1. Coequalizer of the projections of a fibre square
Let X and Y be topological spaces and let f be a continuous mapping of X into Y. Denote by Z the fibre square X×YX and by p1,p2 the two projections of Z into X. Denote by W the fibre product X×YX×YX; for every pair (s,t) of integers equal to 1, 2 or 3, denote by qst:W→Z the mapping defined by qst(x1,x2,x3)=(xs,xt).
Denote by Coeg(f) the groupoid Coeg(ϖ(p1),ϖ(p2)), coequalizer of the two morphisms ϖ(p1),ϖ(p2) of the Poincaré groupoid ϖ(Z) into the Poincaré groupoid ϖ(X). Denote by γ:ϖ(X)→ Coeg(f) the canonical morphism of groupoids. Since f∘p1 = f∘p2, the morphisms of groupoids ϖ(f)∘ϖ(p1) and ϖ(f)∘ϖ(p2) of ϖ(Z) into ϖ(Y) are equal; there thus exists a unique morphism of groupoids ϖ′(f): Coeg(f)→ϖ(Y) such that ϖ′(f)∘γ=ϖ(f).
This property is satisfied under one of the following assumptions:
(i) The spaces X and Y are delacable, the space X×YX is locally arcwise connected, the mapping f is surjective, proper, separated, with locally connected fibres;
(ii) The spaces X and Y are delacable, the mapping f is surjective, proper, separated, with finite fibres, the diagonal ΔX of X×YX is open and its complement is locally connected;
(iii) The spaces X and Y are delacable, the mapping f is surjective, open and possesses the path lifting property;
(iv) Every point of Y possesses a neighbourhood above which there exists a continuous section of the mapping f.
Indeed, under each of these assumptions, the mapping f is surjective; it is also strict by virtue of I, p. 18, example 2. Finally, the morphism ϖ′(f) is an isomorphism, both under the assumptions (i), (ii) or (iii) according to IV, p. 398, th. 1, and under assumption (iv) ( IV, p. 400, th. 2).
The prop. 5 of II, p. 208 describes the isotropy groups of the groupoid Coeg(f). The purpose of this noo is to make explicit, when f satisfies the property (VK), the description of the Poincaré groups of Y which is deduced from it by composition with the isomorphism of groupoids ϖ′(f). The following nosos will be devoted to important particular cases.
Suppose that the mapping f satisfies the property (VK).
Put I=π0(X),J=π0(Z),K=π0(W); for j∈J and s∈{1,2}, put is(j)=π0(ps)(j) ; for k∈K and s,t∈{1,2,3}, put jst(k)=π0(qst)(k). Denote by Γ the skeleton of the pair (ϖ(p1),ϖ(p2)) of morphisms of groupoids from ϖ(Z) into ϖ(X) (II, p. 185, def. 3). It is the quiver (I,J,π0(p1),π0(p2)), since the orbits of the Poincaré groupoid of a topological space are the arcwise connected components of this space.
Suppose moreover that Y is arcwise connected and nonempty. According to II, p. 200, remark 2, π0(Γ) is then in bijection with the set of orbits of the groupoid Coeg(f); since the mapping f satisfies the property (VK), the groupoid Coeg(f) is isomorphic to the groupoid ϖ(Y). The graph Γ is thus connected and nonempty.
Call van Kampen data of f the data of the following elements:
(i) for every i∈I, a point a(i) of the path-connected component i of X ;
(ii) for every j∈J, a point b(j)=(b1(j),b2(j)) of the path-connected component j of Z ;
(iii) for every k∈K, a point c(k)=(c1(k),c2(k),c3(k)) of the path-connected component k of W;
(iv) for every j∈J, the class β1(j) of a path in X joining b1(j) to a(i1(j)) and the class β2(j) of a path in X joining b2(j) to a(i2(j));
(v) for every k∈K and for every pair (s,t) equal to (1,2), (2,3) or (1,3), the class γst(k) of a path in Z joining (cs(k),ct(k)) to b(jst(k));
(vi) a subquiver T of Γ whose associated graph is a maximal tree of the graph Γ;
(vii) an element i0 of I.
Let us choose a van Kampen datum of f. Then, (a,b,β1,β2,T,i0) is a basic equipment of the pair (ϖ(p1),ϖ(p2)) of groupoid morphisms from ϖ(Z) into ϖ(X) (II, p. 192, déf. 4). Moreover, the triples
z=((q12(c(k)),1),(q23(c(k)),1),(q13(c(k)),−1))
and the classes of paths (γ1,2(k),γ2,3(k),γ1,3(k)), where k describes K, define a complementary equipment of the pair (ϖ(p1),ϖ(p2)) (II, p. 208, déf. 3 ; II, p. 205, exemple ; II, p. 205, remarque). We shall say that the complete equipment of the coequalizer Coeg(f) thus defined is deduced from the van Kampen datum of f that we have chosen.
For every j∈J, let φj:π1(Z,b(j))→π1(X,a(i1(j))) and ψj:π1(Z,b(j))→π1(X,a(i2(j))) denote the group homomorphisms defined by
Let τ denote the unique groupoid morphism from Grp(Γ) into ϖ(Y) such that the mapping Som(τ ) maps i∈I onto f(a(i)) and Fl(τ ) maps j∈J onto the class of paths f∗(β1(j))−1f∗(β2(j)) joining f(a(i1(j)))
a(i1(k))
λ1(k)
b1(j12(k))b1(j13(k))
b2(j12(k))b2(j13(k))
c1(k)
c2(k)c3(k)
λ2(k)λ3(k)
a(i2(k))b1(j23(k))b2(j23(k))a(i3(k))
to f(a(i2(j))) in Y. For i∈I, let di∈ Grp(Γ) be the class of the unique path without backtracking joining i0 to i in the tree T and put δi=τ(di).
If S is a set, recall that F(S) denotes the free group on S; the image in F(S) of an element s∈S by the canonical mapping is denoted by [s], or by s if there is no possible confusion.
Suppose that Y is path connected and that f verifies the property (VK). With the preceding notations, there exists a unique group homomorphism
L:(i∗∈Iπ1(X,a(i)))∗F(J)→π1(Y,f(a(i0)))
such that
L(v)=δif∗(v)δi−1 for i∈I and v∈π1(X,a(i)),
L(j)=δi1(j)τ(j)δi2(j)−1 for j∈J.
Moreover, the homomorphism L is surjective and its kernel is the smallest normal subgroup containing the following elements:
(R1)r1(j)=j for j in Fl(T) ;
(R2)r2(j,v)=φj(v)jψj(v)−1j−1 for j∈J and v∈π1(Z,b(j));
(R3)r3(k)=λ1(k)j12(k)λ2(k)j23(k)λ3(k)j13(k)−1, for k∈K.
The existence and uniqueness of the homomorphism L follow from the universal property of free products and free groups. Moreover, this homomorphism is the composite of the homomorphism ϖ′(f)γ(a(i0)) deduced from ϖ′(f) by passing to isotropy groups and of the group homomorphism λ defined by Prop. 5 of II, p. 208, taking into account the fact that the chosen van Kampen datum determines a complete equipment of the pair (ϖ(p1),ϖ(p2)) of groupoid morphisms from ϖ(Z) into ϖ(X). By that proposition, the image of the homomorphism λ is the group Coeg(f)γ(a(i0)) and its kernel is the smallest normal subgroup containing the elements defined by relations (R1),(R2) and (R3). On the other hand, the groupoid morphism ϖ′(f): Coeg(f)→ϖ(Y) is an isomorphism, by definition of property (VK). The theorem is therefore proved.
2. Coverings
Let Y be a nonempty arcwise connected topological space, and let (Ai)i∈I be a covering of Y by nonempty arcwise connected subsets, indexed by a totally ordered set I. Let X=⨆i∈IAi be the sum space of the family (Ai) and let f:X→Y be the mapping deduced from the family of canonical injections of each Ai into Y. Suppose that the mapping f satisfies property (VK). This occurs in particular in the following two cases:
(i) the interiors of the sets Ai, for i∈I, cover Y (cf. IV, p. 402, Example 2);
(ii) the space Y is contractible, as are the spaces Ai, for i∈I, the family (Ai)i∈I is locally finite, the Ai are closed in Y and their pairwise intersections are locally arcwise connected (cf. IV, p. 399, Example 1).
Let J′ be the set of triples (i,i′,V), where i and i′ are elements of I and where V is an arcwise connected component of Ai∩Ai′. If j=(i,i′,V)∈J′, we set i1(j)=i,i2(j)=i′ and j=(i′,i,V). Let J be the subset of J′ formed by the triples such that i<i′. The framework of the covering is the quiver Γ whose set of vertices is I, whose set of arrows is J, and whose origin and end mappings are respectively the mappings j↦i1(j) and j↦i2(j). We identify the graph associated with Γ with the graph Γ whose set of vertices is I, whose set of arrows is J∪J, whose origin and end mappings are the mappings j↦i1(j) and j↦i2(j), and whose involution is the mapping j↦j.
Let p1 and p2 denote the projections of the fibre square X×YX onto X; let Γ be the framework of the pair (ϖ(p1),ϖ(p2)) of groupoid morphisms from ϖ(X×YX) into ϖ(X).
The quiver Γ identifies with a subquiver of Γ; the quivers Γ and Γ are connected.
The arcwise connected components of X are the Ai, for i∈I. The arcwise connected components of X×YX are the (V×{i})×Y(V×{i′}), for (i,i′,V)∈J′. Consequently, the frame Γ of the pair (ϖ(p1),ϖ(p2)) is isomorphic to the quiver whose set of vertices is I, whose set of arrows is J′, the origin and term mappings being respectively the mappings j↦i1(j) and j↦i2(j). The quiver Γ is connected (IV, p. 406). Moreover, this description identifies Γ with a subquiver of Γ. Let us also observe that for every arrow j of Γ, either j∈ Fl(Γ), or there exists i∈I such that j=(i,i,Ai). It follows that the mapping π0(Γ)→π0(Γ) deduced from the injection of Γ into Γ is bijective, so that Γ is connected.
For every element i of I, choose a point a(i) of Ai.
For every element j=(i,i′,V) of J, choose a point b(j) in V, a path B1(j) joining b(j) to a(i) in Ai and a path B2(j)
joiningj=(i′,i,b(jV)) tobelongs toa(i′) in AJi′. Letand let us putj=(i,ib(′,jV)) =an element ofb(j), B1(j)=BJ2; then(j) and B2(j)=B1(j). For j∈J′∪J′, the paths B1(j) and B2(j) in Y are composable. Put
B(j)=B1(j)∗B2(j)(3)
It is a path joining a(i1(j)) to a(i2(j)) in Y; one has the relation B(j)=B(j).
For every j=(i,i′,V)∈J′, denote by pj,1:V→Ai and pj,2:V→Ai′ the canonical injections; also denote by φj:π1(V,b(j))→π1(Ai,a(i)) and ψj:π1(V,b(j))→π1(Ai′,a(i′)) the group homomorphisms defined by
φj(v)=[B1(j)]−1(pj,1)∗(v)[B1(j)]
and
ψj(v)=[B2(j)]−1(pj,2)∗(v)[B2(j)]
for v∈π1(V,b(j)) (cf. IV, p. 407).
Fix an element i0 of I, as well as a subquiver T in the quiver Γ whose associated graph T is a maximal tree of the graph Γ.
For i∈I, let (i0,j1,i1,...,jn,i) be the unique path without backtracking joining i0 to i in the tree T, and put
δ(i)=[B(j1)][B(j2)]...[B(jn)]
this is the class of a path joining a(i0) to a(i) in Y. Let αi denote the homomorphism of π1(Ai,a(i)) into π1(Y,a(i)) deduced from the canonical injection, and let μi:π1(Ai,a(i))→π1(Y,a(i0)) be the group homomorphism defined by
μi(v)=δ(i)αi(v)δ(i)−1
Finally, let μ:F(J)→π1(Y,a(i0)) be the unique group homomorphism such that one has
μ(j)=δ(i1(j))[B(j)]δ(i2(j))−1
for every j∈J. There exists a unique group homomorphism
M:(i∗∈Iπ1(Ai,a(i)))∗F(J)→π1(Y,a(i0))
which coincides with μi on π1(Ai,a(i)), for every i∈ I, and with μ on F(J).
Let K′ be the set of quadruples (i1,i2,i3,U), where i1,i2,i3 are elements of I and where U is a pathwise connected component of Ai1∩Ai2∩Ai3. For every element k=(i1,i2,i3,U) of K′ and every pair (s,t) of elements of {1,2,3}, put jst(k)=(is,it,V), where V is the pathwise connected component of Ais∩Ait which contains U; this is an element of J′.
Let K be the subset of K′ formed by the quadruples (i1,i2,i3,U) such that i1<i2<i3. For every element k=(i1,i2,i3,U) of K, choose a point c(k) of U, together with paths C12(k), C23(k) and C13(k), such that Cst(k) joins c(k) to b(jst(k)) in Ais∩Ait for s,t∈{1,2,3} with s<t.
Let Y be a pathwise connected topological space and let (Ai)i∈I be a covering of Y by non-empty, pathwise connected subsets, indexed by a totally ordered set I. Suppose that the canonical mapping of the sum space of the family (Ai)i∈I into Y satisfies property (VK). Then the homomorphism M introduced above is surjective, and its kernel is the smallest normal subgroup containing the following elements:
Let X be the topological sum of the family (Ai)i∈I and let f:X→Y be the mapping induced by the family of canonical injections of each Ai into Y. The mapping f satisfies property (VK) by hypothesis; we shall therefore apply Theorem 1 of Noo1 to it. We therefore take up again the notation of this Noo and begin by defining a van Kampen datum for the mapping f.
The arcwise connected components of X are the sets Xi=Ai×{i} for i∈I. Thus π0(X) is identified with the set I. For each i∈I, let a(i) denote the point (a(i),i) of Xi.
Put Z=X×YX. The arcwise connected components of Z are the sets Zj=(V×{i})×Y(V×{i′}), where j=(i,i′,V) runs through J′. Thus π0(Z) is identified with the set J′. Let J0 be the set of elements of J of the form (i,i,Ai), so that the family (J0,J,J) is a partition of J′. For i∈I and j=(i,i,Ai)∈J0, put b(j)=(a(i),a(i)) and take for β1(j) and β2(j) the class of the constant path at a(i). For j=(i,i′,V)∈J∪J, let b(j) denote the point ((b(j),i),(b(j),i′)) of Zj and let β1(j) and β2(j) denote the classes of the paths t↦(B1(j)(t),i) and t↦(B2(j)(t),i′) in X.
Put W=X×YX×YX. The arcwise connected components of W are the sets Wk=(U×{i1})×Y(U×{i2})×Y(U×{i3}), where k=(i1,i2,i3,U) runs through K′. Thus π0(W) is identified with the set K′.
Let K0 denote the set of elements of K′ of the form k=(i,i,i,Ai), for i∈I. For such an element k∈K0, put c(k)=(a(i),a(i),a(i)) and choose for γst(k) the class of the constant path at (a(i),a(i)).
Let K1 denote the set of elements of K′ of the form k = (i1,i2,i3,V) for which the set {i1,i2,i3} has two elements. Let k be an element of K′ of the form (i,i,i′,V), so that j=(i,i′,V) belongs to J∪J. We then put
and we take for γ13(k) and γ23(k) the class of the constant path at b(j). One defines analogously c(k),γ12(k),γ13(k),γ23(k) for every element k of K′.
For every element k=(i1,i2,i3,U) of K, put
c(k)=((c(k),i1),(c(k),i2),(c(k),i3))
for every pair (s,t) of distinct elements of {1,2,3}, take for γst(k) the image under the mapping x↦((x,s),(x,t)) of the class of the path Cst(k) in Ais(k)∩Ait(k).
For every point x=(x1,x2,x3) of X×YX×YX and every permutation σ∈S3, put σ(x)=(xσ−1(1),xσ−1(2),xσ−1(3)). Thus an operation of the group S3 on W is defined. For every k=(i1,i2,i3,U)∈K′ and every permutation σ∈S3, put analogously σ(k)=(iσ−1(1),iσ−1(2),iσ−1(3),U); one has
Let k=(i1,i2,i3,U) be an element of K′ such that i1,i2,i3 are pairwise distinct. There exists a unique permutation σ∈S3 such that iσ−1(1)<iσ−1(2)<iσ−1(3), so that σ(k)=(iσ−1(1),iσ−1(2),iσ−1(3),U) belongs to K. For s∈{1,2,3}, one then puts cs(k)=cσ(s)(σ(k)) and c(k)=(c1(k),c2(k),c3(k)), so that c(k)=σ−1(c(σ(k))). For (s,t)∈{(1,2),(1,3),(2,3)}, one defines Cst(k)=Cσ−1(s)σ−1(t)(σ(k)); it is a path joining c(k) to b(jσ(s)σ(t)(σ(k)))=b(jst(k)).
Let g denote the morphism of quivers from Γ into Γ which associates with a vertex i∈I of Γ the vertex Xi=Ai×{i} of Γ and with an arrow j=(i,i′,V)∈J′ of Γ the arrow Zj=(V×{i})×Y(V×{i′}) of Γ. The mapping Som(g) is bijective; the mapping Fl(g) is injective and the graph Γ is connected (IV, p. 410, lemma 1), and the image under g of the subquiver T is a subquiver T of Γ whose associated graph is a maximal tree of the graph Γ.
The points a(i), for i∈I, the points b(j), for j∈J′, the points c(k), for k∈K′, the classes of paths β1(j) and β2(j), for j∈J′, the classes of paths γst(k), for k∈K′, the subquiver g(T) of Γ and the element i0 of I define a van Kampen datum for f.
which induces the isomorphism of π1(X,a(i)) onto π1(Ai,a(i)) deduced from the identification of Ai×{i} and Ai, for all i∈I, and such that one has
ρ(j)=1 for j∈J0
ρ(j)=j,ρ(j)=j−1 for j∈J.
Let L be the group homomorphism defined in Theorem 1 of IV, p. 408. For j=(i,i,Ai)∈J0, one has L(j)=1=M∘ρ(j). Let j=(i,i′,V) be an element of J; one has L(j)=(M∘ρ)(j) by definition. Finally, if j=(i,i′,V) is an element of J,j∈J and one verifies that
L(j)=L(j)−1=M(ρ(j))−1=M(ρ(j))
Consequently, one has M∘ρ=L.
The homomorphism L is surjective (loc. cit.), hence the homomorphism M is also. Since the homomorphism ρ is surjective, the kernel of M is the smallest normal subgroup of (i∗∈Iπ1(Ai,a(i)))∗F(J1) which contains the images under ρ of the elements defined by relations (R1), (R2),(R3) of theorem 1 (IV, p. 408). The proof will be complete once we have verified that these images are, besides the elements defined by relations (R1),(R2),(R3) of Prop. 1, elements conjugate to them, or conjugate to their inverses, together with the identity element.
Elements R1. — An arrow of the oriented tree T is of the form Zj, with j=(i,i′,V)∈J; its image is the element j of F(J).
Elements R2. — Let j=(i,i′,V)∈J′. If i=i′, we have ρ(r2(j,v))=1 for every v∈π1(Ai,a(i)). If j∈J, the image of r2(j,v) is the element r2(j,v)=φj(v)jψj(v)−1j−1, for every v∈π1(Z,b(j)). In the remaining case, we have j∈J and the equality
implies that ρ(r2(j,v)) is conjugate to ρ(r2(j,v−1)).
Elements R3. — Let k=(i1,i2,i3,U) be an element of K′.
If k∈K0,i1 = i2 = i3,λs(k) is the class of the trivial path for every s∈{1,2,3},jst(k)∈J0 for every pair (s,t)∈{(1,2),(1,3),(2,3)}. Then ρ(r3(k)) is the identity element.
Suppose k∈K1. If i1=i2, then j=(i1,i3,U)∈J∪J and we have
r3(k)=β1(j)−1j12β1(j)j23β2(j)−1β2(j)j13−1
whose image under ρ is the identity element. The other cases are dealt with analogously.
Suppose that i1,i2,i3 are pairwise distinct. If i1<i2<i3, k∈K and the image of ρ(r3(k)) is the element r3(k).
Let σ∈S3 be the permutation carrying 1 to 2 and 2 to 3. We have
Since the group S3 is generated by the permutations τ and σ, it follows that, for every k∈K and every σ∈S3,ρ(r3(σ(k))) is conjugate to ρ(r3(k)) or to its inverse.
Under the assumptions of proposition 1, suppose moreover that, for every i∈ I, the image of the homomorphism from π1(Ai,a(i)) into π1(Y,a(i)), induced by the canonical injection of Ai into Y, is trivial. The homomorphism M′:F(J)→π1(Y,a(i0)) induced by M by restriction is then surjective, and its kernel is the smallest distinguished subgroup containing the elements j∈ Fl(T) and the elements j12(k)j23(k)j13(k)−1 for k∈K.
Let π:(i∗∈Iπ1(Ai,a(i)))∗F(J)→F(J) be the unique homomorphism which induces the trivial homomorphism on each π1(Ai,a(i)) and the identity on F(J); it is surjective. Let i∈I. The definition of M and the assumption that the image of the homomorphism from π1(Ai,a(i)) into π1(Y,a(i)) induced by the canonical injection is trivial imply that, for every v∈π1(Ai,a(i)),M(v) is the identity element of π1(Y,a(i0)). Hence M=M′∘π. It follows that the homomorphism M′ is surjective, and its kernel is the smallest distinguished subgroup containing the images under π of the elements r1(j),r2(j,v) and r3(k) defined in Prop. 1. For j∈ Fl(T), we have π(r1(j))=j. For every j∈ J and every v∈π1(V,b(j)), we have π(r2(j,v))=e. Finally, for every k=(i1,i2,i3,U)∈K and every s∈{1,2,3},λs(k) is the class of a loop in Ais; hence π(λs(k))=e, so that π(r3(k))=j12(k)j23(k)j13(k)−1. The corollary follows.
Under the assumptions of proposition 1, suppose moreover that for every i∈I, the group π1(Ai,a(i)) is reduced to the identity element and that, for every triple (i1,i2,i3) of pairwise distinct elements of I, the set Ai1∩Ai2∩Ai3 is empty. The homomorphism M′′: F(J-Fl(T)) →π1(Y,a(i0)) induced by M by restriction is then an isomorphism.
Let π′:F(J)→ F(J - Fl(T)) be the homomorphism which maps [j] to [j] if j∈ J-Fl(T) and which maps [j] to the identity element if j∈ Fl(T); it is surjective and one has M′′∘π′=M′, where M′ is the surjective homomorphism defined in corollary 1. It follows that M′′ is surjective and that its kernel is the smallest distinguished subgroup of F(J-Fl(T)) which contains the images under π′ of the elements described in loc. cit. But, by construction, π′(j)=e if j∈ Fl(T) and the set K is empty, by hypothesis. The homomorphism M′′ is therefore an isomorphism.
Let G be a graph (II, p. 155, definition 1); let S denote the set of vertices of G, A the set of its oriented edges, o and t the origin and term mappings of A into S; for every oriented edge a∈A, let a denote the opposite oriented edge. Endow the sets S and A with the discrete topology; let X be the sum space of the space S and of the space I×A, and let ∼ be the finest equivalence relation in X for which (u,a)∼(1−u,a), (0,a)∼o(a) and (1,a)∼t(a) for every u∈I and every oriented edge a∈A. The quotient space ∣G∣=X/∼ is called the geometric realization of the graph G. Let p denote the canonical projection of X onto ∣G∣.
Let us prove that ∣G∣ is locally contractible. Let s∈S. Let Xs denote the union of {s} and of the sets [0,1[×{a} for a∈o−1(s) and of the
sets ]0,1]×{a} for a∈t−1(s). Let Us be the image of Xs in ∣G∣; it is an open neighbourhood of p(s) in ∣G∣ because Xs is an open saturated neighbourhood of s in X. Let f be the mapping of Xs×I into Xs defined, for u,v∈I and a∈A, by the relations
f(s,v)=s
((1−v)u,a) if 0⩽u<1 and o(a)=s,
f((u,a),v)=
(1−(1−v)(1−u),a) if 0<u⩽1 and t(a)=s.
It is continuous and compatible with the equivalence relation ∼. It therefore defines by passing to the quotient a mapping φs:Us×I→Us which is continuous, since I is locally compact (I, p. 19, Prop. 10). It is a strong contraction of Us onto p(s) (III, p. 237, Def. 6).
Let moreover x=(τ,a)∈]0,1[×A. Let us denote by Xx=]0,1[×{a,a} and let Ux be its image in ∣G∣ by p; it is a neighbourhood of p(x) in ∣G∣, homeomorphic to ]0,1[. The mapping of Xx×I into Xx given by ((u,a),v)↦((1−v)u+vτ,a) and ((u,a),t)↦((1−v)u+v(1−τ),a) is continuous. It defines by passing to the quotient a mapping φx:Ux×I→Ux which is continuous (loc. cit.) and is a strong contraction at x.
Every point of ∣G∣ is the image of a point s∈S or of a point of I×A of the form (τ,a) where 0<τ<1. It follows that every point of ∣G∣ has a neighbourhood contractible at that point. In other words, ∣G∣ is locally contractible and, in particular, unravelable (IV, p. 346, Prop. 5).
Let us choose a total order on S and consider the open covering (Us)s∈S of ∣G∣. Let s and s′ be distinct elements of S. The intersection Us∩Us′ is the union of the p(]0,1[,a) where a runs through the set of oriented edges of G whose extremities are s and s′. Consequently, the framework of the covering (Us)s∈S is identified with the quiver G. Moreover, for every s∈S, Us is contractible at s, hence π1(Us,p(s)) is reduced to the identity element. It then follows from Cor. 2 of IV, p. 416 that for every s∈S,π1(∣G∣,p(s)) is a free group (Nielsen-Schreier theorem).
Under the hypotheses of Prop. 1, suppose in addition that the framework of the covering (Ai)i∈I is an oriented tree. The homomorphism N:i∗∈Iπ1(Ai,a(i))→π1(Y,a(i0)) deduced from M by restriction is then surjective; its kernel is the smallest normal subgroup of i∗∈Iπ1(Ai,a(i)) which contains the elements φj(v)ψj(v)−1, for
every j=(i1,i2,V)∈J and every v∈π1(V,b(j)).
If the arcwise connected components of the intersections Ai∩Ai′, for i=i′ are moreover simply connected, the homomorphism N is then an isomorphism.
Under the hypotheses of the corollary, the graph associated with the quiver Γ is a tree, whence T = Γ. It follows that the image of the group F(J) by the homomorphism M is reduced to the identity element.
Let
ρ:(i∗∈Iπ1(Ai,a(i)))∗F(J)→i∗∈Iπ1(Ai,a(i))
be the unique group homomorphism which induces the identity homomorphism on π1(Ai,a(i)) and whose kernel contains F(J). We have M=N∘ρ. Consequently, the homomorphism N is surjective and its kernel is the smallest normal subgroup of i∗∈Iπ1(Ai,a(i)) which
contains the images under ρ of the elements defined by relations (R1), (R2) and (R3) of Prop. 1. We have ρ(j)=1 for every j∈ Fl(Γ). Since the framework of the covering (Ai)i∈I is a tree, one has Ai1∩Ai2∩Ai3=∅ for every triple (i1,i2,i3) of distinct elements of I. It follows that the kernel of N is the smallest normal subgroup containing the elements φj(v)ψj(v)−1 for every j=(i1,i2,V)∈J and every v∈π1(V,b(j)).
The fundamental group of R2{0} is isomorphic to Z, and the class of the path t↦e2πit is a generator of it (IV, p. 347, corollary). More generally, let n be a natural number and let A={z1,...,zn} be a set of n points of R2. Let Y be the space R2− A; we shall prove that the fundamental group of Y is isomorphic to the free group Fn on n generators. For every i, put zi=(ui,vi). Replacing zi if necessary by f(zi), where f:R2→R2 is a homeomorphism of the form (u,v)↦(u+αv,v), one may suppose that the abscissae of the zi are pairwise distinct. There is likewise no loss in supposing that one has u1<⋯<un.
Put V1=]−∞,u2[×R,Vi=]ui−1,ui+1[×R for 2⩽i⩽n−1, Vn=]un−1,+∞[×R. For 1⩽i⩽n, the set Ui=Vi−{zi} is open in the plane and homeomorphic to R2−{0}. The family (Ui)1⩽i⩽n is an open covering of the space Y=R2− A. The intersection Ui∩Uj is empty for ∣i−j∣⩾2, homeomorphic to R2 for ∣i−j∣=1. By Corollary 3, the fundamental group of Y is isomorphic to the free group Fn.
Let a be a point of Y. For every integer i∈{1,...,n}, let ri be a real number strictly positive and strictly less than the distances from zi to the points zj, for j=i. Let us denote by vi the class of the loop t↦zi+rie2πit at the point zi+ri of Y. Let θi be the class of a path γi joining the point a to the point zi+ri in Y. If the paths γi are injective and if their images meet only at a, one can prove that the unique homomorphism of the free group F(t1,...,tn) into π1(Y,a) such that φ(ti)=θiviθi−1 for every i∈{1,...,n} is a group isomorphism.
An analogous reasoning makes it possible to prove that for every discrete closed subset A of the plane, the fundamental group of R2− A is isomorphic to F(A) (IV, p. 463, Exercise 1).
Under the hypotheses of Proposition 1, suppose that there exists a subset A of Y, connected by arcs and nonempty, such that the intersection Ai∩Ai′ is equal to A for every pair (i,i′) of distinct elements of I. Let a be a point of A. There exists a unique homomorphism φ of the sum of the family of groups (π1(Ai,a))i∈I amalgamated by π1(A,a) into π1(Y,a) which coincides with the homomorphism induced by the canonical injection of Ai into Y, for every i∈I. The homomorphism φ is an isomorphism.
In particular, if the group π1(A,a) is reduced to the identity element, the canonical homomorphism of the free product of the family of groups (π1(Ai,a))i∈I into π1(Y,a) is an isomorphism.
For i∈ I, let us denote by gi:π1(A,a)→π1(Ai,a) and fi:π1(Ai,a)→π1(Y,a) the canonical homomorphisms induced by the inclusions of A into Ai and of Ai into Y. Let us denote by ∗π1(Ai,a) the sum of the groups π1(Ai,a) amalgamated by π1(A,a). Let alsoAp be the unique homomorphism of i∗∈Iπ1(Ai,a) into ∗Aπ1(Ai,a) which induces the identity on π1(Ai,a) (A, I, p. 80, prop. 4). The homomorphism p is surjective and it follows from the definition of the monoid ∗π1(Ai,a) that its kernel is the smallest distinguished subgroup ofiA∗∈Iπ1(Ai,a) which contains the
elements gi(v)gi′(v)−1, for i,i′∈I and v∈π1(A,a).
The homomorphisms fi∘gi:π1(A,a)→π1(Y,a), for i∈I, are equal. It therefore follows from the universal property of amalgamated sums of monoids (A, I, p. 80, Prop. 4) that there exists a unique group homomorphism φ (resp. f) from ∗Aπ1(Ai,a) (resp. from i∗∈Iπ1(Ai,a)) into π1(Y,a) which induces the homomorphism fi on π1(Ai,a). We have f=φ∘p.
For i∈ I, let ui be the canonical injection of A into Ai and vi the canonical injection of Ai into Y. Let also w be the canonical injection of A into Y. For every i∈ I, one has vi∘ui = w, hence π1(vi,a)∘π1(ui,a)=π1(w,a). It then follows from the universal property of amalgamated sums of monoids (A, I, p. 80, Prop. 4) that there exists a unique group homomorphism φ from ∗iπ1(Ai,a) into π1(Y,a) which induces the homomorphism π1(vi,a) on π1(Ai,a). It remains to prove that φ is an isomorphism.
The set J is identified with the set of pairs (i,i′) of elements of I such that i<i′. The set K is identified with the set of triples (i1,i2,i3) of elements of I such that i1<i2<i3. Let us choose all the base points a(i),b(j) and c(k) equal to a and all the paths B(j), Cst(k) equal to the constant path with image a. Let us also fix a point i0∈I.
For every pair (i,i′) of distinct points of I, the framework Γ of the covering (Ai) has exactly one arrow with endpoints i and i′. The arrows of Γ one of whose endpoints is equal to i0 are the arrows of an oriented maximal tree T.
For every element k∈K, one has r3(k)=j12(k)j23(k)j13(k)−1; let R be the normal subgroup of F(J) generated by the elements j∈ Fl(T) and the elements r3(k),k∈ K. Let us prove that R = F(J). It is enough to show that every element j=(i1,i2) of J belongs to R. This is true, by hypothesis, if i1=i0 or i2=i0. Suppose i0<i1 and put k=(i0,i1,i2). This is an element of K such that j=j23(k)=j. Moreover, j12(k) and j13(k) belong to Fl(T). It follows that j belongs to R. The cases where i1<i0<i2 or i2<i0 are treated in an analogous manner.
It then follows from Prop. 1 (IV, p. 412) that the homomorphism f is surjective and that its kernel is the smallest distinguished subgroup containing the relators r2(j,v)=gi(v)gi′(v)−1, for j=(i,i′,A)∈J and v∈π1(A,a). In other words, Ker(f) = Ker(p). It follows that φ is an isomorphism, as was to be proved.
If π1(A,a) is reduced to the identity element, p is an isomorphism, whence the second assertion.
Let ((Xi,xi))i∈I be a family of pointed topological spaces. The bouquet of the family ((Xi,xi))i∈I is by definition, and is denoted by ⋁i∈I(Xi,xi), the quotient topological space of the sum space of the family (Xi)i∈I by the equivalence relation which identifies all the points (xi,i) with one another, for i∈I. Let X denote this topological space and x the common image of the xi. Suppose that, for every i∈ I, the point xi is closed in Xi and that the spaces Xi are simply connected. If I is finite, Corollary 4 implies that the canonical homomorphism i∗∈Iπ1(Xi,xi)→π1(X,x) is an isomorphism.
Remark 1 of IV, p. 429 and Exercise 3, IV, p. 463 give less restrictive conditions under which this homomorphism is an isomorphism. See nevertheless Exercise 4, IV, p. 464.
3. Particular case of a covering formed of two parts
Let X be an arcwise connected topological space, and let B and C be non-empty arcwise connected subsets of X. Assume further one of the two following hypotheses:
(i) The interiors of the sets B and C cover X;
(ii) The sets B and C are closed in X, their union is equal to X, the spaces X, B, C are simply connected, and the space B∩C is locally arcwise connected.
Under these hypotheses, the canonical mapping of the sum space of the family (B,C) onto the space X satisfies property (VK) (cf. IV, p. 409).
Put A=B∩C. Since the space X is connected, the set A is not empty. Let a be a point of A; let j0 denote the arcwise connected component of a in A. For every arcwise connected component j of A distinct from j0, choose a point aj of j, the class βj of a path in B and the class γj of a path in C, both joining aj to a; let φj:π1(A,aj)→π1(B,a) and ψj:π1(A,aj)→π1(C,a) denote the group homomorphisms defined by
φj(v)=βj−1vβj and ψj(v)=γj−1vγj
for v∈π1(A,aj). Let φ0 and ψ0 also denote the homomorphisms of π1(A,a) into π1(B,a) and π1(C,a) respectively induced by the canonical injections. Let ιB and ιC denote the homomorphisms of π1(B,a) and π1(C,a) respectively into π1(X,a) induced by the canonical injections. Finally, let μ be the unique homomorphism of the free group F(π0(A)−{j0}) into π1(X,a) such that μ(j)=βj−1γj for every j∈π0(A)−{j0}.
which coincides with ιB in π1(B,a), ιC in π1(C,a) and μ in F(π0(A){j0}). This homomorphism is surjective and its kernel is the smallest normal subgroup containing the elements
φj(v)jψj(v)−1j−1
for j∈π0(A)−{j0} and v∈π1(A,aj), and the elements
φ0(v)ψ0(v)−1
for v∈π1(A,a).
The framework Γ of the covering of X defined by the family (B,C) has two vertices b and c corresponding to the two sets B and C. The set of its arrows is equal to π0(A); they join the point b to the point c. The graph associated with the subquiver of Γ whose only arrow is j0 is a maximal tree of Γ. The proposition then results from IV, p. 412, prop. 1.
For n⩾1, the sphere Sn is the union of two closed hemispheres, homeomorphic to the closed ball Bn−1 (TG, VI, p. 12), and whose intersection is identified with the sphere Sn−1. For n⩾2, the sphere Sn−1 is arcwise connected; it follows that the Poincaré group of Sn is trivial (cf. I, p. 127, example 3).
The sphere S0 has two arcwise connected components; thus one recovers that the Poincaré group of the circle S1 is isomorphic to a free group on one generator. More precisely, let B and C be the intersections of S1 with the half-planes of equations y⩾0 and y⩽0 in the plane R2. Put a=(1,0),a′=(−1,0); one has B∩C={a,a′}; its connected components are j0={a} and j={a′}. Let β be the class of the path t↦eπit in C; if one identifies C with R2, it joins a to a′ in B. Analogously, let γ be the class of the path t↦e−πit joining a to a′ in C. The path βγ−1 is a loop at a, given by t↦e2πit. By proposition 2, its class generates the group π1(S1,a).
The homomorphism μ is injective. More precisely, there exists a retraction associated with μ which is a group homomorphism.
Let ρ be the unique homomorphism of π1(B,a)∗π1(C,a)∗F(π0(A){j0}) into F(π0(A) - {j0}) which induces the trivial homomorphism on π1(B,a) and π1(C,a) and the identity on F(π0(A)−{j0}). Let N be the kernel of the homomorphism M. By proposition 2,ρ(N) is reduced to the identity element. There therefore exists a unique homomorphism r of π1(X,a) into F(π0(A)−{j0}) such that ρ=r∘M. For every v∈F(π0(A)−{j0}), one has M(v)=μ(v), hence r∘μ is the identity homomorphism. The corollary follows.
If the groups π1(X,a) and π1(A,a) are trivial, the same is true of the groups π1(B,a) and π1(C,a).
By corollary 2, the set A is arcwise connected. The group π1(X,a) is therefore isomorphic to the free product of the groups π1(B,a) and π1(C,a). In particular it contains subgroups isomorphic to the groups π1(B,a) and π1(C,a) (A, I, p. 83). These two groups are therefore trivial if π1(X,a) is.
4. Quotient Spaces
Let X be an arcwise connected topological space endowed with a proper (right) operation of a discrete group G. Put Y=X/G and denote by f:X→Y the canonical mapping. If g∈G and c:I→X is a path in X, let g∗c denote the path t↦c(t)⋅g and g∗[c] its strict homotopy class.
Let o be a point of X. For each g∈G, let βg be the class of a path joining o⋅g to o in X. For each g∈G, let Xg be the set of points x∈X such that x⋅g=x; for each arcwise connected component j of Xg, let aj be a point of j and let γj be the class of a path in X joining aj to o. Let ν:F(G)→π1(Y,f(o)) be the unique group homomorphism such that ν(g)=f∗(βg) for g∈G. Let N:π1(X,o)∗F(G)→π1(Y,f(o)) be the unique group homomorphism which coincides with π1(f,o) on π1(X,o) and with ν on F(G).
Suppose that X is unlaceable. The homomorphism N is then surjective, and its kernel is the smallest distinguished subgroup of π1(X,o)∗F(G) containing the elements
Suppose that X is unlaceable and that the group G is generated by the stabilizers of the points of X. The canonical morphism π1(f,o):π1(X,o)→π1(Y,f(o)) is then surjective. In particular, if X is simply connected by arcs, the same is true of Y.
If X is unlaceable, so is Y (IV, p. 349, prop. 8). Since a connected unlaceable space is simply connected by arcs if and only if it is simply connected (IV, p. 344, corollary 1 of Theorem 1), one thus recovers prop. 11 of I, p. 137.
Let X be an arcwise connected, unlaceable, separated topological space, and let a be a point of X. Let n be an integer ⩾2 and let Y be the quotient of the space Xn by the action of the group Sn operating by permutation of the factors; let f:Xn→Y denote the canonical mapping; let g:X→Y denote the mapping x↦f(x,a,...,a). It follows from the proposition that, for every i, the homomorphism π1(g,a) of π1(X,a) into π1(Y,g(a)) is surjective and that its kernel is the derived group of π1(X,a). In particular, the group π1(Y,g(a)) is abelian.
Suppose that X is unlaceable and that the group G operates freely in X. There exists a unique group homomorphism p:π1(Y,f(o))→G whose kernel contains the image of π1(X,o) and such that p(N(g))=g for every g∈G. Moreover, π1(X,o)→π1(Y,f(o))−→pG is an extension of G by π1(X,o).
Suppose that X is simply connected by arcs. The mapping of G into π1(Y,f(o)) which, to g∈ G, associates the path class f∗(βg) is a surjective group homomorphism; its kernel is the subgroup of G generated by the stabilizers of the points of X.
5. Cones; contraction of a subspace
Let X and Y be non-empty topological spaces, and let f:X→Y be a continuous mapping. Let Cone(f) be the cone of the mapping f, and let s be its vertex. Let αf′:X×I→ Cone(f) and βf′:Y→ Cone(f) denote the canonical mappings. The restriction of αf′ to the subspace X×{0} of X×I is the constant mapping with image {s}. The mapping βf′ induces a homeomorphism of Y onto the base of the cone Cone(f), by which we shall identify these two spaces. Also denote by
σf′:(Cone(f)−{s})×I→ Cone(f)−{s}
the canonical contraction, and by ρf′: Cone(f)−{s}→Y the canonical retraction of the cone deprived of its vertex onto its base.
Let J=π0(X); for each element j of J, let Xj denote the component j of X, let bj be a point of Xj, and let γj denote the class of the path t↦αf′(bj,t) in Côn(f) joining s to f(bj).
Let I be the image of the mapping π0(f); this is the set of arcwise connected components of Y which meet f(X); let φ:J→I denote the mapping deduced from f by passing to arcwise connected components. For each element i∈I, let Yi denote the component i of Y and choose a point ai in Yi.
For each element j of J, choose a path Bj joining the point f(bj) to the point aφ(j) in Yφ(j), and let βj denote its class. Let ψj denote the homomorphism from π1(X,bj) into π1(Y,aφ(j)) defined by
ψj(v)=βj−1f∗(v)βj
for v∈π1(X,aj).
Let σ:I→J be a section of the mapping φ. Put T=σ(I) and τ=σ∘φ; observe that φ∘τ=φ.
Suppose that the arcwise connected components of Y are open. For each i∈ I, let Gi be the quotient of the group π1(Yi,bi) by the smallest normal subgroup containing the image of the homomorphisms ψj, for j∈φ−1(i) ; let pi denote the canonical surjection of π1(Yi,bi) onto Gi.
There exists a unique group homomorphism
P:i∗∈IGi∗F(J−T)→π1(Côn(f),s)
such that
P(pi(v))= Int(γσ(i)βσ(i))(v) for i∈I and v∈π1(Yi,bi),
P(j)=γjβjβτ(j)−1γτ(j)−1 for j∈J−T.
The homomorphism P is an isomorphism.
Let Y′ denote the union of the arcwise connected components of Y which meet f(X), and let f′:X→Y′ be the mapping given by x↦f(x). The set Y′ is an open subset of Y; its arcwise connected components are open. The cone Cone(f′) is identified with the arcwise connected component of s in Cone(f). This makes it possible to suppose that Y=Y′, in other words that the mapping π0(f) is surjective and I=π0(Y).
For each j∈J, let us put Vj=αf′(Xj×]0,1[). On passing to subspaces, the mapping αf′ induces a homeomorphism of X×]0,1[ onto the complement of Y∪{s} in Cone(f). Hence the sets Vj are the arcwise connected components of Cone(f)−(Y∪{s}).
For each i∈I, let us put Ui=(ρf′)−1(Yi); this is an open subset of Cone(f), since Yi is open in Y by hypothesis. For each j∈φ−1(i), one has f(Xj)⊂Yi and
Vj∪Yi=αf′(Xj×]0,1])∪Yi
so that Vj∪Yi is an arcwise connected subset of Cone(f) containing Yi. Since Ui is the union of Yi and of the sets Vj, for j∈φ−1(i), it follows that Ui is arcwise connected.
Finally, the set C′(X)= Cone(f) - Y is an open subset of Cone(f); it is contractible to s, hence arcwise connected.
The set C′(X) and the sets Ui, for i∈I, constitute an open arcwise connected covering of Cone(f), to which we shall apply prop. 1 of IV, p. 412. Let I′ be the set obtained by adjoining s to I; let it be provided with a total ordering for which s is its smallest element.
For distinct elements i,i′ of I, one has Ui∩Ui′=∅. For i∈I, C′(X)∩Ui is the union of the sets Vj, for j∈φ−1(i); they are connected and pairwise disjoint. The intersection of any three distinct sets of this covering is empty.
The framework Γ of the covering considered has the set I′ for its vertices. Its edges are the triplets (s,i,Vj), for j∈J and i=φ(j); thus the set of edges of Γ will be identified with the set J.
For i∈I, one chooses as base point a(i)=ai∈Ui; one also puts a(s)=s∈C′(X).
For j∈ J, put b(j)=αf′(bj,21). Let B1(j) be the path in C′(X) with origin b(j) and endpoint a(s) given by t↦αf′(bj,(1−t)/2). Let B2(j) be the path in Uφ(j) with origin b(j) and endpoint a(φ(j))=aφ(j), obtained by juxtaposing the path t↦αf′(bj,(1+t)/2) and the path Bj. Then the class of the path B(j)=B1(j)∗B2(j) is equal to γjβj.
We choose i0=s.
We take for maximal oriented tree T the unique oriented tree of Γ whose set of arrows is σ(I). We have δ(s)=e, whereas for i∈I,δ(i)=[B(σ(i))]=γσ(i)βσ(i).
Let j∈J and let i=φ(j).
The homomorphism φj of π1(Vj,b(j)) into π1(C′(X),s) is the trivial homomorphism, since C′(X) is contractible to s.
The mapping αf′ induces a homeomorphism of Xj×]0,1[ onto Vj; this homeomorphism induces an isomorphism of the group π1(Vj,b(j)) onto the group π1(Xj,bj)=π1(X,bj). By passing to subspaces, the mapping σf′ induces a strong contraction of Ui onto Yi, which induces an isomorphism of the group π1(Ui,a(i)) onto the group π1(Y,ai). By means of these isomorphisms, the homomorphism
ψj:π1(Vj,b(j))→π1(Uφ(j),a(φ(j)))
is identified with the homomorphism Int(βφ(j)−1)∘f∗ of π1(X,bj) into π1(Y,ai).
Since C′(X) is contractible to s, the homomorphism μs is the trivial homomorphism.
Let i∈I. The homomorphism μi of π1(Ui,a(i)) into π1(Côn(f),s) is identified with the homomorphism of π1(Y,ai) into π1(Côn(f),s) obtained by composing the homomorphism Int(δ(i)) and the homomorphism of π1(Ui,s) into π1(Côn(f),s) deduced from the canonical injection of Ui into Côn(f).
Finally, the homomorphism μ:F(J)→π1(Côn(f),s) is given by
μ(j)=γjβjβτ(j)−1γτ(j)−1
Let P′ be the unique group homomorphism
P′:i∗∈Iπ1(Yi,ai)∗F(J)→π1(Cone(f),s)
which coincides with the homomorphism Int(δ(i)−1) on π1(Yi,ai) and with the homomorphism μ on F(J). By Prop. 1 of IV, p. 412, the homomorphism P′ is surjective and its kernel is the smallest normal subgroup of ∗iπ1(Yi,ai)∗F(J) which contains the elements r1(j), for j∈T, and the elements r2(j,v), for j∈J and v∈π1(Vj,b(j)). (There are no elements r3(k), since the set K is empty.)
For j∈T, one has r1(j)=j. Let j∈J and let v∈π1(Vj,b(j)). In view of the identification of π1(Vj,b(j)) with π1(X,bj), one has r2(j,v)=jψj(v)j−1. Let p denote the canonical surjective homomorphism of ∗π1(Yi,ai)∗F(J) onto ∗Gi∗F(J - T). For j∈T, one has p(r1(j))=ie; for j∈J and v∈π1(Xi,bj), one has p(ψj(v))=e. Hence there exists a unique group homomorphism P of ∗Gi∗F(J−T) into π1(Cone(f),s) such that P′∘p=P; it is an isomorphism.i
Suppose moreover that the spaces X and Y are arcwise connected and let a be a point of X. The canonical mapping of π1(Y,f(a)) into π1(Cone(f),f(a)) is surjective, and its kernel is the smallest normal subgroup which contains the image of the homomorphism π1(f,a).
Suppose moreover that the arcwise connected components of Y are simply arcwise connected. The homomorphism μ:F(J−T)→π1(Cone(f),s) is then an isomorphism.
Let X be a topological space whose arcwise connected components are open. Let A be a closed subspace of X; let ι:A→ X denote the canonical injection, let X/A denote the space deduced from X by contracting A to a point o, and let p:X→X/A denote the canonical mapping. Suppose further that the pair (X,A) has the homotopy extension property. The canonical mapping ρ: Cone(ι)→X/A is then a homotopy equivalence (III, p. 255, Remark 1), and one deduces from Prop. 4 the calculation of the Poincaré group of X/A at its base point o. In particular, if the arcwise connected components of X are simply arcwise connected, the group π1(X/A,o) is a free group.
6. Blowing up and gluing back together
Let C be a topological space and let (Bℓ)ℓ∈L be a finite family of closed subsets of C, pairwise disjoint. Let B be a topological space and, for each ℓ∈L, let hℓ be a homeomorphism of B onto Bℓ. We denote by BL the union of the family (Bℓ)ℓ∈L. We shall suppose that B and L are non-empty. Let R be the equivalence relation on C defined as follows. The class of an element x of C−BL is the set {x}; if x is an element of Bℓ, where ℓ∈L, the class of x is
the set of elements hk(hℓ−1(x)), where k runs through L. Let A be the quotient topological space C/R and f:C→A the canonical surjection. One says that the space A is obtained from the space C by identification of the sets Bℓ by means of the homeomorphisms hℓ.
The mapping f∘hℓ of B into A is independent of the element ℓ of L; it is closed and injective; it therefore induces a homeomorphism of B onto a closed subset of A. Thus we identify B with f∘hℓ(B) by means of the homeomorphism f∘hℓ; the mapping f induces a homeomorphism of C−BL onto A-B.
Suppose moreover that there exists a family (Nℓ)ℓ∈L of open subsets of C, pairwise disjoint, such that Nℓ contains Bℓ for each ℓ∈L. The union U of the family (f(Nℓ))ℓ∈L is open in A and contains B. The set U-B is the union of the pairwise disjoint open sets f(Nℓ−Bℓ), for ℓ∈L.
The mapping f:C→A is proper and separated; its fibres are finite.
Let us prove that f is closed. Let X be a closed subset of C. Let us prove that its image is closed in A. For each ℓ∈L, X∩Bℓ is closed in Bℓ, hence the space Y=⋃ℓ∈Lhℓ−1(X∩Bℓ) is closed in B since L is finite. The saturation X∗ of X for the equivalence relation R is then equal to X∪⋃ℓ∈Lhk(Y), hence is closed in C. Consequently, f(X)=f(X∗) is closed in A, which was to be proved.
The fibres of f are the equivalence classes of the relation R; they are finite. Consequently, the mapping f is proper (TG, I, p. 75, Theorem 1).
Lastly, let us show that f is separated. Let x and y be distinct points of C which have the same image under f. There therefore exists a point b∈B and distinct elements ℓ and m∈L such that x=hℓ(b) and y=hm(b). Consequently, Nℓ and Nm are disjoint neighbourhoods of x and y in C, whence the required assertion by virtue of Prop. 1 of I, p. 25.
Assume in addition the following hypotheses:
– the space A is simply connected and connected;
– the space C is simply connected;
– the space B is connected and locally arcwise connected.
Let us put I=π0(C); if i is an element of I, let Ci denote the connected component i of C. Let η:L→I denote the mapping which associates to an element ℓ∈L the connected component of C containing Bℓ. The mapping η is surjective. For, arguing by contradiction, consider a connected component X of C which meets no set Bℓ. It is an open and closed subset of C, saturated with respect to the equivalence relation R. Hence its image f(X) is an open and closed subset of A, disjoint from B. Since A is assumed connected and B nonempty, f(X) is empty, whence a contradiction.
Let σ:I→L be a section of the mapping η; put τ=σ∘η and T=σ(I).
Choose a point b of B. For every ℓ∈L, put bℓ=hℓ(b) and choose the class βℓ of a path joining bℓ to bτ(ℓ) in C; if ℓ=τ(ℓ), choose for βℓ the class of the constant path whose image is bℓ. For every ℓ∈L, let ϑℓ denote the homomorphism from π1(B,b) into π1(C,bτ(ℓ)) defined by
The homomorphism Q is surjective; its kernel is the smallest normal subgroup containing the elements
ϑℓ0(v)ϑℓ(v)−1 for v∈π1(B,b) and ℓ∈T,
ϑℓ0(v)ℓϑℓ(v)−1ℓ−1 for v∈π1(B,b) and ℓ∈L−T.
The mapping f:C→ A is proper, separated, with finite fibres (IV, p. 430, lemma 2); the spaces A and C are unlaceable. Moreover, C×AC is the union of the diagonal ΔC and of the disjoint subsets Bℓ×ABk=(hℓ,hk)(B), for (ℓ,k)∈L2 with ℓ=k. For such a pair (ℓ,k), we have Bℓ×ABk=Nℓ×ANk; hence this subset is open and closed in C×AC. Thus the diagonal ΔC is open, and its complement in C×AC, a finite union of disjoint subsets homeomorphic to B, is locally connected. This proves that the mapping f satisfies property (VK) (case (ii) of IV, p. 405). With a view to applying theorem 1 of IV, p. 408, we are going to define van Kampen data for f.
For each i∈I, let us choose as base point in Ci the point a(i)=bσ(i).
Let J be the set of path-connected components of C×AC. The sets ΔCi, for i∈I, are the connected components of the diagonal ΔC, which is open and closed in C×AC, therefore these sets belong to J. Analogously, the sets [ℓ1,ℓ2]=Bℓ1×ABℓ2 for ℓ1 and ℓ2 in L with ℓ1=ℓ2, belong to J. Since these sets form a partition of C×AC, they describe the set J.
Let j be an element of J of the form ΔCi, for i∈I. Let us choose as base point b(j) the point (a(i),a(i))=(bσ(i),bσ(i)) and let us take for path classes β1(j) and β2(j) the class of the constant path at bσ(i). Let j be an element of J of the form [ℓ1,ℓ2], where ℓ1 and ℓ2 are distinct elements of L. Let us then put b([ℓ1,ℓ2])=(bℓ1,bℓ2),β1([ℓ1,ℓ2])=βℓ1 and β2([ℓ1,ℓ2])=βℓ2.
Let K=π0(C×AC×AC).
Let ΔC′ denote the diagonal of the space C×AC×AC; this is a closed subset of C×AC×AC, because f is separated, and it is homeomorphic to C. For every i∈ I, let ΔCi′ also denote the image of Ci under the diagonal mapping of C into C×AC×AC; these are open and closed subsets of ΔC′. For every triplet (ℓ1,ℓ2,ℓ3) of elements of L, put also [ℓ1,ℓ2,ℓ3]=Bℓ1×ABℓ2×ABℓ3; these are closed subsets of C×AC×AC, homeomorphic to B. If the set {ℓ1,ℓ2,ℓ3} has at least two elements, one has [ℓ1,ℓ2,ℓ3]=Nℓ1×ANℓ2×ANℓ3, which implies that [ℓ1,ℓ2,ℓ3] is also open in C×AC×AC. Moreover, C×AC×AC is the union of the pairwise disjoint subsets ΔC′ and [ℓ1,ℓ2,ℓ3], where (ℓ1,ℓ2,ℓ3) runs through the set of triplets of elements of L which are not all equal to one and the same element. Thus, ΔC′, then the ΔCi′, for i∈I, are open and closed in C×AC×AC. This implies that the set K of connected components of this space is the union of the following disjoint sets K0 and K1.
The set K0 is the set of components of the form ΔCi′. Let i∈I and let k=ΔCi′. Put c(k)=(bσ(i),bσ(i),bσ(i)). For (s,t)∈{(1,2),(2,3),(1,3)}, choose for γst(k) the class of the constant path at (bσ(i),bσ(i)).
The set K1 consists of the components of the form k = [ℓ1,ℓ2,ℓ3], where ℓ1,ℓ2,ℓ3 are three elements of B such that the set {ℓ1,ℓ2,ℓ3} is of cardinal ⩾2. Put c(k)=(bℓ1,bℓ2,bℓ3). Let (s,t)∈{(1,2),(1,3),(2,3)}. If ℓs=ℓt, take for γst(k) the class (βℓs,βℓt); if ℓs=ℓt, take for γst(k) the class of the constant path at (bℓs,bℓt).
It is then verified that, for every k∈K and for every s∈{1,2,3}, the class of loops λs(k) defined by relation (2) of IV, p. 407, is trivial.
The framework Γ of the pair of groupoid morphisms (ϖ(p1),ϖ(p2)) from ϖ(C×AC) into ϖ(C) has as vertices the set I and as oriented edges the set J. If i∈I, the arrow j=ΔCi has origin and end i; if (ℓ1,ℓ2) is a pair of distinct elements of L, the arrow j=[ℓ1,ℓ2] has origin η(ℓ1) and end η(ℓ2).
Let T be the subquiver of Γ whose set of vertices is I and whose arrows are those of the form [ℓ0,ℓ], for ℓ∈T−{ℓ0}. The associated graph of T is a maximal tree of Γ.
Put i0=η(ℓ0).
If i∈I−{i0}, the unique path in T joining i0 to i is (i0,[ℓ0,σ(i)],i). The groupoid morphism from Grp(Γ) into ϖ(Y) defined on p. 407 (and denoted by τ loc. cit.) maps j to the identity element if j=ΔCi, for i∈I, and maps j=[ℓ1,ℓ2] to f∗(βℓ1)−1f∗(βℓ2), if ℓ1 and ℓ2 are distinct elements of L. For every i∈I, the path class δi defined loc. cit. and joining b=f(bσ(i0)) to b=f(bσ(i)) is given by
δi=f∗(βℓ0)−1f∗(βσ(i))=e
because βℓ=e if ℓ∈T.
Thus a van Kampen datum of the mapping f has been defined. Consider then the unique group homomorphism
Q′:(i∗∈Iπ1(Ci,bσ(i)))∗F(J)→π1(A,b)
which coincides with π1(f,bσ(i)) on π1(C,bσ(i)) and such that
Q′(j)=f∗(β1(j))−1f∗(β2(j))
for j∈J. By IV, p. 408, Theorem 1, this homomorphism is surjective and its kernel is the smallest normal subgroup containing the relators r1(j) (for j∈ Fl(T)), r2(j,v) (for j∈ J and v∈π1(C×AC,b(j))) and r3(k) (for k∈K) defined, loc. cit., by equations (R1),(R2) and (R3).
Let q′ be the unique homomorphism of F(L) into F(L - T) such that q′(ℓ)=ℓ if ℓ∈L−T and q′(ℓ)=e otherwise. Let
be the unique group homomorphism which coincides with the identity on π1(Ci,bσ(i)), for i∈ I and such that one has q(j)=e if j=ΔCi, q([ℓ,ℓ′])=q′(ℓ)−1q′(ℓ′) if ℓ and ℓ′ are distinct elements of L. The homomorphism q is surjective.
If i∈I and j=ΔCi, one has Q′(j)=eb=Q′(q(j)). If j=[ℓ,ℓ′], for ℓ and ℓ′ in L, distinct, one has
Q′(j)=f∗(βℓ)−1f∗(βℓ′)=Q(q′(ℓ))−1Q(q′(ℓ′))
=Q(q′(ℓ)−1q′(ℓ′))=Q∘q([ℓ,ℓ′])
Therefore, Q′ = Q∘q. It follows that the homomorphism Q is surjective and that its kernel is the smallest normal subgroup of (i∗∈Iπ1(Ci,bσ(i)))∗F(L−T) containing the images under q of the relators r1(j) (for j∈ Fl(T)),r2(j,v) (for j∈J and v∈π1(C×AC,b(j))) and r3(k) (for k∈K).
If ℓ∈T−{ℓ0} and j=[ℓ0,ℓ], one has r1(j)=j and q(r1(j))=e.
Let k∈K. One has r3(k)=j12(k)j23(k)j13(k)−1. If k=ΔCi′, for i∈I, let us put j=ΔCi; then one has
q(r3(k))=q(jjj−1)=q(j)=e
Let ℓ and ℓ′ be distinct elements of L. If k=[ℓ,ℓ,ℓ′], one has therefore
q(r3(k))=q(ΔCη(ℓ)[ℓ,ℓ′][ℓ,ℓ′]−1)=q(ΔCη(ℓ))=e
If k=[ℓ,ℓ′,ℓ], one gets
q(r3(k))=q([ℓ,ℓ′][ℓ′,ℓ]ΔCη−1(ℓ))
=q′(ℓ)−1q′(ℓ′)q′(ℓ′)−1q′(ℓ)q(ΔCη(ℓ))−1=e
Moreover, if k=[ℓ′,ℓ,ℓ], one has
q(r3(k))=q([ℓ′,ℓ]ΔCη(ℓ)[ℓ′,ℓ]−1)
=q(ℓ′)−1q(ℓ)q(ΔCη(ℓ))q(ℓ)−1q(ℓ′)=e
Finally, if k=[ℓ1,ℓ2,ℓ3], where ℓ1,ℓ2,ℓ3 are elements of L, pairwise distinct, one has
Let i∈I and put j=ΔCi. The group homomorphisms φj and ψj, from π1(C×AC,b(j))=π1(Ci,a(i)) into π1(C,a(i)) defined by relations (1) of IV, p. 407, are equal respectively to (p1)∗ and to (p2)∗. We have then, for v∈π1(Ci,a(i)),r2(j,v)=vjv−1j−1, whence q(r2(j,v))=e.
Finally, let j=[ℓ,ℓ′], where ℓ and ℓ′ are distinct elements of L. The group homomorphism φj:π1(Bℓ×ABℓ′,b(j))→π1(Cη(ℓ),bτ(ℓ)) defined loc. cit. is the homomorphism ϑℓ, and the homomorphism ψj is the homomorphism ϑℓ′. We have then, for v∈π1(Bℓ×ABℓ′,b(j))
These relations prove, on the one hand, that the elements announced in the proposition belong to the kernel of the homomorphism Q, and, on the other hand, that these elements q(r2(j,v)) all belong to the smallest normal subgroup containing the elements announced in the statement. The proposition follows.
Let A be a topological space, let B be a closed subset of A, and let U be an open neighbourhood of B in A. Suppose that the set U-B is the union of a finite family (Mℓ)ℓ∈L of pairwise disjoint open sets. For each ℓ∈L, put Nℓ′=Mℓ∪B. Let C′ denote the topological sum of the spaces A-B and Nℓ′, for ℓ∈L; let also φ:A−B→C′ and φℓ:Nℓ′→C′, for ℓ∈L, denote the canonical injections. Let C be the quotient topological space of C′ by the finest equivalence relation S for which the points φ(x) and φℓ(x) are equivalent, for each ℓ∈L and each x∈Mℓ; let ρ:C′→C be the canonical mapping.
Let us prove that the space A is recovered from the space C by identification of the sets Bℓ by means of the homeomorphisms ρ∘(φℓ∣B):B→Bℓ.
For every ℓ∈L, the set Mℓ is open in A-B and in Nℓ′. The mappings ρ∘φ and ρ∘φℓ are homeomorphisms of the spaces A-B and Nℓ′, for ℓ∈L, onto open subsets of C (TG, I, p. 17, Prop. 9). For every ℓ∈L, the set φℓ(B) is closed in C′ and saturated for the equivalence relation S. Hence the set Bℓ=ρ(φℓ(B)) is closed in C and the mapping ρ∘φℓ induces a homeomorphism of B onto Bℓ(loc. cit.). Analogously, for every ℓ∈L, the set Nℓ=ρ(φℓ(Nℓ′)) is an open neighbourhood of Bℓ; the sets Nℓ are mutually disjoint.
By passing to the quotient, the mapping f′:C′→A deduced from the canonical injections into A of the spaces A-B and Nℓ, for ℓ∈L, induces a
continuous mapping f:C→A. If x is a point of B, the fibre f−1(x) is the set of points ρ(φℓ(x)), for ℓ∈L. The equivalence relation on C associated with the mapping f is the relation R defined at the beginning of the number. Let BL denote the union of the family (Bℓ)ℓ∈L. By construction, the mapping f induces a homeomorphism of C−BL onto A-B and, for ℓ∈L, a homeomorphism of Nℓ′ onto Nℓ. We shall prove that the mapping f is closed; the topology of A will therefore be the quotient topology of C by the equivalence relation R (I, p. 18, Example 2). For this, let us prove that if F is a subset of A such that F∩(A−B) is closed in A-B and such that F∩Nℓ′ is closed in Nℓ′, for ℓ∈L, the set F is closed in A. The set U, union of the family (Nℓ′)ℓ∈L, is open in A and the sets Nℓ′, for ℓ∈L, constitute a finite closed covering of it. Hence the set F∩U is closed in U (TG, I, p. 18, Prop. 3). The sets U and A-B form an open covering of A, therefore the set F is closed in A (loc. cit.).