Bourbaki

§ 4. Theory of descent

1. Descent data

Let X and Y be topological spaces and let f:X→Yf: X\rightarrow Y be a continuous mapping. Let (Z,p)(Z, p) be an X-space.

Definition 1 0227

A descent data relative to ff on the X-space (Z,p)(Z, p) is called a continuous mapping τ:Z×YX→Z\tau : Z\times_YX\rightarrow Z satisfying the following two properties:

(i) For every pair (x,x′)∈X×YX(x, x')\in X\times_YX, the mapping z↦τ(z,x′)z\mapsto \tau (z, x') induces by restriction a bijection τx,x′\tau_{x,x'} of ZxZ_x onto Zx′Z_{x'};

(ii) For every triplet (x,x′,x′′)(x, x', x'') of points of X such that f(x)=f(x) = f(x′)=f(x′′)f(x') =f(x''), one has

τx,x′′=τx′,x′′∘τx,x′ \tau_{x,x''}=\tau_{x',x''}\circ \tau_{x,x'}

If τ\tau is a descent data on (Z,p)(Z, p), the family (τx,x′)(\tau_{x,x'}), for (x,x′)∈X×YX(x, x')\in X\times_YX, is thus a law of operation (on the right) of the groupoid X×YXX\times_YX on the X-space (Z,p)(Z, p) (II, p. 167). In particular, one has τx,x=\tau_{x,x}= IdZx_{Z_x} for every x∈Xx\in X, which is also written τ(z,p(z))=z\tau (z, p(z)) =z for every z∈Zz\in Z. Conversely, given a law of operation (on the right) of the groupoid X×YXX\times_YX on (Z,p)(Z, p), the map from Z×YXZ\times_YX into Z defined by (z,x)↦z⋅(p(z),x)(z, x)\mapsto z\cdot (p(z), x) verifies the relations (i) and (ii) of the definition.

Let X and Y be topological spaces, f:X→Yf: X\rightarrow Y a continuous mapping and let (Z,p)(Z, p) be an X-space. If τ\tau is a descent data relative to ff on (Z,p)(Z, p), the relation Rτ(z1,z2)R_{\tau}(z_1, z_2) defined by « f(p(z1))=f(p(z_1)) = f(p(z2))f(p(z_2)) and τ(z1,p(z2))=z2\tau (z_1, p(z_2)) =z_2 » is the equivalence relation in Z deduced from the operation of the groupoid X×YXX\times_YX defined by τ\tau; it is compatible with the mapping f∘pf\circ p. It is said that this is the equivalence relation associated with the descent data τ\tau. The canonical bijection (z1,z2)↦(z1,p(z2))(z_1, z_2)\mapsto (z_1, p(z_2)) of the graph Γ of the equivalence relation RτR_{\tau} onto Z×YXZ\times_YX is a homeomorphism, the reciprocal homeomorphism applies an element (z1,x2)∈Z×YX(z_1, x_2)\in Z\times_YX to (z1,τ(z1,x2))(z_1, \tau (z_1, x_2)).

Conversely, let R be an equivalence relation in Z compatible with the mapping f∘p:Z→Yf\circ p: Z\rightarrow Y and let Γ be the graph of R. Suppose moreover that the mapping p2:(z1,z2)↦(z1,p(z2))p_2: (z_1, z_2)\mapsto (z_1, p(z_2)) defines a homeo-morphism of Γ onto Z×YXZ\times_YX. The mapping τ:Z×YX→Z\tau : Z\times_YX\rightarrow Z given by pr2∘p2−1_2\circ p^{-1}_2 is continuous; it is a descent data relative to ff on (Z,p)(Z, p) and the relation R is the equivalence relation associated with τ\tau.

Example 1 0228

Let X and Y be topological spaces, let f:X→Yf: X\rightarrow Y be a continuous mapping and let (T,q)(T, q) be a Y-space. Put Z=X×YTZ = X\times_YT. The mapping τ\tau from Z×YXZ\times_YX into Z which associates with ((x,t),x′)((x, t), x') the element (x′,t)(x', t) is a descent datum relative to ff on the X-space (X×YT(X\times_YT, pr1)_1), called the canonical descent datum. For z1=(x1,t1)z_1= (x_1, t_1) and z2=(x2,t2)∈Zz_2= (x_2, t_2)\in Z, the relation Rτ{z1,z2}R_{\tau}\{z_1, z_2\} is equivalent to t1=t2t_1=t_2.

Example 2 0229

Let Y be a topological space, (Vi)i∈I(V_i)_{i\in I} a family of subsets of Y, and for each i∈Ii\in I, let (Zi,pi)(Z_i, p_i) be a ViV_i-space. Let X denote the sum topological space of the family (Vi)i∈I(V_i)_{i\in I} and (Z,p)(Z, p) the X-space sum of the family (Zi)i∈I(Z_i)_{i\in I}. Let f:X→Yf: X\rightarrow Y be the canonical mapping.

The space X×YXX\times_YX is then identified with the sum space of the family (Vi∩(V_i\cap Vj)(i,j)∈I×I(V_j)_{(i,j)\in I\times I}(I, p. 4, Example 5). Let τ\tau be a descent datum relative to ff on (Z,p)(Z, p). For each pair (i,j)∈I×I(i, j)\in I\times I, define a continuous mapping τi,j:pi−1(Vi∩Vj)→pj−1(Vi∩Vj)\tau_{i,j}:\overset{-1}{p_{i}}(V_i\cap V_j)\rightarrow \overset{-1}{p_{j}}(V_i\cap V_j) by z↦τ(z,(pi(z),j))z\mapsto \tau (z,(p_i(z), j)). The family (τi,j)(\tau_{i,j}) satisfies the following properties:

(i) For each i∈Ii\in I, one has τi,i=\tau_{i,i}= IdZi_{Z_i};

(ii) For each pair (i,j)∈I×I,τi,j(i, j)\in I\times I,\tau_{i,j} is an isomorphism of (Vi∩Vj)(V_i\cap V_j)-spaces ;

(iii) For each triple (i,j,k)∈I×I×I(i, j, k)\in I\times I\times I and each z∈pi−1(Vi∩Vj∩Vk)z\in \overset{-1}{p_{i}}(V_i\cap V_j\cap V_k), one has τj,k(τi,j(z))=τi,k(z)\tau_{j,k}(\tau_{i,j}(z)) =\tau_{i,k}(z).

Conversely, every family (τi,j)(\tau_{i,j}) possessing the above properties arises from a unique descent datum relative to ff on (Z,p)(Z, p).

2. Effective descent data

Let X and Y be topological spaces, f:X→Yf: X\rightarrow Y a continuous mapping and (Z,p)(Z, p) an X-space. Let τ\tau be a descent datum relative to ff on (Z,p)(Z, p), let RτR_{\tau} be the associated equivalence relation and let g:Z→Z/Rτg: Z\rightarrow Z/R_{\tau} be the canonical mapping. Since the relation RτR_{\tau} is compatible with the mapping f∘pf\circ p, there exists a unique continuous mapping q:Z/Rτ→Yq: Z/R_{\tau}\rightarrow Y such that the diagram

Z gZ/Rτ^gZ/R_{\tau}

pq(1) pq \tag{1}

X f^f Y

let it be a commutative square. The Y-space (Z/Rτ,q)(Z/R_{\tau}, q) is called the quotient space of (Z,p)(Z, p) by the descent datum τ\tau. Let us denote by h:Z→X×Yh: Z\rightarrow X\times_Y (Z/Rτ)(Z/R_{\tau}) the mapping defined by h(z)=(p(z),g(z))h(z) = (p(z), g(z)) for z∈Zz\in Z. It is continuous. Let (x,u)∈X×Y(Z/Rτ)(x, u)\in X\times_Y(Z/R_{\tau}) and let z∈Zz\in Z be such that g(z)=ug(z) =u; then we have (z,x)∈Z×YX(z, x)\in Z\times_YX and the point z′=τ(z,x)z'=\tau (z, x) is the unique element of Z such that h(z′)=(x,u)h(z') = (x, u); consequently, the mapping hh is bijective.

One says that the descent datum τ\tau relative to ff on (Z,p)(Z, p) is effective if diagram (1) is a cartesian square, that is to say, if the continuous bijection hh is a homeomorphism. In order that the descent datum τ\tau be effective, it is necessary and sufficient that the sets p−1(U)∩V\overset{-1}{p}(U)\cap V, where U is an open subset of X and V an open subset of Z saturated for RτR_{\tau}, constitute a basis of the topology of Z. In particular, the condition for a descent datum relative to ff to be effective is of a local nature on Y.

Example 1 022A

Let X and Y be topological spaces and f:X→Yf: X\rightarrow Y a continuous mapping. Let (T,q)(T, q) be a Y-space. Let Z be the X-space X×YTX\times_YT, endowed with the mapping pr1_1; let us denote by τ\tau its canonical descent datum relative to ff (IV, p. 383, Example 1). The subsets of Z of the form X×YVX\times_YV, where V is an open subset of T, are open in Z and saturated for the relation Rτ(R_{\tau}(loc. cit.). By definition of the product topology, the sets U×YU\times_Y pr2−1(V)^{-1}_2(V), where U is open in X and V is open in T, form a basis of the topology of X×YZX\times_YZ. Consequently, the canonical descent datum on a fibred product X×YTX\times_YT is effective.

The canonical mapping Z/Rτ→TZ/R_{\tau}\rightarrow T is injective and continuous. It is not, however, necessarily surjective, nor strict (IV, p. 462, exerc. 1).

Example 2 022B

Let us take up again the notation of Example 2 (IV, p. 383). The topological space Z/RτZ/R_{\tau} is then the topological space obtained by gluing the spaces ZiZ_i along the pi−1(Vi∩Vj)\overset{-1}{p_{i}}(V_i\cap V_j) by means of the bijections τi,j\tau_{i,j} (TG, I, p. 16). Consequently, if for every i∈Ii\in I, the set ViV_i is open (resp. closed) in Y, the set g(Zi)g(Z_i) is open (resp. closed) in Z/RτZ/R_{\tau} and the restriction of gg to ZiZ_i induces a homeomorphism of ZiZ_i onto g(Zi)g(Z_i) (TG, I, p. 17, Prop. 9). The space Z is the sum space of the spaces ZiZ_i; the space X×Y(Z/Rτ)X\times_Y(Z/R_{\tau}) is the sum space of the spaces Vi×YV_i\times_Y (Z/Rτ)=g(Zi)(Z/R_{\tau}) =g(Z_i). The mapping hh is identified with the sum mapping of the mappings g∣Zi:Zi→g(Zi)g|Z_i: Z_i\rightarrow g(Z_i). It is therefore a homeomorphism, which shows that the descent datum τ\tau is effective.

Without a particular assumption on the subsets ViV_i, it is not always true that the restriction of gg to ZiZ_i induces a homeomorphism of ZiZ_i onto its image; in this case, the descent datum τ\tau is not effective (IV, p. 462, exerc. 2).

Proposition 1 022C

Let X and Y be topological spaces, let f:X→Yf: X\rightarrow Y be a continuous mapping. Suppose that every point of Y possesses a neighbourhood above which there exists a continuous section of the mapping ff. Then every descent datum relative to ff on an X-space is effective.

Let (Z,p)(Z, p) be an X-space and let τ\tau be a descent datum relative to ff on (Z,p)(Z, p). The assertion that τ\tau is an effective descent datum is local in Y, which allows us to suppose that the mapping ff possesses a continuous section ss. Let g:Z→Z/Rτg: Z\rightarrow Z/R_{\tau} and q:Z/Rτ→Yq: Z/R_{\tau}\rightarrow Y denote the canonical mappings, and let h:Z→X×Y(Z/Rτ)h: Z\rightarrow X\times_Y(Z/R_{\tau}) be the mapping given by z↦(p(z),g(z))z\mapsto (p(z), g(z)). The mapping hh is bijective and continuous; it is enough to prove that it is a homeomorphism.

The mapping from Z into Z which, to zz, associates τ(z,s(f(p(z))))\tau (z, s(f(p(z)))) is continuous and maps every element of Z to the unique element z′z' of Z which is equivalent to it for the relation RτR_{\tau} and such that p(z′)p(z') belongs to the image of ss. It therefore defines, by passing to the quotient, a continuous mapping t:Z/Rτ→Zt: Z/R_{\tau}\rightarrow Z which is a section of the mapping gg. In particular, one has f∘p∘t=q∘g∘t=qf\circ p\circ t=q\circ g\circ t=q.

For every (x,u)∈X×Y(Z/Rτ)(x, u)\in X\times_Y(Z/R_{\tau}), one has f(p(t(u)))=f(x)f(p(t(u))) =f(x); let us then put h′(x,u)=τ(t(u),x)h'(x, u) =\tau (t(u), x). The resulting mapping h′h' from X×Y(Z/Rτ)X\times_Y(Z/R_{\tau}) into Z is continuous. For every (x,u)∈X×Y(Z/Rτ)(x, u)\in X\times_Y(Z/R_{\tau}), one has

h(h′(x,u))=(p(h′(x,u)),g(h′(x,u))) h(h'(x, u)) = (p(h'(x, u)), g(h'(x, u)))
=(p(τ(t(u),x)),g(τ(t(u),x))) = (p(\tau (t(u), x)), g(\tau (t(u), x)))
=(x,u) = (x, u)

because τ(t(u),x)\tau (t(u), x) is equivalent to t(u)t(u) for the relation RτR_{\tau}. This proves that the mapping h∘h′h\circ h' is the identity mapping of X×Y(Z/Rτ)X\times_Y(Z/R_{\tau}). For z∈Zz\in Z, one then has t(g(z))=τ(z,s(f(p(z)))t(g(z)) =\tau (z, s(f(p(z))) and f(p(t(g(z))))=f(p(z))f(p(t(g(z)))) =f(p(z)), whence z=τ(t(g(z)),p(z))z=\tau (t(g(z)), p(z)), by definition of the equivalence relation RτR_{\tau}. Thus one has h′(h(z))=zh'(h(z)) =z and h′∘h=h'\circ h= IdZ_Z. The mapping hh is therefore a homeomorphism, which was to be proved.

Proposition 2 022D

Let f:X→f: X\rightarrow Y be a continuous mapping, let (Z,p)(Z, p) be an X-space and let τ\tau be a descent datum relative to ff on Z. The equivalence relation RτR_{\tau} is closed if ff is proper; it is open if ff is open.

The mapping τ~:Z×YX→\widetilde{\tau}: Z\times_YX\rightarrow X ×Y\times_Y Z defined by (z,x)↦(z, x)\mapsto (p(z),τ(z,x))(p(z), \tau (z, x)) is a homeomorphism, with inverse mapping (x,z)↦(τ(z,x),p(z))(x, z)\mapsto (\tau (z, x), p(z)). We have τ\tau = pr2∘τ~_2\circ \widetilde{\tau}, where pr2:X×YZ→_2: X\times_YZ\rightarrow Z is the second projection. If ff is proper, pr2_2 is proper; if ff is open, pr2_2 is open (I, p. 17, prop. 8). It follows that τ\tau is proper (resp. open) if ff is. The saturation of a subset A of Z with respect to the relation RτR_{\tau} is the image of A×YXA\times_YX under τ\tau. Consequently, if ff is proper, the saturation of a closed subset is closed; if ff is open, the saturation of an open subset is open.

3. Descent of morphisms

Let X and Y be topological spaces and let ff be a continuous mapping of X into Y. Let (Z,p)(Z, p) and (Z′,p′)(Z', p') be X-spaces endowed with descent data relative to ff, denoted respectively by τ\tau and τ′\tau '. One says that an X-morphism φ:Z→Z′\varphi : Z\rightarrow Z' is compatible with the descent data τ\tau and τ′\tau ' if one has

τ′(φ(z),x)=φ(τ(z,x)) \tau '(\varphi (z), x) =\varphi (\tau (z, x))

for every (z,x)∈Z×YX(z, x)\in Z\times_YX. It amounts to the same to say that the images under φ\varphi of two equivalent points with respect to the relation RτR_{\tau} are equivalent with respect to the relation Rτ′R_{\tau'}. Such a morphism φ\varphi defines, by passing to the quotients, a continuous mapping φ‾:Z/Rτ→Z′/Rτ′\overline{\varphi}: Z/R_{\tau}\rightarrow Z'/R_{\tau'}; it is a Y-morphism of spaces.

Let us denote by Cτ,τ′(Z;Z′)\mathscr{C}_{\tau ,\tau'}(Z; Z') the set of X-morphisms of Z into Z′Z' which are compatible with the descent data τ\tau and τ′\tau '.

Proposition 3 022E

Let X and Y be topological spaces and let f:X→f: X\rightarrow Y be a continuous mapping. Let (Z,p)(Z, p) and (Z′,p′)(Z', p') be X-spaces endowed with descent data relative to ff, denoted respectively by τ\tau and τ′\tau '. If the descent datum τ′\tau ' is effective, the mapping φ↦φ‾\varphi \mapsto \overline{\varphi} is a bijection of Cτ,τ′(Z;Z′)\mathscr{C}_{\tau ,\tau'}(Z; Z') onto CY(Z/Rτ;Z′/Rτ′)\mathscr{C}_Y(Z/R_{\tau}; Z'/R_{\tau'}).

For every X-morphism φ\varphi of Z into Z′Z' compatible with the descent data, the mapping φ‾\overline{\varphi} is a Y-morphism. Conversely, let g:Z→Z/Rτg: Z\rightarrow Z/R_{\tau} and g′:Z′→Z′/Rτ′g': Z'\rightarrow Z'/R_{\tau'} denote the canonical mappings, and let ψ:Z/Rτ→Z′/Rτ′\psi : Z/R_{\tau}\rightarrow Z'/R_{\tau'} be a Y-morphism. The mappings p:Z→Xp: Z\rightarrow X and ψ∘g:Z→Z′/Rτ′\psi \circ g: Z\rightarrow Z'/R_{\tau'} are Y-morphisms. The hypothesis that the descent datum τ′\tau ' is effective means that the diagram

Z′g′Z′/Rτ′ {Z'}^{g'}Z'/R_{\tau'}
p′q′p'q'

X f^f Y

is a cartesian square. Hence there exists a unique continuous mapping φ:Z→Z′\varphi : Z\rightarrow Z' such that p′∘φ=pp'\circ \varphi =p and g′∘φ=ψ∘gg'\circ \varphi =\psi \circ g. The first equality means that φ\varphi is an X-morphism, the second equality means that φ\varphi is compatible with the descent data and that φ‾=ψ\overline{\varphi}=\psi, whence the proposition.

4. Descent: case of spread spaces

Let X and Y be topological spaces and let f:X→Yf: X\rightarrow Y be a continuous mapping. Let T and T′T' be Y-spaces; endow the X-spaces X×YTX\times_YT and X×YT′X\times_YT' with their canonical descent data and denote by Cf(X×YT;X×YT′)\mathscr{C}_f(X\times_YT; X\times_YT') the set of X-morphisms of X×YTX\times_YT into X×YT′X\times_YT' which are compatible with these descent data. For every Y-morphism φ:T→T′\varphi : T\rightarrow T', the X-morphism f∗(φ):(x,t)↦f^*(\varphi ): (x, t)\mapsto (x,φ(t))(x, \varphi (t)) of X×YTX\times_YT into X×YT′X\times_YT' is compatible with the canonical descent data. We shall denote by f∗:CY(T;T′)→Cf(X×YT;X×YT′)f^*:\mathscr{C}_Y(T; T')\rightarrow \mathscr{C}_f(X\times_YT; X\times_YT') the resulting mapping.

Proposition 4 022F

Suppose that the mapping ff is strict and surjective and that T is a spread Y-space. Then the mapping f∗:CY(T;T′)→Cf(X×YT;X×YT′)f^*:\mathscr{C}_Y(T; T')\rightarrow \mathscr{C}_f(X\times_YT; X\times_YT') is bijective.

Let τ\tau (resp. τ′\tau ') denote the equivalence relation on X×YTX\times_YT (resp. on X×YT′X\times_YT') associated with the canonical descent datum. Since the mapping ff is surjective, the projection pr2:X×YT→T_2: X\times_YT\rightarrow T is surjective and the canonical mapping (X×YT)/Rτ→T(X\times_YT)/R_{\tau}\rightarrow T is bijective. In particular, the mapping f∗f^* is injective. Let us prove that it is surjective. Let φ:X×YT→X×YT′\varphi : X\times_YT\rightarrow X\times_YT' be an X-morphism compatible with the canonical descent data. Then for (x,t)∈X×YT(x, t)\in X\times_YT, one has φ(x,t)=(x,φ(t))\varphi (x, t) = (x, \varphi (t)), where φ‾\overline{\varphi} is a mapping of T into T′T'.

By definition of φ‾\overline{\varphi}, the mapping φ‾∘\overline{\varphi}\circ pr2:X×YT→T′_2: X\times_YT\rightarrow T' is equal to pr2∘φ_2\circ \varphi, and is therefore continuous. Since the mapping ff is surjective and strict and T is a spread Y-space, the projection pr2:X×YT→T_2: X\times_YT\rightarrow T is strict (I, p. 32, remark 3). By proposition 9 of I, p. 18, the mapping φ‾\overline{\varphi} is therefore continuous. It is a Y-morphism such that f∗(φ‾)=φf^*(\overline{\varphi}) =\varphi, which proves the proposition.

Corollary 022G

Let X and Y be topological spaces and let f:X→Yf: X\rightarrow Y be a strict and surjective mapping. Let T and T′T' be spread Y-spaces. If there exists an X-isomorphism X×YT→X×YT′X\times_YT\rightarrow X\times_YT' compatible with the canonical descent data, the Y-spaces T and T′T' are isomorphic.

Let ψ:X×YT→X×YT′\psi : X\times_YT\rightarrow X\times_YT' be an X-isomorphism of spread spaces. By proposition 4, there exists a unique morphism of Y-spaces φ:T→T′\varphi : T\rightarrow T' such that ψ=f∗(φ)\psi =f^*(\varphi ). Since ff is surjective, the mapping φ\varphi is bijective. It then follows from cor. 2 of I, p. 30 that φ\varphi is an isomorphism, since the Y-spaces T and T′T' are spread.

Proposition 5 022H

Let X and Y be topological spaces and let f:X→Yf: X\rightarrow Y be a continuous mapping. Suppose that the mapping ff is proper and separated, or else open. Every descent datum relative to ff on a spread X-space is effective. Moreover, if ff is surjective, the quotient space is a spread Y-space.

Let (Z,p)(Z, p) be an X-étalé space and let τ\tau be a descent datum relative to ff on (Z,p)(Z, p). Let us denote by RτR_{\tau} the equivalence relation in Z associated with τ\tau, let g:Z→Z/Rτg: Z\rightarrow Z/R_{\tau} be the canonical surjection and let hh denote the mapping of Z into X×Y(Z/Rτ)X\times_Y(Z/R_{\tau}) defined by z↦(p(z),g(z))z\mapsto (p(z), g(z)). It is continuous and bijective; let us prove that it is a homeomorphism.

a) Suppose first that the mapping ff is open.

Let z0z_0 be a point of Z; put x0=p(z0)x_0=p(z_0). Since pp is étale, there exists a neighbourhood U of x0x_0 in X and a continuous section s:U→Zs: U\rightarrow Z of pp over U such that s(x0)=z0s(x_0) =z_0. The mapping p×p:Z×YZ→p\times p: Z\times_YZ\rightarrow X×YXX\times_YX is étale, and the mappings of U×YUU\times_YU into Z×YZZ\times_YZ defined by (x,x′)↦(s(x),s(x′))(x, x')\mapsto (s(x), s(x')) and (x,x′)↦(s(x),τ(s(x),x′))(x, x')\mapsto (s(x), \tau (s(x), x')) are continuous sections of it over U×YYU\times_YY. Since they coincide at every point of U×YUU\times_YU of the form (x,x)(x, x), they therefore coincide on an open neighbourhood V of ΔU\Delta_U contained in U×YUU\times_YU. Let U0U_0 be a neighbourhood of x0x_0 in X such that U0×YU0U_0\times_YU_0 is contained in V. Let (x,u)(x, u) be a point of U0×Yg(s(U0))U_0\times_Yg(s(U_0)); let x′∈U0x'\in U_0 be such that u=g(s(x′))u=g(s(x')). We have therefore s(x)=τ(s(x′),x)s(x) =\tau (s(x'), x), whence Rτ{s(x),s(x′)}R_{\tau}\{s(x), s(x')\} and (x,u)=(x,g(s(x′)))=(x, u) = (x, g(s(x'))) = (x,g(s(x)))=h(s(x))(x, g(s(x))) =h(s(x)).

Since the mapping gg is open (IV, p. 386, prop. 2), the set U0×Yg(s(U0))U_0\times_Yg(s(U_0)) is an open subset of X×Y(Z/Rτ)X\times_Y(Z/R_{\tau}) on which the mapping h−1h^{-1} is equal to the continuous mapping s∘s\circ pr1_1. The mapping hh is therefore a homeomorphism.

b) Suppose now that the mapping ff is proper and separated.

Let z0z_0 be a point of Z; put x0=p(z0)x_0=p(z_0) and y0=f(x0)y_0=f(x_0). The mapping ss given by x↦τ(z0,x)x\mapsto \tau (z_0, x) is a section of pp over f−1(y0)\overset{-1}{f}(y_0).

The set f−1(y0)\overset{-1}{f}(y_0) is compact (TG, I, p. 75, Theorem 1 and I, p. 26, Remark 2), and two distinct points of f−1(y0)\overset{-1}{f}(y_0) have disjoint neighbourhoods in X, since ff is separated (I, p. 25, Proposition 1). By Theorem 2 of I, p. 37, there therefore exists a neighbourhood U0U_0 of f−1(y0)\overset{-1}{f}(y_0) in X and a section s0s_0 of pp over U0U_0 extending ss. The set s0(U0)s_0(U_0) is an open set of Z, since pp is étale (I, p. 30, Corollary 3), and contains the saturated set of {z0}\{z_0\} for the relation RτR_{\tau}. The mapping gg is closed (IV, p. 386, Proposition 2). There then exists an open set V of Z/RτZ/R_{\tau} such that W=g−1(V)∩p−1(U0)W =\overset{-1}{g}(V)\cap \overset{-1}{p}(U_0) is a neighbourhood of z0z_0 contained in s0(U0)s_0(U_0) (I, p. 75, Lemma).

Let now (x,u)∈U0×YV(x, u)\in U_0\times_YV, and let zz be the unique point of Z such that h(z)=(x,u)h(z) = (x, u); by definition, one has z∈Wz\in W. Since W⊂s0(U0)W\subset s_0(U_0), one has z=s0(p(x))z=s_0(p(x)). This shows that the restriction of h−1h^{-1} to the open set U0×YVU_0\times_YV of X×Y(Z/Rτ)X\times_Y(Z/R_{\tau}) is equal to s0∘p∘s_0\circ p\circ pr1_1. Consequently, hh is a homeomorphism.

Thus we have shown that the descent datum τ\tau is effective. The mapping ff is universally strict (I, p. 20, corollary). Under the assumption that ff is surjective, it then follows from Proposition 8 of I, p. 31 that q:Z/Rτ→Yq: Z/R_{\tau}\rightarrow Y is étale.

5. Descent: case of coverings

Let X, Y be topological spaces, and let f:X→Yf: X\rightarrow Y be a surjective continuous mapping. Let (Z,p)(Z, p) be a covering of X, and let τ\tau be a descent datum relative to ff on Z.

If ff is proper and separated (resp. if ff is open), the descent datum τ\tau is effective and the Y-space Z/RτZ/R_{\tau} is an étale space (IV, p. 389, Proposition 5). It is even a covering of Y if ff has a continuous section in a neighbourhood of each point (I, p. 72, Proposition 3) or if Z is a locally finite covering of X (I, p. 77, Corollary 4). This number is devoted to bringing out other conditions under which Z/RτZ/R_{\tau} is a covering of Y.

Let us first prove a lemma.

Lemma 1 022I

Let B, B′B' be topological spaces, and let f:B′→Bf: B'\rightarrow B be a continuous mapping. If the fibred square B′×BB′B'\times_BB' is locally connected, the topological space B′B' is locally connected.

Let aa be a point of B′B' and let V be a neighbourhood of aa. Suppose that the fibre square B′×BB′B'\times_BB' is locally connected and let W be a connected neighbourhood of (a,a)(a, a) in B′×BB′B'\times_BB' which is contained in V×VV\times V. Put U = pr1(W)_1(W). The set U is contained in V, and is connected since the image of a connected set under a continuous mapping is connected. If ΔB′\Delta_{B'} denotes the diagonal of B′×BB′B'\times_BB', the mapping pr1∣ΔB′:ΔB′→B′_1|\Delta_{B'}: \Delta_{B'}\rightarrow B' is a homeomorphism. Since U contains pr1(W∩ΔB′)_1(W\cap \Delta_{B'}) and W∩ΔB′W\cap \Delta_{B'} is a neighbourhood of (a,a)(a, a) in ΔB′\Delta_{B'}, U is a neighbourhood of aa in B′B'. This proves the lemma.

Proposition 6 022J

Let

E′f′{E'}^{f'} E

p′pp'p

B′f{B'}^f B

be a cartesian square. Suppose that the mapping ff is proper, separated and surjective, and make one of the following assumptions:

(i) The fibres of ff are locally connected and the fibre square B′×BB′B'\times_BB' is locally connected.

(ii) The fibres of ff are finite, the diagonal ΔB′\Delta_{B'} of B′×BB′B'\times_BB' is open in B′×BB′B'\times_BB' and B′×BB′−ΔB′B'\times_BB'-\Delta_{B'} is a locally connected space.

Then, if (E′,p′)(E', p') is a covering, (E,p)(E, p) is a covering.

Since the mapping ff is universally strict (I, p. 20, corollary), the mapping pp is étale (I, p. 31, prop. 8) and separated (I, p. 27, prop. 4). We shall suppose, which is permissible, that E′=B′×BEE'= B'\times_BE.

Let aa be a point of B; it is a question of proving that the point aa has a neighbourhood W such that the W-space (EW,pW)(E_W, p_W) is a covering

which is trivializable. Put Ba′B'_a = f−1(a)\overset{-1}{f}(a) and denote by Ea′E'_a = (−p′1)(Ea)(^-{p'}^1)(E_a). The mapping ta:Ea′→Ba′×Eat_a: E'_a\rightarrow B'_a\times E_a defined by ta(y)=(p′(y),f′(y))t_a(y) = (p'(y), f'(y)) is a Ba′B'_a-isomorphism (I, p. 9, prop. 4), hence Ea′E'_a is a trivializable covering of Ba′B'_a and tat_a is a trivialization of this covering.

Let us prove that there exists a neighbourhood V′V' of Ba′B'_a in B′B' and a continuous trivialization tt of the covering (EV′′,pV′)(E'_{V'}, p_{V'}) extending tat_a. Under assumption (ii), Ba′B'_a is finite and its points have pairwise disjoint open neighbourhoods over which the covering E′E' is trivializable, whence the assertion in this case. Under assumption (i), Ba′B'_a is locally connected, and so is B′(B'(IV, p. 390, lemma 1); since the mapping ff is proper and separated, Ba′B'_a is compact and two distinct points have disjoint neighbourhoods in B′B', so that the pair (B′,Ba′)(B',B'_a) satisfies property (PCV) ( I, p. 37, lemma 1). The assertion therefore follows from Cor. 2 of I, p. 90.

As ff is proper, there exists a neighbourhood V of aa in B such that V′V' contains f−1(V)\overset{-1}{f}(V) (lemma, I, p. 75). We may thus suppose that

V′=f−1(V) V'=\overset{-1}{f}(V)

Let us endow the V′V'-spaces V′×EaV'\times E_a and EV′′=V′×BEE'_{V'}= V'\times_BE with their canonical descent data relative to fV:V′→Vf_V: V'\rightarrow V. We shall now prove that, after shrinking V and V′V', the isomorphism of V′V'-spaces t:EV′′→t: E'_{V'}\rightarrow V′×EaV'\times E_a which we have just defined is compatible with the descent data, that is to say that one has t(b1′,x)=t(b2′,x)t(b'_1, x) =t(b'_2, x) if (b1′,b2′)∈V′×VV′(b'_1, b'_2)\in V'\times_VV' and x∈Ef(b1′)x\in E_{f(b'_1)}. Let t~\widetilde{t} denote the mapping pr2∘t:EV′′→Ea_2\circ t: E'_{V'}\rightarrow E_a.

Put V′′=V′×VV′V''= V'\times_VV'; let us consider it as a V′V'-space by means of the first projection. The mapping ((b1′,b2′),x)↦((b1′,b2′),(b1′,x))((b'_1, b'_2), x)\mapsto ((b'_1, b'_2),(b'_1, x)) of V′′×VEV''\times_VE into V′′×V′E′V''\times_{V'}E' is an isomorphism of V′′V''-spaces; this shows that V′′×VEV''\times_VE is a covering of V′′V''.

For i=1i= 1, 2, define a mapping ui:V′′×VE→V′′×Eau_i: V''\times_VE\rightarrow V''\times E_a by setting ui(b1′,b2′,x)=(b1′,b2′,t~(bi′,x))u_i(b'_1, b'_2, x) = (b'_1, b'_2,\widetilde{t}(b'_i, x)); these are trivializations of the covering V′′×VEV''\times_VE. Let W′′W'' be the set of points w∈V′′w\in V'' over which these trivializations u1u_1 and u2u_2 coincide; it contains Ba′×Ba′B'_a\times B'_a, as well as the diagonal ΔV′\Delta_{V'}. Let us prove that W′′W'' is a neighbourhood of Ba′×Ba′B'_a\times B'_a. Under assumption (i), this follows from Cor. 2 of I, p. 80, since B′×B′B'\times B' is locally connected. Under assumption (ii), W′′W'' contains a neighbourhood of (Ba′×Ba′)−ΔBa′(B'_a\times B'_a)-\Delta_{B'_a} in (B′×BB′)−ΔB′(B'\times_BB')-\Delta_{B'}, for this set is locally connected (loc. cit.). As ΔV′\Delta_{V'} is open in B′×BB′,W′′B'\times_BB',W'' is a neighbourhood of Ba′×Ba′B'_a\times B'_a in V′′V''.

As ff is proper, the canonical mapping f′′f'' of B′×BB′B'\times_BB' into B is proper, for it is the composite of the projection pr1:B′×BB′→B′_1: B'\times_BB'\rightarrow B' and the mapping ff.

By the lemma of I, p. 75, there exists a neighbourhood W of aa in V

such that (f′′−1)(W)(f\overset{-1}{''})(W) is contained in W′′W''; let us put W′=(f−1′)(W)W'=(\overset{-1}{f}')(W); it is a subset of V′V' and the isomorphism of étale spaces t:EW′′→W′×Eat: E'_{W'}\rightarrow W'\times E_a is compatible with the canonical descent data relative to the mapping fW:W′→Wf_W: W'\rightarrow W. It follows from corollary (IV, p. 388) that the W-étale spaces EWE_W and W×EaW\times E_a are isomorphic. In particular, EWE_W is a trivializable covering, whence the proposition.

Corollary 1 022K

Let E and B be topological spaces, p:E→Bp: E\rightarrow B a continuous mapping and (Ai)i∈I(A_i)_{i\in I} a locally finite closed covering of B such that for every pair (i,j)∈I×I,i=j̸(i, j)\in I\times I,i=\not j, the intersection Ai∩AjA_i\cap A_j is a locally connected space. Then, in order that the B-space (E,p)(E, p) be a covering, it is necessary and sufficient that, for every i∈Ii\in I, the AiA_i-space (p−1(Ai),pAi)(\overset{-1}{p}(A_i), p_{A_i}) be a covering of AiA_i.

The condition is necessary (cf. I, p. 69). Conversely, let B′B' denote the topological sum of the family (Ai)i∈I(A_i)_{i\in I} and f:B′→Bf: B'\rightarrow B the canonical mapping. The mapping ff is closed (TG, I, p. 6, prop. 4), separated (I, p. 27, remark 5), with finite fibres, hence proper (TG, I, p. 75, th. 1); it is also surjective. The diagonal ΔB′\Delta_{B'}, being equal to ⋃i∈IAi×BAi\bigcup_{i\in I}A_i\times_BA_i, is open in B′×BB′B'\times_BB'. Finally, the space (B′×BB′)−ΔB′(B'\times_BB')-\Delta_{B'} is homeomorphic to the topological sum of the family (Ai∩Aj)(A_i\cap A_j), (i,j)∈I×I(i, j)\in I\times I, i=j̸i=\not j; it is therefore locally connected. Hypothesis (ii) of proposition 6 is satisfied. If for every i,p−1(Ai)i,\overset{-1}{p}(A_i) is a covering of Ai,E′A_i,E' is then a covering of B′B', hence E is a covering of B.

Corollary 2 022L

Let X and Y be topological spaces and let f:X→Yf: X\rightarrow Y be a proper, separated and surjective mapping. Assume in addition one of the following hypotheses:

(i) The fibres of ff, as well as the space X×YXX\times_YX, are locally connected;

(ii) The fibres of ff are finite, the diagonal ΔX\Delta_X of X×YXX\times_YX is open in X×YXX\times_YX and the space (X×YX)−ΔX(X\times_YX)-\Delta_X is locally connected. Then every descent datum relative to ff on a covering of X is effective, and the quotient space is a covering of Y.

Let Z be a covering of X and let τ\tau be a descent datum relative to ff on Z. By proposition 5 (IV, p. 389), the descent datum τ\tau is effective, in other words the square

Z Z/RτZ/R_{\tau}

X f^f Y

is a cartesian square. The hypotheses of proposition 6 (IV, p. 391) are then satisfied. Consequently, Z/RτZ/R_{\tau} is a covering of Y.

Proposition 7 022M

Let X and Y be topological spaces and let f:X→Yf: X\rightarrow Y be a continuous and surjective mapping. Suppose that the space Y is unloopable and that the mapping ff is open and has the path-lifting property. Then every descent datum relative to ff on a covering of X is effective, and the quotient space is a covering of Y.

Let (Z,p)(Z, p) be a covering of X and let τ\tau be a descent datum relative to ff on Z. Since the mapping ff is surjective and open, it follows from prop. 5 of IV, p. 389, that the descent datum τ\tau is effective. Let T=Y/RτT = Y/R_{\tau}; it is the quotient Y-space; its projection qq is étale (loc. cit.); it is also separated (I, p. 27, prop. 4).

By hypothesis, the mapping ff has the path-lifting property, and so does the mapping pp, since Z is a covering of X (III, p. 302, Corollary 2 of Prop. 3). Hence the same is true of the mapping p∘fp\circ f, and therefore of the mapping qq. Consequently (cf. IV, p. 341, Remark 2), T is a covering of Y. The proposition is thus proved.

Remark 022N

Let X and Y be topological spaces and let f:X→Yf: X\rightarrow Y be a mapping. In order that every descent datum relative to ff on a covering of X be effective, and that the quotient space be a covering of Y, it is necessary that ff be strict and that f(X)f(X) be an open and closed subset of X.

In fact, let us identify the X-space X with X×YYX\times_YY and endow it with its canonical descent datum relative to ff. The quotient space is identified with f(X)f(X), endowed with the quotient topology of the topology of X for the equivalence relation defined by ff. If this is a covering of Y, the space f(X)f(X) is then identified with an open and closed subset of Y and the mapping ff is strict.

6. Descent of groupoids

Let X and Y be topological spaces and let f:X→Yf: X\rightarrow Y be a continuous mapping. Let p1p_1 and p2p_2 denote the two projections of X×YXX\times_YX into X, Coeg(ff) the coequalizer groupoid of the pair (ϖ(p1),ϖ(p2))(\varpi (p_1), \varpi (p_2)) of groupoid morphisms from ϖ(X×YX)\varpi (X\times_YX) into ϖ(X)\varpi (X), and γ:ϖ(X)→\gamma :\varpi (X)\rightarrow Coeg(ff) the canonical groupoid morphism (II, p. 199, Def. 2). Since f∘p1=f∘p2f\circ p_1=f\circ p_2, we have ϖ(f)∘ϖ(p1)=ϖ(f)∘ϖ(p2)\varpi (f)\circ \varpi (p_1) =\varpi (f)\circ \varpi (p_2). It then follows from the universal property of coequalizers (II, p. 199, Prop. 3) that there exists a unique groupoid morphism ϖ′(f):\varpi '(f): Coeg(ff)→ϖ(Y)\rightarrow \varpi (Y) such that ϖ(f)=ϖ′(f)∘γ\varpi (f) =\varpi '(f)\circ \gamma.

The set of vertices of Coeg(ff) is the quotient set of the set X = Som(ϖ(X)\varpi (X)) by the equivalence relation defined by f(f(II, p. 200, Remark 1). It is identified with f(X)f(X).

Proposition 8 022O

Let X and Y be non-empty topological spaces and let ff be a continuous mapping of X into Y. Assume that the space X is locally arcwise connected, that the space Y is connected, and that the mapping ff is strict and surjective. Then the groupoid Coeg(ff) is transitive.

Let Γ be the quiver whose set of vertices is X and whose set of arrows is the sum set of Fl(ϖ(X)\varpi (X)) and X×YXX\times_YX, the origin and target mappings being those of ϖ(X)\varpi (X) on Fl(ϖ(X)\varpi (X)) and the mappings p1p_1 and p2p_2 on X×YXX\times_YX. By the definition of the framework of the pair (ϖ(\varpi(pr1_1),ϖ, \varpi(pr2_2)) (II, p. 185, Def. 3) and Remark 2 of II, p. 200, the set of orbits of Coeg(ff) is identified with the set of connected components of the quiver Γ. Since the space X is non-empty, it is enough to prove that the graph Γ is connected.

The connected components of Γ are saturated with respect to the equivalence relation “there exists a path joining xx to x′x'”, and therefore are open in X, since X is locally arcwise connected. They are then also closed. They are also saturated with respect to the equivalence relation R defined by ff. By hypothesis, the mapping ff induces a homeomorphism of X/RX/R onto Y, so that the image under ff of every connected component of Γ is an open and closed subset of Y, and therefore is equal to Y since the space Y is assumed connected.

Let C be a connected component of Γ, and let xx be a point of X. From the preceding, there exists a point x′∈Cx'\in C such that f(x′)=f(x)f(x') =f(x). By definition of the quiver Γ, we have x′∈Cx'\in C. Thus C = X and the quiver Γ is therefore connected.

Proposition 9 022P

Let X and Y be topological spaces and let ff be a continuous mapping of X into Y. Suppose that the space X is locally arcwise connected, that the space Y is simply connected, and that the mapping ff is strict and surjective. Then the groupoid ϖ(Y)\varpi (Y) is generated by the image of ϖ(X)\varpi (X) under ϖ(f)\varpi (f).

We may suppose that the spaces X and Y are non-empty. Let G denote the subgroupoid of ϖ(Y)\varpi (Y) generated by the image of ϖ(X)\varpi (X) under ϖ(f)\varpi (f). Since the mapping ff is surjective, the set of vertices of G is equal to Y. By prop. 8, Coeg(ff) is a transitive groupoid. Since ϖ′(f)\varpi '(f) induces the identity on the vertices, the image of Coeg(ff) under ϖ′(f)\varpi '(f) is a transitive subgroupoid of ϖ(Y)\varpi (Y). The image of ϖ(X)\varpi (X) under γ\gamma generates Coeg(ff) (II, p. 200, corollary); since one has ϖ(f)=\varpi (f) = ϖ′(f)∘γ\varpi '(f)\circ \gamma, the groupoid G is transitive.

Let y0y_0 be a point of Y and let H denote the subgroup Gy0G_{y_0} of π1(Y,y0)\pi_1(Y, y_0). By Th. 1 of IV, p. 342, there exists a connected covering (T,p)(T, p) of Y and a point t0t_0 of the fibre Ty0T_{y_0} of which H is the stabilizer.

If xx is a point of X, the set Fly0,f(x)(G)_{y_0,f(x)}(G) is not empty, since G is transitive. For u∈u\in Fly0,f(x)(G)_{y_0,f(x)}(G), the point t0⋅ut_0\cdot u is a point of the fibre Tf(x)T_{f(x)}, independent of uu since the group H, the isotropy group of G at y0y_0, fixes t0t_0. Let us denote this point by σ(x)\sigma (x); let σ:X→T\sigma : X\rightarrow T denote the resulting mapping and s:X→X×YTs: X\rightarrow X\times_YT the mapping x↦(x,σ(x))x\mapsto (x, \sigma (x)). By construction, the mapping ss is compatible with the canonical operations of ϖ(X)\varpi (X) in X and X×YTX\times_YT. It then follows from lemma 1 of III, p. 312 that ss is continuous. The mapping σ\sigma is therefore continuous; moreover it is compatible with the equivalence relation defined by ff, for σ(x)\sigma (x) depends only on f(x)f(x). Since ff is strict and surjective, there exists a unique continuous mapping σ‾:Y→T\overline{\sigma}: Y\rightarrow T such that σ‾∘f=σ\overline{\sigma}\circ f=\sigma, so that the covering T admits a section. Since T is connected, the mapping p:T→Yp: T\rightarrow Y is a homeomorphism (I, p. 31, cor. 4 of prop. 6), and one has H=π1(Y,y0)H =\pi_1(Y, y_0), whence Gy0=π1(Y,y0)G_{y_0}=\pi_1(Y, y_0). Since G is transitive, it follows that G=ϖ(Y)G =\varpi (Y).

Proposition 10 022Q

Let X and Y be topological spaces and let ff be a continuous mapping of X into Y. Suppose that the space X is simply connected, that the space X×YXX\times_YX is locally arcwise connected, and that every descent datum relative to ff on a covering of X is effective and the quotient space is a covering of Y. Then the groupoid morphism ϖ′(f)\varpi '(f) of Coeg(ff) into ϖ(Y)\varpi (Y) is injective.

Under the hypotheses of the proposition, ff is strict ( IV, p. 394, remark); moreover, one may suppose that it is surjective.

Lemma 2 022R

Let us retain the notation and hypotheses of the proposition. Let T be a set endowed with an operation of the groupoid Coeg(ff) relative to a mapping q:T→Yq: T\rightarrow Y. Then there exists a unique topology on T for which qq makes T a covering of Y, such that for every class of paths with origin cc in X and every point tt of T, one has t⋅γ(c)=t⋅f∗(c)t\cdot \gamma (c) =t\cdot f_*(c). In particular, the operation of Coeg(ff) on T factors through an operation of the groupoid ϖ(Y)\varpi (Y).

The uniqueness of such a topology on T follows from Lemma 1 (III, p. 312); let us prove its existence. Endow the set X×YTX\times_YT with a law of operation of ϖ(X)\varpi (X) by putting (x,t)⋅u=(x⋅u,t⋅γ(u))(x, t)\cdot u= (x\cdot u, t\cdot \gamma (u)) for every (x,t)∈X×YT(x, t)\in X\times_YT and every arrow uu with origin xx in ϖ(X)\varpi (X). Since the space X is simply connected, this law of operation of ϖ(X)\varpi (X) is without local monodromy (III, p. 313, Remark). There therefore exists on the set X×YTX\times_YT a unique topology which makes the X-space (X×YT(X\times_YT, pr1)_1) a covering of X and such that the canonical operation of ϖ(X)\varpi (X) in this covering coincides with the given operation (III, p. 313, Prop. 3). Let us denote this covering by (Z,p)(Z, p). Let us prove that the mapping τ:((x1,t),x2)↦(x2,t)\tau : ((x_1, t), x_2)\mapsto (x_2, t) of Z×YXZ\times_YX into Z is a descent datum on Z relative to the mapping ff. It is enough to verify that it is continuous, the other conditions of Definition 1 of IV, p. 382, being evident.

The mapping τ\tau is a lifting of the mapping pr2:Z×YX→X_2: Z\times_YX\rightarrow X to the covering Z of X. Since the space X×YXX\times_YX is locally arcwise connected, the space Z×YXZ\times_YX, which is a covering of it, is also locally arcwise connected. By the Corollary (III, p. 269), in order to prove that the mapping τ\tau is continuous, it is enough to prove that it is arcwise continuous.

Let therefore c~=((c,g),c′)\widetilde{c}= ((c, g), c') be a path in Z×YXZ\times_YX and let us prove that the mapping τ∘c~:t↦(c′(t),g(t))\tau \circ \widetilde{c}:t\mapsto (c'(t), g(t)) of I\mathbf{I} into Z is continuous. For every s∈[0,1]s\in [0,1], let us denote by csc_s and cs′c'_s the paths in X defined by t↦c(st)t\mapsto c(st) and t↦c′(st)t\mapsto c'(st). For every s∈[0,1]s\in [0,1], one thus has c(s)=c(0)⋅[cs]c(s) =c(0)\cdot [c_s], c′(s)=c′(0)⋅[cs′]c'(s) =c'(0)\cdot [c'_s] and (c,g)(s)=(c(0),g(0))⋅[cs](c, g)(s) = (c(0), g(0))\cdot [c_s], where [u][u] denotes the strict homotopy class of a path u(u(III, p. 304, Remark). The mapping t↦(cs(t),cs′(t))t\mapsto (c_s(t), c'_s(t)) is a path in X×YXX\times_YX; by definition of the coequalizer Coeg(ff), one therefore has γ([cs])=γ([cs′])\gamma ([c_s]) =\gamma ([c'_s]) and g(s)=g(0)⋅γ([cs])=g(s) =g(0)\cdot \gamma ([c_s]) = g(0)⋅γ([cs′])g(0)\cdot \gamma ([c'_s]). By definition of the operation of ϖ(X)\varpi (X) on Z, one therefore has

(τ∘c~)(s)=(c′(s),g(s))=(c′(0)⋅[cs′],g(0)⋅γ([cs′])) (\tau \circ \widetilde{c})(s) = (c'(s), g(s)) = (c'(0)\cdot [c'_s], g(0)\cdot \gamma ([c'_s]))
=(c′(0),g(0))⋅[cs′]=(τ∘c~)(0)⋅[cs′] = (c'(0), g(0))\cdot [c'_s] = (\tau \circ \widetilde{c})(0)\cdot [c'_s]

This shows that τ∘c~\tau \circ \widetilde{c} is a continuous lifting to Z of the path c′(c'(loc. cit.). Consequently, the mapping τ\tau is arcwise continuous, hence continuous; it is a descent datum relative to ff on the X-space Z.

Let RτR_{\tau} denote the equivalence relation defined by the descent datum τ\tau. Since the mapping ff is surjective, the mapping pr2:Z→T_2: Z\rightarrow T induces, by passing to the quotient, a bijection of Z/RτZ/R_{\tau} onto T. Let us endow T with the topology deduced from that of Z/RτZ/R_{\tau} by transport of structure, so that (T,q)(T, q) is a Y-space. By hypothesis, T is therefore a covering of Y and the diagram

Z pr2^{pr_2} T

pqpq

X f^f Y

is a cartesian square.

Let (x,t)(x, t) be a point of Z, cc a path with origin xx in X, and let c~\widetilde{c} be the continuous lifting of cc to Z with origin (x,t)(x, t). The path pr2∘c~_2\circ \widetilde{c} is the lifting with origin tt of the path f∘cf\circ c in Y, so that t⋅γ([c])=t\cdot \gamma ([c]) = t⋅[f∘c]=t⋅f∗([c])t\cdot [f\circ c] =t\cdot f_*([c]). This completes the proof of the lemma.

Let us now prove the proposition. Let uu and vv be two arrows of Coeg(ff) whose images in ϖ(Y)\varpi (Y) are equal. Since the groupoid morphism ϖ′(f)\varpi '(f) is the identity on the vertex sets, the arrows uu and vv have the same origin and the same target. Let us denote by yy the origin of uu; let T be the set of arrows of Coeg(ff) with origin yy, and let q:T→Yq: T\rightarrow Y be the restriction to T of the target mapping. The groupoid Coeg(ff) operates by composition on the right on the set T, relative to qq. By lemma 2, the actions of uu and vv on T are identical. Hence u=u= ey⋅u=ey⋅v=ve_y\cdot u=e_y\cdot v=v. This proves that the groupoid morphism ϖ′(f)\varpi '(f) is injective.

Theorem 1 022S

Let X and Y be topological spaces, and let f:X→f: X\rightarrow Y be a continuous surjective mapping. Suppose that the spaces X and Y are locally path-connected and semi-locally simply connected. Finally, suppose that one of the following properties is satisfied:

(i) The mapping ff is proper, separated, with locally connected fibres, and the space X×YXX\times_YX is locally path-connected;

(ii) The mapping ff is proper, separated, with finite fibres, the diagonal ΔX\Delta_X is open in X×YXX\times_YX, and its complement is locally connected;

(iii) The mapping ff is open and has the path-lifting property. Then the groupoid morphism ϖ′(f)\varpi '(f) is an isomorphism of the groupoid Coeg(ff) onto the Poincaré groupoid ϖ(Y)\varpi (Y).

First let us note that under these hypotheses, the mapping ff is surjective and strict (I, p. 18, example 2). Moreover, every descent datum relative to ff on a covering of X is effective, and the quotient space is a covering of Y; this follows indeed from IV, p. 393, corollary 2 of prop. 4 under hypotheses (i) and (ii), and from prop. 7 of IV, p. 394 under hypothesis (iii). By prop. 10, the groupoid morphism ϖ′(f)\varpi '(f) is therefore injective and its image is a subgroupoid of ϖ(Y)\varpi (Y).

By prop. 9 of IV, p. 395, this image is equal to ϖ(Y)\varpi (Y) under hypotheses (i) and (ii), but also under hypothesis (iii), since the groupoid morphism ϖ(f)\varpi (f) is then surjective.

Consequently, ϖ′(f)\varpi '(f) is an isomorphism.

Example 022T

Here are two examples in which the hypotheses of theorem 1 are satisfied.

1) Let Y be an unlaceable topological space. Let (Ai)i∈I(A_i)_{i\in I} be a locally finite covering of Y by closed sets. Suppose that, for every i∈Ii\in I, the space AiA_i is unlaceable and that, for every pair (i,j)∈I×I(i, j)\in I\times I, the space Ai∩AjA_i\cap A_j is locally path-connected. One may take for X the sum space of the family (Ai)i∈I(A_i)_{i\in I} and for f:X→Yf: X\rightarrow Y the mapping deduced from the family of canonical injections.

2) Let G be a discrete group operating properly in an unlaceable topological space X, put Y=X/GY = X/G and let f:X→Yf: X\rightarrow Y be the canonical mapping. It is open (TG, III, p. 10, lemma 2) and has the path-lifting property by virtue of Theorem 4 of III, p. 287. The space Y is unlaceable by IV, p. 349, Prop. 8, b). By hypothesis, the mapping from G×XG\times X into X×XX\times X given by (g,x)↦(g⋅x,x)(g, x)\mapsto (g\cdot x, x) is proper, hence strict (I, p. 18, Example 2) and its image is X×YXX\times_YX. It then follows from III, p. 261, Prop. 8 that X×YXX\times_YX is locally path-connected.

7. Descent by an étale and surjective mapping

Let X and Y be topological spaces and let ff be a continuous mapping from X into Y. We retain the notation of the preceding No.

Theorem 2 022U

Suppose that every point of Y possesses a neighbourhood above which there exists a continuous section of the mapping ff. The groupoid morphism ϖ′(f)\varpi '(f) from Coeg(ff) into ϖ(Y)\varpi (Y) is an isomorphism.

By hypothesis, there exists a covering (Uj)j∈J(U_j)_{j\in J} of Y by open sets and, for every j∈Jj\in J, a continuous section sjs_j of fUjf_{U_j}.

If cc is a path in X, we denote by [c][c] its strict homotopy class in ϖ(X)\varpi (X) and by {c}\{c\} the image of [c][c] in Coeg(ff) by the groupoid morphism ϖ′(f)\varpi '(f). If cc and c′c' are two paths in X,{c},\{c\} and {c′}\{c'\} are composable in Coeg(ff) if and only if the paths f∘cf\circ c and f∘c′f\circ c' in Y are juxtaposed.

Let c′c' be a path in Y. By Lemma 4 of III, p. 272, applied

to the compact space I\mathbf{I} and to the covering ((−c′1)(Uj)(^-{c'}^1)(U_j))j∈J_{j\in J} of I\mathbf{I}, there exists an integer nn such that, for every integer kk satisfying 1⩽k⩽n1\leqslant k\leqslant n, the image of the interval [k−1n,kn][\frac{k-1}{n},\frac{k}{n}] by c′c' is contained in an open set Uj(k)U_{j(k)}. For every integer k,1⩽k⩽nk,1\leqslant k\leqslant n, let ck′c'_k be the path in Y defined by s↦c′(k+s−1n)s\mapsto c'(\frac{k+s-1}{n}); one has [c′]=[c1′][c2′]...[cn′][c'] = [c'_1][c'_2]. . .[c'_n] (cf. III, p. 291, remark 1), and c′c' is the path denoted by c1′∗c2′∗⋯∗cn′c'_1*c'_2* \cdots *c'_n. For every k∈{1,...,n}k\in \{1, . . . , n\}, let us denote by ckc_k the path sj(k)∘ck′s_{j(k)}\circ c'_k in X and put {ck}=γ([ck])\{c_k\}=\gamma ([c_k]). Since for every kk, the paths ck−1′=f∘ck−1c'_{k-1}=f\circ c_{k-1} and ck′=f∘ckc'_k=f\circ c_k can be juxtaposed, the sequence ({c1},...,{cn})(\{c_1\}, . . . ,\{c_n\}) is composable in Coeg(ff). By construction,

ϖ′(f)({c1}...{cn})=ϖ′(f)({c1})...ϖ′(f)({cn}) \varpi '(f)(\{c_1\}. . .\{c_n\}) =\varpi '(f)(\{c_1\}). . . \varpi '(f)(\{c_n\})
=ϖ(f)([c1])...ϖ(f)([cn]) =\varpi (f)([c_1]). . . \varpi (f)([c_n])
=[f∘c1]...[f∘cn] = [f\circ c_1]. . .[f\circ c_n]
=[c1′]∗⋯∗[cn′]=[c′] = [c'_1]* \cdots *[c'_n] = [c']

which proves that the groupoid morphism ϖ′(f)\varpi '(f) is surjective.

Let uu and vv be arrows of Coeg(ff). Since the groupoid Coeg(ff) is generated by the image of ϖ(X)\varpi (X) (II, p. 200, corollary), there exist finite sequences (c1,...,cn)(c_1, . . . , c_n) and (d1,...,dn)(d_1, . . . , d_n) of paths in X such that one has u={c1}...{cn}u=\{c_1\}. . .\{c_n\} and v={d1}...{dn}v=\{d_1\}. . .\{d_n\}. The paths (f∘c1,...,f∘cn)(f\circ c_1, . . . , f\circ c_n) in Y can then be juxtaposed and one has ϖ′(f)(u)=[(f∘c1)]...[(f∘\varpi '(f)(u) = [(f\circ c_1)]. . .[(f\circ cn)]c_n)]; analogously, ϖ′(f)(v)=[(f∘d1)]...[(f∘dn)]\varpi '(f)(v) = [(f\circ d_1)]. . .[(f\circ d_n)].

Suppose that one has ϖ′(f)(u)=ϖ′(f)(v)\varpi '(f)(u) =\varpi '(f)(v). There then exists a strict homotopy σ\sigma joining (f∘c1)∗⋯∗(f∘cn)(f\circ c_1)* \cdots *(f\circ c_n) to (f∘d1)∗⋯∗(f∘dn)(f\circ d_1)* \cdots *(f\circ d_n). By lemma 4 of III, p. 272, applied to the compact space I×I\mathbf{I}\times \mathbf{I} and to the covering (σ−1(Ui))i∈I)(\overset{-1}{\sigma}(U_i))_{i\in I}) of I×I\mathbf{I}\times \mathbf{I}, there exists an integer m⩾1m\geqslant 1 such that, for every pair of integers (j,k)(j, k) satisfying 1⩽j⩽m1\leqslant j\leqslant m and 1⩽k⩽m1\leqslant k\leqslant m, the image of [j−1m,jm]×[k−1m,km][\frac{j-1}{m},\frac{j}{m}]\times [\frac{k-1}{m},\frac{k}{m}] by σ\sigma is contained in an open set Ui(j,k)U_{i(j,k)} of the covering (Ui)i∈I(U_i)_{i\in I}.

Every path cc in X is of the form c1∗⋯∗cmc_1* \cdots *c_m, where ckc_k is the path t↦c(k−1+tm)t\mapsto c(\frac{k-1+t}{m}). Replacing the integers mm and nn by their product mnmn if necessary, one may therefore suppose that m=nm=n.

For every pair (j,k)(j, k) of integers of {1,...,n}\{1, . . . , n\} and every pair (s,t)∈(s, t)\in I×I\mathbf{I}\times \mathbf{I}, set

σj,k(s,t)=si(j,k)∘σ(s+j−1n,t+k−1n) \sigma_{j,k}(s, t) =s_{i(j,k)}\circ \sigma (\frac{s + j- 1}{n},\frac{t + k- 1}{n})

For t∈It\in \mathbf{I}, also set hj,k0(t)=σj,k(t,0)h^0_{j,k}(t) =\sigma_{j,k}(t,0), hj,k1(t)=σj,k(t,1)h^1_{j,k}(t) =\sigma_{j,k}(t,1), vj,k0(t)=σj,k(0,t)v^0_{j,k}(t) =\sigma_{j,k}(0, t) and vj,k1(t)=σj,k(1,t)v_{j,k}^1(t) =\sigma_{j,k}(1, t). By lemma 1 of III, p. 295, the paths hj,k0∗vj,k1h^0_{j,k}*v_{j,k}^1 and vj,k0∗hj,k1v_{j,k}^0*h^1_{j,k} are strictly homotopic, whence the relation

[hj,k0][vj,k1]=[vj,k0][hj,k1](2) [h^0_{j,k}][v_{j,k}^1] = [v_{j,k}^0][h^1_{j,k}] \tag{2}

in ϖ(X)\varpi (X), for every pair (j,k)∈{1,...,n}2(j, k)\in \{1, . . . , n\}^2. On the other hand, for every pair of integers (j,k)(j, k), with 2 ⩽j⩽n\leqslant j\leqslant n and 1 ⩽k⩽n\leqslant k\leqslant n, one has f∘vj,k0=f∘vj−1,k1f\circ v_{j,k}^0=f\circ v_{j-1,k}^1, whence the relation

{vj,k0}={vj−1,k1}(3) \{v^0_{j,k}\}=\{v_{j-1,k}^1\} \tag{3}

in Coeg(ff). Analogously, for every pair of integers (j,k)(j, k) such that 1⩽j⩽n1\leqslant j\leqslant n and 2⩽k⩽n2\leqslant k\leqslant n, one has

{hj,k0}={hj,k−11}(4) \{h^0_{j,k}\}=\{h^1_{j,k-1}\} \tag{4}

For j∈{1,...,n}j\in \{1, . . . , n\}, the paths f∘cjf\circ c_j and f∘hj,10f\circ h^0_{j,1} coincide. By definition of the coequalizer Coeg(ff), one therefore has {cj}={hj,10}\{c_j\}=\{h^0_{j,1}\}. Analogously, for every j∈{1,...,n},{dj}={hj,n1}j\in \{1, . . . , n\},\{d_j\}=\{h^1_{j,n}\}. Hence one has the relations

u={h1,10}...{hn,10},v={h1,n1}...{hn,n1} u=\{h^0_{1,1}\}. . .\{h^0_{n,1}\},v=\{h^1_{1,n}\}. . .\{h^1_{n,n}\}

By lemma 3 below, one has

{h1,10}...{h1,n0}{v1,n1}...{vn,n1}={v1,10}...{vn,10}{hn,11}...{hn,n1} \{h^0_{1,1}\}. . .\{h^0_{1,n}\}\{v_{1,n}^1\}. . .\{v^1_{n,n}\}=\{v_{1,1}^0\}. . .\{v_{n,1}^0\}\{h^1_{n,1}\}. . .\{h^1_{n,n}\}

Since the paths t↦σ(0,t)t\mapsto \sigma (0, t) and t↦σ(1,t)t\mapsto \sigma (1, t) are constant, one has, for 1⩽k⩽n,{v1,k0}=ea1\leqslant k\leqslant n,\{v^0_{1,k}\}=e_a and {vn,k1}=eb\{v^1_{n,k}\}=e_b where aa and b∈Yb\in Y are the origin and the end of the arrows uu and vv of Coeg(ff). It follows that

{h1,10}...{h1,n0}={hn,10}...{hn,n1} \{h^0_{1,1}\}. . .\{h^0_{1,n}\}=\{h^0_{n,1}\}. . .\{h^1_{n,n}\}

that is, u=vu=v.

Lemma 3 022V

Let G be a groupoid and let p,qp, q be integers ⩾1\geqslant 1. For every pair (j,k)(j, k) of integers such that 0⩽j⩽p0\leqslant j\leqslant p and 0⩽k⩽q0\leqslant k\leqslant q, let xj,kx_{j,k} be a vertex of G; for every pair (j,k)(j, k) such that 1⩽j⩽p1\leqslant j\leqslant p and 0⩽k⩽q0\leqslant k\leqslant q, let hj,kh_{j,k} be an arrow of G joining xj−1,kx_{j-1,k} to xj,kx_{j,k}; for every pair (j,k)(j, k) such that 0⩽j⩽p0\leqslant j\leqslant p and 1⩽k⩽q1\leqslant k\leqslant q, let vj,kv_{j,k} be an arrow of G joining xj,k−1x_{j,k-1} to xj,kx_{j,k}.

Assume that, for every pair (j,k)(j, k) of integers such that 1⩽j⩽p1\leqslant j\leqslant p and 1⩽k⩽q1\leqslant k\leqslant q, the arrows vj−1,kv_{j-1,k} and hj,kh_{j,k} are composable, analogously the arrows hj,k−1h_{j,k-1} and vj,kv_{j,k}, and that one has vj−1,khj,k=hj,k−1vj,kv_{j-1,k}h_{j,k}=h_{j,k-1}v_{j,k}. Then,

h1,0h2,0...hp,0vp,1vp,2...vp,q=v0,1v0,2...v0,qh1,qh2,q...hp,q h_{1,0}h_{2,0}. . . h_{p,0}v_{p,1}v_{p,2}. . . v_{p,q}=v_{0,1}v_{0,2}. . . v_{0,q}h_{1,q}h_{2,q}. . . h_{p,q}

Let us first treat the particular case where q=1q= 1 and prove the result by induction on pp. If p=1p= 1, the assertion to be proved is true by assumption; suppose it true for p−1p-1; one then has

h1,0h2,0...hp,0vp,1=h1,0h2,0...hp−1,0vp−1,1hp,1=v0,1h1,1...hp,1 h_{1,0}h_{2,0}. . . h_{p,0}v_{p,1}=h_{1,0}h_{2,0}. . . h_{p-1,0}v_{p-1,1}h_{p,1}=v_{0,1}h_{1,1}. . . h_{p,1}

by the induction hypothesis, whence the relation for pp.

Let us now prove the result by induction on qq. It is true for q=1q= 1 by what precedes; if it is true for q−1q-1, one then has

h1,0h2,0...hp,0vp,1vp,2...vp,q=v0,1v0,2...v0,q−1h1,q−1...hp,q−1vp,q h_{1,0}h_{2,0}. . . h_{p,0}v_{p,1}v_{p,2}. . . v_{p,q}=v_{0,1}v_{0,2}. . . v_{0,q-1}h_{1,q-1}. . . h_{p,q-1}v_{p,q}

From the case q=1q= 1, one has

h1,q−1...hp,q−1vp,q=v0,qh1,q...hp,q h_{1,q-1}. . . h_{p,q-1}v_{p,q}=v_{0,q}h_{1,q}. . . h_{p,q}

whence the required relation.

Example 1 022W

The theorem applies when the mapping ff is étale and surjective.

Example 2 022X

It also applies when the space X is the sum space of a family (Vi)i∈I(V_i)_{i\in I} of parts of Y whose interiors cover Y, and when ff is the mapping induced by the canonical injections of each of the ViV_i into Y.

8. Poincaré Groupoid of a Quotient Space

Let X be a topological space endowed with a continuous action of a discrete group G; set Y=X/GY = X/G and denote by f:X→f: X\rightarrow Y the canonical mapping. Let ∣G∣|G| denote the groupoid ϖ(G)\varpi (G); its set of vertices is G; for g,g′∈Gg, g'\in G, there exists a unique arrow joining gg to g′g' if g′=g′g'=g', and none otherwise. By passing to the fundamental groupoids, the action m:G×X→Xm: G\times X\rightarrow X induces a morphism of groupoids ϖ(m):∣G∣×ϖ(X)→ϖ(X)\varpi (m):|G| \times \varpi (X)\rightarrow \varpi (X). Let ϖ(X)/G\varpi (X)/G be the coequalizer of the two morphisms of groupoids ϖ(m)\varpi (m) and ϖ\varpi(pr2_2) from ∣G∣×ϖ(X)|G| \times \varpi (X) to ϖ(X)\varpi (X); let β:ϖ(X)→ϖ(X)/G\beta :\varpi (X)\rightarrow \varpi (X)/G denote the canonical morphism of groupoids. We have f∘m=f∘f\circ m=f\circ pr2_2, hence ϖ(f)∘ϖ(m)=ϖ(f)∘ϖ\varpi (f)\circ \varpi (m) =\varpi (f)\circ \varpi(pr2_2). By the universal property of coequalizers, there therefore exists a unique morphism of groupoids ϖ′′(f):ϖ(X)/G→ϖ(Y)\varpi ''(f):\varpi (X)/G\rightarrow \varpi (Y) such that ϖ(f)=ϖ′′(f)∘β\varpi (f) =\varpi ''(f)\circ \beta.

Theorem 3 022Y

Let X be a delaceable topological space and let G be a discrete group operating properly in X; let f:X→X/Gf: X\rightarrow X/G be the canonical surjection. The canonical morphism of groupoids ϖ′′(f):ϖ(X)/G→ϖ(X/G)\varpi ''(f):\varpi (X)/G\rightarrow \varpi (X/G) introduced above is an isomorphism.

Let Coeg(ff) denote the coequalizer of the two morphisms of groupoids from ϖ(X×YX)\varpi (X\times_YX) to ϖ(X)\varpi (X) induced by the projections pr1_1 and pr2_2; let γ:ϖ(X)→\gamma :\varpi (X)\rightarrow Coeg(ff) be the canonical morphism of groupoids. Since the image of the mapping (m(m, pr2):G×X→X×X_2): G\times X\rightarrow X\times X is the subspace X×YXX\times_YX of X×XX\times X, the two morphisms of groupoids γ∘ϖ(m)\gamma \circ \varpi (m) and γ∘ϖ\gamma \circ \varpi(pr2_2), from ∣G∣×ϖ(X)|G|\times \varpi (X) to Coeg(ff), are equal. By the universal property of ϖ(X)/G\varpi (X)/G, there exists a unique morphism of groupoids α:ϖ(X)/G→\alpha :\varpi (X)/G\rightarrow Coeg(ff) such that γ=α∘β\gamma =\alpha \circ \beta.

Let also ϖ′(f)\varpi '(f) denote the unique morphism of groupoids from Coeg(ff) to ϖ(Y)\varpi (Y) such that ϖ(f)=ϖ′(f)∘γ\varpi (f) =\varpi '(f)\circ \gamma. By IV, p. 399, Example 2, the hypotheses of Theorem 1 of IV, p. 398 are satisfied and the morphism ϖ′(f)\varpi '(f) is an isomorphism. Since

ϖ(f)=ϖ′(f)∘γ=ϖ′(f)∘α∘β=ϖ′′(f)∘β \varpi (f) =\varpi '(f)\circ \gamma =\varpi '(f)\circ \alpha \circ \beta =\varpi ''(f)\circ \beta

we have ϖ′′(f)=ϖ′(f)∘α\varpi ''(f) =\varpi '(f)\circ \alpha. Thus, to prove Theorem 3, it is enough to prove that the morphism α\alpha is an isomorphism.

Lemma 4 022Z

For every path c=(c1,c2)c= (c_1, c_2) in X×YXX\times_YX, we have β([c1])=β([c2])\beta ([c_1]) =\beta ([c_2]).

Let cc be such a path.

Let x∈Xx\in X, and let KxK_x be its stabilizer in G. By TG, III, p. 32, prop. 8, there exists an open neighbourhood UxU_x of xx in X such that Kx⋅Ux=K_x\cdot U_x= Ux,g⋅Ux∩UxU_x,g\cdot U_x\cap U_x = ∅\emptyset for every g∈g\in G - KxK_x, and such that the mapping ff induces a homeomorphism of Ux/KxU_x/K_x onto an open neighbourhood VxV_x of f(x)f(x) in Y. As X is unlaceable and the restriction of the mapping ff to UxU_x is open and closed, we may further suppose that UxU_x is connected and that the image of the canonical homomorphism π1(Ux,x)→π1(X,x)\pi_1(U_x, x)\rightarrow \pi_1(X, x) is reduced to the identity element. The open sets (Vx)x∈X(V_x)_{x\in X} thus constructed form an open covering of Y. By lemma 4 of III, p. 272, applied to the compact space I\mathbf{I} and to the open sets (f∘c1)−1(Vx)(f\circ c_1)^{-1}(V_x) for x∈Xx\in X, there exists an integer n⩾1n\geqslant 1 such that for every i∈{1,...,n}i\in \{1, . . . , n\}, there exists a point xix_i in X such that c1([i−1n,in])c_1([\frac{i-1}{n},\frac{i}{n}])

is contained in f−1(Vxi)\overset{-1}{f}(V_{x_i}). As f∘c1=f∘c2,c2([i−1n,in])f\circ c_1=f\circ c_2,c_2([\frac{i-1}{n},\frac{i}{n}]) is also

contained in f−1(Vxi)\overset{-1}{f}(V_{x_i}).

For j=1j= 1 or 2 and for i∈{1,...,n}i\in \{1, . . . , n\}, let cj,ic_{j,i} denote the path in X defined by t↦cj(i+t−1n)t\mapsto c_j(\frac{i+t-1}{n}); we have cj=cj,1∗⋯∗cj,n(c_j=c_{j,1}* \cdots *c_{j,n}(III, p. 291, remark 1). Hence, to prove that β([c1])=β([c2])\beta ([c_1]) =\beta ([c_2]), it is enough to prove that β([c1,i])=β([c2,i])\beta ([c_{1,i}]) =\beta ([c_{2,i}]) for every integer i∈{1,...,n}i\in \{1, . . . , n\}. Replacing the pair of paths (c1,c2)(c_1, c_2) by the pair (c1,i,c2,i)(c_{1,i}, c_{2,i}) if necessary, we thus suppose that there exists x∈Xx\in X such that (f∘c1)([0,1])⊂Vx(f\circ c_1)([0,1])\subset V_x.

The inverse image of VxV_x by ff is the disjoint union of the connected components g⋅Uxg\cdot U_x, where gg runs through a system of representatives in G of G/KxG/K_x. For i=1i= 1 or 2, let gig_i be an element of G such that the point xi=x_i= gi⋅ci(0)g_i\cdot c_i(0) belongs to UxU_x. The image of the path gi⋅cig_i\cdot c_i is then contained in UxU_x. By definition of the groupoid ϖ(X)/G\varpi (X)/G, we have β([gi⋅ci])=β([ci])\beta ([g_i\cdot c_i]) =\beta ([c_i]), which makes it possible to suppose that the images of the paths c1c_1 and c2c_2 are contained in UxU_x and that g1=g2=eg_1=g_2=e.

For s=0s= 0 or 1, let dsd_s be a path in UxU_x joining xx to c1(s)c_1(s) and let gsg_s be an element of KxK_x such that gs⋅c2(s)=c1(s)g_s\cdot c_2(s) =c_1(s). The paths d0∗c1∗d1‾d_0*c_1*\overline{d_1} and d0∗(g0⋅c2)∗(g0g1−1⋅d1‾)d_0*(g_0\cdot c_2)*(g_0g_1^{-1}\cdot \overline{d_1}) are loops at xx in UxU_x. They are therefore strictly homotopic in X to the constant loop at xx, because the image of the canonical homomorphism π1(Ux,x)→π1(X,x)\pi_1(U_x, x)\rightarrow \pi_1(X, x) reduces to the identity element. Their classes have in particular the same image under the groupoid morphism β\beta, whence

β([d0])β([c1])β([d1])−1=β([d0])β([g0⋅c2])β([g0g1−1⋅d1])−1 \beta ([d_0])\beta ([c_1])\beta ([d_1])^{-1}=\beta ([d_0])\beta ([g_0\cdot c_2])\beta ([g_0g_1^{-1}\cdot d_1])^{-1}

Since β([g⋅c])=β([c])\beta ([g\cdot c]) =\beta ([c]) for every element g∈Gg\in G and every path cc in X, it follows that β([c1])=β([c2])\beta ([c_1]) =\beta ([c_2]), as was to be proved.

By the lemma, the two groupoid morphisms β∘ϖ\beta \circ \varpi(pr1_1) and β∘ϖ\beta \circ \varpi(pr2_2) of ϖ(X×YX)\varpi (X\times_YX) into Coeg(ff) are equal. By the universal property of the coequaliser, there therefore exists a unique groupoid morphism α′:\alpha ': Coeg(ff)→ϖ(X)/G\rightarrow \varpi (X)/G such that β=α′∘γ\beta =\alpha '\circ \gamma. The morphism α′∘α\alpha '\circ \alpha is the unique groupoid morphism φ\varphi of ϖ(X)/G\varpi (X)/G into itself such that φ∘β=β\varphi \circ \beta =\beta; hence α′∘α\alpha '\circ \alpha = Idϖ(X)/G_{\varpi(X)/G}. Analogously, α∘α′=\alpha \circ \alpha '= IdCoeg(f)_{Coeg(f)}. Consequently, α\alpha is an isomorphism.

Exercises

See the exercises for § 4.

Machine translation of the English, by gpt-5-6-mini, gpt-5.4, against glossary version 34. Not checked by a person. That model is a cut down version of the one the rest of this language was written by.