun}(P_i,\operatorname{Fun}([m],\mathcal D)). Formation of WmW_m commutes with filtered colimits because [m][m] is represented by a finite simplicial set and taking the core commutes with such colimits. Then use (4). Complete Segal nerves detect equivalences and preserve limits. Since Π(U)\Pi(U) takes colimits to colimits, (12) proves that U↦Rep⁡D(Π(U))U\mapsto\operatorname{Rep}_{\mathcal D}(\Pi(U)) sends colimits in X\mathcal X to limits. All of the spaces Wm(D)W_m(\mathcal D) may be taken in a larger universe if required by the coefficient category.

Lemma 1.1. Suppose D\mathcal D is a full subcategory of spaces and the comparison RVR_V induces an equivalence on mapping spaces between constant objects with values in D\mathcal D for every V∈XV\in\mathcal X. Then RUR_U with coefficients in D\mathcal D is fully faithful for all UU, and its essential image is the subcategory of objects locally constant with values in D\mathcal D.

Proof. Given two stage representations, choose a common stage (T,t)(T,t) and use its effective epimorphism ∐aPa→U\coprod_a P_a\to U as above. In its Čech nerve, a component of an iterated fiber product of the cover is indexed by a tuple of points and maps into one of the PaP_a’s. On that component both representations are constant. The mapping-space comparison is therefore an equivalence on each such component by the assumption. It is an equivalence on their coproduct because both representation and slice categories turn the coproduct into a product. Taking the limit over the Čech nerve gives the original mapping-space comparison. On the representation side this uses (12), and on the topos side it is effective descent.

Now take a locally constant object with the indicated coefficients. Choose an effective epimorphism giving trivializations. On every component in every degree of its Čech nerve the restrictions belong to the essential image, since constant objects have constant stage representations. Each realization functor there is fully faithful, so the coherent descent datum of this object is in the essential image of the limit of the functors. More explicitly, fullness of each image lifts all transition equivalences, and every higher simplex whose vertices are in the image lies in that full subcategory. Descent supplies a representation over UU. ◻

The constant comparison in the lemma has an elementary form. For A,B∈SA,B\in\mathcal S define FU(B)=Map⁡X(U,ΔB)F_U(B)=\operatorname{Map}_{\mathcal X}(U,\Delta B). We have natural equivalences Map⁡Rep⁡S(Π(U))(A,B)≃FU(Map⁡(A,B)),Map⁡X/U(ΔUA,ΔUB)≃Map⁡(A,FU(B)).\begin{aligned} \operatorname{Map}_{\operatorname{Rep}_{\mathcal S}(\Pi(U))}(A,B) &\simeq F_U(\operatorname{Map}(A,B)),\\ \operatorname{Map}_{\mathcal X_{/U}}(\Delta_U A,\Delta_U B) &\simeq \operatorname{Map}(A,F_U(B)). \end{aligned} The first is the formula for maps of constant diagrams at each stage. The second uses ΔUA≃colim⁡AU\Delta_U A\simeq\operatorname{colim}_A U in the slice and then maps out of that colimit. If Π(U)={Pi}\Pi(U)=\{P_i\}, the comparison is consequently colim⁡iMap⁡(A,Map⁡(Pi,B))⟶Map⁡(A,colim⁡iMap⁡(Pi,B)). \operatorname*{colim}_i \operatorname{Map}(A,\operatorname{Map}(P_i,B)) \longrightarrow \operatorname{Map}\bigl(A,\operatorname*{colim}_i \operatorname{Map}(P_i,B)\bigr). This displays the relevant commutation requirement directly.

Two comparison theorems

Theorem 1.2 (Higher local monodromy). Fix n≥0n\geq0, or let n=∞n=\infty. Suppose that the constant-object functor on nn-truncated objects admits a genuine left adjoint Ln ⁣:X≤n⟶S≤n.L_n\colon\mathcal X_{\leq n}\longrightarrow\mathcal S_{\leq n}. Then for each U∈XU\in\mathcal X, realization gives an equivalence Rep⁡S≤n(Π(U))≃LC⁡≤n(X/U).\operatorname{Rep}_{\mathcal S_{\leq n}}(\Pi(U)) \simeq \operatorname{LC}_{\leq n}(\mathcal X_{/U}).

Proof. For an nn-truncated test TT, ΔT\Delta T is nn-truncated. By adjunction, FU(T)≃Map⁡(Lnτ≤nU,T).F_U(T)\simeq \operatorname{Map}(L_n\tau_{\leq n}U,T). If BB is nn-truncated, so is Map⁡(A,B)\operatorname{Map}(A,B) for any space AA. For constant coefficients in the theorem both sides of (13) therefore agree by the exponential law for spaces, with Lnτ≤nUL_n\tau_{\leq n}U as the representing space. Apply Lemma 1.1. For n=∞n=\infty, truncation in this argument means the identity. ◻

The hypothesis of Theorem 1.2 is local nn-connectedness in the convention of [3, Def. 3.2]. For finite nn, the genuine left adjoint represents the relative shape on nn-truncated tests. Even then the monodromy category in general uses the pro-(n+1)(n+1)-truncation of shape. Indeed the objects in W0(S≤n)W_0(\mathcal S_{\leq n}) have loops given by automorphism spaces, hence this classifier is (n+1)(n+1)-truncated. The endpoint map of the arrow classifier has nn-truncated fibers, so its total space is also (n+1)(n+1)-truncated. By (12) and the mapping-space description using the arrow classifier, these two tests suffice to recognize equivalences of the representation categories. For n=0n=0 this recovers why a fundamental groupoid rather than just a space of connected components is needed to classify coverings.

Theorem 1.3 (Finite-coefficient higher Galois theory). For every ∞\infty-topos and every UU, realization induces an equivalence Rep⁡Sπ(Π(U))≃LC⁡π(X/U).\operatorname{Rep}_{\mathcal S_\pi}(\Pi(U)) \simeq \operatorname{LC}_{\pi}(\mathcal X_{/U}). Here local trivializations may have arbitrarily many different π\pi-finite fiber types across the cover.

Proof. For A∈SπA\in\mathcal S_\pi and fixed finite nn, Map⁡(A,−)\operatorname{Map}(A,-) preserves filtered colimits of nn-truncated spaces. Use the finite (n+1)(n+1)-skeleton in the proof of Lemma [lem:pifinitemap]. Mapping out of a finite cell complex is a finite limit of mapping spaces out of points, by repeated use of its cell attachments, and finite limits commute with filtered colimits in spaces. Filtered colimits of nn-truncated spaces also stay nn-truncated. Given B∈SπB\in\mathcal S_\pi, choose nn with BB nn-truncated. All spaces Map⁡(Pi,B)\operatorname{Map}(P_i,B) in (13) are nn-truncated. The comparison is therefore an equivalence. Lemma 1.1 applies without a connectivity hypothesis on X\mathcal X. ◻

Theorem 1.3 is [3, Thm. 4.3]. Its coefficients are π\pi-finite spaces, while its indexing object is the full shape. Proposition [prop:completion-comparison] permits using the profinite completion for the subcategory admitting a finite range of fiber types. The uncompleted shape can still be replaced by its protruncation for the full finite-coefficient theory when a uniform truncation bound is imposed on all fiber types. Without a uniform bound, a single classifier can have components of arbitrarily high degree. In particular, the phrase finite-coefficient classification must specify the category of coefficients and the tests of shape it claims to retain.

Consider the space B=∐m≥2K(Z/2,m).B=\coprod_{m\geq2}K(\mathbb Z/2,m). On each component, the based path fibration has π\pi-finite fiber K(Z/2,m−1)K(\mathbb Z/2,m-1). If the family came from B<∞B_{<\infty}, it would descend to some Postnikov stage τ≤dB\tau_{\leq d}B. For m>dm>d, the image of the mm-th component in that stage is contractible, so the restricted family would have to be constant up to equivalence. The based path fibration has contractible total space, whereas the constant family on K(Z/2,m)K(\mathbb Z/2,m) with this fiber has nontrivial homotopy. The family therefore uses the full shape. This obstruction concerns a collection of higher degrees rather than its fundamental group.

The classifying topos and its limit

For a space TT put β(T)=S/T\beta(T)=\mathcal S_{/T}. The classifying property needed for pro-spaces is Map⁡Top∞(X,β(T))≃Map⁡X(1,ΔT).\operatorname{Map}_{\mathrm{Top}_\infty}(\mathcal X,\beta(T)) \simeq \operatorname{Map}_{\mathcal X}(1,\Delta T). This is the étale slice universal property. The right side classifies a lift of X→S\mathcal X\to\mathcal S through S/T\mathcal S_{/T} by sending the universal point of its constant TT to the chosen point. For a pro-space P={Pi}P=\{P_i\}, define β(P)=lim⁡iβ(Pi)\beta(P)=\lim_i\beta(P_i) in Top∞\mathrm{Top}_\infty. The classifying property and (4) display an adjunction Π∞ ⁣:Top∞⇄Pro⁡(S):β.\Pi_\infty\colon\mathrm{Top}_\infty \rightleftarrows\operatorname{Pro}(\mathcal S) :\beta . Its unit is φ ⁣:X→β(Π∞X)\varphi\colon\mathcal X\to \beta(\Pi_\infty\mathcal X). The realization functor at U=1U=1 factors as Rep⁡S(Π∞X)⟶β(Π∞X)→φ∗X.\operatorname{Rep}_{\mathcal S}(\Pi_\infty\mathcal X) \longrightarrow \beta(\Pi_\infty\mathcal X) \xrightarrow{\varphi^*}\mathcal X . The first arrow is induced by the projections from the topos limit, using their inverse images, while the category underlying that limit is computed by the direct images [3, §2]. Under the local nn-connectedness hypothesis, Hoyois shows that both arrows, with nn-truncated coefficients and objects, are fully faithful. Moreover the image of β(Π∞X)≤n\beta(\Pi_\infty\mathcal X)_{\leq n} in X\mathcal X consists of filtered colimits of locally constant nn-truncated objects [3, Thm. 3.13]. Theorem 1.2 above proves the comparison on stage objects. The further claim about the whole classifying topos uses the fact that in this setting the shape admits a presentation by nn-connected transition maps.

There is a direct description of the topos limit in degree zero for P=BproHP=B_{\mathrm{pro}}H from Chapter [ch:profinite]. An object of its 00-truncation is a coherent family of discrete H/UiH/U_i-sets MiM_i. If Uj⊆UiU_j\subseteq U_i, the direct image takes invariants under Ui/UjU_i/U_j, so the compatibility is Mi≅MjUi/UjM_i\cong M_j^{U_i/U_j}. These give compatible inclusions of inflated HH-sets and their union is continuous. Conversely a continuous discrete HH-set MM yields the family of MUiM^{U_i} and is its union. The family for the union of a compatible system returns that system since a fixed element in the union is represented by a fixed element at a common refinement. The construction on maps is by restriction and union. Thus the topos limit contains all continuous discrete actions, whereas the stage category only contains actions with a common open kernel.

The extension β\beta to all pro-spaces is not fully faithful in general. Return to the tower of tails PmP_m in Chapter [ch:profinite]. The underlying category of β(Pm)\beta(P_m) is a product of copies of spaces, with one coordinate for every element of the tail. The direct images extend a tail-indexed family by terminal objects at the missing coordinates. In a compatible family, each coordinate is eventually missing, so the corresponding objects in earlier stages are terminal. Hence β(P)\beta(P) is the terminal category, the topos β(∅)\beta(\varnothing). But PP and ∅\varnothing were distinguished by their constant empty test. This is the general pro-space obstruction recorded in [3, Rem. 2.8].

A failure of unrestricted stage monodromy

Let C={0,1}NC=\{0,1\}^{\mathbb N} with its product topology, and X=Shv⁡(C;S)\mathcal X=\operatorname{Shv}(C;\mathcal S). Its shape is the system of finite clopen partitions of CC, regarded as discrete spaces and ordered by refinement. Here is a direct computation. A basis is given by the clopen subsets, each of which is compact. A covering of a clopen subset by basis members has a finite disjoint clopen refinement, by choosing a finite subcover and successively subtracting earlier sets. Thus the topology on the basis is generated by finite disjoint clopen covering families. The constant sheaf with value TT takes a clopen VV to colim⁡P∏D∈PT,\operatorname*{colim}_{\mathcal P} \prod_{D\in\mathcal P} T, over finite clopen partitions of VV. This prescription satisfies the disjoint-cover sheaf condition because partitions refining a fixed disjoint cover are cofinal and filtered colimits commute with finite products. Its map from the constant presheaf exhibits sheafification, since any natural map to a sheaf is forced, on a partition, by the maps from its parts and the disjoint-cover sheaf property. Taking V=CV=C computes every constant test of shape.

Consider the two constant sheaves with values N\mathbb N and {0,1}\{0,1\}. A map between them can be specified by a sequence of locally constant functions on CC with values in {0,1}\{0,1\}. Indeed the source is a coproduct of countably many terminal objects. Choose the functions giving the successive coordinates of CC. The sequence is a morphism of the constant sheaves, but no single finite partition simultaneously makes every coordinate function constant on each part. Such a partition would make the identity map of the underlying set of CC factor through a finite set. The sequence therefore lies outside the image of the stage morphism space (13). This gives a failure of full faithfulness for unrestricted set-valued stage monodromy on a topos with profinite 00-dimensional shape. The finite-coefficient theorem still applies to this topos.

Covering spaces and higher transport

For a space BB, the classifying topos S/B≃Fun⁡(B,S)\mathcal S_{/B}\simeq\operatorname{Fun}(B,\mathcal S) has shape BB itself. Its global sections of a constant object TT are Map⁡(B,T)\operatorname{Map}(B,T). All its objects are locally constant, as pullback to the family of points of BB gives their fibers. When BB is connected and pointed, set-valued local systems are π1(B)\pi_1(B)-sets, since their classifying core and arrow core are both 11-truncated. The fiber functor at bb recovers π1(B,b)\pi_1(B,b) as its group of natural automorphisms. For left GG-sets the forgetful functor is Hom⁡G(G,−)\operatorname{Hom}_G(G,-). Equivariant self-maps of the left regular GG-set are right translations, so its automorphism group is GopG^{\mathrm{op}}. The contravariance of the Yoneda embedding in the representing object then gives GG as the automorphism group of the fiber functor. This realizes the fundamental-group construction of [1, §6.7] in the representable case. For ordinary locally path-connected semilocally simply connected spaces, locally constant sheaves of sets are the sheaves of sections of covering spaces, and path lifting gives the same monodromy description on the ordinary fundamental groupoid. These local topological hypotheses ensure that every monodromy action reconstructs a cover.

Positive homotopy groups can enter even for connected base with trivial π1\pi_1. Take B=K(Z,2)B=K(\mathbb Z,2), pointed at its unit. The based path fibration PbB→BP_bB\to B has fiber K(Z,1)K(\mathbb Z,1) and contractible total space. The constant family with this fiber has total space B×K(Z,1)B\times K(\mathbb Z,1), which is not contractible. Thus these are inequivalent families over BB. Their set-valued monodromy cannot distinguish them, while the degree-two shape allows transport of the homotopies in the fiber.

Hypercompletion and shape

Let h ⁣:X∧→Xh\colon\mathcal X^\wedge\to\mathcal X be the hypercompletion. Every truncated object is hypercomplete. Indeed an ∞\infty-connected morphism is left orthogonal to truncated objects over the terminal object, so localizing at those morphisms fixes every truncated object. For each truncated test TT, ΔT\Delta T is already truncated in X\mathcal X, and Map⁡X(1,ΔT)≃Map⁡X∧(1,h∗ΔT).\operatorname{Map}_{\mathcal X}(1,\Delta T) \simeq \operatorname{Map}_{\mathcal X^\wedge} (1,h^*\Delta T). Consequently hypercompletion induces equivalences on protruncated shape and on its profinite completion, as also observed in [3, Rem. 5.3]. That argument only tests truncated targets. Unbounded constant tests and filtered colimits of such tests require separate hypotheses. Likewise an attempted computation of full shape from Postnikov sections cannot exchange the uniformly bounded tests with their limit without establishing the required commutation. The distinction is quantified already by (4), which places a filtered colimit inside the limit over target stages.

For a concrete pro-space with trivial protruncation, use the tower Qi=∏m≥iK(Z,m),Q_i=\prod_{m\geq i}K(\mathbb Z,m), with transitions that insert basepoints in the coordinates new to the target stage. For fixed dd the stages i>di>d are dd-connected, so every dd-truncated test treats QQ as terminal. Yet the compatible inclusions Qi→Q1Q_i\to Q_1 define a point of Map⁡(Q,Q1)\operatorname{Map}(Q,Q_1) that is not in the component of a constant map. At any refinement, the coordinate inclusion followed by projection to that coordinate induces the identity of K(Z,m)K(\mathbb Z,m). A constant homotopy class would yield the zero map on its positive homotopy group. This proves that unbounded tests can see data invisible in every fixed truncation.

Obstructions to higher monodromy

A classical action on a discrete set is determined by its permutation representation. For a space FF, actions on FF are classified by maps into BAut(F)B\mathop{\mathrm{Aut}}(F), whose higher homotopy groups control extensions of the permutation-level information. This chapter describes the lifting problem without assuming that the base has a discrete fundamental group as its entire homotopy type. The argument applies also to maps from a pro-space into a bounded target.

Automorphisms of an Eilenberg–Mac Lane space

For an abelian group AA and n≥1n\geq 1, choose a functorial Eilenberg–Mac Lane space K(A,n)K(A,n) represented by a simplicial abelian group. Transport through another model leaves the following equivalences invariant. Let Aut∗\mathop{\mathrm{Aut}}_* denote the space of based self-equivalences, with the identification at the basepoint included as data.

Lemma 1.1. The map Aut∗(K(A,n))→Aut(A)\mathop{\mathrm{Aut}}_*(K(A,n))\to\mathop{\mathrm{Aut}}(A) is an equivalence onto a discrete group. The full group object of self-equivalences is equivalent to Aut(K(A,n))≃K(A,n)⋊Aut(A),\mathop{\mathrm{Aut}}(K(A,n))\simeq K(A,n)\rtimes\mathop{\mathrm{Aut}}(A), with the translation action on the first factor and the usual action of coefficient automorphisms.

Proof. Pointed maps K(A,n)→K(A,n)K(A,n)\to K(A,n) have components H~n(K(A,n);A)≅Hom⁡(A,A)\widetilde H^n(K(A,n);A)\cong\operatorname{Hom}(A,A). For n≥2n\geq2, this follows from connectivity, Hurewicz, and the universal coefficient theorem. For n=1n=1, the identification uses H1(K(A,1);Z)=AH_1(K(A,1);\mathbb Z)=A and the freeness of H0H_0. In both cases the higher homotopy groups of the pointed mapping space are H~n−i(K(A,n);A)\widetilde H^{n-i}(K(A,n);A) in degree i>0i>0, by suspension and the representing property of Eilenberg–Mac Lane spaces. They vanish by connectivity and reducedness. A pointed map is an equivalence precisely when the induced endomorphism of AA is invertible. The based maps induced by automorphisms of AA therefore give an equivalence from the discrete group.

Evaluation of an unbased self-map at zero has this pointed mapping space as its fiber. Translation by the evaluation value provides a splitting on underlying spaces. In the chosen additive model the assignment (a,α)↦(x↦a+α(x))(a,\alpha)\mapsto(x\mapsto a+\alpha(x)) is a map of group objects, using the semidirect product law. Its effect on the evaluation fibers and bases is an equivalence, which proves the assertion. ◻

Delooping gives a fibration with a distinguished section K(A,n+1)⟶BAut(K(A,n))⟶BAut(A). K(A,n+1)\longrightarrow B\mathop{\mathrm{Aut}}(K(A,n)) \longrightarrow B\mathop{\mathrm{Aut}}(A). The section is induced by the inclusion of coefficient automorphisms in the semidirect product. In the description by families over BAut(A)B\mathop{\mathrm{Aut}}(A), the fibration is the local system of spaces K(A,n+1)K(A,n+1) with its natural transport. Indeed looping the quotient retrieves the stated conjugation action on K(A,n)K(A,n), and the section identifies the neutral point in each fiber.

A local system of abelian groups on a space YY is a functor M:Y→AbM:Y\to\mathrm{Ab}, where Ab\mathrm{Ab} is the ordinary category of abelian groups and all arrows in the image of YY are isomorphisms. Write KY(M,r)→YK_Y(M,r)\to Y for the family associated to the functor y↦K(My,r)y\mapsto K(M_y,r). Define Hr(Y;M)H^r(Y;M) to be the component group of the section space ΓYKY(M,r)\Gamma_Y K_Y(M,r). This agrees with cohomology with local coefficients. In particular the zero section is part of the structure, and translation identifies each section component with the component of zero. For 0≤i≤r0\leq i\leq r, looping the fiberwise group and using that sections preserve limits yields πi(ΓYKY(M,r),0)=Hr−i(Y;M).\pi_i\bigl(\Gamma_Y K_Y(M,r),0\bigr)=H^{r-i}(Y;M). For i>ri>r the homotopy group vanishes. This constructive description needs no choice of cocycle model for the base.

The Postnikov lifting problem

Let ZZ be a connected space, n≥2n\geq2, and Y=τ≤n−1ZY=\tau_{\leq n-1}Z. The map τ≤nZ→Y\tau_{\leq n}Z\to Y has fiber K(A,n)K(A,n) for A=πnZA=\pi_nZ at a chosen basepoint, by the homotopy exact sequence. Transport gives the local coefficient system MM associated to the usual π1(Z)\pi_1(Z)-action on AA. The relative assertion below uses the full section space.

Theorem 1.2 (A Postnikov step as a space of nullhomotopies). The Postnikov extension τ≤nZ→Y\tau_{\leq n} Z\to Y determines a section kk of KY(M,n+1)K_Y(M,n+1) defined up to equivalence with its classifying data. The extension is equivalent over YY to the homotopy pullback of kk and the zero section. Given a map f:T→Yf:T\to Y, the space of lifts of ff to τ≤nZ\tau_{\leq n}Z is equivalent to the space of paths from 00 to f∗kf^*k in ΓTKT(f∗M,n+1)\Gamma_T K_T(f^*M,n+1). It is nonempty exactly when [f∗k]=0[f^*k]=0 in Hn+1(T;f∗M)H^{n+1}(T;f^*M). If a lift is fixed, the full lift space is equivalent to ΓTKT(f∗M,n)\Gamma_T K_T(f^*M,n).

Proof. A family over YY whose fibers are equivalent to K(A,n)K(A,n) is classified by a map Y→BAut(K(A,n))Y\to B\mathop{\mathrm{Aut}}(K(A,n)), by straightening and the component of K(A,n)K(A,n) in the core of spaces. Composing with (14) gives the classifying map of MM. Over that map, a lift to BAut(K(A,n))B\mathop{\mathrm{Aut}}(K(A,n)) is exactly a section kk of KY(M,n+1)K_Y(M,n+1). One can identify the classified family itself by pulling back the bundle of paths that start at zero in the fiber K(A,n+1)K(A,n+1). Such paths have endpoint k(y)k(y) and their path space is a principal K(A,n)K(A,n)-space. Under trivialized coefficients this is the universal path fibration, whose transport consists of translations. Coefficient automorphisms respect zero and act functorially on these paths. Across a change of coefficients the transport thus combines translations with automorphisms by the semidirect product of Lemma 1.1. This is the universal family for the classifying map from which kk was obtained.

Pulling back along ff and taking sections preserves the pullback because sections are represented by the terminal object in the slice. The stated path space therefore gives all lifts and their homotopies. If one path is fixed, composing with its inverse identifies this path space with the loop space at zero. Looping the fiberwise Eilenberg–Mac Lane space and then taking sections proves the final claim. ◻

In particular, when the obstruction is zero the set of equivalence classes of lifts is a torsor for Hn(T;f∗M)H^n(T;f^*M); if a lift is chosen, its higher automorphisms in the lift space are given by the lower cohomology groups. Applying this statement to the Postnikov tower of BAut(F)B\mathop{\mathrm{Aut}}(F) gives successive lifting problems for monodromy with prescribed fiber FF. The first stage records the discrete group π0Aut(F)\pi_0\mathop{\mathrm{Aut}}(F), and later stages can change both existence and the space of choices. The usual Postnikov construction of spaces is reviewed in [18], and the corresponding viewpoint on cohomology in slices is developed in [4].

Example 1.3 (A nonzero obstruction over a torus). Take Y=BZ3Y=B\mathbb Z^3, the homotopy type of the three-torus, M=ZM=\mathbb Z with trivial transport, and n=2n=2. A generator kk of H3(Y;Z)H^3(Y;\mathbb Z) defines a family Z→YZ\to Y as the fiber of k:Y→K(Z,3)k:Y\to K(\mathbb Z,3). Its fibers are K(Z,2)K(\mathbb Z,2) and the homotopy sequence shows that ZZ is a connected 22-type with τ≤1Z≃Y\tau_{\leq1}Z\simeq Y. The identity of YY cannot lift. The map T3→T3\mathbb T^3\to\mathbb T^3 that sends the third coordinate to its identity and leaves the first two coordinates fixed does lift, because it factors through the two-torus. The lift components in the second case form a torsor for H2(T3;Z)=Z3H^2(\mathbb T^3;\mathbb Z)=\mathbb Z^3. At a chosen lift the degree-one and degree-two homotopy groups of the lift space are Z3\mathbb Z^3 and Z\mathbb Z, respectively. All these claims follow from the exterior-algebra cohomology of the torus and Theorem 1.2. They distinguish the vanishing test, the moduli of solutions, and the group of automorphisms of a solution.

Continuous lifting and convergence

Let P=(Pi)P=(P_i) be a pro-space as in Chapter [ch:profinite]. Suppose that a partial monodromy map to a space YY is represented by fi:Pi→Yf_i:P_i\to Y. For jj over ii, write fj:Pj→Yf_j:P_j\to Y for its pullback. The lifting problem for a target map Z→YZ\to Y has a useful stage description.

Proposition 1.4. There is a natural equivalence fib⁡[fi](MapPro(S)(P,Z)⟶MapPro(S)(P,Y))≃colimj→ifib⁡fj(Map(Pj,Z)⟶Map(Pj,Y)).\begin{aligned} &\operatorname{fib}_{[f_i]} \bigl(\mathop{\mathrm{Map}}_{\mathop{\mathrm{Pro}}(\mathcal{S})}(P,Z)\longrightarrow\mathop{\mathrm{Map}}_{\mathop{\mathrm{Pro}}(\mathcal{S})}(P,Y)\bigr)\\ &\qquad\simeq \mathop{\mathrm{colim}}_{j\to i} \operatorname{fib}_{f_j}\bigl(\mathop{\mathrm{Map}}(P_j,Z)\longrightarrow\mathop{\mathrm{Map}}(P_j,Y)\bigr). \end{aligned} For a Postnikov step as in Theorem 1.2, a lift exists exactly when the pulled-back obstruction vanishes at some refining stage.

Proof. Restricting to refinements of the chosen stage is cofinal in the filtered direction. The map in the proposition then compares the filtered colimits of the indicated diagrams of mapping spaces. The compatible basepoint diagram is constant at a point and induces [fi][f_i]. Filtered colimits of spaces commute with the pullback defining its fiber. This proves the equivalence. A filtered colimit of spaces has a component exactly when some stage does, since π0\pi_0 preserves colimits. The result now follows from the vanishing criterion above. ◻

A space ZZ is the limit of its Postnikov tower. There is therefore a comparison of spaces colimilim⁡nMap(Pi,τ≤nZ)⟶lim⁡ncolimiMap(Pi,τ≤nZ). \mathop{\mathrm{colim}}_i\lim_n \mathop{\mathrm{Map}}(P_i,\tau_{\leq n} Z) \longrightarrow \lim_n\mathop{\mathrm{colim}}_i \mathop{\mathrm{Map}}(P_i,\tau_{\leq n} Z). The left side classifies maps from PP to the constant pro-space ZZ. The right side classifies maps from PP to the pro-object consisting of the truncations of ZZ. A lift for each separate obstruction need not use one fixed stage ii. A bounded target, or a constant source, gives an equivalence in (15). The following computation shows how the two sides differ without either condition.

Example 1.5 (A compatible tower with no common stage). Fix an abelian group A≠0A\neq 0, for instance Z\mathbb Z. Let Pi=∏k≥iK(A,k)P_i=\prod_{k\geq i}K(A,k) for i≥2i\geq2, with maps Pi+1→PiP_{i+1}\to P_i given by inserting the basepoint in the iith coordinate. Let Z=P2Z=P_2. Every PiP_i is (i−1)(i-1)-connected. For fixed nn, the map Map(∗,τ≤nZ)→Map(Pi,τ≤nZ)\mathop{\mathrm{Map}}(*,\tau_{\leq n} Z)\to\mathop{\mathrm{Map}}(P_i,\tau_{\leq n}Z) is an equivalence for i>ni>n, since the target is nn-truncated. The right side of (15) is consequently ZZ and is connected.

On the left side, inclusions Pi→ZP_i\to Z represent one compatible element of colimiπ0Map(Pi,Z)\mathop{\mathrm{colim}}_i\pi_0\mathop{\mathrm{Map}}(P_i,Z). This element differs from the component of the constant map. Indeed, for every refining stage PjP_j, restrict the inclusion to K(A,j)K(A,j) using the jjth factor and project ZZ to that factor. The composite is the identity and induces idA\mathop{\mathrm{id}}_A on πj\pi_j, whereas the constant map induces zero. Thus every bounded observation becomes trivial after refinement, while the unbounded monodromy test detects a nontrivial component.

There is a second issue even when the source is an ordinary space. The inverse limit of the spaces of lifts encodes compatibility among choices. Passing to their component sets before forming the limit can lose information. For a countable tower of pointed spaces, the homotopy limit is the homotopy equalizer of the identity and shift on the product. An explicit point of the homotopy equalizer consists of a point in each space and a path from that point to the image of the next point. This model yields a direct example.

Example 1.6 (The degree-pp tower of circles). For a prime pp, take the tower BZB\mathbb Z with transition maps induced by multiplication by pp. It is a tower of abelian group objects in spaces. The difference of the identity and shift on the product is induced on π1\pi_1 by D:∏i≥0Z⟶∏i≥0Z,(ai)⟼(ai−pai+1).D:\prod_{i\geq0}\mathbb Z\longrightarrow\prod_{i\geq0}\mathbb Z, \qquad (a_i)\longmapsto(a_i-pa_{i+1}). The homotopy equalizer is its fiber as a map of group objects, so the long exact sequence identifies its component group with coker⁡D\operatorname{coker}D. The map (bi)↦∑i≥0pibi(b_i)\mapsto \sum_{i\geq0}p^i b_i converges in Zp\mathbb Z_p, is surjective by the pp-adic expansion, and takes the image of DD into the ordinary integers by telescoping. Conversely, if the sum is an integer ss, put ai=p−i(s−∑k<ipkbk)∈Z.a_i=p^{-i}\Big(s-\sum_{k<i}p^k b_k\Big)\in\mathbb Z. Divisibility follows from the pp-adic equality, and ai−pai+1=bia_i-pa_{i+1}=b_i. Thus the component group is Zp/Z\mathbb Z_p/\mathbb Z, although the component set at each stage is a singleton. This is the elementary tower calculation behind the derived inverse limit term.

The two examples locate different coherence requirements. Maps into a pro-target can be compatible even when they have no common source stage. Choices in a tower of lift spaces can have additional components carried by the transition paths. The classification space keeps both effects visible.

Algebraic Galois theories

Pierce spectra and derived étale descent

The ring-theoretic construction in [1, Chapter 4 and Appendix A] changes two ingredients of field-theoretic Galois theory. It uses the entire space of connected components as the space of objects of the Galois groupoid, and it interprets split algebras by sections of a sheaf of rings over a Stone space. Both constructions extend to commutative ring spectra. This extension has a particularly concrete feature. Idempotents determine the topology at every level of the construction, while the algebras of sections retain all the homotopy groups of the ring spectrum. The derived affine category and its Pierce reflection therefore carry different amounts of information.

Throughout this chapter, AA denotes an E∞E_\infty-ring, possibly nonconnective, CAlg⁡A\operatorname{CAlg}_A its ∞\infty-category of commutative algebras, and YY a Stone space, that is, a compact Hausdorff space with a basis of clopen subsets. The empty space and the zero ring are allowed. Write P(A)P(A) for the Stone spectrum of the Boolean algebra Idem⁡(π0A)\operatorname{Idem}(\pi_0 A). Points are ultrafilters. For an idempotent ee, the clopen subset D(e)D(e) consists of ultrafilters containing ee, and every clopen of P(A)P(A) arises uniquely in this way. A ring map A→CA\to C gives a continuous map P(C)→P(A)P(C)\to P(A) by inverse image of ultrafilters. All tensor products between ring spectra below are derived.

Idempotents, localizations, and stalks

Multiplication, complement, and join in Idem⁡(π0A)\operatorname{Idem}(\pi_0 A) are given by efef, 1−e1-e, and e+f−efe+f-ef. These formulas only use degree zero. The localization A[e−1]A[e^{-1}], by contrast, is an algebra over the entire spectrum. Its universal property says that the mapping space from A[e−1]A[e^{-1}] into an AA-algebra CC is contractible when ee maps to a unit of π0C\pi_0 C, and empty otherwise. An idempotent is a unit exactly when it is the element 11. Localization satisfies πt(A[e−1])≅(πtA)[e−1]≅eπtA(t∈Z).\pi_t(A[e^{-1}])\cong (\pi_t A)[e^{-1}]\cong e\pi_t A \qquad(t\in\mathbf Z). Here the last module is viewed over eπ0Ae\pi_0 A, whose unit is ee. These facts are instances of spectral localization [6, Example 7.5.0.7]. They imply A→ ∼ A[e−1]×A[(1−e)−1].A\xrightarrow{\ \sim\ } A[e^{-1}]\times A[(1-e)^{-1}]. Indeed the map induces m↦(em,(1−e)m)m\mapsto(em,(1-e)m) on every homotopy group. The proof also applies to any finite partition of unity by pairwise orthogonal idempotents. Equivalences of ring spectra are detected on the underlying spectra.

Proposition 1.1. The rule D(e)↦A[e−1]D(e)\mapsto A[e^{-1}], with the localization maps as restrictions, defines a sheaf on the clopen basis of P(A)P(A). It extends uniquely, up to equivalence, to a sheaf of E∞E_\infty-rings on all open subsets. Its global sections are AA, and its stalk at an ultrafilter xx is Ax≃colim⁡e∈xA[e−1],A_x\simeq\underset{e\in x}{\operatorname{colim}} A[e^{-1}], where the transition maps restrict along inclusions of smaller clopens. The ring π0Ax\pi_0 A_x has precisely the two idempotents 00 and 11.

Proof. Every cover of a clopen by clopens has a finite subcover and a finite pairwise disjoint clopen refinement formed by intersections and complements. For a disjoint cover the sheaf comparison is exactly the product decomposition above, and intersections of distinct members have the zero ring as sections. The finite refinements yield the sheaf condition on the basis. Explicitly, for a finite cover, decompose every member and every overlap by the same finite Boolean partition. On each nonempty part, the indices of cover members containing it form a nonempty set, with a contractible Čech indexing groupoid. The limit of the restrictions over that part is its section ring. Taking the product over parts recovers the original section ring. The same argument applies to an arbitrary clopen cover by restricting over each part of a finite subordinate disjoint partition. On each part a chosen member contains the entire part, so its restricted cover has a section and the Čech limit is split. The extension and the stalk formula follow by restriction to the basis.

Filtered colimits commute with π0\pi_0, and the polynomial identity defining an idempotent descends to a stage. Every idempotent in eπ0Ae\pi_0 A is an idempotent ff of π0A\pi_0 A with f≤ef\leq e. An ultrafilter containing ee contains either ff or 1−f1-f. Restrict along the former to make the idempotent 11 and along the latter to make it 00. Each ring at a stage has distinct unit and zero because its idempotent belongs to an ultrafilter. They remain distinct in the filtered colimit. ◻

An ultrafilter stalk is indecomposable in the sense of products. The Zariski spectrum instead organizes tests at prime ideals and uses local rings. For degree-zero AA, Proposition 1.1 recovers the Pierce sheaf in [1, Section 4.2]. Its spectral analog shows exactly where the extra information resides. The stalks and restrictions retain the groups eπtAe\pi_t A before passage to the colimit.

A relative Stone adjunction

Let q:Y→P(A)q:Y\to P(A) be continuous. To build its algebra of sections, take pairs (U,W)(\mathcal U,W) where U\mathcal U is a finite partition of YY into nonempty clopens and W=(WU)U∈UW=(W_U)_{U\in\mathcal U} is a family of clopens in P(A)P(A) satisfying q(U)⊆WUq(U)\subseteq W_U. For the empty partition use the empty product, the zero algebra. A refinement partitions into smaller clopens and shrinks each chosen neighborhood inside the neighborhood of its parent. Denote by eWUe_{W_U} the associated idempotent and put A⟨Y⟩=colim⁡(U,W)∏U∈UA[eWU−1]. A\langle Y\rangle =\underset{(\mathcal U,W)}{\operatorname{colim}} \prod_{U\in\mathcal U} A[e_{W_U}^{-1}]. The transition map to a refinement repeats the old factor for each of its children and then restricts it along the smaller neighborhood. Intersections of partitions and neighborhoods give common refinements, so this is a filtered colimit. No point of q(U)q(U) is required to have an open image, which is the reason for using neighborhoods of the images.

Lemma 1.2. For an AA-algebra CC, the space of maps Map⁡CAlg⁡A(∏j=1rA[ej−1],C)\operatorname{Map}_{\operatorname{CAlg}_A} \left(\prod_{j=1}^r A[e_j^{-1}],C\right) is discrete. Its points are ordered clopen partitions P(C)=∐jVjP(C)=\coprod_j V_j, where empty parts are allowed and each VjV_j maps into D(ej)D(e_j) in P(A)P(A).

Proof. The factors in the source define orthogonal idempotents summing to 11, whose images in π0C\pi_0 C give the partition (Vj)(V_j). After this partition is fixed, the target decomposes as ∏jC[vj−1]\prod_j C[v_j^{-1}], for the associated idempotents vjv_j. Mapping to the jj-th factor forces the source idempotent of the jj-th factor to map to 11, so the map factors through the jj-th source factor. The universal property of localization makes this factorwise mapping space contractible exactly when ejvj=vje_j v_j=v_j. To see that no higher ambiguity enters through the partition, one can use [6, Corollary 7.5.4.6]. The source is a spectral étale AA-algebra because its degree-zero algebra is a finite product of idempotent localizations and its higher groups obey the flat base-change formula. Thus mappings to an arbitrary target are determined discretely by degree zero. The Boolean condition ejvj=vje_jv_j=v_j is equivalent to the asserted containment of clopens. ◻

Theorem 1.3 (Relative Pierce adjunction). There is a natural equivalence Map⁡CAlg⁡A(A⟨Y⟩,C)≃Hom⁡Stone/P(A)(P(C),Y).\operatorname{Map}_{\operatorname{CAlg}_A}(A\langle Y\rangle,C) \simeq \operatorname{Hom}_{\mathsf{Stone}/P(A)}(P(C),Y). The canonical continuous map Y→P(A⟨Y⟩)Y\to P(A\langle Y\rangle) is an isomorphism over P(A)P(A). Consequently Y↦A⟨Y⟩Y\mapsto A\langle Y\rangle embeds (Stone/P(A))op(\mathsf{Stone}/P(A))^{\mathrm{op}} fully faithfully into CAlg⁡A\operatorname{CAlg}_A. Its image will be called the Pierce-split AA-algebras. On opposite categories the left adjoint is the functor C↦P(C)C\mapsto P(C) over P(A)P(A). The corresponding reflection is therefore represented on algebras by the map A⟨P(C)⟩→CA\langle P(C)\rangle\to C, and its fixed objects are exactly the Pierce-split algebras.

Proof. For a pair d=(U,W)d=(\mathcal U,W), set Yd=∐U∈UWUY_d=\coprod_{U\in\mathcal U}W_U over P(A)P(A). The basis containment condition in Lemma 1.2 gives the mapping set from P(C)P(C) to YdY_d over P(A)P(A). The finite-stage ring has Pierce spectrum YdY_d, since the Boolean algebra of its idempotents is the finite product of those of its factors. A refinement gives Yd′→YdY_{d'}\to Y_d by inclusion into parent neighborhoods. There is an isomorphism in Stone spaces over P(A)P(A) Y→ ∼ lim⁡dYd.Y\xrightarrow{\ \sim\ }\lim_d Y_d. For nonempty YY, a point of the limit has a single base coordinate x∈P(A)x\in P(A), as is seen by considering the coarsest partition with neighborhood P(A)P(A). The coordinates at the unshrunk partitions choose a compatible member of each finite clopen partition. They determine a unique y∈Yy\in Y by compactness and separation by clopens. For any clopen neighborhood of q(y)q(y), continuity gives a clopen piece about yy mapping into that neighborhood, and the corresponding coordinate forces xx to lie in it. Hence x=q(y)x=q(y). Conversely yy supplies all coordinates. The canonical continuous bijection is a homeomorphism by compactness and Hausdorffness. The statement for the empty space follows from the empty partition.

Mapping out of the colimit in (16) now gives the displayed equivalence since limits of discrete spaces remain discrete. Independently, filtered colimits of E∞E_\infty-rings induce filtered colimits of Boolean algebras of idempotents. An element and the equation making it idempotent occur at a finite stage, and equalities hold after a further refinement. Stone duality identifies the spectrum of that colimit with lim⁡dYd\lim_d Y_d. This proves the second claim and identifies the two constructions naturally. The last statement follows by taking C=A⟨Z⟩C=A\langle Z\rangle. ◻

For an ordinary ring RR, formula (16) gives the ring of sections of the inverse-image Pierce sheaf over YY. A section of the inverse image is locally represented by a section on a clopen neighborhood of its image. Compactness gives a finite clopen partition subordinate to these representatives. If two collections of representatives define the same section, local equality of their germs and a second finite clopen refinement make their representatives equal on sufficiently small clopen neighborhoods. Thus they become equal at a common stage. This identifies the construction with the relative section functor of [1, Section 4.3], including its variance. For a ring spectrum the same formula computes each homotopy group by taking the colimit of the corresponding finite product of modules eWUπtAe_{W_U}\pi_t A. Each finite-stage ring is finite étale over AA, since its ordinary degree-zero algebra is a finite product of direct summands of the unit and its homotopy groups satisfy flat base change. Thus the Pierce-split algebras are filtered colimits of these particular finite étale algebras.

Example 1.4 (A compact parameter space). Suppose π0A\pi_0 A is nonzero with no nontrivial idempotents and Y={0,1}NY=\{0,1\}^{\mathbf N} is the Cantor space. The relative algebra is the filtered colimit of products of copies of AA indexed by clopen partitions. Its degree-tt group is the group of locally constant functions Y→πtAY\to\pi_t A with the target discrete. Every such function has finite image by compactness. For degree zero, even the characteristic function of a single point lies outside this group, since that point is closed and fails to be open. The product ∏y∈YA\prod_{y\in Y}A has all functions as homotopy groups and has the Boolean algebra of all subsets of YY as degree-zero idempotents. The two constructions distinguish continuous parameters from arbitrary parameters.

Corollary 1.5 (Base change). For a map A→CA\to C, there are natural equivalences C⊗AA⟨Y⟩≃C⟨Y×P(A)P(C)⟩,A⟨Y⟩⊗AA⟨Z⟩≃A⟨Y×P(A)Z⟩.C\otimes_A A\langle Y\rangle \simeq C\langle Y\times_{P(A)}P(C)\rangle, \qquad A\langle Y\rangle\otimes_A A\langle Z\rangle \simeq A\langle Y\times_{P(A)}Z\rangle. In particular, PP takes an affine pullback to the corresponding Stone pullback if one leg is the opposite of a Pierce-split algebra.

Proof. Apply the universal property of pushout to maps into an arbitrary CC-algebra DD. The right side of Theorem 1.3 then describes lifts of P(D)→P(A)P(D)\to P(A) to YY, which are precisely maps over P(C)P(C) to the indicated pullback. The first equivalence follows from Yoneda. Taking C=A⟨Z⟩C=A\langle Z\rangle gives the second. ◻

The Galois groupoid of a normal Pierce splitting

We now impose an effectivity hypothesis separately from the classification of split algebras. For A→BA\to B, write Bn=B⊗A(n+1)B_n=B^{\otimes_A(n+1)}, with Amitsur cofaces given by insertion of units and codegeneracies by multiplication. Say the map has affine algebra descent when extension of scalars induces CAlg⁡A→ ∼ lim⁡[n]∈ΔCAlg⁡Bn. \operatorname{CAlg}_A \xrightarrow{\ \sim\ } \lim_{[n]\in\Delta}\operatorname{CAlg}_{B_n}. In this limit, the transition functors are extensions of scalars. Say that the map is normal for Pierce splitting when B⊗ABB\otimes_A B, viewed as a BB-algebra using the first factor, is Pierce-split. Symmetry gives the condition for the second factor. This condition is adapted to the inverse-limit construction above. It refers to one specific reflection, whereas affine algebra descent refers to the ambient affine category.

Theorem 1.6 (Pierce Galois correspondence). If A→BA\to B has affine algebra descent and is normal for Pierce splitting, the simplicial Stone space Gn=P(Bn)G_n =P(B_n) is an internal groupoid. The opposite of the full subcategory of CAlg⁡A\operatorname{CAlg}_A on those DD for which B⊗ADB\otimes_A D is Pierce-split is equivalent to the category of internal G∙G_\bullet-actions in Stone spaces. The underlying space of the action associated to DD is P(B⊗AD)P(B\otimes_A D) over G0=P(B)G_0=P(B).

Proof. The tensor-power algebra in degree nn, with its map from the factor at any vertex, is the nn-fold tensor product over BB of B⊗ABB\otimes_A B with a common selected factor. Normality and Corollary 1.5 make it Pierce-split over that vertex. The usual Segal decomposition of the affine Čech nerve expresses its degree nn as the iterated pullback of nn consecutive arrows over Spec⁡(B)\operatorname{Spec}(B), where Spec⁡\operatorname{Spec} here denotes passage to opposite algebras. Each pullback can use a Pierce-split leg at its vertex. Corollary 1.5 therefore identifies GnG_n with the corresponding Stone pullback of composable arrows. Applying the same corollary to the first and last vertices of a triangle, or to either pair of edges with a common tail or head, gives the outer horn isomorphisms for inverses. Concretely, the exchange of the two factors in B1B_1 supplies inversion, and its identities with multiplication follow from the corresponding identities in the affine Čech groupoid. Thus G∙G_\bullet is a groupoid object.

Under (17), requiring the object at degree zero to be Pierce-split is equivalent to requiring it at all degrees. Indeed each comparison with a vertex identifies the object at degree nn with an extension of scalars of its degree-zero value, and Pierce splitting is preserved by extension of scalars. The equivalence of Theorem 1.3 is natural for precisely these transition functors by Corollary 1.5. Taking opposites therefore identifies the desired category with lim⁡[n]∈ΔStone/Gn.\lim_{[n]\in\Delta} \mathsf{Stone}_{/G_n}. For an internal groupoid this limit is the category of internal actions. Given an object V→G0V\to G_0, an identification of its two pullbacks to G1G_1 is an arrowwise transport map. The degenerate arrows impose its unit law and the pullbacks over G2G_2 impose composition. The Segal isomorphisms identify all further compatibilities with iteration of these two laws. Conversely a unitary associative action gives that identification and all its extensions by pullback. Morphisms are maps over G0G_0 commuting with the transport. This proves the equivalence and the claim about the underlying space. ◻

The condition on B⊗ABB\otimes_A B also gives the closure of the trivial objects under the special base change that transports a BB-algebra along A→BA\to B. For D=B⟨Y⟩D=B\langle Y\rangle, first apply Corollary 1.5 with the second BB-structure on B⊗ABB\otimes_A B. The resulting algebra is Pierce-split over B⊗ABB\otimes_A B. If the latter, with its first BB-structure, corresponds to a Stone space TT over P(B)P(B), the composite is Pierce-split over the first BB. To verify this last assertion, express each algebra over B⊗ABB\otimes_A B using (16). A finite product of clopen factors of B⟨T⟩B\langle T\rangle corresponds over BB to a finite union of clopen subspaces of TT, by the relative mapping formula. Filtered colimits of these algebras correspond to inverse limits of their Stone spaces by that same formula. Conversely, taking D=BD=B in the closure condition recovers normality. In the ordinary ring setting, these are the split-object and groupoid mechanisms in [1, Sections 4.5–4.7].

Flat descent and the geometric étale site

Call A→BA\to B faithfully flat when π0A→π0B\pi_0 A\to\pi_0 B is faithfully flat and πtA⊗π0Aπ0B→ ∼ πtB(t∈Z).\pi_t A\otimes_{\pi_0 A}\pi_0 B\xrightarrow{\ \sim\ }\pi_t B \qquad (t\in\mathbf Z). For any AA-module MM, flat base change gives πt(M⊗AB)=πtM⊗π0Aπ0B\pi_t(M\otimes_A B)=\pi_t M\otimes_{\pi_0 A}\pi_0 B. One way to establish this formula is to compare the two homological functors in MM. The candidate assembled from the homotopy groups and a flat module is exact and preserves coproducts, and the comparison is an isomorphism on all suspended copies of AA. The ordinary ring π0A\pi_0 A acts on the homotopy groups, so the comparison is balanced. The class of modules for which it is an isomorphism in every degree is stable and closed under coproducts. The suspended unit generates Mod⁡A\operatorname{Mod}_A as a localizing subcategory. In particular extension of scalars detects equivalences and concentration of homotopy groups.

Faithfully flat spectral descent identifies Mod⁡A\operatorname{Mod}_A with the Amitsur limit of Mod⁡Bn\operatorname{Mod}_{B_n} by extension of scalars [8, Theorem 3.1]. The functors are symmetric monoidal, so the equivalence is symmetric monoidal with the pointwise structures on the limit. Passing to commutative algebra objects yields (17). Indeed an algebra is given by a functor out of the commutative operad satisfying the inert-arrow conditions, and functor formation and these conditions are checked in every coordinate of a limit. For ordinary rings, concentrate all homotopy groups in degree zero. A differential degree cannot survive base change and then become zero by faithful flatness. This recovers the affine statement of classical fpqc descent [9]. The symmetric monoidal comparison carries the multiplication and unit operations through module descent.

There is a different restriction of the same descent formula. An AA-algebra CC is spectrally étale in the geometric sense when π0C\pi_0 C is étale over π0A\pi_0 A in the ordinary sense and πtC≅πtA⊗π0Aπ0C(t∈Z).\pi_t C\cong\pi_t A\otimes_{\pi_0 A}\pi_0 C \qquad(t\in\mathbf Z). It is finite étale when this ordinary algebra is also finite as a module. Lurie’s étale lifting theorem gives an equivalence CAlg⁡Aet≃N(Alg⁡π0Aet) \operatorname{CAlg}_A^{\mathrm{et}} \simeq N(\operatorname{Alg}_{\pi_0 A}^{\mathrm{et}}) and gives discrete mapping spaces from étale sources into any AA-algebra, with mapping sets computed on π0\pi_0 [6, Theorem 7.5.0.6 and Corollary 7.5.4.6]. Formula (18) applies also to the finite étale subcategory. It is compatible with base change because pushouts of étale maps remain étale and the degree-zero comparison is ordinary base change. For connective bases, the corresponding spectral affine small étale sites agree with the classical sites for π0A\pi_0 A. Their generating covers are tested by joint surjectivity on the same underlying spectra of ordinary rings [5]. The equivalence of sites yields an equivalence of their ∞\infty-topoi of space-valued sheaves, which still have higher sheaf cohomology even though the affine étale probes and their maps are discrete.

Proposition 1.7. If A→BA\to B is faithfully flat, the ∞\infty-categories of étale and finite étale AA-algebras are the respective Amitsur limits of the same subcategories over BnB_n.

Proof. Both properties are stable under extension of scalars. Conversely let CC be an AA-algebra whose base change to BB has either property. Ordinary faithfully flat descent of the étale and finite properties, together with π0(B⊗AC)≅π0B⊗π0Aπ0C\pi_0(B\otimes_A C)\cong\pi_0 B\otimes_{\pi_0 A}\pi_0 C, yields the required condition on degree zero [9]. For each tt, tensor the homotopy comparison map for CC over π0A\pi_0 A with π0B\pi_0 B. The resulting map is the homotopy comparison for B⊗ACB\otimes_A C as a BB-algebra, by flat base change for modules. It is an isomorphism, so the original map is an isomorphism by faithful flatness. Hence the required subcategories are the restrictions of (17) selected in degree zero, which proves the assertion. ◻

What the two restrictions detect

The Pierce functor is built from components, while a small étale site is built from flat unramified maps and contains the Zariski opens. If AA has connected π0\pi_0, the former sends its base to one point regardless of the prime spectrum. A geometrically étale map may have connected source and target with a genuinely different field of functions. Such a map is sent to the identity of a one-point Stone space. It can nevertheless become Pierce-split after a sufficiently large faithfully flat extension, and Theorem 1.6 then records its monodromy through the arrow space of that extension. Component data alone therefore cannot recover the geometric étale site on the base. The descent construction uses component data simultaneously on the cover and on its iterated overlaps.

Here is the field calculation at an infinite level. Let L/KL/K be an infinite algebraic Galois extension, the union of finite Galois subextensions Li/KL_i/K, with Krull group G=lim⁡iGal⁡(Li/K)G=\lim_i \operatorname{Gal}(L_i/K). As LL-algebras, L⊗KL≅colim⁡i∏Gal⁡(Li/K)L≅{continuous functions G→Ldisc}.L\otimes_K L \cong \underset{i}{\operatorname{colim}} \prod_{\operatorname{Gal}(L_i/K)} L \cong \{\text{continuous functions }G\to L_{\mathrm{disc}}\}. At a finite stage the first isomorphism is the splitting of a separable Galois tensor product. For the second, a continuous function to the discrete field has finitely many clopen fibers by compactness, and a common finite-stage refinement detects all of them. Thus HLHL is normal over HKHK. The Pierce image of the Amitsur nerve is the groupoid with one object and arrow group GG, and Theorem 1.6 classifies split algebras by continuous GG-actions on Stone spaces, as in [1, Chapter 3]. All tensor powers here are flat over KK, so passing to Eilenberg–Mac Lane spectra preserves their discreteness.

Spectral Galois extensions and stable monodromy

A categorical Galois theory depends on which objects count as covers. For a commutative ring spectrum, ordinary étaleness of degree zero together with the flat homotopy base-change formula gives one answer. A different answer asks for commutative algebra objects that become products of components of the unit after an effective extension of scalars. The second answer extends the split-algebra construction of [1] into stable homotopy theory. It permits torsors such as the complexification map in periodic topological KK-theory, which fail the homotopy flatness test. The two theories are related by descent, with different input conditions.

Three effectivity inputs

A stable homotopy theory here means a presentable symmetric monoidal stable ∞\infty-category C\mathcal C in which the tensor product preserves colimits separately. Write 1\mathbf 1 for its unit. An algebra B∈CAlg⁡(C)B\in\operatorname{CAlg}(\mathcal C) is faithful if B⊗−B\otimes- is conservative. In a stable category this is equivalent to requiring B⊗M≃0B\otimes M\simeq0 to imply M≃0M\simeq0, by applying the test to cofibers. It is descendable if the thick tensor ideal it generates contains 1\mathbf 1. A thick tensor ideal is closed under finite cofiber sequences, retracts, and tensor products with any object of C\mathcal C.

The following criterion gives the difference between a conservative test and a finite proof of descent. Put I=fib⁡(1→B)I=\operatorname{fib}(\mathbf 1\to B), with its map u:I→1u:I\to\mathbf 1. Write un:I⊗n→1u_n:I^{\otimes n}\to\mathbf 1 for the tensor power of uu.

Proposition 1.1 (Nilpotence criterion). The algebra BB is descendable if and only if unu_n is nullhomotopic for some n≥1n\geq 1.

Proof. Say an object MM belongs to N\mathcal N if un⊗Mu_n\otimes M is nullhomotopic for some nn. It is closed under tensoring with an arbitrary object and under retracts. Suppose M1→M→M2M_1\to M\to M_2 is a cofiber sequence and powers nn and mm annihilate the terms M1M_1 and M2M_2, respectively. The map I⊗m⊗M→MI^{\otimes m}\otimes M\to M becomes null after composition to M2M_2, by naturality. It lifts along M1→MM_1\to M. Tensor that lift with I⊗nI^{\otimes n} and apply unu_n on its target. The resulting composite factors through un⊗M1u_n\otimes M_1, hence vanishes. By naturality this equals un+m⊗Mu_{n+m}\otimes M. This shows extension closure, and suspension, desuspension and rotation of a cofiber sequence give the other two-out-of-three cases. Thus N\mathcal N is thick. It contains BB since multiplication splits B→B⊗BB\to B\otimes B and hence B⊗uB\otimes u is null. If BB is descendable, 1∈N\mathbf 1\in\mathcal N.

Conversely, the cofiber of unu_n is built from the cofibers in the chain I⊗n⟶I⊗(n−1)⟶⋯⟶I⟶1.I^{\otimes n}\longrightarrow I^{\otimes(n-1)} \longrightarrow \cdots\longrightarrow I\longrightarrow\mathbf 1. These cofibers are I⊗j⊗BI^{\otimes j}\otimes B for 0≤j<n0\leq j<n, since the arrows are tensorings of uu. Repeated use of the cofiber sequence for a composite puts the cofiber of unu_n in the thick tensor ideal generated by BB. If unu_n is null, 1\mathbf 1 is a retract of that cofiber. ◻

A descendable algebra is faithful. For a fixed MM, the objects XX satisfying X⊗M≃0X\otimes M\simeq 0 form a thick tensor ideal, so if that ideal contains BB, it contains the unit. Descendability moreover implies effective Amitsur descent for C\mathcal C as a symmetric monoidal category [8, Proposition 3.22]. The decisive step in the comonadicity proof concerns a cosimplicial diagram that becomes split after tensoring with BB. After this tensoring its partial-totalization tower is pro-constant. The objects XX such that the tower for the diagram tensored by XX is pro-constant form a thick tensor ideal. Finite limits commute with tensoring, and constant pro-objects form a stable full subcategory closed under retracts. Tensoring a pro-constant tower produces a pro-constant tower because tensoring defines a levelwise functor on pro-objects. Descendability puts the unit in the ideal. For a pro-constant tower, passage to the limit commutes with tensoring by BB. The dual form of Barr–Beck then gives comonadicity, and the iterated base-change diagram gives the Amitsur limit.

Two other effectivity inputs serve different purposes. Faithfulness plus dualizability suffices for comonadicity without an assumption about compactness of the unit, because tensoring with a dualizable object preserves all limits. When the unit is compact these two conditions also imply descendability [8, Theorem 3.38]. Flat faithful descent of ring spectra, used in Chapter [ch:algebra], is still another criterion. Its hypotheses are expressed on all homotopy groups and it establishes the Amitsur equivalence directly. In particular, a proof of equivalence of module categories via conservative, limit-preserving tensoring is available without checking homotopy flatness.

The Amitsur complex of a Rognes extension

Let GG be a finite group and let A→BA\to B be a map of E∞E_\infty-rings with a coherent action of GG on BB by AA-algebra maps. Its characteristic comparisons are A⟶BhG,γ:B⊗AB⟶∏g∈GB,γg=μ∘(1⊗g). A\longrightarrow B^{hG}, \qquad \gamma:B\otimes_A B\longrightarrow\prod_{g\in G}B, \qquad \gamma_g=\mu\circ(1\otimes g). Here BhGB^{hG} is the homotopy limit of the action diagram over BGBG, computed in ring spectra, and μ\mu is multiplication. If both maps are equivalences, the extension is GG-Galois in the sense of Rognes. It is faithful if BB is faithful as an AA-module [7, Sections 4 and 6]. The faithfulness condition matters to the following categorical comparison. The fixed-point equation only describes the algebra of invariants, while descent must detect every module.

Lemma 1.2 (Bar description of the tensor powers). If γ\gamma is an equivalence, there is an equivalence of cosimplicial algebras B⊗A(∙+1)≃([n]⟼∏(g1,…,gn)∈GnB),B^{\otimes_A(\bullet+1)} \simeq \left([n]\longmapsto\prod_{(g_1,\ldots,g_n)\in G^n} B\right), where the right side is the cochain diagram of the GG-action. The degree-nn component at (g1,…,gn)(g_1,\ldots,g_n) is the map that multiplies the factors after twisting factor ii by g1⋯gig_1\cdots g_i, with the first factor untwisted.

Proof. For fixed nn, use the first factor as a common BB-base and write the tensor power as the nn-fold tensor product over BB of B⊗ABB\otimes_A B. Apply γ\gamma to each factor. Finite products of BB-algebras are taken to finite products by base change, as can be checked on underlying modules, where finite products are also coproducts. This gives an equivalence indexed by independent elements (h1,…,hn)(h_1,\ldots,h_n), with factor ii twisted by hih_i. Reindex by hi=g1⋯gih_i=g_1\cdots g_i, a bijection of indexing sets.

For the first coface, the inserted unit in factor zero makes every later factor share the twist g1g_1. This is the coface applying g1g_1 after deleting it from the coordinate tuple. An interior insertion replaces adjacent group entries by their product. Insertion at the final position drops the last entry. A codegeneracy multiplies two adjacent factors and is identified with inserting the group identity at that place. These formulas are equalities for strict actions and are identities of composites of the multiplication and the action structure maps for a coherent action. Equivalently, the comparison is induced from the map of simplicial diagrams from the action groupoid of GG on the affine represented by BB to its affine Čech nerve, where an arrow and its first object determine the second. This construction gives the whole family with its coherences. ◻

Theorem 1.3 (Stable Galois descent). For a faithful GG-Galois extension of E∞E_\infty-rings, extension of scalars induces symmetric monoidal equivalences Mod⁡A≃(Mod⁡B)hG,CAlg⁡A≃(CAlg⁡B)hG.\operatorname{Mod}_A\simeq(\operatorname{Mod}_B)^{hG}, \qquad \operatorname{CAlg}_A\simeq(\operatorname{CAlg}_B)^{hG}. The group acts on the categories on the right by extension of scalars along its automorphisms of BB. A module or algebra there has semilinear transport with respect to the action on its scalars.

Proof. Rognes’s dualizability theorem makes BB a dualizable AA-module [7, Proposition 6.2.1]. Tensoring with it is therefore both conservative and limit-preserving. To check the latter, regard tensoring as an endofunctor of Mod⁡A\operatorname{Mod}_A and use its left adjoint, tensoring with the dual B∨B^\vee. Limits of BB-modules are computed on their underlying AA-modules, so extension of scalars also preserves them. By comonadicity, the Amitsur module comparison is an equivalence. The Čech comparison is symmetric monoidal, since every extension of scalars is, and the monoidal operations in the limit are pointwise.

Apply Lemma 1.2. Modules over ∏JB\prod_J B with JJ finite are products of modules over the factors. This follows by projecting to the orthogonal central idempotents, giving an inverse to taking the direct sum with its componentwise action. The first coface on the resulting category-valued bar construction is the twist by g1g_1, the middle cofaces compose group elements, and the last drops the last element. Thus its limit is the category of coherent semilinear descent data (Mod⁡B)hG(\operatorname{Mod}_B)^{hG}. Passing to commutative algebra objects in the symmetric monoidal equivalence proves the second assertion. ◻

A fixed-point description follows at the level of individual objects. If DD is a semilinear BB-algebra in the right-hand category, its inverse image is the homotopy fixed-point algebra of the underlying action, under AA. Indeed the right adjoint to extension of scalars with canonical descent data is formed by the limit of the forgetful diagram, including the transport. This is seen by taking mapping spaces out of any AA-algebra and applying extension–restriction adjunction in every degree. The equivalence in Theorem 1.3 identifies this right adjoint with the inverse. The module statement holds by the same argument. Thus reconstructing the underlying algebra or module uses the full homotopy limit even when a classifying category of special algebras has discrete maps.

Intermediate split algebras

Suppose π0B\pi_0 B is nonzero and has only trivial idempotents. A BB-algebra of the form BT=∏t∈TBB^T=\prod_{t\in T} B, with TT a finite set, has a discrete space of maps to another such algebra BUB^U. By Lemma [lem:pierce-split-products], that space is the set of maps U→TU\to T. Extension of scalars along each g:B→Bg:B\to B canonically carries BTB^T to a copy of BTB^T, since it preserves these products and takes the unit to the unit. Every coherent semilinear structure on BTB^T then amounts to a genuine action on TT, by contravariance and discreteness. In formulas one may take the action on functions to be (gφ)(t)=g(φ(g−1t))(g\varphi)(t)=g(\varphi(g^{-1}t)).

Corollary 1.4. Under the hypotheses of Theorem 1.3 and connectedness of π0B\pi_0 B as above, the opposite of the category of AA-algebras becoming BTB^T for finite TT is equivalent to the category of finite GG-sets. Connected objects equipped with a map of algebras to BB correspond to pointed transitive GG-sets, hence to subgroups H≤GH\leq G. For T=G/HT=G/H with point HH, the associated intermediate algebra is BhHB^{hH}.

Proof. The full subcategory in question is preserved by the equivalence of Theorem 1.3 and is characterized after extension by the discrete products just described. The object BB corresponds to the regular GG-set by (19). Hence a map of algebras to BB gives a GG-equivariant map from the regular GG-set, which is specified by its value at the identity. A nonempty split object is connected precisely when the action is transitive, since a nontrivial product decomposition is exactly a disjoint union of invariant subsets. This also agrees with indecomposability of the underlying algebra. An idempotent would give a product decomposition, and each factor would remain split after base change because the idempotents of BTB^T are the characteristic functions of subsets of TT. For transitive TT, the stabilizer of the point yields the unique pointed isomorphism with G/HG/H. The fixed-point construction for BG/HB^{G/H} is a limit over the action groupoid G⋉(G/H)G\ltimes(G/H) with value BB at each object and transport induced by the group action. This groupoid is equivalent to BHBH by inclusion of the stabilizer at HH. Its limit is BhHB^{hH}. Under the same identification, the map to BB is evaluation at the selected object. ◻

The correspondence is formulated in the category of the split algebras and their maps. In spectral algebra there is generally no useful criterion for an intermediate object that simply asks for injectivity of a map on every homotopy group. Connected split objects have a categorical role inside the category defined by base change and decomposition into components. The stabilizer is read from that category.

Corollary 1.5 (Normal subgroups). In the situation of Corollary 1.4, if HH is normal in GG then A→BhHA\to B^{hH} is a faithful (G/H)(G/H)-Galois extension.

Proof. Set Q=G/HQ=G/H and D=BhHD=B^{hH}. The right translations of QQ on G/HG/H commute with the left GG-action. Contravariance transports them to an action by AA-algebra maps on DD. After tensoring with BB, the action is the regular permutation action on ∏QB\prod_Q B. The torsor tensor-square comparison becomes the canonical equivalence obtained from the bijection between pairs of elements of QQ and a point of QQ together with a translation. The operation B⊗A−B\otimes_A- detects equivalences and commutes with limits on modules, hence also on algebras. The fixed-point comparison after that operation reduces to B≃(∏QB)hQB\simeq(\prod_Q B)^{hQ}, which holds because the translation action groupoid on QQ is contractible. Finally if D⊗AM≃0D\otimes_A M\simeq 0, then tensoring with BB makes the finite product of copies of B⊗AMB\otimes_A M zero. Since QQ is nonempty and BB is faithful, M≃0M\simeq0. ◻

The stable Galois category and its range

Mathew’s finite-cover construction makes the preceding restricted classification intrinsic to a stable homotopy theory [8, Definitions 6.1–6.2 and Theorem 6.5]. The Boolean algebra of idempotents of π0End⁡C(1)\pi_0\operatorname{End}_{\mathcal C}(\mathbf 1) plays the role of the Pierce components of the unit. Given a commutative algebra object CC, its localization C[e−1]C[e^{-1}] at an idempotent of the endomorphism ring of the free rank-one CC-module is the corresponding direct factor in its module category. The localization can be constructed using the telescope of multiplication by ee. It is idempotent under relative tensor product and corresponds to the projection onto the summand where ee acts as the identity.

An algebra D∈CAlg⁡(C)D\in\operatorname{CAlg}(\mathcal C) is called a finite cover if some descendable CC makes C⊗DC\otimes D a finite product of the algebras C[e−1]C[e^{-1}], with independently chosen idempotents for the factors. A weak finite cover is defined in the same way with C⊗−C\otimes- conservative and limit-preserving. The idempotents need not partition the unit because several sheets may meet the same component. After a finite Boolean refinement into pairwise disjoint clopens, the product describes a finite number of copies on each clopen component.

The opposite of either finite-cover category is a Galois category in the general sense allowing several components, classified by a profinite groupoid [8, Theorem 5.36]. In particular the covers have discrete mapping spaces in their ambient ∞\infty-category. The two categories agree when the unit is compact. For C=Mod⁡A\mathcal C=\operatorname{Mod}_A this compactness is automatic, since mapping out of AA is computed on the underlying spectra and commutes with filtered colimits. For a connective AA, Mathew’s comparison theorem identifies these covers with the finite étale algebras of π0A\pi_0 A, through their spectral étale lifts [8, Theorem 6.17]. This assertion requires more than the lifting theorem of Chapter [ch:algebra]. It says the tensor-categorical cover condition detects exactly the objects which the lifting theorem starts with in the connective case.

The effect of the definition on torsors can be checked in one direction directly. A faithful Rognes extension A→BA\to B is dualizable and has B⊗AB≃∏GBB\otimes_A B\simeq\prod_G B. Thus BB is its own weak finite-cover splitting witness. In Mod⁡A\operatorname{Mod}_A, compactness of AA upgrades this witness to descendability. Conversely, a GG-torsor in the weak Galois category is a faithful GG-Galois extension, and in the strong category it is a descendable one [8, Proposition 6.13 and Corollary 6.15]. A local trivialization checks the tensor-square comparison and faithfulness by conservativity. The local tensor-square map is the map from a trivial torsor to its product with the group. To check fixed points in the weak case, tensoring by the witness commutes with the fixed-point limit. In the strong case, recover the torsor algebra by Amitsur descent along the witness, then commute its fixed-point limit with that Amitsur limit. At every Amitsur stage the torsor stays trivial and its fixed points are the stage algebra, since the translation action groupoid of the regular action is contractible. The Amitsur limit of these stage algebras recovers the unit.

Periodic rings and higher local systems

Rognes proves that periodic complexification KO→KUKO\to KU, with the complex-conjugation action, is a faithful C2C_2-Galois extension [7, Proposition 5.3.1]. Its faithfulness can also be seen from the cofiber sequence ΣKO→ηKO⟶KU.\Sigma KO \xrightarrow{\eta} KO\longrightarrow KU. If KU⊗KOMKU\otimes_{KO}M vanishes, the image of the stable Hopf element η\eta acts invertibly as a degree-one endomorphism of MM. A sufficiently high power of η\eta is zero, so this invertibility forces M≃0M\simeq0. The map on π0\pi_0 is the identity of Z\mathbf Z, while π1KO≅Z/2\pi_1 KO\cong\mathbf Z/2 and π1KU=0\pi_1 KU=0. Thus this extension fails spectral flatness, yet gives module and algebra descent by Theorem 1.3. It lies in the stable finite-cover category. This explains why the connective hypothesis of the comparison with geometric finite étale algebras cannot simply be removed.

The finite Galois groupoid of a stable category can miss homotopy visible to its module objects. Consider X=K(Z,2)X=K(\mathbf Z,2), an algebraically closed characteristic-zero field kk, and the pointwise tensor category C=Fun⁡(X,Mod⁡Hk)\mathcal C=\operatorname{Fun}(X,\operatorname{Mod}_{Hk}). Its commutative algebras are functors to CAlg⁡Hk\operatorname{CAlg}_{Hk}. Evaluation preserves finite products, localization at idempotents, and the thick tensor ideal operations defining descendability. Every finite cover of C\mathcal C therefore takes values in finite covers of HkHk, which are finite products of HkHk. Its image lands in the maximal groupoid of these products, a 11-truncated space, and XX is simply connected. Thus each such cover is constant up to equivalence. Conversely a constant product is already split by the unit. Maps between the constant products are the opposites of ordinary maps of finite sets, since XX is connected and the value-wise mapping spaces are discrete. The finite Galois groupoid of C\mathcal C is trivial.

For the fiber M=Hk⊕ΣHkM=Hk\oplus\Sigma Hk, its endomorphism spectrum decomposes into two copies of HkHk, one of ΣHk\Sigma Hk, and one of Σ−1Hk\Sigma^{-1}Hk. The space Aut⁡(M)\operatorname{Aut}(M) is a union of components of its endomorphism space, so its identity component has π1≅(k,+)\pi_1\cong(k,+) and no higher homotopy. The simply connected cover of BAut⁡(M)B\operatorname{Aut}(M) is therefore K(k,2)K(k,2). A nonzero additive homomorphism Z→k\mathbf Z\to k gives a map K(Z,2)⟶K(k,2)⟶BAut⁡(M)K(\mathbf Z,2)\longrightarrow K(k,2) \longrightarrow B\operatorname{Aut}(M) which is nonzero on π2\pi_2. It classifies a nonconstant HkHk-module local system. Its higher transport is invisible to the stable finite Galois groupoid, as happens for the fundamental progroupoid and the full shape in the space-valued theory.

Reflections and factorization

Reflective Galois structures and factorization

In Borceux and Janelidze’s categorical Galois theorem, triviality is measured by a slice adjunction and splitting is found by changing the base [1, Chapter 5]. The higher analogue needs two independent ingredients. A slice adjunction specifies the chosen class of trivial objects. Effective descent says which objects and maps over a cover glue across its entire Čech nerve. Separating these ingredients makes it possible to use the same argument with rings, group objects, or spaces while stating their individual descent hypotheses explicitly.

The adjunction seen from a varying base

Let L ⁣:C⇄D ⁣:RL\colon\mathcal C\rightleftarrows\mathcal D\colon R be an adjunction of ∞\infty-categories with unit η\eta and counit ϵ\epsilon. Assume both categories admit pullbacks. Over B∈CB\in\mathcal C, use the functors LB ⁣:C/B⟶D/LB,LB(X→fB)=(LX→LfLB),RB ⁣:D/LB⟶C/B,RB(D→dLB)=(B×RLBRD⟶B).\begin{aligned} L_B&\colon\mathcal C_{/B}\longrightarrow\mathcal D_{/LB},& L_B(X\xrightarrow{f}B)&=(LX\xrightarrow{Lf}LB),\\ R_B&\colon\mathcal D_{/LB}\longrightarrow\mathcal C_{/B},& R_B(D\xrightarrow{d}LB)&=(B\times_{RLB}RD\longrightarrow B). \end{aligned} The pullback uses ηB\eta_B. All slices contain the full mapping spaces over the indicated bases.

Proposition 1.1 (Slice adjunction and base change). There are adjunctions LB⊣RBL_B\dashv R_B. For every h ⁣:B′→Bh\colon B'\to B there is a natural equivalence h∗RB≃RB′(Lh)∗.h^*R_B\simeq R_{B'}(Lh)^*. These equivalences are coherent for chains of base changes. If Σh\Sigma_h denotes postcomposition, there is also an equivalence LBΣh≃ΣLhLB′L_B\Sigma_h\simeq\Sigma_{Lh}L_{B'}.

Proof. Over BB, a map from XX to B×RLBRDB\times_{RLB}RD is the same as a map X→RDX\to RD with a specified identification of the composite to RLBRLB with ηBf\eta_Bf. The original adjunction identifies the space of such maps with the homotopy fiber hofib⁡Lf(Map⁡D(LX,D)⟶Map⁡D(LX,LB)).\operatorname{hofib}_{Lf}\big( \operatorname{Map}_{\mathcal D}(LX,D) \longrightarrow \operatorname{Map}_{\mathcal D}(LX,LB)\big). This is the desired mapping space in D/LB\mathcal D_{/LB}. The unit at ff has underlying map X→B×RLBRLXX\to B\times_{RLB}RLX induced by (f,ηX)(f,\eta_X). To check base change, use the fact that RR preserves pullbacks and substitute R(LB′×LBD)≃RLB′×RLBRDR(LB'\times_{LB}D)\simeq RLB'\times_{RLB}RD in the formula for RB′R_{B'}. Naturality of η\eta and pasting of pullbacks identify the result with B′×RLBRDB'\times_{RLB}RD. Iterated identifications all arise from the limit of the same diagram, so their coherence follows from the universal cones. The assertion about postcomposition follows directly from functoriality of LL. ◻

For a restricted Galois structure, let E\mathcal E in C\mathcal C and F\mathcal F in D\mathcal D be classes of maps containing equivalences and closed under composition and pullback. Write CE/B\mathcal C_{\mathcal E/B} for the full subcategory of C/B\mathcal C_{/B} spanned by maps in E\mathcal E, and similarly for DF/LB\mathcal D_{\mathcal F/LB}. If LL takes E\mathcal E to F\mathcal F and RR takes F\mathcal F to E\mathcal E, the displayed slice adjunction and its base-change equivalences restrict to these full subcategories. The full-subcategory condition matters. Maps between the selected objects over a base retain their higher morphisms whether or not their underlying arrows belong to E\mathcal E.

Definition 1.2 (Admissibility and splitting). The structure is admissible when the restricted RBR_B is fully faithful at each base. A map over BB is trivial if the unit of the restricted slice adjunction at that map is an equivalence. A map over BB is split by p ⁣:U→Bp\colon U\to B if its pullback to UU is trivial. It is a covering relative to a specified class of descent maps if one of those maps splits it.

Full faithfulness is equivalent to invertibility of the slice counit. It makes the full subcategory Triv⁡(B)\operatorname{Triv}(B) of trivial maps the essential image of RBR_B. A map is trivial exactly when its unit square X→ηXRLX↓↓B→ηBRLB\begin{array}{ccc} X&\xrightarrow{\eta_X}&RLX\\ \downarrow&&\downarrow\\ B&\xrightarrow{\eta_B}&RLB \end{array} is a pullback. Proposition 1.1 shows that trivial maps are stable under base change. Their defining square is computed using the global unit, but the proof of stability uses the right slice adjoints. There is no need to assume that LL preserves all pullbacks.

Lemma 1.3 (A practical admissibility criterion). If RR is fully faithful, the unrestricted structure is admissible precisely when LL preserves every pullback of the form B×RLBRDB\times_{RLB}RD where the first arrow is ηB\eta_B and the other is the image of D→LBD\to LB. In this assertion a preserved pullback has comparison map L(B×RLBRD)→DL(B\times_{RLB}RD)\to D.

Proof. Under the counit equivalences LRD≃DLRD\simeq D and LRLB≃LBLRLB\simeq LB, the image of the cospan has first arrow equivalent to the identity of LBLB by the triangle identity. Its limit is DD. The comparison is precisely the counit of LB⊣RBL_B\dashv R_B. Invertibility of these counits for all B,DB,D is the assertion. ◻

Admissibility can be checked with a smaller class F\mathcal F when the problem concerns only extensions. For instance, a reflection may be useful on slices of surjections even if its behavior on slices of all arrows has not been established. Effective descent on the selected slices requires both unrestricted descent and detection of the selected arrow condition, or a descent proof directly on the selected slices.

The comparison over one descent map

For p ⁣:U→Bp\colon U\to B in E\mathcal E, let U∙U_\bullet be its Čech nerve, with Uk=U×B(k+1)U_k=U^{\times_B(k+1)}. Use the pullback functors along its simplicial structure maps to form a cosimplicial diagram of restricted slices. Say that pp is an effective descent map for E\mathcal E if the comparison CE/B⟶lim⁡[k]∈ΔCE/Uk\mathcal C_{\mathcal E/B}\longrightarrow \lim_{[k]\in\Delta}\mathcal C_{\mathcal E/U_k} is an equivalence. When every map is allowed and C\mathcal C is an ∞\infty-topos, every effective epimorphism satisfies this condition by the van Kampen property of its Čech realization [2, Section 6.1]. More generally, if unrestricted descent holds and membership in E\mathcal E is detected after pullback along pp, restricting the equivalence gives the indicated condition.

Theorem 1.4 (Relative higher Galois comparison). In an admissible structure, let p ⁣:U→Bp\colon U\to B be an effective descent map for E\mathcal E. The full subcategory Split⁡(p)\operatorname{Split}(p) of maps split by pp satisfies Split⁡(p)≃lim⁡[k]∈ΔDF/LUk.\operatorname{Split}(p)\simeq \lim_{[k]\in\Delta}\mathcal D_{\mathcal F/LU_k}. The diagram on the right is formed with pullbacks along maps of L(U∙)L(U_\bullet).

Proof. Restrict the unrestricted comparison for the selected slices to the full subcategories Triv⁡(Uk)\operatorname{Triv}(U_k). A point of the limit of the slices is a coherent family whose object in any degree kk is equivalent to the pullback of the degree-zero object along any vertex. Consequently, if its object in degree zero is trivial, all its objects are trivial. Conversely a family consisting of trivial objects has a trivial degree-zero object. Limits of diagrams of full subcategories are full in the limit of the ambient diagrams. These observations give the full essential image of Split⁡(p)\operatorname{Split}(p) and preserve its mapping spaces. Finally the equivalences RUk ⁣:DF/LUk≃Triv⁡(Uk)R_{U_k}\colon\mathcal D_{\mathcal F/LU_k}\simeq\operatorname{Triv}(U_k) respect restriction by Proposition 1.1. Taking their limit gives the result. ◻

At degree one the family has an equivalence between two pullbacks of a single object over LULU. At degree two it gives a specified homotopy between composite gluing and direct gluing. The comparison retains the higher data as well. An ordinary category of strict descent isomorphisms is sufficient in a nerve of an ordinary category, where the mapping spaces are discrete. In that case the degree-two relation is an equality and determines all the higher compatibility for an object. In categories with nondiscrete mapping spaces the totalization is intrinsic to the assertion.

There are two distinct possible simplifications of this comparison. If LL preserves the pullbacks that form U∙U_\bullet, the simplicial object L(U∙)L(U_\bullet) is the Čech nerve of LpLp. If it further presents an effective groupoid in a suitable category D\mathcal D, the totalization can be expressed as its category of actions. If unrestricted descent also holds for LpLp in D\mathcal D, then the result is the entire unrestricted slice over LBLB. Each step uses a separate property. Applying a reflection termwise to a Čech nerve is meaningful even if its image fails the groupoid Segal conditions, and the comparison theorem still holds in the form given above.

Proposition 1.5 (The monadic comparison). Suppose that p∗p^* on the selected slices is monadic and admits the restricted left adjoint Σp\Sigma_p. Let T=p∗ΣpT=p^*\Sigma_p be its monad on CE/U\mathcal C_{\mathcal E/U}. If TT sends trivial objects to trivial objects, it restricts to a monad TtrivT_{\mathrm{triv}}, and Split⁡(p)≃Alg⁡Ttriv(Triv⁡(U)).\operatorname{Split}(p)\simeq \operatorname{Alg}_{T_{\mathrm{triv}}}(\operatorname{Triv}(U)).

Proof. The full subcategory of trivial objects contains every iterate of TT on any one of its objects. Hence the unit and multiplication of the monad restrict to it with their coherences. A coherent algebra structure on such an object uses only these iterates and maps between them. Full faithfulness gives the same spaces for the operations, compatibility diagrams, and algebra morphisms in the smaller and larger categories. The restricted algebra category is thus the full subcategory of Alg⁡T(CE/U)\operatorname{Alg}_{T}(\mathcal C_{\mathcal E/U}) on algebras whose underlying objects are trivial. By monadic comparison, underlying objects correspond to pullbacks of objects over BB. This is exactly the splitting condition. ◻

The stability hypothesis is equivalent to requiring that ΣpRU(D)\Sigma_p R_U(D) be split by pp for every allowed D→LUD\to LU. This recovers the extra condition in the monadic Galois-descent version of the ordinary theorem [1, Section 5.1]. The Čech comparison uses a different stated hypothesis, namely an effective descent comparison to the full cosimplicial diagram. This formulation avoids assuming that every monad arising from a chosen class already has an identified groupoid model.

Proposition 1.6 (Change of splitting cover). Suppose the specified descent maps are stable under pullback and composition. Coverings are stable under base change. If q ⁣:V→Bq\colon V\to B is a specified descent map and q∗fq^*f is a covering over VV, then ff is a covering whenever ff is allowed. If V→UV\to U is a map over BB, then Split⁡(p)⊆Split⁡(V→B)\operatorname{Split}(p)\subseteq\operatorname{Split}(V\to B).

Proof. For a new base B′→BB'\to B, pull back a splitting map U→BU\to B and use stability of triviality. If W→VW\to V splits q∗fq^*f, its composite W→BW\to B is a specified descent map that splits ff. The refinement statement follows from the same stability of triviality. All assertions concern full subcategories and respect the coherent pullback functors between their comparisons. ◻

An allowed effective descent map pp is normal for this structure when p∗pp^*p is trivial. This belongs to the relative terminology. A normal splitting cover is a descent map that splits itself, and its pullbacks are again normal whenever the descent class is stable under pullback. The definition has no dependence on a choice of geometric point or separable closure. Chapter [ch:central] relates it to centrality in the ordinary group calculation.

Two factorization systems arising from truncation

Return to a reflection, using all arrows and identifying RR with inclusion. Write LL also for the endofunctor RLRL under this convention. Let M\mathcal M be the trivial maps and W\mathcal W the maps inverted by LL.

Every orthogonality assertion below concerns a contractible space of fillers in every coherent commutative square. Equivalently, a ⁣:A→Ba\colon A\to B is orthogonal to m ⁣:X→Ym\colon X\to Y if the natural map Map⁡(B,X)⟶Map⁡(A,X)×Map⁡(A,Y)Map⁡(B,Y)\operatorname{Map}(B,X) \longrightarrow \operatorname{Map}(A,X)\times_{\operatorname{Map}(A,Y)} \operatorname{Map}(B,Y) is an equivalence. This definition retains the space of commutativity data of a square as well as the space of diagonals.

Theorem 1.7 (Admissible reflective factorization). For an unrestricted admissible reflection there is an orthogonal factorization system (W,M)(\mathcal W,\mathcal M) with factorization A⟶B×LBLA⟶B(f ⁣:A→B).A\longrightarrow B\times_{LB}LA\longrightarrow B \qquad (f\colon A\to B). The factorizations are functorial and W\mathcal W is stable under pullback along M\mathcal M.

Proof. The right arrow is RB(LBf)R_B(L_Bf). Admissibility makes its LL-image the arrow LA→LBLA\to LB up to equivalence, with counit inverse to the image of the left arrow by the triangle identity. Hence the left arrow lies in W\mathcal W. To see orthogonality, take a lifting problem against X=Y×LYLX→YX=Y\times_{LY}LX\to Y. For a fixed bottom arrow B→YB\to Y, lifts amount to compatible maps B→LXB\to LX over LYLY. For a∈Wa\in\mathcal W, restriction along aa is an equivalence on maps into either of the local objects LX,LYLX,LY. It is therefore an equivalence on their homotopy fibers over the prescribed composite to LYLY, proving orthogonality. The class W\mathcal W is closed under retracts, and so is M\mathcal M, since the natural comparison from an object to its slice reflection carries retract diagrams to retract diagrams. The usual orthogonality-retract argument now identifies each class with the full orthogonal of the other. Functoriality follows from the units and pullback universal properties.

For the stability statement let m ⁣:C→Bm\colon C\to B be trivial and a ⁣:A→Ba\colon A\to B inverted. Write C≃B×LBDC\simeq B\times_{LB}D for D∈D/LBD\in\mathcal D_{/LB}. Then A×BC≃A×LA(LA×LBD)A\times_BC\simeq A\times_{LA}(LA\times_{LB}D). Applying admissibility at AA and BB gives the comparison LA×LBD→DLA\times_{LB}D\to D under LL, an equivalence. ◻

Stable units mean that LL preserves pullbacks over every local object in the ambient category, including pullbacks in which neither of the objects over that base is local. This implies admissibility by the practical criterion. Left exactness requires finite-limit preservation over arbitrary bases and is stronger. If LL is left exact, W\mathcal W is stable under arbitrary pullback. Conversely, if C\mathcal C has finite limits and W\mathcal W is pullback stable, LL is left exact. Indeed the reflection fixes the terminal object, since it is local. For a pullback A×BCA\times_BC, the map to A×LBCA\times_{LB}C is a pullback of the diagonal of B→LBB\to LB. That diagonal lies in W\mathcal W by pullback stability and two-out-of-three, since either projection B×LBB→BB\times_{LB}B\to B lies in W\mathcal W and the composite with the diagonal is the identity. The map A×LBC⟶LA×LBLCA\times_{LB}C\longrightarrow LA\times_{LB}LC is the composite of pullbacks of the two units, so is also in W\mathcal W. The final object is local, since the right adjoint preserves limits. This proves preservation of pullbacks.

Corollary 1.8 (Left-exact localizations). Let L ⁣:X→YL\colon\mathcal X\to\mathcal Y be an accessible left-exact localization of an ∞\infty-topos, and allow all arrows. For any effective epimorphism p ⁣:U→Bp\colon U\to B, a map over BB is split by pp if and only if it is already trivial.

Proof. The localized category Y\mathcal Y is an ∞\infty-topos. Left exactness identifies L(U∙)L(U_\bullet) with the Čech nerve of LpLp. Since LL is a left adjoint, it preserves the colimit of the original nerve, so LpLp is also an effective epimorphism. The right-hand side of Theorem 1.4 is therefore equivalent to Y/LB\mathcal Y_{/LB} by effective descent there. Under that equivalence the trivial object RB(D)R_B(D) corresponds to the descent datum of DD, since the base-change formula of Proposition 1.1 identifies its restriction in every degree. Thus the equivalence on slices realizes the inclusion of the trivial objects into the split objects as an equivalence. ◻

Definition 1.9 (Relative truncation). For n≥−1n\geq -1 and an ∞\infty-topos X\mathcal X, the relative nn-truncation of f ⁣:A→Bf\colon A\to B is the reflection of AA into the full subcategory of nn-truncated objects in X/B\mathcal X_{/B}. Write its total object as τ≤nBA\tau_{\leq n}^{B}A. A map is nn-connected if its relative nn-truncation is terminal in its target slice.

The pair of nn-connected and nn-truncated maps is an orthogonal factorization system in an ∞\infty-topos, with both classes stable under pullback [2, Sections 5.5.6 and 6.5.1]. A lifting problem with an nn-truncated right arrow over BB depends only on the relative nn-truncation of the left arrow by the slice universal property. For A→BA\to B, the first map in the relative truncation factorization is nn-connected, and the second is nn-truncated. The factorization and lifting conditions are often called the truncation modality [17]. The indexing here pairs nn-connected maps with nn-truncated maps, and an effective epimorphism is (−1)(-1)-connected.

Global nn-truncation, with codomain the nn-truncated objects, has stable units. Indeed, given A→D←CA\to D\leftarrow C with DD nn-truncated, factor the comparison A×DC⟶τ≤nA×Dτ≤nCA\times_D C\longrightarrow \tau_{\leq n} A\times_D \tau_{\leq n} C as two pullbacks of nn-connected units. The target is nn-truncated because the full subcategory of truncated objects is closed under limits, and the composite is nn-connected. This identifies the comparison with a truncation. The reflective factorization of Theorem 1.7 therefore applies. Its intermediate object B×τ≤nBτ≤nAB\times_{\tau_{\leq n}B}\tau_{\leq n}A is constructed from the global truncations. Relative truncation uses τ≤nBA\tau_{\leq n}^{B}A. They have different universal properties over the unreplaced BB.

Example 1.10 (Relative components of paths). Take X=S\mathcal X=\mathcal{S}, B=S1B=S^1 and A=∗A=* with the basepoint. The global 00-truncations of AA and BB coincide, so the global reflective factorization has intermediate object S1S^1. The fiber of ∗→S1*\to S^1 over the basepoint is ΩS1≃Z\Omega S^1\simeq\mathbb Z, where Z\mathbb Z is discrete. Thus ∗→S1*\to S^1 is already 00-truncated over S1S^1, and the relative factorization has intermediate object ∗*. Since this fiber has infinitely many components, the map cannot be 00-connected. Pullback to the basepoint also exhibits failure of pullback stability for the maps inverted by global 00-truncation.

Localizing the reflective pair

For global nn-truncation, its trivial maps are the base changes of maps between nn-truncated objects. Their fibers are nn-truncated. When the base admits a cover by an nn-truncated object, the converse holds after passing to that cover. This includes every base space for n≥0n\geq0, since the coproduct of basepoints of its components gives a discrete effective epimorphism.

Proposition 1.11 (Local triviality and stable inversion). Fix n≥−1n\geq-1 in an ∞\infty-topos X\mathcal X. Every nn-connected arrow is inverted by global nn-truncation after arbitrary pullback. Every arrow that becomes trivial for global nn-truncation over an effective epimorphism is nn-truncated. Over a base admitting an effective epimorphism from an nn-truncated object, both converses hold.

Proof. If ff is nn-connected, its pullbacks are nn-connected. For any such pullback a ⁣:X→Ya\colon X\to Y and any globally nn-truncated object DD, the projection Y×D→YY\times D\to Y is an nn-truncated arrow. Orthogonality then identifies maps X→DX\to D with maps Y→DY\to D. Global truncation thus inverts aa.

For the reverse implication over a base of the indicated kind, write ff as A→eT→mBA\xrightarrow{e}T\xrightarrow{m}B in the relative nn-connected, nn-truncated factorization. If every pullback of ff is inverted by global truncation, the same holds for mm since it holds for ee. Pull back to an effective epimorphism U→BU\to B with UU nn-truncated. The pullback of mm has nn-truncated domain and codomain, so inversion by global truncation makes it an equivalence already. Then mm is an equivalence by effective descent, proving ff is nn-connected.

Trivial arrows are nn-truncated since relative truncation is stable under pullback. This property descends along effective epimorphisms by induction on the truncation of the relative diagonal, starting with descent of equivalences. Descent of equivalences is immediate from the slice comparison over the Čech nerve. Finally, if U→BU\to B is an effective epimorphism with UU nn-truncated and ff is nn-truncated, then U×BAU\times_B A is nn-truncated by composition with U→1U\to 1. Both domain and base of the new arrow are local, so its unit square is a pullback. ◻

Over spaces the last result gives a homotopical form of monotone–light factorization. For n=0n=0, the left class consists of maps with nonempty connected homotopy fibers, and the right class consists of maps with discrete homotopy fibers. This concerns the homotopy theory of spaces. The original result concerns compact Hausdorff spaces and ordinary point inverses [1, Section 5.8]. Those settings have different base change, and passing from a topological map to its homotopy type need not preserve its point inverse. The corresponding constructions agree in the formal operation of testing connectedness after base change and testing the right class by local splitting.

Central extensions and higher bands

Categorical centrality in the ordinary group calculation arises by splitting for abelianization. A higher central extension also has an interpretation as a principal bundle over a delooping, where its band keeps track of all the homotopy of the kernel. The constructions are compatible for discrete groups. To compare them in higher degrees, it is useful to prove the ordinary result first and to keep the higher notion of a band explicit.

The ordinary abelianization comparison

Use Eab=E/[E,E]E_{\mathrm{ab}}=E/[E,E] and write aE ⁣:E→Eaba_E\colon E\to E_{\mathrm{ab}}. For groups, admit surjections in both the source and target of the abelianization reflection. Its slice right adjoint over GG takes a surjection D→GabD\to G_{\mathrm{ab}} of abelian groups to G×GabD→GG\times_{G_{\mathrm{ab}}}D\to G. The following computation works for this class and makes the admissibility criterion concrete.

Lemma 1.1 (Abelianization over a discrete base). Let P=G×GabDP=G\times_{G_{\mathrm{ab}}}D, with D→GabD\to G_{\mathrm{ab}} surjective and DD abelian. Then [P,P]=[G,G]×{1}andPab≅D.[P,P]=[G,G]\times\{1\}\qquad\text{and}\qquad P_{\mathrm{ab}}\cong D. In particular, abelianization is admissible on these surjective slices.

Proof. The two projections from PP are surjective. Every commutator in PP has second coordinate 11 since DD is abelian. Conversely, given x,y∈Gx,y\in G, choose lifts (x,d),(y,d′)∈P(x,d),(y,d')\in P. Their commutator is ([x,y],1)([x,y],1). Since such commutators generate [G,G][G,G], this proves the first equality. The second projection has kernel [G,G]×{1}[G,G]\times\{1\} and is surjective, yielding the second assertion. This identifies the slice counit with an isomorphism. ◻

Every surjection of ordinary groups is an effective descent map for surjective arrows. One can see this directly by viewing a group as a reduced simplicial set satisfying the Segal conditions and the invertibility condition. A surjection U→GU\to G induces a pointwise surjection between their group nerves, since the map in degree rr is Ur→GrU^r\to G^r. Ordinary effective descent for a surjection of sets comes from forming equivalence classes under its compatible gluing relation. Applied degree by degree, this gives descent for diagrams of sets, hence a descended simplicial set over the nerve of GG. The Segal comparisons are maps over the products GrG^r. Their pullbacks to UrU^r are isomorphisms, so they are isomorphisms. The same argument shows that the shear comparison (d2,d1)(d_2,d_1) in degree two is an isomorphism, which recovers invertibility in the descended monoid. Surjectivity in degree one is also detected after pullback. Thus the descended object is an allowed group surjection, with morphisms descending by the same degreewise argument.

Theorem 1.2 (Trivial, split, and central group extensions). For a group surjection e ⁣:E→Ge\colon E\to G with kernel KK, the following statements hold.

  1. The surjection is trivial for abelianization exactly when K∩[E,E]=1K\cap[E,E]=1.

  2. The surjection is split by some surjection exactly when KK is central in EE.

  3. A central surjection is split by itself and thus is normal for this Galois structure.

The reflection of ee into central extensions over GG is the quotient E/[K,E]→GE/[K,E]\to G.

Proof. Consider the unit comparison ϕ ⁣:E⟶G×GabEab.\phi\colon E\longrightarrow G\times_{G_{\mathrm{ab}}}E_{\mathrm{ab}}. It is surjective. To lift (g,u)(g,u), first choose xx lifting gg. The difference between uu and aE(x)a_E(x) lies in the kernel of Eab→GabE_{\mathrm{ab}}\to G_{\mathrm{ab}}. This kernel is the image of KK because the surjection [E,E]→[G,G][E,E]\to[G,G] allows any representative mapping to [G,G][G,G] to be corrected to an element of KK without changing its abelianized class. Multiplying xx by a lift of the difference in KK gives the desired lift. The kernel of ϕ\phi is visibly K∩[E,E]K\cap[E,E]. This proves the first statement. It also implies that any trivial extension is central, since [K,E]⊆K∩[E,E][K,E]\subseteq K\cap[E,E] by normality of the kernel.

Suppose p ⁣:U→Gp\colon U\to G splits ee. The kernel of U×GE→UU\times_GE\to U is canonically KK and is central in U×GEU\times_GE by the first statement. The second projection U×GE→EU\times_GE\to E is surjective, so each y∈Ey\in E is the image of an element commuting with every element of the lifted KK. Therefore KK is central in EE.

Conversely, centrality makes the map E×K⟶E×GE,(x,k)⟼(x,xk)E\times K\longrightarrow E\times_GE,\qquad (x,k)\longmapsto(x,xk) a group isomorphism over EE. The projection E×K→EE\times K\to E is trivial by the first statement, since KK is abelian and [E×K,E×K]=[E,E]×{1}[E\times K,E\times K]=[E,E]\times\{1\}. Hence ee is split by itself.

Finally, [K,E][K,E] is a normal subgroup of EE contained in KK. In the quotient by this subgroup the image of KK commutes with the image of EE. Every map from ee to a central extension over GG kills each commutator [k,x][k,x], since kk maps into the central kernel of the target. It therefore factors uniquely through this quotient. This proves the universal property. ◻

This gives the ordinary relative comparison and its algebraic content [1, Section 5.2]. A general treatment of categorical centrality for admissible reflections appears in [19]. The split object in the sense of Chapter [ch:reflections] is an extension that becomes trivial after a chosen change of base. An ordinary group splitting by a section is a separate property. For instance, a central extension that has a group-homomorphic section is isomorphic over its base to a direct product with its kernel, whereas the classification below also includes nonsplit group extensions.

Higher central extensions as pointed principal bundles

Work in an ∞\infty-topos X\mathcal X. Write BGBG for the delooping of a group object GG. The group object and pointed connected object descriptions are equivalent via BB and Ω\Omega [2, Section 6.1]. If AA is a commutative group object, BABA itself has a commutative group structure, so its delooping B2AB^2A is defined. One way to construct this structure is to apply the bar construction to the grouplike commutative monoid structure on AA. Realization preserves the finite products in the bar diagram, so the commutative structure survives and its underlying monoid is grouplike because it is connected.

Definition 1.3 (Framed central extension). Fix a group object GG and a commutative group object AA. A framed central extension of GG by AA is a principal BABA-bundle P→BGP\to BG with a point over the basepoint of BGBG. Equivalences preserve the principal structure and the chosen point, including their coherences. The associated group extension is A→ΩP→GA\to\Omega P\to G.

The choice of a principal action is the band in this definition. The basepoint trivialization identifies the fiber over the basepoint with BABA using that action. Taking loops then identifies the kernel with AA. The structure is thus a higher analogue of prescribing an abelian coefficient group and keeping its identification with the central kernel fixed.

Theorem 1.4 (Classifying space of framed extensions). The space of framed central extensions of GG by AA is Map⁡X∗(BG,B2A).\operatorname{Map}_{\mathcal X_*}(BG,B^2A). For the extension specified by c ⁣:BG→B2Ac\colon BG\to B^2A, there is a pointed fiber sequence BA⟶hofib⁡(c)⟶BG→cB2A.BA\longrightarrow \operatorname{hofib}(c)\longrightarrow BG \xrightarrow{c} B^2A. Its associated group map has π0\pi_0-surjective augmentation π0Ωhofib⁡(c)→π0G\pi_0\Omega\operatorname{hofib}(c)\to\pi_0G, and the image of π0A\pi_0 A is central in π0Ωhofib⁡(c)\pi_0\Omega\operatorname{hofib}(c).

Proof. The universal principal BABA-bundle is 1→B2A1\to B^2A. Its Čech nerve is the bar construction of BABA. Pullback classifies principal bundles, with mapping spaces, by the principal-bundle comparison, applied to BABA [4]. At the basepoint of BGBG a lift to the pullback of the universal bundle is exactly a specified identification between the classifying map at that basepoint and the basepoint of B2AB^2A. This gives the asserted equivalence of spaces and the fiber sequence.

The fibers of P→BGP\to BG are locally BABA, which is connected. Relative connectedness can be checked after an effective epimorphism by relative truncation and effective descent. Consequently P→BGP\to BG is 00-connected and PP is connected, since BGBG is connected. Taking loops preserves the pointed pullback describing the fiber and gives the group kernel AA. The long exact homotopy sequence, or its internal sheaf form, gives the asserted surjectivity because BABA has terminal π0\pi_0.

To see centrality, loop the principal action BA×P→PBA\times P\to P and use preservation of products by loops. The resulting map A×ΩP→ΩPA\times\Omega P\to\Omega P is a morphism of group objects and its restrictions to the two factors are the kernel inclusion and the identity. The images of those factors commute on π0\pi_0, since the source is a product of groups there. The principal action records a choice of coherence beyond this equality on components. ◻

For a general higher group GG, the kernel of π0A→π0ΩP\pi_0 A\to\pi_0\Omega P is the image of the connecting homomorphism π1G→π0A\pi_1G\to\pi_0 A. The homotopy fiber assertion incorporates this correction to the component-level group sequence.

Proposition 1.5 (Recognition from a coherent action). Let p ⁣:E→Gp\colon E\to G be a map of group objects surjective on the sheaves of components. Suppose a coherent action of AA on EE over GG is supplied, with the simplicial action diagram A∙×EA^\bullet\times E formed in group objects and with AA carrying its specified commutative group structure. If the action shear A×E→E×GEA\times E\to E\times_GE is an equivalence, the resulting pointed group extension has a framed central structure.

Proof. Deloop each term of the action diagram. The equivalence between group objects and pointed connected objects preserves products, so the resulting simplicial action diagram is BA∙×BEBA^\bullet\times BE. The map BE→BGBE\to BG is 00-connected, since the long exact sequence and surjectivity of π0p\pi_0p give connected homotopy fibers at the basepoint, and the basepoint 1→BG1\to BG is an effective epimorphism. In particular BE×BGBEBE\times_{BG}BE is connected. The delooped shear BA×BE→BE×BGBEBA\times BE\to BE\times_{BG}BE is therefore a map between pointed connected objects. Its loop comparison is the given equivalence, so it is an equivalence as well. This proves that BE→BGBE\to BG is a principal BABA-bundle, with the frame given by the basepoint of BEBE. ◻

Corollary 1.6 (Ordinary central extensions and their automorphisms). For ordinary groups GG and ordinary abelian groups AA regarded as discrete objects of S\mathcal{S}, framed extensions give exactly the ordinary central extensions of GG with kernel identified with AA. Their isomorphism classes are H2(G;A)H^2(G;A) for the trivial action, and the automorphism group of an extension with kernel and quotient fixed is H1(G;A)H^1(G;A).

Proof. The fiber sequence BA→P→BGBA\to P\to BG shows that PP is a connected 11-type. Thus ΩP\Omega P is an ordinary group. Its surjectivity, injectivity of the kernel, and centrality follow from the theorem and the long exact sequence. Conversely an ordinary central extension A→E→GA\to E\to G yields the group homomorphism A×E→EA\times E\to E by multiplication of the central inclusion with the identity. Delooping this homomorphism gives a left BABA-action on BEBE over BGBG. The map BA×BE⟶BE×BGBEBA\times BE\longrightarrow BE\times_{BG}BE is an equivalence. Both sides are pointed and connected, the right side by surjectivity of E→GE\to G, and looping identifies the map with the ordinary torsor comparison A×E→E×GEA\times E\to E\times_GE. The latter is bijective by uniqueness of the kernel element relating two lifts. Also BE→BGBE\to BG is an effective epimorphism since both spaces are connected. This gives the principal bundle with its point. The constructions are inverse on objects and isomorphisms, using the equivalence between groups and pointed connected 11-types. The central action in this degree is forced by the kernel inclusion and the group multiplication, so it adds no independent data.

Finally B2A=K(A,2)B^2A=K(A,2). Homotopy groups of Map⁡∗(BG,K(A,2))\operatorname{Map}_*(BG,K(A,2)) are the reduced cohomology groups in degrees 2−i2-i for 0≤i≤20\leq i\leq2. Here degree zero agrees with unreduced cohomology in positive cohomological degree, and H~0(BG;A)=0\widetilde H^0(BG;A)=0 because BGBG is connected. Translation by an extension class identifies its based automorphism group with the first homotopy group at the zero class. This gives the two identifications. ◻

One can read this classification directly in a group bar resolution. A section of sets s ⁣:G→Es\colon G\to E with s(1)=1s(1)=1 determines α(g,h)∈A\alpha(g,h)\in A by s(g)s(h)=α(g,h)s(gh).s(g)s(h)=\alpha(g,h)s(gh). Associativity and centrality give α(g,h)+α(gh,k)=α(h,k)+α(g,hk).\alpha(g,h)+\alpha(gh,k)=\alpha(h,k)+\alpha(g,hk). Conversely a normalized solution defines the multiplication (a,g)(b,h)=(a+b+α(g,h),gh)(a,g)(b,h)=(a+b+\alpha(g,h),gh) on A×GA\times G. The identity and inverses follow from the normalization and the cocycle equality, and its projection has central kernel AA. Changing the section modifies α\alpha by the group coboundary of a normalized function G→AG\to A. The bar nerve of GG has rr-simplices GrG^r, so its ordinary normalized cochain differential in degree two is exactly that equation. This comparison identifies the group cocycle with the classifying map and explains which operation in the proof is replaced by coherent descent in higher degrees.

Higher cohomology and the role of trivialization

If A0A_0 is an ordinary abelian group object in an ∞\infty-topos, let K(A0,r)K(A_0,r) denote its Eilenberg–Mac Lane object in degree rr. For the commutative group object A=K(A0,n)A=K(A_0,n) the classifying space in Theorem 1.4 is Map⁡X∗(BG,K(A0,n+2)).\operatorname{Map}_{\mathcal X_*}(BG,K(A_0,n+2)). The shifts come from using one delooping to pass from a group object to its classifying object and a second to classify the principal action on that object. In spaces, components are Hn+2(BG;A0)H^{n+2}(BG;A_0) and higher homotopy groups are the reduced cohomology in the corresponding lower degrees. In a general topos, the pointed mapping space is the homotopy fiber of restriction along 1→BG1\to BG over the zero section. This description records the chosen trivialization over the basepoint even when cohomology of the terminal object is nontrivial.

A variable coefficient object can instead be specified over BGBG. Let A\mathcal A be an abelian group object in the 00-truncated part of X/BG\mathcal X_{/BG} and choose an identification of its restriction at the basepoint with A0A_0. Eilenberg–Mac Lane objects can be formed in this slice. A principal K/BG(A,n+1)K_{/BG}(\mathcal A,n+1)-bundle in the slice is classified by a section of K/BG(A,n+2)K_{/BG}(\mathcal A,n+2). The proof is the same principal-bundle comparison, now with terminal object BG→BGBG\to BG. Taking the homotopy fiber of restriction to the basepoint over the zero section adds a point in the pulled-back torsor. Looping the resulting pointed total object gives a group map with kernel K(A0,n)K(A_0,n). The action over BGBG is a relative action and the coefficient object may vary along loops. For the constant coefficient object it is the global principal action in the theorem. Thus the slice construction also tracks coefficient twisting without imposing centrality on an arbitrary variable band.

Example 1.7 (An extension with a higher obstruction class). In spaces take G=ΩS3G=\Omega S^3 and A=K(Z,1)A=K(\mathbb Z,1). Let c ⁣:S3→B2A=K(Z,3)c\colon S^3\to B^2A=K(\mathbb Z,3) represent the integral generator. The map on π3\pi_3 induced by cc is an isomorphism, so the boundary π3(S3)→π2(BA)\pi_3(S^3)\to\pi_2(BA) in the fiber sequence BA→P→S3BA\to P\to S^3 is an isomorphism. This gives π2(P)=0\pi_2(P)=0. For the zero class the total object is BA×S3BA\times S^3 and has π2≅Z\pi_2\cong\mathbb Z. Both looped extensions have trivial component groups. The principal class distinguishes them and is detected by higher homotopy.

Normalization of the band is essential to the comparison with abelianization. For ordinary central group extensions the conjugation action on AA is trivial, with no higher choices of coherence. In the higher case it is possible to ask for a neutralizable coefficient system, a chosen null homotopy of an action, or a specified principal BABA-bundle. Definition 1.3 chooses the principal-bundle condition. The classifying theorem concerns that structure and its morphisms.

Abelianization after passing to group objects in spaces

Two extensions of the abelianization functor illustrate why the band construction has an independent role. The first applies ordinary abelianization to π0\pi_0. The second uses the left adjoint to the forgetful functor from infinite loop spaces to group objects in spaces. The first is a reflection onto a discrete subcategory. The second has a counit that can change the positive homotopy groups of a connective spectrum.

Proposition 1.8 (A discrete reflection on higher group objects). The functor from group objects in spaces to ordinary abelian groups given by G↦(π0G)abG\mapsto(\pi_0G)_{\mathrm{ab}} is an admissible reflection when the allowed maps are surjective on π0\pi_0 in the source and surjective in the target. A map split for this structure by an allowed effective descent map has 00-truncated fibers.

Proof. Maps into a discrete abelian group factor through π0\pi_0 and abelianization by the two respective universal properties. This proves the reflection assertion. Given an allowed abelian group surjection D→(π0G)abD\to(\pi_0G)_{\mathrm{ab}}, the underlying space of the slice right adjoint is the pullback G×(π0G)abDG\times_{(\pi_0G)_{\mathrm{ab}}}D. Since its base is discrete, its component set is π0G×(π0G)abD\pi_0G\times_{(\pi_0G)_{\mathrm{ab}}}D. Lemma 1.1 then computes the counit. Closure of the allowed maps under pullback follows from stability of effective epimorphisms of spaces. Surjectivity of the projection to GG follows from the same stability. The sliced adjunction is admissible.

Every trivial map is a pullback from a map of discrete abelian groups, so has discrete fibers. If an allowed map becomes trivial along a descent map U→GU\to G, the underlying map of spaces U→GU\to G is an effective epimorphism since it is surjective on components. Truncation of fibers descends along such a map by Chapter [ch:reflections]. ◻

In fact, in this structure each allowed base change map is effective for the restricted Čech comparison. In the functor ∞\infty-topos of simplicial spaces, the map of group nerves associated to U→GU\to G is an effective epimorphism. This holds pointwise because the map in simplicial degree rr is Ur→GrU^r\to G^r. Descent in that functor category provides a simplicial object over the group nerve of GG. The degree-zero condition is unchanged under pullback since both group nerves have terminal object in degree zero. The Segal comparisons over GrG^r and the invertibility comparison over G2G^2 become equivalences when pulled back along the effective epimorphisms Ur→GrU^r\to G^r and U2→G2U^2\to G^2, respectively. Here the invertibility comparison is the shear (d2,d1)(d_2,d_1), transported along the shear equivalence for the base group nerve. They are therefore equivalences prior to pullback. The descended object is a group nerve, and the required surjectivity on π0\pi_0 in degree one is detected after base change. The whole comparison is an equivalence on the selected full subcategories, with all spaces of morphisms.

Consequently a higher principal extension with a nondiscrete kernel, such as the preceding example, may have no interpretation as a covering for this discrete Galois structure. Its fibers already prevent that comparison. The discrete reflection retains the ordinary group case proved above and provides one admissible choice when the intended fibers have no higher homotopy.

Proposition 1.9 (Infinite-loop abelianization in spaces). Write UU for the forgetful functor from commutative group objects to group objects in spaces. Under the equivalence between commutative group objects and connective spectra, a left adjoint to UU is F(G)=Σ−1Σ∞BG.F(G)=\Sigma^{-1}\Sigma^\infty BG. Here Σ∞\Sigma^\infty is reduced suspension spectrum. The zeroth homotopy group of F(G)F(G) is (π0G)ab(\pi_0G)_{\mathrm{ab}}. The counit is noninvertible at H(Z2)H(\mathbb Z^2), so full faithfulness fails for UU.

Proof. Under infinite-loop recognition, a connective spectrum VV gives the infinite loop space Ω∞V\Omega^\infty V, with delooping Ω∞ΣV\Omega^\infty\Sigma V [6, Remark 5.2.6.26]. For pointed connected BGBG, the adjunction between reduced suspension spectrum and infinite loop space gives Map⁡Grp⁡(S)(G,U(V))≃Map⁡S∗(BG,Ω∞ΣV)≃Map⁡Sp⁡(Σ∞BG,ΣV)≃Map⁡Sp⁡(Σ−1Σ∞BG,V).\begin{split} \operatorname{Map}_{\operatorname{Grp}(\mathcal{S})}(G,U(V)) &\simeq \operatorname{Map}_{\mathcal{S}_*}(BG,\Omega^\infty\Sigma V)\\ &\simeq \operatorname{Map}_{\operatorname{Sp}}(\Sigma^\infty BG,\Sigma V)\\ &\simeq \operatorname{Map}_{\operatorname{Sp}}(\Sigma^{-1}\Sigma^\infty BG,V). \end{split} The spectrum in the last source is connective because BGBG is connected. Its zeroth homotopy group is the degree-one stable homotopy group of BGBG, and stable Hurewicz identifies this with H1(BG;Z)≅(π0G)abH_1(BG;\mathbb Z)\cong(\pi_0G)_{\mathrm{ab}}.

For a counterexample to full faithfulness, use the connective Eilenberg–Mac Lane spectrum H(Z2)H(\mathbb Z^2). Its underlying group object is discrete Z2\mathbb Z^2, with delooping S1×S1S^1\times S^1. There is a stable splitting Σ∞(S1×S1)≃Σ∞S1⊕Σ∞S1⊕Σ∞S2.\Sigma^\infty(S^1\times S^1) \simeq \Sigma^\infty S^1\oplus\Sigma^\infty S^1\oplus\Sigma^\infty S^2. Indeed the stable cofiber of the wedge inclusion is the suspension spectrum of the smash product, and the two projections split the wedge inclusion in spectra. Thus F(U(H(Z2)))F(U(H(\mathbb Z^2))) has a sphere spectrum suspended once as a summand. It has nonzero first homotopy group, while H(Z2)H(\mathbb Z^2) has zero first homotopy group. The counit cannot be an equivalence, contrary to what full faithfulness would imply. ◻

The two constructions agree on the zeroth homotopy group and differ even when the input is discrete and already abelian. The higher band theorem uses delooping and effective groupoid descent directly, so its classifying space requires neither an admissible infinite-loop reflection nor a presentation by fiberwise discrete data.

Directed Galois theories

Directed symmetries and generalized descent

A passage from a splitting object to its groupoid of changes of trivialization discards no information when all transports are invertible. There are also settings where a comparison has an intrinsic direction. A specialization of points can induce a map of fibers which is far from an equivalence. The directed analog of a Galois action is then a functor on a category with genuine noninvertible morphisms. This chapter distinguishes that construction from the generalized descent on precategories in [1, Chapter 7], and gives a higher form of both constructions.

Complete Segal objects and groupoid objects

Let B\mathcal B be an ∞\infty-topos. A simplicial object C∙C_\bullet in B\mathcal B is a Segal object if the canonical map Cn⟶C1×C0C1×C0⋯×C0C1C_n\longrightarrow C_1\times_{C_0}C_1 \times_{C_0}\cdots\times_{C_0}C_1 is an equivalence for n≥2n\geq 2. The maps in each pullback are the target and the source of successive arrows. Here d1d_1 is the source and d0d_0 the target. The simplicial identities give identity and composition laws with all their coherences. In particular choosing a representative composition at degree two gives a contractible space of choices, and degree three provides the associativity homotopy for the compatible choices.

There is an object C1eqC_1^{\mathrm{eq}} of equivalences. A generalized arrow over T∈BT\in\mathcal B belongs to this subobject when it is invertible in the category encoded by the Segal space Map⁡B(T,C∙)\operatorname{Map}_{\mathcal B}(T,C_\bullet). Equivalence is an intrinsic property of an arrow. The object can be built by taking the object of coherent inverse data for the walking isomorphism. The space of choices of coherent inverses to an equivalence is contractible. Completeness means that the degeneracy map specifying identities induces an equivalence C0≃C1eqC_0\simeq C_1^{\mathrm{eq}}. In contexts TT this says that equivalences of objects are modeled by paths in the object space. Complete Segal objects are one model of small ∞\infty-categories internal to B\mathcal B [11].

Remark 1.1. An internal groupoid object can have different object and arrow objects, as happens for a nontrivial Čech nerve. Such a groupoid object is a Segal object with all arrows invertible. It usually fails completeness in the complete Segal sense above. A complete Segal object with all arrows invertible is equivalent to a constant simplicial object. An effective groupoid object in an ∞\infty-topos is a presentation of its quotient, while a constant simplicial object on the quotient is already complete.

Working contextually is important. Global elements of objects of an ∞\infty-topos need not detect equivalences. For example, a statement about fibers over all c:T→C0c:T\to C_0 can be tested at the generic point c=id⁡C0c=\operatorname{id}_{C_0} in context C0C_0. A statement about fibers over all 1→C01\to C_0 can hold vacuously while the assertion over C0C_0 fails. The recovery of internal constructions from their underlying global categories requires its own hypotheses.

Coherent presheaves and direction

Definition 1.2. A right fibration over an internal category C∙C_\bullet is a map q:E∙→C∙q:E_\bullet\to C_\bullet of complete Segal objects such that each terminal-vertex map gives a pullback square En⟶E0↓↓Cn⟶C0.\begin{array}{ccc} E_n &\longrightarrow& E_0\\ \downarrow && \downarrow\\ C_n &\longrightarrow& C_0 . \end{array} A left fibration uses the initial vertex instead. Write RFib⁡B(C)\operatorname{RFib}_{\mathcal B}(C) and LFib⁡B(C)\operatorname{LFib}_{\mathcal B}(C) for the respective full subcategories over CC.

For a generalized edge f:x→yf:x\to y with parameters in TT, the right-fibration square in degree one identifies a lift ending at an element of EyE_y with that element. Taking the source of such lifts defines a map f∗:Ey→Exf^*:E_y\to E_x in B/T\mathcal B_{/T}. In degree two, lifting the two-simplex ending at zz identifies (gf)∗(gf)^* with f∗g∗f^*g^* on the fiber over zz. Higher terminal-vertex squares provide the compatible homotopies for longer strings. A left fibration instead gives an operation f!:Ex→Eyf_!:E_x\to E_y. The object E∙E_\bullet stores the transport as a coherent system, even when no levelwise representatives of its homotopies have been chosen.

There is a large internal universe ΩB\Omega_{\mathcal B} of objects in slices of B\mathcal B, taking TT to B/T\mathcal B_{/T} with reindexing by pullback, in a suitable enlargement of universes. The internal Grothendieck construction identifies the internal category of small right fibrations with the internal functor category from CopC^{\mathrm{op}} to ΩB\Omega_{\mathcal B}. The left-fibration version is obtained by taking the opposites of the total category and the base. This theorem and internal Yoneda are developed in [11, §§4.5–4.7]. For a contextual point c:T→C0c:T\to C_0, the representable right fibration is the slice C/c→C∣TC_{/c}\to C|_T, formed in B/T\mathcal B_{/T}. The Yoneda comparison evaluates a map of right fibrations on the identity of cc. The lift of an arrow into cc then recovers its value by pullback transport. Coherence makes this a universal description of the mapping object, including in families over TT.

Externally the constructions specialize to straightening in spaces, RFib⁡S(C)≃Fun⁡(Cop,S),LFib⁡S(C)≃Fun⁡(C,S).\operatorname{RFib}_{\mathcal S}(C)\simeq \operatorname{Fun}(C^{\mathrm{op}},\mathcal S),\qquad \operatorname{LFib}_{\mathcal S}(C)\simeq \operatorname{Fun}(C,\mathcal S). We use the right-hand convention for ordinary presheaves and the second convention for covariant representations. An equivalence C→DC\to D induces equivalent categories of either sort. In the special case of a groupoid, inversion identifies the two conventions. For a general CC, reversing CC has mathematical content, and the relevant variance has to be part of any classification theorem [2, §2.2].

Proposition 1.3 (Inverting the transport). Let WW be a set of arrows of a small ∞\infty-category CC. Restriction along C→C[W−1]C\to C[W^{-1}] identifies Fun⁡(C[W−1]op,S)\operatorname{Fun}(C[W^{-1}]^{\mathrm{op}},\mathcal S) with the full subcategory of right fibrations on CC in which transport along WW is by equivalences. The corresponding statement holds for left fibrations.

Proof. The universal property of localization applies to CopC^{\mathrm{op}} and the reversals of the arrows in WW. It identifies its entire functor category into S\mathcal S with the indicated full subcategory of Fun⁡(Cop,S)\operatorname{Fun}(C^{\mathrm{op}},\mathcal S). Under straightening, sending a reversed arrow to an equivalence is precisely the transport condition. The covariant proof uses CC itself. ◻

For WW consisting of all arrows, C[W−1]C[W^{-1}] is the groupoidification CgpdC^{\mathrm{gpd}}. It represents the subtheory with invertible transport. In particular it does not generally classify all right or all left fibrations on CC.

Two factorization systems

The factorization developed in [1, §7.3] first sends a functor to its essential image, then embeds that image fully faithfully. This has a direct higher analog. For an ∞\infty-category BB, write B≃B^\simeq for its maximal subgroupoid.

Proposition 1.4. Every functor f:A→Bf:A\to B factors as an essentially surjective functor followed by a fully faithful functor. These classes form an orthogonal factorization system on Cat⁡∞\operatorname{Cat}_\infty.

Proof. Choose the full subcategory I⊆BI\subseteq B on the union of the components of B≃B^\simeq meeting f(A≃)f(A^\simeq). It contains the image of ff, every object is equivalent to one from AA, and inclusion is fully faithful.

For orthogonality, take an essentially surjective e:A→Be:A\to B and a fully faithful m:X→Ym:X\to Y. Replace XX over YY by its full essential image using the preceding construction and full faithfulness. Given a square with ee on the left and mm on the right, the map B→YB\to Y sends every object into this full subcategory since every object is equivalent to one in the image of AA. A full subcategory is determined by its allowed vertices and includes every simplex of YY with those vertices in a quasicategory model. The map B→YB\to Y thus lifts. Postcomposition with a full inclusion gives a fully faithful functor Fun⁡(B,X)→Fun⁡(B,Y)\operatorname{Fun}(B,X)\to\operatorname{Fun}(B,Y). On cores the space of a lift with its prescribed identification is empty or contractible. The existence just proved, and the same full faithfulness for AA, show that imposing the rest of the square leaves a contractible space of lifts. Closure under retracts follows from the intrinsic definitions. Existence and orthogonality give the factorization system. ◻

A second factorization has a different right class. Given f:A→Bf:A\to B, let (b↓f)(b\downarrow f) be the comma category of objects (a,b→f(a))(a,b\to f(a)), with arrows over bb. Write ∣D∣|D| for the groupoidification of DD. We call ff final if every ∣b↓f∣|b\downarrow f| is contractible. By the higher form of Quillen’s Theorem A, this is equivalent to ff preserving colimit computations by restriction [2, §4.1.3]. An initial functor is one whose opposite is final. Its test uses the comma categories of arrows f(a)→bf(a)\to b.

Theorem 1.5 (Comprehensive factorization). There is an orthogonal factorization system on Cat⁡∞\operatorname{Cat}_\infty with left class the final functors and right class the right fibrations, taken up to equivalence of arrows. The right-fibration factor of f:A→Bf:A\to B is classified by the presheaf Lf=Lan⁡fop1A,Lf(b)≃∣b↓f∣.L_f=\operatorname{Lan}_{f^{\mathrm{op}}}\mathbf 1_A,\qquad L_f(b)\simeq |b\downarrow f|. Dually, the initial functors and the left fibrations form an orthogonal factorization system.

Proof. Here 1A\mathbf 1_A is the terminal presheaf. The pointwise formula for left Kan extension has indexing category (b↓f)op(b\downarrow f)^{\mathrm{op}}. The colimit of the constant point diagram is its groupoidification, canonically equivalent to that of the opposite category. In these terms the action of LfL_f is precomposition on comma categories. Let p:Ef→Bp:E_f\to B be the right fibration classified by LfL_f. The Kan unit 1A→(fop)∗Lf\mathbf 1_A\to (f^{\mathrm{op}})^*L_f corresponds to a section of f∗Eff^*E_f over AA, giving j:A→Efj:A\to E_f with pj=fpj=f. Under the comma description, j(a)j(a) is the point represented by (a,id⁡f(a))(a,\operatorname{id}_{f(a)}).

To prove finality of jj, fix e=(b,z)e=(b,z) in EfE_f. A lift of β:b→f(a)\beta:b\to f(a) from ee to j(a)j(a) is specified by a path between zz and the transported point Lf(β)(j(a))L_f(\beta)(j(a)) in Lf(b)L_f(b). This description is natural in (a,β)(a,\beta), since an arrow of this comma category supplies a homotopy relating the two transported points. Consequently (e↓j)(e\downarrow j) is classified over (b↓f)(b\downarrow f) by the family of path spaces Map⁡Lf(b)(z,λ(a,β))\operatorname{Map}_{L_f(b)}(z,\lambda(a,\beta)), where λ:(b↓f)→∣b↓f∣\lambda:(b\downarrow f)\to |b\downarrow f| is the localization map. This family sends every arrow to an equivalence. Its total groupoidification is the colimit of the family in spaces by straightening and the universal property of a colimit. Universality of colimits in spaces computes this colimit by pulling the colimit of the points λ(a,β)\lambda(a,\beta) in S/Lf(b)\mathcal S_{/L_f(b)} back along zz. That colimit is id⁡Lf(b)\operatorname{id}_{L_f(b)}, so the pullback is contractible. This is the comma criterion for jj.

For orthogonality, pull a right fibration back along the bottom map B→CB\to C of a square and let the resulting fibration over BB correspond to FF. A top map lifting a given arrow e:A→Be:A\to B gives a section over AA, equivalently an element of Map⁡(1A,(eop)∗F)\operatorname{Map}(\mathbf 1_A,(e^{\mathrm{op}})^*F). The space of fillers is the fiber of the restriction map from Map⁡(1B,F)\operatorname{Map}(\mathbf 1_B,F) over this element. The left Kan extension adjunction identifies restriction with precomposition by Le→1BL_e\to\mathbf 1_B. This arrow is an equivalence when ee is final, so the fiber is contractible. Finally final functors and right fibrations are each closed under retracts. For the right class this follows directly from the defining pullback condition, or from its lifting tests against terminal-vertex inclusions. Thus the factorizations and mutual orthogonality identify both classes. Taking opposites proves the dual assertion. ◻

The distinction between the two factorizations is visible on a category with indefinitely composable arrows. Take Z\mathbb Z with its usual order, viewed as an ∞\infty-category. The inclusion of the even integers is final since the even integers above each given bb form a poset with a least element. It is fully faithful and fails essential surjectivity. Conversely Zdisc→Z\mathbb Z_{\mathrm{disc}}\to\mathbb Z is essentially surjective, whereas (b↓f)(b\downarrow f) is a discrete infinite ray for this ff and has noncontractible groupoidification. Even a groupoidification equivalence need not be final. The inclusion of 00 in Z\mathbb Z induces an equivalence on groupoidifications because Z\mathbb Z is filtered and nonempty, yet (1↓f)(1\downarrow f) is empty.

The internal versions of the comprehensive factorization and of the comma criterion are established in [11, §4.4]. The tests there use every context TT and take groupoidification internally. Applying the preceding external criterion only to global points of BB would not give that result.

Descent along an augmented diagram

The precategories of [1, §§7.1–7.5] are diagrams retaining a small amount of categorical incidence data, before enforcing the Segal pullback conditions. Their coefficient presheaves are formed by a two-dimensional limit. The corresponding higher construction can start from an arbitrary indexing ∞\infty-category. This preserves the distinction the original book makes between a limit of change-of-base categories and the category of ordinary directed presheaves on an internal category.

Let II be small, let D:I▹→AD:I^\triangleright\to\mathcal A be an augmented diagram with cone point aa, and let F,G:Aop→Cat⁡∞F,G:\mathcal A^{\mathrm{op}}\to\operatorname{Cat}_\infty be contravariant coefficient systems. The natural transformations between them are homotopy coherent. We call the augmentation effective for GG when the canonical comparison G(a)⟶lim⁡i∈IopG(Di)G(a)\longrightarrow \operatorname*{lim}_{i\in I^{\mathrm{op}}}G(D_i) is an equivalence. An object of the limit includes objects in the coefficient categories and equivalences relating every pullback of them, together with higher coherences. The maps between the objects of a compatible family are likewise compared in the entire mapping spaces.

Theorem 1.6 (Relative splitting along a diagram). Suppose F→GF\to G is pointwise fully faithful, the augmentation DD is effective for GG, and there is an object i0i_0 such that every ii admits an arrow i→i0i\to i_0 in II. The choices of those arrows are immaterial. Let Split⁡F(D)⊆G(a)\operatorname{Split}_F(D)\subseteq G(a) consist of those xx whose restriction to Di0D_{i_0} belongs to the essential image of F(Di0)F(D_{i_0}). Then comparison induces an equivalence Split⁡F(D)≃lim⁡i∈IopF(Di).\operatorname{Split}_F(D)\simeq \operatorname*{lim}_{i\in I^{\mathrm{op}}} F(D_i).

Proof. A limit of full inclusions is fully faithful. Its image consists of those coherent families whose entries lie in the respective essential images. One way to see this last assertion is to replace each F(Di)F(D_i) by the full subcategory of G(Di)G(D_i) on its image. Naturality ensures that each reindexing map restricts to the relevant full subcategories. In the model of a limit as the category of Cartesian sections of the associated Cartesian fibration over II, the section belongs to the subfibration precisely when its objects do. Fullness in each fiber implies that every compatibility equivalence and coherent simplex of that section belongs to the subfibration as well.

For a section x∙x_\bullet, membership of xi0x_{i_0} implies membership of xix_i because an arrow i→i0i\to i_0 yields xi≃G(Di→Di0)(xi0)x_i\simeq G(D_i\to D_{i_0})(x_{i_0}). Naturality of F→GF\to G places this pullback in the image of F(Di)F(D_i). The converse follows by evaluation at i0i_0. This identifies the limit of the F(Di)F(D_i) with the pullback of F(Di0)⟶G(Di0)alonglim⁡G(Di)⟶G(Di0).F(D_{i_0})\longrightarrow G(D_{i_0}) \quad\text{along}\quad \lim G(D_i)\longrightarrow G(D_{i_0}). Using effectivity, this pullback is the defining full subcategory of G(a)G(a). ◻

Taking II to be the shape of the precategory in [1, §7.2], each of its object, arrow, and composable-pair indices maps to the object index. In the ordinary category-valued case the coherent pullback equivalences are isomorphisms in the coefficient categories, and the limit computes the pseudolimit used in the book. The relative splitting criterion is then exactly the requirement at the object index in [1, Theorem 7.5.2]. The higher theorem allows the whole simplicial shape and the resulting mapping spaces, with no reduction to comparison isomorphisms on a truncated shape.

If a diagram has several distinguished objects collectively receiving arrows from each index, the same argument imposes the condition at all distinguished objects. If D∙D_\bullet is simplicial, I=ΔopI=\Delta^{\mathrm{op}} and i0=[0]i_0=[0] can be used. In that case a vertex inclusion in Δ\Delta supplies the arrow required by the proof. For the Čech nerve of an effective epimorphism in an ∞\infty-topos, the slice coefficient system G(T)=B/TG(T)=\mathcal B_{/T} has effective descent. The theorem then restricts ordinary descent to a full class of objects that is stable under reindexing. Its naturality requirement on F→GF\to G is explicit. It does not imply that every adjunction on slice categories determines such a subsystem.

There is a precise interpretation of the slice construction on an arbitrary simplicial object D∙D_\bullet. Every colimit in an ∞\infty-topos is van Kampen [2, §6.1.3], so lim⁡[n]∈ΔB/Dn≃B/∣D∙∣B,∣D∙∣B=colim⁡[n]∈ΔopDn.\operatorname*{lim}_{[n]\in\Delta}\mathcal B_{/D_n} \simeq \mathcal B_{/|D_\bullet|_{\mathcal B}}, \qquad |D_\bullet|_{\mathcal B}= \operatorname*{colim}_{[n]\in\Delta^{\mathrm{op}}}D_n . Indeed a compatible family amounts to a map of simplicial objects E∙→D∙E_\bullet\to D_\bullet for which the naturality square for every simplicial operator is Cartesian. If D∙D_\bullet is Segal, this includes both vertex conditions, so transports from one endpoint to the other are equivalences. Over a complete Segal space the realization ∣D∙∣S|D_\bullet|_{\mathcal S} models the groupoidification. It counts simplices as coherent paths in a space, and in that target every edge becomes invertible. Thus these particular limit coefficients generalize the precategorical construction, while noninvertible directed transport requires the left- or right-fibration construction.

Descent of directed representations

Two useful forms of directed gluing keep track of which category contains the colimit. The first has its presentation in Cat⁡∞\operatorname{Cat}_\infty, where arrows remain directed. The second is indexed by objects in a base topos and glues all simplicial degrees using pullback.

Proposition 1.7 (Gluing over categorical colimits). Let C=colim⁡i∈ICiC=\operatorname*{colim}_{i\in I}C_i in Cat⁡∞\operatorname{Cat}_\infty, with all CiC_i small, and let E\mathcal E be an ∞\infty-category in the coefficient universe. Then Fun⁡(C,E)≃lim⁡i∈IopFun⁡(Ci,E).\operatorname{Fun}(C,\mathcal E)\simeq \operatorname*{lim}_{i\in I^{\mathrm{op}}}\operatorname{Fun}(C_i,\mathcal E). If a collection WiW_i of arrows is specified in each CiC_i compatibly with the diagram, the equivalence restricts to functors inverting the union WW of their images in CC. The contravariant version follows by taking opposites of the indexing categories CiC_i.

Proof. For any ∞\infty-category KK of the appropriate size, K×−K\times- preserves colimits, as its right adjoint is Fun⁡(K,−)\operatorname{Fun}(K,-). Mapping from KK to the comparison functor therefore yields Map⁡(K×C,E)≃lim⁡iMap⁡(K×Ci,E).\operatorname{Map}(K\times C,\mathcal E) \simeq \lim_i\operatorname{Map}(K\times C_i,\mathcal E). Testing against all KK proves the equivalence of ∞\infty-categories. Inverting an image arrow is a pointwise property of an object of the functor category. A full subcategory of the limit defined by these conditions consequently corresponds to the specified full subcategory on the left. The operation C↦CopC\mapsto C^{\mathrm{op}} is an autoequivalence of Cat⁡∞\operatorname{Cat}_\infty and preserves the colimit in question. ◻

Proposition 1.8 (Changing the base of an internal category). Let X∈BX\in\mathcal B, let U→XU\to X be an effective epimorphism, and let CC be a small internal ∞\infty-category in B/X\mathcal B_{/X}. For Un=U×X(n+1)U_n=U^{\times_X(n+1)} let CUnC_{U_n} be the pullback internal category. Then there is an equivalence RFib⁡B/X(C)≃lim⁡[n]∈ΔRFib⁡B/Un(CUn).\operatorname{RFib}_{\mathcal B_{/X}}(C)\simeq \operatorname*{lim}_{[n]\in\Delta} \operatorname{RFib}_{\mathcal B_{/U_n}}(C_{U_n}). The left-fibration version is obtained by the same argument with initial vertices.

Proof. Effective descent in the topos gives B/X≃lim⁡B/Un\mathcal B_{/X}\simeq\lim\mathcal B_{/U_n}. Taking simplicial diagrams commutes with this limit, and taking arrows with target the prescribed CC also commutes with the limit. This glues the underlying map of simplicial objects, functorially on morphisms. An arrow in B/X\mathcal B_{/X} is an equivalence if its pullback to UU is an equivalence. To see this, its pullbacks to each UnU_n are equivalences by further reindexing, and the limit of slice categories detects equivalences on entries.

The pullback functors between all these slices preserve limits. Consequently the Segal equations for the glued source can be checked after pullback to UU, where they hold. The same reasoning applies to the completeness comparison by computing the space of coherent inverse data using limits. Finally the terminal-vertex pullback conditions amount to equivalences with finite pullbacks and hold after reindexing to UU. The glued map is therefore a right fibration. This restricts the equivalence on underlying maps to precisely the subcategories in the statement. ◻

Exit paths and constructible monodromy

A covering of a connected space transports its fibers along paths, with inverse paths producing inverse equivalences. A constructible object on a stratified space has locally constant fibers on each stratum, together with comparisons across incident strata. An exit path starts in a lower stratum and may move toward higher strata. The reverse path can violate the stratification condition. This asymmetry supplies the directed information in constructible monodromy.

The topological classification

Give a poset PP its Alexandrov topology of upward-closed subsets. A PP-stratified space is a continuous map s:X→Ps:X\to P, and Xp=s−1(p)X_p=s^{-1}(p). An exit simplex σ:∣Δn∣→X\sigma:|\Delta^n|\to X is a singular simplex for which there is a chain p0≤⋯≤pnp_0\leq\cdots\leq p_n in PP such that σ(t0,…,tn)\sigma(t_0,\ldots,t_n) belongs to XpiX_{p_i} when ii is the largest index for which ti>0t_i>0. Face and degeneracy operations respect this condition. A conically stratified space admits open neighborhoods of points x∈Xpx\in X_p given by stratified charts Z×C(L)Z\times C(L), where Z×{vertex}Z\times\{\text{vertex}\} lies in XpX_p and the link LL is stratified by the elements above pp. Here C(L)C(L) is the open cone, with a basis of vertex neighborhoods given by the cones of uniformly bounded radius. Conicality ensures that the exit simplicial set is an ∞\infty-category, denoted Exit⁡P(X)\operatorname{Exit}_P(X). We use the conical chart and exit-path conventions in [12, §2.1].

The homotopy type Π∞Xp\Pi_\infty X_p is the fiber of Exit⁡P(X)→P\operatorname{Exit}_P(X)\to P over pp. The inclusion of any individual stratum in the exit category is fully faithful. Indeed a simplex whose vertices are all over pp has constant stratum label, so lies entirely in XpX_p. All endomorphisms in Exit⁡P(X)\operatorname{Exit}_P(X) are equivalences for the same reason. More generally the functor to PP is conservative. For a one-stratum space the simplicial model is the ordinary singular complex, with no directed constraints.

A hypersheaf of spaces on XX is a sheaf which satisfies descent for hypercovers. Call it PP-hyperconstructible if its restriction, as a hypersheaf, to every XpX_p is locally constant. This is a property, so the hyperconstructible objects form a full subcategory. In particular the construction does not prescribe a single common space as the value of all stalks. Each locally constant restriction can carry its own monodromy.

Theorem 1.1 (Topological exodromy, space coefficients). If X→PX\to P is conically stratified and each XpX_p admits a basis of weakly contractible open neighborhoods, there is an equivalence Cons⁡Phyp(X;S)≃Fun⁡(Exit⁡P(X),S).\operatorname{Cons}^{\mathrm{hyp}}_P(X;\mathcal S) \simeq \operatorname{Fun}(\operatorname{Exit}_P(X),\mathcal S). Evaluation of a representation at xx corresponds to the stalk at xx.

This is the space-coefficient case of the theorem of Porta and Teyssier [12, Theorem 5.4.1 and Corollary 6.2.5]. Their coefficient-general statement assumes joint conservativity of the hyper-restrictions to the strata. For spaces this conservativity follows from the stalk criterion for hypersheaves. The stalks of the restriction to a stratum are the stalks on XX at those same points, and the strata contain every point. For the full theorem the extra conservativity condition is substantive. The statement above uses hyperdescent throughout. Passing to ordinary sheaves requires a comparison with hypercompletion in the setting at hand.

The proof architecture of the cited theorem clarifies the role of the hypotheses. The exodromy adjunction reconstructs a hypersheaf from a representation by assigning to an open UU the limit over its exit-path category. Conical charts give cofinal tests for computing its inverse image on a stratum. In these tests the restriction of both functors and of the adjunction transformations agrees with the corresponding monodromy adjunction on the stratum. For a locally weakly contractible stratum, that adjunction is an equivalence on local systems. Hyper-restriction detects the unit on constructible hypersheaves, and evaluation at every point detects the counit on representations. This is where the theorem uses the local geometry beyond just the existence of a topological stratification [12, §§4–5].

By the convention of Chapter [ch:directed], the representations in Theorem 1.1 are classified by left fibrations over Exit⁡P(X)\operatorname{Exit}_P(X). If one uses entrance paths and denotes the resulting opposite category by Ent⁡P(X)\operatorname{Ent}_P(X), the same data are ordinary presheaves on Ent⁡P(X)\operatorname{Ent}_P(X). This change of orientation applies to the indexing category. The stalk maps attached to exit paths are still covariant.

Proposition 1.2 (Sections and open restriction). Under the hypotheses of Theorem 1.1, let FF be the hypersheaf corresponding to M:Exit⁡P(X)→SM:\operatorname{Exit}_P(X)\to\mathcal S. For an open U⊆XU\subseteq X with the induced stratification, F(U)≃lim⁡Exit⁡P(U)M∣Exit⁡P(U).F(U)\simeq \lim_{\operatorname{Exit}_P(U)} M|_{\operatorname{Exit}_P(U)}. The limits retain every compatibility with the noninvertible arrows of the exit category.

Proof. The strata of an open subspace inherit the required local bases and conical charts. Naturality of exodromy for open inclusions identifies hypersheaf restriction with precomposition by Exit⁡P(U)→Exit⁡P(X)\operatorname{Exit}_P(U)\to\operatorname{Exit}_P(X) [12, Proposition 6.2.3]. Sections are maps from the terminal hypersheaf. The terminal hypersheaf is constructible, and under an equivalence of categories it corresponds to the terminal functor, of constant value ∗*. The mapping space out of that functor is the limit being asserted, by the defining adjunction of the constant-diagram functor with the limit functor. ◻

A conical calculation with higher stalks

Let MM be the plane with the discrete closed subset Z×{0}\mathbb Z\times\{0\} removed, and let L=S2L=S^2. Both are locally contractible manifolds, MM with infinitely many punctures. On X=M×C(L)X=M\times C(L), put the lower stratum Z=M×{vertex}Z=M\times\{\text{vertex}\} and the upper stratum U=M×L×R>0U=M\times L\times\mathbb R_{>0}. The lower stratum has conical charts obtained by taking disks in MM times cones of uniformly bounded radius. The upper stratum is open and is an ordinary manifold. This gives explicit conical and weak local contractibility hypotheses for exodromy.

Write K=Π∞(M)K=\Pi_\infty(M) and T=Π∞(L)T=\Pi_\infty(L). The exit category of the cone is T◃T^\triangleleft, the category obtained by adjoining an initial object to TT. To justify this, the nonvertex subcategory is full and equivalent to TT by product with R>0\mathbb R_{>0}, while the vertex is initial by the cone lemma [12, Lemma 2.1.8]. An initial vertex determines, up to contractible choice, a coherent cone on the inclusion of the nonvertex subcategory. The resulting functor from T◃T^\triangleleft is fully faithful. The mapping spaces out of the initial vertex are contractible and there are no arrows from a nonvertex object to the vertex, since such arrows decrease the stratum label. Mapping spaces between nonvertex objects agree by fullness. The functor is therefore an equivalence. Exit simplices of a product with an unstratified space decompose into a singular simplex in that space and an exit simplex in the stratified factor. It follows that Exit⁡P(X)≃K×T◃.\operatorname{Exit}_P(X)\simeq K\times T^\triangleleft .

Write π:K×T→K\pi:K\times T\to K for projection. A representation is equivalently the full coherent data A:K→S,B:K×T→S,η:π∗A⟶B.A:K\to\mathcal S,\qquad B:K\times T\to\mathcal S,\qquad \eta:\pi^*A\longrightarrow B . Indeed currying gives a KK-indexed family of diagrams on T◃T^\triangleleft. A diagram on T◃T^\triangleleft is a diagram on TT equipped with a cone, which gives precisely the displayed data in each fiber. Natural transformations are compatible maps of the two diagrams and the comparison; the compatibility square is specified in the functor ∞\infty-category. At a given kk, adjunction also writes the comparison as A(k)→lim⁡t∈TB(k,t)A(k)\to \lim_{t\in T} B(k,t). This last description shows why keeping only the monodromy on each stratum can lose a comparison even when both restrictions are known exactly.

The inclusion of KK at the initial vertex is initial. For an object (k,t)(k,t) of the entire exit category, the category of arrows from (k′,vertex)(k',\text{vertex}) to it is equivalent to K/kK_{/k}, which has a terminal object. Therefore the global sections of the hypersheaf corresponding to (A,B,η)(A,B,\eta) are lim⁡KA\lim_K A by Proposition 1.2 and the initiality criterion. The comparison η\eta still controls the behavior on neighborhoods intersecting the upper stratum.

For example, take AA and BB to be constant of value the Eilenberg–Mac Lane space Q=K(Z,2)Q=K(\mathbb Z,2). The identity comparison and the comparison constant at the base point give inequivalent constructible hypersheaves with equivalent restrictions to both strata. A natural equivalence between the two representations would give a commutative square up to homotopy on any radial exit arrow, with equivalences on its vertical sides. On π2(Q)\pi_2(Q), one horizontal side is the identity and the other is zero. No such square can have invertible vertical sides.

There is also a link contribution in the direct image j∗Bj_*B for j:U↪Xj:U\hookrightarrow X. In the representation category j∗j_* corresponds to right Kan extension along K×T↪K×T◃K\times T\hookrightarrow K\times T^\triangleleft. This can be seen directly from j∗B(V)=B(V∩U)j_*B(V)=B(V\cap U) and the section formula. At (k,vertex)(k,\text{vertex}), the comma category for right Kan extension is equivalent to Kk/×TK_{k/}\times T. In this product, restricting a limit to the slice at id⁡k\operatorname{id}_k is allowed because the identity is initial in Kk/K_{k/}. Therefore the value on the lower stratum is k↦lim⁡TB(k,−)k\mapsto\lim_T B(k,-). For the constant diagram with value QQ, this gives Map⁡S(S2,Q)≃Q×Zdisc.\operatorname{Map}_{\mathcal S}(S^2,Q) \simeq Q\times \mathbb Z_{\mathrm{disc}}. Evaluation at a chosen point of S2S^2 has fiber Ω2Q≃Zdisc\Omega^2 Q\simeq\mathbb Z_{\mathrm{disc}}. Taking QQ in its abelian group space model separates a map into its value at the chosen point and a pointed map by translation, proving the product decomposition. In particular the direct image has extra components on the lower stratum although the initial value QQ is connected. The higher homotopy type of the link is used in this computation. Its fundamental group is trivial.

Inverting all exit paths in this example sends K×T◃K\times T^\triangleleft to KK. Groupoidification preserves products for these factors, as can be seen either from the existence of an initial object in the cone or by taking iterated colimits of constant points. Accordingly a functor comes from ordinary monodromy on XX precisely when η\eta is an equivalence. In that case B≃π∗AB\simeq\pi^* A by η\eta, which already implies the factorization. Conversely any representation from a groupoid sends the exit comparisons to equivalences. This agrees with the deformation retraction of XX to the lower stratum.

A van Kampen principle with directed coefficients

The right colimit for local presentations of a stratified space is a colimit of exit categories. Let {Uα}\{U_\alpha\} be an open cover of a conically stratified XX with locally weakly contractible strata, and let V∙→XV_\bullet\to X be its open Čech presentation. Its degree nn can be read as the disjoint union of the intersections Uα0∩⋯∩UαnU_{\alpha_0}\cap\cdots\cap U_{\alpha_n}. The exit-path van Kampen theorem gives Exit⁡P(X)≃colim⁡[n]∈ΔopExit⁡P(Vn)in Cat⁡∞.\operatorname{Exit}_P(X)\simeq \operatorname*{colim}_{[n]\in\Delta^{\mathrm{op}}}\operatorname{Exit}_P(V_n) \qquad\text{in }\operatorname{Cat}_\infty . For the application here this follows from the colimit-preserving exit-path functor on the hypersheaf topos, obtained by extending its values on opens [12, §4]. Indeed the coproducts of intersections give the Čech nerve in that topos, and its colimit is terminal because the cover map is an effective epimorphism. The exit functor sends this colimit to the displayed colimit. It sends a coproduct of the opens as hypersheaves to a coproduct of their exit categories, which agrees with the exit category of the disjoint union.

Proposition [prop:directed-colim-glue] now identifies constructible hypersheaves with coherent families of representations on these local exit categories. This operation compares local assignments on intersections by equivalences of representations. Their values along individual exit arrows inside an open set can be arbitrary maps. The equivalence relating two restrictions of a representation on an overlap and the map that the representation assigns to an exit arrow inside an open carry different parts of the data.

Suppose the strata are coarsened by a map of posets Q→PQ\to P, and both stratifications on XX are conical with locally weakly contractible strata. Let v:Exit⁡Q(X)→Exit⁡P(X)v:\operatorname{Exit}_Q(X)\to\operatorname{Exit}_P(X) be the comparison. Under these hypotheses vv is the localization at the arrows whose images are equivalences [12, Corollary 6.2.7]. Therefore a QQ-hyperconstructible hypersheaf is PP-hyperconstructible if and only if the corresponding representation sends every such arrow to an equivalence. One implication follows by restriction from a PP-representation. In the other direction the universal property of localization descends the representation to Exit⁡P(X)\operatorname{Exit}_P(X), and naturality of exodromy identifies its pullback with the original hypersheaf. The test is about coherent functors, so the comparison includes mapping spaces of classes that survive the coarsening. With the one-stratum coarsening, the test forces every transport to be invertible and recovers locally constant hypersheaves.

Recovering the directed category from its coefficients

Call an object aa of a presentable ∞\infty-category atomic if Map⁡(a,−)\operatorname{Map}(a,-) preserves all small colimits. This includes preservation of the empty colimit and is stronger than compactness, which asks only for preservation of filtered colimits. An atomic object is one on which it is possible to test arbitrary colimits by mapping out of it.

Proposition 1.3. Let CC be a small ∞\infty-category admitting a conservative functor to a poset. There is a canonical equivalence C≃(Fun⁡(C,S)at)op,C \simeq \left(\operatorname{Fun}(C,\mathcal S)_{\mathrm{at}}\right)^{\mathrm{op}}, where the subscript denotes the full subcategory of atomic objects.

Proof. Covariant representables hc=Map⁡C(c,−)h_c=\operatorname{Map}_C(c,-) are atomic since mapping out of hch_c is evaluation and colimits are pointwise. Conversely every functor F:C→SF:C\to\mathcal S is a colimit of covariant representables over its category of elements, by density. For FF atomic, apply Map⁡(F,−)\operatorname{Map}(F,-) to one such colimit presentation of FF. The identity of FF is in the image on components of some stage, since π0\pi_0 preserves colimits of spaces and every element of a colimit of sets comes from a stage. Thus FF is a retract of a representable, with maps F→hc→FF\to h_c\to F whose composite is homotopic to the identity.

Every endomorphism of an object of CC maps to an identity in the poset and is an equivalence by conservativity. By Yoneda the induced endomorphism of hch_c is also an equivalence. With one composite the identity and the other invertible, the two maps in the retract diagram are equivalences. Every atom is therefore representable. Covariant Yoneda is fully faithful on CopC^{\mathrm{op}}, giving the claimed variance and equivalence. ◻

Combining this proposition with Theorem 1.1 recovers the exit category as the opposite of the atomic constructible hypersheaves. Under this identification the atom associated to xx corepresents the stalk at xx. A map of atoms has variance reversed relative to an exit path. This follows from the natural transformation between evaluation functors and the contravariance of corepresenting objects. In particular the directed category can be recovered from the intrinsic colimits and arrows of the coefficient category, without a choice of a base point. If CC does not admit the conservativity assumption, the general statement from the proof recovers the idempotent completion, since atoms are then retracts of representables [2, §5.1.6].

Atomic generation is also a route to exodromy for stratified ∞\infty-topoi whose geometric origin need not be conical. Take a geometric morphism to Fun⁡(P,S)\operatorname{Fun}(P,\mathcal S), define the stratum topoi by geometric base change along the point topoi at pp, and define constructibility by locally constant restrictions. The formulation of [13, Definition 2.2.10] assumes atomic generation of the constructible subcategory, its closure under limits and colimits in the ambient topos, and preservation of limits by the inverse image of the stratification. The first property identifies the constructible category with a presheaf category by the free colimit completion of its atoms. The limit compatibility gives the left adjoint to stratification pullback on constructible coefficients, and this adjoint carries atoms to atoms because its right adjoint preserves colimits. Taking opposites then supplies the functor from the abstract exit category to PP. Thus a theorem proving these properties for a given stratified topos supplies a classifier without assuming that its simplices are represented by topological exit paths.

The profinite directed case

The framework of Barwick, Glasman and Haine supplies an arithmetic counterpart [10]. For a coherent scheme YY, in particular a quasicompact quasiseparated scheme, their category Gal⁡(Y)\operatorname{Gal}(Y) has geometric points as objects and étale specializations as morphisms. The construction equips it with a profinite categorical structure. A specialization carries incidence information, whereas endomorphisms at a geometric point account for the absolute Galois action on its residue-field data. Keeping both features is necessary to reconstruct the associated constructible coefficients.

For space coefficients in the sense of [10, Corollary 0.5.2], the theorem takes the form Cons⁡(Yeˊt;Sπ)≃Fun⁡cts(Gal⁡(Y),Sπ),\operatorname{Cons}(Y_{\textnormal{\'et}};\mathcal S_\pi) \simeq \operatorname{Fun}_{\mathrm{cts}} (\operatorname{Gal}(Y),\mathcal S_\pi), where on the left the constructibility condition is that of the stratified étale topos, and Sπ\mathcal S_\pi denotes the π\pi-finite spaces. Such spaces are truncated, have finitely many components, and have finite homotopy groups in each component. The right side uses the profinite stratified shape and its continuous representation category. When a field is viewed as a one-point scheme, a chosen separable closure supplies a single geometric base point up to isomorphism, and the loop automorphisms in this classification recover its absolute Galois group with its profinite structure. General schemes may also have noninvertible specializations between points.

The classifying pro-space of the profinite stratified shape recovers the ordinary profinite shape for spectral stratified topoi [10, Example 0.4.8]. This forgetful operation is the arithmetic counterpart of discarding directed transport by groupoidifying an exit category. The arithmetic theorem retains a continuity condition on representations, while Theorem 1.1 uses an ordinary small exit ∞\infty-category without an added profinite topology. The comparison thus relates the two generalizations through the role of coherence and direction, with the appropriate topology determined by the coefficient problem.

Higher topoi and presentation

Geometric morphisms and the classification of sections

The final chapter of Borceux and Janelidze passes from objects in a topos to morphisms between topoi. Their descent groupoid is a kernel construction in a two-dimensional category. It records identifications between geometric morphisms and thereby differs from the kernel pair of a map of objects in a fixed topos. In higher dimensions this distinction persists. The kernel of an étale geometric morphism can be computed from an internal Čech nerve. An external atlas of a topos requires separate information about its overlaps. We first prove the representation theorem for étale geometric morphisms, then use it to identify both the scope and the limits of the internal construction. The classical background is [1, §§7.7–7.10].

The category of geometric morphisms

An ∞\infty-topos is a presentable ∞\infty-category admitting a presentation by an accessible left exact localization of an ∞\infty-category of presheaves of spaces. Write P(C)=Fun⁡(Cop,S)\mathcal P(C)=\operatorname{Fun}(C^{\mathrm{op}},\mathcal S) for the presheaves on a small ∞\infty-category. A geometric morphism f ⁣:Y→Xf\colon\mathcal Y\to\mathcal X is specified by an inverse image f∗ ⁣:X→Yf^*\colon\mathcal X\to\mathcal Y. It preserves finite limits and small colimits, so the adjoint functor theorem supplies its right adjoint f∗f_*. Thus a morphism of topoi reverses the direction of the functor that performs substitution. In particular a localization L ⁣:X→X′L\colon\mathcal X\to\mathcal X' gives a geometric morphism X′→X\mathcal X'\to\mathcal X when LL is left exact. In the inverse image convention, a morphism from ff to gg in the category of geometric morphisms is a natural transformation f∗→g∗f^*\to g^*. Passing to mates reverses this direction if one works with direct images.

For each pair Y,X\mathcal Y,\mathcal X, put Geom⁡(Y,X)=Fun⁡lex,L(X,Y).\operatorname{Geom}(\mathcal Y,\mathcal X) =\operatorname{Fun}^{\mathrm{lex},\mathrm L} (\mathcal X,\mathcal Y). We take a full subcategory of the indicated functor category, with all natural transformations between geometric inverse images. Composition of functors and their transformations yields the (∞,2)(\infty,2)-category of geometric morphisms in this convention. Its underlying (∞,1)(\infty,1)-category, written Top∞\mathbf{Top}_{\infty}, keeps the maximal subgroupoids of the hom ∞\infty-categories. A comparison saying that two diagrams of inverse images are identified in Top∞\mathbf{Top}_{\infty} means that their comparison transformation is invertible. The noninvertible transformations can still be useful. For instance, a classifier for a theory can have models as the objects of a hom category and homomorphisms as its noninvertible transformations. The core records the space of models and their equivalences.

The terminal ∞\infty-topos in Top∞\mathbf{Top}_{\infty} is S\mathcal S. For each X\mathcal X, the inverse image of its structure map to S\mathcal S sends a space KK to ΔXK=colim⁡K1X\Delta_{\mathcal X}K=\operatorname{colim}_{K}1_{\mathcal X}. Indeed, S\mathcal S is the free cocompletion of the terminal category, and the requirement to preserve the terminal object specifies its image. The resulting functor preserves finite limits by the finite limit extension theorem for presheaves [2, Prop. 6.1.5.2]. Its right adjoint is global sections Map⁡X(1X,−)\operatorname{Map}_{\mathcal X}(1_{\mathcal X},-). The entire hom category Geom⁡(X,S)\operatorname{Geom}(\mathcal X,\mathcal S) is contractible. This argument uses the terminal-object condition on the inverse image, so it applies equally when maps of topoi carry higher homotopies.

An inverse image preserves effective epimorphisms because it preserves Čech nerves by left exactness and preserves their realizations. It also preserves monomorphisms by the diagonal test. An image factorization is an effective epimorphism followed by a monomorphism, so an inverse image preserves images as well. In particular the family of images of generating covers transported by the inverse image remains jointly covering. This elementary property is what links geometric morphisms to a site presentation.

The universal property of a site

Suppose CC has a Grothendieck topology. A sieve RR on cc is regarded as a monomorphism R↪y(c)R\hookrightarrow y(c) in P(C)\mathcal P(C). Sheafification P(C)→Shv⁡(C;S)\mathcal P(C)\to\operatorname{Shv}(C;\mathcal S) is the localization inverting covering sieve inclusions. If a family {ci→c}\{c_i\to c\} generates RR, the image of ∐iy(ci)→y(c)\coprod_i y(c_i)\to y(c) is RR. The sheaf condition at RR for FF is the equivalence F(c)→∼Map⁡P(C)(R,F).F(c)\xrightarrow{\sim} \operatorname{Map}_{\mathcal P(C)}(R,F). It imposes a coherent limit condition on the whole sieve. When there are pullbacks of the covering arrows, the condition also admits a Čech form by replacing RR by the realization of the Čech nerve of ∐iy(ci)→R\coprod_i y(c_i)\to R. This identification takes place already in presheaves, where effective epimorphisms and colimits are calculated pointwise.

Theorem 1.1 (Site description of inverse images). Let CC be a small ∞\infty-category with finite limits and a Grothendieck topology. Let j ⁣:C→Shv⁡(C;S)j\colon C\to\operatorname{Shv}(C;\mathcal S) be Yoneda followed by sheafification. For every ∞\infty-topos Y\mathcal Y, precomposition by jj gives an equivalence of Geom⁡(Y,Shv⁡(C;S))\operatorname{Geom}(\mathcal Y,\operatorname{Shv}(C;\mathcal S)) with the full subcategory of Fun⁡(C,Y)\operatorname{Fun}(C,\mathcal Y) on functors AA satisfying the following conditions.

  1. The functor AA preserves finite limits.

  2. For every family {ci→c}\{c_i\to c\} generating a covering sieve, the map ∐iA(ci)→A(c)\coprod_i A(c_i)\to A(c) is an effective epimorphism.

Proof. The free cocompletion property identifies AA with its colimit-preserving extension A‾ ⁣:P(C)→Y\overline A\colon\mathcal P(C)\to\mathcal Y. Under this identification the extension is left exact precisely when AA is left exact. Necessity uses the left exactness of Yoneda. For sufficiency apply the finite limit extension theorem [2, Prop. 6.1.5.2] to the finitely complete domain CC.

The extension then takes the image factorization ∐iy(ci)⟶R⟶y(c)\coprod_i y(c_i)\longrightarrow R\longrightarrow y(c) to an image factorization. Thus the condition on the generating family is equivalent to A‾\overline A inverting the covering sieve. By the universal property of localization, A‾\overline A factors if and only if all these inclusions are inverted. The descended functor is left exact since it is the restriction of A‾\overline A along the right adjoint inclusion of sheaves, which preserves limits. It preserves colimits by the localization property or by taking the corresponding right adjoint on presheaves, whose image lies in sheaves. These constructions are equivalences of full functor subcategories and preserve natural transformations. This proves the equivalence including its morphisms [2, Prop. 6.2.3.20]. ◻

For a presentation by an arbitrary accessible left exact localization L ⁣:P(C)→XL\colon\mathcal P(C)\to\mathcal X, the first part of the argument survives without any topology on CC. Restriction to CC identifies inverse images with those functors for which A‾\overline A is left exact and sends the LL-equivalences to equivalences. When LL is displayed by a set of generators, testing those generators suffices since A‾\overline A preserves colimits. An arbitrary localization can impose relations beyond those specified by covering sieves. Thus the general inverse image theorem starts with the actual localization, which may include a condition at a map between presheaves that is not a sieve inclusion.

For an ordinary site (C,J)(C,J) this construction provides a canonical lift of its ordinary topos. The 00-truncated objects of Shv⁡(C;S)\operatorname{Shv}(C;\mathcal S) are exactly its sheaves of sets. To see this, evaluation on the sheafified representables tests whether a sheaf takes 00-truncated values. Conversely, mapping into a sheaf taking such values can be tested on a colimit of sheafified representables and then gives a limit of 00-truncated spaces. The canonical lift depends only on the ordinary topos. In the terminology of higher topos theory it is the 11-localic lift, and inverse images from an ∞\infty-topos into it are detected by the corresponding functors on 00-truncated objects [2, Prop. 6.4.5.7]. One may therefore extend an ordinary geometric morphism between Grothendieck topoi to the 11-localic lifts. Recovering an arbitrary ∞\infty-topos from its 00-truncated objects is a stronger assertion and generally fails.

Étale maps and dependent substitution

For a map u ⁣:U→Vu\colon U\to V in an ∞\infty-topos X\mathcal X, base change has adjoints on both sides u!⊣u∗⊣u∗(u∗ ⁣:X/V⟶X/U).u_!\dashv u^*\dashv u_* \qquad \left( u^*\colon\mathcal X_{/V}\longrightarrow\mathcal X_{/U} \right). The first adjoint composes an arrow to UU with uu. The second exists because pullback preserves colimits. Pullback preserves finite limits as well. The resulting map of topoi X/U→X/V\mathcal X_{/U}\to\mathcal X_{/V} is étale. In particular πU ⁣:X/U→X\pi_U\colon\mathcal X_{/U}\to\mathcal X denotes the map obtained from U→1XU\to1_{\mathcal X}. More generally a geometric morphism to X\mathcal X is called étale when it is equivalent over X\mathcal X to some πU\pi_U. In this use of the word UU may have higher homotopy in its fibers.

Composition and pullback satisfy a relative projection formula. For A→BA\to B in X/V\mathcal X_{/V} and D→u∗BD\to u^*B in X/U\mathcal X_{/U}, there is a natural equivalence u!(u∗A×u∗BD)≃A×Bu!D.u_!\big(u^*A\times_{u^*B}D\big) \simeq A\times_B u_!D. Indeed the underlying object on either side is A×BDA\times_B D in X\mathcal X, with its composite map to VV. If a square in X\mathcal X is a pullback and the parallel horizontal maps are g′ ⁣:U′→Ug'\colon U'\to U and g ⁣:V′→Vg\colon V'\to V, then g∗u!≃u!′g′∗g^*u_!\simeq u'_!g'^* by the same calculation. For the right adjoints the comparison g∗u∗≃u∗′g′∗g^*u_*\simeq u'_*g'^* follows by testing against T∈X/V′T\in\mathcal X_{/V'} and using u∗g!T≃g!′(u′∗T)u^* g_!T\simeq g'_!(u'^*T). These identities describe the two ways to integrate a family after change of base. They extend Frobenius identities on subobjects, where the relative (−1)(-1)-truncation of dependent sum is read as existential quantification.

Theorem 1.2 (The slice classifies sections). For U∈XU\in\mathcal X, the category of geometric morphisms Y→X/U\mathcal Y\to\mathcal X_{/U} over the hom category Geom⁡(Y,X)\operatorname{Geom}(\mathcal Y,\mathcal X) is represented by the left fibration classified by Geom⁡(Y,X)⟶S,f⟼Map⁡Y(1Y,f∗U).\operatorname{Geom}(\mathcal Y,\mathcal X)\longrightarrow\mathcal S, \qquad f\longmapsto \operatorname{Map}_{\mathcal Y}(1_{\mathcal Y},f^*U). In particular, a lift of ff to a geometric morphism Y→X/U\mathcal Y\to\mathcal X_{/U} over X\mathcal X amounts to a section of f∗Uf^*U, with all paths between sections.

Proof. The object πU∗U\pi_U^*U is U×U→UU\times U\to U and carries the diagonal section δ\delta over UU. A geometric morphism hh from Y\mathcal Y to X/U\mathcal X_{/U} gives the pair (h∗πU∗,h∗δ)(h^*\pi_U^*,h^*\delta). Conversely, a pair (f∗,s)(f^*,s) with s ⁣:1Y→f∗Us\colon1_{\mathcal Y}\to f^*U induces the functor (A→aU)⟼1Y×f∗Uf∗A.(A\xrightarrow{a}U)\longmapsto 1_{\mathcal Y}\times_{f^*U} f^*A. Here the first step applies f∗f^* in the slice. The second step is pullback along ss. Both steps preserve small colimits and finite limits. For the first step, finite limits follow from left exactness and colimits in a slice are computed in its ambient category. For the second step, colimits follow from universality and limits from the right adjoint description of pullback as right adjoint to composition. Consequently this formula defines a geometric inverse image.

For any A→UA\to U there is a pullback square in the slice expressing it as πU∗A×πU∗U1X/U\pi_U^*A\times_{\pi_U^*U}1_{\mathcal X_{/U}}, with the right-hand map determined by δ\delta. An inverse image preserves this square. Applying the construction to (h∗πU∗,h∗δ)(h^*\pi_U^*,h^*\delta) therefore recovers h∗h^* naturally on objects and maps. The reverse composite recovers (f∗,s)(f^*,s) by evaluating on objects C×UC\times U and then on the diagonal. These natural comparisons also hold for transformations. In detail, a transformation between pairs consists of a transformation α ⁣:f∗→g∗\alpha\colon f^*\to g^* and a specified path αUs≃t\alpha_U s\simeq t. It induces the transformation of the pullback formulas by the universal property of the pullback. Restriction to πU∗\pi_U^* and evaluation at δ\delta recovers the original pair of data. Since these operations preserve parametrized families of transformations, the comparison is an equivalence of ∞\infty-categories. The category of these pairs is the Grothendieck construction of the displayed space-valued functor. See also [2, Prop. 6.3.5.5 and Cor. 6.3.5.6]. ◻

The section classifier gives two useful refinements. First, if g ⁣:Y→Xg\colon\mathcal Y\to\mathcal X is geometric, then base change of πU\pi_U has source Y/g∗U\mathcal Y_{/g^*U}. A test geometric morphism f ⁣:Z→Yf\colon\mathcal Z\to\mathcal Y lifts to that slice exactly by specifying a section of f∗g∗Uf^*g^*U, which also describes a compatible lift of gfgf to X/U\mathcal X_{/U}. Second, there is an equivalence of ∞\infty-categories X≃(Top∞/X)et,U⟼πU.\mathcal X \simeq \big(\mathbf{Top}_{\infty}/\mathcal X\big)_{\mathrm{et}}, \qquad U\longmapsto \pi_U. Essential surjectivity is the definition of étale. For full faithfulness use the section classifier with Y=X/V\mathcal Y=\mathcal X_{/V} and f=πVf=\pi_V. A section of V×U→VV\times U\to V in that slice is precisely a map V→UV\to U. Spaces of such sections compute the relative mapping spaces. This result identifies a whole category of geometric tests over X\mathcal X with the objects of X\mathcal X, without choosing a site or a point.

Internal classifying objects and external classifiers

An internal group GG in X\mathcal X admits a delooping BG∈XBG\in\mathcal X. For every f ⁣:Y→Xf\colon\mathcal Y\to\mathcal X, f∗BG≃B(f∗G)f^*BG\simeq B(f^*G) because the bar construction uses finite products and geometric realization. Combining the slice classifier with torsor classification gives the space of lifts of ff through X/BG\mathcal X_{/BG} as the space of f∗Gf^*G-torsors in Y\mathcal Y. The universal torsor there is classified by the diagonal section of πBG∗BG\pi_{BG}^*BG. For an arrow V→1V\to1 in X\mathcal X the same rule describes GG-torsors in the slice X/V\mathcal X_{/V} by maps V→BGV\to BG. For a general geometric ff, its inverse image replaces every coefficient object before the section is taken.

The object BGBG represents torsors on objects of X\mathcal X. The slice X/BG\mathcal X_{/BG} represents the choice of a torsor over a geometric map into X\mathcal X. In both statements the group is fixed in the base. The category of GG-objects includes arbitrary equivariant maps and will figure in the presentation of the next chapter.

Groupoid presentations of higher topoi

A presentation gives two kinds of information. It identifies a domain on which objects can be described locally, and it says how to identify their pullbacks on overlaps. The ordinary Joyal–Tierney theorem provides both pieces using locales. The higher construction below presents slices by étale geometric morphisms and keeps track of the type of each overlap. The difference becomes visible even when the atlas itself is the topos of spaces.

The descent formula for a diagram of slices

A colimit in an ∞\infty-topos is van Kampen in the following sense. For a small diagram U ⁣:I→XU\colon I\to\mathcal X with colimit VV, restriction along the colimit cone gives an equivalence of ∞\infty-categories X/V ≃ lim⁡i∈IopX/Ui.\mathcal X_{/V}\ \simeq\ \lim_{i\in I^{\mathrm{op}}}\mathcal X_{/U_i}. Transition functors on the right are pullbacks. Equivalently, a Cartesian family Ai→UiA_i\to U_i, meaning one whose transport squares are pullbacks, has a realization colim⁡iAi→V\operatorname{colim}_i A_i\to V whose pullbacks are the given family. This is the descent form of the higher Giraud axioms [2, Thm. 6.1.3.9 and Prop. 6.1.3.10]. The realization description also shows why preserving pullbacks by colimits of slices is a stronger condition than mere universality of colimits. Universality gives the recovery of a slice object from an already existing pullback family. Descent asserts that a compatible family given independently of a global object descends.

Theorem 1.1 (Colimits of slices as higher topoi). With the preceding hypotheses, there is an equivalence colim⁡i∈ITop∞X/Ui ≃ X/V.\operatorname*{colim}_{i\in I}^{\mathbf{Top}_{\infty}} \mathcal X_{/U_i}\ \simeq\ \mathcal X_{/V}. For every ∞\infty-topos Z\mathcal Z, the induced comparison Geom⁡(X/V,Z) ⟶ lim⁡i∈IopGeom⁡(X/Ui,Z)\operatorname{Geom}(\mathcal X_{/V},\mathcal Z) \ \longrightarrow\ \lim_{i\in I^{\mathrm{op}}} \operatorname{Geom}(\mathcal X_{/U_i},\mathcal Z) is an equivalence of ∞\infty-categories.

Proof. Apply Fun⁡(Z,−)\operatorname{Fun}(\mathcal Z,-) to (20). The limit of functor categories classifies a coherent family of functors to the slices. It remains to restrict to geometric inverse images. The projections from the limit preserve finite limits and small colimits, since all the transition functors do. For example, a diagram of compatible families has its componentwise colimit formed in each slice. A transition functor carries each component colimit to the corresponding colimit, so the colimit inherits the required compatibility and satisfies the universal property. The same argument applies to finite limits. Joint conservativity of projections then says that a functor into the limit is left exact and colimit-preserving exactly when each component is. Natural transformations are restricted through the same equivalence of functor categories. Taking cores yields the mapping space universal property of the colimit of topoi. This is the fixed base case of the more general colimit theorem for étale geometric morphisms [2, Thm. 6.3.5.13]. ◻

From a groupoid to its quotient

Let G∙ ⁣:Δop→XG_\bullet\colon\Delta^{\mathrm{op}}\to\mathcal X be a groupoid object, in the Segal sense with all arrows invertible. Put Q=∣G∙∣Q=|G_\bullet|. The Segal identities and invertibility concern an internal diagram in X\mathcal X. They imply that G∙G_\bullet is the Čech nerve of G0→QG_0\to Q. We recall a proof using the van Kampen property because its argument makes explicit the role of invertibility.

Proposition 1.2 (Effectiveness of internal groupoids). The map G0→QG_0\to Q is an effective epimorphism, and there are compatible equivalences Gn≃G0×Q⋯×QG0⏟n+1.G_n \simeq \underbrace{G_0\times_Q\cdots\times_QG_0}_{n+1}.

Proof. Define a simplicial object E∙E_\bullet by En=Gn+1E_n=G_{n+1}, where the simplicial operators retain the first vertex and act on the remaining vertices. In degree nn, its map to GnG_n forgets that first vertex. There is also an augmentation E∙→G0E_\bullet\to G_0 given by the retained vertex. Duplicating the retained vertex is an extra degeneracy contracting this augmentation, so ∣E∙∣≃G0|E_\bullet|\simeq G_0.

The map E∙→G∙E_\bullet\to G_\bullet is Cartesian. To check this, first pull it back along a vertex of a simplex GnG_n. Over the first vertex of the remaining chain, the assertion is a Segal identity since the extra data is the one incoming arrow. Over any other vertex, the incoming arrow can be transported forward along the chosen segment of that chain. Composition by an invertible arrow is an equivalence on the corresponding space of arrows, with inverse given by composition by its inverse. This argument is functorial on generalized points and proves the pullback identity for any chosen vertex. To check an arbitrary simplicial operator Gn→GmG_n\to G_m, evaluate the two candidate pullbacks over a vertex of GmG_m. Both identify with the same pullback of E0→G0E_0\to G_0 along the composite vertex of GnG_n.

Descent now identifies E0E_0 with the pullback of ∣E∙∣|E_\bullet| along G0→QG_0\to Q. Its two projections are the endpoints of the arrow in G1G_1. Hence G1≃G0×QG0G_1\simeq G_0\times_Q G_0. Compatibility with longer chains follows by the Segal maps and their composition coherences. The augmented Čech nerve therefore realizes to QQ, proving effectiveness. This is the shift argument of [15, §6.10] in the notation of internal groupoids. ◻

A G∙G_\bullet-object is a Cartesian transformation A∙→G∙A_\bullet\to G_\bullet. It is given first by an object A0→G0A_0\to G_0, then by transport of this object along arrows of G1G_1, with composition homotopies on G2G_2 and their higher compatibility. Equivariant morphisms are maps of such transformations. The corresponding ∞\infty-category will be denoted Act⁡X(G∙)\operatorname{Act}_{\mathcal X}(G_\bullet). Applying (20) gives Act⁡X(G∙) ≃ lim⁡[n]∈ΔX/Gn ≃ X/Q.\operatorname{Act}_{\mathcal X}(G_\bullet) \ \simeq\ \lim_{[n]\in\Delta}\mathcal X_{/G_n} \ \simeq\ \mathcal X_{/Q}. The inverse sends A→QA\to Q to the family A×QGnA\times_QG_n. The forward map sends a Cartesian family to its realization. Under Theorem 1.1, the right-hand side is also the colimit of the simplicial diagram of étale geometric morphisms X/Gn\mathcal X_{/G_n}. The two variance conventions agree because the geometric morphisms point in the direction opposite to their inverse image functors.

For the special case of U→1XU\to1_{\mathcal X} effective epi, let Gn=Un+1G_n=U^{n+1}. Then Q≃1XQ\simeq1_{\mathcal X} and (21) recovers X\mathcal X itself from the Čech diagram of X/U→X\mathcal X_{/U}\to\mathcal X. Indeed Theorem [thm:geo-sections] identifies X/U×X⋯×XX/U⏟n+1≃X/Un+1,\underbrace{\mathcal X_{/U}\times_{\mathcal X}\cdots \times_{\mathcal X}\mathcal X_{/U}}_{n+1} \simeq \mathcal X_{/U^{n+1}}, where the fiber products on the left are in Top∞\mathbf{Top}_{\infty}. The inverse image of this projection is conservative since evaluation of a Cartesian family at degree zero detects equivalences. Conversely, if πU∗\pi_U^* is conservative, it reflects the equivalence obtained by pulling back im⁡(U→1)→1\operatorname{im}(U\to1)\to1 along UU. This image must be all of 11, so U→1U\to1 is an effective epimorphism. For étale geometric morphisms, surjectivity in the sense of conservative inverse image thus agrees with being induced by an effective epimorphism of objects. All the formulas hold in a slice over any specified base object.

Presentations here are invariant under refinement in a concrete sense. If W→U→1XW\to U\to1_{\mathcal X} are effective epimorphisms, their Čech nerves have a map W∙+1→U∙+1W^{\bullet+1}\to U^{\bullet+1}. Pullback on actions is an equivalence because under (21) it identifies with the identity functor of X\mathcal X. Two covers U→1XU\to1_{\mathcal X} and V→1XV\to1_{\mathcal X} admit a common refinement by U×VU\times V. Both projections are covers by pullback stability. There are typically many different simplicial diagrams presenting the same topos.

The classical localic theorem

For a locale LL, let O(L)\mathcal O(L) be its frame. A map of locales f ⁣:L→Mf\colon L\to M is open when f∗ ⁣:O(M)→O(L)f^*\colon\mathcal O(M)\to\mathcal O(L) admits a left adjoint f!f_! satisfying the Frobenius identity f!(a∧f∗b)=f!(a)∧b.f_!(a\wedge f^*b)=f_!(a)\wedge b. An étale locale over MM is one locally isomorphic over MM to open sublocales of MM. Its sheaf of sections is an ordinary sheaf of sets on MM. An open localic groupoid consists of locales of objects and arrows, groupoid operations in locales, and open source and target maps.

The Joyal–Tierney theorem provides, for every ordinary Grothendieck topos E\mathcal E, an open localic groupoid GG with E\mathcal E equivalent to the category of étale GG-objects. These are étale locales over G0G_0 with an action along G1G_1, or equivalently sheaves of sets on G0G_0 with the corresponding equivariance. An ingredient of the theorem supplies a locale LL with an open effective-descent geometric morphism Shv⁡(L;Set⁡)→E\operatorname{Shv}(L;\operatorname{Set})\to\mathcal E. Take its Čech nerve by pseudopullbacks of ordinary topoi. The map from a localic topos to an arbitrary Grothendieck topos is localic in the relative sense. Indeed its source has subterminal generators, and these are subobjects of the inverse image of the terminal object. Base change preserves relative localicity. Consequently the iterated overlaps are localic over the localic source and therefore localic. Openness under base change gives the openness of the groupoid. Effective descent identifies E\mathcal E with the category of sheaves on G0G_0 endowed with the required action. The existence of the localic cover and the effective-descent theorem are substantive parts of the classical result [14, Chs. VII–VIII]; see also [1, Thms. 7.10.1 and 7.10.7].

There are two uses of étaleness here. In the ordinary localic result, étale locales carry the sheaves on which the groupoid acts. The atlas of the target topos is supplied as an open geometric morphism of effective descent. In the higher relative result, an atlas X/U→X\mathcal X_{/U}\to\mathcal X is itself an étale geometric morphism. Its source need not be localic. The theory of internal groupoid actions proves the relative presentation once a cover is chosen. In the classical theorem, the existence of an external atlas with localic overlaps is additional content.

When a higher étale atlas has localic overlaps

Call an ∞\infty-topos 00-localic if it is equivalent to Shv⁡(L;S)\operatorname{Shv}(L;\mathcal S) for a locale LL. Its locale is recovered from the subobjects of 11 [2, §7.3.2]. For a space AA, the subobjects of the terminal object in S/A\mathcal S_{/A} are indexed by subsets of π0A\pi_0 A. To see this, a subobject has fiber either empty or contractible, and this property is invariant under transport along paths. Conversely each union of path components gives such a subobject.

Lemma 1.3. The ∞\infty-topos S/A\mathcal S_{/A} is 00-localic exactly when AA is 00-truncated.

Proof. For a discrete space AA, a family over AA is a sheaf of spaces on the discrete locale with set of points AA. For the converse, the reconstructed locale would have frame the power set of π0A\pi_0 A. Over each nonempty connected component A0A_0 the corresponding open slice would then be equivalent to S\mathcal S. Such an equivalence must identify the constant-object functor S→S/A0\mathcal S\to\mathcal S_{/A_0} with an equivalence, since that functor is the unique geometric inverse image out of S\mathcal S up to equivalence. For a∈A0a\in A_0, the object {a}→A0\{a\}\to A_0 has nonempty fiber at aa. A constant object with nonempty fiber has a global section. A section of {a}→A0\{a\}\to A_0 makes id⁡A0\operatorname{id}_{A_0} homotopic to the constant map at aa, so A0A_0 is contractible. This holds for each connected component, giving a discrete homotopy type. ◻

Theorem 1.4 (Localic étale Čech presentations over spaces). Let BB be a small space. The ∞\infty-topos S/B\mathcal S_{/B} admits a surjective étale geometric morphism for which both the source and the double overlap are 00-localic if and only if BB is 11-truncated. In that case every higher overlap of the chosen atlas can also be taken 00-localic.

Proof. Write the covering source as S/U\mathcal S_{/U}, with map to S/B\mathcal S_{/B} induced by U→BU\to B. Surjectivity implies U→BU\to B is effective epi, hence surjective on path components. The source is 00-localic only if UU is discrete by Lemma 1.3. For a discrete UU, the space U×BUU\times_B U is the disjoint union, over pairs u,v∈Uu,v\in U, of spaces of paths in BB from the image of uu to the image of vv. Since the double overlap is S/(U×BU)\mathcal S_{/(U\times_B U)}, its localicity forces these path spaces to be discrete. Using u=vu=v and one point over each component shows that each loop space of BB is 00-truncated. Equivalently, BB is 11-truncated.

Conversely, take UU discrete with one chosen point of BB in each component. This is an effective epimorphism, including the empty case. Its higher fiber products over BB have discrete homotopy types, since a tuple of points is accompanied by a tuple of paths in the 11-truncated space BB. Lemma 1.3 identifies all their slices with 00-localic ∞\infty-topoi. ◻

For an infinite discrete group GG and B=BGB=BG, the atlas coming from ∗→BG*\to BG has nnth term S/Gn\mathcal S_{/G^n}. Equation (21) gives its category of actions as Fun⁡(BG,S)\operatorname{Fun}(BG,\mathcal S). Its 00-truncated objects are ordinary GG-sets. Taking instead B=K(Z,2)B=K(\mathbb Z,2) yields the atlas S→S/B\mathcal S\to\mathcal S_{/B} whose first overlap is S/K(Z,1)\mathcal S_{/K(\mathbb Z,1)}. That overlap is 11-localic since it is a presheaf category on an ordinary groupoid, and it fails to be 00-localic by Lemma 1.3. Moreover S/B\mathcal S_{/B} and S\mathcal S have the same ordinary topos of 00-truncated objects. Straightening identifies the first ordinary topos with set-valued functors on BB, which factor through the fundamental groupoid. For K(Z,2)K(\mathbb Z,2) that groupoid is contractible. The two higher topoi are inequivalent by Lemma 1.3. Theorem 1.4 concerns the specific class of surjective étale geometric morphisms and does not assume a theorem about all other notions of higher atlas.

Hyperdescent and the role of truncation

In any ∞\infty-topos an augmented simplicial object V∙→VV_\bullet\to V is a hypercover if each map to its relative matching object is an effective epimorphism. In degree zero the matching object is VV. In degree one it is V0×VV0V_0\times_VV_0. In higher degrees it is the object of compatible boundaries relative to VV. Each matching condition asks for local existence of a lift. When the augmentation is a Čech nerve, there is a stronger identity. The levels are specified iterated pullbacks and the geometric realization of a cover is exactly VV.

The realization of an arbitrary hypercover maps to VV by an ∞\infty-connective morphism [2, Lem. 6.5.3.11]. This means its relative nn-truncation is terminal for every finite nn. Applying (20) to V∙V_\bullet identifies lim⁡ΔX/Vn\lim_\Delta\mathcal X_{/V_n} with X/∣V∙∣\mathcal X_{/|V_\bullet|}. Consequently the restriction comparison for all objects over VV is an equivalence precisely when ∣V∙∣→V|V_\bullet|\to V is an equivalence. Indeed, if the restriction g∗g^* for a map gg is an equivalence, so is its left adjoint g!g_!. Applying g!g_! to the terminal object in the source slice gives gg as an object of the target slice, so gg is terminal there. Conversely an equivalence gg induces an equivalence of slices.

An ∞\infty-topos is hypercomplete if every ∞\infty-connective map is an equivalence. Hypercompletion is the left exact localization which forces these equivalences. Each ∞\infty-connective morphism gives a constant simplicial hypercover in its target slice, according to the equivalent iterated-diagonal test for ∞\infty-connectivity [2, Lem. 6.5.3.5]. Thus effectiveness of all hypercovers is equivalent to hypercompleteness [2, Thm. 6.5.3.12]. A bounded truncated object already satisfies the extra locality condition. Mapping into an nn-truncated object cannot distinguish an ∞\infty-connective map from an equivalence, by orthogonality to nn-truncated maps. In particular, passing to ordinary sheaves of sets cannot test whether a given ∞\infty-topos is hypercomplete.

A concrete instance of the distinction is the canonical 11-localic lift of continuous Zp\mathbb Z_p-sets. One can present it by the category of transitive quotients Zp/pnZp\mathbb Z_p/p^n\mathbb Z_p with its atomic topology. This is an ∞\infty-topos of ordinary sheaves of spaces in the sense of covering sieves, and it fails to be hypercomplete by Wieland’s example [2, Warning 7.2.2.31]. The site is ordinary and its underlying ordinary topos already has a classical group-theoretic presentation. Neither fact removes the hyperdescent distinction. The higher quotient (21) uses an effective groupoid object, so it works in that topos before hypercompletion as well as in hypercomplete examples.

Proposition 1.5 (Relative hypercompletion). Let L ⁣:X→X∧L\colon\mathcal X\to\mathcal X^\wedge be hypercompletion. For every U∈XU\in\mathcal X, the induced functor L/U ⁣:X/U⟶(X∧)/LUL_{/U}\colon\mathcal X_{/U}\longrightarrow (\mathcal X^\wedge)_{/LU} is the hypercompletion of the slice. In particular, applying LL to a hypercover V∙→VV_\bullet\to V gives a slice comparison (X∧)/LV ≃ lim⁡[n]∈Δ(X∧)/LVn.(\mathcal X^\wedge)_{/LV} \ \simeq\ \lim_{[n]\in\Delta} (\mathcal X^\wedge)_{/LV_n}.

Proof. Let i ⁣:X∧↪Xi\colon\mathcal X^\wedge\hookrightarrow\mathcal X be the fully faithful inclusion. For T→LUT\to LU, the right adjoint of L/UL_{/U} is the object U×iLUiT→UU\times_{iLU}iT\to U. Indeed, a map into this pullback over UU amounts under L⊣iL\dashv i to a map into TT over LULU. The counit is invertible by left exactness of LL and the counit equivalence LiT≃TLiT\simeq T. It follows that L/UL_{/U} is a localization. A morphism in the slice is inverted precisely when its underlying morphism in X\mathcal X is inverted by LL. This is the class of ∞\infty-connective maps by the strong saturation theorem [2, Prop. 6.5.2.8]. Connectivity in the slice agrees with the connectivity of the underlying map. For example, the slice over its codomain in X/U\mathcal X_{/U} identifies with the slice over that same codomain in X\mathcal X, so the relative truncation tests are identical. This identifies the localization as the desired hypercompletion.

Finally, LL preserves the realization of V∙V_\bullet and sends its ∞\infty-connective augmentation to an equivalence. Apply (20) inside X∧\mathcal X^\wedge. The result can also be read as descent for relative hypercomplete objects. The base object in the formula is LVLV, even when VV itself has not been assumed hypercomplete. ◻

In the Zp\mathbb Z_p example the constant hypercover of a nonequivalence supplied by infinite connectivity has a slice comparison that fails before hypercompletion. The proposition makes the comparison an equivalence after applying LL to its entire augmentation. The higher Čech theorem continues to apply to actual groupoid quotients in the original topos.

Reconstruction and comparison

A classification theorem identifies a category of forms or coverings from a symmetry object and chosen coefficients. A reconstruction theorem asks what symmetry can be read back from that category. This direction is sensitive to the fiber functor, the availability of representing objects, and the tests used to define equivalence. We finish by comparing the different reconstruction problems.

Recovering the space of fibers

Let BB be a small space and put L(B)=Fun(B,S)\mathcal L(B)=\mathop{\mathrm{Fun}}(B,\mathcal{S}). Write evb:L(B)→S\mathrm{ev}_b:\mathcal L(B)\to\mathcal{S} for evaluation. A path from bb to cc induces a natural equivalence evb→evc\mathrm{ev}_b\to\mathrm{ev}_c. The space of such natural equivalences remembers all paths and their homotopies.

Theorem 1.1 (Reconstruction from the evaluations). The functor B⟶Fun(L(B),S),b⟼evbB\longrightarrow\mathop{\mathrm{Fun}}(\mathcal L(B),\mathcal{S}), \qquad b\longmapsto\mathrm{ev}_b is fully faithful. In particular, as group objects in spaces, Aut(evb)≃ΩbB.\mathop{\mathrm{Aut}}(\mathrm{ev}_b)\simeq\Omega_b B. These assertions concern natural transformations in the full functor category.

Proof. For b∈Bb\in B, the covariant representable Rb=MapB(b,−)R_b=\mathop{\mathrm{Map}}_B(b,-) is an object of L(B)\mathcal L(B). The covariant version of Yoneda gives evb≃MapL(B)(Rb,−)\mathrm{ev}_b\simeq\mathop{\mathrm{Map}}_{\mathcal L(B)}(R_b,-). A natural transformation from evaluation at bb to evaluation at cc is, by a second application of Yoneda, a point of MapL(B)(Rc,Rb)≃Rb(c)\mathop{\mathrm{Map}}_{\mathcal L(B)}(R_c,R_b)\simeq R_b(c). The equivalence is natural in b,cb,c, and it respects composition because the two appearances of the contravariant representing-object assignment reverse order twice. Equivalently, a path u:b→cu:b\to c acts on each FF by F(u)F(u), which agrees with this identification. Every arrow in BB is invertible, so the equivalence of mapping spaces identifies the full space of natural transformations between the two evaluations with its subspace of equivalences. ◻

This proof explains the regular-action formula. On writing B=BGB=BG, the representable R∗R_* is the regular left GG-space. Its endomorphisms are right multiplications, which compose in the opposite order. Passing from the representing object to the evaluation functor reverses the order again. The group on the right side of Theorem 1.1 is therefore compatible with covariant monodromy. For the groupoid BB, the pointed reconstruction only uses one evaluation, while the full family of evaluations retains the components and the paths between them.

Proposition 1.2 (Reconstruction from all geometric points). The core of the category of colimit-preserving, finite-limit-preserving functors S/B→S\mathcal{S}_{/B}\to\mathcal{S} is equivalent to BB. Under straightening, its objects are the evaluations of Theorem 1.1.

Proof. The universal property of the slice topos, as in Chapter [ch:geometric], identifies geometric morphisms T→S/B\mathcal T\to\mathcal{S}_{/B} with pairs consisting of a geometric morphism a:T→Sa:\mathcal T\to\mathcal{S} and a section of a∗Ba^*B. For T=S\mathcal T=\mathcal{S}, the space of choices of aa is contractible. A colimit-preserving functor from spaces is determined by the image of the point, and left exactness forces that image to be terminal. The section space of BB is BB. The inverse image associated to bb pulls back a family along b:∗→Bb:*\to B, and that is evaluation under straightening. ◻

The distinction between this result and a general theorem on points of topoi is substantial. The slice of spaces has specially chosen geometric points, determined by the family classifier BB. A general higher topos has a shape defined by constant-object tests. That definition does not assert that its points are the components of its shape, or that its points distinguish all objects.

What truncation preserves

The preceding calculation is also valid with truncated coefficients. For an integer n≥0n\geq0, let Ln(B)=Fun(B,S≤n)\mathcal L_n(B)=\mathop{\mathrm{Fun}}(B,\mathcal{S}_{\leq n}). Denote the evaluation functor on Ln(B)\mathcal L_n(B) by evbn\mathrm{ev}_b^n, giving its output as a space.

Proposition 1.3 (Truncated reconstruction). There is a natural equivalence MapFun(Ln(B),S)(evbn,evcn)≃τ≤nMapB(b,c).\mathop{\mathrm{Map}}_{\mathop{\mathrm{Fun}}(\mathcal L_n(B),\mathcal{S})} (\mathrm{ev}_b^n,\mathrm{ev}_c^n) \simeq\tau_{\leq n}\mathop{\mathrm{Map}}_B(b,c). Every component of the left side consists of equivalences. Restricted evaluation consequently defines a fully faithful functor τ≤n+1B→Fun(Ln(B),S)\tau_{\leq n+1}B\to\mathop{\mathrm{Fun}}(\mathcal L_n(B),\mathcal{S}).

Proof. Objectwise truncation is left adjoint to the inclusion Ln(B)↪L(B)\mathcal L_n(B)\hookrightarrow\mathcal L(B). Hence evbn\mathrm{ev}_b^n is represented by Rbn=τ≤nRbR_b^n=\tau_{\leq n}R_b in Ln(B)\mathcal L_n(B). The two Yoneda identifications give Map(evbn,evcn)≃MapLn(B)(Rcn,Rbn)≃Rbn(c)\mathop{\mathrm{Map}}(\mathrm{ev}_b^n,\mathrm{ev}_c^n) \simeq \mathop{\mathrm{Map}}_{\mathcal L_n(B)}(R_c^n,R_b^n) \simeq R_b^n(c). The components of Rbn(c)R_b^n(c) come from actual paths in BB, so all corresponding natural transformations are equivalences. The full subcategory spanned by the evaluations is thus an (n+1)(n+1)-truncated groupoid. The evaluation functor factors through τ≤n+1B\tau_{\leq n+1}B by its truncation universal property. The mapping-space description identifies the factorization as fully faithful. ◻

In particular, set-valued monodromy recovers the fundamental groupoid and its groups of loops at chosen points. A space of coefficients truncated in a higher degree records one additional degree in the classifying groupoid. The regular representable need not lie in the subcategory of finite coefficients. The proof above therefore gives no reconstruction of the uncompleted base from finite tests. The pro-space mapping formula and finite-range hypotheses from Chapters [ch:profinite] and [ch:shape] are the relevant comparison in that setting.

For infinite discrete groups, the loss can already occur with set coefficients. The additive group Q\mathbb Q has no nontrivial finite quotient. If a map Q→H\mathbb Q\to H has finite image of order dividing mm, write any xx as mymy. Its image is the mmth power of the image of yy, hence the identity. Any action of Q\mathbb Q on a finite set is trivial, since it factors through a finite permutation group. In contrast the regular action recovers Q\mathbb Q itself by Proposition 1.3. The choice of coefficient category therefore changes the reconstruction even for an ordinary group.

A comparison of the classification results

The constructions used throughout the book can be read in the following order. Choose a base and the objects considered over it. Specify the splitting condition. Determine the overlap diagram under which such objects can descend. Then decide whether the overlap diagram has a smaller groupoid, pro-object, or directed category as a presentation. The outputs below indicate which of these steps require hypotheses.

Input Key hypothesis Classification
Input Key hypothesis Classification
An internal cover U→BU\to B Effective epimorphism in an ∞\infty-topos Objects over BB as cartesian actions of the overlap groupoid. Chapter [ch:descent].
An adjunction L⊣RL\dashv R and a cover U→BU\to B Admissibility in slices and effective descent for the chosen maps Objects split by UU as a limit of slices over the image overlap diagram. Additional exactness makes that diagram a groupoid. Chapter [ch:reflections].
A space, a pro-space, or the shape of a topos Representable base for ordinary straightening or the coefficient and locality hypotheses of higher Galois theory Monodromy on the space, stage representations, or locally constant objects, respectively. Chapters [ch:profinite]–[ch:shape].
A spectral algebra R→SR\to S The flat or stable descent hypotheses appropriate to the objects; a Galois tensor identity for action comparison Amitsur descent and, when the tensor identity holds, coherent semilinear actions. Chapters [ch:algebra]–[ch:spectral].
A stratified base The constructibility and geometric hypotheses of the corresponding exodromy theorem Stratumwise systems with transport along exit paths or an étale profinite category. Chapter [ch:exodromy].
A topos and a geometric atlas Effectivity of the chosen presentation in the specified category of topoi or slices A category of sheaves or descent objects associated to that presentation. Chapter [ch:presentations].

These implications have several useful reversals and limitations. A trivializing cover gives a subgroup correspondence if the classification of intermediate objects imposes connectedness, pointing, and the appropriate truncation. Dropping these conditions leads to actions on families, with a different notion of map. The totalization formula for a relative reflection requires compatible coefficient categories. Promoting a general image simplicial diagram to a groupoid requires a check on the Segal pullbacks. A theorem on profinite completion concerns the tests retained by that completion. Finally, specialization maps for a stratified space can fail to be invertible, so passing to the groupoid completion has the effect of requiring additional inverses from its representations.

The common result is a homotopy-coherent descent statement whose strength is measured by its mapping spaces. The classical correspondence can then be obtained by restriction, truncation, or reversal of variance in those cases where the intermediate mathematical objects have the requisite form. The higher theorems keep the coherent data visible and make it possible to say exactly which features of the original symmetry object can be recovered.

References

  1. Francis Borceux and George Janelidze. Galois Theories. Cambridge Studies in Advanced Mathematics 72. Cambridge University Press, 2001. Publisher edition.

  2. Jacob Lurie. Higher Topos Theory. Annals of Mathematics Studies 170. Princeton University Press, 2009. arXiv:math/0608040.

  3. Marc Hoyois. Higher Galois theory. Journal of Pure and Applied Algebra 222 (2018), 1859–1877. arXiv:1506.07155.

  4. Thomas Nikolaus, Urs Schreiber, and Danny Stevenson. Principal infinity-bundles—General theory. Journal of Homotopy and Related Structures 10 (2015), 749–801. arXiv:1207.0248.

  5. Jacob Lurie. Derived Algebraic Geometry VII: Spectral Schemes, 2011. Author’s manuscript.

  6. Jacob Lurie. Higher Algebra, 2017. Author’s manuscript.

  7. John Rognes. Galois extensions of structured ring spectra. Memoirs of the American Mathematical Society 192 (2008), no. 898. arXiv:math/0502183.

  8. Akhil Mathew. The Galois group of a stable homotopy theory. Advances in Mathematics 291 (2016), 403–541. arXiv:1404.2156.

  9. The Stacks Project Authors. The Stacks Project. Tag 0244 and Tag 023R.

  10. Clark Barwick, Saul Glasman, and Peter Haine. Exodromy. arXiv:1807.03281.

  11. Louis Martini. Yoneda’s lemma for internal higher categories. arXiv:2103.17141.

  12. Mauro Porta and Jean-Baptiste Teyssier. Topological exodromy with coefficients. arXiv:2211.05004.

  13. Peter J. Haine, Mauro Porta, and Jean-Baptiste Teyssier. Exodromy beyond conicality. arXiv:2401.12825.

  14. André Joyal and Myles Tierney. An extension of the Galois theory of Grothendieck. Memoirs of the American Mathematical Society 51 (1984), no. 309. DOI 10.1090/memo/0309.

  15. Charles Rezk. Toposes and homotopy toposes, v0.15, 2010. Author’s manuscript.

  16. Daniel Dugger, Sharon Hollander, and Daniel C. Isaksen. Hypercovers and simplicial presheaves. Mathematical Proceedings of the Cambridge Philosophical Society 136 (2004), 9–51. arXiv:math/0205027.

  17. Egbert Rijke, Michael Shulman, and Bas Spitters. Modalities in homotopy type theory. Logical Methods in Computer Science 16 (2020), 2:2. arXiv:1706.07526.

  18. Jacob Lurie. Kerodon, §3.5. Truncations and Postnikov towers.

  19. George Janelidze and G. Max Kelly. Galois theory and a general notion of central extension. Journal of Pure and Applied Algebra 97 (1994), 135–161. DOI 10.1016/0022-4049(94)90057-4.

Complete flattened LaTeX source
\documentclass[11pt,twoside,openany]{book}
\usepackage[T1]{fontenc}

\usepackage{lmodern}

\usepackage{charter}

\usepackage{amsmath,amssymb,amsthm,mathtools}

\usepackage{booktabs,array,tabularx,longtable}

\usepackage{enumitem,microtype}

\usepackage{tikz-cd}

\usepackage{graphicx,xcolor,titlesec,fancyhdr}

\usepackage[paperwidth=6.75in,paperheight=9.7in,inner=0.83in,outer=0.78in,
  top=0.78in,bottom=0.82in,headsep=0.22in,footskip=0.40in,headheight=15pt]{geometry}

\usepackage{makeidx}

\usepackage[unicode,colorlinks=true]{hyperref}

\usepackage{bookmark}

\usepackage{fancyvrb}

\definecolor{inkblue}{HTML}{203654}
\definecolor{sea}{HTML}{266B70}
\definecolor{softtext}{HTML}{425366}
\hypersetup{pdftitle={Galois Theory in Higher Categories},
  pdfsubject={Descent, monodromy, and higher topoi},
  pdfkeywords={higher Galois theory, infinity-topos, descent, monodromy, exodromy},
  linkcolor=inkblue,citecolor=sea,urlcolor=sea,bookmarksnumbered=true,
  pdfdisplaydoctitle=true}
\urlstyle{same}
\setlist{itemsep=2pt,topsep=4pt}
\setcounter{tocdepth}{1}
\setcounter{secnumdepth}{2}
\setlength{\parindent}{1.15em}
\setlength{\emergencystretch}{2em}
\raggedbottom
\allowdisplaybreaks[2]
\makeindex
\renewcommand{\indexname}{Subject index}
\makeatletter
\renewcommand{\@pnumwidth}{2.05em}
\renewcommand{\@tocrmarg}{3em}
\makeatother
\titleformat{\chapter}[display]
  {\normalfont\color{inkblue}\bfseries}
  {\sffamily\footnotesize\MakeUppercase{\chaptertitlename}\hspace{0.8em}\thechapter}
  {0.9ex}{\huge\raggedright}[\vspace{0.8ex}{\color{sea}\titlerule}]
\titlespacing*{\chapter}{0pt}{8pt}{21pt}
\titleformat{\section}{\normalfont\large\bfseries\color{inkblue}}{\thesection}{0.65em}{}
\titleformat{\subsection}{\normalfont\normalsize\bfseries\color{inkblue}}{\thesubsection}{0.65em}{}
\pagestyle{fancy}
\fancyhf{}
\fancyhead[LE,RO]{\sffamily\small\thepage}
\fancyhead[RE]{\sffamily\small\color{softtext}\nouppercase{\leftmark}}
\fancyhead[LO]{\sffamily\small\color{softtext}\nouppercase{\rightmark}}
\renewcommand{\headrulewidth}{0.25pt}
\renewcommand{\chaptermark}[1]{\markboth{\thechapter\enspace #1}{}}
\renewcommand{\sectionmark}[1]{\markright{\thesection\enspace #1}}
\fancypagestyle{plain}{\fancyhf{}\fancyfoot[C]{\sffamily\small\thepage}\renewcommand{\headrulewidth}{0pt}}
\newtheorem{theorem}{Theorem}[chapter]
\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}
\newcommand{\SSet}{\mathcal{S}}
\newcommand{\CatInf}{\mathrm{Cat}_{\infty}}
\newcommand{\op}{\mathrm{op}}
\newcommand{\X}{\mathcal{X}}
\DeclareRobustCommand{\Cech}{\ifmmode\text{\v{C}ech}\else\v{C}ech\fi}
\DeclareMathOperator{\Map}{Map}
\DeclareMathOperator{\Fun}{Fun}
\DeclareMathOperator{\id}{id}
\DeclareMathOperator*{\colim}{colim}
\DeclareMathOperator*{\Tot}{Tot}
\DeclareMathOperator{\Eq}{Eq}
\DeclareMathOperator{\Pro}{Pro}
\DeclareMathOperator{\Aut}{Aut}
\pdfstringdefDisableCommands{\def\({}\def\){}\def\ensuremath#1{#1}\def\infty{infinity}\def\Cech{Cech}}
\begin{document}
\frontmatter
\begin{titlepage}
\thispagestyle{empty}
\vspace*{0.48in}
{\color{sea}\rule{0.8in}{1.3pt}}\par
\vspace{0.48in}
{\sffamily\small\bfseries\color{sea}RESEARCH MONOGRAPH\par}
\vspace{0.27in}
{\fontsize{38}{42}\selectfont\bfseries\color{inkblue}
 Galois Theory\\[0.03in]in Higher Categories\par}
\vspace{0.30in}
{\fontsize{17}{22}\selectfont\color{softtext}
 Descent, monodromy, and higher topoi\par}
\vfill
{\large\color{inkblue}Manuscript for \textit{Higher Categorical Semantics}\par}
\vspace{0.10in}
{\small\color{softtext}November 2026\par}
\vspace{0.38in}
\end{titlepage}

\chapter*{Preface}
\markboth{Preface}{}
\addcontentsline{toc}{chapter}{Preface}

The ordinary expression \(L\otimes_K L\) in Galois theory serves two purposes.
It describes the overlap of the extension with itself, and, when splitting
holds, it encodes the symmetries by which copies of a descended object are
identified. The higher analogue is the entire simplicial object of overlaps.
Its realization recovers the base under effective descent. Its Segal
operators describe multiplication in a groupoid, internal composition in a
category, or a monad on a category of objects over a splitting base. These
three conclusions require different hypotheses. Much of the work below is
to identify them and prove the comparisons between their outputs.

The point of departure is Francis Borceux and George Janelidze's
\textit{Galois Theories}~\cite{bj}. Their sequence of generalizations passes
from subgroup correspondences to split algebras, profinite symmetry,
Pierce spectra, relative reflections, coverings, and generalized descent.
This monograph follows those questions through the passage from ordinary
categories to \((\infty,1)\)-categories. The organizing results are
homotopy-coherent descent in an \(\infty\)-topos and the classification of
families by transport. They are joined by versions in stable and directed
settings where the relevant additional structure survives.

\section*{Reading the book}

Part I develops the correspondence between families and actions and proves
the restriction of effective descent to chosen coefficients. Part II treats
continuity of monodromy, the shape of a higher topos, and obstructions to
lifting actions through truncations. Part III compares Pierce descent,
connective spectral \'{e}tale descent, and Galois extensions in stable
homotopy theory. Part IV studies admissible reflections in slices,
factorization, and centrality. Part V turns from invertible transport to
right and left fibrations, directed descent, and exodromy. Part VI returns
to toposes themselves and compares internal, geometric, and external
presentation theorems.

The correspondence with the original book is thematic. Its Chapters 1--2
are treated in Part I, its Chapter 3 in Part II, and its Chapter 4 and
appendix in Part III. Its Chapter 5 chiefly informs Part IV. Its Chapter 6
is distributed over Parts I, II, and IV, where the notions of covering,
connectedness, and local triviality can be stated independently. Its
Chapter 7 motivates Parts V and VI. The exposition assumes familiarity
with adjunctions, slices, and elementary homotopy theory. It gives
universe and variance conventions before using any classification
theorem. Deep inputs receive references at their points of use.

\section*{Abstract}

We develop higher analogues of the categorical Galois theories of Borceux
and Janelidze. Effective descent in an \(\infty\)-topos yields actions of
\v{C}ech groupoids and classifies torsors by delooping. Pro-spaces encode
continuous monodromy under coefficient hypotheses, while Postnikov
extensions give obstructions to lifting actions. We compare Pierce and
spectral descent, admissible reflections and relative factorization,
directed transport and exodromy, and groupoid presentations of topoi. The
ordinary and higher statements are connected by specified truncations,
duality, and base-change comparisons.

\section*{Manuscript disclosure}

This manuscript is an independent mathematical treatment with the source
book explicitly cited. Generative AI assisted with drafting, bibliography
research, mathematical consistency checks, and typesetting. The journal
submission metadata provides the responsible author or research group.

\tableofcontents
\mainmatter
\part{Torsors and descent}
\chapter{The language of coherent descent}\label{ch:language}
\index{descent!coherent}\index{universes}

In its simplest form a Galois correspondence compares subobjects of a
given extension with subgroups of its symmetries. A stronger statement
recovers the category of objects that split after a base change from
trivial objects on that base and their gluing data. When gluing takes
place in an \(\infty\)-category, a gluing map comes with compatibilities
among homotopies and coherences among those compatibilities. A
simplicial overlap diagram organizes this information without requiring
a separate definition at each level.

\section{Conventions and the ambient category}

Fix a universe of small spaces and write \(\SSet\) for their
\(\infty\)-category. Functor categories, categories of large coefficients,
and pro-categories are formed in larger universes when needed. A space
means a homotopy type. An \(\infty\)-category means an
\((\infty,1)\)-category, so the morphisms between two given objects form
a space \(\Map(-,-)\). For a topological space \(T\), we distinguish its
topology, its singular homotopy type, and the \(\infty\)-topos of sheaves
on its open-set site. These inputs agree on certain classifications under
the local hypotheses given below.

Limits, colimits, slices, and fibers are understood in the higher
categorical sense. The space
\(\Map_{\CatInf}(\mathcal C,\mathcal D)=
\Fun(\mathcal C,\mathcal D)^{\simeq}\) is the maximal subgroupoid of
the functor category. The functor category itself uses all natural
transformations, an important distinction when we study geometric
morphisms in Chapter~\ref{ch:geometric}.

\begin{definition}
An \(\infty\)-topos is an accessible left exact localization of a
presheaf \(\infty\)-category \(\Fun(C^{\op},\SSet)\) for a small
\(\infty\)-category \(C\). A geometric morphism
\(f:\mathcal Y\to\mathcal X\) is specified by its inverse-image
functor \(f^*:\mathcal X\to\mathcal Y\), which preserves colimits
and finite limits. Its right adjoint is denoted \(f_*\).
\end{definition}

A geometric morphism points in the opposite direction from its
inverse image. For a map \(p:U\to B\) inside \(\X\), the same
notation \(p^*\) means base change from \(\X_{/B}\) to
\(\X_{/U}\). The left adjoint \(p_!\) composes a map with \(p\).
The right adjoint \(p_*\) is dependent product. Its existence
follows from local cartesian closure. If \(p\) is a map of spaces
and \(V\to U\) is a family, the fiber of \(p_!V\) over \(b\) is
the total space over the homotopy fiber \(U_b\). The fiber of
\(p_*V\) over \(b\) is the space of sections over \(U_b\).

\begin{lemma}[Mapping from a section]\label{lem:grammar-sections}
Given \(q:E\to B\) and \(a:A\to B\) in \(\X\), one has
\[
\Map_{\X_{/B}}(A,E)
\simeq
\operatorname{fib}_{a}\bigl(\Map_{\X}(A,E)
                          \xrightarrow{q\circ-}\Map_{\X}(A,B)\bigr).
\]
This is also the space of sections of \(a^*E\to A\).
\end{lemma}

\begin{proof}
A point of the fiber consists of a map to \(E\) and a homotopy
identifying its composite with \(a\). These are the data in the
slice mapping space, with the same assertion for families
parametrized by a test space. The pullback universal property
rewrites that homotopy fiber as maps from the terminal
object of \(\X_{/A}\) to \(a^*E\).
\end{proof}

If \(F\) is a fixed fiber, its space of automorphisms \(\Aut(F)\)
consists of the equivalence components of \(\Map(F,F)\).
Composition makes it a group object in spaces. Its group of
components \(\pi_0\Aut(F)\) classifies automorphisms up to homotopy.
A family with fiber \(F\) can depend on homotopies within a single
component of \(\Aut(F)\), even when the group of components of
the automorphism object acts trivially.

\section{Cartesianness and realizations}\label{sec:grammar-van-kampen}

A cartesian family over a diagram \(U:I\to\X\) is a natural
transformation \(V\to U\) whose square over every arrow is a
pullback. Let \(B=\colim_I U\). Pullback defines a comparison
\begin{equation}\label{eq:grammar-slices}
\X_{/B}\longrightarrow\lim_{i\in I^{\op}}\X_{/U_i}.
\end{equation}
The target consists of cartesian families. The identifications
over arrows and their higher coherences are part of the data.
A colimit for which this comparison is an equivalence is called
van Kampen.
\index{colimit!van Kampen}

\begin{proposition}[A test for descent of a diagram]\label{prop:grammar-descent}
Suppose the ambient category admits pullbacks and the required
colimits, and these colimits are preserved by base change. The
comparison \eqref{eq:grammar-slices} is fully faithful. It is an
equivalence exactly when for each cartesian \(V\to U\), the squares
\[
\begin{tikzcd}[column sep=large]
V_i \arrow[r] \arrow[d] & \colim_j V_j \arrow[d] \\
U_i \arrow[r] & B
\end{tikzcd}
\]
are pullbacks.
\end{proposition}

\begin{proof}
Realization takes \(V\) to \(\colim_I V\to B\). For any \(W\to B\),
maps from the realization to \(W\) are the limit of maps from
\(V_i\to B\) to \(W\), by the universal property of the colimit in
the slice. Base change rewrites the latter mapping spaces as maps
from \(V_i\to U_i\) to \(W\times_B U_i\). Their limit is the space
of maps of cartesian families into the comparison image of \(W\).
Realization is therefore left adjoint to the comparison.
The counit \(\colim_I(W\times_B U_i)\to W\) is an equivalence by
the base-change assumption, so the right adjoint is fully faithful.
For each \(V\), the unit is an equivalence exactly when its components
\(V_i\to U_i\times_B\colim_j V_j\) are equivalences.
\end{proof}

A candidate over the base is determined by its restrictions as
soon as the comparison is fully faithful. Existence of a candidate
for every cartesian family requires the unit condition. Each
covering theorem below uses a specific argument for existence.

\begin{theorem}[Van Kampen property of higher topoi]\label{thm:grammar-vk}
Small colimits in an \(\infty\)-topos are van Kampen.
\end{theorem}

\begin{proof}
In spaces, straightening over a space \(K\) gives
\(\SSet_{/K}\simeq\Fun(K,\SSet)\), covariantly for a family
viewed by transport. The inclusion of spaces into
\(\CatInf\) preserves colimits because it is left adjoint to
the maximal subgroupoid functor. The internal-hom universal
property turns colimits of bases into limits of the functor
categories. In a presheaf \(\infty\)-category, the two criteria
in Proposition~\ref{prop:grammar-descent} then hold objectwise.

Write \(L:\mathcal P\rightleftarrows\X:i\) for a left exact
localization. A diagram of local objects and its cartesian
families can be viewed in \(\mathcal P\), since the inclusion
preserves limits. Realizing there gives pullback squares as in
Proposition~\ref{prop:grammar-descent}. Applying \(L\) preserves
those pullbacks and gives their realizations in \(\X\).
Universality of colimits in \(\X\) follows by the same localization
argument. If \(B_P\) is the colimit of the diagram in \(\mathcal P\),
pull back that colimit along \(iW\times_{iB}B_P\to B_P\) and apply
\(L\). This identifies \(W\) as the colimit of the diagram base
changed along \(W\to B\). The two criteria now apply in \(\X\).
The straightening input is developed in \cite[Ch.~2]{htt}.
\end{proof}

\section{The overlap diagram}

The \(\Cech\) nerve of \(p:U\to B\) has \(k\)-simplices
\[
U_k=\underbrace{U\times_B\cdots\times_B U}_{k+1}.
\]
Its operators omit or repeat factors. It is a groupoid object because
the edges encode pairs of local lifts and the inverse operation
reverses a pair. The map \(p\) is an effective epimorphism if
the augmentation \(U_\bullet\to B\) exhibits \(B\) as its
realization. In spaces this amounts to surjectivity on components.
In a topos of sheaves it expresses local existence, so a section
over the full base is unnecessary.
\index{effective epimorphism}\index{Cech nerve@\v{C}ech nerve}

\begin{corollary}[Effective descent over one cover]\label{cor:grammar-cover}
For an effective epimorphism \(p:U\to B\) in \(\X\),
the natural comparison is an equivalence
\[
\X_{/B}\simeq\Tot\bigl([k]\longmapsto \X_{/U_k}\bigr).
\]
Moreover \(p^*\) is conservative.
\end{corollary}

\begin{proof}
Apply Theorem~\ref{thm:grammar-vk} to the \(\Cech\) nerve.
For a morphism in the limit whose degree-zero component is an
equivalence, each degree-\(k\) component is an equivalence by
restriction along a vertex map \([0]\to[k]\).
Equivalences in the limit are detected componentwise.
\end{proof}

A stable category or a category of algebras can require a different
proof. Theorem~\ref{thm:grammar-vk} concerns slices of an
\(\infty\)-topos. A comparison for modules over a spectral ring
computes a totalization of module categories using scalar extension.
Its conservativity and continuity require hypotheses on that
extension. Chapter~\ref{ch:spectral} gives such hypotheses.

\section{Restricting coefficients}

Let \(D^\bullet\) be a cosimplicial diagram of
\(\infty\)-categories. Let \(E^k\) be replete full subcategories of
\(D^k\), preserved by all structure functors.

\begin{lemma}[Restricting full coefficients]\label{lem:grammar-restrict}
The functor \(\Tot E^\bullet\to\Tot D^\bullet\) is fully faithful.
An object of \(\Tot D^\bullet\) lies in its essential image exactly
when its degree-zero component lies in \(E^0\).
\end{lemma}

\begin{proof}
For any pair of compatible families, the mapping space is the limit
of the corresponding levelwise mapping-space diagram. Fullness makes
the two diagrams equivalent in every degree. The essential image
consists of objects whose components lie in the replete full
subcategories. Each structure equivalence and coherence of the given
diagram belongs to the full subcategory as soon as its vertices do.
Applying a vertex map \([0]\to[k]\) to the degree-zero component
puts the degree-\(k\) component in \(E^k\).
\end{proof}

In applications one must establish the unrestricted comparison and
then check stability of the coefficients or construct a comparison
on which stability holds. In algebraic Galois theory the second
step can follow from a relative left exactness condition. In
stratified geometry the comparison must respect which strata are
connected by directed arrows. An equivalence of cores classifies
objects and their equivalences. We use a comparison of full
categories when morphisms between the coefficients also matter.

\chapter[Torsors and monodromy]{Torsors, monodromy, and intermediate objects}
\label{ch:torsors}

The first two chapters of Borceux and Janelidze pass from intermediate fields to a category of split algebras. The Galois group acts on a set of embeddings. Inside the resulting action category, the quotients of the regular action recover the intermediate fields, with arrows reversed by the passage from algebras to actions \cite[Chapters 1--2]{bj}. The argument suggests two distinct operations. One classifies objects trivialized by a cover. The other identifies which pointed intermediate objects are governed by subgroups. In an \(\infty\)-category these operations have different answers unless a truncation condition is imposed.

\section{Pointed connected objects}
\index{group object}\index{delooping}\index{loop object}

Write \(\mathcal S\) for the \(\infty\)-category of spaces and \(\mathcal X\) for an \(\infty\)-topos. Pullbacks, colimits, and fibers are taken in the \(\infty\)-categorical sense. We write \(1\) for a terminal object and call a map \(1\to B\) a point of \(B\) internal to \(\mathcal X\). This means a section of an object in the topos. A geometric morphism \(\mathcal S\to\mathcal X\) is a different operation.

A simplicial object \(D_\bullet\) is a groupoid object if its Segal maps are equivalences and the arrows are invertible in the internal sense. For example, the \v{C}ech nerve of a map \(q:U\to B\) is the groupoid object with
\[
   \check C(q)_n=\underbrace{U\times_B\cdots\times_B U}_{n+1\ \mathrm{factors}}.
\]
Its face maps delete vertices and its degeneracies repeat vertices. The map \(q\) is an effective epimorphism if the augmentation \(|\check C(q)_\bullet|\to B\) is an equivalence. Effectivity of groupoids in an \(\infty\)-topos says that for every groupoid object \(D_\bullet\), the canonical map \(D_0\to |D_\bullet|\) is an effective epimorphism and its \v{C}ech nerve identifies with \(D_\bullet\) \cite[Section 6.1.3]{htt}.

A group object is a groupoid object \(G_\bullet\) with terminal object of objects. Its arrow object is denoted by \(G=G_1\). In degree \(n\) the Segal maps identify \(G_n\) with \(G^n\). The two ways of composing three arrows agree through the simplicial structure, which also supplies all higher coherences. The delooping is \(BG=|G_\bullet|\), pointed by the degree-zero inclusion. A pointed object \(b:1\to B\) is connected if \(b\) is an effective epimorphism. In this situation,
\[
   \Omega_b B=1\times_B 1
\]
is the based loop group object, with simplicial description \(\check C(b)_\bullet\). The definition of connectedness here is equivalent to \(\tau_{\leq 0}B\simeq 1\). In fact a subobject of \(B\) is classified by a map to the subobject classifier, which is a \(0\)-truncated object. If \(\tau_{\leq 0}B\simeq 1\), any subobject containing \(b\) is therefore all of \(B\), so the image of \(b\) is all of \(B\). Conversely, if \(b\) is an effective epimorphism, a map from \(B\) to a \(0\)-truncated object is determined by its restriction along \(b\). Every map from \(1\) to that object extends along \(B\to 1\). The resulting universal property identifies \(\tau_{\leq 0}B\) with \(1\).

\begin{proposition}[Loop and delooping]
\label{prop:loop-delooping}
Taking loops and taking delooping define inverse equivalences between the \(\infty\)-category of group objects in \(\mathcal X\) and the \(\infty\)-category of pointed connected objects of \(\mathcal X\).
\end{proposition}

\begin{proof}
A map of group objects induces a pointed map on geometric realizations. A pointed map induces a map on \v{C}ech nerves of its points. For a pointed connected object the augmentation of its \v{C}ech nerve realizes to the object. For a group object its simplicial groupoid is the \v{C}ech nerve of the point of its realization by effectivity. Both identifications are canonical maps of diagrams. They give unit and counit equivalences of the two functors, including on mapping spaces.
\end{proof}

These operations also work in every slice \(\mathcal X_{/B}\). The terminal object in that slice is the identity of \(B\). A point is a section, and the corresponding notion of connectedness says that the section is an effective epimorphism. Slicing is necessary when a group varies with parameters. A single external group of global sections need not recover an internal group object. The family of mapping spaces \(\operatorname{Map}_{\mathcal X}(T,G)\), with its action by change of \(T\), does recover it by the \(\infty\)-categorical Yoneda lemma.

\section{Actions and classifying maps}
\index{action!homotopy-coherent}\index{torsor!classification}

A left \(G\)-object can be presented as a simplicial object \(E_\bullet\) over \(G_\bullet\) such that restriction to the initial vertex induces
\[
    E_n\xrightarrow{\ \sim\ }E_0\times G^n .
\]
The map to the final vertex at \(n=1\) supplies the action. With this orientation, successive arrows \(g,h\) act by \(h\cdot(g\cdot x)=(hg)\cdot x\). Here and below formulas in elements specify maps of objects and their coherent simplicial identities. The category of actions is the full subcategory of the category of simplicial objects over \(G_\bullet\) specified by the displayed pullback property. In particular its morphisms are coherent equivariant morphisms. The quotient \(E_0//G\) means \(|E_\bullet|\).

\begin{theorem}[Actions as objects over a delooping]
\label{thm:actions-delooping}
The functor sending \(E_\bullet\to G_\bullet\) to \(|E_\bullet|\to BG\) is an equivalence
\[
       \operatorname{Act}_{G}(\mathcal X)\simeq \mathcal X_{/BG}.
\]
Its inverse sends \(Y\to BG\) to the action diagram \(G_\bullet\times_{BG}Y\).
\end{theorem}

\begin{proof}
For a groupoid, cartesianness over the initial vertex implies cartesianness over every vertex. In degree one, transport to the other vertex has inverse induced by inversion. In higher degrees the Segal maps express transport along a string of arrows as successive transport in degree one. The resulting squares are cartesian for all simplex operators. The Segal pullbacks for \(G_\bullet\), together with the cartesian vertex squares, then give the Segal pullbacks for \(E_\bullet\). Invertibility is inherited from the base groupoid by transport. Thus \(E_\bullet\) is a groupoid object. Put \(Y=|E_\bullet|\). Effectivity gives \(E_1\simeq E_0\times_Y E_0\) and \(G_1\simeq 1\times_{BG}1\). The square
\[
\begin{array}{ccc}
 E_0&\longrightarrow&1\\
 \downarrow&&\downarrow\\
 Y&\longrightarrow&BG
\end{array}
\]
is cartesian. To see this, test the comparison \(E_0\to Y\times_{BG}1\) after pullback along the effective epimorphism \(E_0\to Y\). The resulting map is
\[
 E_0\times_Y E_0\longrightarrow E_0\times_{BG}1.
\]
Under the identifications above it is exactly the degree-one vertex equivalence \(E_1\simeq E_0\times G_1\), using the first projection to \(E_0\). Equivalences can be tested after an effective epimorphism, as follows by reconstructing both source and target by the base changes of its \v{C}ech nerve and realizing. The vertex equivalence now gives \(E_\bullet\simeq G_\bullet\times_{BG}Y\) in all degrees. In the other direction, universality of colimits implies \(|G_\bullet\times_{BG}Y|\simeq Y\). All comparisons are natural in the diagrams, so they give inverse equivalences of \(\infty\)-categories.
\end{proof}

A \(G\)-torsor over \(T\) is an action in \(\mathcal X_{/T}\) on an object \(P\to T\) such that \(P\to T\) is an effective epimorphism and the transport map
\[
       P\times G\longrightarrow P\times_T P,
       \qquad (x,g)\longmapsto(x,g\cdot x)
\]
is an equivalence over \(T\). The universal torsor has underlying object \(1\to BG\). Its action is specified by the identification of its \v{C}ech groupoid with the bar groupoid of \(G\). Conventions using right actions give the same classifier via inversion.

\begin{corollary}[Torsor classification]
\label{cor:torsor-classification}
The space of \(G\)-torsors over \(T\), with equivariant equivalences as arrows, is naturally equivalent to \(\operatorname{Map}_{\mathcal X}(T,BG)\). Pullback in \(T\) corresponds to precomposition with the classifying map.
\end{corollary}

\begin{proof}
The classifying equivalence of Theorem~\ref{thm:actions-delooping} in \(\mathcal X_{/T}\) sends an action on \(P\) to an object \(P//G\to T\times BG\) over \(T\). The comparison from its action groupoid to the \v{C}ech nerve of \(P\to T\) is an equivalence in degree one by the torsor condition. The two Segal properties make it an equivalence in every degree. Since \(P\to T\) is effective, this means \(P//G\simeq T\) over \(T\). Conversely, if \(P//G\simeq T\), effectivity identifies the action groupoid with the \v{C}ech nerve of \(P\to T\), giving both torsor conditions. The core of objects \(Y\to T\times BG\) over \(T\) with \(Y\to T\) an equivalence is \(\operatorname{Map}_{\mathcal X}(T,BG)\), by replacing the source by \(T\). This identification and the resulting pullback of the universal torsor commute with base change.
\end{proof}

This formulation retains automorphisms of a torsor and their higher homotopies. At a classifying map \(c:T\to BG\), the loop space of the mapping space in Corollary~\ref{cor:torsor-classification} is the space of equivariant automorphisms of the classified torsor. Choosing a section of the torsor identifies it with the trivial torsor through the transport equivalence. Changing the section can change that identification. For an ordinary nonabelian group the induced change is conjugation. For an internal abelian group \(A\) with Eilenberg--Mac Lane deloopings, the same corollary applied to \(G=K(A,m)\) identifies torsors with maps to \(K(A,m+1)\). With the convention \(H^i(\mathcal X,A)=\pi_0\operatorname{Map}_{\mathcal X}(1,K(A,i))\), global isomorphism classes form \(H^{m+1}(\mathcal X,A)\). The based homotopy groups of the torsor space at the trivial class recover lower degrees, because looping mapping spaces commutes with looping the target.

\section{Transport over a space}
\index{monodromy}\index{homotopy orbits}\index{homotopy fixed points}

For a space \(B\) there is a canonical equivalence
\begin{equation}
\label{eq:transport-space}
       \mathcal S_{/B}\simeq \operatorname{Fun}(B,\mathcal S).
\end{equation}
The functor associated to \(Y\to B\) sends \(b\) to \(Y_b=\{b\}\times_B Y\). Its maps are transport along paths, and homotopies of paths act through the higher coherences of transport. One construction uses straightening of left fibrations \cite[Section 2.2.1]{htt}. Over a groupoidal base, the Grothendieck construction of a space-valued functor has only invertible arrows. Its total space is the homotopy sum \(\operatorname{colim}_{b\in B}F(b)\) over \(\operatorname{colim}_{b\in B}*\simeq B\). In the other direction an arbitrary map of spaces can be represented by a Kan fibration, which in particular is a left fibration. Thus the straightening correspondence restricts to the displayed equivalence.

For connected \(B\) the point \(b\) is an effective epimorphism in \(\mathcal S\), and the space \(B\) is the delooping of \(\Omega_bB\). Equation~\eqref{eq:transport-space} and Theorem~\ref{thm:actions-delooping} identify evaluation at \(b\) with the forgetful functor from \(\Omega_bB\)-actions. The point \(b: *\to B\) has fiber \(\Omega_bB\). Under the equivalence it is the regular action, also characterized as the free action on the point. Indeed maps \(*\to Y\) over \(B\) are points of \(Y_b\), and the Yoneda lemma in \(\operatorname{Fun}(B,\mathcal S)\) characterizes the free action by the representable functor \(\operatorname{Map}_B(b,-)\).

Pullback along a map \(f:C\to B\) corresponds to restriction of representations. Both adjoints to restriction exist and are computed by Kan extension. At \(b\), left Kan extension of \(F:C\to\mathcal S\) takes the homotopy sum of \(F\) over the groupoid of pairs \((c,f(c)\to b)\). Right Kan extension takes the limit over pairs \((c,b\to f(c))\). Inversion identifies the two indexing groupoids. The left adjoint is composition of total-space maps over the base, since giving a map from that composite is equivalent to giving a map to the pullback by the universal property of a fiber product. In the right adjoint the fiber over \(b\) is the space of sections of the restricted family over the fiber of \(f\). In particular the two adjoints along \(B\to *\) are homotopy orbits and homotopy fixed points.

If a connected space \(B\) has a fiber \(F\) at \(b\), then after choosing an identification with a specified space \(K\), the action can be expressed by a based map
\[
        B\longrightarrow B\operatorname{Aut}(K).
\]
Here \(\operatorname{Aut}(K)\) is the space of equivalences \(K\to K\) with composition. To verify the classification, take the core \(\mathcal S^\simeq\) in a universe containing the given fibers, choose the component containing \(K\), and use Proposition~\ref{prop:loop-delooping} to identify the component with \(B\operatorname{Aut}(K)\). Every functor from a space lands in the core, since all its arrows are invertible, and connectedness forces the selected component. Taking the homotopy fiber of evaluation at \(b\) gives the based mapping space and records the chosen identification. Forgetting the identification gives the unpointed mapping space. Taking \(\pi_0\) of \(\operatorname{Aut}(K)\) loses the higher transition data. For instance, let \(K=S^1\). The identity component of \(\operatorname{Aut}(S^1)\) is equivalent to \(S^1\), as follows by evaluating a degree-one map at a point and noting that the degree-one component of the based mapping space is contractible. Consequently
\[
       \pi_2\bigl(B\operatorname{Aut}(S^1)\bigr)\cong \mathbb Z.
\]
There are based families with fiber \(S^1\) over \(S^2\) whose classifying maps represent a nonzero integer, even though \(\pi_1(S^2)\) is trivial.

\section{The subgroup correspondence and its higher replacement}
\index{subgroup correspondence}\index{covering!pointed}

Fix a pointed connected space \((B,b)\). A pointed intermediate object is a factorization
\[
                  *\xrightarrow{y}Y\xrightarrow{q}B
\]
of \(b\), including an identification of the composite with \(b\). Its morphisms are maps of such factorizations, with coherent triangles. By \eqref{eq:transport-space} the category of factorizations is the category of actions equipped with a map from the regular action, or equivalently the category of actions with a specified point of the fiber over \(b\). This statement is true whether or not \(Y\) is connected. It follows because taking the undercategory of a given object preserves equivalences of \(\infty\)-categories.

\begin{proposition}[Connected intermediates]
\label{prop:connected-intermediates}
The full subcategory of pointed intermediate objects with connected \(Y\) is equivalent to the slice of group objects in spaces over \(G=\Omega_b B\). Under the equivalence \(Y\to B\) corresponds to \(H=\Omega_yY\to G\). Its fiber over \(b\), with the distinguished point, is the homotopy quotient \(G//H\) of the right \(H\)-action on \(G\) through \(H\to G\).
\end{proposition}

\begin{proof}
Slicing the equivalence of Proposition~\ref{prop:loop-delooping} gives the categorical claim. For the fiber, model the map as \(BH\to BG\). Pull back the family \(*\to BG\) to \(BH\). Its fiber at the point of \(BH\) is \(G\). Choose the path coordinate of a point in this family to run from the image of the moving point of \(BH\) to the fixed point of \(BG\). If \(\phi:H\to G\) is the homomorphism, transport in this coordinate along \(h\) sends \(g\) to \(g\phi(h)^{-1}\). This is the covariant presentation of the right action \(g\cdot h=g\phi(h)\). The total space is its homotopy quotient \(G//H\), and the chosen point comes from the identity in \(G\). Equivalently, this quotient is the realization obtained by pulling the augmented bar diagram of \(H\) back along \(*\to BG\), so its description fixes the same map and basepoint without choosing path coordinates.
\end{proof}

The connectedness condition on the total space is the homotopy-theoretic transitivity condition. In a pointed factorization, the map from the regular action to the fiber is the orbit map induced by the distinguished point. It is an effective epimorphism exactly when \(Y\) is connected. In one direction the orbit map is the base change of \(*\to Y\) along \(Y_b\to Y\), and in the other direction effective epimorphisms descend along the effective epimorphism \(Y_b\to Y\). The latter map is a base change of \(*\to B\). To justify the descent assertion, factor a map into its effective epimorphism and monomorphism. The subobject given by the second factor is terminal after base change precisely if it was terminal before base change, since base change along an effective epimorphism detects equivalences. For spaces this criterion says that the regular orbit is surjective on components and places no discreteness condition on the fiber.

\begin{theorem}[Subgroups at the covering level]
\label{thm:subgroup-cover}
The category of pointed factorizations over \((B,b)\) with \(Y\) connected and \(q:Y\to B\) \(0\)-truncated is equivalent to the poset of subgroups of \(\pi_1(B,b)\), ordered by inclusion.
\end{theorem}

\begin{proof}
Under \eqref{eq:transport-space}, the condition on \(q\) says every value of the functor is discrete. Such functors factor through the ordinary fundamental groupoid of \(B\). This follows from the universal property of the homotopy category, since the category of sets is ordinary and a functor on a groupoid takes its arrows to bijections. Choosing the object \(b\) of the connected fundamental groupoid gives a \(\pi_1(B,b)\)-action on the fiber. The distinguished point \(x\) and connectedness make the action transitive, by the orbit criterion above. The stabilizer \(K\) of \(x\) is a subgroup, and the pointed action is isomorphic to the left action on \(\pi_1(B,b)/K\) via \(gK\mapsto g\cdot x\). A pointed equivariant map exists from the quotient by \(K\) to the quotient by \(L\) precisely when \(K\subseteq L\), and is then uniquely determined by the displayed formula. Mapping spaces between the pointed actions are discrete because maps between discrete fibers form discrete spaces and naturality is imposed by a limit of such spaces. The result is the stated equivalence of categories.
\end{proof}

This is the regular-quotient argument of \cite[Proposition 2.4.1]{bj} at the covering level. The opposite category reverses its containment order, as for intermediate fields. Subgroups up to conjugacy classify connected covers up to equivalence when the point in each intermediate cover is forgotten. For example, the equivariant endomorphisms of the transitive set \(\pi_1(B,b)/K\) correspond to the cosets fixed under \(K\). The group of invertible endomorphisms is \(N_{\pi_1(B,b)}(K)/K\), up to inversion in the convention for right translation.

The loop homomorphism for a connected covering is the inclusion of a union of components of \(G\). Indeed its fiberwise description as the stabilizer is the pullback of the orbit map \(G\to Y_b\) along the point \(x\), and \(Y_b\) is discrete. For a general intermediate object the same pullback is a homotopy fiber with a chosen nullhomotopy of motion of \(x\). Such a fiber need not be a union of components. Take \(B=*\) and \(Y=K(\mathbb Z,2)\). The map of loop groups is \(K(\mathbb Z,1)\to *\), whose only homotopy fiber is neither empty nor contractible. Hence it fails to be a monomorphism in spaces. Proposition~\ref{prop:connected-intermediates} accommodates this data by allowing arbitrary homomorphisms into \(G\).

\begin{corollary}[Relative truncation of the stabilizer]
For \(n\geq 0\), a map \(BH\to BG\) of pointed connected spaces is \(n\)-truncated if and only if \(H\to G\) is \((n-1)\)-truncated.
\end{corollary}

\begin{proof}
Set \(F=BH\times_{BG}*\), pointed by the chosen points. The homotopy fiber of \(H\to G\) over the identity is \(\Omega F\), since based loops preserve limits. Every fiber of \(H\to G\) over a point in its image on components is equivalent to the identity fiber by translation, and all other fibers are empty. The degree-one orbit argument also shows that \(G\to F\) meets every component of \(F\), and the left \(G\)-action on \(F\) gives equivalences carrying the distinguished point into each component. Thus \(F\) is \(n\)-truncated if and only if its loop space at that point is \((n-1)\)-truncated. The fibers of \(BH\to BG\) are all equivalent to \(F\) since \(BG\) is connected. Combining these observations proves the claim.
\end{proof}

For \(n=0\), this identifies the covering condition with monomorphisms of loop groups. For \(n=1\), the stabilizer map can have nontrivial discrete homotopy fibers, as for \(Y=B\mathbb Z\) over \(B=*\). A classification by subgroup inclusions already fails at this level. The full slice of group objects records how the stabilizer maps into the ambient group, including the coherent data through which it acts.

\section{Regular intermediates and delooped quotients}
\index{covering!regular}\index{quotient!delooped}

For a pointed connected cover classified by a subgroup \(K\subseteq\pi_1(B,b)\), the deck transformations act transitively on the fiber precisely when \(K\) is normal. Indeed the orbit of the distinguished coset under deck transformations is indexed by \(N_{\pi_1(B,b)}(K)/K\), so it is all of \(\pi_1(B,b)/K\) precisely when the normalizer is all of \(\pi_1(B,b)\). In that case the quotient is itself a group and the cover is the pullback of
\[
    *\longrightarrow B\bigl(\pi_1(B,b)/K\bigr)
\]
along the map \(B\to B(\pi_1(B,b)/K)\) induced by the quotient homomorphism. The path space construction gives left transport by the quotient group, and the right deck transformations commute with it.

At higher truncation levels the homotopy quotient \(G//H\) is always defined. It can have higher homotopy groups even when \(G\) is discrete. The existence of a delooped quotient is a separate requirement and can be expressed in terms of the pointed action already obtained from the factorization.

\begin{proposition}[Criterion for a delooped quotient]
\label{prop:delooped-quotient}
Let \(i:H\to G\) be a map of group objects in spaces and let \(Q\) be another group object. For a homomorphism \(\phi:G\to Q\), there is a pointed equivalence over \(BG\)
\[
         BH\simeq BG\times_{BQ}*
\]
if and only if \(G//H\), as a pointed left \(G\)-space, is equivalent to the regular left \(Q\)-space restricted along \(\phi\). In the pointed actions the specified points are those induced by the identities.
\end{proposition}

\begin{proof}
The object \(*\to BQ\) corresponds to the regular action by \eqref{eq:transport-space}. Pullback along \(B\phi\) restricts the action. Proposition~\ref{prop:connected-intermediates} computes the fiber of \(BH\to BG\) as \(G//H\). Transport from the fixed base along loops of \(BG\) acts on this fiber by left multiplication. This action commutes coherently with the right \(H\)-action used to form the quotient. Specifying a point in the fiber is the same as giving a pointed map from the regular action of \(G\). Hence \eqref{eq:transport-space}, applied also to the undercategory of the point, gives the stated equivalence of conditions.
\end{proof}

For a solution \(\phi\), the action on components of \(Q\) is transitive because \(BH\) is connected. Thus \(\pi_0 G\to \pi_0 Q\) is surjective. This condition is automatic for the ordinary quotient by a normal subgroup, but the full equivalence of pointed actions in Proposition~\ref{prop:delooped-quotient} probes every homotopy group of the fiber and its transport.

\chapter{Effective descent and \v{C}ech actions}
\label{ch:descent}

The split-algebra equivalence in \cite[Chapter 2]{bj} classifies objects by transport among their trivializations. Later chapters of that book make the descent morphism explicit and replace a single group by an internal groupoid. In a higher topos this construction starts with the entire \v{C}ech nerve of an effective epimorphism. Each composable string of changes of trivialization comes with its own coherence, and these coherences are part of the gluing datum.

\section{Descent data and their realization}
\index{descent!effective}\index{effective epimorphism}\index{Cech nerve@\v{C}ech nerve}

For a map \(q:U\to B\), put \(U_n=U^{\times_B(n+1)}\). The cosimplicial diagram of slices
\[
        [n]\longmapsto \mathcal X_{/U_n}
\]
acts on a simplex operator by pullback. Define \(\operatorname{Desc}(q)\) to be its limit in \(\infty\)-categories. This limit can be described as the category of simplicial objects \(V_\bullet\to U_\bullet\) such that for every map \([m]\to[n]\) in the simplex category the square
\[
\begin{array}{ccc}
 V_n&\longrightarrow&V_m\\
 \downarrow&&\downarrow\\
 U_n&\longrightarrow&U_m
\end{array}
\]
is cartesian. Indeed the straightening of a diagram taking an object \(A\) to \(\mathcal X_{/A}\) is the target projection of the arrow category with its cartesian morphisms the pullback squares. A limit of such a diagram classifies its cartesian sections. In this description the two vertex identifications in degree one give an equivalence of the two pullbacks of \(V_0\) to \(U_1\). The higher simplices specify coherent composition of those equivalences.

\begin{theorem}[Effective descent]
\label{thm:effective-descent}
For an effective epimorphism \(q:U\to B\) in an \(\infty\)-topos, the comparison
\[
       \mathcal X_{/B}\longrightarrow \operatorname{Desc}(q),
       \qquad E\longmapsto (E\times_B U_n)_{n\geq 0},
\]
is an equivalence. Its inverse is geometric realization over \(B\).
\end{theorem}

\begin{proof}
The realization of the displayed diagram for \(E\) is \(E\), by universality of colimits and \(|U_\bullet|\simeq B\). For the other composite let \(V_\bullet\) be a descent datum and put \(Y=|V_\bullet|\). First, \(V_\bullet\) is a groupoid object. The Segal squares can be checked using a vertex at the beginning of a string together with cartesian transport to each next vertex. A composable string of arrows from \(V_0\) then consists of its initial point and the underlying composable string in \(U_\bullet\). Transport along the inverse of a base arrow gives the inverse arrow of \(V_\bullet\), again by the cartesian vertex condition.

By effectivity \(V_0\to Y\) is an effective epimorphism and \(V_1\simeq V_0\times_Y V_0\). There is a canonical comparison \(V_0\to Y\times_B U\) over \(Y\). After pullback along \(V_0\to Y\), this comparison becomes
\[
       V_0\times_Y V_0\longrightarrow V_0\times_B U.
\]
The source is \(V_1\), and the target is \(V_0\times_U U_1\) where \(U_1\to U\) uses the initial vertex. The comparison is therefore the cartesian vertex equivalence of the datum. To descend that equivalence, pull it and both its domain and codomain to the entire \v{C}ech nerve of \(V_0\to Y\). The comparison is an equivalence in every such degree. Each original object is recovered by realizing its pullback diagram, by universality of colimits. Hence \(V_0\simeq Y\times_B U\). The vertex comparison in each degree identifies \(V_n\simeq Y\times_B U_n\), compatibly with all simplex operators. This gives the natural equivalences of the two composites.
\end{proof}

This gives the underlying mechanism of the action equivalence of Theorem~\ref{thm:actions-delooping}. When \(U=1\) and \(B=BG\), the \v{C}ech nerve is the bar groupoid of \(G\). At the other extreme, there may be no distinguished global point of \(U\) or \(B\). The internal groupoid then describes transitions between varying objects of objects. Replacing it by the loops at a chosen object would require extra hypotheses that are absent for most covers in a topos.

\begin{corollary}[What a general map sees]
\label{cor:descent-image}
For any \(q:U\to B\), let \(i:I\hookrightarrow B\) denote its image and \(p:U\to I\) the associated effective epimorphism. There is a canonical equivalence \(\operatorname{Desc}(q)\simeq\mathcal X_{/I}\), under which comparison from \(\mathcal X_{/B}\) is restriction along \(i\). Comparison is an equivalence if and only if \(q\) is an effective epimorphism.
\end{corollary}

\begin{proof}
The \\v{C}ech diagrams for \(p\) and \(q\) coincide because \(i\) is a monomorphism. More explicitly \(I\times_B I\simeq I\), so the comparison \(U\times_I U\to U\times_B U\) is an equivalence, and the same holds for every iterated pullback. Theorem~\ref{thm:effective-descent} applied to \(p\) gives the conclusion about restriction. If \(i^*\) is an equivalence, it reflects equivalences and carries \((I\to B)\to(B\to B)\) to an equivalence. Thus \(i\) is an equivalence.
\end{proof}

\section{Mapping spaces and properties local on the base}
\index{descent!mapping spaces}\index{truncated map}

Descent reconstructs morphisms as well as objects. For \(E,F\to B\) and \(q\) effective, it identifies
\begin{equation}
\label{eq:descent-mapping}
 \operatorname{Map}_{\mathcal X_{/B}}(E,F)
 \simeq
 \lim_{[n]\in\Delta}
 \operatorname{Map}_{\mathcal X_{/U_n}}
       (E\times_B U_n,F\times_B U_n).
\end{equation}
To check this directly, realize the augmented diagram \(E\times_B U_\bullet\to E\) in the slice over \(B\). Mapping into \(F\) turns its colimit into a limit of mapping spaces. The adjunction between composition and pullback along \(U_n\to B\) identifies each term of that limit with the corresponding term on the right of \eqref{eq:descent-mapping}. An element of the right-hand side specifies a local map together with compatibility homotopies over intersections and higher coherent comparisons on all iterated intersections.

Let \(\mathcal D(B)\subseteq \mathcal X_{/B}\) be a full subcategory closed under equivalences, for each \(B\), and assume pullback takes objects of \(\mathcal D\) to objects of \(\mathcal D\). Call membership local on the base if \(E\to B\) belongs to \(\mathcal D(B)\) whenever its pullback belongs to \(\mathcal D(U)\) along an effective epimorphism \(U\to B\).

\begin{proposition}[Descent for full subcategories]
\label{prop:relative-full-descent}
If membership in \(\mathcal D\) is local on the base, then for every effective epimorphism \(q:U\to B\),
\[
    \mathcal D(B)\simeq \lim_{[n]\in\Delta}\mathcal D(U_n).
\]
Conversely, if this comparison is an equivalence for all effective epimorphisms, membership is local on the base.
\end{proposition}

\begin{proof}
The proposed target is a full subcategory of \(\operatorname{Desc}(q)\), since limits preserve full inclusions of \(\infty\)-categories. A cartesian section lies in it exactly when its degree-zero object is in \(\mathcal D(U)\), since the other entries are equivalent to pullbacks along a vertex. Theorem~\ref{thm:effective-descent} identifies such sections with the objects over \(B\) whose pullback along \(q\) belongs to \(\mathcal D(U)\). This proves both directions.
\end{proof}

For example, the class of \(n\)-truncated maps is local on the base for each integer \(n\geq -2\). The case \(n=-2\) is the conservativity of pullback along an effective epimorphism. For the induction step, a map \(E\to B\) is \(n\)-truncated precisely if its diagonal \(E\to E\times_B E\) is \((n-1)\)-truncated. Forming the diagonal commutes with base change. The cover on its codomain after pullback along \(U\to B\) is effective, so the induction hypothesis applies. In particular this yields descent for monomorphisms and for covering objects with discrete fibers. Effective epimorphisms themselves are local, by applying the equivalence test to the image inclusion as in the orbit argument of Chapter~\ref{ch:torsors}. The locality proof uses finite-level properties and is valid in a nonhypercomplete topos.

\section{The descent monad}
\index{monad!descent}\index{Barr--Beck theorem}

Effective descent also has a monadic formulation. The functor \(q^*:\mathcal X_{/B}\to\mathcal X_{/U}\) has left adjoint \(q_!\) given by composition with \(q\). It preserves colimits by universality and detects equivalences by the \v{C}ech argument. The \(\infty\)-categorical Barr--Beck theorem then gives
\[
      \mathcal X_{/B}\simeq
      \operatorname{Alg}_{T}(\mathcal X_{/U}),
      \qquad T=q^*q_!.
\]
In the language of the Barr--Beck hypotheses, preserving all colimits implies preservation of realizations of those simplicial objects that become split after \(q^*\) \cite[Theorem 4.7.3.5]{ha}. This proves monadicity without assuming that the underlying morphism \(q\) has a section.

The monad can be expressed in terms of the two projections \(s,t:U_1\rightrightarrows U\), with \(s\) the source and \(t\) the target. For \(A\to U\),
\[
       T(A)\simeq t_!s^* A \simeq U\times_B A .
\]
In the last expression the first coordinate is the target in \(U\). The unit is induced by the diagonal of \(U\) over \(B\). The multiplication takes \((u_2,u_1,a)\) to \((u_2,a)\) with the prescribed composition of paths in \(B\). Its simplicial powers are indexed by strings in the \v{C}ech groupoid. For \(A=E\times_B U\), the monad action identifies the point \(e\) over the preceding vertex with the same point over the new vertex. Thus the map \(T(A)\to A\) is the transport correspondence from the canonical descent datum. The monadic comparison and the \v{C}ech comparison agree on every pullback from \(B\), and both comparisons are equivalences. This relates their coherent algebra and coherent transport descriptions without an extra hypothesis on \(q\).

As with the classical treatment in \cite[Section 5.1]{bj}, the monad lives on a slice over the splitting object. If a geometric problem begins with algebras and replaces them by opposite affine objects, geometric pullback usually becomes extension of scalars. Passing to opposites converts the preceding monadic expression into a comonadic one with the corresponding variance of the adjunction. In a general algebraic category, extension of scalars still must be proved to satisfy a descent theorem. The argument in this chapter uses groupoid effectivity and universal colimits in a specific ambient topos, which are the hypotheses making arbitrary effective epimorphisms work here.

\section{Refinements and changes of groupoid presentation}
\index{groupoid!change of presentation}

Every effective internal groupoid presents a base. Let \(D_\bullet\) be an internal groupoid, let \(B=|D_\bullet|\), and apply Theorem~\ref{thm:effective-descent} to \(D_0\to B\). This identifies cartesian objects over \(D_\bullet\) with \(\mathcal X_{/B}\). A map of groupoids induces a pullback functor between their cartesian action categories, and under this equivalence it is pullback along the map of realizations. In fact, pulling back \(E\times_{|D_\bullet|}D_\bullet\) along the map of groupoids gives its pullback from \(E\) degree by degree. Refinement of an effective cover is a particular case, in which the map of realizations is the identity. Restricting a datum to a refinement changes the representatives for its local objects and their transition maps while preserving its realized object.

The following criterion gives a specific test for changing presentations. For a map \(F_\bullet\to D_\bullet\) of internal groupoids, call it fully faithful if the comparison
\[
 F_1\longrightarrow (F_0\times F_0)\times_{D_0\times D_0}D_1
\]
is an equivalence. Call it essentially surjective if the map
\[
 F_0\times_{D_0,s}D_1\xrightarrow{\,t\,}D_0
\]
is an effective epimorphism.

\begin{proposition}[Change of presentation]
If a map of internal groupoids is fully faithful and essentially surjective, it induces an equivalence of realizations and an equivalence of cartesian action categories. Conversely, a map inducing an equivalence of realizations is fully faithful and essentially surjective.
\end{proposition}

\begin{proof}
Write \(A=|F_\bullet|\) and \(B=|D_\bullet|\). Effectivity gives
\[
 (F_0\times F_0)\times_{D_0\times D_0}D_1\simeq F_0\times_B F_0.
\]
Full faithfulness thus identifies the degree-one terms of the \v{C}ech nerves of \(F_0\to A\) and \(F_0\to B\), compatibly with both vertices. By the Segal identities their entire nerves agree. Their realizations identify \(A\) with the image of \(F_0\to B\). On the other hand \(F_0\times_{D_0,s}D_1\simeq F_0\times_B D_0\), so essential surjectivity is equivalent to \(F_0\to B\) being effective after base change along \(D_0\to B\). Effectivity descends along that cover. The image is therefore \(B\). If \(A\to B\) is an equivalence, the same formulas give full faithfulness, and essential surjectivity follows by base change from the effective map \(F_0\to A\). The result for actions follows from the descent comparisons.
\end{proof}

An equivalence here allows the objects of objects to change. In later applications one can replace a groupoid associated to a cover by a more convenient presentation as long as the change of realization, or the two conditions in the proposition, can be verified.

\section{Fixed split type and the torsor of frames}
\index{torsor!frames}\index{split object}

Fix \(K\in\mathcal X\). An object \(E\to B\) is a form of \(K\) if there exists an effective epimorphism \(U\to B\) and an equivalence \(E\times_B U\simeq U\times K\) over \(U\). For a specified effective epimorphism \(q\), the objects it splits with this type correspond in \(\operatorname{Desc}(q)\) to those data with degree-zero entry equivalent to \(U\times K\). The category of these objects, with all maps over \(B\), is full by Theorem~\ref{thm:effective-descent}. To compute the space of forms we restrict to equivalences of total objects.

The group object \(G=\underline{\operatorname{Aut}}(K)\) is the object of self-equivalences with its coherent composition. Concretely, a section of \(G\) over \(W\) is an equivalence \(W\times K\to W\times K\) over \(W\), and these spaces of sections are represented by the equivalence part of the internal hom. For a form \(E\), its frame object over \(B\) represents equivalences \(W\times K\to E\times_B W\) over \(W\), for test objects \(W\to B\). Local cartesian closedness gives the internal hom over \(B\), and the equivalence locus can be represented by requiring a two-sided inverse. More precisely, the object of maps with left and right inverses and witnesses for each gives this locus. At a map that is an equivalence the space of choices of each witness together with its inverse is contractible, by cancellation along an equivalence. At any other map at least one of the two spaces of choices is empty.

\begin{proposition}[Classification of forms]
\label{prop:forms-torsors}
The frame construction identifies the space of forms of \(K\) over \(B\) with the space of right \(G\)-torsors over \(B\), hence with \(\operatorname{Map}_{\mathcal X}(B,BG)\).
\end{proposition}

\begin{proof}
Precomposition equips the frame object \(\operatorname{Fr}(E)\) with a right \(G\)-action. A choice of local equivalence with \(U\times K\) gives a section over \(U\) and identifies the action there with the regular right action. The torsor comparison
\[
 \operatorname{Fr}(E)\times G\longrightarrow
 \operatorname{Fr}(E)\times_B\operatorname{Fr}(E)
\]
is an equivalence after pulling back along \(U\to B\), and is therefore an equivalence globally. The frame map to \(B\) is effective since it has a section over \(U\). Conversely, for a right \(G\)-torsor \(P\) form \(P\times^G K\) by realizing the diagonal action groupoid, where \(g\) acts by \((p,x)\mapsto(p g^{-1},g x)\). Over the effective epimorphism \(P\to B\) the pulled-back torsor has its tautological section. The associated object then identifies with the product with \(K\). The map from \(P\) to the frames of this associated object sends a local point \(p\) to \(x\mapsto[p,x]\). It is equivariant by the relation \([p g,x]=[p,g x]\), and becomes an equivalence where \(P\) is trivialized, so it is an equivalence. The evaluation of a frame on \(K\) gives \(\operatorname{Fr}(E)\times^G K\to E\), which becomes the product equivalence on the same trivializing cover and is thus an equivalence. Both maps are natural, giving inverse equivalences on cores. Corollary~\ref{cor:torsor-classification} gives the mapping-space description.
\end{proof}

If a finite-level truncation of \(K\) is imposed, its automorphism object also has bounded truncation degree, because mapping into an \(n\)-truncated object gives \(n\)-truncated mapping spaces. The classification then uses a correspondingly truncated target \(BG\). For general \(K\), it involves coherence at every degree. A section of the frame torsor is a chosen trivialization, and two sections may differ by a nontrivial automorphism. This distinguishes a property, being split, from the moduli of its witnesses.

\section{The boundary between \v{C}ech descent and hyperdescent}
\index{hypercover}\index{hypercompleteness}

A hypercover \(V_\bullet\to B\) is an augmented simplicial object whose relative matching maps are effective epimorphisms, including \(V_0\to B\). The \v{C}ech nerve of an effective epimorphism is a hypercover, with the matching maps in positive degrees equivalences. A general hypercover may have additional local choices in higher degrees. It need not be a groupoid, so the effectivity argument for Theorem~\ref{thm:effective-descent} does not prove descent along it.

For an object \(E\) of \(\mathcal X\), hypercompleteness means that mapping into \(E\) inverts all infinitely connective maps. The hypercover criterion identifies this with the requirement that the augmentation induce an equivalence
\[
     \operatorname{Map}_{\mathcal X}(B,E)
       \longrightarrow
     \lim_{[n]\in\Delta}\operatorname{Map}_{\mathcal X}(V_n,E)
\]
for all hypercovers \(V_\bullet\to B\) \cite[Theorem 6.5.3.12]{htt}. This uses the fact that \(|V_\bullet|\to B\) is infinitely connective. In a hypercomplete topos it is an equivalence. A truncated target \(E\) also satisfies the displayed condition in an arbitrary \(\infty\)-topos, because maps into a target of fixed finite truncation cannot detect an infinitely connective discrepancy. The distinction matters when local trivializations or coefficients are given as unbounded towers. Categorical descent for one effective epimorphism here avoids any completeness hypothesis because the augmentation of that particular nerve is already an equivalence.

\part{Monodromy and shape}
\chapter{Pro-spaces and continuous monodromy}
\label{ch:profinite}
\index{pro-space}\index{monodromy!continuous}

The passage from a single splitting field to a directed family of splitting
fields in \cite[Ch.~3]{bj} changes the meaning of an action. At each finite
stage the Galois group acts discretely. An orbit in the limit need only be
controlled by a quotient large enough to see that orbit. A continuous
action on an entire space can require a more restrictive or a more
flexible definition, depending on whether one demands a single
trivializing stage for the object or permits a filtered assembly of
objects with that property. This chapter distinguishes those constructions
before passing to the shape of a topos.

\section{Pro-objects and the role of the test space}

Write \(\mathcal S\) for the \(\infty\)-category of spaces. A pro-space
\(P\) is represented by a small cofiltered diagram
\(i\mapsto P_i\). For a second pro-space \(Q=\{Q_j\}_{j\in J}\),
the mapping formula is
\begin{equation}
\label{eq:pro-map}
 \operatorname{Map}_{\operatorname{Pro}(\mathcal S)}(P,Q)
 \simeq
 \lim_{j\in J}\operatorname*{colim}_{i\in I^{\mathrm{op}}}
             \operatorname{Map}_{\mathcal S}(P_i,Q_j).
\end{equation}
We place \(\operatorname{colim}_{i}\) over the filtered direction
throughout this chapter. Formula \eqref{eq:pro-map} comes from the
embedding of \(\operatorname{Pro}(\mathcal S)^{\mathrm{op}}\) in
\(\operatorname{Fun}(\mathcal S,\mathcal S)\) taking \(P\) to
\[
 F_P(T)=\operatorname*{colim}_{i}\operatorname{Map}(P_i,T).
\]
Indeed the representable \(\operatorname{Map}(Q_j,-)\) is a term
in the filtered presentation of \(F_Q\). A natural transformation from
\(F_Q\) to \(F_P\) is a compatible family of transformations from those
representables. The Yoneda lemma identifies the family with the right
side of \eqref{eq:pro-map}. In particular, an equivalence of pro-spaces
can be checked against every constant test \(T\). Here and below an
ambient size enlargement may be used for the categories of test functors.

The realized inverse limit \(\lim_i P_i\) loses this test property. Take
\(P_m=\{m,m+1,\dots\}\) with the inclusions as transition maps. As
discrete spaces their inverse limit is empty. Yet \(F_P(\varnothing)\)
is empty because every \(P_m\) is nonempty. The constant empty
pro-space has \(F_{\varnothing}(\varnothing)\simeq *\). Thus even
this tower of discrete spaces cannot be replaced by its realized
inverse limit for a classification problem.

\begin{definition}
\index{profinite completion}\index{protruncation}\index{pi-finite space@$\pi$-finite space}
A space is \emph{\(\pi\)-finite} if it is truncated, has finitely
many components, and has finite positive homotopy groups at every
basepoint. Let \(\mathcal S_\pi\) be the full subcategory on these
spaces, and let \(\mathcal S_{<\infty}\) be the full subcategory of
all truncated spaces. The corresponding completions of \(P\) are
written \(P^{\wedge}_\pi\in\operatorname{Pro}(\mathcal S_\pi)\)
and \(P_{<\infty}\in\operatorname{Pro}(\mathcal S_{<\infty})\).
\end{definition}

These completions are characterized by retaining exactly the tests of
the indicated kind. In the functor presentation, restrict
\(F_P\) to the indicated full subcategory. Each subcategory admits
finite limits. In the \(\pi\)-finite case, this follows from the
homotopy exact sequence for a pullback and the finiteness of the
component sets. Both restrictions preserve finite limits because
filtered colimits of spaces do so. The pro-object characterization by
accessible finite-limit-preserving functors then gives the completions.
For truncated spaces the accessibility issue can be handled by the
cofiltered diagram \(\{\tau_{\leq d}P_i\}_{i,d}\), since maps to any fixed
truncated test stabilize in the Postnikov coordinate. A particular consequence that will be
used repeatedly is
\begin{equation}
\label{eq:pro-completion}
 \operatorname{Map}(P^{\wedge}_\pi,T)
 \simeq\operatorname{Map}(P,T)
 \qquad (T\in\mathcal S_\pi).
\end{equation}
There is an analogous formula for \(P_{<\infty}\) with
\(T\in\mathcal S_{<\infty}\). Either construction is left adjoint
to the inclusion of the corresponding pro-category. This follows
for a general target by taking the limit of the equivalences for
its constant stages. The same argument in degree \(0\) recovers
the equivalence between pro-finite sets and Stone spaces used in
\cite[Ch.~3]{bj}. The higher completion tests Eilenberg--Mac Lane
spaces, classifying spaces, and their finite Postnikov combinations.

\section{A representation category at stages}
\index{representation!at a stage}

For a coefficient \(\infty\)-category \(\mathcal D\), set
\begin{equation}
\label{eq:stage-representations}
 \operatorname{Rep}_{\mathcal D}(P)=
 \operatorname*{colim}_{i}\operatorname{Fun}(P_i,\mathcal D),
\end{equation}
where the colimit is in \(\operatorname{Cat}_{\infty}\), and
the arrows are given by restriction. A representative of an object is
a family and all its transport coherence at a single stage.
A representative of a morphism is a natural transformation after
passing to a common stage. More precisely, if \(F\) and \(G\)
are already represented at a common stage \(i\), their mapping
space is
\begin{equation}
\label{eq:rep-map}
 \operatorname{Map}_{\operatorname{Rep}_{\mathcal D}(P)}(F,G)
 \simeq \operatorname*{colim}_{(k\to i)}
        \operatorname{Map}_{\operatorname{Fun}(P_k,\mathcal D)}
                     (F|_{P_k},G|_{P_k}).
\end{equation}
This follows by computing mapping spaces as fibers of the core of
the arrow category over the core of pairs of objects. Filtered colimits
of \(\infty\)-categories commute with forming these cores and arrow
categories, and filtered colimits of spaces commute with the fibers.
The reason for the first commutation can also be seen in simplicial
models: every prescribed simplex and every witness that an edge is
invertible involves a finite diagram, so it is already present at a
single refinement. Two stages that represent a prescribed datum
become comparable after a further refinement. This proves invariance
under cofiltered change of presentation and recovers the functorial
construction of \cite[\S2]{hoyois}.

The coefficient category may have infinitely many objects up to
equivalence. A finite stage can still carry an arbitrary object of
\(\mathcal D\). The crucial finiteness for completion is instead the
size and truncation of the spaces classifying possible transport
and coherent maps.

\begin{lemma}
\label{lem:pifinitemap}
If \(A\) and \(B\) are \(\pi\)-finite, then
\(\operatorname{Map}(A,B)\) is \(\pi\)-finite.
The component of \(A\) in the core of spaces is \(B\operatorname{Aut}(A)\)
and is \(\pi\)-finite.
\end{lemma}

\begin{proof}
A \(\pi\)-finite space has a simplicial model with finitely many
simplices in every degree. For a connected \(1\)-type use the nerve
of its finite fundamental group. At a higher Postnikov step, the
fiber \(K(M,n)\) for a finite abelian group \(M\) has a simplicial
abelian model with finitely many simplices in each degree. Represent
the step by its twisted \(k\)-invariant into the
\(K(M,n+1)\)-fibration over the nerve of the finite fundamental
group. This target fibration can be modeled by the Borel construction
for that group acting on \(K(M,n+1)\). Its total space and the
fiberwise path fibration are finite in each simplicial degree.
Pulling the path fibration back along the twisted invariant gives
a degreewise finite model for the extension. This inductive construction
and a finite disjoint union give the asserted model.

If \(B\) is \(n\)-truncated with \(n\geq 0\), the inclusion of the
\((n+1)\)-skeleton of this model into \(A\) induces an equivalence
on maps to \(B\), since the cells subsequently attached have dimension
at least \(n+2\). The skeleton is a finite cell complex. Mapping from
a cell attachment into \(B\) forms a finite homotopy pullback of
mapping spaces, starting with \(B\) and a point. Closure under
finite homotopy limits gives the claim. A \((-1)\)-truncated
target is either empty or contractible and the conclusion holds
directly. The equivalences \(A\to A\) form a union of components
of \(\operatorname{Map}(A,A)\), hence a \(\pi\)-finite group-like
space. Its delooping is connected and shifts its homotopy groups
by one degree, proving the second assertion.
\end{proof}

To specify the completion comparison for all morphisms, let
\(\operatorname{Rep}_{\mathcal S_\pi}^{\mathrm{fr}}(P)\) consist
of representations admitting a representative at a stage with
a finite range of equivalence classes of fibers. Here
\(\mathrm{fr}\) refers to the range of isomorphism types,
and the superscript imposes a condition on objects only.
Thus all maps in \eqref{eq:rep-map} between those objects are retained.

\begin{proposition}
\label{prop:completion-comparison}
Precomposition with \(P\to P^{\wedge}_\pi\) gives an equivalence
\[
 \operatorname{Rep}_{\mathcal S_\pi}(P^{\wedge}_\pi)
 \simeq \operatorname{Rep}_{\mathcal S_\pi}^{\mathrm{fr}}(P).
\]
When \(P\) admits a presentation by connected spaces,
the finite-range condition on the right is automatic.
\end{proposition}

\begin{proof}
A representation over a \(\pi\)-finite stage has finite range because
the stage has finitely many components. Conversely, for a finite
list \(D\) of equivalence classes of fibers, let \(C_D\) be the
union of those components in the core of \(\mathcal S_\pi\).
It is \(\pi\)-finite by Lemma \ref{lem:pifinitemap}. The moduli
space of stage objects with range in \(D\) is
\(\operatorname{Map}(P,C_D)\), because a functor from a space
factors through the core of its target. Thus every object lifts
through the completion by \eqref{eq:pro-completion}.

For the map space between two such objects choose finite lists
\(D_0,D_1\) for the source and target. Form the core of the arrow
category and restrict its endpoints to \(C_{D_0}\times C_{D_1}\).
The resulting space \(E\) has a projection to that product with
fiber \(\operatorname{Map}(A,B)\) at \((A,B)\).
Each fiber and the base are \(\pi\)-finite. A uniform truncation
bound for the fibers exists because the lists are finite.
The homotopy exact sequence shows that the total space is also
\(\pi\)-finite. The space of morphisms for stage representations
with given endpoints is the fiber of
\[
 \operatorname{Map}(P,E)\longrightarrow
     \operatorname{Map}(P,C_{D_0}\times C_{D_1}).
\]
This fiber description follows at each stage and passes through the
filtered colimit. Both its source and target are preserved under
\eqref{eq:pro-completion}, which proves full faithfulness.
A connected stage maps to one component of the core, proving the final
assertion.
\end{proof}

For example, take the constant pro-space \(\mathbb N\), with
\(\mathbb N\) discrete. A family of finite discrete spaces with
fiber cardinality \(m+1\) over \(m\in\mathbb N\) belongs to
\(\operatorname{Rep}_{\mathcal S_\pi}(\mathbb N)\).
It cannot descend from its profinite completion by
Proposition \ref{prop:completion-comparison}. Each individual fiber
satisfies the requisite finiteness condition. The failure comes from the
infinitely many equivalence classes occurring in a single family.

\section{From one-stage actions to continuous discrete actions}
\index{profinite group!continuous action}\index{ind-completion}

Let \(H=\lim_i H/U_i\) be a profinite group, where the \(U_i\) form
the directed system of open normal subgroups, and set
\[
   B_{\mathrm{pro}}H=\{B(H/U_i)\}_{i}.
\]
A left action on a discrete set \(M\) is continuous exactly when
the stabilizer of each element is open. One direction follows by
taking the preimage of the singleton under each orbit map. For the
converse, the action map restricted to \(H\times\{m\}\) is locally
constant for each \(m\), and these pieces are open in
\(H\times M\).

The stage category \(\operatorname{Rep}_{\mathrm{Set}}(B_{\mathrm{pro}}H)\)
realizes the full subcategory of discrete actions with a common
open normal subgroup acting trivially. An action at one stage has
that property. Conversely the common subgroup allows the action
to be written over a quotient. If two actions come from
\(H/U_i\) and \(H/U_j\), then a function between them is
\(H\)-equivariant exactly when it is equivariant for
\(H/(U_i\cap U_j)\), using surjectivity of the quotient map.
Therefore passing to stages identifies their mapping sets. Every
finite continuous \(H\)-set belongs to this subcategory, as a
finite intersection of open stabilizers is open.

Every continuous discrete \(H\)-set \(M\) is a filtered union of
finite continuous \(H\)-sets. Every orbit is finite by the compactness
of \(H\) and the openness of its stabilizer, and one can take the
finite unions of orbits. A finite \(H\)-set is compact among
continuous discrete \(H\)-sets. To see this, transport a map
into a filtered colimit to some stage as a function on its finite
underlying set. Choose an open normal subgroup fixing all elements
of the source and their chosen representatives at that stage.
Equivariance can now be checked against the finitely many elements
of the quotient by that subgroup. The required equalities hold after
some further refinement, by the description of filtered colimits
of sets. The same argument with finitely many pointwise equalities
proves uniqueness in the colimit of mapping sets. Thus the category
of all continuous discrete actions is the ind-completion of its
finite subcategory.

The difference between stage representations and this ind-completion
already occurs for \(H=\mathbb Z_p\). On the disjoint union
\(\coprod_{m\geq 1}\mathbb Z/p^m\mathbb Z\), the translation
action is continuous and requires arbitrarily deep quotients.
Its kernel is trivial, which is a closed subgroup that is
not open. This object is built from stage actions by taking
a filtered union, but no single stage represents it.

A different construction first discards the topology and takes
\(B(H^{\delta})\), the delooping of the underlying discrete group.
For \(H=\prod_{m\geq 1}\mathbb F_2\), there is an abstract
homomorphism \(H\to\mathbb F_2\) that annihilates
\(\bigoplus_{m\geq 1}\mathbb F_2\) and takes the all-ones sequence
to \(1\). Indeed, that sequence defines a nonzero vector in the
quotient by the direct sum. Extend this vector to an algebraic basis
and project to its coordinate. The resulting homomorphism is
discontinuous because its kernel is dense and proper. Its action
by translations on \(\mathbb F_2\) is an action of
\(H^{\delta}\) with no continuous counterpart. Thus using the
abstract classifying space instead of the pro-classifying space can
change even the finite-coefficient theory.

\section[Closed stabilizers]{Closed stabilizers and the infinite correspondence}
\index{profinite group!closed subgroup}\index{Galois correspondence!infinite}

The compact Hausdorff replacement of a finite discrete transitive
action is a transitive action on a profinite space. For every
closed subgroup \(L\leq H\), there is a homeomorphism of
\(H\)-spaces
\begin{equation}
\label{eq:closed-stabilizer}
 H/L\longrightarrow\lim_i (H/U_i)/(LU_i/U_i).
\end{equation}
Here \(H/L\) denotes the left \(H\)-space of cosets \(hL\), with
the quotient topology. To prove injectivity use
\(L=\bigcap_i LU_i\). Indeed, if \(h\notin L\), some coset
neighborhood \(hU_i\) is disjoint from \(L\), by closedness and
the neighborhood basis at the identity. A compatible point of the
inverse limit specifies a family of closed cosets of the groups
\(LU_i\). Any finite subfamily intersects by refining the indices.
Compactness gives a point in their intersection, proving surjectivity.
A continuous bijection from a compact space to a Hausdorff space
is a homeomorphism.

Conversely, given a transitive continuous action on a nonempty
profinite space \(Y\) and a point \(y\in Y\), its stabilizer \(L\)
is closed as the inverse image of \(\{y\}\) under the orbit map.
The induced continuous bijection \(H/L\to Y\) is again a
homeomorphism. Between pointed homogeneous spaces \((H/L,L)\)
and \((H/M,M)\) there is a unique pointed equivariant map precisely
when \(L\subseteq M\). In fact its value at \(hL\) is forced to be
\(hM\), and the subgroup condition makes this formula well defined
and continuous. An unpointed equivariant map can send \(L\) to any
coset \(hM\) with \(L\subseteq hMh^{-1}\); the map is then
determined by its value there. In the discrete continuous theory,
transitive objects instead correspond to open stabilizers.

For a Galois field extension \(K\subseteq L\), the ordinary
subextension correspondence uses closed subgroups of
\(H=\operatorname{Gal}(L/K)\). The equality of a subgroup
with the limit of its finite images in \eqref{eq:closed-stabilizer}
shows why closedness is the condition visible to all finite
subextensions. In greater algebraic generality the stages of the
classical correspondence can likewise assemble into profinite
\(H\)-spaces, as in \cite[\S3.5]{bj}. Continuous actions on
profinite spaces and continuous actions on discrete sets therefore
enter at different points of infinitary Galois theory. Replacing
both by one category of \(H^\delta\)-sets identifies constructions
that have different stabilizers and different colimits.

\section{Continuous cochains and twisted coefficients}
\index{cohomology!continuous}\index{local coefficients}

For a discrete continuous \(H\)-module \(M\), consider the cochain
groups
\[
 C^q_{\mathrm{cts}}(H,M)=\operatorname{Cont}(H^q,M)
 \qquad(q\geq0)
\]
with the group-cohomology differential. For a fixed cochain the
image is finite, since a compact space has finite image in a discrete
space. Moreover continuity implies that the cochain factors as a
function through some \((H/U_i)^q\). Pulling back its finitely many
fibers gives a clopen partition of \(H^q\). Each point has an open
product of cosets contained in its partition member. A finite
subcover by such products exists, and refining all the coset
neighborhoods to one \(U_i\) proves the factorization.
After shrinking \(U_i\) further, it fixes every element of the image.
Thus the cochain corresponds to an element of
\(C^q(H/U_i,M^{U_i})\). In degree \(0\) the identical conclusion
holds by openness of the stabilizer of the single value. The
differentials respect refinement of the quotient and inclusion of
the fixed submodules, giving an isomorphism of complexes
\[
 C^\bullet_{\mathrm{cts}}(H,M)
   \cong \operatorname*{colim}_{i}
                     C^\bullet(H/U_i,M^{U_i}).
\]
Filtered colimits are exact in abelian groups. Taking cohomology
therefore gives
\begin{equation}
\label{eq:continuous-cohomology}
 H^q_{\mathrm{cts}}(H,M)
   \cong \operatorname*{colim}_{i} H^q(H/U_i,M^{U_i}).
\end{equation}
If the action on \(M\) factors through a fixed stage, the
subgroups contained in its kernel form a cofinal subsystem and
one can use \(M\) throughout on the right.

Let \(M\) be finite with continuous action, and take \(q\geq1\).
The stage module actions induce families of pointed spaces
\(K(M,q)\) over \(B(H/U_i)\) for all sufficiently deep \(i\).
The mapping space of sections at a stage is the homotopy fixed
point space of \(K(M,q)\), modeled by totalizing the usual bar
cosimplicial object. Its set of components is
\(H^q(H/U_i,M)\). Taking components commutes with filtered
colimits, so the stage mapping space has components
\(H^q_{\mathrm{cts}}(H,M)\) by
\eqref{eq:continuous-cohomology}. This identifies pro-monodromy
with twisted cohomology directly from the stage dynamics. An
untwisted test of the form \(\operatorname{Map}(P,K(M,q))\)
only computes the version with trivial action. In particular,
truncation of an underlying homotopy type does not justify erasing
the transport of its coefficient groups.

The three continuous constructions above use the same inverse system
in different ways. Discrete ind-objects permit each orbit to descend
separately. Profinite spaces permit closed stabilizers and keep the
successive quotients of an orbit. Stagewise higher representations
retain the full space of local coefficient transports at every stage,
with refinement identifying equivalent presentations. The next
chapter explains how a single \(\infty\)-topos supplies the pro-space
that indexes such stages.

\chapter{The shape of a topos and its local systems}
\label{ch:shape}
\index{shape}\index{topos!shape}

Covering theory in \cite[Ch.~6]{bj} extracts a fundamental
group from a category of coverings and a fiber functor. A higher
topos contains families with fibers of all homotopy types.
The role of the fundamental invariant is then played by a space
or pro-space, and it can be constructed before deciding which
coefficients the classification ought to allow.

\section{Shape as a functor of constant tests}
\index{shape!relative}\index{constant object}

Let \(\mathcal X\) be an \(\infty\)-topos with constant-object
functor \(\Delta\colon\mathcal S\to\mathcal X\) and
global sections \(\Gamma\). For \(U\in\mathcal X\)
define \(\Pi(U)\) by
\begin{equation}
\label{eq:relative-shape}
 \operatorname{Map}_{\operatorname{Pro}(\mathcal S)}(\Pi(U),T)
  \simeq \operatorname{Map}_{\mathcal X}(U,\Delta T).
\end{equation}
The right side preserves finite limits in \(T\).
It is also accessible, since \(\Delta\) preserves colimits and
\(U\) is compact for some regular cardinal in the presentable
category \(\mathcal X\). The pro-representability criterion
from Chapter~\ref{ch:profinite} applies. Equivalently
\(\Pi(U)\) is the shape of the slice \(\mathcal X_{/U}\),
whose constant-object functor is
\(\Delta_U T=(U\times\Delta T\to U)\).
The shape of \(\mathcal X\) is \(\Pi_\infty\mathcal X=\Pi(1)\)
and corepresents the functor \(T\mapsto\Gamma\Delta T\).
These are the relative and absolute constructions of
\cite[\S7.1.6]{htt} and \cite[\S2]{hoyois}.

One presentation of \(\Pi(U)\) uses the category of pairs
\((T,t)\) with \(t\colon U\to\Delta T\).
The maps are maps of test spaces together with specified coherent
compatibilities with \(t\). The category of pairs is cofiltered
since the functor of tests preserves finite limits. Indeed, a
finite diagram of such pairs determines a diagram of test spaces
whose limit inherits a section by finite-limit preservation.
The resulting pair supplies a cone. An essentially small
cofiltered presentation exists by accessibility, and the
projection to \(T\) pro-represents \eqref{eq:relative-shape}.
Using all pairs is convenient for comparisons even when the
class of tests is large.

There is a relation to actions internal to \(\mathcal X\).
For a point \(a\in T\) put
\[
  P_a=U\times_{\Delta T}\Delta\{a\}.
\]
The pullbacks inherit transport along \(T\), producing a
functor \(a\mapsto P_a\) in \(\mathcal X_{/U}\).
The colimit of this functor is the terminal object of the
slice. Indeed the family of all points of \(T\), with all
transport, has colimit the identity of \(T\) in
\(\mathcal S_{/T}\). Apply \(\Delta\) and then pull back
along \(t\), both of which preserve colimits. Conversely a
functor \(T\to\mathcal X_{/U}\) with colimit the terminal
object is cartesian over the constant diagram of terminal
objects. The correspondence between families over that constant
diagram and objects over its colimit follows from universal
descent in an \(\infty\)-topos. It gives a section
\(U\to U\times\Delta T\) in the slice. This is the sense
in which the indexing pairs are principal \(T\)-actions,
even when \(T\) has many components or has higher homotopy.

The functor \(U\mapsto\Pi(U)\) sends colimits to colimits.
For \(U\simeq\operatorname{colim}_{d} U_d\), the right
side of \eqref{eq:relative-shape} is the limit of the corresponding
right sides. Constant tests are jointly conservative, hence the
comparison from the colimit of the \(\Pi(U_d)\) is an equivalence.
For a geometric morphism, pullback of constant objects along \(f\colon\mathcal Y\to\mathcal X\)
induces a map from \(\Pi_\infty\mathcal Y\) to
\(\Pi_\infty\mathcal X\). It is an equivalence exactly when
pullback gives equivalences on the spaces of all constant tests.

\section{From stage representations to objects}
\index{locally constant object}\index{complete Segal nerve}

For \((T,t)\) as above and a family \(E\to T\) of spaces,
put
\begin{equation}
\label{eq:realize-stage}
 E_t=U\times_{\Delta T}\Delta E\longrightarrow U.
\end{equation}
This operation respects both the reindexing of tests and
natural transformations of families. Straightening over \(T\)
and then taking the filtered colimit of stage representations
gives a functor
\[
  R_U\colon\operatorname{Rep}_{\mathcal S}(\Pi(U))
              \longrightarrow\mathcal X_{/U}.
\]
Its values are locally constant. By definition this means that
there is an effective epimorphism \(\coprod_\alpha U_\alpha\to U\)
such that the restriction to every \(U_\alpha\) is equivalent
in the slice to a constant object. In this case use the fibers
\(P_a\to U\) for one point \(a\) of each component of \(T\).
The sum of the selected points is an effective epimorphism
onto \(T\), and \(\Delta\) and pullback preserve effective
epimorphisms. Over \(P_a\), the object \(E_t\) is constant
with fiber \(\operatorname{fib}_a(E)\).
The same argument applies with \(\mathcal S_{\leq n}\)
or \(\mathcal S_\pi\) as the coefficient category.

A useful reduction in proving converse classification theorems
is that stage representations satisfy descent on \(U\). To
prove this, let \(\mathcal D\) be any chosen coefficient
category, and write \(W_m(\mathcal D)\) for the core of
\(\operatorname{Fun}([m],\mathcal D)\). The complete Segal
nerve of \(\operatorname{Rep}_{\mathcal D}(P)\) has
\(m\)-th space
\begin{equation}
\label{eq:rezk-stage}
 W_m(\operatorname{Rep}_{\mathcal D}(P))
     \simeq \operatorname{Map}(P,W_m(\mathcal D)).
\end{equation}
Indeed, at a stage this follows by taking the core of
\(\operatorname{Fun}(P_i,\operatorname{Fun}([m],\mathcal D))\).
Formation of \(W_m\) commutes with filtered colimits because
\([m]\) is represented by a finite simplicial set and taking
the core commutes with such colimits. Then use
\eqref{eq:pro-map}. Complete Segal nerves detect equivalences
and preserve limits. Since \(\Pi(U)\) takes colimits to
colimits, \eqref{eq:rezk-stage} proves that
\(U\mapsto\operatorname{Rep}_{\mathcal D}(\Pi(U))\)
sends colimits in \(\mathcal X\) to limits. All of the spaces
\(W_m(\mathcal D)\) may be taken in a larger universe
if required by the coefficient category.

\begin{lemma}
\label{lem:stage-comparison-local}
Suppose \(\mathcal D\) is a full subcategory of spaces and the
comparison \(R_V\) induces an equivalence on mapping spaces
between constant objects with values in \(\mathcal D\) for
every \(V\in\mathcal X\). Then \(R_U\) with coefficients
in \(\mathcal D\) is fully faithful for all \(U\), and its
essential image is the subcategory of objects locally constant
with values in \(\mathcal D\).
\end{lemma}

\begin{proof}
Given two stage representations, choose a common stage
\((T,t)\) and use its effective epimorphism
\(\coprod_a P_a\to U\) as above. In its \v{C}ech nerve, a
component of an iterated fiber product of the cover is indexed by
a tuple of points and maps into one of the \(P_a\)'s.
On that component both representations are constant.
The mapping-space comparison is therefore an equivalence on
each such component by the assumption. It is an equivalence
on their coproduct because both representation and slice
categories turn the coproduct into a product.
Taking the limit over the \v{C}ech nerve gives the original
mapping-space comparison. On the representation side this
uses \eqref{eq:rezk-stage}, and on the topos side it is
effective descent.

Now take a locally constant object with the indicated
coefficients. Choose an effective epimorphism giving
trivializations. On every component in every degree of its
\v{C}ech nerve the restrictions belong to the essential image,
since constant objects have constant stage representations.
Each realization functor there is fully faithful, so the coherent
descent datum of this object is in the essential image of the
limit of the functors. More explicitly, fullness of each image
lifts all transition equivalences, and every higher simplex
whose vertices are in the image lies in that full
subcategory. Descent supplies a representation over \(U\).
\end{proof}

The constant comparison in the lemma has an elementary form.
For \(A,B\in\mathcal S\) define
\(F_U(B)=\operatorname{Map}_{\mathcal X}(U,\Delta B)\).
We have natural equivalences
\[
\begin{aligned}
 \operatorname{Map}_{\operatorname{Rep}_{\mathcal S}(\Pi(U))}(A,B)
    &\simeq F_U(\operatorname{Map}(A,B)),\\
 \operatorname{Map}_{\mathcal X_{/U}}(\Delta_U A,\Delta_U B)
    &\simeq \operatorname{Map}(A,F_U(B)).
\end{aligned}
\]
The first is the formula for maps of constant diagrams at each
stage. The second uses \(\Delta_U A\simeq\operatorname{colim}_A U\)
in the slice and then maps out of that colimit. If
\(\Pi(U)=\{P_i\}\), the comparison is consequently
\begin{equation}
\label{eq:constant-stage-comparison}
 \operatorname*{colim}_i
      \operatorname{Map}(A,\operatorname{Map}(P_i,B))
    \longrightarrow
      \operatorname{Map}\bigl(A,\operatorname*{colim}_i
                          \operatorname{Map}(P_i,B)\bigr).
\end{equation}
This displays the relevant commutation requirement directly.

\section{Two comparison theorems}

\begin{theorem}[Higher local monodromy]
\label{thm:shape-local}
\index{topos!locally connected}
Fix \(n\geq0\), or let \(n=\infty\).
Suppose that the constant-object functor on \(n\)-truncated
objects admits a genuine left adjoint
\[
 L_n\colon\mathcal X_{\leq n}\longrightarrow\mathcal S_{\leq n}.
\]
Then for each \(U\in\mathcal X\), realization gives an equivalence
\[
 \operatorname{Rep}_{\mathcal S_{\leq n}}(\Pi(U))
     \simeq \operatorname{LC}_{\leq n}(\mathcal X_{/U}).
\]
\end{theorem}

\begin{proof}
For an \(n\)-truncated test \(T\), \(\Delta T\) is
\(n\)-truncated. By adjunction,
\[
 F_U(T)\simeq
   \operatorname{Map}(L_n\tau_{\leq n}U,T).
\]
If \(B\) is \(n\)-truncated, so is \(\operatorname{Map}(A,B)\)
for any space \(A\). For constant coefficients in the theorem
both sides of \eqref{eq:constant-stage-comparison} therefore
agree by the exponential law for spaces, with
\(L_n\tau_{\leq n}U\) as the representing space.
Apply Lemma \ref{lem:stage-comparison-local}.
For \(n=\infty\), truncation in this argument means the
identity.
\end{proof}

The hypothesis of Theorem \ref{thm:shape-local} is local
\(n\)-connectedness in the convention of \cite[Def.~3.2]{hoyois}.
For finite \(n\), the genuine left adjoint represents the relative
shape on \(n\)-truncated tests. Even then the monodromy category
in general uses the \emph{pro-\((n+1)\)-truncation} of shape.
Indeed the objects in \(W_0(\mathcal S_{\leq n})\) have loops
given by automorphism spaces, hence this classifier is
\((n+1)\)-truncated. The endpoint map of the arrow classifier has
\(n\)-truncated fibers, so its total space is also
\((n+1)\)-truncated. By \eqref{eq:rezk-stage} and the
mapping-space description using the arrow classifier, these
two tests suffice to recognize equivalences of the representation
categories. For \(n=0\) this recovers why a fundamental groupoid
rather than just a space of connected components is needed
to classify coverings.

\begin{theorem}[Finite-coefficient higher Galois theory]
\label{thm:shape-finite}
\index{locally constant object!finite coefficient}
For every \(\infty\)-topos and every \(U\), realization induces an
equivalence
\[
 \operatorname{Rep}_{\mathcal S_\pi}(\Pi(U))
       \simeq \operatorname{LC}_{\pi}(\mathcal X_{/U}).
\]
Here local trivializations may have arbitrarily many different
\(\pi\)-finite fiber types across the cover.
\end{theorem}

\begin{proof}
For \(A\in\mathcal S_\pi\) and fixed finite \(n\),
\(\operatorname{Map}(A,-)\) preserves filtered colimits of
\(n\)-truncated spaces. Use the finite \((n+1)\)-skeleton in the
proof of Lemma \ref{lem:pifinitemap}. Mapping out of a
finite cell complex is a finite limit of mapping spaces out of
points, by repeated use of its cell attachments, and finite
limits commute with filtered colimits in spaces. Filtered
colimits of \(n\)-truncated spaces also stay \(n\)-truncated.
Given \(B\in\mathcal S_\pi\), choose \(n\) with
\(B\) \(n\)-truncated. All spaces
\(\operatorname{Map}(P_i,B)\) in
\eqref{eq:constant-stage-comparison} are \(n\)-truncated.
The comparison is therefore an equivalence.
Lemma \ref{lem:stage-comparison-local} applies without a
connectivity hypothesis on \(\mathcal X\).
\end{proof}

Theorem \ref{thm:shape-finite} is \cite[Thm.~4.3]{hoyois}.
Its coefficients are \(\pi\)-finite spaces, while its indexing
object is the full shape. Proposition
\ref{prop:completion-comparison} permits using the profinite
completion for the subcategory admitting a finite range of
fiber types. The uncompleted shape can still be replaced by its
\emph{protruncation} for the full finite-coefficient theory when a
uniform truncation bound is imposed on all fiber types. Without
a uniform bound, a single classifier can have components of
arbitrarily high degree. In particular, the phrase
\emph{finite-coefficient classification} must specify the
category of coefficients and the tests of shape it claims to
retain.

Consider the space
\[
   B=\coprod_{m\geq2}K(\mathbb Z/2,m).
\]
On each component, the based path fibration has
\(\pi\)-finite fiber \(K(\mathbb Z/2,m-1)\). If the family came from
\(B_{<\infty}\), it would descend to some Postnikov stage
\(\tau_{\leq d}B\). For \(m>d\), the image of the
\(m\)-th component in that stage is contractible, so the
restricted family would have to be constant up to equivalence.
The based path fibration has contractible total space, whereas
the constant family on \(K(\mathbb Z/2,m)\) with this fiber
has nontrivial homotopy. The family therefore uses the full
shape. This obstruction concerns a collection of higher
degrees rather than its fundamental group.

\section[The classifying topos]{The classifying topos and its limit}
\index{topos!classifying}

For a space \(T\) put \(\beta(T)=\mathcal S_{/T}\).
The classifying property needed for pro-spaces is
\[
 \operatorname{Map}_{\mathrm{Top}_\infty}(\mathcal X,\beta(T))
       \simeq \operatorname{Map}_{\mathcal X}(1,\Delta T).
\]
This is the étale slice universal property. The right side
classifies a lift of \(\mathcal X\to\mathcal S\) through
\(\mathcal S_{/T}\) by sending the universal point of its
constant \(T\) to the chosen point. For a pro-space
\(P=\{P_i\}\), define \(\beta(P)=\lim_i\beta(P_i)\)
in \(\mathrm{Top}_\infty\). The classifying property and
\eqref{eq:pro-map} display an adjunction
\[
    \Pi_\infty\colon\mathrm{Top}_\infty
                  \rightleftarrows\operatorname{Pro}(\mathcal S)
                      :\beta .
\]
Its unit is \(\varphi\colon\mathcal X\to
\beta(\Pi_\infty\mathcal X)\).
The realization functor at \(U=1\) factors as
\[
 \operatorname{Rep}_{\mathcal S}(\Pi_\infty\mathcal X)
 \longrightarrow \beta(\Pi_\infty\mathcal X)
 \xrightarrow{\varphi^*}\mathcal X .
\]
The first arrow is induced by the projections from the
topos limit, using their inverse images, while the category
underlying that limit is computed by the direct images
\cite[\S2]{hoyois}. Under the local \(n\)-connectedness
hypothesis, Hoyois shows that both arrows, with \(n\)-truncated
coefficients and objects, are fully faithful. Moreover the image
of \(\beta(\Pi_\infty\mathcal X)_{\leq n}\) in
\(\mathcal X\) consists of filtered colimits of locally
constant \(n\)-truncated objects
\cite[Thm.~3.13]{hoyois}. Theorem \ref{thm:shape-local}
above proves the comparison on stage objects. The further
claim about the whole classifying topos uses the fact that
in this setting the shape admits a presentation by
\(n\)-connected transition maps.

There is a direct description of the topos limit in degree zero
for \(P=B_{\mathrm{pro}}H\) from Chapter~\ref{ch:profinite}.
An object of its \(0\)-truncation is a coherent family of
discrete \(H/U_i\)-sets \(M_i\). If \(U_j\subseteq U_i\), the
direct image takes invariants under \(U_i/U_j\), so the compatibility
is \(M_i\cong M_j^{U_i/U_j}\). These give compatible inclusions
of inflated \(H\)-sets and their union is continuous. Conversely
a continuous discrete \(H\)-set \(M\) yields the family
of \(M^{U_i}\) and is its union. The family for the union of
a compatible system returns that system since a fixed
element in the union is represented by a fixed element
at a common refinement. The construction on maps is
by restriction and union. Thus the topos limit contains
all continuous discrete actions, whereas the stage category
only contains actions with a common open kernel.

The extension \(\beta\) to all pro-spaces is not fully
faithful in general. Return to the tower of tails \(P_m\) in
Chapter~\ref{ch:profinite}. The underlying category of
\(\beta(P_m)\) is a product of copies of spaces, with one
coordinate for every element of the tail. The direct images
extend a tail-indexed family by terminal objects at the
missing coordinates. In a compatible family, each coordinate
is eventually missing, so the corresponding objects in
earlier stages are terminal. Hence \(\beta(P)\) is the
terminal category, the topos \(\beta(\varnothing)\).
But \(P\) and \(\varnothing\) were distinguished by their
constant empty test. This is the general pro-space obstruction
recorded in \cite[Rem.~2.8]{hoyois}.

\section{A failure of unrestricted stage monodromy}
\index{Cantor space}

Let \(C=\{0,1\}^{\mathbb N}\) with its product topology,
and \(\mathcal X=\operatorname{Shv}(C;\mathcal S)\).
Its shape is the system of finite clopen partitions of \(C\),
regarded as discrete spaces and ordered by refinement. Here is a
direct computation. A basis is given by the clopen subsets,
each of which is compact. A covering of a clopen subset by
basis members has a finite disjoint clopen refinement, by choosing
a finite subcover and successively subtracting earlier sets.
Thus the topology on the basis is generated by finite disjoint
clopen covering families. The constant sheaf with value \(T\)
takes a clopen \(V\) to
\[
       \operatorname*{colim}_{\mathcal P}
                  \prod_{D\in\mathcal P} T,
\]
over finite clopen partitions of \(V\). This prescription
satisfies the disjoint-cover sheaf condition because partitions
refining a fixed disjoint cover are cofinal and filtered
colimits commute with finite products. Its map from the
constant presheaf exhibits sheafification, since any natural
map to a sheaf is forced, on a partition, by the maps from
its parts and the disjoint-cover sheaf property. Taking
\(V=C\) computes every constant test of shape.

Consider the two constant sheaves with values \(\mathbb N\)
and \(\{0,1\}\). A map between them can be specified by a
sequence of locally constant functions on \(C\) with values
in \(\{0,1\}\). Indeed the source is a coproduct of countably
many terminal objects. Choose the functions giving the
successive coordinates of \(C\). The sequence is a morphism
of the constant sheaves, but no single finite partition
simultaneously makes every coordinate function constant on
each part. Such a partition would make the identity map of
the underlying set of \(C\) factor through a finite set.
The sequence therefore lies outside the image of the
stage morphism space \eqref{eq:constant-stage-comparison}.
This gives a failure of full faithfulness for unrestricted
set-valued stage monodromy on a topos with profinite
\(0\)-dimensional shape. The finite-coefficient theorem
still applies to this topos.

\section[Higher transport]{Covering spaces and higher transport}
\index{covering space}\index{fundamental group!fiber functor}

For a space \(B\), the classifying topos
\(\mathcal S_{/B}\simeq\operatorname{Fun}(B,\mathcal S)\)
has shape \(B\) itself. Its global sections of a constant
object \(T\) are \(\operatorname{Map}(B,T)\).
All its objects are locally constant, as pullback to the family
of points of \(B\) gives their fibers. When \(B\) is
connected and pointed, set-valued local systems are
\(\pi_1(B)\)-sets, since their classifying core and arrow
core are both \(1\)-truncated. The fiber functor at \(b\)
recovers \(\pi_1(B,b)\) as its group of natural automorphisms.
For left \(G\)-sets the forgetful functor is
\(\operatorname{Hom}_G(G,-)\). Equivariant self-maps
of the left regular \(G\)-set are right translations,
so its automorphism group is \(G^{\mathrm{op}}\).
The contravariance of the Yoneda embedding in the representing
object then gives \(G\) as the automorphism group of
the fiber functor. This realizes the fundamental-group
construction of \cite[\S6.7]{bj} in the representable case.
For ordinary locally path-connected semilocally simply
connected spaces, locally constant sheaves of sets are the
sheaves of sections of covering spaces, and path lifting gives
the same monodromy description on the ordinary fundamental
groupoid. These local topological hypotheses ensure that
every monodromy action reconstructs a cover.

Positive homotopy groups can enter even for connected
base with trivial \(\pi_1\). Take
\(B=K(\mathbb Z,2)\), pointed at its unit.
The based path fibration \(P_bB\to B\) has fiber
\(K(\mathbb Z,1)\) and contractible total space.
The constant family with this fiber has total space
\(B\times K(\mathbb Z,1)\), which is not contractible.
Thus these are inequivalent families over \(B\).
Their set-valued monodromy cannot distinguish them,
while the degree-two shape allows transport of the
homotopies in the fiber.

\section{Hypercompletion and shape}
\index{hypercompletion}\index{shape!protruncated}

Let \(h\colon\mathcal X^\wedge\to\mathcal X\) be the
hypercompletion. Every truncated object is hypercomplete.
Indeed an \(\infty\)-connected morphism is left orthogonal
to truncated objects over the terminal object, so localizing
at those morphisms fixes every truncated object.
For each truncated test \(T\), \(\Delta T\) is already
truncated in \(\mathcal X\), and
\[
 \operatorname{Map}_{\mathcal X}(1,\Delta T)
       \simeq \operatorname{Map}_{\mathcal X^\wedge}
                     (1,h^*\Delta T).
\]
Consequently hypercompletion induces equivalences on
protruncated shape and on its profinite completion, as
also observed in \cite[Rem.~5.3]{hoyois}. That argument
only tests truncated targets. Unbounded constant tests
and filtered colimits of such tests require separate
hypotheses. Likewise an attempted computation of full
shape from Postnikov sections cannot exchange the
uniformly bounded tests with their limit without
establishing the required commutation. The distinction
is quantified already by \eqref{eq:pro-map}, which places
a filtered colimit inside the limit over target stages.

For a concrete pro-space with trivial protruncation, use the
tower
\[
  Q_i=\prod_{m\geq i}K(\mathbb Z,m),
\]
with transitions that insert basepoints in the coordinates
new to the target stage. For fixed \(d\) the stages
\(i>d\) are \(d\)-connected, so every \(d\)-truncated
test treats \(Q\) as terminal. Yet the compatible
inclusions \(Q_i\to Q_1\) define a point of
\(\operatorname{Map}(Q,Q_1)\) that is not in the
component of a constant map. At any refinement, the coordinate inclusion followed
by projection to that coordinate induces the identity of \(K(\mathbb Z,m)\).
A constant homotopy class would yield the zero map on its
positive homotopy group. This proves that unbounded tests can
see data invisible in every fixed truncation.

\chapter{Obstructions to higher monodromy}\label{ch:obstructions}
\index{monodromy!obstruction theory}\index{Postnikov tower}

A classical action on a discrete set is determined by its permutation
representation. For a space \(F\), actions on \(F\) are classified by
maps into \(B\Aut(F)\), whose higher homotopy groups control extensions
of the permutation-level information. This chapter describes the
lifting problem without assuming that the base has a discrete
fundamental group as its entire homotopy type. The argument applies
also to maps from a pro-space into a bounded target.

\section{Automorphisms of an Eilenberg--Mac Lane space}

For an abelian group \(A\) and \(n\geq 1\), choose a functorial
Eilenberg--Mac Lane space \(K(A,n)\) represented by a simplicial
abelian group. Transport through another model leaves the following
equivalences invariant. Let \(\Aut_*\) denote the space of based
self-equivalences, with the identification at the basepoint included
as data.

\begin{lemma}\label{lem:ob-aut-em}
The map \(\Aut_*(K(A,n))\to\Aut(A)\) is an equivalence onto a discrete
group. The full group object of self-equivalences is equivalent to
\[
\Aut(K(A,n))\simeq K(A,n)\rtimes\Aut(A),
\]
with the translation action on the first factor and the usual action
of coefficient automorphisms.
\end{lemma}

\begin{proof}
Pointed maps \(K(A,n)\to K(A,n)\) have components
\(\widetilde H^n(K(A,n);A)\cong\operatorname{Hom}(A,A)\).
For \(n\geq2\), this follows from connectivity, Hurewicz, and the
universal coefficient theorem. For \(n=1\), the identification uses
\(H_1(K(A,1);\mathbb Z)=A\) and the freeness of \(H_0\). In both cases
the higher homotopy groups of the pointed mapping space are
\(\widetilde H^{n-i}(K(A,n);A)\) in degree \(i>0\), by suspension
and the representing property of Eilenberg--Mac Lane spaces.
They vanish by connectivity and reducedness. A pointed map is
an equivalence precisely when the induced endomorphism of \(A\)
is invertible. The based maps induced by automorphisms of \(A\)
therefore give an equivalence from the discrete group.

Evaluation of an unbased self-map at zero has this pointed mapping
space as its fiber. Translation by the evaluation value provides
a splitting on underlying spaces. In the chosen additive model the
assignment \((a,\alpha)\mapsto(x\mapsto a+\alpha(x))\) is a map
of group objects, using the semidirect product law. Its effect on
the evaluation fibers and bases is an equivalence, which proves
the assertion.
\end{proof}

Delooping gives a fibration with a distinguished section
\begin{equation}\label{eq:ob-exact}
 K(A,n+1)\longrightarrow B\Aut(K(A,n))
       \longrightarrow B\Aut(A).
\end{equation}
The section is induced by the inclusion of coefficient
automorphisms in the semidirect product. In the description by
families over \(B\Aut(A)\), the fibration is the local system of
spaces \(K(A,n+1)\) with its natural transport. Indeed looping the
quotient retrieves the stated conjugation action on \(K(A,n)\),
and the section identifies the neutral point in each fiber.

A local system of abelian groups on a space \(Y\) is a functor
\(M:Y\to\mathrm{Ab}\), where \(\mathrm{Ab}\) is the ordinary
category of abelian groups and all arrows in the image of \(Y\)
are isomorphisms. Write \(K_Y(M,r)\to Y\) for the family
associated to the functor \(y\mapsto K(M_y,r)\).
Define \(H^r(Y;M)\) to be the component group of the section
space \(\Gamma_Y K_Y(M,r)\). This agrees with cohomology with
local coefficients. In particular the zero section is part of
the structure, and translation identifies each section component
with the component of zero. For \(0\leq i\leq r\), looping the
fiberwise group and using that sections preserve limits yields
\[
\pi_i\bigl(\Gamma_Y K_Y(M,r),0\bigr)=H^{r-i}(Y;M).
\]
For \(i>r\) the homotopy group vanishes. This constructive
description needs no choice of cocycle model for the base.
\index{local coefficients}\index{Eilenberg--Mac Lane space}

\section{The Postnikov lifting problem}

Let \(Z\) be a connected space, \(n\geq2\), and
\(Y=\tau_{\leq n-1}Z\). The map \(\tau_{\leq n}Z\to Y\)
has fiber \(K(A,n)\) for \(A=\pi_nZ\) at a chosen basepoint,
by the homotopy exact sequence. Transport gives the local
coefficient system \(M\) associated to the usual
\(\pi_1(Z)\)-action on \(A\). The relative assertion below
uses the full section space.

\begin{theorem}[A Postnikov step as a space of nullhomotopies]
\label{thm:ob-postnikov}
The Postnikov extension \(\tau_{\leq n} Z\to Y\) determines a
section \(k\) of \(K_Y(M,n+1)\) defined up to equivalence with
its classifying data. The extension is equivalent over \(Y\)
to the homotopy pullback of \(k\) and the zero section. Given a
map \(f:T\to Y\), the space of lifts of \(f\) to
\(\tau_{\leq n}Z\) is equivalent to the space of paths from
\(0\) to \(f^*k\) in \(\Gamma_T K_T(f^*M,n+1)\).
It is nonempty exactly when \([f^*k]=0\) in
\(H^{n+1}(T;f^*M)\). If a lift is fixed, the full lift space
is equivalent to \(\Gamma_T K_T(f^*M,n)\).
\end{theorem}

\begin{proof}
A family over \(Y\) whose fibers are equivalent to \(K(A,n)\)
is classified by a map \(Y\to B\Aut(K(A,n))\), by
straightening and the component of \(K(A,n)\) in the core of
spaces. Composing with \eqref{eq:ob-exact} gives the classifying
map of \(M\). Over that map, a lift to \(B\Aut(K(A,n))\)
is exactly a section \(k\) of \(K_Y(M,n+1)\). One can identify
the classified family itself by pulling back the bundle of
paths that start at zero in the fiber \(K(A,n+1)\). Such paths
have endpoint \(k(y)\) and their path space is a principal
\(K(A,n)\)-space. Under trivialized coefficients this is the
universal path fibration, whose transport consists of
translations. Coefficient automorphisms respect zero and
act functorially on these paths. Across a change of coefficients
the transport thus combines translations with automorphisms by
the semidirect product of Lemma~\ref{lem:ob-aut-em}. This is
the universal family for the classifying map from which \(k\)
was obtained.

Pulling back along \(f\) and taking sections preserves the
pullback because sections are represented by the terminal
object in the slice. The stated path space therefore gives
all lifts and their homotopies. If one path is fixed, composing
with its inverse identifies this path space with the loop
space at zero. Looping the fiberwise Eilenberg--Mac Lane space
and then taking sections proves the final claim.
\end{proof}

In particular, when the obstruction is zero the set of
equivalence classes of lifts is a torsor for \(H^n(T;f^*M)\);
if a lift is chosen, its higher automorphisms in the lift space
are given by the lower cohomology groups. Applying this statement
to the Postnikov tower of \(B\Aut(F)\) gives successive lifting
problems for monodromy with prescribed fiber \(F\). The first
stage records the discrete group \(\pi_0\Aut(F)\), and later
stages can change both existence and the space of choices.
The usual Postnikov construction of spaces is reviewed
in~\cite{kerodon}, and the corresponding viewpoint on
cohomology in slices is developed in~\cite{nss}.

\begin{example}[A nonzero obstruction over a torus]
Take \(Y=B\mathbb Z^3\), the homotopy type of the three-torus,
\(M=\mathbb Z\) with trivial transport, and \(n=2\).
A generator \(k\) of \(H^3(Y;\mathbb Z)\) defines a family
\(Z\to Y\) as the fiber of \(k:Y\to K(\mathbb Z,3)\).
Its fibers are \(K(\mathbb Z,2)\) and the homotopy sequence
shows that \(Z\) is a connected \(2\)-type with
\(\tau_{\leq1}Z\simeq Y\).
The identity of \(Y\) cannot lift. The map
\(\mathbb T^3\to\mathbb T^3\) that sends the third coordinate
to its identity and leaves the first two coordinates fixed
does lift, because it factors through the two-torus.
The lift components in the second case form a torsor for
\(H^2(\mathbb T^3;\mathbb Z)=\mathbb Z^3\). At a chosen lift
the degree-one and degree-two homotopy groups of the lift
space are \(\mathbb Z^3\) and \(\mathbb Z\), respectively.
All these claims follow from the exterior-algebra cohomology
of the torus and Theorem~\ref{thm:ob-postnikov}. They distinguish
the vanishing test, the moduli of solutions, and the group of
automorphisms of a solution.
\end{example}

\section{Continuous lifting and convergence}

Let \(P=(P_i)\) be a pro-space as in Chapter~\ref{ch:profinite}.
Suppose that a partial monodromy map to a space \(Y\) is
represented by \(f_i:P_i\to Y\). For \(j\) over \(i\), write
\(f_j:P_j\to Y\) for its pullback. The lifting problem for a
target map \(Z\to Y\) has a useful stage description.

\begin{proposition}\label{prop:ob-pro-lifts}
There is a natural equivalence
\[
\begin{aligned}
 &\operatorname{fib}_{[f_i]}
     \bigl(\Map_{\Pro(\SSet)}(P,Z)\longrightarrow\Map_{\Pro(\SSet)}(P,Y)\bigr)\\
 &\qquad\simeq
 \colim_{j\to i}
   \operatorname{fib}_{f_j}\bigl(\Map(P_j,Z)\longrightarrow\Map(P_j,Y)\bigr).
\end{aligned}
\]
For a Postnikov step as in Theorem~\ref{thm:ob-postnikov}, a
lift exists exactly when the pulled-back obstruction vanishes
at some refining stage.
\end{proposition}

\begin{proof}
Restricting to refinements of the chosen stage is cofinal in the
filtered direction. The map in the proposition then compares
the filtered colimits of the indicated diagrams of mapping
spaces. The compatible basepoint diagram is constant at a point
and induces \([f_i]\). Filtered colimits of spaces commute with
the pullback defining its fiber. This proves the equivalence.
A filtered colimit of spaces has a component exactly when some
stage does, since \(\pi_0\) preserves colimits. The result now
follows from the vanishing criterion above.
\end{proof}

A space \(Z\) is the limit of its Postnikov tower. There is
therefore a comparison of spaces
\begin{equation}\label{eq:ob-pro-tower}
 \colim_i\lim_n \Map(P_i,\tau_{\leq n} Z)
 \longrightarrow
 \lim_n\colim_i \Map(P_i,\tau_{\leq n} Z).
\end{equation}
The left side classifies maps from \(P\) to the constant pro-space
\(Z\). The right side classifies maps from \(P\) to the pro-object
consisting of the truncations of \(Z\). A lift for each separate
obstruction need not use one fixed stage \(i\). A bounded target,
or a constant source, gives an equivalence in
\eqref{eq:ob-pro-tower}. The following computation shows how the
two sides differ without either condition.

\begin{example}[A compatible tower with no common stage]
Fix an abelian group \(A\neq 0\), for instance \(\mathbb Z\).
Let
\(P_i=\prod_{k\geq i}K(A,k)\) for \(i\geq2\), with maps
\(P_{i+1}\to P_i\) given by inserting the basepoint in the \(i\)th
coordinate. Let \(Z=P_2\). Every \(P_i\) is \((i-1)\)-connected.
For fixed \(n\), the map
\(\Map(*,\tau_{\leq n} Z)\to\Map(P_i,\tau_{\leq n}Z)\)
is an equivalence for \(i>n\), since the target is \(n\)-truncated.
The right side of \eqref{eq:ob-pro-tower} is consequently \(Z\)
and is connected.

On the left side, inclusions \(P_i\to Z\) represent one compatible
element of \(\colim_i\pi_0\Map(P_i,Z)\). This element differs
from the component of the constant map. Indeed, for every
refining stage \(P_j\), restrict the inclusion to
\(K(A,j)\) using the \(j\)th factor and project \(Z\) to
that factor. The composite is the identity and induces
\(\id_A\) on \(\pi_j\), whereas the constant map induces zero.
Thus every bounded observation becomes trivial after refinement,
while the unbounded monodromy test detects a nontrivial component.
\end{example}

There is a second issue even when the source is an ordinary
space. The inverse limit of the spaces of lifts encodes
compatibility among choices. Passing to their component sets
before forming the limit can lose information. For a countable
tower of pointed spaces, the homotopy limit is the homotopy
equalizer of the identity and shift on the product. An explicit
point of the homotopy equalizer consists of a point in each
space and a path from that point to the image of the next
point. This model yields a direct example.

\begin{example}[The degree-\(p\) tower of circles]
For a prime \(p\), take the tower \(B\mathbb Z\) with transition
maps induced by multiplication by \(p\). It is a tower of
abelian group objects in spaces. The difference of the identity
and shift on the product is induced on \(\pi_1\) by
\[
 D:\prod_{i\geq0}\mathbb Z\longrightarrow\prod_{i\geq0}\mathbb Z,
 \qquad (a_i)\longmapsto(a_i-pa_{i+1}).
\]
The homotopy equalizer is its fiber as a map of group objects,
so the long exact sequence identifies its component group with
\(\operatorname{coker}D\). The map
\((b_i)\mapsto \sum_{i\geq0}p^i b_i\) converges in
\(\mathbb Z_p\), is surjective by the \(p\)-adic expansion,
and takes the image of \(D\) into the ordinary integers by
telescoping. Conversely, if the sum is an integer \(s\), put
\[
 a_i=p^{-i}\Big(s-\sum_{k<i}p^k b_k\Big)\in\mathbb Z.
\]
Divisibility follows from the \(p\)-adic equality, and
\(a_i-pa_{i+1}=b_i\). Thus the component group is
\(\mathbb Z_p/\mathbb Z\), although the component set at
each stage is a singleton. This is the elementary tower
calculation behind the derived inverse limit term.
\end{example}

The two examples locate different coherence requirements.
Maps into a pro-target can be compatible even when they have
no common source stage. Choices in a tower of lift spaces can
have additional components carried by the transition paths.
The classification space keeps both effects visible.

\part{Algebraic Galois theories}
\chapter{Pierce spectra and derived \'{e}tale descent}\label{ch:algebra}

The ring-theoretic construction in \cite[Chapter~4 and Appendix~A]{bj} changes two ingredients of field-theoretic Galois theory. It uses the entire space of connected components as the space of objects of the Galois groupoid, and it interprets split algebras by sections of a sheaf of rings over a Stone space. Both constructions extend to commutative ring spectra. This extension has a particularly concrete feature. Idempotents determine the topology at every level of the construction, while the algebras of sections retain all the homotopy groups of the ring spectrum. The derived affine category and its Pierce reflection therefore carry different amounts of information.

Throughout this chapter, \(A\) denotes an \(E_\infty\)-ring, possibly nonconnective, \(\operatorname{CAlg}_A\) its \(\infty\)-category of commutative algebras, and \(Y\) a Stone space, that is, a compact Hausdorff space with a basis of clopen subsets. The empty space and the zero ring are allowed. Write \(P(A)\) for the Stone spectrum of the Boolean algebra \(\operatorname{Idem}(\pi_0 A)\). Points are ultrafilters. For an idempotent \(e\), the clopen subset \(D(e)\) consists of ultrafilters containing \(e\), and every clopen of \(P(A)\) arises uniquely in this way. A ring map \(A\to C\) gives a continuous map \(P(C)\to P(A)\) by inverse image of ultrafilters. All tensor products between ring spectra below are derived.

\section{Idempotents, localizations, and stalks}
\index{Pierce spectrum}\index{idempotent!spectral localization}\index{Pierce sheaf}

Multiplication, complement, and join in \(\operatorname{Idem}(\pi_0 A)\) are given by \(ef\), \(1-e\), and \(e+f-ef\). These formulas only use degree zero. The localization \(A[e^{-1}]\), by contrast, is an algebra over the entire spectrum. Its universal property says that the mapping space from \(A[e^{-1}]\) into an \(A\)-algebra \(C\) is contractible when \(e\) maps to a unit of \(\pi_0 C\), and empty otherwise. An idempotent is a unit exactly when it is the element \(1\). Localization satisfies
\[
 \pi_t(A[e^{-1}])\cong (\pi_t A)[e^{-1}]\cong e\pi_t A
 \qquad(t\in\mathbf Z).
\]
Here the last module is viewed over \(e\pi_0 A\), whose unit is \(e\). These facts are instances of spectral localization \cite[Example~7.5.0.7]{ha}. They imply
\[
 A\xrightarrow{\ \sim\ } A[e^{-1}]\times A[(1-e)^{-1}].
\]
Indeed the map induces \(m\mapsto(em,(1-e)m)\) on every homotopy group. The proof also applies to any finite partition of unity by pairwise orthogonal idempotents. Equivalences of ring spectra are detected on the underlying spectra.

\begin{proposition}\label{prop:pierce-sheaf}
The rule \(D(e)\mapsto A[e^{-1}]\), with the localization maps as restrictions, defines a sheaf on the clopen basis of \(P(A)\). It extends uniquely, up to equivalence, to a sheaf of \(E_\infty\)-rings on all open subsets. Its global sections are \(A\), and its stalk at an ultrafilter \(x\) is
\[
 A_x\simeq\underset{e\in x}{\operatorname{colim}} A[e^{-1}],
\]
where the transition maps restrict along inclusions of smaller clopens. The ring \(\pi_0 A_x\) has precisely the two idempotents \(0\) and \(1\).
\end{proposition}

\begin{proof}
Every cover of a clopen by clopens has a finite subcover and a finite pairwise disjoint clopen refinement formed by intersections and complements. For a disjoint cover the sheaf comparison is exactly the product decomposition above, and intersections of distinct members have the zero ring as sections. The finite refinements yield the sheaf condition on the basis. Explicitly, for a finite cover, decompose every member and every overlap by the same finite Boolean partition. On each nonempty part, the indices of cover members containing it form a nonempty set, with a contractible \Cech{} indexing groupoid. The limit of the restrictions over that part is its section ring. Taking the product over parts recovers the original section ring. The same argument applies to an arbitrary clopen cover by restricting over each part of a finite subordinate disjoint partition. On each part a chosen member contains the entire part, so its restricted cover has a section and the \Cech{} limit is split. The extension and the stalk formula follow by restriction to the basis.

Filtered colimits commute with \(\pi_0\), and the polynomial identity defining an idempotent descends to a stage. Every idempotent in \(e\pi_0 A\) is an idempotent \(f\) of \(\pi_0 A\) with \(f\leq e\). An ultrafilter containing \(e\) contains either \(f\) or \(1-f\). Restrict along the former to make the idempotent \(1\) and along the latter to make it \(0\). Each ring at a stage has distinct unit and zero because its idempotent belongs to an ultrafilter. They remain distinct in the filtered colimit.
\end{proof}

An ultrafilter stalk is indecomposable in the sense of products. The Zariski spectrum instead organizes tests at prime ideals and uses local rings. For degree-zero \(A\), Proposition~\ref{prop:pierce-sheaf} recovers the Pierce sheaf in \cite[Section~4.2]{bj}. Its spectral analog shows exactly where the extra information resides. The stalks and restrictions retain the groups \(e\pi_t A\) before passage to the colimit.

\section{A relative Stone adjunction}
\index{Stone space!relative Pierce adjunction}\index{split algebra!Pierce}

Let \(q:Y\to P(A)\) be continuous. To build its algebra of sections, take pairs \((\mathcal U,W)\) where \(\mathcal U\) is a finite partition of \(Y\) into nonempty clopens and \(W=(W_U)_{U\in\mathcal U}\) is a family of clopens in \(P(A)\) satisfying \(q(U)\subseteq W_U\). For the empty partition use the empty product, the zero algebra. A refinement partitions into smaller clopens and shrinks each chosen neighborhood inside the neighborhood of its parent. Denote by \(e_{W_U}\) the associated idempotent and put
\begin{equation}\label{eq:pierce-relative}
  A\langle Y\rangle
  =\underset{(\mathcal U,W)}{\operatorname{colim}}
     \prod_{U\in\mathcal U} A[e_{W_U}^{-1}].
\end{equation}
The transition map to a refinement repeats the old factor for each of its children and then restricts it along the smaller neighborhood. Intersections of partitions and neighborhoods give common refinements, so this is a filtered colimit. No point of \(q(U)\) is required to have an open image, which is the reason for using neighborhoods of the images.

\begin{lemma}\label{lem:pierce-split-products}
For an \(A\)-algebra \(C\), the space of maps
\[
 \operatorname{Map}_{\operatorname{CAlg}_A}
       \left(\prod_{j=1}^r A[e_j^{-1}],C\right)
\]
is discrete. Its points are ordered clopen partitions \(P(C)=\coprod_j V_j\), where empty parts are allowed and each \(V_j\) maps into \(D(e_j)\) in \(P(A)\).
\end{lemma}

\begin{proof}
The factors in the source define orthogonal idempotents summing to \(1\), whose images in \(\pi_0 C\) give the partition \((V_j)\). After this partition is fixed, the target decomposes as \(\prod_j C[v_j^{-1}]\), for the associated idempotents \(v_j\). Mapping to the \(j\)-th factor forces the source idempotent of the \(j\)-th factor to map to \(1\), so the map factors through the \(j\)-th source factor. The universal property of localization makes this factorwise mapping space contractible exactly when \(e_j v_j=v_j\). To see that no higher ambiguity enters through the partition, one can use \cite[Corollary~7.5.4.6]{ha}. The source is a spectral \'etale \(A\)-algebra because its degree-zero algebra is a finite product of idempotent localizations and its higher groups obey the flat base-change formula. Thus mappings to an arbitrary target are determined discretely by degree zero. The Boolean condition \(e_jv_j=v_j\) is equivalent to the asserted containment of clopens.
\end{proof}

\begin{theorem}[Relative Pierce adjunction]\label{thm:pierce-relative}
There is a natural equivalence
\[
  \operatorname{Map}_{\operatorname{CAlg}_A}(A\langle Y\rangle,C)
   \simeq \operatorname{Hom}_{\mathsf{Stone}/P(A)}(P(C),Y).
\]
The canonical continuous map \(Y\to P(A\langle Y\rangle)\) is an isomorphism over \(P(A)\). Consequently \(Y\mapsto A\langle Y\rangle\) embeds \((\mathsf{Stone}/P(A))^{\mathrm{op}}\) fully faithfully into \(\operatorname{CAlg}_A\). Its image will be called the \emph{Pierce-split} \(A\)-algebras. On opposite categories the left adjoint is the functor \(C\mapsto P(C)\) over \(P(A)\). The corresponding reflection is therefore represented on algebras by the map \(A\langle P(C)\rangle\to C\), and its fixed objects are exactly the Pierce-split algebras.
\end{theorem}

\begin{proof}
For a pair \(d=(\mathcal U,W)\), set \(Y_d=\coprod_{U\in\mathcal U}W_U\) over \(P(A)\). The basis containment condition in Lemma~\ref{lem:pierce-split-products} gives the mapping set from \(P(C)\) to \(Y_d\) over \(P(A)\). The finite-stage ring has Pierce spectrum \(Y_d\), since the Boolean algebra of its idempotents is the finite product of those of its factors. A refinement gives \(Y_{d'}\to Y_d\) by inclusion into parent neighborhoods. There is an isomorphism in Stone spaces over \(P(A)\)
\[
 Y\xrightarrow{\ \sim\ }\lim_d Y_d.
\]
For nonempty \(Y\), a point of the limit has a single base coordinate \(x\in P(A)\), as is seen by considering the coarsest partition with neighborhood \(P(A)\). The coordinates at the unshrunk partitions choose a compatible member of each finite clopen partition. They determine a unique \(y\in Y\) by compactness and separation by clopens. For any clopen neighborhood of \(q(y)\), continuity gives a clopen piece about \(y\) mapping into that neighborhood, and the corresponding coordinate forces \(x\) to lie in it. Hence \(x=q(y)\). Conversely \(y\) supplies all coordinates. The canonical continuous bijection is a homeomorphism by compactness and Hausdorffness. The statement for the empty space follows from the empty partition.

Mapping out of the colimit in \eqref{eq:pierce-relative} now gives the displayed equivalence since limits of discrete spaces remain discrete. Independently, filtered colimits of \(E_\infty\)-rings induce filtered colimits of Boolean algebras of idempotents. An element and the equation making it idempotent occur at a finite stage, and equalities hold after a further refinement. Stone duality identifies the spectrum of that colimit with \(\lim_d Y_d\). This proves the second claim and identifies the two constructions naturally. The last statement follows by taking \(C=A\langle Z\rangle\).
\end{proof}

For an ordinary ring \(R\), formula \eqref{eq:pierce-relative} gives the ring of sections of the inverse-image Pierce sheaf over \(Y\). A section of the inverse image is locally represented by a section on a clopen neighborhood of its image. Compactness gives a finite clopen partition subordinate to these representatives. If two collections of representatives define the same section, local equality of their germs and a second finite clopen refinement make their representatives equal on sufficiently small clopen neighborhoods. Thus they become equal at a common stage. This identifies the construction with the relative section functor of \cite[Section~4.3]{bj}, including its variance. For a ring spectrum the same formula computes each homotopy group by taking the colimit of the corresponding finite product of modules \(e_{W_U}\pi_t A\). Each finite-stage ring is finite \'etale over \(A\), since its ordinary degree-zero algebra is a finite product of direct summands of the unit and its homotopy groups satisfy flat base change. Thus the Pierce-split algebras are filtered colimits of these particular finite \'etale algebras.

\begin{example}[A compact parameter space]\label{ex:pierce-cantor}
Suppose \(\pi_0 A\) is nonzero with no nontrivial idempotents and \(Y=\{0,1\}^{\mathbf N}\) is the Cantor space. The relative algebra is the filtered colimit of products of copies of \(A\) indexed by clopen partitions. Its degree-\(t\) group is the group of locally constant functions \(Y\to\pi_t A\) with the target discrete. Every such function has finite image by compactness. For degree zero, even the characteristic function of a single point lies outside this group, since that point is closed and fails to be open. The product \(\prod_{y\in Y}A\) has all functions as homotopy groups and has the Boolean algebra of all subsets of \(Y\) as degree-zero idempotents. The two constructions distinguish continuous parameters from arbitrary parameters.
\end{example}

\begin{corollary}[Base change]\label{cor:pierce-base-change}
For a map \(A\to C\), there are natural equivalences
\[
 C\otimes_A A\langle Y\rangle
 \simeq C\langle Y\times_{P(A)}P(C)\rangle,
 \qquad
 A\langle Y\rangle\otimes_A A\langle Z\rangle
 \simeq A\langle Y\times_{P(A)}Z\rangle.
\]
In particular, \(P\) takes an affine pullback to the corresponding Stone pullback if one leg is the opposite of a Pierce-split algebra.
\end{corollary}

\begin{proof}
Apply the universal property of pushout to maps into an arbitrary \(C\)-algebra \(D\). The right side of Theorem~\ref{thm:pierce-relative} then describes lifts of \(P(D)\to P(A)\) to \(Y\), which are precisely maps over \(P(C)\) to the indicated pullback. The first equivalence follows from Yoneda. Taking \(C=A\langle Z\rangle\) gives the second.
\end{proof}

\section{The Galois groupoid of a normal Pierce splitting}
\index{Galois groupoid!Pierce}

We now impose an effectivity hypothesis separately from the classification of split algebras. For \(A\to B\), write \(B_n=B^{\otimes_A(n+1)}\), with Amitsur cofaces given by insertion of units and codegeneracies by multiplication. Say the map has \emph{affine algebra descent} when extension of scalars induces
\begin{equation}\label{eq:affine-alg-desc}
  \operatorname{CAlg}_A \xrightarrow{\ \sim\ }
   \lim_{[n]\in\Delta}\operatorname{CAlg}_{B_n}.
\end{equation}
In this limit, the transition functors are extensions of scalars. Say that the map is \emph{normal for Pierce splitting} when \(B\otimes_A B\), viewed as a \(B\)-algebra using the first factor, is Pierce-split. Symmetry gives the condition for the second factor. This condition is adapted to the inverse-limit construction above. It refers to one specific reflection, whereas affine algebra descent refers to the ambient affine category.

\begin{theorem}[Pierce Galois correspondence]\label{thm:pierce-galois}
If \(A\to B\) has affine algebra descent and is normal for Pierce splitting, the simplicial Stone space
\[
  G_n =P(B_n)
\]
is an internal groupoid. The opposite of the full subcategory of \(\operatorname{CAlg}_A\) on those \(D\) for which \(B\otimes_A D\) is Pierce-split is equivalent to the category of internal \(G_\bullet\)-actions in Stone spaces. The underlying space of the action associated to \(D\) is \(P(B\otimes_A D)\) over \(G_0=P(B)\).
\end{theorem}

\begin{proof}
The tensor-power algebra in degree \(n\), with its map from the factor at any vertex, is the \(n\)-fold tensor product over \(B\) of \(B\otimes_A B\) with a common selected factor. Normality and Corollary~\ref{cor:pierce-base-change} make it Pierce-split over that vertex. The usual Segal decomposition of the affine \Cech{} nerve expresses its degree \(n\) as the iterated pullback of \(n\) consecutive arrows over \(\operatorname{Spec}(B)\), where \(\operatorname{Spec}\) here denotes passage to opposite algebras. Each pullback can use a Pierce-split leg at its vertex. Corollary~\ref{cor:pierce-base-change} therefore identifies \(G_n\) with the corresponding Stone pullback of composable arrows. Applying the same corollary to the first and last vertices of a triangle, or to either pair of edges with a common tail or head, gives the outer horn isomorphisms for inverses. Concretely, the exchange of the two factors in \(B_1\) supplies inversion, and its identities with multiplication follow from the corresponding identities in the affine \Cech{} groupoid. Thus \(G_\bullet\) is a groupoid object.

Under \eqref{eq:affine-alg-desc}, requiring the object at degree zero to be Pierce-split is equivalent to requiring it at all degrees. Indeed each comparison with a vertex identifies the object at degree \(n\) with an extension of scalars of its degree-zero value, and Pierce splitting is preserved by extension of scalars. The equivalence of Theorem~\ref{thm:pierce-relative} is natural for precisely these transition functors by Corollary~\ref{cor:pierce-base-change}. Taking opposites therefore identifies the desired category with
\[
   \lim_{[n]\in\Delta} \mathsf{Stone}_{/G_n}.
\]
For an internal groupoid this limit is the category of internal actions. Given an object \(V\to G_0\), an identification of its two pullbacks to \(G_1\) is an arrowwise transport map. The degenerate arrows impose its unit law and the pullbacks over \(G_2\) impose composition. The Segal isomorphisms identify all further compatibilities with iteration of these two laws. Conversely a unitary associative action gives that identification and all its extensions by pullback. Morphisms are maps over \(G_0\) commuting with the transport. This proves the equivalence and the claim about the underlying space.
\end{proof}

The condition on \(B\otimes_A B\) also gives the closure of the trivial objects under the special base change that transports a \(B\)-algebra along \(A\to B\). For \(D=B\langle Y\rangle\), first apply Corollary~\ref{cor:pierce-base-change} with the second \(B\)-structure on \(B\otimes_A B\). The resulting algebra is Pierce-split over \(B\otimes_A B\). If the latter, with its first \(B\)-structure, corresponds to a Stone space \(T\) over \(P(B)\), the composite is Pierce-split over the first \(B\). To verify this last assertion, express each algebra over \(B\otimes_A B\) using \eqref{eq:pierce-relative}. A finite product of clopen factors of \(B\langle T\rangle\) corresponds over \(B\) to a finite union of clopen subspaces of \(T\), by the relative mapping formula. Filtered colimits of these algebras correspond to inverse limits of their Stone spaces by that same formula. Conversely, taking \(D=B\) in the closure condition recovers normality. In the ordinary ring setting, these are the split-object and groupoid mechanisms in \cite[Sections~4.5--4.7]{bj}.

\section{Flat descent and the geometric \'etale site}
\index{descent!faithfully flat}\index{etale algebra@\'etale algebra!spectral}

Call \(A\to B\) faithfully flat when \(\pi_0 A\to\pi_0 B\) is faithfully flat and
\[
 \pi_t A\otimes_{\pi_0 A}\pi_0 B\xrightarrow{\ \sim\ }\pi_t B
 \qquad (t\in\mathbf Z).
\]
For any \(A\)-module \(M\), flat base change gives \(\pi_t(M\otimes_A B)=\pi_t M\otimes_{\pi_0 A}\pi_0 B\). One way to establish this formula is to compare the two homological functors in \(M\). The candidate assembled from the homotopy groups and a flat module is exact and preserves coproducts, and the comparison is an isomorphism on all suspended copies of \(A\). The ordinary ring \(\pi_0 A\) acts on the homotopy groups, so the comparison is balanced. The class of modules for which it is an isomorphism in every degree is stable and closed under coproducts. The suspended unit generates \(\operatorname{Mod}_A\) as a localizing subcategory. In particular extension of scalars detects equivalences and concentration of homotopy groups.

Faithfully flat spectral descent identifies \(\operatorname{Mod}_A\) with the Amitsur limit of \(\operatorname{Mod}_{B_n}\) by extension of scalars \cite[Theorem~3.1]{mathew}. The functors are symmetric monoidal, so the equivalence is symmetric monoidal with the pointwise structures on the limit. Passing to commutative algebra objects yields \eqref{eq:affine-alg-desc}. Indeed an algebra is given by a functor out of the commutative operad satisfying the inert-arrow conditions, and functor formation and these conditions are checked in every coordinate of a limit. For ordinary rings, concentrate all homotopy groups in degree zero. A differential degree cannot survive base change and then become zero by faithful flatness. This recovers the affine statement of classical fpqc descent \cite{stacks}. The symmetric monoidal comparison carries the multiplication and unit operations through module descent.

There is a different restriction of the same descent formula. An \(A\)-algebra \(C\) is \emph{spectrally \'etale} in the geometric sense when \(\pi_0 C\) is \'etale over \(\pi_0 A\) in the ordinary sense and
\[
 \pi_t C\cong\pi_t A\otimes_{\pi_0 A}\pi_0 C
 \qquad(t\in\mathbf Z).
\]
It is finite \'etale when this ordinary algebra is also finite as a module. Lurie's \'etale lifting theorem gives an equivalence
\begin{equation}\label{eq:etale-pi0}
 \operatorname{CAlg}_A^{\mathrm{et}}
   \simeq N(\operatorname{Alg}_{\pi_0 A}^{\mathrm{et}})
\end{equation}
and gives discrete mapping spaces from \'etale sources into any \(A\)-algebra, with mapping sets computed on \(\pi_0\) \cite[Theorem~7.5.0.6 and Corollary~7.5.4.6]{ha}. Formula \eqref{eq:etale-pi0} applies also to the finite \'etale subcategory. It is compatible with base change because pushouts of \'etale maps remain \'etale and the degree-zero comparison is ordinary base change. For connective bases, the corresponding spectral affine small \'etale sites agree with the classical sites for \(\pi_0 A\). Their generating covers are tested by joint surjectivity on the same underlying spectra of ordinary rings \cite{dag}. The equivalence of sites yields an equivalence of their \(\infty\)-topoi of space-valued sheaves, which still have higher sheaf cohomology even though the affine \'etale probes and their maps are discrete.

\begin{proposition}\label{prop:spectral-etale-desc}
If \(A\to B\) is faithfully flat, the \(\infty\)-categories of \'etale and finite \'etale \(A\)-algebras are the respective Amitsur limits of the same subcategories over \(B_n\).
\end{proposition}

\begin{proof}
Both properties are stable under extension of scalars. Conversely let \(C\) be an \(A\)-algebra whose base change to \(B\) has either property. Ordinary faithfully flat descent of the \'etale and finite properties, together with \(\pi_0(B\otimes_A C)\cong\pi_0 B\otimes_{\pi_0 A}\pi_0 C\), yields the required condition on degree zero \cite{stacks}. For each \(t\), tensor the homotopy comparison map for \(C\) over \(\pi_0 A\) with \(\pi_0 B\). The resulting map is the homotopy comparison for \(B\otimes_A C\) as a \(B\)-algebra, by flat base change for modules. It is an isomorphism, so the original map is an isomorphism by faithful flatness. Hence the required subcategories are the restrictions of \eqref{eq:affine-alg-desc} selected in degree zero, which proves the assertion.
\end{proof}

\section{What the two restrictions detect}

The Pierce functor is built from components, while a small \'etale site is built from flat unramified maps and contains the Zariski opens. If \(A\) has connected \(\pi_0\), the former sends its base to one point regardless of the prime spectrum. A geometrically \'etale map may have connected source and target with a genuinely different field of functions. Such a map is sent to the identity of a one-point Stone space. It can nevertheless become Pierce-split after a sufficiently large faithfully flat extension, and Theorem~\ref{thm:pierce-galois} then records its monodromy through the arrow space of that extension. Component data alone therefore cannot recover the geometric \'etale site on the base. The descent construction uses component data simultaneously on the cover and on its iterated overlaps.

Here is the field calculation at an infinite level. Let \(L/K\) be an infinite algebraic Galois extension, the union of finite Galois subextensions \(L_i/K\), with Krull group \(G=\lim_i \operatorname{Gal}(L_i/K)\). As \(L\)-algebras,
\[
 L\otimes_K L
   \cong \underset{i}{\operatorname{colim}}
        \prod_{\operatorname{Gal}(L_i/K)} L
   \cong \{\text{continuous functions }G\to L_{\mathrm{disc}}\}.
\]
At a finite stage the first isomorphism is the splitting of a separable Galois tensor product. For the second, a continuous function to the discrete field has finitely many clopen fibers by compactness, and a common finite-stage refinement detects all of them. Thus \(HL\) is normal over \(HK\). The Pierce image of the Amitsur nerve is the groupoid with one object and arrow group \(G\), and Theorem~\ref{thm:pierce-galois} classifies split algebras by continuous \(G\)-actions on Stone spaces, as in \cite[Chapter~3]{bj}. All tensor powers here are flat over \(K\), so passing to Eilenberg--Mac Lane spectra preserves their discreteness.

\chapter{Spectral Galois extensions and stable monodromy}\label{ch:spectral}

A categorical Galois theory depends on which objects count as covers. For a commutative ring spectrum, ordinary \'etaleness of degree zero together with the flat homotopy base-change formula gives one answer. A different answer asks for commutative algebra objects that become products of components of the unit after an effective extension of scalars. The second answer extends the split-algebra construction of \cite{bj} into stable homotopy theory. It permits torsors such as the complexification map in periodic topological \(K\)-theory, which fail the homotopy flatness test. The two theories are related by descent, with different input conditions.

\section{Three effectivity inputs}
\index{descendability}\index{nilpotence!tensor}

A \emph{stable homotopy theory} here means a presentable symmetric monoidal stable \(\infty\)-category \(\mathcal C\) in which the tensor product preserves colimits separately. Write \(\mathbf 1\) for its unit. An algebra \(B\in\operatorname{CAlg}(\mathcal C)\) is \emph{faithful} if \(B\otimes-\) is conservative. In a stable category this is equivalent to requiring \(B\otimes M\simeq0\) to imply \(M\simeq0\), by applying the test to cofibers. It is \emph{descendable} if the thick tensor ideal it generates contains \(\mathbf 1\). A thick tensor ideal is closed under finite cofiber sequences, retracts, and tensor products with any object of \(\mathcal C\).

The following criterion gives the difference between a conservative test and a finite proof of descent. Put \(I=\operatorname{fib}(\mathbf 1\to B)\), with its map \(u:I\to\mathbf 1\). Write \(u_n:I^{\otimes n}\to\mathbf 1\) for the tensor power of \(u\).

\begin{proposition}[Nilpotence criterion]\label{prop:spectral-nil}
The algebra \(B\) is descendable if and only if \(u_n\) is nullhomotopic for some \(n\geq 1\).
\end{proposition}

\begin{proof}
Say an object \(M\) belongs to \(\mathcal N\) if \(u_n\otimes M\) is nullhomotopic for some \(n\). It is closed under tensoring with an arbitrary object and under retracts. Suppose \(M_1\to M\to M_2\) is a cofiber sequence and powers \(n\) and \(m\) annihilate the terms \(M_1\) and \(M_2\), respectively. The map \(I^{\otimes m}\otimes M\to M\) becomes null after composition to \(M_2\), by naturality. It lifts along \(M_1\to M\). Tensor that lift with \(I^{\otimes n}\) and apply \(u_n\) on its target. The resulting composite factors through \(u_n\otimes M_1\), hence vanishes. By naturality this equals \(u_{n+m}\otimes M\). This shows extension closure, and suspension, desuspension and rotation of a cofiber sequence give the other two-out-of-three cases. Thus \(\mathcal N\) is thick. It contains \(B\) since multiplication splits \(B\to B\otimes B\) and hence \(B\otimes u\) is null. If \(B\) is descendable, \(\mathbf 1\in\mathcal N\).

Conversely, the cofiber of \(u_n\) is built from the cofibers in the chain
\[
 I^{\otimes n}\longrightarrow I^{\otimes(n-1)}
     \longrightarrow \cdots\longrightarrow I\longrightarrow\mathbf 1.
\]
These cofibers are \(I^{\otimes j}\otimes B\) for \(0\leq j<n\), since the arrows are tensorings of \(u\). Repeated use of the cofiber sequence for a composite puts the cofiber of \(u_n\) in the thick tensor ideal generated by \(B\). If \(u_n\) is null, \(\mathbf 1\) is a retract of that cofiber.
\end{proof}

A descendable algebra is faithful. For a fixed \(M\), the objects \(X\) satisfying \(X\otimes M\simeq 0\) form a thick tensor ideal, so if that ideal contains \(B\), it contains the unit. Descendability moreover implies effective Amitsur descent for \(\mathcal C\) as a symmetric monoidal category \cite[Proposition~3.22]{mathew}. The decisive step in the comonadicity proof concerns a cosimplicial diagram that becomes split after tensoring with \(B\). After this tensoring its partial-totalization tower is pro-constant. The objects \(X\) such that the tower for the diagram tensored by \(X\) is pro-constant form a thick tensor ideal. Finite limits commute with tensoring, and constant pro-objects form a stable full subcategory closed under retracts. Tensoring a pro-constant tower produces a pro-constant tower because tensoring defines a levelwise functor on pro-objects. Descendability puts the unit in the ideal. For a pro-constant tower, passage to the limit commutes with tensoring by \(B\). The dual form of Barr--Beck then gives comonadicity, and the iterated base-change diagram gives the Amitsur limit.

Two other effectivity inputs serve different purposes. Faithfulness plus dualizability suffices for comonadicity without an assumption about compactness of the unit, because tensoring with a dualizable object preserves all limits. When the unit is compact these two conditions also imply descendability \cite[Theorem~3.38]{mathew}. Flat faithful descent of ring spectra, used in Chapter~\ref{ch:algebra}, is still another criterion. Its hypotheses are expressed on all homotopy groups and it establishes the Amitsur equivalence directly. In particular, a proof of equivalence of module categories via conservative, limit-preserving tensoring is available without checking homotopy flatness.

\section{The Amitsur complex of a Rognes extension}
\index{Amitsur complex!coherent group action}\index{Galois extension!spectral}

Let \(G\) be a finite group and let \(A\to B\) be a map of \(E_\infty\)-rings with a coherent action of \(G\) on \(B\) by \(A\)-algebra maps. Its characteristic comparisons are
\begin{equation}\label{eq:rognes-galois-comparisons}
 A\longrightarrow B^{hG},
 \qquad
 \gamma:B\otimes_A B\longrightarrow\prod_{g\in G}B,
 \qquad
 \gamma_g=\mu\circ(1\otimes g).
\end{equation}
Here \(B^{hG}\) is the homotopy limit of the action diagram over \(BG\), computed in ring spectra, and \(\mu\) is multiplication. If both maps are equivalences, the extension is \(G\)-Galois in the sense of Rognes. It is \emph{faithful} if \(B\) is faithful as an \(A\)-module \cite[Sections~4 and~6]{rognes}. The faithfulness condition matters to the following categorical comparison. The fixed-point equation only describes the algebra of invariants, while descent must detect every module.

\begin{lemma}[Bar description of the tensor powers]\label{lem:spectral-bar}
If \(\gamma\) is an equivalence, there is an equivalence of cosimplicial algebras
\[
 B^{\otimes_A(\bullet+1)} \simeq
 \left([n]\longmapsto\prod_{(g_1,\ldots,g_n)\in G^n} B\right),
\]
where the right side is the cochain diagram of the \(G\)-action. The degree-\(n\) component at \((g_1,\ldots,g_n)\) is the map that multiplies the factors after twisting factor \(i\) by \(g_1\cdots g_i\), with the first factor untwisted.
\end{lemma}

\begin{proof}
For fixed \(n\), use the first factor as a common \(B\)-base and write the tensor power as the \(n\)-fold tensor product over \(B\) of \(B\otimes_A B\). Apply \(\gamma\) to each factor. Finite products of \(B\)-algebras are taken to finite products by base change, as can be checked on underlying modules, where finite products are also coproducts. This gives an equivalence indexed by independent elements \((h_1,\ldots,h_n)\), with factor \(i\) twisted by \(h_i\). Reindex by \(h_i=g_1\cdots g_i\), a bijection of indexing sets.

For the first coface, the inserted unit in factor zero makes every later factor share the twist \(g_1\). This is the coface applying \(g_1\) after deleting it from the coordinate tuple. An interior insertion replaces adjacent group entries by their product. Insertion at the final position drops the last entry. A codegeneracy multiplies two adjacent factors and is identified with inserting the group identity at that place. These formulas are equalities for strict actions and are identities of composites of the multiplication and the action structure maps for a coherent action. Equivalently, the comparison is induced from the map of simplicial diagrams from the action groupoid of \(G\) on the affine represented by \(B\) to its affine \Cech{} nerve, where an arrow and its first object determine the second. This construction gives the whole family with its coherences.
\end{proof}

\begin{theorem}[Stable Galois descent]\label{thm:spectral-galois-descent}
For a faithful \(G\)-Galois extension of \(E_\infty\)-rings, extension of scalars induces symmetric monoidal equivalences
\[
 \operatorname{Mod}_A\simeq(\operatorname{Mod}_B)^{hG},
 \qquad
 \operatorname{CAlg}_A\simeq(\operatorname{CAlg}_B)^{hG}.
\]
The group acts on the categories on the right by extension of scalars along its automorphisms of \(B\). A module or algebra there has semilinear transport with respect to the action on its scalars.
\end{theorem}

\begin{proof}
Rognes's dualizability theorem makes \(B\) a dualizable \(A\)-module \cite[Proposition~6.2.1]{rognes}. Tensoring with it is therefore both conservative and limit-preserving. To check the latter, regard tensoring as an endofunctor of \(\operatorname{Mod}_A\) and use its left adjoint, tensoring with the dual \(B^\vee\). Limits of \(B\)-modules are computed on their underlying \(A\)-modules, so extension of scalars also preserves them. By comonadicity, the Amitsur module comparison is an equivalence. The \Cech{} comparison is symmetric monoidal, since every extension of scalars is, and the monoidal operations in the limit are pointwise.

Apply Lemma~\ref{lem:spectral-bar}. Modules over \(\prod_J B\) with \(J\) finite are products of modules over the factors. This follows by projecting to the orthogonal central idempotents, giving an inverse to taking the direct sum with its componentwise action. The first coface on the resulting category-valued bar construction is the twist by \(g_1\), the middle cofaces compose group elements, and the last drops the last element. Thus its limit is the category of coherent semilinear descent data \((\operatorname{Mod}_B)^{hG}\). Passing to commutative algebra objects in the symmetric monoidal equivalence proves the second assertion.
\end{proof}

A fixed-point description follows at the level of individual objects. If \(D\) is a semilinear \(B\)-algebra in the right-hand category, its inverse image is the homotopy fixed-point algebra of the underlying action, under \(A\). Indeed the right adjoint to extension of scalars with canonical descent data is formed by the limit of the forgetful diagram, including the transport. This is seen by taking mapping spaces out of any \(A\)-algebra and applying extension--restriction adjunction in every degree. The equivalence in Theorem~\ref{thm:spectral-galois-descent} identifies this right adjoint with the inverse. The module statement holds by the same argument. Thus reconstructing the underlying algebra or module uses the full homotopy limit even when a classifying category of special algebras has discrete maps.

\section{Intermediate split algebras}
\index{homotopy fixed points!intermediate algebra}

Suppose \(\pi_0 B\) is nonzero and has only trivial idempotents. A \(B\)-algebra of the form \(B^T=\prod_{t\in T} B\), with \(T\) a finite set, has a discrete space of maps to another such algebra \(B^U\). By Lemma~\ref{lem:pierce-split-products}, that space is the set of maps \(U\to T\). Extension of scalars along each \(g:B\to B\) canonically carries \(B^T\) to a copy of \(B^T\), since it preserves these products and takes the unit to the unit. Every coherent semilinear structure on \(B^T\) then amounts to a genuine action on \(T\), by contravariance and discreteness. In formulas one may take the action on functions to be \((g\varphi)(t)=g(\varphi(g^{-1}t))\).

\begin{corollary}\label{cor:spectral-intermediate}
Under the hypotheses of Theorem~\ref{thm:spectral-galois-descent} and connectedness of \(\pi_0 B\) as above, the opposite of the category of \(A\)-algebras becoming \(B^T\) for finite \(T\) is equivalent to the category of finite \(G\)-sets. Connected objects equipped with a map of algebras to \(B\) correspond to pointed transitive \(G\)-sets, hence to subgroups \(H\leq G\). For \(T=G/H\) with point \(H\), the associated intermediate algebra is \(B^{hH}\).
\end{corollary}

\begin{proof}
The full subcategory in question is preserved by the equivalence of Theorem~\ref{thm:spectral-galois-descent} and is characterized after extension by the discrete products just described. The object \(B\) corresponds to the regular \(G\)-set by \eqref{eq:rognes-galois-comparisons}. Hence a map of algebras to \(B\) gives a \(G\)-equivariant map from the regular \(G\)-set, which is specified by its value at the identity. A nonempty split object is connected precisely when the action is transitive, since a nontrivial product decomposition is exactly a disjoint union of invariant subsets. This also agrees with indecomposability of the underlying algebra. An idempotent would give a product decomposition, and each factor would remain split after base change because the idempotents of \(B^T\) are the characteristic functions of subsets of \(T\). For transitive \(T\), the stabilizer of the point yields the unique pointed isomorphism with \(G/H\). The fixed-point construction for \(B^{G/H}\) is a limit over the action groupoid \(G\ltimes(G/H)\) with value \(B\) at each object and transport induced by the group action. This groupoid is equivalent to \(BH\) by inclusion of the stabilizer at \(H\). Its limit is \(B^{hH}\). Under the same identification, the map to \(B\) is evaluation at the selected object.
\end{proof}

The correspondence is formulated in the category of the split algebras and their maps. In spectral algebra there is generally no useful criterion for an intermediate object that simply asks for injectivity of a map on every homotopy group. Connected split objects have a categorical role inside the category defined by base change and decomposition into components. The stabilizer is read from that category.


\begin{corollary}[Normal subgroups]\label{cor:spectral-normal}
In the situation of Corollary~\ref{cor:spectral-intermediate}, if \(H\) is normal in \(G\) then \(A\to B^{hH}\) is a faithful \((G/H)\)-Galois extension.
\end{corollary}

\begin{proof}
Set \(Q=G/H\) and \(D=B^{hH}\). The right translations of \(Q\) on \(G/H\) commute with the left \(G\)-action. Contravariance transports them to an action by \(A\)-algebra maps on \(D\). After tensoring with \(B\), the action is the regular permutation action on \(\prod_Q B\). The torsor tensor-square comparison becomes the canonical equivalence obtained from the bijection between pairs of elements of \(Q\) and a point of \(Q\) together with a translation. The operation \(B\otimes_A-\) detects equivalences and commutes with limits on modules, hence also on algebras. The fixed-point comparison after that operation reduces to \(B\simeq(\prod_Q B)^{hQ}\), which holds because the translation action groupoid on \(Q\) is contractible. Finally if \(D\otimes_A M\simeq 0\), then tensoring with \(B\) makes the finite product of copies of \(B\otimes_A M\) zero. Since \(Q\) is nonempty and \(B\) is faithful, \(M\simeq0\).
\end{proof}

\section{The stable Galois category and its range}
\index{Galois category!stable homotopy theory}\index{finite cover!weak}

Mathew's finite-cover construction makes the preceding restricted classification intrinsic to a stable homotopy theory \cite[Definitions~6.1--6.2 and Theorem~6.5]{mathew}. The Boolean algebra of idempotents of \(\pi_0\operatorname{End}_{\mathcal C}(\mathbf 1)\) plays the role of the Pierce components of the unit. Given a commutative algebra object \(C\), its localization \(C[e^{-1}]\) at an idempotent of the endomorphism ring of the free rank-one \(C\)-module is the corresponding direct factor in its module category. The localization can be constructed using the telescope of multiplication by \(e\). It is idempotent under relative tensor product and corresponds to the projection onto the summand where \(e\) acts as the identity.

An algebra \(D\in\operatorname{CAlg}(\mathcal C)\) is called a finite cover if some descendable \(C\) makes \(C\otimes D\) a finite product of the algebras \(C[e^{-1}]\), with independently chosen idempotents for the factors. A weak finite cover is defined in the same way with \(C\otimes-\) conservative and limit-preserving. The idempotents need not partition the unit because several sheets may meet the same component. After a finite Boolean refinement into pairwise disjoint clopens, the product describes a finite number of copies on each clopen component.

The opposite of either finite-cover category is a Galois category in the general sense allowing several components, classified by a profinite groupoid \cite[Theorem~5.36]{mathew}. In particular the covers have discrete mapping spaces in their ambient \(\infty\)-category. The two categories agree when the unit is compact. For \(\mathcal C=\operatorname{Mod}_A\) this compactness is automatic, since mapping out of \(A\) is computed on the underlying spectra and commutes with filtered colimits. For a connective \(A\), Mathew's comparison theorem identifies these covers with the finite \'etale algebras of \(\pi_0 A\), through their spectral \'etale lifts \cite[Theorem~6.17]{mathew}. This assertion requires more than the lifting theorem of Chapter~\ref{ch:algebra}. It says the tensor-categorical cover condition detects exactly the objects which the lifting theorem starts with in the connective case.

The effect of the definition on torsors can be checked in one direction directly. A faithful Rognes extension \(A\to B\) is dualizable and has \(B\otimes_A B\simeq\prod_G B\). Thus \(B\) is its own weak finite-cover splitting witness. In \(\operatorname{Mod}_A\), compactness of \(A\) upgrades this witness to descendability. Conversely, a \(G\)-torsor in the weak Galois category is a faithful \(G\)-Galois extension, and in the strong category it is a descendable one \cite[Proposition~6.13 and Corollary~6.15]{mathew}. A local trivialization checks the tensor-square comparison and faithfulness by conservativity. The local tensor-square map is the map from a trivial torsor to its product with the group. To check fixed points in the weak case, tensoring by the witness commutes with the fixed-point limit. In the strong case, recover the torsor algebra by Amitsur descent along the witness, then commute its fixed-point limit with that Amitsur limit. At every Amitsur stage the torsor stays trivial and its fixed points are the stage algebra, since the translation action groupoid of the regular action is contractible. The Amitsur limit of these stage algebras recovers the unit.

\section{Periodic rings and higher local systems}
\index{local system!stable modules}

Rognes proves that periodic complexification \(KO\to KU\), with the complex-conjugation action, is a faithful \(C_2\)-Galois extension \cite[Proposition~5.3.1]{rognes}. Its faithfulness can also be seen from the cofiber sequence
\[
 \Sigma KO \xrightarrow{\eta} KO\longrightarrow KU.
\]
If \(KU\otimes_{KO}M\) vanishes, the image of the stable Hopf element \(\eta\) acts invertibly as a degree-one endomorphism of \(M\). A sufficiently high power of \(\eta\) is zero, so this invertibility forces \(M\simeq0\). The map on \(\pi_0\) is the identity of \(\mathbf Z\), while \(\pi_1 KO\cong\mathbf Z/2\) and \(\pi_1 KU=0\). Thus this extension fails spectral flatness, yet gives module and algebra descent by Theorem~\ref{thm:spectral-galois-descent}. It lies in the stable finite-cover category. This explains why the connective hypothesis of the comparison with geometric finite \'etale algebras cannot simply be removed.

The finite Galois groupoid of a stable category can miss homotopy visible to its module objects. Consider \(X=K(\mathbf Z,2)\), an algebraically closed characteristic-zero field \(k\), and the pointwise tensor category \(\mathcal C=\operatorname{Fun}(X,\operatorname{Mod}_{Hk})\). Its commutative algebras are functors to \(\operatorname{CAlg}_{Hk}\). Evaluation preserves finite products, localization at idempotents, and the thick tensor ideal operations defining descendability. Every finite cover of \(\mathcal C\) therefore takes values in finite covers of \(Hk\), which are finite products of \(Hk\). Its image lands in the maximal groupoid of these products, a \(1\)-truncated space, and \(X\) is simply connected. Thus each such cover is constant up to equivalence. Conversely a constant product is already split by the unit. Maps between the constant products are the opposites of ordinary maps of finite sets, since \(X\) is connected and the value-wise mapping spaces are discrete. The finite Galois groupoid of \(\mathcal C\) is trivial.

For the fiber \(M=Hk\oplus\Sigma Hk\), its endomorphism spectrum decomposes into two copies of \(Hk\), one of \(\Sigma Hk\), and one of \(\Sigma^{-1}Hk\). The space \(\operatorname{Aut}(M)\) is a union of components of its endomorphism space, so its identity component has \(\pi_1\cong(k,+)\) and no higher homotopy. The simply connected cover of \(B\operatorname{Aut}(M)\) is therefore \(K(k,2)\). A nonzero additive homomorphism \(\mathbf Z\to k\) gives a map
\[
 K(\mathbf Z,2)\longrightarrow K(k,2)
                    \longrightarrow B\operatorname{Aut}(M)
\]
which is nonzero on \(\pi_2\). It classifies a nonconstant \(Hk\)-module local system. Its higher transport is invisible to the stable finite Galois groupoid, as happens for the fundamental progroupoid and the full shape in the space-valued theory.

\part{Reflections and factorization}
\chapter{Reflective Galois structures and factorization}
\label{ch:reflections}
\index{Galois structure}\index{reflection!Galois}

In Borceux and Janelidze's categorical Galois theorem, triviality is measured by a slice adjunction and splitting is found by changing the base \cite[Chapter~5]{bj}. The higher analogue needs two independent ingredients. A slice adjunction specifies the chosen class of trivial objects. Effective descent says which objects and maps over a cover glue across its entire \v{C}ech nerve. Separating these ingredients makes it possible to use the same argument with rings, group objects, or spaces while stating their individual descent hypotheses explicitly.

\section{The adjunction seen from a varying base}

Let $L\colon\mathcal C\rightleftarrows\mathcal D\colon R$ be an adjunction of $\infty$-categories with unit $\eta$ and counit $\epsilon$. Assume both categories admit pullbacks. Over $B\in\mathcal C$, use the functors
\[
\begin{aligned}
 L_B&\colon\mathcal C_{/B}\longrightarrow\mathcal D_{/LB},&
 L_B(X\xrightarrow{f}B)&=(LX\xrightarrow{Lf}LB),\\
 R_B&\colon\mathcal D_{/LB}\longrightarrow\mathcal C_{/B},&
 R_B(D\xrightarrow{d}LB)&=(B\times_{RLB}RD\longrightarrow B).
\end{aligned}
\]
The pullback uses $\eta_B$. All slices contain the full mapping spaces over the indicated bases.

\begin{proposition}[Slice adjunction and base change]
\label{prop:relative-slice-adjunction}
There are adjunctions $L_B\dashv R_B$. For every $h\colon B'\to B$ there is a natural equivalence
\[
 h^*R_B\simeq R_{B'}(Lh)^*.
\]
These equivalences are coherent for chains of base changes. If $\Sigma_h$ denotes postcomposition, there is also an equivalence $L_B\Sigma_h\simeq\Sigma_{Lh}L_{B'}$.
\end{proposition}

\begin{proof}
Over $B$, a map from $X$ to $B\times_{RLB}RD$ is the same as a map $X\to RD$ with a specified identification of the composite to $RLB$ with $\eta_Bf$. The original adjunction identifies the space of such maps with the homotopy fiber
\[
 \operatorname{hofib}_{Lf}\big(
 \operatorname{Map}_{\mathcal D}(LX,D)
 \longrightarrow \operatorname{Map}_{\mathcal D}(LX,LB)\big).
\]
This is the desired mapping space in $\mathcal D_{/LB}$. The unit at $f$ has underlying map $X\to B\times_{RLB}RLX$ induced by $(f,\eta_X)$. To check base change, use the fact that $R$ preserves pullbacks and substitute
\[
 R(LB'\times_{LB}D)\simeq RLB'\times_{RLB}RD
\]
in the formula for $R_{B'}$. Naturality of $\eta$ and pasting of pullbacks identify the result with $B'\times_{RLB}RD$. Iterated identifications all arise from the limit of the same diagram, so their coherence follows from the universal cones. The assertion about postcomposition follows directly from functoriality of $L$.
\end{proof}

For a restricted Galois structure, let $\mathcal E$ in $\mathcal C$ and $\mathcal F$ in $\mathcal D$ be classes of maps containing equivalences and closed under composition and pullback. Write $\mathcal C_{\mathcal E/B}$ for the full subcategory of $\mathcal C_{/B}$ spanned by maps in $\mathcal E$, and similarly for $\mathcal D_{\mathcal F/LB}$. If $L$ takes $\mathcal E$ to $\mathcal F$ and $R$ takes $\mathcal F$ to $\mathcal E$, the displayed slice adjunction and its base-change equivalences restrict to these full subcategories. The full-subcategory condition matters. Maps between the selected objects over a base retain their higher morphisms whether or not their underlying arrows belong to $\mathcal E$.

\index{admissibility}\index{splitting!relative}
\begin{definition}[Admissibility and splitting]
The structure is admissible when the restricted $R_B$ is fully faithful at each base. A map over $B$ is \emph{trivial} if the unit of the restricted slice adjunction at that map is an equivalence. A map over $B$ is \emph{split by} $p\colon U\to B$ if its pullback to $U$ is trivial. It is a \emph{covering} relative to a specified class of descent maps if one of those maps splits it.
\end{definition}

Full faithfulness is equivalent to invertibility of the slice counit. It makes the full subcategory $\operatorname{Triv}(B)$ of trivial maps the essential image of $R_B$. A map is trivial exactly when its unit square
\[
\begin{array}{ccc}
 X&\xrightarrow{\eta_X}&RLX\\
 \downarrow&&\downarrow\\
 B&\xrightarrow{\eta_B}&RLB
\end{array}
\]
is a pullback. Proposition~\ref{prop:relative-slice-adjunction} shows that trivial maps are stable under base change. Their defining square is computed using the global unit, but the proof of stability uses the right slice adjoints. There is no need to assume that $L$ preserves all pullbacks.

\begin{lemma}[A practical admissibility criterion]
If $R$ is fully faithful, the unrestricted structure is admissible precisely when $L$ preserves every pullback of the form
\[
 B\times_{RLB}RD
\]
where the first arrow is $\eta_B$ and the other is the image of $D\to LB$. In this assertion a preserved pullback has comparison map $L(B\times_{RLB}RD)\to D$.
\end{lemma}

\begin{proof}
Under the counit equivalences $LRD\simeq D$ and $LRLB\simeq LB$, the image of the cospan has first arrow equivalent to the identity of $LB$ by the triangle identity. Its limit is $D$. The comparison is precisely the counit of $L_B\dashv R_B$. Invertibility of these counits for all $B,D$ is the assertion.
\end{proof}

Admissibility can be checked with a smaller class $\mathcal F$ when the problem concerns only extensions. For instance, a reflection may be useful on slices of surjections even if its behavior on slices of all arrows has not been established. Effective descent on the selected slices requires both unrestricted descent and detection of the selected arrow condition, or a descent proof directly on the selected slices.

\section{The comparison over one descent map}

For $p\colon U\to B$ in $\mathcal E$, let $U_\bullet$ be its \v{C}ech nerve, with $U_k=U^{\times_B(k+1)}$. Use the pullback functors along its simplicial structure maps to form a cosimplicial diagram of restricted slices. Say that $p$ is an effective descent map for $\mathcal E$ if the comparison
\[
 \mathcal C_{\mathcal E/B}\longrightarrow
 \lim_{[k]\in\Delta}\mathcal C_{\mathcal E/U_k}
\]
is an equivalence. When every map is allowed and $\mathcal C$ is an $\infty$-topos, every effective epimorphism satisfies this condition by the van Kampen property of its \v{C}ech realization \cite[Section~6.1]{htt}. More generally, if unrestricted descent holds and membership in $\mathcal E$ is detected after pullback along $p$, restricting the equivalence gives the indicated condition.

\begin{theorem}[Relative higher Galois comparison]
\label{thm:relative-higher-galois}
In an admissible structure, let $p\colon U\to B$ be an effective descent map for $\mathcal E$. The full subcategory $\operatorname{Split}(p)$ of maps split by $p$ satisfies
\[
 \operatorname{Split}(p)\simeq
 \lim_{[k]\in\Delta}\mathcal D_{\mathcal F/LU_k}.
\]
The diagram on the right is formed with pullbacks along maps of $L(U_\bullet)$.
\end{theorem}

\begin{proof}
Restrict the unrestricted comparison for the selected slices to the full subcategories $\operatorname{Triv}(U_k)$. A point of the limit of the slices is a coherent family whose object in any degree $k$ is equivalent to the pullback of the degree-zero object along any vertex. Consequently, if its object in degree zero is trivial, all its objects are trivial. Conversely a family consisting of trivial objects has a trivial degree-zero object. Limits of diagrams of full subcategories are full in the limit of the ambient diagrams. These observations give the full essential image of $\operatorname{Split}(p)$ and preserve its mapping spaces. Finally the equivalences $R_{U_k}\colon\mathcal D_{\mathcal F/LU_k}\simeq\operatorname{Triv}(U_k)$ respect restriction by Proposition~\ref{prop:relative-slice-adjunction}. Taking their limit gives the result.
\end{proof}

At degree one the family has an equivalence between two pullbacks of a single object over $LU$. At degree two it gives a specified homotopy between composite gluing and direct gluing. The comparison retains the higher data as well. An ordinary category of strict descent isomorphisms is sufficient in a nerve of an ordinary category, where the mapping spaces are discrete. In that case the degree-two relation is an equality and determines all the higher compatibility for an object. In categories with nondiscrete mapping spaces the totalization is intrinsic to the assertion.

There are two distinct possible simplifications of this comparison. If $L$ preserves the pullbacks that form $U_\bullet$, the simplicial object $L(U_\bullet)$ is the \v{C}ech nerve of $Lp$. If it further presents an effective groupoid in a suitable category $\mathcal D$, the totalization can be expressed as its category of actions. If unrestricted descent also holds for $Lp$ in $\mathcal D$, then the result is the entire unrestricted slice over $LB$. Each step uses a separate property. Applying a reflection termwise to a \v{C}ech nerve is meaningful even if its image fails the groupoid Segal conditions, and the comparison theorem still holds in the form given above.

\index{descent!monadic}
\begin{proposition}[The monadic comparison]
Suppose that $p^*$ on the selected slices is monadic and admits the restricted left adjoint $\Sigma_p$. Let $T=p^*\Sigma_p$ be its monad on $\mathcal C_{\mathcal E/U}$. If $T$ sends trivial objects to trivial objects, it restricts to a monad $T_{\mathrm{triv}}$, and
\[
 \operatorname{Split}(p)\simeq
 \operatorname{Alg}_{T_{\mathrm{triv}}}(\operatorname{Triv}(U)).
\]
\end{proposition}

\begin{proof}
The full subcategory of trivial objects contains every iterate of $T$ on any one of its objects. Hence the unit and multiplication of the monad restrict to it with their coherences. A coherent algebra structure on such an object uses only these iterates and maps between them. Full faithfulness gives the same spaces for the operations, compatibility diagrams, and algebra morphisms in the smaller and larger categories. The restricted algebra category is thus the full subcategory of $\operatorname{Alg}_{T}(\mathcal C_{\mathcal E/U})$ on algebras whose underlying objects are trivial. By monadic comparison, underlying objects correspond to pullbacks of objects over $B$. This is exactly the splitting condition.
\end{proof}

The stability hypothesis is equivalent to requiring that $\Sigma_p R_U(D)$ be split by $p$ for every allowed $D\to LU$. This recovers the extra condition in the monadic Galois-descent version of the ordinary theorem \cite[Section~5.1]{bj}. The \v{C}ech comparison uses a different stated hypothesis, namely an effective descent comparison to the full cosimplicial diagram. This formulation avoids assuming that every monad arising from a chosen class already has an identified groupoid model.

\begin{proposition}[Change of splitting cover]
Suppose the specified descent maps are stable under pullback and composition. Coverings are stable under base change. If $q\colon V\to B$ is a specified descent map and $q^*f$ is a covering over $V$, then $f$ is a covering whenever $f$ is allowed. If $V\to U$ is a map over $B$, then $\operatorname{Split}(p)\subseteq\operatorname{Split}(V\to B)$.
\end{proposition}

\begin{proof}
For a new base $B'\to B$, pull back a splitting map $U\to B$ and use stability of triviality. If $W\to V$ splits $q^*f$, its composite $W\to B$ is a specified descent map that splits $f$. The refinement statement follows from the same stability of triviality. All assertions concern full subcategories and respect the coherent pullback functors between their comparisons.
\end{proof}

An allowed effective descent map $p$ is \emph{normal} for this structure when $p^*p$ is trivial. This belongs to the relative terminology. A normal splitting cover is a descent map that splits itself, and its pullbacks are again normal whenever the descent class is stable under pullback. The definition has no dependence on a choice of geometric point or separable closure. Chapter~\ref{ch:central} relates it to centrality in the ordinary group calculation.

\section{Two factorization systems arising from truncation}

Return to a reflection, using all arrows and identifying $R$ with inclusion. Write $L$ also for the endofunctor $RL$ under this convention. Let $\mathcal M$ be the trivial maps and $\mathcal W$ the maps inverted by $L$.

Every orthogonality assertion below concerns a contractible space of fillers in every coherent commutative square. Equivalently, $a\colon A\to B$ is orthogonal to $m\colon X\to Y$ if the natural map
\[
 \operatorname{Map}(B,X) \longrightarrow
 \operatorname{Map}(A,X)\times_{\operatorname{Map}(A,Y)}
 \operatorname{Map}(B,Y)
\]
is an equivalence. This definition retains the space of commutativity data of a square as well as the space of diagonals.

\index{factorization system!reflective}\index{orthogonality}
\begin{theorem}[Admissible reflective factorization]
\label{thm:admissible-reflective-factorization}
For an unrestricted admissible reflection there is an orthogonal factorization system $(\mathcal W,\mathcal M)$ with factorization
\[
 A\longrightarrow B\times_{LB}LA\longrightarrow B
 \qquad (f\colon A\to B).
\]
The factorizations are functorial and $\mathcal W$ is stable under pullback along $\mathcal M$.
\end{theorem}

\begin{proof}
The right arrow is $R_B(L_Bf)$. Admissibility makes its $L$-image the arrow $LA\to LB$ up to equivalence, with counit inverse to the image of the left arrow by the triangle identity. Hence the left arrow lies in $\mathcal W$. To see orthogonality, take a lifting problem against $X=Y\times_{LY}LX\to Y$. For a fixed bottom arrow $B\to Y$, lifts amount to compatible maps $B\to LX$ over $LY$. For $a\in\mathcal W$, restriction along $a$ is an equivalence on maps into either of the local objects $LX,LY$. It is therefore an equivalence on their homotopy fibers over the prescribed composite to $LY$, proving orthogonality. The class $\mathcal W$ is closed under retracts, and so is $\mathcal M$, since the natural comparison from an object to its slice reflection carries retract diagrams to retract diagrams. The usual orthogonality-retract argument now identifies each class with the full orthogonal of the other. Functoriality follows from the units and pullback universal properties.

For the stability statement let $m\colon C\to B$ be trivial and $a\colon A\to B$ inverted. Write $C\simeq B\times_{LB}D$ for $D\in\mathcal D_{/LB}$. Then $A\times_BC\simeq A\times_{LA}(LA\times_{LB}D)$. Applying admissibility at $A$ and $B$ gives the comparison $LA\times_{LB}D\to D$ under $L$, an equivalence.
\end{proof}

\index{stable units}Stable units mean that $L$ preserves pullbacks over every local object in the ambient category, including pullbacks in which neither of the objects over that base is local. This implies admissibility by the practical criterion. Left exactness requires finite-limit preservation over arbitrary bases and is stronger. If $L$ is left exact, $\mathcal W$ is stable under arbitrary pullback. Conversely, if $\mathcal C$ has finite limits and $\mathcal W$ is pullback stable, $L$ is left exact. Indeed the reflection fixes the terminal object, since it is local. For a pullback $A\times_BC$, the map to $A\times_{LB}C$ is a pullback of the diagonal of $B\to LB$. That diagonal lies in $\mathcal W$ by pullback stability and two-out-of-three, since either projection $B\times_{LB}B\to B$ lies in $\mathcal W$ and the composite with the diagonal is the identity. The map
\[
 A\times_{LB}C\longrightarrow LA\times_{LB}LC
\]
is the composite of pullbacks of the two units, so is also in $\mathcal W$. The final object is local, since the right adjoint preserves limits. This proves preservation of pullbacks.

\begin{corollary}[Left-exact localizations]
Let $L\colon\mathcal X\to\mathcal Y$ be an accessible left-exact localization of an $\infty$-topos, and allow all arrows. For any effective epimorphism $p\colon U\to B$, a map over $B$ is split by $p$ if and only if it is already trivial.
\end{corollary}

\begin{proof}
The localized category $\mathcal Y$ is an $\infty$-topos. Left exactness identifies $L(U_\bullet)$ with the \v{C}ech nerve of $Lp$. Since $L$ is a left adjoint, it preserves the colimit of the original nerve, so $Lp$ is also an effective epimorphism. The right-hand side of Theorem~\ref{thm:relative-higher-galois} is therefore equivalent to $\mathcal Y_{/LB}$ by effective descent there. Under that equivalence the trivial object $R_B(D)$ corresponds to the descent datum of $D$, since the base-change formula of Proposition~\ref{prop:relative-slice-adjunction} identifies its restriction in every degree. Thus the equivalence on slices realizes the inclusion of the trivial objects into the split objects as an equivalence.
\end{proof}

\index{truncation!relative}\index{modality!truncation}
\begin{definition}[Relative truncation]
For $n\geq -1$ and an $\infty$-topos $\mathcal X$, the relative $n$-truncation of $f\colon A\to B$ is the reflection of $A$ into the full subcategory of $n$-truncated objects in $\mathcal X_{/B}$. Write its total object as $\tau_{\leq n}^{B}A$. A map is $n$-connected if its relative $n$-truncation is terminal in its target slice.
\end{definition}

The pair of $n$-connected and $n$-truncated maps is an orthogonal factorization system in an $\infty$-topos, with both classes stable under pullback \cite[Sections~5.5.6 and~6.5.1]{htt}. A lifting problem with an $n$-truncated right arrow over $B$ depends only on the relative $n$-truncation of the left arrow by the slice universal property. For $A\to B$, the first map in the relative truncation factorization is $n$-connected, and the second is $n$-truncated. The factorization and lifting conditions are often called the truncation modality \cite{modalities}. The indexing here pairs $n$-connected maps with $n$-truncated maps, and an effective epimorphism is $(-1)$-connected.

Global $n$-truncation, with codomain the $n$-truncated objects, has stable units. Indeed, given $A\to D\leftarrow C$ with $D$ $n$-truncated, factor the comparison
\[
 A\times_D C\longrightarrow \tau_{\leq n} A\times_D \tau_{\leq n} C
\]
as two pullbacks of $n$-connected units. The target is $n$-truncated because the full subcategory of truncated objects is closed under limits, and the composite is $n$-connected. This identifies the comparison with a truncation. The reflective factorization of Theorem~\ref{thm:admissible-reflective-factorization} therefore applies. Its intermediate object
\[
 B\times_{\tau_{\leq n}B}\tau_{\leq n}A
\]
is constructed from the global truncations. Relative truncation uses $\tau_{\leq n}^{B}A$. They have different universal properties over the unreplaced $B$.

\begin{example}[Relative components of paths]
Take $\mathcal X=\SSet$, $B=S^1$ and $A=*$ with the basepoint. The global $0$-truncations of $A$ and $B$ coincide, so the global reflective factorization has intermediate object $S^1$. The fiber of $*\to S^1$ over the basepoint is $\Omega S^1\simeq\mathbb Z$, where $\mathbb Z$ is discrete. Thus $*\to S^1$ is already $0$-truncated over $S^1$, and the relative factorization has intermediate object $*$. Since this fiber has infinitely many components, the map cannot be $0$-connected. Pullback to the basepoint also exhibits failure of pullback stability for the maps inverted by global $0$-truncation.
\end{example}

\section{Localizing the reflective pair}

For global $n$-truncation, its trivial maps are the base changes of maps between $n$-truncated objects. Their fibers are $n$-truncated. When the base admits a cover by an $n$-truncated object, the converse holds after passing to that cover. This includes every base space for $n\geq0$, since the coproduct of basepoints of its components gives a discrete effective epimorphism.

\begin{proposition}[Local triviality and stable inversion]
\label{prop:localizing-reflective-pair}
Fix $n\geq-1$ in an $\infty$-topos $\mathcal X$. Every $n$-connected arrow is inverted by global $n$-truncation after arbitrary pullback. Every arrow that becomes trivial for global $n$-truncation over an effective epimorphism is $n$-truncated. Over a base admitting an effective epimorphism from an $n$-truncated object, both converses hold.
\end{proposition}

\begin{proof}
If $f$ is $n$-connected, its pullbacks are $n$-connected. For any such pullback $a\colon X\to Y$ and any globally $n$-truncated object $D$, the projection $Y\times D\to Y$ is an $n$-truncated arrow. Orthogonality then identifies maps $X\to D$ with maps $Y\to D$. Global truncation thus inverts $a$.

For the reverse implication over a base of the indicated kind, write $f$ as $A\xrightarrow{e}T\xrightarrow{m}B$ in the relative $n$-connected, $n$-truncated factorization. If every pullback of $f$ is inverted by global truncation, the same holds for $m$ since it holds for $e$. Pull back to an effective epimorphism $U\to B$ with $U$ $n$-truncated. The pullback of $m$ has $n$-truncated domain and codomain, so inversion by global truncation makes it an equivalence already. Then $m$ is an equivalence by effective descent, proving $f$ is $n$-connected.

Trivial arrows are $n$-truncated since relative truncation is stable under pullback. This property descends along effective epimorphisms by induction on the truncation of the relative diagonal, starting with descent of equivalences. Descent of equivalences is immediate from the slice comparison over the \v{C}ech nerve. Finally, if $U\to B$ is an effective epimorphism with $U$ $n$-truncated and $f$ is $n$-truncated, then $U\times_B A$ is $n$-truncated by composition with $U\to 1$. Both domain and base of the new arrow are local, so its unit square is a pullback.
\end{proof}

\index{monotone--light factorization}Over spaces the last result gives a homotopical form of monotone--light factorization. For $n=0$, the left class consists of maps with nonempty connected homotopy fibers, and the right class consists of maps with discrete homotopy fibers. This concerns the homotopy theory of spaces. The original result concerns compact Hausdorff spaces and ordinary point inverses \cite[Section~5.8]{bj}. Those settings have different base change, and passing from a topological map to its homotopy type need not preserve its point inverse. The corresponding constructions agree in the formal operation of testing connectedness after base change and testing the right class by local splitting.

\chapter{Central extensions and higher bands}
\label{ch:central}
\index{central extension}

Categorical centrality in the ordinary group calculation arises by splitting for abelianization. A higher central extension also has an interpretation as a principal bundle over a delooping, where its band keeps track of all the homotopy of the kernel. The constructions are compatible for discrete groups. To compare them in higher degrees, it is useful to prove the ordinary result first and to keep the higher notion of a band explicit.

\section{The ordinary abelianization comparison}\index{abelianization!ordinary}

Use $E_{\mathrm{ab}}=E/[E,E]$ and write $a_E\colon E\to E_{\mathrm{ab}}$. For groups, admit surjections in both the source and target of the abelianization reflection. Its slice right adjoint over $G$ takes a surjection $D\to G_{\mathrm{ab}}$ of abelian groups to $G\times_{G_{\mathrm{ab}}}D\to G$. The following computation works for this class and makes the admissibility criterion concrete.

\begin{lemma}[Abelianization over a discrete base]
\label{lem:abelianization-admissible}
Let $P=G\times_{G_{\mathrm{ab}}}D$, with $D\to G_{\mathrm{ab}}$ surjective and $D$ abelian. Then
\[
 [P,P]=[G,G]\times\{1\}\qquad\text{and}\qquad P_{\mathrm{ab}}\cong D.
\]
In particular, abelianization is admissible on these surjective slices.
\end{lemma}

\begin{proof}
The two projections from $P$ are surjective. Every commutator in $P$ has second coordinate $1$ since $D$ is abelian. Conversely, given $x,y\in G$, choose lifts $(x,d),(y,d')\in P$. Their commutator is $([x,y],1)$. Since such commutators generate $[G,G]$, this proves the first equality. The second projection has kernel $[G,G]\times\{1\}$ and is surjective, yielding the second assertion. This identifies the slice counit with an isomorphism.
\end{proof}

Every surjection of ordinary groups is an effective descent map for surjective arrows. One can see this directly by viewing a group as a reduced simplicial set satisfying the Segal conditions and the invertibility condition. A surjection $U\to G$ induces a pointwise surjection between their group nerves, since the map in degree $r$ is $U^r\to G^r$. Ordinary effective descent for a surjection of sets comes from forming equivalence classes under its compatible gluing relation. Applied degree by degree, this gives descent for diagrams of sets, hence a descended simplicial set over the nerve of $G$. The Segal comparisons are maps over the products $G^r$. Their pullbacks to $U^r$ are isomorphisms, so they are isomorphisms. The same argument shows that the shear comparison $(d_2,d_1)$ in degree two is an isomorphism, which recovers invertibility in the descended monoid. Surjectivity in degree one is also detected after pullback. Thus the descended object is an allowed group surjection, with morphisms descending by the same degreewise argument.

\begin{theorem}[Trivial, split, and central group extensions]
\label{thm:ordinary-central}
For a group surjection $e\colon E\to G$ with kernel $K$, the following statements hold.
\begin{enumerate}
\item The surjection is trivial for abelianization exactly when $K\cap[E,E]=1$.
\item The surjection is split by some surjection exactly when $K$ is central in $E$.
\item A central surjection is split by itself and thus is normal for this Galois structure.
\end{enumerate}
The reflection of $e$ into central extensions over $G$ is the quotient $E/[K,E]\to G$.
\end{theorem}

\begin{proof}
Consider the unit comparison
\[
 \phi\colon E\longrightarrow G\times_{G_{\mathrm{ab}}}E_{\mathrm{ab}}.
\]
It is surjective. To lift $(g,u)$, first choose $x$ lifting $g$. The difference between $u$ and $a_E(x)$ lies in the kernel of $E_{\mathrm{ab}}\to G_{\mathrm{ab}}$. This kernel is the image of $K$ because the surjection $[E,E]\to[G,G]$ allows any representative mapping to $[G,G]$ to be corrected to an element of $K$ without changing its abelianized class. Multiplying $x$ by a lift of the difference in $K$ gives the desired lift. The kernel of $\phi$ is visibly $K\cap[E,E]$. This proves the first statement. It also implies that any trivial extension is central, since $[K,E]\subseteq K\cap[E,E]$ by normality of the kernel.

Suppose $p\colon U\to G$ splits $e$. The kernel of $U\times_GE\to U$ is canonically $K$ and is central in $U\times_GE$ by the first statement. The second projection $U\times_GE\to E$ is surjective, so each $y\in E$ is the image of an element commuting with every element of the lifted $K$. Therefore $K$ is central in $E$.

Conversely, centrality makes the map
\[
 E\times K\longrightarrow E\times_GE,\qquad
 (x,k)\longmapsto(x,xk)
\]
a group isomorphism over $E$. The projection $E\times K\to E$ is trivial by the first statement, since $K$ is abelian and $[E\times K,E\times K]=[E,E]\times\{1\}$. Hence $e$ is split by itself.

Finally, $[K,E]$ is a normal subgroup of $E$ contained in $K$. In the quotient by this subgroup the image of $K$ commutes with the image of $E$. Every map from $e$ to a central extension over $G$ kills each commutator $[k,x]$, since $k$ maps into the central kernel of the target. It therefore factors uniquely through this quotient. This proves the universal property.
\end{proof}

This gives the ordinary relative comparison and its algebraic content \cite[Section~5.2]{bj}. A general treatment of categorical centrality for admissible reflections appears in \cite{jkcentral}. The split object in the sense of Chapter~\ref{ch:reflections} is an extension that becomes trivial after a chosen change of base. An ordinary group splitting by a section is a separate property. For instance, a central extension that has a group-homomorphic section is isomorphic over its base to a direct product with its kernel, whereas the classification below also includes nonsplit group extensions.

\section{Higher central extensions as pointed principal bundles}

Work in an $\infty$-topos $\mathcal X$. Write $BG$ for the delooping of a group object $G$. The group object and pointed connected object descriptions are equivalent via $B$ and $\Omega$ \cite[Section~6.1]{htt}. If $A$ is a commutative group object, $BA$ itself has a commutative group structure, so its delooping $B^2A$ is defined. One way to construct this structure is to apply the bar construction to the grouplike commutative monoid structure on $A$. Realization preserves the finite products in the bar diagram, so the commutative structure survives and its underlying monoid is grouplike because it is connected.

\index{central extension!framed}\index{band}
\begin{definition}[Framed central extension]
\label{def:framed-higher-central}
Fix a group object $G$ and a commutative group object $A$. A framed central extension of $G$ by $A$ is a principal $BA$-bundle $P\to BG$ with a point over the basepoint of $BG$. Equivalences preserve the principal structure and the chosen point, including their coherences. The associated group extension is $A\to\Omega P\to G$.
\end{definition}

The choice of a principal action is the band in this definition. The basepoint trivialization identifies the fiber over the basepoint with $BA$ using that action. Taking loops then identifies the kernel with $A$. The structure is thus a higher analogue of prescribing an abelian coefficient group and keeping its identification with the central kernel fixed.

\begin{theorem}[Classifying space of framed extensions]
\label{thm:higher-central-classification}
The space of framed central extensions of $G$ by $A$ is
\[
 \operatorname{Map}_{\mathcal X_*}(BG,B^2A).
\]
For the extension specified by $c\colon BG\to B^2A$, there is a pointed fiber sequence
\[
 BA\longrightarrow \operatorname{hofib}(c)\longrightarrow BG
     \xrightarrow{c} B^2A.
\]
Its associated group map has $\pi_0$-surjective augmentation $\pi_0\Omega\operatorname{hofib}(c)\to\pi_0G$, and the image of $\pi_0 A$ is central in $\pi_0\Omega\operatorname{hofib}(c)$.
\end{theorem}

\begin{proof}
The universal principal $BA$-bundle is $1\to B^2A$. Its \v{C}ech nerve is the bar construction of $BA$. Pullback classifies principal bundles, with mapping spaces, by the principal-bundle comparison, applied to $BA$ \cite{nss}. At the basepoint of $BG$ a lift to the pullback of the universal bundle is exactly a specified identification between the classifying map at that basepoint and the basepoint of $B^2A$. This gives the asserted equivalence of spaces and the fiber sequence.

The fibers of $P\to BG$ are locally $BA$, which is connected. Relative connectedness can be checked after an effective epimorphism by relative truncation and effective descent. Consequently $P\to BG$ is $0$-connected and $P$ is connected, since $BG$ is connected. Taking loops preserves the pointed pullback describing the fiber and gives the group kernel $A$. The long exact homotopy sequence, or its internal sheaf form, gives the asserted surjectivity because $BA$ has terminal $\pi_0$.

To see centrality, loop the principal action $BA\times P\to P$ and use preservation of products by loops. The resulting map $A\times\Omega P\to\Omega P$ is a morphism of group objects and its restrictions to the two factors are the kernel inclusion and the identity. The images of those factors commute on $\pi_0$, since the source is a product of groups there. The principal action records a choice of coherence beyond this equality on components.
\end{proof}

For a general higher group $G$, the kernel of $\pi_0 A\to\pi_0\Omega P$ is the image of the connecting homomorphism $\pi_1G\to\pi_0 A$. The homotopy fiber assertion incorporates this correction to the component-level group sequence.

\begin{proposition}[Recognition from a coherent action]
Let $p\colon E\to G$ be a map of group objects surjective on the sheaves of components. Suppose a coherent action of $A$ on $E$ over $G$ is supplied, with the simplicial action diagram $A^\bullet\times E$ formed in group objects and with $A$ carrying its specified commutative group structure. If the action shear $A\times E\to E\times_GE$ is an equivalence, the resulting pointed group extension has a framed central structure.
\end{proposition}

\begin{proof}
Deloop each term of the action diagram. The equivalence between group objects and pointed connected objects preserves products, so the resulting simplicial action diagram is $BA^\bullet\times BE$. The map $BE\to BG$ is $0$-connected, since the long exact sequence and surjectivity of $\pi_0p$ give connected homotopy fibers at the basepoint, and the basepoint $1\to BG$ is an effective epimorphism. In particular $BE\times_{BG}BE$ is connected. The delooped shear $BA\times BE\to BE\times_{BG}BE$ is therefore a map between pointed connected objects. Its loop comparison is the given equivalence, so it is an equivalence as well. This proves that $BE\to BG$ is a principal $BA$-bundle, with the frame given by the basepoint of $BE$.
\end{proof}

\begin{corollary}[Ordinary central extensions and their automorphisms]
For ordinary groups $G$ and ordinary abelian groups $A$ regarded as discrete objects of $\SSet$, framed extensions give exactly the ordinary central extensions of $G$ with kernel identified with $A$. Their isomorphism classes are $H^2(G;A)$ for the trivial action, and the automorphism group of an extension with kernel and quotient fixed is $H^1(G;A)$.
\end{corollary}

\begin{proof}
The fiber sequence $BA\to P\to BG$ shows that $P$ is a connected $1$-type. Thus $\Omega P$ is an ordinary group. Its surjectivity, injectivity of the kernel, and centrality follow from the theorem and the long exact sequence. Conversely an ordinary central extension $A\to E\to G$ yields the group homomorphism $A\times E\to E$ by multiplication of the central inclusion with the identity. Delooping this homomorphism gives a left $BA$-action on $BE$ over $BG$. The map
\[
 BA\times BE\longrightarrow BE\times_{BG}BE
\]
is an equivalence. Both sides are pointed and connected, the right side by surjectivity of $E\to G$, and looping identifies the map with the ordinary torsor comparison $A\times E\to E\times_GE$. The latter is bijective by uniqueness of the kernel element relating two lifts. Also $BE\to BG$ is an effective epimorphism since both spaces are connected. This gives the principal bundle with its point. The constructions are inverse on objects and isomorphisms, using the equivalence between groups and pointed connected $1$-types. The central action in this degree is forced by the kernel inclusion and the group multiplication, so it adds no independent data.

Finally $B^2A=K(A,2)$. Homotopy groups of $\operatorname{Map}_*(BG,K(A,2))$ are the reduced cohomology groups in degrees $2-i$ for $0\leq i\leq2$. Here degree zero agrees with unreduced cohomology in positive cohomological degree, and $\widetilde H^0(BG;A)=0$ because $BG$ is connected. Translation by an extension class identifies its based automorphism group with the first homotopy group at the zero class. This gives the two identifications.
\end{proof}

One can read this classification directly in a group bar resolution. A section of sets $s\colon G\to E$ with $s(1)=1$ determines $\alpha(g,h)\in A$ by
\[
 s(g)s(h)=\alpha(g,h)s(gh).
\]
Associativity and centrality give
\[
 \alpha(g,h)+\alpha(gh,k)=\alpha(h,k)+\alpha(g,hk).
\]
Conversely a normalized solution defines the multiplication $(a,g)(b,h)=(a+b+\alpha(g,h),gh)$ on $A\times G$. The identity and inverses follow from the normalization and the cocycle equality, and its projection has central kernel $A$. Changing the section modifies $\alpha$ by the group coboundary of a normalized function $G\to A$. The bar nerve of $G$ has $r$-simplices $G^r$, so its ordinary normalized cochain differential in degree two is exactly that equation. This comparison identifies the group cocycle with the classifying map and explains which operation in the proof is replaced by coherent descent in higher degrees.

\section{Higher cohomology and the role of trivialization}\index{cohomology!central extensions}\index{local coefficients}

If $A_0$ is an ordinary abelian group object in an $\infty$-topos, let $K(A_0,r)$ denote its Eilenberg--Mac Lane object in degree $r$. For the commutative group object $A=K(A_0,n)$ the classifying space in Theorem~\ref{thm:higher-central-classification} is
\[
 \operatorname{Map}_{\mathcal X_*}(BG,K(A_0,n+2)).
\]
The shifts come from using one delooping to pass from a group object to its classifying object and a second to classify the principal action on that object. In spaces, components are $H^{n+2}(BG;A_0)$ and higher homotopy groups are the reduced cohomology in the corresponding lower degrees. In a general topos, the pointed mapping space is the homotopy fiber of restriction along $1\to BG$ over the zero section. This description records the chosen trivialization over the basepoint even when cohomology of the terminal object is nontrivial.

A variable coefficient object can instead be specified over $BG$. Let $\mathcal A$ be an abelian group object in the $0$-truncated part of $\mathcal X_{/BG}$ and choose an identification of its restriction at the basepoint with $A_0$. Eilenberg--Mac Lane objects can be formed in this slice. A principal $K_{/BG}(\mathcal A,n+1)$-bundle in the slice is classified by a section of $K_{/BG}(\mathcal A,n+2)$. The proof is the same principal-bundle comparison, now with terminal object $BG\to BG$. Taking the homotopy fiber of restriction to the basepoint over the zero section adds a point in the pulled-back torsor. Looping the resulting pointed total object gives a group map with kernel $K(A_0,n)$. The action over $BG$ is a relative action and the coefficient object may vary along loops. For the constant coefficient object it is the global principal action in the theorem. Thus the slice construction also tracks coefficient twisting without imposing centrality on an arbitrary variable band.

\begin{example}[An extension with a higher obstruction class]
In spaces take $G=\Omega S^3$ and $A=K(\mathbb Z,1)$. Let $c\colon S^3\to B^2A=K(\mathbb Z,3)$ represent the integral generator. The map on $\pi_3$ induced by $c$ is an isomorphism, so the boundary $\pi_3(S^3)\to\pi_2(BA)$ in the fiber sequence $BA\to P\to S^3$ is an isomorphism. This gives $\pi_2(P)=0$. For the zero class the total object is $BA\times S^3$ and has $\pi_2\cong\mathbb Z$. Both looped extensions have trivial component groups. The principal class distinguishes them and is detected by higher homotopy.
\end{example}

Normalization of the band is essential to the comparison with abelianization. For ordinary central group extensions the conjugation action on $A$ is trivial, with no higher choices of coherence. In the higher case it is possible to ask for a neutralizable coefficient system, a chosen null homotopy of an action, or a specified principal $BA$-bundle. Definition~\ref{def:framed-higher-central} chooses the principal-bundle condition. The classifying theorem concerns that structure and its morphisms.

\section{Abelianization after passing to group objects in spaces}

Two extensions of the abelianization functor illustrate why the band construction has an independent role. The first applies ordinary abelianization to $\pi_0$. The second uses the left adjoint to the forgetful functor from infinite loop spaces to group objects in spaces. The first is a reflection onto a discrete subcategory. The second has a counit that can change the positive homotopy groups of a connective spectrum.

\index{abelianization!discrete reflection}
\begin{proposition}[A discrete reflection on higher group objects]
The functor from group objects in spaces to ordinary abelian groups given by $G\mapsto(\pi_0G)_{\mathrm{ab}}$ is an admissible reflection when the allowed maps are surjective on $\pi_0$ in the source and surjective in the target. A map split for this structure by an allowed effective descent map has $0$-truncated fibers.
\end{proposition}

\begin{proof}
Maps into a discrete abelian group factor through $\pi_0$ and abelianization by the two respective universal properties. This proves the reflection assertion. Given an allowed abelian group surjection $D\to(\pi_0G)_{\mathrm{ab}}$, the underlying space of the slice right adjoint is the pullback $G\times_{(\pi_0G)_{\mathrm{ab}}}D$. Since its base is discrete, its component set is $\pi_0G\times_{(\pi_0G)_{\mathrm{ab}}}D$. Lemma~\ref{lem:abelianization-admissible} then computes the counit. Closure of the allowed maps under pullback follows from stability of effective epimorphisms of spaces. Surjectivity of the projection to $G$ follows from the same stability. The sliced adjunction is admissible.

Every trivial map is a pullback from a map of discrete abelian groups, so has discrete fibers. If an allowed map becomes trivial along a descent map $U\to G$, the underlying map of spaces $U\to G$ is an effective epimorphism since it is surjective on components. Truncation of fibers descends along such a map by Chapter~\ref{ch:reflections}.
\end{proof}

In fact, in this structure each allowed base change map is effective for the restricted \v{C}ech comparison. In the functor $\infty$-topos of simplicial spaces, the map of group nerves associated to $U\to G$ is an effective epimorphism. This holds pointwise because the map in simplicial degree $r$ is $U^r\to G^r$. Descent in that functor category provides a simplicial object over the group nerve of $G$. The degree-zero condition is unchanged under pullback since both group nerves have terminal object in degree zero. The Segal comparisons over $G^r$ and the invertibility comparison over $G^2$ become equivalences when pulled back along the effective epimorphisms $U^r\to G^r$ and $U^2\to G^2$, respectively. Here the invertibility comparison is the shear $(d_2,d_1)$, transported along the shear equivalence for the base group nerve. They are therefore equivalences prior to pullback. The descended object is a group nerve, and the required surjectivity on $\pi_0$ in degree one is detected after base change. The whole comparison is an equivalence on the selected full subcategories, with all spaces of morphisms.

Consequently a higher principal extension with a nondiscrete kernel, such as the preceding example, may have no interpretation as a covering for this discrete Galois structure. Its fibers already prevent that comparison. The discrete reflection retains the ordinary group case proved above and provides one admissible choice when the intended fibers have no higher homotopy.

\index{abelianization!infinite-loop}
\begin{proposition}[Infinite-loop abelianization in spaces]
Write $U$ for the forgetful functor from commutative group objects to group objects in spaces. Under the equivalence between commutative group objects and connective spectra, a left adjoint to $U$ is
\[
 F(G)=\Sigma^{-1}\Sigma^\infty BG.
\]
Here $\Sigma^\infty$ is reduced suspension spectrum. The zeroth homotopy group of $F(G)$ is $(\pi_0G)_{\mathrm{ab}}$. The counit is noninvertible at $H(\mathbb Z^2)$, so full faithfulness fails for $U$.
\end{proposition}

\begin{proof}
Under infinite-loop recognition, a connective spectrum $V$ gives the infinite loop space $\Omega^\infty V$, with delooping $\Omega^\infty\Sigma V$ \cite[Remark~5.2.6.26]{ha}. For pointed connected $BG$, the adjunction between reduced suspension spectrum and infinite loop space gives
\[
 \begin{split}
 \operatorname{Map}_{\operatorname{Grp}(\SSet)}(G,U(V))
 &\simeq \operatorname{Map}_{\SSet_*}(BG,\Omega^\infty\Sigma V)\\
 &\simeq \operatorname{Map}_{\operatorname{Sp}}(\Sigma^\infty BG,\Sigma V)\\
 &\simeq \operatorname{Map}_{\operatorname{Sp}}(\Sigma^{-1}\Sigma^\infty BG,V).
 \end{split}
\]
The spectrum in the last source is connective because $BG$ is connected. Its zeroth homotopy group is the degree-one stable homotopy group of $BG$, and stable Hurewicz identifies this with $H_1(BG;\mathbb Z)\cong(\pi_0G)_{\mathrm{ab}}$.

For a counterexample to full faithfulness, use the connective Eilenberg--Mac Lane spectrum $H(\mathbb Z^2)$. Its underlying group object is discrete $\mathbb Z^2$, with delooping $S^1\times S^1$. There is a stable splitting
\[
 \Sigma^\infty(S^1\times S^1)
 \simeq \Sigma^\infty S^1\oplus\Sigma^\infty S^1\oplus\Sigma^\infty S^2.
\]
Indeed the stable cofiber of the wedge inclusion is the suspension spectrum of the smash product, and the two projections split the wedge inclusion in spectra. Thus $F(U(H(\mathbb Z^2)))$ has a sphere spectrum suspended once as a summand. It has nonzero first homotopy group, while $H(\mathbb Z^2)$ has zero first homotopy group. The counit cannot be an equivalence, contrary to what full faithfulness would imply.
\end{proof}

The two constructions agree on the zeroth homotopy group and differ even when the input is discrete and already abelian. The higher band theorem uses delooping and effective groupoid descent directly, so its classifying space requires neither an admissible infinite-loop reflection nor a presentation by fiberwise discrete data.

\part{Directed Galois theories}
\chapter{Directed symmetries and generalized descent}
\label{ch:directed}

A passage from a splitting object to its groupoid of changes of trivialization discards no information when all transports are invertible. There are also settings where a comparison has an intrinsic direction. A specialization of points can induce a map of fibers which is far from an equivalence. The directed analog of a Galois action is then a functor on a category with genuine noninvertible morphisms. This chapter distinguishes that construction from the generalized descent on precategories in \cite[Chapter~7]{bj}, and gives a higher form of both constructions.

\section{Complete Segal objects and groupoid objects}\index{complete Segal object}

Let \(\mathcal B\) be an \(\infty\)-topos. A simplicial object \(C_\bullet\) in \(\mathcal B\) is a Segal object if the canonical map
\[
 C_n\longrightarrow C_1\times_{C_0}C_1
              \times_{C_0}\cdots\times_{C_0}C_1
\]
is an equivalence for \(n\geq 2\). The maps in each pullback are the target and the source of successive arrows. Here \(d_1\) is the source and \(d_0\) the target. The simplicial identities give identity and composition laws with all their coherences. In particular choosing a representative composition at degree two gives a contractible space of choices, and degree three provides the associativity homotopy for the compatible choices.

There is an object \(C_1^{\mathrm{eq}}\) of equivalences. A generalized arrow over \(T\in\mathcal B\) belongs to this subobject when it is invertible in the category encoded by the Segal space \(\operatorname{Map}_{\mathcal B}(T,C_\bullet)\). Equivalence is an intrinsic property of an arrow. The object can be built by taking the object of coherent inverse data for the walking isomorphism. The space of choices of coherent inverses to an equivalence is contractible. Completeness means that the degeneracy map specifying identities induces an equivalence \(C_0\simeq C_1^{\mathrm{eq}}\). In contexts \(T\) this says that equivalences of objects are modeled by paths in the object space. Complete Segal objects are one model of small \(\infty\)-categories internal to \(\mathcal B\) \cite{martini}.

\begin{remark}
An internal groupoid object can have different object and arrow objects, as happens for a nontrivial \Cech{} nerve. Such a groupoid object is a Segal object with all arrows invertible. It usually fails completeness in the complete Segal sense above. A complete Segal object with all arrows invertible is equivalent to a constant simplicial object. An effective groupoid object in an \(\infty\)-topos is a presentation of its quotient, while a constant simplicial object on the quotient is already complete.
\end{remark}

Working contextually is important. Global elements of objects of an \(\infty\)-topos need not detect equivalences. For example, a statement about fibers over all \(c:T\to C_0\) can be tested at the generic point \(c=\operatorname{id}_{C_0}\) in context \(C_0\). A statement about fibers over all \(1\to C_0\) can hold vacuously while the assertion over \(C_0\) fails. The recovery of internal constructions from their underlying global categories requires its own hypotheses.

\section{Coherent presheaves and direction}\index{right fibration}\index{left fibration}

\begin{definition}
A right fibration over an internal category \(C_\bullet\) is a map \(q:E_\bullet\to C_\bullet\) of complete Segal objects such that each terminal-vertex map gives a pullback square
\[
 \begin{array}{ccc}
 E_n &\longrightarrow& E_0\\
 \downarrow && \downarrow\\
 C_n &\longrightarrow& C_0 .
 \end{array}
\]
A left fibration uses the initial vertex instead. Write \(\operatorname{RFib}_{\mathcal B}(C)\) and \(\operatorname{LFib}_{\mathcal B}(C)\) for the respective full subcategories over \(C\).
\end{definition}

For a generalized edge \(f:x\to y\) with parameters in \(T\), the right-fibration square in degree one identifies a lift ending at an element of \(E_y\) with that element. Taking the source of such lifts defines a map \(f^*:E_y\to E_x\) in \(\mathcal B_{/T}\). In degree two, lifting the two-simplex ending at \(z\) identifies \((gf)^*\) with \(f^*g^*\) on the fiber over \(z\). Higher terminal-vertex squares provide the compatible homotopies for longer strings. A left fibration instead gives an operation \(f_!:E_x\to E_y\). The object \(E_\bullet\) stores the transport as a coherent system, even when no levelwise representatives of its homotopies have been chosen.

There is a large internal universe \(\Omega_{\mathcal B}\) of objects in slices of \(\mathcal B\), taking \(T\) to \(\mathcal B_{/T}\) with reindexing by pullback, in a suitable enlargement of universes. The internal Grothendieck construction identifies the internal category of small right fibrations with the internal functor category from \(C^{\mathrm{op}}\) to \(\Omega_{\mathcal B}\). The left-fibration version is obtained by taking the opposites of the total category and the base. This theorem and internal Yoneda are developed in \cite[\S\S4.5--4.7]{martini}. For a contextual point \(c:T\to C_0\), the representable right fibration is the slice \(C_{/c}\to C|_T\), formed in \(\mathcal B_{/T}\). The Yoneda comparison evaluates a map of right fibrations on the identity of \(c\). The lift of an arrow into \(c\) then recovers its value by pullback transport. Coherence makes this a universal description of the mapping object, including in families over \(T\).

Externally the constructions specialize to straightening in spaces,
\[
 \operatorname{RFib}_{\mathcal S}(C)\simeq
       \operatorname{Fun}(C^{\mathrm{op}},\mathcal S),\qquad
 \operatorname{LFib}_{\mathcal S}(C)\simeq
       \operatorname{Fun}(C,\mathcal S).
\]
We use the right-hand convention for ordinary presheaves and the second convention for covariant representations. An equivalence \(C\to D\) induces equivalent categories of either sort. In the special case of a groupoid, inversion identifies the two conventions. For a general \(C\), reversing \(C\) has mathematical content, and the relevant variance has to be part of any classification theorem \cite[\S2.2]{htt}.

\begin{proposition}[Inverting the transport]
\label{prop:directed-localization}
Let \(W\) be a set of arrows of a small \(\infty\)-category \(C\). Restriction along \(C\to C[W^{-1}]\) identifies
\[
 \operatorname{Fun}(C[W^{-1}]^{\mathrm{op}},\mathcal S)
\]
with the full subcategory of right fibrations on \(C\) in which transport along \(W\) is by equivalences. The corresponding statement holds for left fibrations.
\end{proposition}

\begin{proof}
The universal property of localization applies to \(C^{\mathrm{op}}\) and the reversals of the arrows in \(W\). It identifies its entire functor category into \(\mathcal S\) with the indicated full subcategory of \(\operatorname{Fun}(C^{\mathrm{op}},\mathcal S)\). Under straightening, sending a reversed arrow to an equivalence is precisely the transport condition. The covariant proof uses \(C\) itself.
\end{proof}

For \(W\) consisting of all arrows, \(C[W^{-1}]\) is the groupoidification \(C^{\mathrm{gpd}}\). It represents the subtheory with invertible transport. In particular it does not generally classify all right or all left fibrations on \(C\).

\section{Two factorization systems}

The factorization developed in \cite[\S7.3]{bj} first sends a functor to its essential image, then embeds that image fully faithfully. This has a direct higher analog. For an \(\infty\)-category \(B\), write \(B^\simeq\) for its maximal subgroupoid.

\begin{proposition}
\label{prop:ess-image-factor}
Every functor \(f:A\to B\) factors as an essentially surjective functor followed by a fully faithful functor. These classes form an orthogonal factorization system on \(\operatorname{Cat}_\infty\).
\end{proposition}

\begin{proof}
Choose the full subcategory \(I\subseteq B\) on the union of the components of \(B^\simeq\) meeting \(f(A^\simeq)\). It contains the image of \(f\), every object is equivalent to one from \(A\), and inclusion is fully faithful.

For orthogonality, take an essentially surjective \(e:A\to B\) and a fully faithful \(m:X\to Y\). Replace \(X\) over \(Y\) by its full essential image using the preceding construction and full faithfulness. Given a square with \(e\) on the left and \(m\) on the right, the map \(B\to Y\) sends every object into this full subcategory since every object is equivalent to one in the image of \(A\). A full subcategory is determined by its allowed vertices and includes every simplex of \(Y\) with those vertices in a quasicategory model. The map \(B\to Y\) thus lifts. Postcomposition with a full inclusion gives a fully faithful functor \(\operatorname{Fun}(B,X)\to\operatorname{Fun}(B,Y)\). On cores the space of a lift with its prescribed identification is empty or contractible. The existence just proved, and the same full faithfulness for \(A\), show that imposing the rest of the square leaves a contractible space of lifts. Closure under retracts follows from the intrinsic definitions. Existence and orthogonality give the factorization system.
\end{proof}

A second factorization has a different right class. Given \(f:A\to B\), let \((b\downarrow f)\) be the comma category of objects \((a,b\to f(a))\), with arrows over \(b\). Write \(|D|\) for the groupoidification of \(D\). We call \(f\) final if every \(|b\downarrow f|\) is contractible. By the higher form of Quillen's Theorem~A, this is equivalent to \(f\) preserving colimit computations by restriction \cite[\S4.1.3]{htt}. An initial functor is one whose opposite is final. Its test uses the comma categories of arrows \(f(a)\to b\).

\index{comprehensive factorization}\begin{theorem}[Comprehensive factorization]
\label{thm:comprehensive-factor}
There is an orthogonal factorization system on \(\operatorname{Cat}_\infty\) with left class the final functors and right class the right fibrations, taken up to equivalence of arrows. The right-fibration factor of \(f:A\to B\) is classified by the presheaf
\[
 L_f=\operatorname{Lan}_{f^{\mathrm{op}}}\mathbf 1_A,\qquad
 L_f(b)\simeq |b\downarrow f|.
\]
Dually, the initial functors and the left fibrations form an orthogonal factorization system.
\end{theorem}

\begin{proof}
Here \(\mathbf 1_A\) is the terminal presheaf. The pointwise formula for left Kan extension has indexing category \((b\downarrow f)^{\mathrm{op}}\). The colimit of the constant point diagram is its groupoidification, canonically equivalent to that of the opposite category. In these terms the action of \(L_f\) is precomposition on comma categories. Let \(p:E_f\to B\) be the right fibration classified by \(L_f\). The Kan unit \(\mathbf 1_A\to (f^{\mathrm{op}})^*L_f\) corresponds to a section of \(f^*E_f\) over \(A\), giving \(j:A\to E_f\) with \(pj=f\). Under the comma description, \(j(a)\) is the point represented by \((a,\operatorname{id}_{f(a)})\).

To prove finality of \(j\), fix \(e=(b,z)\) in \(E_f\). A lift of \(\beta:b\to f(a)\) from \(e\) to \(j(a)\) is specified by a path between \(z\) and the transported point \(L_f(\beta)(j(a))\) in \(L_f(b)\). This description is natural in \((a,\beta)\), since an arrow of this comma category supplies a homotopy relating the two transported points. Consequently \((e\downarrow j)\) is classified over \((b\downarrow f)\) by the family of path spaces \(\operatorname{Map}_{L_f(b)}(z,\lambda(a,\beta))\), where \(\lambda:(b\downarrow f)\to |b\downarrow f|\) is the localization map. This family sends every arrow to an equivalence. Its total groupoidification is the colimit of the family in spaces by straightening and the universal property of a colimit. Universality of colimits in spaces computes this colimit by pulling the colimit of the points \(\lambda(a,\beta)\) in \(\mathcal S_{/L_f(b)}\) back along \(z\). That colimit is \(\operatorname{id}_{L_f(b)}\), so the pullback is contractible. This is the comma criterion for \(j\).

For orthogonality, pull a right fibration back along the bottom map \(B\to C\) of a square and let the resulting fibration over \(B\) correspond to \(F\). A top map lifting a given arrow \(e:A\to B\) gives a section over \(A\), equivalently an element of \(\operatorname{Map}(\mathbf 1_A,(e^{\mathrm{op}})^*F)\). The space of fillers is the fiber of the restriction map from \(\operatorname{Map}(\mathbf 1_B,F)\) over this element. The left Kan extension adjunction identifies restriction with precomposition by \(L_e\to\mathbf 1_B\). This arrow is an equivalence when \(e\) is final, so the fiber is contractible. Finally final functors and right fibrations are each closed under retracts. For the right class this follows directly from the defining pullback condition, or from its lifting tests against terminal-vertex inclusions. Thus the factorizations and mutual orthogonality identify both classes. Taking opposites proves the dual assertion.
\end{proof}

The distinction between the two factorizations is visible on a category with indefinitely composable arrows. Take \(\mathbb Z\) with its usual order, viewed as an \(\infty\)-category. The inclusion of the even integers is final since the even integers above each given \(b\) form a poset with a least element. It is fully faithful and fails essential surjectivity. Conversely \(\mathbb Z_{\mathrm{disc}}\to\mathbb Z\) is essentially surjective, whereas \((b\downarrow f)\) is a discrete infinite ray for this \(f\) and has noncontractible groupoidification. Even a groupoidification equivalence need not be final. The inclusion of \(0\) in \(\mathbb Z\) induces an equivalence on groupoidifications because \(\mathbb Z\) is filtered and nonempty, yet \((1\downarrow f)\) is empty.

The internal versions of the comprehensive factorization and of the comma criterion are established in \cite[\S4.4]{martini}. The tests there use every context \(T\) and take groupoidification internally. Applying the preceding external criterion only to global points of \(B\) would not give that result.

\section{Descent along an augmented diagram}

The precategories of \cite[\S\S7.1--7.5]{bj} are diagrams retaining a small amount of categorical incidence data, before enforcing the Segal pullback conditions. Their coefficient presheaves are formed by a two-dimensional limit. The corresponding higher construction can start from an arbitrary indexing \(\infty\)-category. This preserves the distinction the original book makes between a limit of change-of-base categories and the category of ordinary directed presheaves on an internal category.

Let \(I\) be small, let \(D:I^\triangleright\to\mathcal A\) be an augmented diagram with cone point \(a\), and let \(F,G:\mathcal A^{\mathrm{op}}\to\operatorname{Cat}_\infty\) be contravariant coefficient systems. The natural transformations between them are homotopy coherent. We call the augmentation effective for \(G\) when the canonical comparison
\[
 G(a)\longrightarrow \operatorname*{lim}_{i\in I^{\mathrm{op}}}G(D_i)
\]
is an equivalence. An object of the limit includes objects in the coefficient categories and equivalences relating every pullback of them, together with higher coherences. The maps between the objects of a compatible family are likewise compared in the entire mapping spaces.

\index{relative splitting}\begin{theorem}[Relative splitting along a diagram]
\label{thm:relative-indexed-split}
Suppose \(F\to G\) is pointwise fully faithful, the augmentation \(D\) is effective for \(G\), and there is an object \(i_0\) such that every \(i\) admits an arrow \(i\to i_0\) in \(I\). The choices of those arrows are immaterial. Let \(\operatorname{Split}_F(D)\subseteq G(a)\) consist of those \(x\) whose restriction to \(D_{i_0}\) belongs to the essential image of \(F(D_{i_0})\). Then comparison induces an equivalence
\[
 \operatorname{Split}_F(D)\simeq
       \operatorname*{lim}_{i\in I^{\mathrm{op}}} F(D_i).
\]
\end{theorem}

\begin{proof}
A limit of full inclusions is fully faithful. Its image consists of those coherent families whose entries lie in the respective essential images. One way to see this last assertion is to replace each \(F(D_i)\) by the full subcategory of \(G(D_i)\) on its image. Naturality ensures that each reindexing map restricts to the relevant full subcategories. In the model of a limit as the category of Cartesian sections of the associated Cartesian fibration over \(I\), the section belongs to the subfibration precisely when its objects do. Fullness in each fiber implies that every compatibility equivalence and coherent simplex of that section belongs to the subfibration as well.

For a section \(x_\bullet\), membership of \(x_{i_0}\) implies membership of \(x_i\) because an arrow \(i\to i_0\) yields \(x_i\simeq G(D_i\to D_{i_0})(x_{i_0})\). Naturality of \(F\to G\) places this pullback in the image of \(F(D_i)\). The converse follows by evaluation at \(i_0\). This identifies the limit of the \(F(D_i)\) with the pullback of
\[
 F(D_{i_0})\longrightarrow G(D_{i_0})
 \quad\text{along}\quad
 \lim G(D_i)\longrightarrow G(D_{i_0}).
\]
Using effectivity, this pullback is the defining full subcategory of \(G(a)\).
\end{proof}

Taking \(I\) to be the shape of the precategory in \cite[\S7.2]{bj}, each of its object, arrow, and composable-pair indices maps to the object index. In the ordinary category-valued case the coherent pullback equivalences are isomorphisms in the coefficient categories, and the limit computes the pseudolimit used in the book. The relative splitting criterion is then exactly the requirement at the object index in \cite[Theorem~7.5.2]{bj}. The higher theorem allows the whole simplicial shape and the resulting mapping spaces, with no reduction to comparison isomorphisms on a truncated shape.

If a diagram has several distinguished objects collectively receiving arrows from each index, the same argument imposes the condition at all distinguished objects. If \(D_\bullet\) is simplicial, \(I=\Delta^{\mathrm{op}}\) and \(i_0=[0]\) can be used. In that case a vertex inclusion in \(\Delta\) supplies the arrow required by the proof. For the \Cech{} nerve of an effective epimorphism in an \(\infty\)-topos, the slice coefficient system \(G(T)=\mathcal B_{/T}\) has effective descent. The theorem then restricts ordinary descent to a full class of objects that is stable under reindexing. Its naturality requirement on \(F\to G\) is explicit. It does not imply that every adjunction on slice categories determines such a subsystem.

There is a precise interpretation of the slice construction on an arbitrary simplicial object \(D_\bullet\). Every colimit in an \(\infty\)-topos is van Kampen \cite[\S6.1.3]{htt}, so
\[
 \operatorname*{lim}_{[n]\in\Delta}\mathcal B_{/D_n}
   \simeq \mathcal B_{/|D_\bullet|_{\mathcal B}},
 \qquad
 |D_\bullet|_{\mathcal B}=
       \operatorname*{colim}_{[n]\in\Delta^{\mathrm{op}}}D_n .
\]
Indeed a compatible family amounts to a map of simplicial objects \(E_\bullet\to D_\bullet\) for which the naturality square for every simplicial operator is Cartesian. If \(D_\bullet\) is Segal, this includes both vertex conditions, so transports from one endpoint to the other are equivalences. Over a complete Segal space the realization \(|D_\bullet|_{\mathcal S}\) models the groupoidification. It counts simplices as coherent paths in a space, and in that target every edge becomes invertible. Thus these particular limit coefficients generalize the precategorical construction, while noninvertible directed transport requires the left- or right-fibration construction.

\section{Descent of directed representations}

Two useful forms of directed gluing keep track of which category contains the colimit. The first has its presentation in \(\operatorname{Cat}_\infty\), where arrows remain directed. The second is indexed by objects in a base topos and glues all simplicial degrees using pullback.

\begin{proposition}[Gluing over categorical colimits]
\label{prop:directed-colim-glue}
Let \(C=\operatorname*{colim}_{i\in I}C_i\) in \(\operatorname{Cat}_\infty\), with all \(C_i\) small, and let \(\mathcal E\) be an \(\infty\)-category in the coefficient universe. Then
\[
 \operatorname{Fun}(C,\mathcal E)\simeq
       \operatorname*{lim}_{i\in I^{\mathrm{op}}}\operatorname{Fun}(C_i,\mathcal E).
\]
If a collection \(W_i\) of arrows is specified in each \(C_i\) compatibly with the diagram, the equivalence restricts to functors inverting the union \(W\) of their images in \(C\). The contravariant version follows by taking opposites of the indexing categories \(C_i\).
\end{proposition}

\begin{proof}
For any \(\infty\)-category \(K\) of the appropriate size, \(K\times-\) preserves colimits, as its right adjoint is \(\operatorname{Fun}(K,-)\). Mapping from \(K\) to the comparison functor therefore yields
\[
 \operatorname{Map}(K\times C,\mathcal E)
  \simeq \lim_i\operatorname{Map}(K\times C_i,\mathcal E).
\]
Testing against all \(K\) proves the equivalence of \(\infty\)-categories. Inverting an image arrow is a pointwise property of an object of the functor category. A full subcategory of the limit defined by these conditions consequently corresponds to the specified full subcategory on the left. The operation \(C\mapsto C^{\mathrm{op}}\) is an autoequivalence of \(\operatorname{Cat}_\infty\) and preserves the colimit in question.
\end{proof}

\begin{proposition}[Changing the base of an internal category]
\label{prop:rfib-internal-descent}
Let \(X\in\mathcal B\), let \(U\to X\) be an effective epimorphism, and let \(C\) be a small internal \(\infty\)-category in \(\mathcal B_{/X}\). For \(U_n=U^{\times_X(n+1)}\) let \(C_{U_n}\) be the pullback internal category. Then there is an equivalence
\[
 \operatorname{RFib}_{\mathcal B_{/X}}(C)\simeq
 \operatorname*{lim}_{[n]\in\Delta}
       \operatorname{RFib}_{\mathcal B_{/U_n}}(C_{U_n}).
\]
The left-fibration version is obtained by the same argument with initial vertices.
\end{proposition}

\begin{proof}
Effective descent in the topos gives \(\mathcal B_{/X}\simeq\lim\mathcal B_{/U_n}\). Taking simplicial diagrams commutes with this limit, and taking arrows with target the prescribed \(C\) also commutes with the limit. This glues the underlying map of simplicial objects, functorially on morphisms. An arrow in \(\mathcal B_{/X}\) is an equivalence if its pullback to \(U\) is an equivalence. To see this, its pullbacks to each \(U_n\) are equivalences by further reindexing, and the limit of slice categories detects equivalences on entries.

The pullback functors between all these slices preserve limits. Consequently the Segal equations for the glued source can be checked after pullback to \(U\), where they hold. The same reasoning applies to the completeness comparison by computing the space of coherent inverse data using limits. Finally the terminal-vertex pullback conditions amount to equivalences with finite pullbacks and hold after reindexing to \(U\). The glued map is therefore a right fibration. This restricts the equivalence on underlying maps to precisely the subcategories in the statement.
\end{proof}

\chapter{Exit paths and constructible monodromy}
\label{ch:exodromy}

A covering of a connected space transports its fibers along paths, with inverse paths producing inverse equivalences. A constructible object on a stratified space has locally constant fibers on each stratum, together with comparisons across incident strata. An exit path starts in a lower stratum and may move toward higher strata. The reverse path can violate the stratification condition. This asymmetry supplies the directed information in constructible monodromy.

\section{The topological classification}\index{exodromy}\index{exit-path category}

Give a poset \(P\) its Alexandrov topology of upward-closed subsets. A \(P\)-stratified space is a continuous map \(s:X\to P\), and \(X_p=s^{-1}(p)\). An exit simplex \(\sigma:|\Delta^n|\to X\) is a singular simplex for which there is a chain \(p_0\leq\cdots\leq p_n\) in \(P\) such that \(\sigma(t_0,\ldots,t_n)\) belongs to \(X_{p_i}\) when \(i\) is the largest index for which \(t_i>0\). Face and degeneracy operations respect this condition. A conically stratified space admits open neighborhoods of points \(x\in X_p\) given by stratified charts \(Z\times C(L)\), where \(Z\times\{\text{vertex}\}\) lies in \(X_p\) and the link \(L\) is stratified by the elements above \(p\). Here \(C(L)\) is the open cone, with a basis of vertex neighborhoods given by the cones of uniformly bounded radius. Conicality ensures that the exit simplicial set is an \(\infty\)-category, denoted \(\operatorname{Exit}_P(X)\). We use the conical chart and exit-path conventions in \cite[\S2.1]{portateyssier}.

The homotopy type \(\Pi_\infty X_p\) is the fiber of \(\operatorname{Exit}_P(X)\to P\) over \(p\). The inclusion of any individual stratum in the exit category is fully faithful. Indeed a simplex whose vertices are all over \(p\) has constant stratum label, so lies entirely in \(X_p\). All endomorphisms in \(\operatorname{Exit}_P(X)\) are equivalences for the same reason. More generally the functor to \(P\) is conservative. For a one-stratum space the simplicial model is the ordinary singular complex, with no directed constraints.

A hypersheaf of spaces on \(X\) is a sheaf which satisfies descent for hypercovers. Call it \(P\)-hyperconstructible if its restriction, as a hypersheaf, to every \(X_p\) is locally constant. This is a property, so the hyperconstructible objects form a full subcategory. In particular the construction does not prescribe a single common space as the value of all stalks. Each locally constant restriction can carry its own monodromy.

\begin{theorem}[Topological exodromy, space coefficients]
\label{thm:exit-space-coefficients}
If \(X\to P\) is conically stratified and each \(X_p\) admits a basis of weakly contractible open neighborhoods, there is an equivalence
\[
 \operatorname{Cons}^{\mathrm{hyp}}_P(X;\mathcal S)
     \simeq \operatorname{Fun}(\operatorname{Exit}_P(X),\mathcal S).
\]
Evaluation of a representation at \(x\) corresponds to the stalk at \(x\).
\end{theorem}

This is the space-coefficient case of the theorem of Porta and Teyssier \cite[Theorem~5.4.1 and Corollary~6.2.5]{portateyssier}. Their coefficient-general statement assumes joint conservativity of the hyper-restrictions to the strata. For spaces this conservativity follows from the stalk criterion for hypersheaves. The stalks of the restriction to a stratum are the stalks on \(X\) at those same points, and the strata contain every point. For the full theorem the extra conservativity condition is substantive. The statement above uses hyperdescent throughout. Passing to ordinary sheaves requires a comparison with hypercompletion in the setting at hand.

The proof architecture of the cited theorem clarifies the role of the hypotheses. The exodromy adjunction reconstructs a hypersheaf from a representation by assigning to an open \(U\) the limit over its exit-path category. Conical charts give cofinal tests for computing its inverse image on a stratum. In these tests the restriction of both functors and of the adjunction transformations agrees with the corresponding monodromy adjunction on the stratum. For a locally weakly contractible stratum, that adjunction is an equivalence on local systems. Hyper-restriction detects the unit on constructible hypersheaves, and evaluation at every point detects the counit on representations. This is where the theorem uses the local geometry beyond just the existence of a topological stratification \cite[\S\S4--5]{portateyssier}.

By the convention of Chapter~\ref{ch:directed}, the representations in Theorem~\ref{thm:exit-space-coefficients} are classified by left fibrations over \(\operatorname{Exit}_P(X)\). If one uses entrance paths and denotes the resulting opposite category by \(\operatorname{Ent}_P(X)\), the same data are ordinary presheaves on \(\operatorname{Ent}_P(X)\). This change of orientation applies to the indexing category. The stalk maps attached to exit paths are still covariant.

\begin{proposition}[Sections and open restriction]
\label{prop:exit-sections}
Under the hypotheses of Theorem~\ref{thm:exit-space-coefficients}, let \(F\) be the hypersheaf corresponding to \(M:\operatorname{Exit}_P(X)\to\mathcal S\). For an open \(U\subseteq X\) with the induced stratification,
\[
 F(U)\simeq \lim_{\operatorname{Exit}_P(U)} M|_{\operatorname{Exit}_P(U)}.
\]
The limits retain every compatibility with the noninvertible arrows of the exit category.
\end{proposition}

\begin{proof}
The strata of an open subspace inherit the required local bases and conical charts. Naturality of exodromy for open inclusions identifies hypersheaf restriction with precomposition by \(\operatorname{Exit}_P(U)\to\operatorname{Exit}_P(X)\) \cite[Proposition~6.2.3]{portateyssier}. Sections are maps from the terminal hypersheaf. The terminal hypersheaf is constructible, and under an equivalence of categories it corresponds to the terminal functor, of constant value \(*\). The mapping space out of that functor is the limit being asserted, by the defining adjunction of the constant-diagram functor with the limit functor.
\end{proof}

\section{A conical calculation with higher stalks}
\label{sec:exit-cone}

Let \(M\) be the plane with the discrete closed subset \(\mathbb Z\times\{0\}\) removed, and let \(L=S^2\). Both are locally contractible manifolds, \(M\) with infinitely many punctures. On \(X=M\times C(L)\), put the lower stratum \(Z=M\times\{\text{vertex}\}\) and the upper stratum \(U=M\times L\times\mathbb R_{>0}\). The lower stratum has conical charts obtained by taking disks in \(M\) times cones of uniformly bounded radius. The upper stratum is open and is an ordinary manifold. This gives explicit conical and weak local contractibility hypotheses for exodromy.

Write \(K=\Pi_\infty(M)\) and \(T=\Pi_\infty(L)\). The exit category of the cone is \(T^\triangleleft\), the category obtained by adjoining an initial object to \(T\). To justify this, the nonvertex subcategory is full and equivalent to \(T\) by product with \(\mathbb R_{>0}\), while the vertex is initial by the cone lemma \cite[Lemma~2.1.8]{portateyssier}. An initial vertex determines, up to contractible choice, a coherent cone on the inclusion of the nonvertex subcategory. The resulting functor from \(T^\triangleleft\) is fully faithful. The mapping spaces out of the initial vertex are contractible and there are no arrows from a nonvertex object to the vertex, since such arrows decrease the stratum label. Mapping spaces between nonvertex objects agree by fullness. The functor is therefore an equivalence. Exit simplices of a product with an unstratified space decompose into a singular simplex in that space and an exit simplex in the stratified factor. It follows that
\[
 \operatorname{Exit}_P(X)\simeq K\times T^\triangleleft .
\]

Write \(\pi:K\times T\to K\) for projection. A representation is equivalently the full coherent data
\[
 A:K\to\mathcal S,\qquad
 B:K\times T\to\mathcal S,\qquad
 \eta:\pi^*A\longrightarrow B .
\]
Indeed currying gives a \(K\)-indexed family of diagrams on \(T^\triangleleft\). A diagram on \(T^\triangleleft\) is a diagram on \(T\) equipped with a cone, which gives precisely the displayed data in each fiber. Natural transformations are compatible maps of the two diagrams and the comparison; the compatibility square is specified in the functor \(\infty\)-category. At a given \(k\), adjunction also writes the comparison as \(A(k)\to \lim_{t\in T} B(k,t)\). This last description shows why keeping only the monodromy on each stratum can lose a comparison even when both restrictions are known exactly.

The inclusion of \(K\) at the initial vertex is initial. For an object \((k,t)\) of the entire exit category, the category of arrows from \((k',\text{vertex})\) to it is equivalent to \(K_{/k}\), which has a terminal object. Therefore the global sections of the hypersheaf corresponding to \((A,B,\eta)\) are \(\lim_K A\) by Proposition~\ref{prop:exit-sections} and the initiality criterion. The comparison \(\eta\) still controls the behavior on neighborhoods intersecting the upper stratum.

For example, take \(A\) and \(B\) to be constant of value the Eilenberg--Mac~Lane space \(Q=K(\mathbb Z,2)\). The identity comparison and the comparison constant at the base point give inequivalent constructible hypersheaves with equivalent restrictions to both strata. A natural equivalence between the two representations would give a commutative square up to homotopy on any radial exit arrow, with equivalences on its vertical sides. On \(\pi_2(Q)\), one horizontal side is the identity and the other is zero. No such square can have invertible vertical sides.

There is also a link contribution in the direct image \(j_*B\) for \(j:U\hookrightarrow X\). In the representation category \(j_*\) corresponds to right Kan extension along \(K\times T\hookrightarrow K\times T^\triangleleft\). This can be seen directly from \(j_*B(V)=B(V\cap U)\) and the section formula. At \((k,\text{vertex})\), the comma category for right Kan extension is equivalent to \(K_{k/}\times T\). In this product, restricting a limit to the slice at \(\operatorname{id}_k\) is allowed because the identity is initial in \(K_{k/}\). Therefore the value on the lower stratum is \(k\mapsto\lim_T B(k,-)\). For the constant diagram with value \(Q\), this gives
\[
 \operatorname{Map}_{\mathcal S}(S^2,Q)
     \simeq Q\times \mathbb Z_{\mathrm{disc}}.
\]
Evaluation at a chosen point of \(S^2\) has fiber \(\Omega^2 Q\simeq\mathbb Z_{\mathrm{disc}}\). Taking \(Q\) in its abelian group space model separates a map into its value at the chosen point and a pointed map by translation, proving the product decomposition. In particular the direct image has extra components on the lower stratum although the initial value \(Q\) is connected. The higher homotopy type of the link is used in this computation. Its fundamental group is trivial.

Inverting all exit paths in this example sends \(K\times T^\triangleleft\) to \(K\). Groupoidification preserves products for these factors, as can be seen either from the existence of an initial object in the cone or by taking iterated colimits of constant points. Accordingly a functor comes from ordinary monodromy on \(X\) precisely when \(\eta\) is an equivalence. In that case \(B\simeq\pi^* A\) by \(\eta\), which already implies the factorization. Conversely any representation from a groupoid sends the exit comparisons to equivalences. This agrees with the deformation retraction of \(X\) to the lower stratum.

\section{A van Kampen principle with directed coefficients}

The right colimit for local presentations of a stratified space is a colimit of exit categories. Let \(\{U_\alpha\}\) be an open cover of a conically stratified \(X\) with locally weakly contractible strata, and let \(V_\bullet\to X\) be its open \Cech{} presentation. Its degree \(n\) can be read as the disjoint union of the intersections \(U_{\alpha_0}\cap\cdots\cap U_{\alpha_n}\). The exit-path van Kampen theorem gives
\[
 \operatorname{Exit}_P(X)\simeq
 \operatorname*{colim}_{[n]\in\Delta^{\mathrm{op}}}\operatorname{Exit}_P(V_n)
       \qquad\text{in }\operatorname{Cat}_\infty .
\]
For the application here this follows from the colimit-preserving exit-path functor on the hypersheaf topos, obtained by extending its values on opens \cite[\S4]{portateyssier}. Indeed the coproducts of intersections give the \Cech{} nerve in that topos, and its colimit is terminal because the cover map is an effective epimorphism. The exit functor sends this colimit to the displayed colimit. It sends a coproduct of the opens as hypersheaves to a coproduct of their exit categories, which agrees with the exit category of the disjoint union.

Proposition~\ref{prop:directed-colim-glue} now identifies constructible hypersheaves with coherent families of representations on these local exit categories. This operation compares local assignments on intersections by equivalences of representations. Their values along individual exit arrows inside an open set can be arbitrary maps. The equivalence relating two restrictions of a representation on an overlap and the map that the representation assigns to an exit arrow inside an open carry different parts of the data.

Suppose the strata are coarsened by a map of posets \(Q\to P\), and both stratifications on \(X\) are conical with locally weakly contractible strata. Let \(v:\operatorname{Exit}_Q(X)\to\operatorname{Exit}_P(X)\) be the comparison. Under these hypotheses \(v\) is the localization at the arrows whose images are equivalences \cite[Corollary~6.2.7]{portateyssier}. Therefore a \(Q\)-hyperconstructible hypersheaf is \(P\)-hyperconstructible if and only if the corresponding representation sends every such arrow to an equivalence. One implication follows by restriction from a \(P\)-representation. In the other direction the universal property of localization descends the representation to \(\operatorname{Exit}_P(X)\), and naturality of exodromy identifies its pullback with the original hypersheaf. The test is about coherent functors, so the comparison includes mapping spaces of classes that survive the coarsening. With the one-stratum coarsening, the test forces every transport to be invertible and recovers locally constant hypersheaves.

\section{Recovering the directed category from its coefficients}\index{atomic object}

Call an object \(a\) of a presentable \(\infty\)-category atomic if \(\operatorname{Map}(a,-)\) preserves all small colimits. This includes preservation of the empty colimit and is stronger than compactness, which asks only for preservation of filtered colimits. An atomic object is one on which it is possible to test arbitrary colimits by mapping out of it.

\begin{proposition}
\label{prop:exit-atomic}
Let \(C\) be a small \(\infty\)-category admitting a conservative functor to a poset. There is a canonical equivalence
\[
 C \simeq
 \left(\operatorname{Fun}(C,\mathcal S)_{\mathrm{at}}\right)^{\mathrm{op}},
\]
where the subscript denotes the full subcategory of atomic objects.
\end{proposition}

\begin{proof}
Covariant representables \(h_c=\operatorname{Map}_C(c,-)\) are atomic since mapping out of \(h_c\) is evaluation and colimits are pointwise. Conversely every functor \(F:C\to\mathcal S\) is a colimit of covariant representables over its category of elements, by density. For \(F\) atomic, apply \(\operatorname{Map}(F,-)\) to one such colimit presentation of \(F\). The identity of \(F\) is in the image on components of some stage, since \(\pi_0\) preserves colimits of spaces and every element of a colimit of sets comes from a stage. Thus \(F\) is a retract of a representable, with maps \(F\to h_c\to F\) whose composite is homotopic to the identity.

Every endomorphism of an object of \(C\) maps to an identity in the poset and is an equivalence by conservativity. By Yoneda the induced endomorphism of \(h_c\) is also an equivalence. With one composite the identity and the other invertible, the two maps in the retract diagram are equivalences. Every atom is therefore representable. Covariant Yoneda is fully faithful on \(C^{\mathrm{op}}\), giving the claimed variance and equivalence.
\end{proof}

Combining this proposition with Theorem~\ref{thm:exit-space-coefficients} recovers the exit category as the opposite of the atomic constructible hypersheaves. Under this identification the atom associated to \(x\) corepresents the stalk at \(x\). A map of atoms has variance reversed relative to an exit path. This follows from the natural transformation between evaluation functors and the contravariance of corepresenting objects. In particular the directed category can be recovered from the intrinsic colimits and arrows of the coefficient category, without a choice of a base point. If \(C\) does not admit the conservativity assumption, the general statement from the proof recovers the idempotent completion, since atoms are then retracts of representables \cite[\S5.1.6]{htt}.

Atomic generation is also a route to exodromy for stratified \(\infty\)-topoi whose geometric origin need not be conical. Take a geometric morphism to \(\operatorname{Fun}(P,\mathcal S)\), define the stratum topoi by geometric base change along the point topoi at \(p\), and define constructibility by locally constant restrictions. The formulation of \cite[Definition~2.2.10]{beyondconical} assumes atomic generation of the constructible subcategory, its closure under limits and colimits in the ambient topos, and preservation of limits by the inverse image of the stratification. The first property identifies the constructible category with a presheaf category by the free colimit completion of its atoms. The limit compatibility gives the left adjoint to stratification pullback on constructible coefficients, and this adjoint carries atoms to atoms because its right adjoint preserves colimits. Taking opposites then supplies the functor from the abstract exit category to \(P\). Thus a theorem proving these properties for a given stratified topos supplies a classifier without assuming that its simplices are represented by topological exit paths.

\section{The profinite directed case}\index{profinite exodromy}

The framework of Barwick, Glasman and Haine supplies an arithmetic counterpart \cite{exodromy}. For a coherent scheme \(Y\), in particular a quasicompact quasiseparated scheme, their category \(\operatorname{Gal}(Y)\) has geometric points as objects and \'etale specializations as morphisms. The construction equips it with a profinite categorical structure. A specialization carries incidence information, whereas endomorphisms at a geometric point account for the absolute Galois action on its residue-field data. Keeping both features is necessary to reconstruct the associated constructible coefficients.

For space coefficients in the sense of \cite[Corollary~0.5.2]{exodromy}, the theorem takes the form
\[
 \operatorname{Cons}(Y_{\textnormal{\'et}};\mathcal S_\pi)
      \simeq \operatorname{Fun}_{\mathrm{cts}}
             (\operatorname{Gal}(Y),\mathcal S_\pi),
\]
where on the left the constructibility condition is that of the stratified \'etale topos, and \(\mathcal S_\pi\) denotes the \(\pi\)-finite spaces. Such spaces are truncated, have finitely many components, and have finite homotopy groups in each component. The right side uses the profinite stratified shape and its continuous representation category. When a field is viewed as a one-point scheme, a chosen separable closure supplies a single geometric base point up to isomorphism, and the loop automorphisms in this classification recover its absolute Galois group with its profinite structure. General schemes may also have noninvertible specializations between points.

The classifying pro-space of the profinite stratified shape recovers the ordinary profinite shape for spectral stratified topoi \cite[Example~0.4.8]{exodromy}. This forgetful operation is the arithmetic counterpart of discarding directed transport by groupoidifying an exit category. The arithmetic theorem retains a continuity condition on representations, while Theorem~\ref{thm:exit-space-coefficients} uses an ordinary small exit \(\infty\)-category without an added profinite topology. The comparison thus relates the two generalizations through the role of coherence and direction, with the appropriate topology determined by the coefficient problem.

\part{Higher topoi and presentation}
\chapter{Geometric morphisms and the classification of sections}
\label{ch:geometric}

The final chapter of Borceux and Janelidze passes from objects in a
topos to morphisms between topoi. Their descent groupoid is a kernel
construction in a two-dimensional category. It records identifications
between geometric morphisms and thereby differs from the kernel pair
of a map of objects in a fixed topos. In higher dimensions this
distinction persists. The kernel of an étale geometric morphism can
be computed from an internal Čech nerve. An external atlas of a topos
requires separate information about its overlaps. We first prove
the representation theorem for étale geometric morphisms, then use
it to identify both the scope and the limits of the internal
construction. The classical background is \cite[\S\S7.7--7.10]{bj}.

\section{The category of geometric morphisms}
\index{geometric morphism}
\index{inverse image}

An $\infty$-topos is a presentable $\infty$-category admitting a
presentation by an accessible left exact localization of an
$\infty$-category of presheaves of spaces. Write
$\mathcal P(C)=\operatorname{Fun}(C^{\mathrm{op}},\mathcal S)$ for the
presheaves on a small $\infty$-category. A geometric morphism
$f\colon\mathcal Y\to\mathcal X$ is specified by an inverse image
$f^*\colon\mathcal X\to\mathcal Y$. It preserves finite limits and
small colimits, so the adjoint functor theorem supplies its right
adjoint $f_*$. Thus a morphism of topoi reverses the direction of
the functor that performs substitution. In particular a localization
$L\colon\mathcal X\to\mathcal X'$ gives a geometric morphism
$\mathcal X'\to\mathcal X$ when $L$ is left exact. In the inverse
image convention, a morphism from $f$ to $g$ in the category of
geometric morphisms is a natural transformation $f^*\to g^*$.
Passing to mates reverses this direction if one works with direct
images.

For each pair $\mathcal Y,\mathcal X$, put
\[
 \operatorname{Geom}(\mathcal Y,\mathcal X)
  =\operatorname{Fun}^{\mathrm{lex},\mathrm L}
       (\mathcal X,\mathcal Y).
\]
We take a full subcategory of the indicated functor category,
with all natural transformations between geometric inverse images.
Composition of functors and their transformations yields the
$(\infty,2)$-category of geometric morphisms in this convention.
Its underlying $(\infty,1)$-category, written
$\mathbf{Top}_{\infty}$, keeps the maximal subgroupoids of the
hom $\infty$-categories. A comparison saying that two diagrams
of inverse images are identified in $\mathbf{Top}_{\infty}$ means
that their comparison transformation is invertible. The
noninvertible transformations can still be useful. For instance,
a classifier for a theory can have models as the objects of a hom
category and homomorphisms as its noninvertible transformations.
The core records the space of models and their equivalences.

The terminal $\infty$-topos in $\mathbf{Top}_{\infty}$ is
$\mathcal S$. For each $\mathcal X$, the inverse image of its structure
map to $\mathcal S$ sends a space $K$ to
$\Delta_{\mathcal X}K=\operatorname{colim}_{K}1_{\mathcal X}$. Indeed,
$\mathcal S$ is the free cocompletion of the terminal category,
and the requirement to preserve the terminal object specifies its
image. The resulting functor preserves finite limits by the finite
limit extension theorem for presheaves
\cite[Prop.~6.1.5.2]{htt}. Its right adjoint is global sections
$\operatorname{Map}_{\mathcal X}(1_{\mathcal X},-)$.
The entire hom category
$\operatorname{Geom}(\mathcal X,\mathcal S)$ is contractible. This
argument uses the terminal-object condition on the inverse image,
so it applies equally when maps of topoi carry higher homotopies.

An inverse image preserves effective epimorphisms because it
preserves Čech nerves by left exactness and preserves their
realizations. It also preserves monomorphisms by the diagonal
test. An image factorization is an effective epimorphism followed
by a monomorphism, so an inverse image preserves images as well.
In particular the family of images of generating covers transported
by the inverse image remains jointly covering. This elementary
property is what links geometric morphisms to a site presentation.

\section{The universal property of a site}
\index{Grothendieck topology}
\index{sheafification}
\label{sec:higher-site-classifier}

Suppose $C$ has a Grothendieck topology. A sieve $R$ on $c$ is regarded
as a monomorphism $R\hookrightarrow y(c)$ in $\mathcal P(C)$.
Sheafification
$\mathcal P(C)\to\operatorname{Shv}(C;\mathcal S)$ is the
localization inverting covering sieve inclusions. If a family
$\{c_i\to c\}$ generates $R$, the image of
$\coprod_i y(c_i)\to y(c)$ is $R$. The sheaf condition at $R$ for
$F$ is the equivalence
\[
 F(c)\xrightarrow{\sim}
 \operatorname{Map}_{\mathcal P(C)}(R,F).
\]
It imposes a coherent limit condition on the whole sieve. When
there are pullbacks of the covering arrows, the condition also
admits a Čech form by replacing $R$ by the realization of the
Čech nerve of $\coprod_i y(c_i)\to R$. This identification
takes place already in presheaves, where effective epimorphisms
and colimits are calculated pointwise.

\begin{theorem}[Site description of inverse images]
\label{thm:geo-site-inverse}
Let $C$ be a small $\infty$-category with finite limits and a
Grothendieck topology. Let
$j\colon C\to\operatorname{Shv}(C;\mathcal S)$ be Yoneda followed
by sheafification. For every $\infty$-topos $\mathcal Y$,
precomposition by $j$ gives an equivalence of
$\operatorname{Geom}(\mathcal Y,\operatorname{Shv}(C;\mathcal S))$
with the full subcategory of
$\operatorname{Fun}(C,\mathcal Y)$ on functors $A$ satisfying
the following conditions.
\begin{enumerate}
\item The functor $A$ preserves finite limits.
\item For every family $\{c_i\to c\}$ generating a covering sieve,
the map $\coprod_i A(c_i)\to A(c)$ is an effective epimorphism.
\end{enumerate}
\end{theorem}

\begin{proof}
The free cocompletion property identifies $A$ with its
colimit-preserving extension
$\overline A\colon\mathcal P(C)\to\mathcal Y$. Under this
identification the extension is left exact precisely when $A$ is
left exact. Necessity uses the left exactness of Yoneda. For
sufficiency apply the finite limit extension theorem
\cite[Prop.~6.1.5.2]{htt} to the finitely complete domain $C$.

The extension then takes the image factorization
\[
 \coprod_i y(c_i)\longrightarrow R\longrightarrow y(c)
\]
to an image factorization. Thus the condition on the generating
family is equivalent to $\overline A$ inverting the covering sieve.
By the universal property of localization, $\overline A$ factors
if and only if all these inclusions are inverted. The descended
functor is left exact since it is the restriction of $\overline A$
along the right adjoint inclusion of sheaves, which preserves limits.
It preserves colimits by the localization property or by taking
the corresponding right adjoint on presheaves, whose image lies in
sheaves. These constructions are equivalences of full functor
subcategories and preserve natural transformations. This proves
the equivalence including its morphisms
\cite[Prop.~6.2.3.20]{htt}.
\end{proof}

For a presentation by an arbitrary accessible left exact
localization $L\colon\mathcal P(C)\to\mathcal X$, the first part
of the argument survives without any topology on $C$. Restriction
to $C$ identifies inverse images with those functors for which
$\overline A$ is left exact and sends the $L$-equivalences to
equivalences. When $L$ is displayed by a set of generators,
testing those generators suffices since $\overline A$
preserves colimits. An arbitrary localization can impose
relations beyond those specified by covering sieves. Thus the
general inverse image theorem starts with the actual localization,
which may include a condition at a map between presheaves that
is not a sieve inclusion.

For an ordinary site $(C,J)$ this construction provides a
canonical lift of its ordinary topos. The $0$-truncated objects
of $\operatorname{Shv}(C;\mathcal S)$ are exactly its sheaves
of sets. To see this, evaluation on the sheafified representables
tests whether a sheaf takes $0$-truncated values. Conversely,
mapping into a sheaf taking such values can be tested on
a colimit of sheafified representables and then gives a limit of
$0$-truncated spaces. The canonical lift depends only on the
ordinary topos. In the terminology of higher topos theory it
is the $1$-localic lift, and inverse images from an
$\infty$-topos into it are detected by the corresponding functors
on $0$-truncated objects
\cite[Prop.~6.4.5.7]{htt}. One may therefore extend an ordinary
geometric morphism between Grothendieck topoi to the
$1$-localic lifts. Recovering an arbitrary $\infty$-topos from its
$0$-truncated objects is a stronger assertion and generally fails.

\section{Étale maps and dependent substitution}
\index{etale geometric morphism@étale geometric morphism}
\index{Beck--Chevalley comparison}

For a map $u\colon U\to V$ in an $\infty$-topos $\mathcal X$,
base change has adjoints on both sides
\[
 u_!\dashv u^*\dashv u_*
 \qquad
 \left(
 u^*\colon\mathcal X_{/V}\longrightarrow\mathcal X_{/U}
 \right).
\]
The first adjoint composes an arrow to $U$ with $u$.
The second exists because pullback preserves colimits. Pullback
preserves finite limits as well. The resulting map of topoi
$\mathcal X_{/U}\to\mathcal X_{/V}$ is étale. In particular
$\pi_U\colon\mathcal X_{/U}\to\mathcal X$ denotes the map obtained
from $U\to1_{\mathcal X}$. More generally a geometric morphism
to $\mathcal X$ is called étale when it is equivalent over
$\mathcal X$ to some $\pi_U$. In this use of the word $U$
may have higher homotopy in its fibers.

Composition and pullback satisfy a relative projection formula.
For $A\to B$ in $\mathcal X_{/V}$ and $D\to u^*B$ in
$\mathcal X_{/U}$, there is a natural equivalence
\[
 u_!\big(u^*A\times_{u^*B}D\big)
   \simeq A\times_B u_!D.
\]
Indeed the underlying object on either side is
$A\times_B D$ in $\mathcal X$, with its composite map to $V$.
If a square in $\mathcal X$ is a pullback and the parallel
horizontal maps are $g'\colon U'\to U$ and $g\colon V'\to V$,
then $g^*u_!\simeq u'_!g'^*$ by the same calculation. For
the right adjoints the comparison
$g^*u_*\simeq u'_*g'^*$ follows by testing against
$T\in\mathcal X_{/V'}$ and using
$u^* g_!T\simeq g'_!(u'^*T)$. These identities describe the
two ways to integrate a family after change of base. They extend
Frobenius identities on subobjects, where the relative
$(-1)$-truncation of dependent sum is read as existential
quantification.

\begin{theorem}[The slice classifies sections]
\label{thm:geo-sections}
For $U\in\mathcal X$, the category of geometric morphisms
$\mathcal Y\to\mathcal X_{/U}$ over the hom category
$\operatorname{Geom}(\mathcal Y,\mathcal X)$ is represented by
the left fibration classified by
\[
 \operatorname{Geom}(\mathcal Y,\mathcal X)\longrightarrow\mathcal S,
 \qquad
 f\longmapsto
 \operatorname{Map}_{\mathcal Y}(1_{\mathcal Y},f^*U).
\]
In particular, a lift of $f$ to a geometric morphism
$\mathcal Y\to\mathcal X_{/U}$ over $\mathcal X$ amounts to
a section of $f^*U$, with all paths between sections.
\end{theorem}

\begin{proof}
The object $\pi_U^*U$ is $U\times U\to U$ and carries the
diagonal section $\delta$ over $U$. A geometric morphism $h$
from $\mathcal Y$ to $\mathcal X_{/U}$ gives the pair
$(h^*\pi_U^*,h^*\delta)$. Conversely, a pair $(f^*,s)$ with
$s\colon1_{\mathcal Y}\to f^*U$ induces the functor
\[
 (A\xrightarrow{a}U)\longmapsto
    1_{\mathcal Y}\times_{f^*U} f^*A.
\]
Here the first step applies $f^*$ in the slice. The second
step is pullback along $s$. Both steps preserve small
colimits and finite limits. For the first step, finite limits
follow from left exactness and colimits in a slice are computed
in its ambient category. For the second step, colimits follow
from universality and limits from the right adjoint description
of pullback as right adjoint to composition. Consequently this
formula defines a geometric inverse image.

For any $A\to U$ there is a pullback square in the slice
expressing it as
$\pi_U^*A\times_{\pi_U^*U}1_{\mathcal X_{/U}}$, with the
right-hand map determined by $\delta$. An inverse image preserves
this square. Applying the construction to $(h^*\pi_U^*,h^*\delta)$
therefore recovers $h^*$ naturally on objects and maps.
The reverse composite recovers $(f^*,s)$ by evaluating on
objects $C\times U$ and then on the diagonal. These natural
comparisons also hold for transformations. In detail, a
transformation between pairs consists of a transformation
$\alpha\colon f^*\to g^*$ and a specified path
$\alpha_U s\simeq t$. It induces the transformation of the
pullback formulas by the universal property of the pullback.
Restriction to $\pi_U^*$ and evaluation at $\delta$ recovers
the original pair of data. Since these operations preserve
parametrized families of transformations, the comparison is
an equivalence of $\infty$-categories. The category of these
pairs is the Grothendieck construction of the displayed
space-valued functor. See also \cite[Prop.~6.3.5.5 and
Cor.~6.3.5.6]{htt}.
\end{proof}

The section classifier gives two useful refinements. First, if
$g\colon\mathcal Y\to\mathcal X$ is geometric, then base change
of $\pi_U$ has source $\mathcal Y_{/g^*U}$. A test geometric
morphism $f\colon\mathcal Z\to\mathcal Y$ lifts to that slice
exactly by specifying a section of $f^*g^*U$, which also
describes a compatible lift of $gf$ to $\mathcal X_{/U}$. Second,
there is an equivalence of $\infty$-categories
\[
 \mathcal X \simeq
 \big(\mathbf{Top}_{\infty}/\mathcal X\big)_{\mathrm{et}},
 \qquad U\longmapsto \pi_U.
\]
Essential surjectivity is the definition of étale. For full
faithfulness use the section classifier with
$\mathcal Y=\mathcal X_{/V}$ and $f=\pi_V$.
A section of $V\times U\to V$ in that slice is precisely a
map $V\to U$. Spaces of such sections compute the relative
mapping spaces. This result identifies a whole category of
geometric tests over $\mathcal X$ with the objects of
$\mathcal X$, without choosing a site or a point.

\section{Internal classifying objects and external classifiers}
\index{classifying topos}

An internal group $G$ in $\mathcal X$ admits a delooping
$BG\in\mathcal X$. For every $f\colon\mathcal Y\to\mathcal X$,
$f^*BG\simeq B(f^*G)$ because the bar construction uses finite
products and geometric realization. Combining the slice
classifier with torsor classification gives the space of lifts
of $f$ through $\mathcal X_{/BG}$ as the space of
$f^*G$-torsors in $\mathcal Y$. The universal torsor there is
classified by the diagonal section of $\pi_{BG}^*BG$.
For an arrow $V\to1$ in $\mathcal X$ the same rule describes
$G$-torsors in the slice $\mathcal X_{/V}$ by maps $V\to BG$.
For a general geometric $f$, its inverse image replaces every
coefficient object before the section is taken.

The object \(BG\) represents torsors on objects of \(\mathcal X\).
The slice \(\mathcal X_{/BG}\) represents the choice of a torsor
over a geometric map into \(\mathcal X\). In both statements the group
is fixed in the base. The category of \(G\)-objects includes arbitrary
equivariant maps and will figure in the presentation of the next chapter.

\chapter{Groupoid presentations of higher topoi}
\label{ch:presentations}

A presentation gives two kinds of information. It identifies a
domain on which objects can be described locally, and it says
how to identify their pullbacks on overlaps. The ordinary
Joyal--Tierney theorem provides both pieces using locales.
The higher construction below presents slices by étale
geometric morphisms and keeps track of the type of each overlap.
The difference becomes visible even when the atlas itself is
the topos of spaces.

\section{The descent formula for a diagram of slices}
\index{van Kampen colimit}

A colimit in an $\infty$-topos is van Kampen in the following
sense. For a small diagram $U\colon I\to\mathcal X$ with colimit
$V$, restriction along the colimit cone gives an equivalence
of $\infty$-categories
\begin{equation}
 \mathcal X_{/V}\ \simeq\
 \lim_{i\in I^{\mathrm{op}}}\mathcal X_{/U_i}.
 \label{eq:higher-slice-vankampen}
\end{equation}
Transition functors on the right are pullbacks. Equivalently,
a Cartesian family $A_i\to U_i$, meaning one whose transport
squares are pullbacks, has a realization
$\operatorname{colim}_i A_i\to V$ whose pullbacks are the given family.
This is the descent form of the higher Giraud axioms
\cite[Thm.~6.1.3.9 and Prop.~6.1.3.10]{htt}.
The realization description also shows why preserving pullbacks
by colimits of slices is a stronger condition than mere
universality of colimits. Universality gives the recovery of a
slice object from an already existing pullback family. Descent
asserts that a compatible family given independently of a global
object descends.

\begin{theorem}[Colimits of slices as higher topoi]
\label{thm:geometric-colimit-slices}
With the preceding hypotheses, there is an equivalence
\[
 \operatorname*{colim}_{i\in I}^{\mathbf{Top}_{\infty}}
         \mathcal X_{/U_i}\ \simeq\ \mathcal X_{/V}.
\]
For every $\infty$-topos $\mathcal Z$, the induced comparison
\[
 \operatorname{Geom}(\mathcal X_{/V},\mathcal Z)
 \ \longrightarrow\
 \lim_{i\in I^{\mathrm{op}}}
        \operatorname{Geom}(\mathcal X_{/U_i},\mathcal Z)
\]
is an equivalence of $\infty$-categories.
\end{theorem}

\begin{proof}
Apply $\operatorname{Fun}(\mathcal Z,-)$ to
\eqref{eq:higher-slice-vankampen}. The limit of functor categories
classifies a coherent family of functors to the slices.
It remains to restrict to geometric inverse images. The
projections from the limit preserve finite limits and
small colimits, since all the transition functors do.
For example, a diagram of compatible families has its
componentwise colimit formed in each slice. A transition
functor carries each component colimit to the corresponding
colimit, so the colimit inherits the required compatibility
and satisfies the universal property. The same argument
applies to finite limits. Joint conservativity of projections
then says that a functor into the limit is left exact and
colimit-preserving exactly when each component is. Natural
transformations are restricted through the same equivalence
of functor categories. Taking cores yields the mapping space
universal property of the colimit of topoi. This is the fixed
base case of the more general colimit theorem for étale
geometric morphisms \cite[Thm.~6.3.5.13]{htt}.
\end{proof}

\section{From a groupoid to its quotient}
\index{groupoid object!effectiveness}
\index{groupoid object!actions}

Let $G_\bullet\colon\Delta^{\mathrm{op}}\to\mathcal X$ be
a groupoid object, in the Segal sense with all arrows
invertible. Put $Q=|G_\bullet|$. The Segal identities and invertibility concern an internal diagram in $\mathcal X$.
They imply that $G_\bullet$ is the Čech nerve of
$G_0\to Q$. We recall a proof using the van Kampen
property because its argument makes explicit the role of
invertibility.

\begin{proposition}[Effectiveness of internal groupoids]
\label{prop:geo-effective-groupoid}
The map $G_0\to Q$ is an effective epimorphism, and there
are compatible equivalences
\[
 G_n \simeq
 \underbrace{G_0\times_Q\cdots\times_QG_0}_{n+1}.
\]
\end{proposition}

\begin{proof}
Define a simplicial object $E_\bullet$ by $E_n=G_{n+1}$,
where the simplicial operators retain the first vertex and
act on the remaining vertices. In degree $n$, its map to
$G_n$ forgets that first vertex. There is also an augmentation
$E_\bullet\to G_0$ given by the retained vertex. Duplicating
the retained vertex is an extra degeneracy contracting this
augmentation, so $|E_\bullet|\simeq G_0$.

The map $E_\bullet\to G_\bullet$ is Cartesian. To check this,
first pull it back along a vertex of a simplex $G_n$.
Over the first vertex of the remaining chain, the assertion
is a Segal identity since the extra data is the one incoming
arrow. Over any other vertex, the incoming arrow can be
transported forward along the chosen segment of that chain.
Composition by an invertible arrow is an equivalence on the
corresponding space of arrows, with inverse given by
composition by its inverse. This argument is functorial on
generalized points and proves the pullback identity for any
chosen vertex. To check an arbitrary simplicial operator
$G_n\to G_m$, evaluate the two candidate pullbacks over a
vertex of $G_m$. Both identify with the same pullback of
$E_0\to G_0$ along the composite vertex of $G_n$.

Descent now identifies $E_0$ with the pullback of
$|E_\bullet|$ along $G_0\to Q$. Its two projections are the
endpoints of the arrow in $G_1$. Hence
$G_1\simeq G_0\times_Q G_0$. Compatibility with longer
chains follows by the Segal maps and their composition
coherences. The augmented Čech nerve therefore realizes to
$Q$, proving effectiveness. This is the shift argument
of \cite[\S6.10]{rezk} in the notation of internal groupoids.
\end{proof}

A $G_\bullet$-object is a Cartesian transformation
$A_\bullet\to G_\bullet$. It is given first by an object
$A_0\to G_0$, then by transport of this object along
arrows of $G_1$, with composition homotopies on $G_2$
and their higher compatibility. Equivariant morphisms are
maps of such transformations. The corresponding
$\infty$-category will be denoted $\operatorname{Act}_{\mathcal X}(G_\bullet)$.
Applying \eqref{eq:higher-slice-vankampen} gives
\begin{equation}
 \operatorname{Act}_{\mathcal X}(G_\bullet)
 \ \simeq\ \lim_{[n]\in\Delta}\mathcal X_{/G_n}
 \ \simeq\ \mathcal X_{/Q}.
 \label{eq:higher-quotient-actions}
\end{equation}
The inverse sends $A\to Q$ to the family $A\times_QG_n$.
The forward map sends a Cartesian family to its realization.
Under Theorem~\ref{thm:geometric-colimit-slices}, the
right-hand side is also the colimit of the simplicial
diagram of étale geometric morphisms $\mathcal X_{/G_n}$.
The two variance conventions agree because the geometric
morphisms point in the direction opposite to their inverse
image functors.

For the special case of $U\to1_{\mathcal X}$ effective epi,
let $G_n=U^{n+1}$. Then $Q\simeq1_{\mathcal X}$ and
\eqref{eq:higher-quotient-actions} recovers $\mathcal X$ itself
from the Čech diagram of $\mathcal X_{/U}\to\mathcal X$.
Indeed Theorem~\ref{thm:geo-sections} identifies
\[
 \underbrace{\mathcal X_{/U}\times_{\mathcal X}\cdots
                 \times_{\mathcal X}\mathcal X_{/U}}_{n+1}
 \simeq \mathcal X_{/U^{n+1}},
\]
where the fiber products on the left are in
$\mathbf{Top}_{\infty}$. The inverse image of this projection
is conservative since evaluation of a Cartesian family at
degree zero detects equivalences. Conversely, if
$\pi_U^*$ is conservative, it reflects the equivalence
obtained by pulling back $\operatorname{im}(U\to1)\to1$
along $U$. This image must be all of $1$, so $U\to1$ is
an effective epimorphism. For étale geometric morphisms,
surjectivity in the sense of conservative inverse image
thus agrees with being induced by an effective epimorphism
of objects. All the formulas hold in a slice over any
specified base object.

Presentations here are invariant under refinement in a
concrete sense. If $W\to U\to1_{\mathcal X}$ are effective
epimorphisms, their Čech nerves have a map
$W^{\bullet+1}\to U^{\bullet+1}$. Pullback on actions is
an equivalence because under \eqref{eq:higher-quotient-actions}
it identifies with the identity functor of $\mathcal X$.
Two covers $U\to1_{\mathcal X}$ and $V\to1_{\mathcal X}$
admit a common refinement by $U\times V$. Both projections
are covers by pullback stability. There are typically many
different simplicial diagrams presenting the same topos.

\section{The classical localic theorem}
\index{Joyal--Tierney theorem}
\index{localic groupoid}

For a locale $L$, let $\mathcal O(L)$ be its frame. A map
of locales $f\colon L\to M$ is open when
$f^*\colon\mathcal O(M)\to\mathcal O(L)$ admits a left adjoint
$f_!$ satisfying the Frobenius identity
\[
 f_!(a\wedge f^*b)=f_!(a)\wedge b.
\]
An étale locale over $M$ is one locally isomorphic over $M$
to open sublocales of $M$. Its sheaf of sections is an
ordinary sheaf of sets on $M$. An open localic groupoid
consists of locales of objects and arrows, groupoid
operations in locales, and open source and target maps.

The Joyal--Tierney theorem provides, for every ordinary
Grothendieck topos $\mathcal E$, an open localic groupoid $G$
with $\mathcal E$ equivalent to the category of étale
$G$-objects. These are étale locales over $G_0$ with an
action along $G_1$, or equivalently sheaves of sets on
$G_0$ with the corresponding equivariance. An ingredient
of the theorem supplies a locale $L$ with an open
effective-descent geometric morphism
$\operatorname{Shv}(L;\operatorname{Set})\to\mathcal E$.
Take its Čech nerve by pseudopullbacks of ordinary
topoi. The map from a localic topos to an arbitrary
Grothendieck topos is localic in the relative sense. Indeed
its source has subterminal generators, and these are
subobjects of the inverse image of the terminal object.
Base change preserves relative localicity. Consequently the
iterated overlaps are localic over the localic source and
therefore localic. Openness under base change gives the
openness of the groupoid. Effective descent identifies
$\mathcal E$ with the category of sheaves on $G_0$ endowed
with the required action. The existence of the localic cover
and the effective-descent theorem are substantive parts of
the classical result \cite[Chs.~VII--VIII]{joyal-tierney};
see also \cite[Thms.~7.10.1 and 7.10.7]{bj}.

There are two uses of étaleness here. In the ordinary localic
result, étale locales carry the sheaves on which the groupoid
acts. The atlas of the target topos is supplied as an open
geometric morphism of effective descent. In the higher
relative result, an atlas $\mathcal X_{/U}\to\mathcal X$ is
itself an étale geometric morphism. Its source need not be
localic. The theory of internal groupoid actions proves the
relative presentation once a cover is chosen. In the
classical theorem, the existence of an external atlas with
localic overlaps is additional content.

\section{When a higher étale atlas has localic overlaps}
\index{localic topos}

Call an $\infty$-topos $0$-localic if it is equivalent to
$\operatorname{Shv}(L;\mathcal S)$ for a locale $L$. Its locale
is recovered from the subobjects of $1$
\cite[\S7.3.2]{htt}. For a space $A$,
the subobjects of the terminal object in $\mathcal S_{/A}$
are indexed by subsets of $\pi_0 A$. To see this, a subobject
has fiber either empty or contractible, and this property
is invariant under transport along paths. Conversely each
union of path components gives such a subobject.

\begin{lemma}
\label{lem:geo-localic-space}
The $\infty$-topos $\mathcal S_{/A}$ is $0$-localic exactly when
$A$ is $0$-truncated.
\end{lemma}
\begin{proof}
For a discrete space $A$, a family over $A$ is a sheaf of spaces
on the discrete locale with set of points $A$. For the converse,
the reconstructed locale would have frame the power set of
$\pi_0 A$. Over each nonempty connected component $A_0$ the
corresponding open slice would then be equivalent to $\mathcal S$.
Such an equivalence must identify the constant-object functor
$\mathcal S\to\mathcal S_{/A_0}$ with an equivalence, since that
functor is the unique geometric inverse image out of $\mathcal S$
up to equivalence. For $a\in A_0$, the object $\{a\}\to A_0$ has
nonempty fiber at $a$. A constant object with nonempty fiber
has a global section. A section of $\{a\}\to A_0$ makes
$\operatorname{id}_{A_0}$ homotopic to the constant map at $a$,
so $A_0$ is contractible. This holds for each connected
component, giving a discrete homotopy type.
\end{proof}

\begin{theorem}[Localic étale Čech presentations over spaces]
\label{thm:geo-localic-bound}
Let $B$ be a small space. The $\infty$-topos $\mathcal S_{/B}$
admits a surjective étale geometric morphism for which both
the source and the double overlap are $0$-localic if and
only if $B$ is $1$-truncated. In that case every higher overlap
of the chosen atlas can also be taken $0$-localic.
\end{theorem}
\begin{proof}
Write the covering source as $\mathcal S_{/U}$, with map to
$\mathcal S_{/B}$ induced by $U\to B$. Surjectivity implies
$U\to B$ is effective epi, hence surjective on path components.
The source is $0$-localic only if $U$ is discrete by
Lemma~\ref{lem:geo-localic-space}. For a discrete $U$, the
space $U\times_B U$ is the disjoint union, over pairs $u,v\in U$,
of spaces of paths in $B$ from the image of $u$ to the image
of $v$. Since the double overlap is
$\mathcal S_{/(U\times_B U)}$, its localicity forces these path
spaces to be discrete. Using $u=v$ and one point over each
component shows that each loop space of $B$ is $0$-truncated.
Equivalently, $B$ is $1$-truncated.

Conversely, take $U$ discrete with one chosen point of $B$ in each
component. This is an effective epimorphism, including the
empty case. Its higher fiber products over $B$ have discrete
homotopy types, since a tuple of points is accompanied by a
tuple of paths in the $1$-truncated space $B$.
Lemma~\ref{lem:geo-localic-space} identifies all their slices
with $0$-localic $\infty$-topoi.
\end{proof}

For an infinite discrete group $G$ and $B=BG$, the atlas
coming from $*\to BG$ has $n$th term $\mathcal S_{/G^n}$.
Equation~\eqref{eq:higher-quotient-actions} gives its category
of actions as $\operatorname{Fun}(BG,\mathcal S)$. Its
$0$-truncated objects are ordinary $G$-sets. Taking instead
$B=K(\mathbb Z,2)$ yields the atlas
$\mathcal S\to\mathcal S_{/B}$ whose first overlap is
$\mathcal S_{/K(\mathbb Z,1)}$. That overlap is $1$-localic
since it is a presheaf category on an ordinary groupoid,
and it fails to be $0$-localic by Lemma~\ref{lem:geo-localic-space}.
Moreover $\mathcal S_{/B}$ and $\mathcal S$ have the same
ordinary topos of $0$-truncated objects. Straightening
identifies the first ordinary topos with set-valued functors
on $B$, which factor through the fundamental groupoid.
For $K(\mathbb Z,2)$ that groupoid is contractible. The two
higher topoi are inequivalent by Lemma~\ref{lem:geo-localic-space}.
Theorem~\ref{thm:geo-localic-bound} concerns the specific
class of surjective étale geometric morphisms and does not
assume a theorem about all other notions of higher atlas.

\section{Hyperdescent and the role of truncation}
\index{hypercover}
\index{hypercompletion}

In any $\infty$-topos an augmented simplicial object
$V_\bullet\to V$ is a hypercover if each map to its
relative matching object is an effective epimorphism.
In degree zero the matching object is $V$. In degree one
it is $V_0\times_VV_0$. In higher degrees it is the
object of compatible boundaries relative to $V$.
Each matching condition asks for local existence of a lift.
When the augmentation is a Čech nerve, there is a stronger
identity. The levels are specified iterated pullbacks and
the geometric realization of a cover is exactly $V$.

The realization of an arbitrary hypercover maps to $V$ by an
$\infty$-connective morphism
\cite[Lem.~6.5.3.11]{htt}. This means its relative
$n$-truncation is terminal for every finite $n$. Applying
\eqref{eq:higher-slice-vankampen} to $V_\bullet$ identifies
$\lim_\Delta\mathcal X_{/V_n}$ with
$\mathcal X_{/|V_\bullet|}$. Consequently the restriction
comparison for all objects over $V$ is an equivalence
precisely when $|V_\bullet|\to V$ is an equivalence.
Indeed, if the restriction $g^*$ for a map $g$ is an
equivalence, so is its left adjoint $g_!$. Applying $g_!$
to the terminal object in the source slice gives $g$ as
an object of the target slice, so $g$ is terminal there.
Conversely an equivalence $g$ induces an equivalence of slices.

An $\infty$-topos is hypercomplete if every $\infty$-connective
map is an equivalence. Hypercompletion is the left exact
localization which forces these equivalences. Each
$\infty$-connective morphism gives a constant simplicial
hypercover in its target slice, according to the equivalent
iterated-diagonal test for $\infty$-connectivity
\cite[Lem.~6.5.3.5]{htt}. Thus effectiveness of
all hypercovers is equivalent to hypercompleteness
\cite[Thm.~6.5.3.12]{htt}. A bounded truncated object
already satisfies the extra locality condition. Mapping
into an $n$-truncated object cannot distinguish an
$\infty$-connective map from an equivalence, by orthogonality
to $n$-truncated maps. In particular,
passing to ordinary sheaves of sets cannot test whether
a given $\infty$-topos is hypercomplete.

A concrete instance of the distinction is the canonical
$1$-localic lift of continuous $\mathbb Z_p$-sets. One can
present it by the category of transitive quotients
$\mathbb Z_p/p^n\mathbb Z_p$ with its atomic topology.
This is an $\infty$-topos of ordinary sheaves of spaces
in the sense of covering sieves, and it fails to be
hypercomplete by Wieland's example
\cite[Warning~7.2.2.31]{htt}. The site is ordinary and
its underlying ordinary topos already has a classical
group-theoretic presentation. Neither fact removes the
hyperdescent distinction. The higher quotient
\eqref{eq:higher-quotient-actions} uses an effective
groupoid object, so it works in that topos before
hypercompletion as well as in hypercomplete examples.

\begin{proposition}[Relative hypercompletion]
\label{prop:relative-hypercompletion}
Let $L\colon\mathcal X\to\mathcal X^\wedge$ be hypercompletion.
For every $U\in\mathcal X$, the induced functor
\[
 L_{/U}\colon\mathcal X_{/U}\longrightarrow
                  (\mathcal X^\wedge)_{/LU}
\]
is the hypercompletion of the slice. In particular, applying
$L$ to a hypercover $V_\bullet\to V$ gives a slice
comparison
\[
 (\mathcal X^\wedge)_{/LV}
 \ \simeq\
 \lim_{[n]\in\Delta}
              (\mathcal X^\wedge)_{/LV_n}.
\]
\end{proposition}

\begin{proof}
Let $i\colon\mathcal X^\wedge\hookrightarrow\mathcal X$ be
the fully faithful inclusion. For $T\to LU$, the right
adjoint of $L_{/U}$ is the object
$U\times_{iLU}iT\to U$. Indeed, a map into this pullback
over $U$ amounts under $L\dashv i$ to a map into $T$
over $LU$. The counit is invertible by left exactness of
$L$ and the counit equivalence $LiT\simeq T$. It follows
that $L_{/U}$ is a localization. A morphism in the slice
is inverted precisely when its underlying morphism in
$\mathcal X$ is inverted by $L$. This is the class of
$\infty$-connective maps by the strong saturation theorem
\cite[Prop.~6.5.2.8]{htt}. Connectivity in the slice agrees
with the connectivity of the underlying map. For example,
the slice over its codomain in $\mathcal X_{/U}$ identifies
with the slice over that same codomain in $\mathcal X$,
so the relative truncation tests are identical. This
identifies the localization as the desired hypercompletion.

Finally, $L$ preserves the realization of $V_\bullet$
and sends its $\infty$-connective augmentation to an
equivalence. Apply
\eqref{eq:higher-slice-vankampen} inside $\mathcal X^\wedge$.
The result can also be read as descent for relative
hypercomplete objects. The base object in the formula
is $LV$, even when $V$ itself has not been assumed
hypercomplete.
\end{proof}

In the $\mathbb Z_p$ example the constant hypercover of a
nonequivalence supplied by infinite connectivity has a slice
comparison that fails before hypercompletion. The proposition
makes the comparison an equivalence after applying $L$ to
its entire augmentation. The higher Čech theorem continues
to apply to actual groupoid quotients in the original topos.

\chapter{Reconstruction and comparison}\label{ch:comparison}
\index{reconstruction}\index{fiber functor}

A classification theorem identifies a category of forms or coverings
from a symmetry object and chosen coefficients. A reconstruction
theorem asks what symmetry can be read back from that category.
This direction is sensitive to the fiber functor, the availability
of representing objects, and the tests used to define equivalence.
We finish by comparing the different reconstruction problems.

\section{Recovering the space of fibers}

Let \(B\) be a small space and put
\(\mathcal L(B)=\Fun(B,\SSet)\). Write
\(\mathrm{ev}_b:\mathcal L(B)\to\SSet\) for evaluation.
A path from \(b\) to \(c\) induces a natural equivalence
\(\mathrm{ev}_b\to\mathrm{ev}_c\). The space of such natural
equivalences remembers all paths and their homotopies.

\begin{theorem}[Reconstruction from the evaluations]
\label{thm:comparison-evaluation}
The functor
\[
  B\longrightarrow\Fun(\mathcal L(B),\SSet),
  \qquad b\longmapsto\mathrm{ev}_b
\]
is fully faithful. In particular, as group objects in spaces,
\[
  \Aut(\mathrm{ev}_b)\simeq\Omega_b B.
\]
These assertions concern natural transformations in the full
functor category.
\end{theorem}

\begin{proof}
For \(b\in B\), the covariant representable
\(R_b=\Map_B(b,-)\) is an object of \(\mathcal L(B)\).
The covariant version of Yoneda gives
\(\mathrm{ev}_b\simeq\Map_{\mathcal L(B)}(R_b,-)\).
A natural transformation from evaluation at \(b\) to evaluation
at \(c\) is, by a second application of Yoneda, a point of
\(\Map_{\mathcal L(B)}(R_c,R_b)\simeq R_b(c)\).
The equivalence is natural in \(b,c\), and it respects
composition because the two appearances of the contravariant
representing-object assignment reverse order twice.
Equivalently, a path \(u:b\to c\) acts on each \(F\) by \(F(u)\),
which agrees with this identification. Every arrow in \(B\)
is invertible, so the equivalence of mapping spaces identifies
the full space of natural transformations between the two
evaluations with its subspace of equivalences.
\end{proof}

This proof explains the regular-action formula. On writing
\(B=BG\), the representable \(R_*\) is the regular left
\(G\)-space. Its endomorphisms are right multiplications,
which compose in the opposite order. Passing from the
representing object to the evaluation functor reverses the
order again. The group on the right side of
Theorem~\ref{thm:comparison-evaluation} is therefore compatible
with covariant monodromy. For the groupoid \(B\), the
pointed reconstruction only uses one evaluation, while the
full family of evaluations retains the components and the
paths between them.

\begin{proposition}[Reconstruction from all geometric points]
\label{prop:comparison-points}
The core of the category of colimit-preserving, finite-limit-preserving
functors \(\SSet_{/B}\to\SSet\) is equivalent to \(B\). Under
straightening, its objects are the evaluations of
Theorem~\ref{thm:comparison-evaluation}.
\end{proposition}

\begin{proof}
The universal property of the slice topos, as in
Chapter~\ref{ch:geometric}, identifies geometric morphisms
\(\mathcal T\to\SSet_{/B}\) with pairs consisting of a geometric
morphism \(a:\mathcal T\to\SSet\) and a section of \(a^*B\).
For \(\mathcal T=\SSet\), the space of choices of \(a\) is
contractible. A colimit-preserving functor from spaces is
determined by the image of the point, and left exactness forces
that image to be terminal. The section space of \(B\) is \(B\).
The inverse image associated to \(b\) pulls back a family along
\(b:*\to B\), and that is evaluation under straightening.
\end{proof}

The distinction between this result and a general theorem on
points of topoi is substantial. The slice of spaces has
specially chosen geometric points, determined by the family
classifier \(B\). A general higher topos has a shape defined
by constant-object tests. That definition does not assert that
its points are the components of its shape, or that its
points distinguish all objects.

\section{What truncation preserves}

The preceding calculation is also valid with truncated
coefficients. For an integer \(n\geq0\), let
\(\mathcal L_n(B)=\Fun(B,\SSet_{\leq n})\). Denote the
evaluation functor on \(\mathcal L_n(B)\) by
\(\mathrm{ev}_b^n\), giving its output as a space.

\begin{proposition}[Truncated reconstruction]
\label{prop:comparison-trunc}
There is a natural equivalence
\[
\Map_{\Fun(\mathcal L_n(B),\SSet)}
    (\mathrm{ev}_b^n,\mathrm{ev}_c^n)
   \simeq\tau_{\leq n}\Map_B(b,c).
\]
Every component of the left side consists of equivalences.
Restricted evaluation consequently defines a fully faithful
functor \(\tau_{\leq n+1}B\to\Fun(\mathcal L_n(B),\SSet)\).
\end{proposition}

\begin{proof}
Objectwise truncation is left adjoint to the inclusion
\(\mathcal L_n(B)\hookrightarrow\mathcal L(B)\).
Hence \(\mathrm{ev}_b^n\) is represented by
\(R_b^n=\tau_{\leq n}R_b\) in \(\mathcal L_n(B)\).
The two Yoneda identifications give
\(\Map(\mathrm{ev}_b^n,\mathrm{ev}_c^n)
\simeq \Map_{\mathcal L_n(B)}(R_c^n,R_b^n)
\simeq R_b^n(c)\).
The components of \(R_b^n(c)\) come from actual paths in \(B\),
so all corresponding natural transformations are equivalences.
The full subcategory spanned by the evaluations is thus an
\((n+1)\)-truncated groupoid. The evaluation functor factors
through \(\tau_{\leq n+1}B\) by its truncation universal
property. The mapping-space description identifies the
factorization as fully faithful.
\end{proof}

In particular, set-valued monodromy recovers the fundamental
groupoid and its groups of loops at chosen points. A space of
coefficients truncated in a higher degree records one additional
degree in the classifying groupoid. The regular representable
need not lie in the subcategory of finite coefficients. The
proof above therefore gives no reconstruction of the uncompleted
base from finite tests. The pro-space mapping formula and
finite-range hypotheses from Chapters~\ref{ch:profinite}
and~\ref{ch:shape} are the relevant comparison in that setting.

For infinite discrete groups, the loss can already occur with
set coefficients. The additive group \(\mathbb Q\) has no
nontrivial finite quotient. If a map \(\mathbb Q\to H\) has
finite image of order dividing \(m\), write any \(x\) as \(my\).
Its image is the \(m\)th power of the image of \(y\), hence
the identity. Any action of \(\mathbb Q\) on a finite set
is trivial, since it factors through a finite permutation group.
In contrast the regular action recovers \(\mathbb Q\) itself by
Proposition~\ref{prop:comparison-trunc}. The choice of coefficient
category therefore changes the reconstruction even for an ordinary
group.

\section{A comparison of the classification results}

The constructions used throughout the book can be read in the
following order. Choose a base and the objects considered over
it. Specify the splitting condition. Determine the overlap
diagram under which such objects can descend. Then decide whether
the overlap diagram has a smaller groupoid, pro-object, or directed
category as a presentation. The outputs below indicate which
of these steps require hypotheses.

\begingroup
\small
\renewcommand{\arraystretch}{1.19}
\begin{longtable}{@{}
>{\raggedright\arraybackslash}p{0.205\textwidth}
>{\raggedright\arraybackslash}p{0.347\textwidth}
>{\raggedright\arraybackslash}p{0.365\textwidth}@{}}
\toprule
Input & Key hypothesis & Classification \\
\midrule
\endfirsthead
\toprule
Input & Key hypothesis & Classification \\
\midrule
\endhead
An internal cover \(U\to B\) &
Effective epimorphism in an \(\infty\)-topos &
Objects over \(B\) as cartesian actions of the overlap groupoid.
Chapter~\ref{ch:descent}. \\
An adjunction \(L\dashv R\) and a cover \(U\to B\) &
Admissibility in slices and effective descent for the chosen maps &
Objects split by \(U\) as a limit of slices over the image overlap
diagram. Additional exactness makes that diagram a groupoid.
Chapter~\ref{ch:reflections}. \\
A space, a pro-space, or the shape of a topos &
Representable base for ordinary straightening or the coefficient
and locality hypotheses of higher Galois theory &
Monodromy on the space, stage representations, or locally constant
objects, respectively. Chapters~\ref{ch:profinite}--\ref{ch:shape}. \\
A spectral algebra \(R\to S\) &
The flat or stable descent hypotheses appropriate to the objects;
a Galois tensor identity for action comparison &
Amitsur descent and, when the tensor identity holds, coherent
semilinear actions. Chapters~\ref{ch:algebra}--\ref{ch:spectral}. \\
A stratified base &
The constructibility and geometric hypotheses of the corresponding
exodromy theorem &
Stratumwise systems with transport along exit paths or an
\'etale profinite category. Chapter~\ref{ch:exodromy}. \\
A topos and a geometric atlas &
Effectivity of the chosen presentation in the specified
category of topoi or slices &
A category of sheaves or descent objects associated to that
presentation. Chapter~\ref{ch:presentations}. \\
\bottomrule
\end{longtable}
\endgroup

These implications have several useful reversals and limitations.
A trivializing cover gives a subgroup correspondence if the
classification of intermediate objects imposes connectedness,
pointing, and the appropriate truncation. Dropping these
conditions leads to actions on families, with a different
notion of map. The totalization formula for a relative
reflection requires compatible coefficient categories.
Promoting a general image simplicial diagram to a groupoid
requires a check on the Segal pullbacks. A theorem on
profinite completion concerns the tests retained by that
completion. Finally, specialization maps for a stratified
space can fail to be invertible, so passing to the groupoid
completion has the effect of requiring additional inverses
from its representations.

The common result is a homotopy-coherent descent statement
whose strength is measured by its mapping spaces. The
classical correspondence can then be obtained by restriction,
truncation, or reversal of variance in those cases where the
intermediate mathematical objects have the requisite form.
The higher theorems keep the coherent data visible and make it
possible to say exactly which features of the original
symmetry object can be recovered.

\backmatter
\begingroup
\small
\renewcommand{\chaptermark}[1]{\markboth{#1}{}}
\phantomsection
\addcontentsline{toc}{chapter}{Bibliography}
\begin{thebibliography}{99}
\setlength{\itemsep}{2pt}
\bibitem{bj} Francis Borceux and George Janelidze.
\emph{Galois Theories}. Cambridge Studies in Advanced Mathematics 72.
Cambridge University Press, 2001.
\href{https://doi.org/10.1017/CBO9780511619939}{Publisher edition}.
\bibitem{htt} Jacob Lurie. \emph{Higher Topos Theory}.
Annals of Mathematics Studies 170. Princeton University Press, 2009.
\href{https://arxiv.org/abs/math/0608040}{arXiv:math/0608040}.
\bibitem{hoyois} Marc Hoyois. Higher Galois theory.
\emph{Journal of Pure and Applied Algebra} 222 (2018), 1859--1877.
\href{https://arxiv.org/abs/1506.07155}{arXiv:1506.07155}.
\bibitem{nss} Thomas Nikolaus, Urs Schreiber, and Danny Stevenson.
Principal infinity-bundles---General theory.
\emph{Journal of Homotopy and Related Structures} 10 (2015), 749--801.
\href{https://arxiv.org/abs/1207.0248}{arXiv:1207.0248}.
\bibitem{dag} Jacob Lurie. \emph{Derived Algebraic Geometry VII: Spectral
Schemes}, 2011. \href{https://www.math.ias.edu/~lurie/papers/DAG-VII.pdf}{Author's manuscript}.
\bibitem{ha} Jacob Lurie. \emph{Higher Algebra}, 2017.
\href{https://people.math.harvard.edu/~lurie/papers/HA.pdf}{Author's manuscript}.
\bibitem{rognes} John Rognes. Galois extensions of structured ring
spectra. \emph{Memoirs of the American Mathematical Society} 192 (2008),
no.~898. \href{https://arxiv.org/abs/math/0502183}{arXiv:math/0502183}.
\bibitem{mathew} Akhil Mathew. The Galois group of a stable homotopy theory.
\emph{Advances in Mathematics} 291 (2016), 403--541.
\href{https://arxiv.org/abs/1404.2156}{arXiv:1404.2156}.
\bibitem{stacks} The Stacks Project Authors. \emph{The Stacks Project}.
\href{https://stacks.math.columbia.edu/tag/0244}{Tag 0244}
and \href{https://stacks.math.columbia.edu/tag/023R}{Tag 023R}.
\bibitem{exodromy} Clark Barwick, Saul Glasman, and Peter Haine.
\emph{Exodromy}. \href{https://arxiv.org/abs/1807.03281}{arXiv:1807.03281}.
\bibitem{martini} Louis Martini. Yoneda's lemma for internal higher categories.
\href{https://arxiv.org/abs/2103.17141}{arXiv:2103.17141}.
\bibitem{portateyssier} Mauro Porta and Jean-Baptiste Teyssier.
Topological exodromy with coefficients.
\href{https://arxiv.org/abs/2211.05004}{arXiv:2211.05004}.
\bibitem{beyondconical} Peter J. Haine, Mauro Porta, and Jean-Baptiste
Teyssier. Exodromy beyond conicality.
\href{https://arxiv.org/abs/2401.12825}{arXiv:2401.12825}.
\bibitem{joyal-tierney} Andr\'e Joyal and Myles Tierney.
An extension of the Galois theory of Grothendieck.
\emph{Memoirs of the American Mathematical Society} 51 (1984), no.~309.
\href{https://doi.org/10.1090/memo/0309}{DOI 10.1090/memo/0309}.
\bibitem{rezk} Charles Rezk. \emph{Toposes and homotopy toposes}, v0.15, 2010.
\href{https://rezk.web.illinois.edu/homotopy-topos-sketch.pdf}{Author's manuscript}.
\bibitem{dhi} Daniel Dugger, Sharon Hollander, and Daniel C. Isaksen.
Hypercovers and simplicial presheaves.
\emph{Mathematical Proceedings of the Cambridge Philosophical Society}
136 (2004), 9--51. \href{https://arxiv.org/abs/math/0205027}{arXiv:math/0205027}.
\bibitem{modalities} Egbert Rijke, Michael Shulman, and Bas Spitters.
Modalities in homotopy type theory. \emph{Logical Methods in Computer
Science} 16 (2020), 2:2. \href{https://arxiv.org/abs/1706.07526}{arXiv:1706.07526}.
\bibitem{kerodon} Jacob Lurie. \emph{Kerodon}, \S3.5.
\href{https://kerodon.net/tag/0513}{Truncations and Postnikov towers}.
\bibitem{jkcentral} George Janelidze and G.~Max Kelly.
Galois theory and a general notion of central extension.
\emph{Journal of Pure and Applied Algebra} 97 (1994), 135--161.
\href{https://doi.org/10.1016/0022-4049(94)90057-4}{DOI 10.1016/0022-4049(94)90057-4}.
\end{thebibliography}
\endgroup

\clearpage
\phantomsection
\addcontentsline{toc}{chapter}{Subject index}
\printindex
\end{document}

End of manuscript · Back to top ↑