Bourbaki

§ 3. Groupoids

1. Categories

Definition 1 01SZ

Let C be a quiver. Let J denote the set of pairs (f,g)(f, g) of arrows of C such that t(f)=o(g)t(f) =o(g). A law of composition in C is a mapping m:J→m: J\rightarrow Fl(C) such that, for every pair (f,g)∈J(f, g)\in J, the origin of the arrow m(f,g)m(f, g) is that of ff and its end that of gg.

Let C be a quiver endowed with a law of composition mm. Two arrows ff and gg such that t(f)=o(g)t(f) =o(g) are said to be composable; the arrow m(f,g)m(f, g) is called their composite, or their product, and is often denoted by f⋅gf\cdot g, or even by fgf g.

For every vertex aa of C, the set Ca,aC_{a,a} = Fla,a(C)_{a,a}(C), endowed with the mapping deduced from mm by passing to subsets, is a magma (A, I, p. 1, Def. 1).

One says that a family (fi)i∈I(f_i)_{i\in I} of arrows of C, indexed by a finite nonempty totally ordered set I, is composable if, for every pair (i,j)(i, j) of consecutive elements of I, the arrows fif_i and fjf_j are composable. The product ∏i∈Ifi\prod_{i\in I}f_i of such a family is defined by induction on the cardinal of I in the following way (cf. A, I, p. 3):

(i) if I has a single element ω,∏i∈Ifi=fω\omega ,\prod_{i\in I}f_i=f_{\omega};

(ii) if I has at least two elements and if ω\omega is the smallest element of I, one sets ∏i∈Ifi=fω⋅∏i∈I−{ω}fi\prod_{i\in I}f_i=f_{\omega}\cdot \prod_{i\in I-\{\omega\}}f_i.

The law of composition mm is said to be associative if one has f⋅(g⋅h)=f\cdot (g\cdot h) = (f⋅g)⋅h(f\cdot g)\cdot h for every composable triple (f,g,h)(f, g, h) of arrows of C. When the law mm is associative, the product of a sequence (f1,...,fn)(f_1, . . . , f_n) of nn composable arrows, where nn is an integer ⩾1\geqslant 1, is denoted by f1f2...fnf_1f_2. . . f_n.

Let aa be a vertex of C. One says that an arrow e∈Ca,ae\in C_{a,a} is an identity element of mm at aa if one has ef=fef=f for every arrow ff with origin aa and ge=gge=g for every arrow gg with end aa. There exists at most one identity element of mm at a:a: if ee and e′e' are two such elements, one has e′=ee′=ee'=ee'=e. When such an identity element exists, it is often denoted by eae_a; it is then an identity element of the magma Ca,aC_{a,a} (A, I, p. 12).

Definition 2 01T0

A category is a quiver endowed with an associative law of composition for which there exists at each vertex an identity element.

Example 1 01T1

Let C be a quiver. Let Ch(C) denote the set of paths in C and o:o: Ch(C) →\rightarrow Som(C),t:,t: Ch(C) →\rightarrow Som(C) the mappings which associate with a path its origin and its end. The quadruple ΩC=\Omega_C= (Som(C), Ch(C),o,t, o, t) is a quiver. Endow it with the law of composition defined by juxtaposition of paths. This law of composition is associative; for every vertex aa of C, the constant loop ea=(a)e_a= (a) of origin aa is an identity element at aa. Thus, ΩC\Omega_C is a category. It is said to be the category of paths of the quiver C.

Example 2 01T2

Let Σ be a Species of structures in a theory T\mathscr{T} stronger than set theory and let σ(x,y,s,u)\sigma (x, y, s, u) be a term of T\mathscr{T} satisfying conditions (MOI_I), (MOII_{II}) and (MOIII_{III}) of E, IV, p. 11. Let S be a set; suppose that every element of S is a pair (x,s)(x, s), where xx is a set and ss a structure of species Σ on xx. Let F be the set of mappings ff such that there exist (x,s)(x, s) and (y,t)(y, t) in S such that ff is a σ\sigma-morphism of xx, endowed with the structure ss, into yy, endowed with the structure tt; one then sets o(f)=(x,s)o(f) = (x, s) and t(f)=(y,t)t(f) = (y, t). Endowed with the law of composition given by m(f,g)=g∘fm(f, g) =g\circ f, the quiver (S,F,o,t)(S,F, o, t) is a category. In this context, the elements of S are rather called objects.

Let C be a category.

For every vertex aa of C, the set Ca,aC_{a,a}, endowed with the law of composition (f,g)↦fg(f, g)\mapsto f g, is a monoid with identity element eae_a.

A map ff of C is said to be invertible if there exists a map gg of C such that o(g)=t(f),t(g)=o(f)o(g) =t(f),t(g) =o(f) and such that fgf g and gfgf are identity elements at o(f)o(f) and t(f)t(f) respectively. Such a map gg is unique (if fg=eo(f)f g=e_{o(f)} and g′f=et(f)g'f=e_{t(f)}, one has g=et(f)g=(g′f)g=g=e_{t(f)}g= (g'f)g= g′(fg)=g′eo(f)=g′g'(f g) =g'e_{o(f)}=g') and is called the inverse of ff; the inverse of an invertible map ff is denoted by f−1f^{-1}.

2. Functors

Definition 3 01T3

Let C and C′C' be categories. A functor from C into C′C' is a morphism of quivers φ:C→C′\varphi : C\rightarrow C' such that φ(fg)=\varphi (f g) = φ(f)φ(g)\varphi (f)\varphi (g) for every pair (f,g)(f, g) of composable maps of C.

Let C and C′C' be categories and let φ:C→C′\varphi : C\rightarrow C' be a functor. Let ff be an arrow of C,a,a its origin and bb its end; then, φ(f)\varphi (f) is an arrow of C′C' with origin φ(a)\varphi (a) and end φ(b)\varphi (b).

Let C, C′,C′′C',C'' be categories and let φ:C→C′\varphi : C\rightarrow C' and φ′:C′→C′′\varphi ': C'\rightarrow C'' be functors. Then, φ′∘φ\varphi '\circ \varphi is a functor.

Let C be a category. Then the quiver morphism IdC_C is a functor.

Let φ:C→C′\varphi : C\rightarrow C' be a functor. In order that φ\varphi be an isomorphism, it is necessary and sufficient that the mappings Som(φ\varphi ) and Fl(φ\varphi ) be bijective.

Thus, if one calls a category structure on sets S and F the data of a quiver structure on these sets and of an associative composition law for which there exists at each vertex an identity element, one can take the functors as morphisms of the category structure (E, IV, p. 11).

3. Groupoids

Definition 4 01T4

A groupoid is a category each arrow of which is invertible.

Let G be a groupoid and let aa be a vertex of G. The monoid Ga,aG_{a,a} is a group, denoted by GaG_a and called the isotropy group of G at aa.

Let a,ba,b be vertices of G and let f∈Ga,bf\in G_{a,b} be an arrow joining aa to bb. The mapping g↦fgf−1g\mapsto f gf^{-1} is a group isomorphism of GbG_b onto GaG_a, denoted by Int(ff). If ff and f′f' are composable arrows of G, one has Int(ff′f f') == Int(ff)∘\circ Int(f′f') ; the inverse of the isomorphism Int(ff) is the isomorphism Int(f−1f^{-1}).

A morphism of groupoids φ:G→G′\varphi : G\rightarrow G' is a morphism of categories. Moreover, φ\varphi is an isomorphism of groupoids if and only if it is an isomorphism of categories.

Let φ:G→G′\varphi : G\rightarrow G' be a morphism of groupoids. For every vertex aa of G, the mapping f↦φ(f)f\mapsto \varphi (f) of GaG_a into (G′)φ(a)(G')_{\varphi(a)} is a group homomorphism denoted by φa\varphi_a. In particular, one has φ(ea)=eφ(a)\varphi (e_a) =e_{\varphi(a)}.

For every arrow ff of G,φ(f−1),\varphi (f^{-1}) is the inverse of φ(f)\varphi (f). If ff is an arrow of G joining a vertex aa to a vertex bb, one thus has, for every element g∈Gbg\in G_b, the relation Int(φ(f)\varphi (f))(φ(g))=φ(\varphi (g)) =\varphi(Int(ff)(g)(g)).

4. Orbits of a groupoid

Let G be a groupoid. The connected components of the underlying quiver of G are called the orbits of G. The set of the orbits of G is denoted by Orb(G); if φ:G→G′\varphi : G\rightarrow G' is a morphism of groupoids, the mapping π0(φ)\pi_0(\varphi ) deduced from φ\varphi by passing to the connected components of the associated graphs is in general denoted by Orb(φ\varphi ) and called the mapping deduced from φ\varphi by passing to the orbits.

A groupoid which has exactly one orbit is said to be transitive. If G is a transitive groupoid, the groups GaG_a, for a∈a\in Som(G), are isomorphic. One says that the groupoid G is simply transitive if it is transitive and if, moreover, the isotropy groups GaG_a are all reduced to their identity element.

Proposition 1 01T5

Let G be a groupoid. The relation "there exists an arrow of G joining aa to bb" is an equivalence relation in the set of vertices of G whose equivalence classes are the orbits of G.

Let a,b,ca,b,c be vertices of G. One has ea∈Ga,ae_a\in G_{a,a}; if f∈Ga,bf\in G_{a,b}, f−1∈Gb,af^{-1}\in G_{b,a}; if f∈Ga,bf\in G_{a,b} and g∈Gb,cg\in G_{b,c}, one has fg∈Ga,cf g\in G_{a,c}. This shows that the indicated relation is reflexive, symmetric and transitive; it is therefore an equivalence relation. The second assertion then results from the definition of the connected components of a quiver.

5. Examples of groupoids

1) Let G be a group. Let us denote by G\mathscr{G} the quiver whose set of vertices is reduced to one element and whose set of arrows is G. Endowed with the law of composition induced by that of G,G,\mathscr{G} is a groupoid, called the groupoid associated with the group G.

2) Let X be a set. Let G be the quiver (X,X×X(X,X\times X, pr1_1, pr2)_2); define a law of composition in G by putting (x,x′)⋅(x′,x′′)=(x,x′′)(x, x')\cdot (x', x'') = (x, x''), if x,x′,x′′x, x', x'' are elements of X. This law is associative; for every x∈Xx\in X, (x,x)(x, x) is the identity element at xx; the inverse of the arrow (x,x′)(x, x') is the arrow (x′,x)(x', x). The groupoid thus defined is called the groupoid of pairs of the set X. If X is not empty, it is simply transitive.

3) Let X and S be sets and let p:X→Sp: X\rightarrow S be a mapping. For a∈Sa\in S, let Xa=p−1(a)X_a=\overset{-1}{p}(a). For aa and bb in S, let Ga,bG_{a,b} be the set B(Xa;Xb)\mathscr{B}(X_a; X_b) of bijections from XaX_a onto XbX_b and let G be the sum of the sets Ga,bG_{a,b}. Let oo and tt be the mappings from G into S such that o(f)=ao(f) =a and t(f)=bt(f) =b for every element f∈Ga,bf\in G_{a,b}. The quadruple (S,G,o,t)(S,G, o, t) is a quiver. Endow it with the law of composition defined by m(f,g)=g∘fm(f, g) =g\circ f if f∈B(Xa;Xb)f\in \mathscr{B}(X_a; X_b) and g∈B(Xb;Xc)g\in \mathscr{B}(X_b; X_c), where a,b,ca, b, c are points of S. It is a groupoid; it is denoted by B(X,p)\mathscr{B}(X, p) and called the groupoid of permutations of X relative to pp.

4) Let (Gi)i∈I(G_i)_{i\in I} be a family of groupoids. Let us denote by G the product quiver of the family of underlying quivers; for each i∈Ii\in I, let pri:G→Gi_i: G\rightarrow G_i be the canonical quiver morphism. There exists a unique law of composition on G for which G is a groupoid and such that, for each i∈Ii\in I, pri_i is a groupoid morphism. Let f=(fi)i∈If= (f_i)_{i\in I} and g=(gi)i∈Ig= (g_i)_{i\in I} be arrows of G; they are composable if and only if the arrows fif_i and gig_i are composable, for i∈Ii\in I. One then has fg=(figi)i∈If g= (f_ig_i)_{i\in I}.

The quiver G, endowed with this law of composition, is called the product groupoid of the family (Gi)i∈I(G_i)_{i\in I}. It satisfies the following universal property: for every groupoid G′G' and every family (φi)i∈I(\varphi_i)_{i\in I}, where φi\varphi_i is a groupoid morphism of G′G' into GiG_i, there exists a unique groupoid morphism φ:G′→G\varphi : G'\rightarrow G such that φi=\varphi_i= pri∘φ_i\circ \varphi for each i∈Ii\in I.

5) Let ((Gi)i∈I,(φi,j)i≺j)((G_i)_{i\in I},(\varphi_{i,j})_{i\prec j}) be an inductive system of groupoids, indexed by a right filtering preordered set (I,≺)(I,\prec ), the φi,j\varphi_{i,j} being groupoid morphisms. Let G be the quiver whose set of vertices is lim⟶i\longrightarrow_i Som(Gi_i), whose set of arrows is lim⟶i\longrightarrow_i Fl(Gi_i), the source and target mappings being the mappings lim⟶ioGi\longrightarrow_io_{G_i} and lim⟶itGi\longrightarrow_it_{G_i} respectively. The laws of composition of the GiG_i induce a law of composition on G which makes it a groupoid (cf. A, I, p. 114). The canonical mappings Som(Gi_i)→\rightarrow Som(G) and Fl(Gi_i)→\rightarrow Fl(G) define a groupoid morphism φi\varphi_i of GiG_i into G. If i,ji,j are elements of I such that i≺ji\prec j, one has φj∘φi,j=φi\varphi_j\circ \varphi_{i,j}=\varphi_i. The groupoid G is called the inductive limit of the family of groupoids GiG_i and is denoted by lim⟶iGi\longrightarrow_iG_i. It satisfies the following universal property: for every groupoid H and every family (ψi)i∈I(\psi_i)_{i\in I}, where, for i∈I,ψi:Gi→Hi\in I,\psi_i: G_i\rightarrow H is a groupoid morphism, such that ψj∘φi,j=ψi\psi_j\circ \varphi_{i,j}=\psi_i for every pair (i,j)(i, j) of elements of I such that i≺ji\prec j, there exists a unique groupoid morphism ψ:G→H\psi : G\rightarrow H such that ψi=ψ∘φi\psi_i=\psi \circ \varphi_i.

For the existence of a groupoid satisfying this universal property when one does not suppose that the set I is right filtering, cf. II, p. 228, exerc. 3.

6. Subgroupoids

Definition 5 01T6

Let C be a category. A subcategory of C is a subquiver D of C satisfying the following conditions:

(i) For every vertex aa of D,ea,e_a is an arrow of D;

(ii) For every composable pair (f,g)(f, g) of arrows of C belonging to D, the product fgf g is an arrow of D;

Let C be a category and let D be a subcategory of D; endowed with the law of composition deduced from that of C by passing to subsets, D is a category.

Every full subquiver of C is a subcategory of C. The intersection of a family of subcategories of C is a subcategory of C.

Definition 6 01T7

Let G be a groupoid. A subgroupoid of G is a subcategory H of G such that the inverse of every arrow of G belonging to H is an arrow of H.

Let G be a groupoid. Every subgroupoid of G is a groupoid. Every full subquiver of G is a subgroupoid of G. The intersection of a family of subgroupoids of G is a subgroupoid of G.

Let H be a subquiver of G. The intersection of the subgroupoids of G of which H is a subquiver is called the subgroupoid of G generated by H. The set of its vertices is equal to that of H and its orbits are the connected components of H.

Example 1 01T8

Let X be a set; let us denote by X×XX\times X the groupoid of pairs of X (II, p. 163, Example 2). Let R be an equivalence relation on X. The subquiver of the groupoid X×XX\times X whose set of vertices is X and whose set of arrows is the graph of the equivalence relation R is a subgroupoid of X×XX\times X. Its orbits are the equivalence classes of the relation R.

Conversely, every subgroupoid of X×XX\times X whose set of vertices is X is of this form.

Let X and S be sets and let p:X→p: X\rightarrow S be a mapping. The mapping pp defines an equivalence relation in X (E, II, p. 41). One denotes by X×SXX\times_SX the subgroupoid of X×XX\times X defined by this equivalence relation, and one calls it the groupoid of pairs of (X,p)(X, p).

Example 2 01T9

Let G and G′G' be groupoids, let φ\varphi and ψ\psi be groupoid morphisms from G into G′G'. The equalizer of φ\varphi and ψ(\psi (cf. II, p. 153) is a subgroupoid of G.

Example 3 01TA

Let X be a set and Γ a group. Denote by X×Γ×XX\times \Gamma \times X the graph whose set of vertices is X, whose set of arrows is X×Γ×XX\times \Gamma \times X, the origin and end mappings being respectively pr1_1 and pr3_3. Endowed with the composition law (x,γ,x′)⋅(x′,γ′,x′′)=(x,γγ′,x′′)(x, \gamma , x')\cdot (x', \gamma ', x'') = (x, \gamma \gamma ', x''), it is a groupoid. It is moreover the groupoid deduced by transport of structure from the groupoid (X×X)×Γ(X\times X)\times \Gamma, product of the groupoid of pairs X×XX\times X and of the groupoid associated with the group Γ (II, p. 163, example 1), by means of the bijection (x,x′,γ)↦(x,γ,x′)(x, x', \gamma )\mapsto (x, \gamma , x') of X×X×ΓX\times X\times \Gamma onto X×Γ×XX\times \Gamma \times X.

Let m:X×Γ→Xm: X\times \Gamma \rightarrow X be a right operation of the group Γ on X. The graph F of the mapping mm, that is to say the set of triples (x,γ,x′)∈X×Γ×X(x, \gamma , x')\in X\times \Gamma \times X such that x′=xγx'=x\gamma, is the set of arrows of a unique subgroupoid G of X×Γ×XX\times \Gamma \times X. The orbits of this groupoid are the orbits of Γ in X; if x∈Xx\in X, the isotropy group of G at xx is the set of (x,γ,x)(x, \gamma , x), where γ\gamma runs through the stabilizer of xx in Γ.

Conversely, every subgroupoid G of the groupoid X×Γ×XX\times \Gamma \times X such that the mapping (pr1_1, pr2_2) of Fl(G) into X×ΓX\times \Gamma is a bijection is of this form.

Example 4 01TB

Let G be a groupoid, let X be a set and let φ:X→\varphi : X\rightarrow Som(G) be a mapping. Define a graph G′G' in the following way. The set Som(G′') is the set X; for every pair (x,y)(x, y) of elements of X, the set Flx,y(G′)_{x,y}(G') is the set of triples (x,f,y)∈X×(x, f, y)\in X\times Fl(G) ×X\times X, where ff is an element of Flφ(x),φ(y)(G)_{\varphi(x),\varphi(y)}(G). Let x,y,zx,y,z be elements of X,f∈,f\in Flφ(x),φ(y)(G)_{\varphi(x),\varphi(y)}(G) and g∈g\in Flφ(y),φ(z)(G)_{\varphi(y),\varphi(z)}(G) arrows of G, and put (x,f,y)⋅(y,g,z)=(x,fg,z)(x, f, y)\cdot (y, g, z) = (x, f g, z). This defines a composition law in the graph G′G' which makes it a groupoid. It is called the inverse image groupoid of the groupoid G by the mapping φ\varphi and is denoted by φ∗(G)\varphi^*(G).

The pair (φ,ψ)(\varphi , \psi ), where ψ:\psi : Fl(φ∗(G)\varphi^*(G))→\rightarrow Fl(G) is the mapping defined by (x,f,y)↦f(x, f, y)\mapsto f is a groupoid morphism from φ∗(G)\varphi^*(G) into G, called the canonical morphism.

Example 5 01TC

Let φ:G→G′\varphi : G\rightarrow G' be a morphism of groupoids. Let H be the subquiver of G with set of vertices Som(G) and whose set of arrows consists of the f∈f\in Fl(G) such that φ(f)\varphi (f) is an identity element of G′G'. For every a∈a\in Som(G),ea∈Ha,a,e_a\in H_{a,a}. For every f∈f\in Fl(H), φ(f−1)=φ(f)−1\varphi (f^{-1}) =\varphi (f)^{-1}, hence f−1∈f^{-1}\in Fl(H). Further, if ff and gg are composable arrows of H, one has φ(fg)=φ(f)φ(g)=et(φ(f))eo(φ(g))=\varphi (f g) =\varphi (f)\varphi (g) =e_{t(\varphi(f))}e_{o(\varphi(g))}= et(φ(f))e_{t(\varphi(f))}, hence fgf g is an arrow of H. This shows that H is a subgroupoid of G. It is called the kernel of φ\varphi and is denoted by Ker(φ\varphi ).

Example 6 01TD

Let φ:G→G′\varphi : G\rightarrow G' be a morphism of groupoids. The quiver φ(G)\varphi (G) whose set of vertices is φ\varphi(Som(G)) and whose set of arrows is φ\varphi(Fl(G)) is not in general a subgroupoid of G′G'. This is nevertheless the case if the mapping Som(φ\varphi ) is injective (II, p. 225, exerc. 5).

7. Operations of a groupoid

Let G be a groupoid; let S denote the set of its vertices. Let X be a set and let p:X→Sp: X\rightarrow S be a mapping. A (right) operation φ\varphi of the groupoid G on the set X, relative to pp, is a morphism of groupoids φ\varphi from G into the groupoid B(X,p)\mathscr{B}(X, p) of permutations of the set X relative to p(p(II, p. 163) which induces the identity on the set of vertices. In other words, φ\varphi consists, for each pair (a,b)(a, b) of points of S and for each arrow g∈Gg\in G joining aa to bb, of a mapping φ(g):Xa→Xb\varphi (g): X_a\rightarrow X_b, such that, for every pair (f,g)(f, g) of composable arrows of G,

φ(fg)=φ(g)∘φ(f) \varphi (f g) =\varphi (g)\circ \varphi (f)

and such that, for every a∈S,φ(ea)a\in S,\varphi (e_a) is the identity of XaX_a.

Let φ\varphi and φ′\varphi ' be operations of the groupoid G on a set X, relative to a mapping p:X→Sp: X\rightarrow S. In order that one have φ=φ′\varphi =\varphi ', it is enough that there exist a subquiver H of G, generating G, such that φ(f)=φ′(f)\varphi (f) =\varphi '(f) for every f∈f\in Fl(H). In fact, the set of arrows ff of G such that φ(f)=φ′(f)\varphi (f) =\varphi '(f) is the set of arrows of a subgroupoid of G, with set of vertices S.

Let G be a groupoid with set of vertices S, let X be a set, let p:X→Sp: X\rightarrow S be a mapping, and let φ\varphi be an operation of G on X relative to pp. Let GφG_{\varphi} be the subquiver of p∗(G)p^*(G) whose set of vertices is X and whose set of arrows is the set of triples (x,f,y)∈X×(x, f, y)\in X\times Fl(G) ×X\times X such that φ(f)(x)=y\varphi (f)(x) =y. It is a subgroupoid of p∗(G)p^*(G).

Conversely, let Γ be a subgroupoid of p∗(G)p^*(G); suppose that the mapping (x,f,y)↦(x,f)(x, f, y)\mapsto (x, f) of Fl(Γ) into X×X\times Fl(G) is injective and that its image is the set of pairs (x,f)(x, f) such that p(x)=o(f)p(x) =o(f). Then there exists a unique operation φ\varphi of the groupoid G on X such that Γ=Gφ\Gamma = G_{\varphi}.

The orbits of the groupoid GφG_{\varphi} are called the orbits of the operation of G on X. By definition, they are the equivalence classes of the relation in X defined by R{x,y}R\{x, y\} if and only if there exists f∈f\in Fl(G) such that φ(f)(x)=y\varphi (f)(x) =y.

Example 1 01TE

Let G be a group, and let G\mathscr{G} be the groupoid associated with G (II, p. 163, Example 1). If X is a set, an operation (on the right) of the groupoid G\mathscr{G} on X is nothing other than an operation (on the right) of the group G on X. Moreover, the orbits of the operation of G\mathscr{G} coincide with those of the group G.

Example 2 01TF

Let G=(S,F,o,t)G = (S,F, o, t) be a groupoid. There exists a unique operation of G on the set S, relative to IdS_S. It is given by φ(f)(a)=b\varphi (f)(a) =b if f∈Ga,bf\in G_{a,b}. The orbits for this operation are the orbits of G in the sense defined on p. 162.

Let G be a groupoid, let S be the set of its vertices, let X be a set, and let p:X→Sp: X\rightarrow S be a mapping. Let φ\varphi be an operation of G on X relative to pp. One says that the groupoid G operates without monodromy on X if, for every point a∈Sa\in S and every element f∈Ga,φ(f)f\in G_a,\varphi (f) is the identity of XaX_a. If the groupoid G is transitive, it is enough that this be so for one point of S.

Let aa be a point of S and suppose that one has φ(f)=\varphi (f) = IdXa_{X_a} for every f∈Gaf\in G_a. Let bb be a point of S belonging to the orbit of aa and let gg be an arrow of G joining aa to bb. For every f∈Gbf\in G_b, Int(gg)(f)=gfg−1(f) =gf g^{-1} is an element of GaG_a; hence one has IdXa=φ(gfg−1)=φ(g)φ(f)φ(g)−1_{X_a}=\varphi (gf g^{-1}) =\varphi (g)\varphi (f)\varphi (g)^{-1}, so that φ(f)=\varphi (f) = IdXb_{X_b}. Let bb be a point of S belonging to the orbit of aa and let g,g′g,g' be arrows of G joining aa to bb; then g′g−1g'g^{-1} is a loop at aa, whence φ(g′g−1)=\varphi (g'g^{-1}) = IdXa_{X_a} and φ(g)=φ(g′)\varphi (g) =\varphi (g').

8. Normal Subgroupoids; Quotient Groupoids

Definition 7 01TG

Let G be a groupoid. One says that a subgroupoid H of G is normal if Som(H) = Som(G) and if for every pair (a,b)(a, b) of vertices of H and every arrow f∈Ga,bf\in G_{a,b}, one has Int(ff)(Hb)=Ha(H_b) = H_a.

Let G be a groupoid and let H be a subgroupoid of G whose set of vertices is equal to Som(G). To verify that H is distinguished, it is enough to prove that fHbf−1⊂HafH_bf^{-1}\subset H_a for every a∈a\in Som(G), every b∈b\in Som(G) and every f∈Ga,bf\in G_{a,b}. In fact, if this is so, one also has f−1Haf⊂Hbf^{-1}H_af\subset H_b for every arrow f∈Gb,af\in G_{b,a}, that is, Ha⊂fHbf−1H_a\subset fH_bf^{-1}, and consequently fHbf−1=HafH_bf^{-1}= H_a.

Let G be a groupoid and let H be a distinguished subgroupoid of G. For every vertex aa of G, the subgroup HaH_a of GaG_a is a distinguished subgroup of GaG_a.

Let φ:G→G′\varphi : G\rightarrow G' be a morphism of groupoids. The kernel of φ(\varphi (II, p. 166, example 5) is a distinguished subgroupoid of G. In fact, let ff be an arrow in G joining aa to bb and let g∈g\in Ker(φ\varphi )b_b; then one has φ(fgf−1)=φ(f)φ(g)φ(f−1)=φ(f)eφ(b)φ(f)−1=eφ(a)\varphi (f gf^{-1}) =\varphi (f)\varphi (g)\varphi (f^{-1}) =\varphi (f)e_{\varphi(b)}\varphi (f)^{-1}=e_{\varphi(a)}, whence the inclusion ffKer(φ\varphi )bf−1⊂_bf^{-1}\subset Ker(φ\varphi )a_a.

Let G be a groupoid. The groupoid G and the subgroupoid of G whose set of arrows is the set of identity elements of G are distinguished subgroupoids of G. The intersection of a family of distinguished subgroupoids of G is a distinguished subgroupoid of G. In particular, for every subset F⊂F\subset Fl(G), there exists a smallest distinguished subgroupoid of G whose set of arrows contains F. It is called the distinguished subgroupoid generated by F.

Let H be a distinguished subgroupoid of G. Let R\mathscr{R} be the equivalence relation in Fl(G) defined by R{f,g}\mathscr{R}\{f, g\} if and only if there exist arrows xx and yy in Fl(H) such that f=xgyf=xgy. If ff and gg are arrows of G equivalent modulo R\mathscr{R}, their origins (resp. their ends) belong to the same orbit of H. Put F′=F'= Fl(G)/R/\mathscr{R} and denote by o′o' and t′t' the mappings of F′F' into Orb(H) deduced from oo and tt by passing to the quotients. Denote by G/HG/H the quiver (Orb(H),F′,o′,t′,F', o', t').

Lemma 01TH

Given composable arrows uu and vv of G/HG/H, one can choose representatives ff and gg of uu and vv in Fl(G) which are composable. The class modulo R\mathscr{R} of the product fgf g does not depend on the choice of ff and gg.

Let ff and gg be arbitrary representatives of uu and vv in Fl(G). The end of ff and the origin of gg belong to one and the same orbit of H, hence can be joined by an arrow xx of H (II, p. 162, prop. 1). The arrows fxf x and gg are then representatives of uu and vv in F which are composable in G. This proves the first assertion.

Let now f,gf,g on the one hand, and f′,g′f',g' on the other hand, be representatives of uu and vv which are composable in G. By hypothesis, there exist arrows x,y,z,tx,y,z,t in H such that one has f′=xfyf'=xf y and g′=zgtg'=zgt. Then one has yz∈Ht(f)yz\in H_{t(f)} and f′g′=xfyzgt=xf(yz)f−1fgtf'g'=xf yzgt=xf(yz)f^{-1}f gt. Since H is a normal subgroupoid of G, the arrow f(yz)f−1f(yz)f^{-1} is an arrow of H. The same is therefore true of the arrow xf(yz)f−1xf(yz)f^{-1}, which proves that f′g′f'g' and fgf g are equivalent modulo R\mathscr{R}.

By virtue of this lemma, there exists a unique law of composition mm in the quiver G/HG/H such that, for every pair (u,v)(u, v) of composable arrows, m(u,v)m(u, v) is the class modulo R\mathscr{R} of the product fgf g, for every pair (f,g)(f, g) of composable arrows of G such that uu is the class of ff and vv that of gg. Endowed with this law of composition, G/HG/H is a groupoid. Let p1:p_1: Som(G) →\rightarrow Som(G/H/H) and p2:p_2: Fl(G) →\rightarrow Fl(G/H/H) be the canonical surjections. The pair p=(p1,p2)p= (p_1, p_2) is a morphism of groupoids of G into G/HG/H whose kernel is the normal subgroupoid H.

Definition 8 01TI

One says that G/HG/H is the quotient groupoid of G by H and that p:G→G/Hp: G\rightarrow G/H is the canonical morphism.

Remark 1 01TJ

The mapping Som(pp) defines, by passing to quotients, a bijection of the set of orbits of G onto the set of orbits of G/HG/H. In particular, G is transitive if and only if G/HG/H is.

Remark 2 01TK

Let aa and bb be vertices of G. The mapping of Ga,bG_{a,b} into (G/H)p(a),p(b)(G/H)_{p(a),p(b)} deduced from pp is surjective. Let in fact uu be an arrow of G/HG/H joining p(a)p(a) to p(b)p(b); it is the class modulo R\mathscr{R} of an arrow ff of G. The origin o(f)o(f) of ff belongs to the orbit of aa in H; hence there exists an arrow xx in H joining aa to o(f)o(f). Analogously, there exists an arrow yy in H joining t(f)t(f) to bb. Then f′=xfyf'=xf y is an element of Ga,bG_{a,b} whose class modulo R\mathscr{R} is uu.

Proposition 2 01TL

Let aa be a vertex of G. The group homomorphism pa:Ga→(G/H)p(a)p_a: G_a\rightarrow (G/H)_{p(a)} deduced from pp by passing to the isotropy groups is surjective; its kernel is HaH_a.

The homomorphism pap_a is surjective by virtue of the preceding remark. Its kernel contains HaH_a by construction of the groupoid G/HG/H. Let ff be an element of GaG_a such that pa(f)=ep(a)p_a(f) =e_{p(a)}. This means that ff and eae_a are equivalent modulo R\mathscr{R}; there then exist arrows xx and yy of H such that f=xeayf=xe_ay. Necessarily, xx and yy belong to HaH_a, whence f∈Haf\in H_a.

Proposition 3 01TM

Let G be a groupoid, let H be a distinguished subgroupoid of G and let p:G→G/Hp: G\rightarrow G/H be the canonical morphism. Let φ:G→G′\varphi : G\rightarrow G' be a groupoid morphism such that H⊂H\subset Ker(φ\varphi ). There exists a unique groupoid morphism φ‾:G/H→G′\overline{\varphi}: G/H\rightarrow G' such that φ‾∘p=φ\overline{\varphi}\circ p=\varphi.

One says that φ‾\overline{\varphi} is the groupoid morphism deduced from φ\varphi by passing to the quotient.

The uniqueness of such a morphism is obvious, since the mappings Som(pp) and Fl(pp) are surjective.

Let aa and bb be vertices of G. If they are in the same orbit of H, there exists an arrow ff joining aa to bb in H, and one has φ(f)=eφ(a)\varphi (f) =e_{\varphi(a)}. In particular, φ(a)=φ(b)\varphi (a) =\varphi (b). Consequently, the mapping Som(φ\varphi ) defines, by passing to the quotient, a mapping φ‾1:\overline{\varphi}_1: Orb(H) →\rightarrow Som(G′'). Let ff and gg be arrows in G. If ff and gg are equivalent modulo R\mathscr{R}, there exist arrows xx and yy in H such that ff = xgyxgy. Consequently, φ(f)=φ(x)φ(g)φ(y)=φ(g)\varphi (f) =\varphi (x)\varphi (g)\varphi (y) =\varphi (g) since φ(x)\varphi (x) and φ(y)\varphi (y) are identity elements. The mapping Fl(φ\varphi ) therefore defines, by passing to the quotient, a mapping φ‾2:\overline{\varphi}_2: Fl(G)/R→/\mathscr{R}\rightarrow Fl(G′').

Let ff and gg be composable arrows of G; let us denote by uu and vv their classes in Fl(G/H/H). One has φ‾2(uv)=φ(fg)=φ(f)φ(g)=\overline{\varphi}_2(uv) =\varphi (f g) =\varphi (f)\varphi (g) = φ‾2(u)φ‾2(v)\overline{\varphi}_2(u)\overline{\varphi}_2(v).

The pair φ‾=(φ‾1,φ2)\overline{\varphi}= (\overline{\varphi}_1, \varphi_2) is a groupoid morphism from G/HG/H into G′G' and one has φ‾∘p=φ\overline{\varphi}\circ p=\varphi.

9. Groupoid of classes of paths of a graph

Let G be a graph. Let us denote by ΩG\Omega_G the category of paths of the underlying quiver of G (II, p. 160, example 1). Consider the finest equivalence relation R\mathscr{R} in Ch(G) such that one has:

(i) For every pair (a,b)(a, b) of vertices of G and every arrow ff in G joining aa to bb, the paths (a,f,b,f‾,a)(a, f, b,\overline{f}, a) and eae_a are equivalent modulo R\mathscr{R};

(ii) If (c,d)(c, d) and (c′,d′)(c', d') are pairs of juxtaposed paths in G such that R{c,c′}\mathscr{R}\{c, c'\} and R{d,d′}\mathscr{R}\{d, d'\}, the paths c∗dc*d and c′∗d′c'*d' are equivalent modulo R\mathscr{R}.

Two paths equivalent modulo R\mathscr{R} have the same origin and the same end. The mappings oo and tt from Ch(G) into Som(G) define, by passing to the quotient, mappings o′o' and t′t' from Ch(G)/R/\mathscr{R} into Som(G). Let us denote by ϖG\varpi_G the quiver (Som(G), Ch(G)/R,o′,t′/\mathscr{R}, o', t'). The pair φ\varphi formed by the identity mapping of Som(G) and the canonical projection of Ch(G) onto Ch(G)/R/\mathscr{R} is a morphism of quivers from ΩG\Omega_G into ϖG\varpi_G. The juxtaposition of paths in Ch(G) defines, by passing to the quotients, a law of composition in ϖG\varpi_G.

Proposition 4 01TN

Endowed with this law of composition, ϖG\varpi_G is a groupoid.

By construction, one has the relation φ(cc′)=φ(c)φ(c′)\varphi (cc') =\varphi (c)\varphi (c'), for every pair of juxtaposed paths (c,c′)(c, c') in G. Every arrow of ϖG\varpi_G is of the form φ(c)\varphi (c), where cc is a path in G. It follows that the law of composition of ϖG\varpi_G is associative and that φ(ea)\varphi (e_a) is an identity element at aa, for every vertex aa of ϖG\varpi_G. It remains to prove that every arrow of ϖG\varpi_G is invertible. Let cc be a path in G, and let us prove by induction on the length of cc that one has φ(c)φ(c‾)=φ(eo(c))\varphi (c)\varphi (\overline{c}) =\varphi (e_{o(c)}). This equality is true if cc is of length 0. If cc is of length n⩾1n\geqslant 1, we may write c=c1c2c=c_1c_2, with c1c_1 of length 1 and c2c_2 of length n−1n-1. Then, c‾=c2‾c1‾\overline{c}=\overline{c_2}\overline{c_1} and one has

φ(c)φ(c‾)=φ(c1)φ(c2)φ(c2‾)φ(c1‾)=φ(c1)φ(eo(c2))φ(c1‾)=φ(c1)φ(c1‾) \varphi (c)\varphi (\overline{c}) =\varphi (c_1)\varphi (c_2)\varphi (\overline{c_2})\varphi (\overline{c_1}) =\varphi (c_1)\varphi (e_{o(c_2)})\varphi (\overline{c_1}) =\varphi (c_1)\varphi (\overline{c_1})
=φ(c1c1‾)=φ(eo(c1)) =\varphi (c_1\overline{c_1}) =\varphi (e_{o(c_1)})

by definition of the relation R\mathscr{R}.

Applying this equality to the path c‾\overline{c}, we see that φ(c‾)φ(c)=\varphi (\overline{c})\varphi (c) = φ(et(c))\varphi (e_{t(c)}). Thus, φ(c)\varphi (c) is invertible, with inverse φ(c‾)\varphi (\overline{c}).

The equivalence classes of the relation R\mathscr{R} are called the path classes in the graph G; the groupoid ϖG\varpi_G is called the groupoid of path classes of the graph G. It has the same set of vertices as G, and its orbits are the connected components of the graph G.

Let us denote by vv the mapping which associates with an arrow ff of G the class of the path (o(f),f,t(f))(o(f), f, t(f)) of G. The pair jj = (IdSom(G),v_{Som(G)}, v) is a morphism, called canonical, of quivers from G into ϖG\varpi_G. Its image generates ϖG\varpi_G; for every arrow ff of G, one has j(f‾)=j(f)−1j(\overline{f}) =j(f)^{-1}.

Let G and G′G' be graphs, let φ:G→G′\varphi : G\rightarrow G' be a graph morphism. Let us denote by j:G→ϖGj: G\rightarrow \varpi_G and j′:G′→ϖG′j': G'\rightarrow \varpi_{G'} the canonical morphisms. If c=(a0,f1,a1,...,an)c= (a_0, f_1, a_1, . . . , a_n) is a path in G, the class of the path φ(c)=(φ(a0),φ(f1),φ(a1),...,φ(an))\varphi (c) = (\varphi (a_0), \varphi (f_1), \varphi (a_1), . . . , \varphi (a_n)) depends only on the class of cc. Thus, by passing to equivalence classes, one defines a quiver morphism ϖ(φ):ϖG→ϖG′\varpi (\varphi ):\varpi_G\rightarrow \varpi_{G'} such that ϖ(φ)∘j=j′∘φ\varpi (\varphi )\circ j=j'\circ \varphi. It is a groupoid morphism.

Proposition 5 01TO

Let G be a graph; let us denote by jj the canonical quiver morphism of G into ϖG\varpi_G. Let φ\varphi be a quiver morphism of G into a groupoid G′G' such that φ(f‾)=φ(f)−1\varphi (\overline{f}) =\varphi (f)^{-1} for every arrow ff of G. Then there exists a unique groupoid morphism φ′\varphi ' of ϖG\varpi_G into G′G' such that φ′∘j=φ\varphi '\circ j=\varphi.

One defines a mapping u:u: Ch(G) →\rightarrow Fl(G′') by setting, for every path c=(a0,f1,a1,...,fn,an)c= (a_0, f_1, a_1, . . . , f_n, a_n) in G,u(c)=ea0φ(f1)...φ(fn),u(c) =e_{a_0}\varphi (f_1). . . \varphi (f_n). For every pair (c,c′)(c, c') of composable paths in G, one has u(cc′)=u(cc') = u(c)u(c′)u(c)u(c'). The mapping uu is compatible with the equivalence relation R\mathscr{R} defined above, whence, by passing to the quotient, a mapping u′:u': Ch(G)/R→/\mathscr{R}\rightarrow Fl(G′'). Let us then set φ′=\varphi '= (Som(φ\varphi ),u′, u'). It is a groupoid morphism of ϖG\varpi_G into G′G' such that φ′∘j=φ\varphi '\circ j=\varphi.

Let ψ\psi be a groupoid morphism of ϖG\varpi_G into G′G' such that ψ∘j=φ\psi \circ j=\varphi. The equalizer of φ′\varphi ' and ψ\psi is a subgroupoid of ϖG\varpi_G which contains j(G)j(G). Hence it is equal to ϖG\varpi_G, because ϖG\varpi_G is generated by j(G)j(G), and one has ψ=φ′\psi =\varphi '.

Corollary 1 01TP

In a graph, every class of paths contains a unique path without retracing.

Let G be a graph. Let us denote by j:G→ϖGj: G\rightarrow \varpi_G the canonical quiver morphism.

The existence, for every path cc of G, of an equivalent path without retracing is immediate by induction on the length of cc (cf. II, p. 157).

Let A be an orientation of G and let G′G' be the groupoid associated with the free group F(A) constructed on A (II, p. 163, Example 1). Let ψ\psi be the morphism of quivers from G into G′G' such that, for every f∈A,ψ(f)f\in A,\psi (f) is the element ff of F(A) and ψ(f‾)\psi (\overline{f}) is the element f−1f^{-1} of F(A). Let ψ′\psi ' be the unique morphism of groupoids from ϖG\varpi_G into G′G' such that ψ′∘j=ψ\psi '\circ j=\psi.

Let c,c′c,c' be paths without backtracking equivalent modulo R\mathscr{R}. They have the same source and the same end. To prove that they are equal, it is enough to prove that the sequences (f1,...,fn)(f_1, . . . , f_n) and (g1,...,gm)(g_1, . . . , g_m) of their edges are equal. The image under ψ′\psi ' of the common class of cc and c′c' is equal to ψ(f1)...ψ(fn)\psi (f_1). . . \psi (f_n) and to ψ(g1)...ψ(gm)\psi (g_1). . . \psi (g_m). Now the terms of the sequences (ψ(f1),...,ψ(fn))(\psi (f_1), . . . , \psi (f_n)) and (ψ(g1),...,ψ(gm))(\psi (g_1), . . . , \psi (g_m)) belong to the subset A∪A−1A\cup A^{-1} of F(A) and two consecutive elements of these sequences are not inverses of one another. By A, I, p. 84, Prop. 7, these two sequences are equal. It follows that the sequences (f1,...,fn)(f_1, . . . , f_n) and (g1,...,gm)(g_1, . . . , g_m) are equal, and therefore that c=c′c=c', whence uniqueness.

Corollary 2 01TQ

The canonical morphism from G into ϖG\varpi_G is injective.

This results immediately from Corollary 1.

Corollary 3 01TR

Let G′G' be a subgraph of G. The morphism from ϖG′\varpi_{G'} into ϖG\varpi_G deduced from the injection of G′G' into G is injective.

Let us denote by ii the inclusion morphism of G′G' into G and by ϖ(i)\varpi (i) the morphism from ϖG′\varpi_{G'} into ϖG\varpi_G deduced from it. The mappings Som(ii) and Som(ϖ(i)\varpi (i)) coincide. Moreover, if cc is a path without backtracking in G′G', the path i(c)i(c) is a path without backtracking in G. The assertion therefore results from Corollary 1.

Corollary 4 01TS

A nonempty graph G is a tree if and only if the groupoid ϖG\varpi_G is simply transitive.

Let aa be a point of G. By Corollary 1, the following properties are equivalent:

(i) The isotropy group of ϖG\varpi_G at aa is reduced to the identity element;

(ii) The only loop in G with origin aa which is without backtracking is the constant loop with origin aa.

The groupoid ϖG\varpi_G is simply transitive if and only if it is transitive and has property (i) for every aa. The graph G is a tree if and only if it is connected and has property (ii) for every point aa. Since the orbits of ϖG\varpi_G are identified with the connected components of G, the corollary follows.

10. Free Groupoids

Definition 9 01TT

Let C be a quiver. The groupoid ϖC~\varpi_{\widetilde{C}} of classes of paths of the graph C~\widetilde{C} associated with C is called the free groupoid constructed on C, and is denoted by Grp(C).

The set of vertices of Grp(C) is equal to that of the vertices of C; the orbits of Grp(C) are the connected components of C.

Let C be a nonempty quiver. By Corollary 4, p. 174, the graph C~\widetilde{C} associated with C is a tree if and only if the free groupoid Grp(C) constructed on C is simply transitive.

The composite of the canonical quiver morphisms i:C→C~i: C\rightarrow \widetilde{C} and j:C~→ϖCj:\widetilde{C}\rightarrow \varpi_C is a quiver morphism from C into Grp(C) which is called the canonical morphism of C into Grp(C). Let us denote it by θ\theta. For every vertex aa of C, one has θ(a)=a\theta (a) =a. If ff is an arrow of C,θ(f),\theta (f) is the class of the path (o(f),(f,1),t(f))(o(f),(f,1), t(f)) of C~\widetilde{C}. The morphism θ\theta is injective (II, p. 174, Corollary 2) and its image generates the groupoid Grp(C).

Let φ:C→C′\varphi : C\rightarrow C' be a quiver morphism. Let us denote by θC\theta_C (resp. θC′\theta_{C'}) the canonical morphism of C into Grp(C) (resp. of C′C' into Grp(C′')). The morphism ϖ(φ~)\varpi (\widetilde{\varphi}) is the unique groupoid morphism ψ\psi from Grp(C) into Grp(C′') such that ψ∘θC=θC′∘φ\psi \circ \theta_C=\theta_{C'}\circ \varphi.

Proposition 6 01TU

Let C be a quiver, let G be a groupoid, and let φ\varphi be a quiver morphism from C into G. There exists a unique groupoid morphism φ′\varphi ' from Grp(C) into G such that φ′∘θC=φ\varphi '\circ \theta_C=\varphi.

Let ψ\psi be the quiver morphism from C~\widetilde{C} into G such that ψ(a)=φ(a)\psi (a) =\varphi (a) for every vertex aa of C and ψ(f,ε)=φ(f)ε\psi (f, \varepsilon ) =\varphi (f)^{\varepsilon} for every arrow ff of C and every ε∈{−1,1}\varepsilon \in \{-1,1\}. One has ψ∘i=φ\psi \circ i=\varphi. By Proposition 5, p. 173, there exists a morphism φ′\varphi ' from Grp(C) =ϖC=\varpi_C into G such that φ′∘j=ψ\varphi '\circ j=\psi. One then has φ′∘θC=φ′∘j∘i=ψ∘i=~φ\varphi '\circ \theta_C=\varphi '\circ j\circ i=\psi \circ i\widetilde{=}\varphi. As in the proof of Prop. 5, the uniqueness of φ′\varphi ' results from the fact that θC(C)\theta_C(C) generates the groupoid Grp(C).

Under the hypotheses of Prop. 6, one sometimes says that φ′\varphi ' is the groupoid morphism from Grp(C) into G extending the quiver morphism φ\varphi.

Example 01TV

Let C be a quiver having a unique vertex ss and let A be the set of its arrows. The groupoid Grp(C) is then the groupoid associated with the free group F(A) constructed on A. In fact, it has a unique vertex ss and it results immediately from Prop. 6 that the canonical mapping from A into Grp(C)s_s deduced from the canonical morphism of C into Grp(C) satisfies the universal property of free groups (A, I, p. 85, Prop. 8).

11. Contraction of arrows of a groupoid

Let G be a groupoid and let F be a subset of the set of arrows of G. Let H be the normal subgroupoid of G generated by F. The groupoid G/HG/H is called the groupoid deduced from G by contraction of the set of arrows F and is denoted by G/FG/F. If pp denotes the canonical morphism from G into G/FG/F, one has p(o(f))=p(t(f))p(o(f)) =p(t(f)) and p(f)=ep(o(f))p(f) =e_{p(o(f))} for every arrow f∈Ff\in F.

This morphism satisfies the following universal property.

Proposition 7 01TW

For every groupoid morphism φ:G→G′\varphi : G\rightarrow G' which maps every arrow belonging to F onto an identity element of G′G', there exists a unique groupoid morphism φ′:G/F→G′\varphi ': G/F\rightarrow G' such that φ′∘p=φ\varphi '\circ p=\varphi.

The kernel of φ\varphi is a normal subgroupoid of G and the set of its arrows contains F. By definition, H is the smallest normal subgroupoid of G whose set of arrows contains F; it is therefore contained in the kernel of φ\varphi. The proposition then results from II, p. 170, Prop. 3.

Let us denote by Γ the subquiver of G whose set of vertices is Som(G) and whose set of arrows is F. The orbits of H are identified with the connected components of the quiver Γ. By definition of the quotient of a groupoid by a normal subgroupoid, the set of vertices of G/FG/F is identified with the set of connected components of Γ; in other words, it is the quotient set of Som(G) by the finest equivalence relation such that o(f)o(f) is equivalent to t(f)t(f) for every arrow f∈Ff\in F.

The mapping pp induces, by passing to quotients, a bijection of the set of orbits of G onto the set of orbits of G/FG/F (II, p. 170, remark 1).

The purpose of this Noo^o is to calculate the homomorphisms induced by pp by passing to the isotropy groups, which amounts, according to prop. 2 (II, p. 170), to calculating the isotropy groups of the subgroupoid H.

Proposition 8 01TX

Let δ\delta be the canonical morphism of the free groupoid Grp(Γ) into G extending the canonical injective morphism of Γ into G. Let aa be a vertex of G and let A be the orbit of aa in G. For every b∈Ab\in A, let fabf_{ab} be an arrow of G joining aa to bb. The isotropy group HaH_a is the distinguished subgroup of GaG_a generated by the elements Int(fabf_{ab})(δ(c))(\delta (c)), where bb runs through the set A and cc runs through the elements of Grp(Γ)b_b.

If xx and yy are vertices of G belonging to the same orbit, with x=a̸x=\not a, let us moreover fix an arrow fxyf_{xy} of G joining xx to yy. For every point xx of G, let NxN_x be the distinguished subgroup of GxG_x generated by the elements Int(fxyf_{xy})(δ(c))(\delta (c)), where yy runs through the orbit of xx in G and cc runs through the group Grp(Γ)y_y. For every f∈Gxyf\in G_{xy} and every g∈Nyg\in N_y, one has fgf−1∈Nxf gf^{-1}\in N_x. Let x∈x\in Som(G). One has δ\delta(Grp(Γ)x_x)⊂Nx\subset N_x, by definition of NxN_x. By definition of the groupoid H,δ(f),\delta (f) is an arrow of H for every arrow ff of Grp(Γ) ; consequently, δ\delta(Grp(Γ)x_x) is contained in HxH_x. Since H is a distinguished subgroupoid of G, one then has

Int(fxyf_{xy})(δ(c))∈(\delta (c))\in Int(fxyf_{xy})(Ny)⊂(N_y)\subset Int(fxyf_{xy})(Hy)⊂Hx(H_y)\subset H_x

for every y∈y\in Som(G), and hence Nx⊂HxN_x\subset H_x.

Let xx and yy be vertices of Γ. Put Nx,y=∅N_{x,y}=\emptyset if xx and yy do not belong to the same connected component of Γ. In the contrary case, let cc and c′c' be classes of paths joining xx to yy in the quiver Γ~\widetilde{\Gamma} associated with Γ. Let zz be a vertex of G belonging to the orbit of yy and let ℓ\ell be an element of Grp(Γ)z_z; in the group GxG_x, we have the equality:

δ(c)fyzδ(ℓ)fyz−1δ(c′)−1 \delta (c)f_{yz}\delta (\ell )f_{yz}^{-1}\delta (c')^{-1}
=δ(c)fyzfxz−1(fxzδ(ℓ)fxz−1)fxzfyz−1δ(c)−1δ(c(c′)−1) =\delta (c)f_{yz}f_{xz}^{-1}(f_{xz}\delta (\ell )f_{xz}^{-1})f_{xz}f_{yz}^{-1}\delta (c)^{-1}\delta (c(c')^{-1})

The arrow δ(c(c′)−1)\delta (c(c')^{-1}) is the class of a loop at xx in the quiver Γ, hence belongs to NxN_x, as does the arrow fxzδ(ℓ)fxz−1f_{xz}\delta (\ell )f_{xz}^{-1}. Since δ(c)fyzfxz−1\delta (c)f_{yz}f_{xz}^{-1} belongs to GxG_x and NxN_x is a distinguished subgroup of GxG_x, it follows that δ(c)fyzδ(ℓ)fyz−1δ(c′)−1\delta (c)f_{yz}\delta (\ell )f_{yz}^{-1}\delta (c')^{-1} belongs to NxN_x. Consequently, δ(c)Ny⊂Nxδ(c′)\delta (c)N_y\subset N_x\delta (c'). By symmetry, we have δ(c)Ny=Nxδ(c′)\delta (c)N_y= N_x\delta (c'). This implies that this set does not depend on the choice of cc and c′c'; we denote it by Nx,yN_{x,y}.

Let N be the subquiver of G whose set of vertices is Som(G) and whose set of arrows joining xx to yy is equal to Nx,yN_{x,y} for every pair (x,y)(x, y) of vertices of G. It is a distinguished subgroupoid of G whose set of arrows contains F; hence H is a subgroupoid of N. In particular, Ha⊂NaH_a\subset N_a, whence the equality.

Corollary 1 01TY

Suppose further that, for every arrow ff of F, we have o(f)=t(f)o(f) =t(f). Then the origin and the end of every path in the quiver Γ coincide. Let aa be a vertex of G and let A be its orbit in G. For every element cc of F whose origin bb belongs to A, let us fix an arrow fcf_c of G joining aa to bb and put κ(c)=\kappa (c) = Int(fcf_c)(δ(c))=(\delta (c)) = fcδ(c)fc−1f_c\delta (c)f_c^{-1}.

The group HaH_a is then the distinguished subgroup of GaG_a generated by the arrows κ(c)\kappa (c).

By Prop. 8, the arrows κ(c)\kappa (c) are contained in the group HaH_a. Let NaN_a denote the smallest normal subgroup of GaG_a containing them. By that proposition, it is enough to prove that, for every vertex bb of G belonging to A, every arrow fabf_{ab} joining aa to bb in G, and every element cc of Grp(Γ)b_b, Int(fabf_{ab})(δ(c))(\delta (c)) belongs to NaN_a. Since the origin and the term of every element of F are equal, a loop at a vertex bb in the graph Γ~\widetilde{\Gamma} is of the form (b,(f1,ε1),b,...,(fn,εn),b)(b,(f_1, \varepsilon_1), b, . . . ,(f_n, \varepsilon_n), b), where for every i∈{1,...,n},fii\in \{1, . . . , n\},f_i is an element of F with origin bb and εi∈{−1,1}\varepsilon_i\in \{-1,1\}. We have

Int(fabf_{ab})(δ(c))=fabfc−1κ(f1)ε1...κ(fn)εnfcfab−1(\delta (c)) =f_{ab}f_c^{-1}\kappa (f_1)^{\varepsilon_1}. . . \kappa (f_n)^{\varepsilon_n}f_cf_{ab}^{-1}.

Since fabfc−1∈Gaf_{ab}f_c^{-1}\in G_a, it follows that Int(fabf_{ab})(δ(c))(\delta (c)) belongs to NaN_a.

Corollary 2 01TZ

If the graph associated with the quiver Γ is a forest, the homomorphism pap_a of GaG_a into (G/F)p(a)(G/F)_{p(a)} is bijective, for every vertex aa of G.

In fact, under this assumption, one has Grp(Γ)b={eb}_b=\{e_b\} for every vertex bb of G (Corollary 1 of II, p. 173). It then follows from Proposition 8 that, for every vertex aa of G, the group HaH_a reduces to the identity element.

12. Poincaré Group of a Graph

Let G be a graph. Let aa be a vertex of G; the isotropy group at aa of the groupoid Grp(G) is called the Poincaré group of G at aa and is denoted by π1(G,a)\pi_1(G, a). Let cc be a class of paths in G, and let aa be its origin and bb its term. The map Int(cc):π1(G,b)→π1(G,a):\pi_1(G, b)\rightarrow \pi_1(G, a) defined by c′↦cc′c−1c'\mapsto cc'c^{-1} is a group isomorphism. Let φ:G→H\varphi : G\rightarrow H be a graph morphism. Let θG:G→\theta_G: G\rightarrow Grp(G) and θH:H→\theta_H: H\rightarrow Grp(H) denote the canonical morphisms. If φ‾\overline{\varphi} denotes the unique groupoid morphism Grp(G) →\rightarrow Grp(H) such that φ‾∘θG=θH∘φ\overline{\varphi}\circ \theta_G=\theta_H\circ \varphi, the group homomorphism φ‾a:π1(G,a)→π1(H,φ(a))\overline{\varphi}_a:\pi_1(G, a)\rightarrow \pi_1(H, \varphi (a)) is denoted by π1(φ,a)\pi_1(\varphi , a).

Let G be a connected graph, let S be an orientation of G and let A be a maximal oriented tree of G (II, p. 157, prop. 1). Given vertices aa and bb of G, there exists a unique class of paths γa,b\gamma_{a,b} in the graph A~\widetilde{A} associated with A which joins aa to b(b(II, p. 174, cor. 4 of prop. 5). If aa, bb and cc are vertices of G, one has γa,bγb,c=γa,c\gamma_{a,b}\gamma_{b,c}=\gamma_{a,c}, these two classes of paths being equal to the unique class of paths joining aa to cc in A~\widetilde{A}.

Proposition 9 01U0

Let aa be a vertex of G. There exists a unique homomorphism λ\lambda of the free group F(S-Fl(A)) into π1(G,a)\pi_1(G, a) such that

λ(f)=γa,o(f)⋅f⋅γt(f),a(1) \lambda (f) =\gamma_{a,o(f)}\cdot f\cdot \gamma_{t(f),a} \tag{1}

for f∈Sf\in S - Fl(A). The homomorphism λ\lambda is a group isomorphism.

Let us denote by L the group F(S-Fl(A)). The existence and uniqueness of the homomorphism λ:L→π1(G,a)\lambda : L\rightarrow \pi_1(G, a) satisfying relations (1) follows from A, I, p. 85, prop. 8. Let L\mathscr{L} be the groupoid associated with the group L (II, p. 163, example 1); let us denote by ss its unique vertex. There exists a unique groupoid morphism μ:\mu: Grp(G) →L\rightarrow \mathscr{L} such that μ(θG(f))=f\mu(\theta_G(f)) =f for every arrow f∈f\in S-Fl(A) and μ(θG(f))=es\mu(\theta_G(f)) =e_s for every arrow f∈f\in Fl(A) (II, p. 173, prop. 5). For f∈f\in S-Fl(A), one has μ(θG(f))=f\mu(\theta_G(f)) =f; the homomorphism μa∘λ\mu_a\circ \lambda is therefore the identity isomorphism of the group L (A, I, p. 85, prop. 8). It follows that λ\lambda is injective.

Let ϖG/A\varpi_G/A be the groupoid deduced from ϖG\varpi_G by contraction of the arrows of A and let p:ϖG→ϖG/Ap:\varpi_G\rightarrow \varpi_G/A be the canonical groupoid morphism. It is surjective and the group homomorphism pap_a deduced from pp by passing to the isotropy groups is an isomorphism (II, p. 178, cor. 2 of prop. 8). Since the graph G is assumed connected and A is a maximal tree in it, the groupoid ϖG/A\varpi_G/A has a unique vertex (II, p. 176). Moreover, ϖG\varpi_G is generated by θG(G)\theta_G(G); consequently, ϖG/A\varpi_G/A is generated by the loops p(f)p(f), for f∈Sf\in S, and the group (ϖG/A)p(a)(\varpi_G/A)_{p(a)} is generated by the elements of the form p(θG(f))p(\theta_G(f)), for f∈f\in S-Fl(A). The homomorphism pa∘λp_a\circ \lambda is therefore surjective. Consequently, λ\lambda is surjective.

Remark 1 01U1

The homomorphism μa\mu_a is the inverse isomorphism of λ\lambda.

Remark 2 01U2

There exists a unique homomorphism λ:\lambda : F(Fl(G)) →π1(G,a)\rightarrow \pi_1(G, a) defined by relations (1) for every f∈f\in Fl(G). It follows from proposition 8 that the homomorphism λ\lambda is surjective and that its kernel is the smallest normal subgroup of F(Fl(G)) containing the elements ff, for f∈f\in Fl(A), and the elements f⋅f‾f\cdot \overline{f}, for f∈f\in Fl(G).

Exercises

See the exercises for § 3.

Machine translation of the English, by gpt-5.4, against glossary version 34. Not checked by a person.