Bourbaki

3. Subspaces; quotient spaces

1. Subspaces of a topological space

Let AA be a subset of a topological space XX. We have defined the topology induced on AA by the topology of XX as the inverse image of the latter under the canonical injection A→XA \to X (§ 2, no. 3, Example 1). An equivalent definition is as follows:

Definition 1

Let AA be a subset of a topological space XX. The topology induced on AA by the topology of XX is that in which the open sets are the intersections with AA of open sets of XX. The set AA with this topology is called a subspace of XX.

Example

The topology induced on the set Z\mathbf{Z} of rational integers by the topology of the rational line is the discrete topology, for the intersection of Z\mathbf{Z} and the open interval ]n−1/2,n+1/2[]n - 1/2, n + 1/2[ is the set {n}\{ n \}.

By Proposition 5 of § 2, no. 3 (or directly from Definition 1) we see that, if B⊂A⊂XB \subset A \subset X, the subspace BB of XX is identical with the subspace BB of the subspace AA of XX (transitivity of induced topologies). If G\mathfrak{G} is a subbase (resp. a base) of the topology of XX (§2§ 2, no. 3, Example 3) its trace GA\mathfrak{G}_A on AA is a subbase (resp. a base) of the topology induced on AA.

In all questions which involve the elements or subsets of AA, it is essential to distinguish carefully between their properties as points (resp. subsets) of XX, and their properties as points (resp. subsets) of the subspace AA. We shall make this distinction by using the phrases "in AA", "with respect to AA", or "relative to AA" to refer to properties in the latter category (possibly contrasting them with the phrases "in XX", "with respect to XX", "relative to XX").

An open set of the subspace AA need not be open in XX; in order that every set which is open in AA should be open in XX it is necessary and sufficient that AA is open in XX. The condition is necessary, since AA is open in AA, and it is sufficient by virtue of (OII)(O_{II}) and Definition 1.

The sets which are closed in AA are the intersections with AA of the closed sets in XX (§2§ 2, no. 3, Example 1); as above we see that every set which is closed in AA is closed in XX if and only if AA is closed in XX.

The neighbourhoods of a point x∈Ax \in A relative to AA are the intersections with AA of neighbourhoods of xx relative to XX. Every neighbourhood of xx relative to AA is a neighbourhood of xx relative to XX if and only if AA is a neighbourhood of xx in XX.

Proposition 1

If AA and BB are two subsets of a topological space XX, and B⊂AB \subset A, then the closure of BB in the subspace AA is the intersection with AA of the closure B‾\overline{B} of BB in XX.

If x∈Ax \in A, every neighbourhood of xx in AA is of the form V∩AV \cap A, where VV is a neighbourhood of xx in XX. Since V∩B=(V∩A)∩BV \cap B = (V \cap A) \cap B, it follows that xx lies in the closure of BB with respect to AA if and only if xx lies in the closure of BB with respect to XX.

Corollary

A subset BB of AA is dense in AA if and only if B‾=A‾\overline{B} = \overline{A} in XX (i.e. if and only if A⊂B‾A \subset \overline{B}).

It follows that if A,B,CA, B, C are three subsets of XX such that A⊃B⊃CA \supset B \supset C, and if BB is dense in AA, and CC is dense in BB, then CC is dense in AA (transitivity of density), for we have A‾=B‾=C‾\overline{A} = \overline{B} = \overline{C} in XX.

Proposition 2

Let AA be a dense subset of a topological space XX; then for each x∈Ax \in A and each neighbourhood VV of xx relative to AA, the closure V‾\overline{V} of VV in XX is a neighbourhood of xx in XX.

For VV contains U∩AU \cap A, where UU is an open subset of XX which contains xx, hence V‾\overline{V} contains U∩A‾=UU \cap \overline{A} = U (§ 1, no. 6, Proposition 5).

Proposition 3

Let (Ai)i∈I(A_i)_{i \in I} be a family of subsets of a topological space XX, such that one of the following properties holds: a) The interiors of the AiA_i cover XX. b) (Ai)i∈I(A_i)_{i \in I} is a locally finite closed covering of XX (§ 1, no. 5).

Under these conditions, a subset BB of XX is open (resp. closed) in XX if and only if each of the sets B∩AiB \cap A_i is open (resp. closed) in AiA_i.

Clearly if BB is open (resp. closed) in XX, then B∩AiB \cap A_i is open (resp. closed) in AiA_i. Conversely, suppose first that condition a) is satisfied; since (∁B)∩Ai=Ai−(B∩Ai)(\complement B) \cap A_i = A_i - (B \cap A_i), it is enough, by duality, to consider the case in which each of the B∩AiB \cap A_i is open with respect to AiA_i. In this case B∩A˙iB \cap \dot{A}_i is open in A˙i\dot{A}_i for each i∈Ii \in I, and therefore open in XX; and since B=⋃i(B∩A˙i)B = \bigcup_i (B \cap \dot{A}_i) by hypothesis, it follows that BB is open in XX.

Now suppose that b) is satisfied; by duality again, we need only consider the case in which each of the B∩AiB \cap A_i is closed in AiA_i, and therefore closed in XX. Since the family (B∩Ai)(B \cap A_i) is locally finite and B=⋃i(B∩Ai)B = \bigcup_i (B \cap A_i), it follows from § 1, no. 5, Proposition 4 that BB is closed in XX.

Remark

Let (Ui)i∈I(U_i)_{i \in I} be an open covering of a topological space XX, and for each i∈Ii \in I let Bi\mathfrak{B}_i be a base of the topology of the subspace UiU_i of XX; then it is clear that B=⋃i∈IBi\mathfrak{B} = \bigcup_{i \in I} \mathfrak{B}_i is a base of the topology of XX.

2. Continuity with respect to a subspace

Let XX and YY be two topological spaces, ff a mapping of XX into YY, BB a subset of YY which contains f(X)f(X). The definition of the induced topology as an initial topology (§ 2, no. 3, Proposition 4) shows that ff is continuous at x∈Xx \in X if and only if the mapping of XX into the subspace BB of YY, having the same graph as ff, is continuous at xx.

Now let AA be a subset of XX; if ff is continuous at x∈Ax \in A (resp. continuous on XX), its restriction f∣Af|A is a mapping of the subspace AA into YY, which is continuous at xx (resp. continuous on AA) by Proposition 2 of § 2, no. 1. We shall sometimes say that a mapping f:X→Yf : X \to Y is continuous relative to AA at x∈Ax \in A (resp. continuous relative to AA) if its restriction f∣Af|A is continuous at xx (resp. continuous on AA).

It should be noted that f∣Af|A can be continuous without ff being continuous at any point of XX; an example of this phenomenon is provided by the characteristic function φA\varphi_A of a subset AA of XX which is such that both AA and its complement are dense in XX (§2§ 2, Exercise 11), φA\varphi_A being regarded as a mapping of XX into the discrete space {0,1}\{0, 1\}. φA\varphi_A is not continuous at any point of XX, but its restriction to AA is constant and therefore continuous.

If AA is a neighbourhood in XX of a point x∈Ax \in A, and if f:X→Yf : X \to Y is such that f∣Af|A is continuous at xx, then ff is continuous at xx; for each neighbourhood of xx in AA is a neighbourhood of xx in XX (local character of continuity).

Proposition 4

Let* (Ai)i∈I(A_i)_{i \in I} be a family of subsets of a topological space XX whose interiors cover XX, or which is a locally finite closed covering of XX. Let ff be a mapping of XX into a topological space X′X'. If the restriction of ff to each of the subspaces AiA_i is continuous, then ff *is continuous.

For if F′F' is a closed subset of X′X' and if F=f‾(F′)F = \overline{f}(F'), then F∩AiF \cap A_i is closed in AiA_i for each i∈Ii \in I (§2§ 2, no. 1, Theorem 1) and therefore FF is closed in XX by Proposition 3 of no. 1; the result now follows from Theorem 1 of §2§ 2, no. 1.

3. Locally closed subspaces

Definition 2

A subset* LL of a topological space XX is said to be locally closed at a point x∈Lx \in L if there is a neighbourhood VV of xx in XX such that L∩VL \cap V is a closed subset of the subspace VV. L is said to be locally closed in XX if it is locally closed at each x∈Lx \in L.

Remark

Let FF be a subset of XX such that for each point xx of XX there is a neighbourhood VV of xx such that V∩FV \cap F is closed in the subspace VV; then it follows from Proposition 3 of no. 1 that FF is closed in XX. On the other hand, Proposition 5 below shows that in general there are locally closed sets which are not closed in XX.

Proposition 5

For a subset* LL of a topological space XX, the following properties are equivalent:

a) LL is locally closed in XX.

b) LL is the intersection of an open subset and a closed subset of XX.

c) LL is open in its closure L‾\overline{L} in XX.

If LL is locally closed, then, for each x∈Lx \in L, there is an open neighbourhood VxV_x of xx in XX such that L∩VxL \cap V_x is closed in VxV_x; U=⋃x∈LVxU = \bigcup_{x \in L} V_x is open in XX, and Proposition 3 of no. 1 shows that LL is closed in UU; therefore a) implies b). If L=U∩FL = U \cap F, where UU is open and FF closed in XX, we have L⊂FL \subset F; hence L⊂U∩L⊂U∩F=LL \subset U \cap L \subset U \cap F = L, which shows that L=U∩LL = U \cap L is open in LL, so that b) implies c). Finally, if L=U∩LL = U \cap L, where UU is open in XX, LL is closed in UU, hence locally closed, and thus c) implies a).

Corollary

Let f:X→X′f : X \to X' be a continuous map and L′L' a locally closed subset of X′X'; then f−1(L′)f^{-1}(L') is locally closed in XX*.

This follows immediately from Proposition 5 above and Theorem 1 of § 2, no. 1.

4. Quotient spaces

Definition 3

Let XX be a topological space, RR an equivalence relation on XX. The quotient space of XX by RR is the quotient set X/RX/R with the topology which is the quotient of the topology of XX by the relation RR (§ 2, no. 4, Example 1)*.

Unless the contrary is expressly stated, whenever we speak of X/RX/R as a topological space, it is to be understood that we mean the quotient space of XX by RR. We shall often say that this topological space is the space obtained by identifying the points of XX which belong to the same equivalence class mod RR.

Let φ\varphi be the canonical mapping X→X/RX \to X/R. By definition (§ 2, no. 4, Proposition 6 and its corollary) the open (resp. closed) sets in X/RX/R are the sets AA such that φ−1(A)\varphi^{-1}(A) is open (resp. closed) in XX; in other words, the open (resp. closed) sets in X/RX/R are in one-to-one correspondence with the open (resp. closed) subsets of XX which are saturated with respect to RR and are the canonical images of these subsets.

Proposition 6

Let XX be a topological space, RR an equivalence relation on XX, φ\varphi the canonical mapping of XX onto X/RX/R; then a mapping ff of X/RX/R into a topological space YY is continuous if and only if f∘φf \circ \varphi is continuous on XX*.

This is a particular case of § 2, no. 4, Proposition 6; it expresses the fact that the quotient topology is the final topology for the mapping φ\varphi.

Proposition 6 shows that there is a one-to-one correspondence between the continuous mappings of X/RX/R into YY and the continuous maps of XX into YY which are constant on each equivalence class mod RR.

Example

Consider the equivalence relation x≡y(mod1)x \equiv y \pmod{1} on the real line R\mathbf{R}; the quotient space of R\mathbf{R} by this relation is called the one-dimensional torus and is denoted by TT. The equivalence class of a point x∈Rx \in \mathbf{R} consists of all the points x+nx + n, where nn runs through the set Z\mathbf{Z} of rational integers. By Proposition 6 there is a one-to-one correspondence between the continuous functions on TT and the continuous functions on R\mathbf{R} which are periodic with period 1. We shall return to this important example in Chapter V, § 1.

Corollary

Let X,YX, Y be two topological spaces, RR (resp. SS) an equivalence relation on XX (resp. YY), and let f:X→Yf : X \to Y be a continuous mapping which is compatible with the equivalence relations RR and SS (Set Theory R, § 5, no. 8); then the mapping g:X/R→Y/Sg : X/R \to Y/S induced by ff (Set Theory R, § 5, no. 8) is continuous.

This is a particular case of a general property of quotient structures (Set Theory, Chapter IV, § 2, no. 6, criterion CST 20).

Proposition 7 (Transitivity of quotient spaces)

Let RR and SS be two equivalence relations on a topological space XX such that RR implies SS, and let S/RS/R be the quotient equivalence relation on the quotient space X/RX/R (Set Theory R, § 5, no. 9). Then the canonical bijection (X/R)/(S/R)→X/S(X/R)/(S/R) \to X/S is a homeomorphism.

This is a particular case of the transitivity of final topologies (§ 2, no. 4, Proposition 7. Cf. Set Theory, Chapter IV, § 2, no. 3, criterion CST 21).

5. Canonical decomposition of a continuous mapping

Let XX and YY be two topological spaces, f:X→Yf : X \to Y a continuous map, RR the equivalence relation f(x)=f(y)f(x) = f(y) on XX. Consider the canonical decomposition

f:X→φX/R→gf(X)→ψY f : X \xrightarrow{\varphi} X/R \xrightarrow{g} f(X) \xrightarrow{\psi} Y

where φ\varphi is the canonical (surjective) mapping of XX onto the quotient space X/RX/R, ψ\psi is the canonical injection of the subspace f(X)f(X) into YY, and gg is the bijection associated with ff (Set Theory R, § 5, no. 3). It is immediately seen that gg is continuous (by Proposition 6 of no. 4); this is also a particular case of a general result on quotient structures. (Cf. Set Theory, Chapter IV, § 2, no. 6). But the bijection gg is not necessarily a homeomorphism.

Proposition 8

Let f=ψ∘g∘φf = \psi \circ g \circ \varphi be the canonical decomposition of a continuous mapping f:X→Yf : X \to Y, and let RR denote the equivalence relation

f(x)=f(y). f(x) = f(y).
Then the following three conditions are equivalent: a) gg is a homeomorphism of X/RX/R onto f(X)f(X). b) The image under ff of every open set which is saturated with respect to RR is an open set in the subspace f(X)f(X). c) The image under ff of every closed set which is saturated with respect to RR is a closed set in the subspace f(X)f(X).

For the condition b) [resp. c)] expresses that the image under gg of every open (resp. closed) set in X/RX/R is an open (resp. closed) set in f(X)f(X).

Example

Let XX be a topological space, (Xi)i∈I(X_i)_{i \in I} a covering of XX, YY the sum of the subspaces XiX_i of XX; then there is a partition (Yi)i∈I(Y_i)_{i \in I} of YY into subspaces which are both open and closed, and for each i∈Ii \in I there is a homeomorphism fi:Yi→Xif_i : Y_i \to X_i. Let f:Y→Xf : Y \to X be the continuous mapping which agrees with fif_i on YiY_i for each i∈Ii \in I, and let RR be the equivalence relation f(x)=f(y)f(x) = f(y); the quotient space Y/RY/R is thus obtained by "pasting together" the YiY_i (§ 2, no. 5). Consider the bijection g:Y/R→Xg : Y/R \to X associated with ff; in general gg is not a homeomorphism, as is shown by the example in which each XiX_i consists of a single point and XX is not discrete. However, if the interiors of the XiX_i cover XX, or if (Xi)(X_i) is a locally finite closed covering of XX, then gg is a homeomorphism: for if UU is any open subset of YY which is saturated with respect to RR, then for each i∈Ii \in I the set

f(U)∩Xi=fi(U∩Yi) f(U) \cap X_i = f_i(U \cap Y_i)
is open in XiX_i, and the assertion follows from Proposition 3 of no. 1.

The following proposition gives a simple sufficient condition for gg to be a homeomorphism:

Proposition 9

Let f:X→Yf : X \to Y be a continuous surjection, and let RR denote the equivalence relation f(x)=f(y)f(x) = f(y). If there is a continuous section s:Y→Xs : Y \to X associated with ff (Set Theory, Chapter II, § 3, no. 8, Definition 11), then the mapping g:X/R→Yg : X/R \to Y associated with ff is a homeomorphism, and ss is a homeomorphism of YY onto the subspace s(Y)s(Y) of XX.

For if φ:X→X/R\varphi : X \to X/R is the canonical mapping, then gg and φ∘s\varphi \circ s are bijective, continuous and inverse to each other; likewise ss and the restriction of ff to s(Y)s(Y) are bijective, continuous and inverse to each other.

If RR is an equivalence relation on a topological space XX and

φ:X→X/R \varphi : X \to X/R
is the canonical mapping, a continuous section s:X/R→Xs : X/R \to X associated with φ\varphi is also called a continuous section of XX with respect to RR (cf. Set Theory, Chapter II, § 6, no. 2); the subspace s(X/R)s(X/R) of XX is then homeomorphic to X/RX/R. If we are given s(X/R)s(X/R), then ss is uniquely determined; s(X/R)s(X/R) is often called, by abuse of language, a (continuous) section of XX with respect to RR.

A continuous section of a topological space with respect to an equivalence relation need not exist (Exercise 12).

6. Quotient space of a subspace

Let XX be a topological space, AA a subspace of XX, RR an equivalence relation on XX, ff the canonical map X→X/RX \to X/R, gg the restriction of ff to AA. The equivalence relation g(x)=g(y)g(x) = g(y) on AA is just the relation RAR_A induced by RR on AA (Set Theory R, § 5, no. 5). Let g=ψ∘h∘φg = \psi \circ h \circ \varphi be the canonical decomposition of gg, so that if jj is the canonical injection of AA into XX we have a commutative diagram (*)

A→φA/RA→hf(A)→ψX/R.↘j↗sX \begin{array}{ccc} A & \xrightarrow{\varphi} & A/R_A \xrightarrow{h} f(A) \xrightarrow{\psi} X/R. \\ & & \\ & \searrow_j & \nearrow_s \\ & & X \end{array}

Proposition 10

The canonical bijection* h:A/RA→f(A)h : A/R_A \to f(A) is continuous. Furthermore, the following three statements are equivalent:

a) hh is a homeomorphism.

b) Every open subset of AA which is saturated with respect to RAR_A is the intersection with AA of an open subset of XX which is saturated with respect to RR.

c) Every closed subset of AA which is saturated with respect to RAR_A is the intersection with AA of a closed subset of XX which is saturated with respect to RR.

The first part of the proposition is immediate (no. 5). The second part follows from Proposition 8 of no. 5: if BB is an open (resp. closed) subset of AA which is saturated with respect to RAR_A, and g(B)=f(B)g(B) = f(B) is the intersection with f(A)f(A) of an open (resp. closed) subset CC of X/RX/R, then BB is the intersection with AA of the open (resp. closed) subset f−1(C)f^{-1}(C)

(*) This expression means that f∘j=ψ∘h∘φf \circ j = \psi \circ h \circ \varphi.

of X, which is saturated with respect to R; and conversely, if B is the intersection with A of an open (resp. closed) subset D which is saturated with respect to R, then f(B)f(B) is the intersection of f(A)f(A) and f(D)f(D), which is open (resp. closed) in X/RX/R.

Corollary 1

If A is an open (resp. closed) subset of X which is saturated with respect to R, then the canonical mapping h:A/RA→f(A)h : A/R_A \to f(A) is a homeomorphism.

For if A is open (resp. closed) in X and saturated with respect to R, and if B⊂AB \subset A is open (resp. closed) in A and saturated with respect to RAR_A, then B is open (resp. closed) in X and saturated with respect to R.

Corollary 2

If there is a continuous mapping u:X→Au : X \to A such that u(x)u(x) is congruent to xx mod R for each x∈Xx \in X, then f(A)=X/Rf(A) = X/R and the canonical mapping h:A/RA→X/Rh : A/R_A \to X/R is a homeomorphism.

Since each equivalence class mod R meets A, the canonical image of A/RAA/R_A in X/RX/R is the whole of X/RX/R; on the other hand, if U is open in A and is saturated with respect to RAR_A, it follows from the hypothesis that uˉ1(U)\bar{u}^1(U) is the set obtained by saturating U with respect to R; since uu is continuous, uˉ1(U)\bar{u}^1(U) is open in X (§ 2, no. 1, Theorem 1). The corollary follows from this fact by virtue of Proposition 10.

Example

Let R denote the equivalence relation x≡y(mod1)x \equiv y \pmod{1} on the real line R\mathbf{R} (no. 4, Example) and let A denote the closed interval [0,1][0, 1]; A contains at least one point of each equivalence class mod R. The canonical mapping of A/RAA/R_A onto the torus T is a homeomorphism; for if F is closed in A (and hence in R), in order to saturate F with respect to the relation R we have to take the union of the closed sets F+nF + n (for all n∈Zn \in \mathbf{Z}), which evidently form a locally finite family, so that their union is closed (§ 1, no. 5, Proposition 4); the assertion follows from this. We remark that A/RAA/R_A is obtained by identifying the points 0 and 1 in A. \

Exercises

See the exercises for § 3.

Transcribed from the printed volume. Not translated.