Bourbaki

2. Continuous functions

1. Continuous functions

Definition 1

A mapping ff of a topological space XX into a topological space X′X' is said to be continuous at a point x0∈Xx_0 \in X if, given any neighbourhood V′V' of f(x0)f(x_0) in X′X', there is a neighbourhood VV of x0x_0 in XX such that the relation x∈Vx \in V implies f(x)∈V′f(x) \in V'.

Definition 1 may be restated in the following more intuitive form: to say that ff is continuous at the point x0x_0 means that f(x)f(x) is as near as we please to f(x0)f(x_0) whenever xx is sufficiently near x0x_0.

The relation "for each x∈V,f(x)∈V′x \in V, f(x) \in V'" is equivalent to f(V)⊂V′f(V) \subset V' or again to V⊂f‾−1(V′)V \subset \overline{f}^{-1}(V'); in view of the neighbourhood axiom (VIV_I), we see that Definition 1 is equivalent to the following: f:X→X′f : X \to X' is said to be continuous at the point x0x_0 if, for each neighbourhood V′V' of f(x0)f(x_0) in X′X', f‾−1(V′)\overline{f}^{-1}(V') is a neighbourhood of x0x_0 in XX. Moreover, it is sufficient that f‾−1(V′)\overline{f}^{-1}(V') is a neighbourhood of x0x_0 for each neighbourhood V′V' belonging to a fundamental system of neighbourhoods of f(x0)f(x_0) in X′X' (§1§ 1, no. 3).

Proposition 1

Let ff be a mapping of a topological space XX into a topological space X′X'. If ff is continuous at xx, and if xx lies in the closure of a subset AA of XX, then f(x)f(x) lies in the closure of f(A)f(A).

Let V′V' be a neighbourhood of f(x)f(x) in X′X'. Since ff is continuous at xx, f‾−1(V′)\overline{f}^{-1}(V') is a neighbourhood of xx in XX. Hence f‾−1(V′)\overline{f}^{-1}(V') meets AA, from which it follows that V′V' meets f(A)f(A), and therefore f(x)f(x) is in the closure of f(A)f(A).

Proposition 2

Let X,X′,X′′X, X', X'' be three topological spaces; let ff be a mapping of XX into X′X', continuous at x∈Xx \in X; let gg be a mapping of X′X' into X′′X'', continuous at f(x)f(x). Then the composition h=g∘f:X→X′′h = g \circ f : X \to X'' is continuous at xx.

Let V′′V'' be a neighbourhood of h(x)=g(f(x))h(x) = g(f(x)) in X′′X''. Since gg is continuous at f(x)f(x) it follows that g‾−1(V′′)\overline{g}^{-1}(V'') is a neighbourhood of f(x)f(x) in X′X'. But ff is continuous at xx, hence f‾−1(g‾−1(V′′))=h‾−1(V′′)\overline{f}^{-1}(\overline{g}^{-1}(V'')) = \overline{h}^{-1}(V'') is a neighbourhood of xx in XX, and therefore hh is continuous at xx.

Definition 2

A mapping of a topological space XX into a topological space X′X' is said to be continuous on XX (or just continuous) if it is continuous at each point of XX.

Example 1

The identity mapping of a topological space XX onto itself is continuous.

Example 2

A constant map of a topological space into a topological space is continuous.

Example 3

Every mapping of a discrete space into a topological space is continuous.

Theorem 1

Let ff be a mapping of a topological space XX into a topological space X′X'. Then the following statements are equivalent:

a) ff is continuous in XX.

b) For every subset AA of XX, f(A‾)⊂f(A)‾f(\overline{A}) \subset \overline{f(A)}.

c) The inverse image under ff of every closed subset of X′X' is a closed subset of XX.

d) The inverse image under ff of every open subset of X′X' is an open subset of XX.

We have already seen that a) implies b) (Proposition 1). To show that b) implies c), let F′F' be a closed subset of X′X' and let F=f−1(F′)‾F = \overline{f^{-1}(F')}; then by hypothesis f(F‾)⊂f(F)‾⊂F′‾=F′f(\overline{F}) \subset \overline{f(F)} \subset \overline{F'} = F', hence F‾⊂f−1(F′)‾=F⊂F‾\overline{F} \subset \overline{f^{-1}(F')} = F \subset \overline{F}, so that F=F‾F = \overline{F} and FF is closed. By virtue of the relation f−1(A′)‾=f−1(CA′)‾\overline{f^{-1}(A')} = \overline{f^{-1}(CA')} for every subset A′A' of X′X', c) implies d). Finally, suppose that d) is satisfied. Let xx be any point of XX and let V′V' be any neighbourhood of f(x)f(x) in X′X'; then there is an open set A′A' in X′X' for which

f(x)∈A′⊂V′ f(x) \in A' \subset V'

and hence x∈f−1(A′)‾⊂f−1(V′)‾x \in \overline{f^{-1}(A')} \subset \overline{f^{-1}(V')}. By hypothesis, f−1(A′)‾\overline{f^{-1}(A')} is open in XX, so that f−1(V′)‾\overline{f^{-1}(V')} is a neighbourhood of xx in XX. Thus d) implies a).

Remarks. 1) Let B\mathfrak{B} be a base (§ 1, no. 3) of the topology of X′X'; then for f:X→X′f : X \to X' to be continuous, it is necessary and sufficient that f−1(U′)‾\overline{f^{-1}(U')} is open in XX for every U′∈BU' \in \mathfrak{B}.

Example

Let aa be any rational number. The mapping x→a+xx \to a + x of the rational line Q\mathbf{Q} into itself is continuous on Q\mathbf{Q}, for the inverse image under this mapping of an open interval ]b,c[]b, c[ is the open interval

]b−a,c−a[. ]b - a, c - a[.

Likewise, the mapping x→axx \to ax is continuous on Q\mathbf{Q}; this is clear if a=0a = 0, for then ax=0ax = 0 for all xx; if a≠0a \neq 0 then the inverse image under this mapping of the open interval ]b,c[]b, c[ is the open interval with end-points b/ab/a and c/ac/a.

2) The direct image of an open (resp. closed) set of XX under a continuous mapping f:X→X′f : X \to X' is not necessarily open (resp. closed) in X′X' (cf. § 5).

Example

\* The mapping f:x→1/(1+x2)f : x \to 1/(1 + x^2) of R\mathbf{R} into itself is continuous, but f(R)f(\mathbf{R}) is the half-open interval ]0,1]]0, 1], which is neither open nor closed in R\mathbf{R}. \

Theorem 2

1) If f:X→X′f : X \to X' and g:X′→X′′g : X' \to X'' are two continuous mappings, then g∘f:X→X′′g \circ f : X \to X'' is continuous.

2) For a bijection ff of a topological space XX onto a topological space X′X' to be a homeomorphism, it is necessary and sufficient that ff and the inverse of ff are continuous (or, as is also said, that ff is bicontinuous).

The first assertion is an immediate consequence of Proposition 2; the second follows from Theorem 1, d) and the definition of a homeomorphism (§ 1, no. 1).

Remark 1

It is possible to have a continuous bijection of a topological space XX onto a topological space X′X' which is not bicontinuous: for example, take X′X' to be the rational line Q\mathbf{Q}, and XX to be the set Q\mathbf{Q} with the discrete topology; then the identity map X→X′X \to X' is continuous but is not a homeomorphism.

Remark 2

To verify that a continuous bijection f:X→X′f : X \to X' is a homeomorphism, it is enough to show that for each x∈Xx \in X and each neighbourhood VV of xx, f(V)f(V) is a neighbourhood of f(x)f(x) in X′X'.

Remark 3

Let XX be a topological space, and for each x∈Xx \in X let B(x)\mathfrak{B}(x) be the set of all neighbourhoods of xx. Let x0x_0 be a point of XX; for each x∈Xx \in X, define a set B0(x)\mathfrak{B}_0(x) of subsets of XX as follows: B0(x0)=B(x0)\mathfrak{B}_0(x_0) = \mathfrak{B}(x_0), and if x≠x0x \neq x_0 then B0(x)\mathfrak{B}_0(x) is to be the set of all subsets of XX which contain xx. It is immediately verified (§ 1, no. 2, Proposition 2) that the sets B0(x)\mathfrak{B}_0(x) are sets of neighbourhoods of points of XX for a topology on XX; let X0X_0 denote the topological space thus obtained, and let j:X0→Xj : X_0 \to X denote the identity map, which is continuous but not in general bicontinuous. A mapping ff of XX into a topological space X′X' is continuous at the point x0x_0 if and only if the composition X0→jX→fX′X_0 \xrightarrow{j} X \xrightarrow{f} X' is continuous on X0X_0; this follows immediately from the definitions.

2. Comparison of topologies

Theorem 2 of no. 1 shows that we may take the continuous mappings as morphisms of topological structures (Set Theory, Chapter IV, § 2, no. 1); from now on, we shall assume that we have made this choice of morphisms. In accordance with the general definitions (Set Theory, Chapter IV, § 2, no. 2), this allows us to define an ordering on the set of topologies on a given set XX:

Definition 3

Given two topologies T1,T2\mathfrak{T}_1, \mathfrak{T}_2 on the same set XX, we say that T1\mathfrak{T}_1 is finer than T2\mathfrak{T}_2 (and that T2\mathfrak{T}_2 is coarser than T1\mathfrak{T}_1) if, denoting by XiX_i the set XX with the topology Ti\mathfrak{T}_i (i=1,2i = 1, 2), the identity mapping X1→X2X_1 \to X_2 is continuous. If in addition T1≠T2\mathfrak{T}_1 \neq \mathfrak{T}_2, we say that T1\mathfrak{T}_1 is strictly finer than T2\mathfrak{T}_2 (and that T2\mathfrak{T}_2 is strictly coarser than T1\mathfrak{T}_1).

Two topologies, one of which is finer than the other, are said to be comparable.

The criteria for a mapping to be continuous (no. 1, Definition 1 and Theorem 1) give the following proposition:

Proposition 3

Given two topologies T1,T2\mathfrak{T}_1, \mathfrak{T}_2 on a set XX, the following statements are equivalent: a) T1\mathfrak{T}_1 is finer than T2\mathfrak{T}_2. b) For each x∈Xx \in X, each neighbourhood of xx for T2\mathfrak{T}_2 is a neighbourhood of xx for T1\mathfrak{T}_1. c) For each subset AA of XX, the closure of AA in the topology T1\mathfrak{T}_1 is contained in the closure of AA in the topology T2\mathfrak{T}_2. d) Every subset of XX which is closed in T2\mathfrak{T}_2 is closed in T1\mathfrak{T}_1. e) Every subset of XX which is open in T2\mathfrak{T}_2 is open in T1\mathfrak{T}_1.

Example

\* In Hilbert space HH consisting of sequences x=(xn)x = (x_n) of real numbers such that

∥x∥2=∑n=0∞xn2<+∞, \|x\|^2 = \sum_{n=0}^\infty x_n^2 < +\infty,
the neighbourhoods of a point x0x_0 in the strong topology on HH are the sets which contain an open ball ∥x−x0∥<α\|x - x_0\| < \alpha centred at x0x_0; the neighbourhoods of x0x_0 in the weak topology on HH are the sets containing a set defined by a relation of the form sup⁡1≤i≤n∣(x−x0∣ai)∣≤1\sup_{1 \leq i \leq n} |(x - x_0|a_i)| \leq 1, where the aia_i are points of HH and
(x∣y)=∑n=0∞xnyn (x|y) = \sum_{n=0}^\infty x_n y_n
if x=(xn)x = (x_n) and y=(yn)y = (y_n). Now if β=sup⁡1≤i≤n∥ai∥\beta = \sup_{1 \leq i \leq n} \|a_i\|, the relation ∥x−x0∥≤1/β\|x - x_0\| \leq 1 / \beta implies that ∣(x−x0∣ai)∣≤∥x−x0∥⋅∥ai∥≤1|(x - x_0|a_i)| \leq \|x - x_0\| \cdot \|a_i\| \leq 1 for 1≤i≤n1 \leq i \leq n; hence the strong topology on HH is finer than the weak topology. On the other hand, given any finite family (ai)1≤i≤n(a_i)_{1 \leq i \leq n} of points of HH, there are points xx in HH such that (x−x0∣ai)=0(x - x_0|a_i) = 0 for 1≤i≤n1 \leq i \leq n and such that ∥x−x0∥\|x - x_0\| is arbitrarily large; this shows that the strong topology is strictly finer than the weak topology. \

Remark 1

In the ordered set of all topologies on a set XX, the topology in which the only open sets are ∅\varnothing and XX is the coarsest and the discrete topology is the finest.

Remark 2

The finer the topology, the more open sets, closed sets and neighbourhoods; the finer the topology, the smaller (resp. the larger) the closure (resp. the interior) of a set; the finer the topology, the fewer dense sets.

Remark 3

If f:X→X′f : X \to X' is a continuous mapping, it remains continuous if the topology of XX is replaced by a finer topology and the topology of X′X' is replaced by a coarser topology (no. 1, Theorem 2). In other words, the finer the topology of XX and the coarser the topology of X′X', the more continuous mappings there are of XX into X′X'.

3. Initial topologies

Proposition 4

Let XX be a set, let (Yi)i∈I(Y_i)_{i \in I} be a family of topological spaces, and for each i∈Ii \in I let fif_i be a mapping of XX into YiY_i. Let S\mathscr{S} be the set of subsets of XX of the form f‾i−1(Ui)\overline{f}_i^{-1}(U_i) (i∈Ii \in I, UiU_i open in YiY_i), and let B\mathfrak{B} be the set of finite intersections of sets of S\mathscr{S}. Then B\mathfrak{B} is a base of a topology T\mathfrak{T} on XX which is the initial topological structure on XX for the family (fi)(f_i) (Set Theory, Chapter IV, § 2, no. 3) and in particular is the coarsest topology on XX for which the mappings fif_i are continuous. More precisely, if gg is a mapping of a topological space ZZ into XX, then gg is continuous at a point z∈Zz \in Z (X carrying the topology T\mathfrak{T}) if and only if each of the functions fi∘gf_i \circ g is continuous at zz.

Let D\mathscr{D} be the set of all unions of sets belonging to B\mathfrak{B}; clearly D\mathscr{D} satisfies axiom (O1)(O_1) since formation of unions is associative; and D\mathscr{D} satisfies axiom (OII)(O_{II}) by reason of the definition of B\mathfrak{B} and the fact that finite intersection is distributive over arbitrary union [Set Theory, R § 4, formula (37)]. The set D\mathscr{D} is therefore the set of open subsets of XX for a topology T\mathfrak{T} of which B\mathfrak{B} is a base. We shall prove the last assertion of the proposition, which implies the others by reason of the general properties of initial structures (Set Theory, Chapter IV, § 2, no. 3, criterion CST 9). In the first place, the definition of S\mathscr{S} shows that the fif_i are continuous on XX (no. 1, Theorem 1); hence, if gg is continuous at zz, so are the mappings fi∘gf_i \circ g (no. 1, Proposition 2). Conversely, suppose that all the mappings fi∘gf_i \circ g are continuous at zz, and let VV be a neighbourhood of g(z)g(z) in XX; by definition, there is a finite subset JJ of II, and for each i∈Ji \in J an open subset UiU_i of YiY_i such that VV contains the set ⋂i∈Jf‾i−1(Ui)\bigcap_{i \in J} \overline{f}_i^{-1}(U_i) and g(z)g(z) belongs to this set. It follows that

g‾−1(V)⊃⋂i∈Jg‾−1(f‾i−1(Ui)), \overline{g}^{-1}(V) \supset \bigcap_{i \in J} \overline{g}^{-1}(\overline{f}_i^{-1}(U_i)),
and the hypothesis implies that each of the sets g‾−1(f‾i−1(Ui))\overline{g}^{-1}(\overline{f}_i^{-1}(U_i)) is a neighbourhood of zz in ZZ; hence g‾−1(V)\overline{g}^{-1}(V) is also a neighbourhood of zz in ZZ. This completes the proof.

Let Bi\mathfrak{B}_i be a base of the topology of Yi(i∈I)Y_i (i \in I); let S′\mathscr{S}' denote the set of subsets of XX of the form f‾i−1(Ui)\overline{f}_i^{-1}(U_i) for i∈Ii \in I and Ui∈BiU_i \in \mathfrak{B}_i for each i∈Ii \in I; if B′\mathfrak{B}' is the set of finite intersections of sets of S′\mathscr{S}', it is evident that B′\mathfrak{B}' is a base of the topology T\mathfrak{T}.

The general properties of initial structures (Set Theory, Chapter IV, § 2, no. 3, criterion CST 10) imply in particular the following transitivity property (the direct proof of which is quite straightforward):

Proposition 5

Let XX be a set, (Zi)i∈I(Z_i)_{i \in I} a family of topological spaces, (Jλ)λ∈L(J_\lambda)_{\lambda \in L} a partition of II and (Yλ)λ∈L(Y_\lambda)_{\lambda \in L} a family of sets indexed by LL. Also for each λ∈L\lambda \in L let hλh_\lambda be a mapping of XX into YλY_\lambda; for each λ∈L\lambda \in L and each i∈Jλi \in J_\lambda let giλg_{i\lambda} be a mapping of YλY_\lambda into ZiZ_i, and put fi=giλ∘hλf_i = g_{i\lambda} \circ h_\lambda. If each YλY_\lambda carries the coarsest topology for which the mappings giλ(i∈Jλ)g_{i\lambda} (i \in J_\lambda) are continuous, then the coarsest topology on XX for which the fif_i are continuous is the same as the coarsest topology for which the hλh_\lambda are continuous.

Example 1

Inverse image of a topology. Let XX be a set, YY a topological space, ff a mapping of XX into YY; the coarsest topology T\mathfrak{T} on XX for which ff is continuous is called the inverse image under ff of the topology of YY. It follows from Proposition 4 and the formulae for the inverse image of a union and an intersection [Set Theory, R, § 4, formulae (34) and (46)] that the open (resp. closed) sets in the topology T\mathfrak{T} are the inverse images under ff of the open (resp. closed) sets of YY; consequently, for each x∈Xx \in X, the sets f−1(W)f^{-1}(W), where WW runs through a fundamental system of neighbourhoods of f(x)f(x) in YY, form a fundamental system of neighbourhoods of xx in the topology T\mathfrak{T}. In § 3 we shall study, under the name of induced topology, the particular case in which XX is a subset of YY and ff is the canonical injection X→YX \to Y; XX, with the induced topology, is then called a subspace of YY.

For a mapping ff of a topological space XX into a topological space X′X' to be continuous it is necessary and sufficient that the topology of XX is finer than the inverse image under ff of the topology of X′X'.

Example 2

Least upper bound of a set of topologies. Every family (Ti)i∈I(\mathfrak{T}_i)_{i \in I} of topologies on a set XX has a least upper bound T\mathfrak{T} in the ordered set of all topologies on XX, i.e. there exists a topology on XX which is coarsest among all the topologies on XX which are finer than each of the Ti\mathfrak{T}_i. To see this we may apply Proposition 4, taking YiY_i to be the set XX with the topology Ti\mathfrak{T}_i, and fif_i to be the identity map X→YiX \to Y_i; T\mathfrak{T} is the coarsest topology for which all the mappings fif_i are continuous.

Let S\mathfrak{S} be an arbitrary set of subsets of a set XX; amongst the topologies T\mathfrak{T} on XX for which the sets of S\mathfrak{S} are open, there is a topology T0\mathfrak{T}_0 which is coarser than all the others and which is called the topology generated by S\mathfrak{S}. For each set U∈SU \in \mathfrak{S} let TU\mathfrak{T}_U be the topology whose open sets are ∅,U\varnothing, U and XX [it is clear that this set of subsets of XX satisfies (O_I) and (O_{II})]; then T0\mathfrak{T}_0 is just the least upper bound of the topologies TU\mathfrak{T}_U. By Proposition 4, if B\mathfrak{B} is the set of finite intersections of sets belonging to S\mathfrak{S}, then B\mathfrak{B} is a base of the topology T0\mathfrak{T}_0. We say that S\mathfrak{S} is a subbase of T0\mathfrak{T}_0.

Example 3

Product topology. Let (Xi)i∈I(X_i)_{i \in I} be a family of topological spaces. The coarsest topology on the product set X=∏i∈IXiX = \prod_{i \in I} X_i for which the projections pri:X→Xi\mathrm{pr}_i : X \to X_i are continuous mappings is called the product of the topologies of the XiX_i; we shall study it in more detail in

4. Final topologies

Proposition 6

Let XX be a set, let (Yi)i∈I(Y_i)_{i \in I} be a family of topological spaces, and for each i∈Ii \in I let fif_i be a mapping of YiY_i into XX. Let D\mathscr{D} be the set of subsets UU of XX such that f‾i−1(U)\overline{f}_i^{-1}(U) is open in YiY_i for each i∈Ii \in I; then D\mathscr{D} is the set of open subsets of XX in a topology T\mathfrak{T} on XX which is the final structure on XX for the family (fi)(f_i) (Set Theory, Chapter IV, § 2, no. 5), and in particular T\mathfrak{T} is the finest topology on XX for which the mappings fif_i are continuous. In other words, if gg is a mapping of XX into a topological space ZZ, then gg is continuous (XX carrying the topology T\mathfrak{T}) if and only if each of the mappings g∘fig \circ f_i is continuous.

It is immediately verified that D\mathscr{D} satisfies the axioms (OI)(O_I) and (OII)(O_{II}). [Set Theory, R, § 4, formulae (34) and (46)]. We shall prove the last assertion of the proposition, which implies the other assertions by reason of the general properties of final structures (Set Theory, Chapter IV, § 2, no. 5, criterion CST 18). It is clear that the fif_i are continuous in the topology T\mathfrak{T}, by the definition of D\mathscr{D} (no. 1, Theorem 1); hence if gg is continuous, so is each mapping g∘fig \circ f_i (no. 1, Theorem 2). Conversely, suppose that each g∘fig \circ f_i is continuous, and let VV be an open set in ZZ; by hypothesis, f‾i−1(g‾−1(V))\overline{f}_i^{-1}(\overline{g}^{-1}(V)) is open in YiY_i for each i∈Ii \in I; hence g‾−1(V)∈D\overline{g}^{-1}(V) \in \mathscr{D}, and the proof is complete.

Corollary

Under the hypotheses of Proposition 6, a subset FF of XX is closed in the topology T\mathfrak{T} if and only if f‾i−1(F)\overline{f}_i^{-1}(F) is closed in YiY_i for each i∈Ii \in I.

This follows from the definition of the open sets in the topology T\mathfrak{T} by taking complements.

The general properties of final structures (Set Theory, Chapter IV, § 2, no. 5, criterion CST 19) imply the following transitivity property (the direct proof of which is also straightforward):

Proposition 7

Let XX be a set, (Zi)i∈I(Z_i)_{i \in I} a family of topological spaces, (Jλ)λ∈L(J_\lambda)_{\lambda \in L} a partition of II, and (Yλ)λ∈L(Y_\lambda)_{\lambda \in L} a family of sets indexed by LL. Also, for each λ∈L\lambda \in L let hλh_\lambda be a mapping of YλY_\lambda into XX; for each λ∈L\lambda \in L and each i∈Jλi \in J_\lambda let gλig_{\lambda i} be a mapping of ZiZ_i into YλY_\lambda, and put fi=hλ∘gλif_i = h_\lambda \circ g_{\lambda i}. If each YλY_\lambda carries the finest topology for which the mappings gλi(i∈Jλ)g_{\lambda i} (i \in J_\lambda) are continuous, then the finest topology on XX for which the fif_i are continuous is the same as the finest topology for which the hλh_\lambda are continuous.

Examples

1) Quotient topology. Let XX be a topological space, RR an equivalence relation on XX, Y=X/RY = X/R the quotient set of XX with respect to the relation RR, φ:X→Y\varphi : X \to Y the canonical mapping. The finest topology on YY for which φ\varphi is continuous is called the quotient of the topology of XX by the relation RR; we shall study it in more detail in § 3.

2) Greatest lower bound of a set of topologies. Every family (Ti)i∈I(\mathfrak{T}_i)_{i \in I} of topologies on a set XX has a greatest lower bound T\mathfrak{T} in the set of all topologies on XX, i.e. T\mathfrak{T} is the finest of all the topologies on XX which are coarser than each of the Ti\mathfrak{T}_i. To see this we may apply Proposition 6, taking YiY_i to be the set XX with the topology Ti\mathfrak{T}_i, and fif_i to be the identity mapping Yi→XY_i \to X. If Di\mathscr{D}_i is the set of subsets of XX which are open in the topology Ti\mathfrak{T}_i, then the set ⋂i∈IDi\bigcap_{i \in I} \mathscr{D}_i is the set of subsets of XX which are open in T\mathfrak{T}. T\mathfrak{T} is also called the intersection of the topologies Ti\mathfrak{T}_i.

3) Sum of topological spaces. Let (Xi)i∈I(X_i)_{i \in I} be a family of topological spaces, XX the set which is the sum of the XiX_i (Set Theory, Chapter II, § 4, no. 8, Definition 8); for each i∈Ii \in I, let jij_i be the canonical (injective) mapping of XiX_i into XX. The finest topology T\mathfrak{T} on XX for which the mappings jij_i are continuous is called the sum of the topologies of the XiX_i, and XX with this topology is said to be the sum of the topological spaces XiX_i. Let us identify each of the XiX_i with a subset of XX by means of jij_i; then a set A⊂XA \subset X is open (resp. closed) in the topology T\mathfrak{T} if and only if each of the sets A∩XiA \cap X_i is open (resp. closed) in XiX_i. In particular, each of the XiX_i is both open and closed.

The following proposition generalizes the situation of Example 3:

Proposition 8

Let XX be a set, (Xλ)λ∈L(X_\lambda)_{\lambda \in L} a family of subsets of XX. Suppose each XλX_\lambda carries a topology Tλ\mathfrak{T}_\lambda such that, for each pair of indices (λ,μ)(\lambda, \mu): 1) Xλ∩XμX_\lambda \cap X_\mu is open (resp. closed) in each of the topologies Tλ,Tμ\mathfrak{T}_\lambda, \mathfrak{T}_\mu. 2) The topologies induced on Xλ∩XμX_\lambda \cap X_\mu by Tλ\mathfrak{T}_\lambda and Tμ\mathfrak{T}_\mu coincide. Let T\mathfrak{T} be the finest topology on XX for which the injections jλ:Xλ→Xj_\lambda : X_\lambda \to X are continuous. Then, for each λ∈L\lambda \in L, XλX_\lambda is open (resp. closed) in XX in the topology T\mathfrak{T}, and the topology induced by T\mathfrak{T} on XλX_\lambda coincides with Tλ\mathfrak{T}_\lambda.

In view of Proposition 6 and its corollary, it is enough to show that for each λ\lambda and each subset AλA_\lambda of XλX_\lambda the following statements are equivalent: (i) AλA_\lambda is open (resp. closed) in the topology Tλ\mathfrak{T}_\lambda. (ii) For each μ∈L\mu \in L, Aλ∩XμA_\lambda \cap X_\mu is open (resp. closed) in the topology Tμ\mathfrak{T}_\mu.

It is clear that (ii) implies (i) by taking μ=λ\mu = \lambda. Conversely, if (i) is satisfied, then Aλ∩XμA_\lambda \cap X_\mu is open (resp. closed) in Xλ∩XμX_\lambda \cap X_\mu for the topology Tλμ\mathfrak{T}_{\lambda \mu} induced on Xλ∩XμX_{\lambda} \cap X_{\mu} by Tλ\mathfrak{T}_{\lambda}; but Tλμ\mathfrak{T}_{\lambda \mu} is also the topology induced on Xλ∩XμX_{\lambda} \cap X_{\mu} by Tμ\mathfrak{T}_{\mu}; hence Aλ∩XμA_{\lambda} \cap X_{\mu} is also the intersection of Xλ∩XμX_{\lambda} \cap X_{\mu} with a subset BμB_{\mu} of XμX_{\mu} which is open (resp. closed) in the topology Tμ\mathfrak{T}_{\mu}; since Xλ∩XμX_{\lambda} \cap X_{\mu} is open (resp. closed) in Tμ\mathfrak{T}_{\mu}, so is Aλ∩XμA_{\lambda} \cap X_{\mu}. This completes the proof.

We remark that if the union of the XλX_{\lambda} is different from XX, then the topology induced by T\mathfrak{T} on X−(⋃λ∈LXλ)X - \left( \bigcup_{\lambda \in L} X_{\lambda} \right) is discrete. For if x∈Xx \in X belongs to none of the XλX_{\lambda}, then {x}∩Xλ=∅\{ x \} \cap X_{\lambda} = \varnothing is open in each topology Tλ\mathfrak{T}_{\lambda} and therefore {x}\{ x \} is open in the topology T\mathfrak{T}.

5. Pasting together of topological spaces

Let (Xλ)λ∈L(X_{\lambda})_{\lambda \in L} be a family of sets, and let XX be the set which is the sum of the XλX_{\lambda} (Set Theory, Chapter II, § 4, no. 8, Definition 8); we shall identify each XλX_{\lambda} with a subset of XX by means of the canonical injection jλ:Xλ→Xj_{\lambda} : X_{\lambda} \to X.

Let RR be an equivalence relation on XX such that each equivalence class of RR has at most one element in each XλX_{\lambda}; for each pair of indices (λ,μ)(\lambda, \mu) let AλμA_{\lambda \mu} be the subset of XλX_{\lambda} consisting of the elements xx for which there is an element y∈Xμy \in X_{\mu} which belongs to the equivalence class of XX. Clearly to each x∈Aλμx \in A_{\lambda \mu} corresponds a unique y∈Aμλy \in A_{\mu \lambda} which is congruent to xx mod RR; the mappings hμλ:Aλμ→Aμλh_{\mu \lambda} : A_{\lambda \mu} \to A_{\mu \lambda} so defined satisfy the following conditions:

(i) For each λ∈L\lambda \in L, hλλh_{\lambda \lambda} is the identity mapping of Aλλ=XλA_{\lambda \lambda} = X_{\lambda}.

(ii) For each triple of indices (λ,μ,ν)(\lambda, \mu, \nu) of LL and each x∈Aλμ∩Aλνx \in A_{\lambda \mu} \cap A_{\lambda \nu}, we have hμλ(x)∈Aμνh_{\mu \lambda}(x) \in A_{\mu \nu} and

hνλ(x)=hνμ(hμλ(x)). h_{\nu \lambda}(x) = h_{\nu \mu}(h_{\mu \lambda}(x)).

Conversely, suppose that for each pair of indices (λ,μ)(\lambda, \mu) we are given a subset AλμA_{\lambda \mu} of XλX_{\lambda} and a mapping hμλ:Aλμ→Aμλh_{\mu \lambda} : A_{\lambda \mu} \to A_{\mu \lambda} satisfying the conditions (i) and (ii) above. It follows first of all from (ii) applied to the triples (λ,μ,λ)(\lambda, \mu, \lambda) and (μ,λ,μ)(\mu, \lambda, \mu) that hλμ∘hμλh_{\lambda \mu} \circ h_{\mu \lambda} (resp. hμλ∘hλμh_{\mu \lambda} \circ h_{\lambda \mu}) is the restriction of hλλh_{\lambda \lambda} (resp. hμμh_{\mu \mu}) to AλμA_{\lambda \mu} (resp. AμλA_{\mu \lambda}); hence we deduce from (i) that hλμh_{\lambda \mu} and hμλh_{\mu \lambda} are bijections which are inverses of each other. Now let R{x,y}R \{ x, y \} be the relation "there exist λ,μ\lambda, \mu such that x∈Aλμx \in A_{\lambda \mu}, y∈Aμλy \in A_{\mu \lambda} and y=hμλ(x)y = h_{\mu \lambda}(x)". It follows from (i) and from what precedes that RR is reflexive and symmetric; on the other hand, if x∈Aλμx \in A_{\lambda \mu},

y=hμλ(x)∈Aμλ∩Aμνandz=hνμ(y), y = h_{\mu \lambda}(x) \in A_{\mu \lambda} \cap A_{\mu \nu} \quad \text{and} \quad z = h_{\nu \mu}(y),

then also x=hλμ(y)x = h_{\lambda \mu}(y) and therefore, by (ii), x∈Aλμ∩Aλνx \in A_{\lambda \mu} \cap A_{\lambda \nu}; the relation (i) thus shows that RR is transitive, and therefore RR is an equivalence relation on XX. It follows also from (i) and from the definition of RR that each equivalence class mod RR has at most one element in each of the sets XλX_\lambda, and that AλμA_{\lambda \mu} is the set of all x∈Xλx \in X_\lambda for which there is an element y∈Xμy \in X_\mu congruent to xx mod RR. We say that the quotient set X/RX/R is obtained by pasting together the XλX_\lambda along the AλμA_{\lambda \mu} by means of the bijections hμλh_{\mu \lambda}. If φ:X→X/R\varphi : X \to X/R is the canonical mapping, the restriction of φ\varphi to each XλX_\lambda is a bijection of XλX_\lambda onto φ(Xλ)\varphi(X_\lambda).

Now suppose that each XλX_\lambda is a topological space, and let Tλ\mathfrak{T}_\lambda be its topology. Let T\mathfrak{T} be the finest topology on the set X/RX/R for which the mappings φ∘jλ\varphi \circ j_\lambda are continuous; T\mathfrak{T} is the quotient by RR of the topology on XX which is the sum of the topologies Tλ\mathfrak{T}_\lambda. We say that the topological space X/RX/R (with the topology T\mathfrak{T}) is obtained by pasting together the topological spaces XλX_\lambda along the AλμA_{\lambda \mu} by means of the bijections hμλh_{\mu \lambda}. The open (resp. closed) subsets of X/RX/R are thus the canonical images of the subsets BB of XX which are saturated with respect to RR and are such that B∩XλB \cap X_\lambda is open (resp. closed) in XλX_\lambda for each λ∈L\lambda \in L.

Since the restriction of φ\varphi to each XλX_\lambda is a bijection onto the subset Xλ′=φ(Xλ)X'_\lambda = \varphi(X_\lambda) of X/RX/R, we can transport the topology Tλ\mathfrak{T}_\lambda to Xλ′X'_\lambda by means of this bijection, so that Xλ′X'_\lambda carries a topology Tλ′\mathfrak{T}'_\lambda; and the topology T\mathfrak{T} on X/RX/R is the finest for which the canonical injections Xλ′→X/RX'_\lambda \to X/R are continuous. In general, the topology induced by T\mathfrak{T} on Xλ′X'_\lambda is coarser than Tλ′\mathfrak{T}'_\lambda, but not identical with the latter; even if the hμλh_{\mu \lambda} are homeomorphisms (§ 3, Exercise 15). However, it follows from no. 4, Proposition 8 that, with the preceding notation:

Proposition 9

Suppose that the* hμλh_{\mu \lambda} are homeomorphisms and that each AλμA_{\lambda \mu} is open (resp. closed) in XλX_\lambda; then each φ(Xλ)\varphi(X_\lambda) is open (resp. closed) in X/RX/R and the restriction of φ\varphi to XλX_\lambda is a homeomorphism of XλX_\lambda onto the subspace φ(Xλ)\varphi(X_\lambda) of X/RX/R.

Exercises

See the exercises for § 2.

Transcribed from the printed volume. Not translated.