Bourbaki

3. Special function spaces

1. Spaces of mappings into a metric space

Let XX be a set, YY a uniform space, (fi)i∈I(f_i)_{i \in I} a family of pseudometrics defining the uniform structure of YY (Chapter IX, § 1, no. 4), and let S\mathscr{S} be a set of subsets of XX. For each i∈Ii \in I, each set A∈SA \in \mathscr{S}, and each pair (u,v)(u, v) of mappings of XX into YY, write

gi,A(u,v)=sup⁡x∈Afi(u(x),v(x)); g_{i, A}(u, v) = \sup_{x \in A} f_i(u(x), v(x));

it follows immediately that gi,Ag_{i, A} is a pseudometric on F(X;Y)\mathscr{F}(X; Y) and that the family of pseudometrics (gi,A)i∈I,A∈S(g_{i, A})_{i \in I, A \in \mathscr{S}} defines the uniformity of S\mathscr{S}-*convergence* on F(X;Y)\mathscr{F}(X; Y). In particular:

Proposition 1

If YY is a metrizable uniform space, the uniformity of uniform convergence on F(X;Y)\mathscr{F}(X; Y) is metrizable.

For if dd is a metric on YY compatible with its uniform structure, the structure of uniform convergence on F(X;Y)\mathscr{F}(X; Y) is defined by the single pseudometric

ε(u,v)=sup⁡x∈Xd(u(x),v(x)); \varepsilon(u, v) = \sup_{x \in X} d(u(x), v(x));

in general this pseudometric is not finite, but it is equivalent to a finite one (Chapter IX, § 1, no. 2), and since the uniformity of uniform convergence is Hausdorff (§ 1, no. 2, Proposition 1), it is metrizable.

Corollary

Let XX be a topological space and let YY be a metrizable uniform space. Suppose that there is a sequence (Kn)(K_n) of compact subsets of XX such that every compact subset of XX is contained in some KnK_n. Then the uniformity of compact convergence on F(X;Y)\mathscr{F}(X; Y) is metrizable.

Since the KnK_n cover XX, Fc(X;Y)\mathscr{F}_c(X; Y) is isomorphic to a uniform subspace of ∏nFu(Kn;Y)\prod_n \mathscr{F}_u(K_n; Y) (§1§ 1, no. 2, Remark 3), and the corollary therefore follows from Proposition 1 (Chapter IX, § 2, no. 4, Theorem 1, Corollary 2).

Note that this corollary applies in particular if XX is locally compact and σ-compact (Chapter I, § 9, no. 9, Proposition 15, Corollary 1).

Now let YY be a metric space and let dd be its metric. If XX is any set and S\mathscr{S} any set of subsets of XX, we shall denote by BS(X;Y)\mathscr{B}_{\mathscr{S}}(X; Y) the set of all mappings u:X→Yu : X \to Y such that u(A)u(A) is bounded for each A∈SA \in \mathscr{S}. Unless the contrary is expressly stated we shall regard BS(X;Y)\mathscr{B}_{\mathscr{S}}(X; Y) as endowed with the uniformity of S\mathscr{S}-convergence, which is defined by the following family of pseudometrics on BS(X;Y)\mathscr{B}_{\mathscr{S}}(X; Y):

dA(u,v)=sup⁡x∈Ad(u(x),v(x))(A∈S) d_A(u, v) = \sup_{x \in A} d(u(x), v(x)) \tag{$A \in \mathscr{S}$}

which are finite by hypothesis. When S={X}\mathscr{S} = \{X\}, we write B(X;Y)\mathscr{B}(X; Y) in place of BS(X;Y)\mathscr{B}_{\mathscr{S}}(X; Y). A mapping u:X→Yu : X \to Y is said to be bounded if it belongs to B(X;Y)\mathscr{B}(X; Y), i.e. if u(X)u(X) is a bounded subset of YY.

Proposition 2

Let XX be a set and YY a metric space. The set B(X;Y)\mathscr{B}(X; Y) of bounded mappings is both open and closed in the space Fu(X;Y)\mathscr{F}_u(X; Y).

If uu is bounded, then every mapping v:X→Yv : X \to Y such that for all x∈Xx \in X, we have d(u(x),v(x))≤1d(u(x), v(x)) \leq 1 is bounded, because

d(v(x),v(x0))≤d(u(x),u(x0))+2; d(v(x), v(x_0)) \leq d(u(x), u(x_0)) + 2;

hence B(X;Y)\mathscr{B}(X; Y) is open. On the other hand, if uu lies in the closure of B(X;Y)\mathscr{B}(X; Y) in Fu(X;Y)\mathscr{F}_u(X; Y), there is a mapping u0∈B(X;Y)u_0 \in \mathscr{B}(X; Y) such that d(u(x),u0(x))≤1d(u(x), u_0(x)) \leq 1 for all x∈Xx \in X; hence uu is bounded.

Corollary 1

Let XX be a set and YY a metric space. Then BS(X;Y)\mathscr{B}_{\mathscr{S}}(X; Y) is closed in FS(X;Y)\mathscr{F}_{\mathscr{S}}(X; Y). In particular, if YY is complete then BS(X;Y)\mathscr{B}_{\mathscr{S}}(X; Y) is complete with respect to the uniformity of S\mathscr{S}-convergence.

For BS(X;Y)\mathscr{B}_\mathfrak{S}(X; Y) is the inverse image of the subset ∏A∈SB(A;Y)\prod_{A \in \mathfrak{S}} \mathscr{B}(A; Y) of the product ∏A∈SFu(X;Y)\prod_{A \in \mathfrak{S}} \mathscr{F}_u(X; Y) under the canonical mapping of FS(X;Y)\mathscr{F}_\mathfrak{S}(X; Y) into ∏A∈SFu(A;Y)\prod_{A \in \mathfrak{S}} \mathscr{F}_u(A; Y); the first assertion therefore follows from § 1, no. 2, Remark 3, and the second follows from the first, if we take account of Theorem 1 of § 1, no. 5.

Corollary 2

Let X be a topological space and Y a metric space. Then the space of all bounded continuous mappings of X into Y is both open and closed in Cu(X;Y)C_u(X; Y); it is complete if Y is complete.

The space in question is B(X;Y)∩Cu(X;Y)\mathscr{B}(X; Y) \cap C_u(X; Y); the first assertion follows from Proposition 2; the second follows from the first (§1§ 1, no. 6, Theorem 2, Corollary 1).

2. Spaces of mappings into a normed space

Consider, more particularly, the situation in which Y is a normed vector space over a non-discrete valued division ring K (Chapter IX, § 3, no. 3). Let us denote by ∣∣y∣∣||y|| the norm of y∈Yy \in Y. The set F(X;Y)=YX\mathscr{F}(X; Y) = Y^X is then canonically endowed with a K-vector space structure. A mapping u:X→Yu : X \to Y is bounded if and only if the real-valued function x↦∣∣u(x)∣∣x \mapsto ||u(x)|| is bounded in X. If u,vu, v are bounded mappings of X into Y, it is clear that u+vu + v and λu(λ∈K)\lambda u (\lambda \in K) are bounded; in other words, B(X;Y)\mathscr{B}(X; Y) is a vector subspace of F(X;Y)\mathscr{F}(X; Y). Moreover, ∣∣u∣∣=sup⁡x∈X∣∣u(x)∣∣||u|| = \sup_{x \in X} ||u(x)|| is a norm on B(X;Y)\mathscr{B}(X; Y); for it satisfies the triangle inequality and ∣∣u∣∣=0||u|| = 0 implies u=0u = 0, and for each λ∈K\lambda \in K we have

∣∣λu∣∣=sup⁡x∈X∣∣λu(x)∣∣=sup⁡x∈X∣λ∣⋅∣∣u(x)∣∣=∣λ∣⋅sup⁡x∈X∣∣u(x)∣∣=∣λ∣⋅∣∣u∣∣. ||\lambda u|| = \sup_{x \in X} ||\lambda u(x)|| = \sup_{x \in X} |\lambda| \cdot ||u(x)|| = |\lambda| \cdot \sup_{x \in X} ||u(x)|| = |\lambda| \cdot ||u||.

Moreover, it is immediately verified that the uniformity on B(X;Y)\mathscr{B}(X; Y) defined by this norm is the uniformity of uniform convergence. Unless the contrary is expressly stated, whenever B(X;Y)\mathscr{B}(X; Y) is considered as a normed space, it is the norm defined above which is in question.

Proposition 3

If the normed space Y is complete, then every series (un)(u_n) of bounded mappings of X into Y which is absolutely convergent in the normed space B(X;Y)\mathscr{B}(X; Y) (i.e. which is such that ∑n=0∞∣∣un∣∣<+∞\sum_{n=0}^\infty ||u_n|| < +\infty; cf. Chapter IX, § 3, no. 6) is uniformly convergent in X.

For since B(X;Y)\mathscr{B}(X; Y) is complete (no. 1, Proposition 2, Corollary 1), the result follows from Chapter IX, § 3, no. 6, Proposition 11 and the definition of a uniformly convergent series.

Remark 1

If ∑n=0∞∥un∥<+∞\sum_{n=0}^{\infty} \|u_n\| < +\infty, then ∑n=0∞∣un(x)∣≤∑n=0∞∣un∣<+∞\sum_{n=0}^{\infty} |u_n(x)| \leq \sum_{n=0}^{\infty} |u_n| < +\infty for each x∈Xx \in X; in other words, for each x∈Xx \in X the series with general term un(x)u_n(x) is absolutely convergent in the space YY. The converse is false. To avoid all confusion we shall sometimes say that the series with general term unu_n is normally convergent, meaning that the series with general term ∥un∥\|u_n\| is convergent. A series can be uniformly convergent in XX without being normally convergent; this is the case, for example, for the series (un)(u_n) in the space B(R,R)\mathscr{B}(\mathbf{R}, \mathbf{R}), defined as follows: un(x)=(1/n)sin⁡xu_n(x) = (1/n) \sin x if x∈[nπ,(n+1)π]x \in [n\pi, (n+1)\pi], un(x)=0u_n(x) = 0 otherwise.

When YY is a normed algebra (Chapter IX, § 3, no. 7) over a non-discrete valued field KK, then B(X;Y)\mathscr{B}(X; Y) is a KK-algebra, and the norm ∥u∥\|u\| is compatible with the algebra structure, since

∥uv∥=sup⁡x∈X∥u(x)v(x)∥≤sup⁡x∈X∥u(x)∥⋅∥v(x)∥≤sup⁡x∈X∥u(x)∥⋅sup⁡x∈X∥v(x)∥=∥u∥⋅∥v∥. \|uv\| = \sup_{x \in X} \|u(x)v(x)\| \leq \sup_{x \in X} \|u(x)\| \cdot \|v(x)\| \leq \sup_{x \in X} \|u(x)\| \cdot \sup_{x \in X} \|v(x)\| = \|u\| \cdot \|v\|.
Thus B(X;Y)\mathscr{B}(X; Y) is now a normed algebra over KK.

Proposition 4

Let XiX_i (1≤i≤n1 \leq i \leq n) and YY be normed vector spaces over a non-discrete valued division ring KK, and let X=∏i=1nXiX = \prod_{i=1}^n X_i. Then the set of all multilinear mappings of XX into YY is closed in the space Fs(X;Y)\mathscr{F}_s(X; Y).

This set consists of all u∈F(X;Y)u \in F(X; Y) which satisfy all the relations

u(x1,…,xi′+xi′′,…,xn)=u(x1,…,xi′,…,xn)+u(x1,…,xi′′,…,xn),u(x1,…,λxi,…,xn)=λu(x1,…,xi,…,xn) \begin{align*} u(x_1, \ldots, x_i' + x_i'', \ldots, x_n) &= u(x_1, \ldots, x_i', \ldots, x_n) \\ &\quad + u(x_1, \ldots, x_i'', \ldots, x_n), \\ u(x_1, \ldots, \lambda x_i, \ldots, x_n) &= \lambda u(x_1, \ldots, x_i, \ldots, x_n) \end{align*}
(1≤i≤n, xi, xi′, xi′′1 \leq i \leq n,\ x_i,\ x_i',\ x_i'' arbitrary elements of XiX_i, λ\lambda an arbitrary element of KK); since both sides of the relations (1) are continuous functions of uu on Fs(X;Y)\mathscr{F}_s(X; Y) (§ 1, no. 2, Remark 6), the result follows (Chapter I, § 8, no. 1, Proposition 2).

Proposition 5

Under the hypotheses of Proposition 4, the set L(X1,…,Xn;Y)\mathscr{L}(X_1, \ldots, X_n; Y) of continuous multilinear mappings of XX into YY is closed in F(X;Y)\mathscr{F}(X; Y) with respect to the topology of bounded convergence; it is complete with respect to the uniformity of bounded convergence if YY is complete.

For if G\mathscr{G} is the set of all bounded subsets of XX, L(X1,…,Xn;Y)\mathscr{L}(X_1, \ldots, X_n; Y) is the intersection of the set of all multilinear mappings of XX into YY and the set BG(X;Y)\mathscr{B}_\mathscr{G}(X; Y) (Chapter IX, § 3, no. 5, Theorem 1); the result thus follows from Proposition 4 and Proposition 2, Corollary 1.

For the remainder of this subsection, KK denotes a non-discrete valued field.

Then L(X1,…,Xn;Y)\mathscr{L}(X_1, \ldots, X_n; Y) is a vector subspace of F(X;Y)\mathscr{F}(X; Y). Let BB be the unit ball in XX, the set of all (xi)1≤i≤n(x_i)_{1 \leq i \leq n} such that sup⁡1≤i≤n∥xi∥≤1\sup_{1 \leq i \leq n} \|x_i\| \leq 1.

Then the mapping u→u∣Bu \to u|B of L(X1,…,Xn;Y)\mathscr{L}(X_1, \ldots, X_n; Y) into B(B;Y)\mathscr{B}(B; Y) is injective; moreover, the inverse image, under this mapping, of the uniformity of uniform convergence on B(B;Y)\mathscr{B}(B; Y) is the uniformity of bounded convergence on L(X1,…,Xn;Y)\mathscr{L}(X_1, \ldots, X_n; Y). For every bounded subset of XX is contained in a set of the form μB\mu B (for some μ∈K∗\mu \in K^*), and if uu is an element of L(X1,…,Xn;Y)\mathscr{L}(X_1, \ldots, X_n; Y), to say that ∥u(z)∥≤a\|u(z)\| \leq a for all z∈μBz \in \mu B is equivalent to saying that ∥u(z)∥≤a/∣μ∣n\|u(z)\| \leq a/|\mu|^n for all z∈Bz \in B. It is easily verified that the number

∥u∥=sup⁡z≠0∥u(z)∥∥z∥ \|u\| = \sup_{z \neq 0} \frac{\|u(z)\|}{\|z\|}
is a norm on L(X1,…,Xn;Y)\mathscr{L}(X_1, \ldots, X_n; Y) and defines the uniformity of bounded convergence on this set, and clearly we have
(2)∥u(x1,…,xn)∥≤∥u∥⋅∥x1∥⋯∥xn∥. (2) \quad \|u(x_1, \ldots, x_n)\| \leq \|u\| \cdot \|x_1\| \cdots \|x_n\|.

Unless the contrary is expressly stated, whenever L(X1,…,Xn;Y)\mathscr{L}(X_1, \ldots, X_n; Y) is considered as a normed space, it is the norm defined above which is in question.

Proposition 6

The multilinear mapping*

(u,x1,…,xn)→u(x1,…,xn) (u, x_1, \ldots, x_n) \to u(x_1, \ldots, x_n)
of the normed space L(X1,…,Xn;Y)×X1×⋯×Xn\mathscr{L}(X_1, \ldots, X_n; Y) \times X_1 \times \cdots \times X_n into YY is continuous.

This is an immediate consequence of the inequality (2) (Chapter IX, § 3, no. 5, Theorem 1).

Proposition 7

Let* X,Y,ZX, Y, Z be three normed spaces over KK. The canonical mapping of the normed space L(X,Y;Z)\mathscr{L}(X, Y; Z) into the space of linear mappings of XX into L(Y;Z)\mathscr{L}(Y; Z) which sends each u∈L(X,Y;Z)u \in \mathscr{L}(X, Y; Z) to the mapping x→u(x,.)x \to u(x, .) is an isometry of L(X;Y;Z)\mathscr{L}(X; Y; Z) onto L(X;L(Y;Z))\mathscr{L}(X; \mathscr{L}(Y; Z)).

This follows immediately from the definitions and the relation

sup⁡∥x∥≤1(sup⁡∥y∥≤1∥u(x,y)∥)=sup⁡∥x∥≤1,∥y∥≤1∥u(x,y)∥. \sup_{\|x\| \leq 1} \left( \sup_{\|y\| \leq 1} \|u(x, y)\| \right) = \sup_{\|x\| \leq 1, \|y\| \leq 1} \|u(x, y)\|.

Proposition 8

Let X,Y,ZX, Y, Z be three normed spaces over KK. The bilinear mapping (u,v)→v∘u(u, v) \to v \circ u of L(X;Y)×L(Y;Z)\mathscr{L}(X; Y) \times \mathscr{L}(Y; Z) into L(X;Z)\mathscr{L}(X; Z) is continuous.

For if u∈L(X;Y)u \in \mathscr{L}(X; Y) and v∈L(Y;Z)v \in \mathscr{L}(Y; Z) we have

∣∣v∘u∣∣≤∣∣u∣∣⋅∣∣v∣∣, ||v \circ u|| \leq ||u|| \cdot ||v||,
since for all x∈Xx \in X we have ∣∣v(u(x))∣∣≤∣∣v∣∣⋅∣∣u(x)∣∣≤∣∣v∣∣⋅∣∣u∣∣⋅∣∣x∣∣||v(u(x))|| \leq ||v|| \cdot ||u(x)|| \leq ||v|| \cdot ||u|| \cdot ||x|| by reason of (2).

In particular, on the set L(X)\mathscr{L}(X) of continuous endomorphisms of a normed space XX over KK, the norm ∣∣u∣∣||u|| is compatible with the KK-algebra structure of L(X)\mathscr{L}(X).

Remark 2

The set L(Rm;Rn)\mathscr{L}(\mathbf{R}^m; \mathbf{R}^n) of linear (necessarily continuous) mappings of Rm\mathbf{R}^m into Rn\mathbf{R}^n can be identified with the set Mn,m(R)M_{n,m}(\mathbf{R}) of matrices with nn rows and mm columns with coefficients in R\mathbf{R} and hence can be identified with Rmn\mathbf{R}^{mn}; on L(Rm;Rn)\mathscr{L}(\mathbf{R}^m; \mathbf{R}^n), the uniformity of bounded convergence (with respect to the Euclidean metric on Rm\mathbf{R}^m), of compact convergence, and of pointwise convergence are then identified with the additive uniformity on Rmn\mathbf{R}^{mn}. Take the norm of x=(xi)∈Rnx = (x_i) \in \mathbf{R}^n to be

∣∣x∣∣=sup⁡i∣xi∣, ||x|| = \sup_i |x_i|,
and let (ej)(e_j) be the canonical basis of Rm\mathbf{R}^m; if uu and vv are two linear mappings of Rm\mathbf{R}^m into Rn\mathbf{R}^n such that ∥u(ej)−v(ej)∥≤ϵ\|u(e_j) - v(e_j)\| \leq \epsilon for 1≤j≤m1 \leq j \leq m, then we have ∣αij−βij∣≤ϵ|\alpha_{ij} - \beta_{ij}| \leq \epsilon for each pair (i,j)(i, j) [U=(αij)U = (\alpha_{ij}) and V=(βij)V = (\beta_{ij}) being the matrices of u,vu, v respectively]; and conversely, if these inequalities are satisfied, we have ∥u(x)−v(x)∥≤maϵ\|u(x) - v(x)\| \leq ma\epsilon for every point xx of a cube of centre o and side aa in Rm\mathbf{R}^m.

3. Countability properties of spaces of continuous functions

Theorem 1

Let XX be a compact space. a) If XX is metrizable and if YY is any metrizable uniform space of countable type (Chapter IX, § 2, no. 8), then the metrizable space Cu(X;Y)C_u(X; Y) of continuous mappings of XX into YY, endowed with the topology of uniform convergence, is of countable type. b) Conversely, if the metrizable space Cu(X;R)C_u(X; \mathbf{R}) is of countable type, then XX is metrizable.

a) Let dd (resp. d′d') be a metric compatible with the topology of XX (resp. with the uniformity of YY); then δ(f,g)=sup⁡x∈Xd′(f(x),g(x))\delta(f, g) = \sup_{x \in X} d'(f(x), g(x)) is a metric defining the uniformity of uniform convergence on the space C(X;Y)C(X; Y), the functions of C(X;Y)C(X; Y) being bounded because XX is compact (no. 1). For each pair of integers m>0,n>0m > 0, n > 0, let GmnG_{mn} be the set of functions f∈C(X;Y)f \in \mathscr{C}(X; Y) such that the relation d(x,x′)≤1/md(x, x') \leq 1/m implies d(f(x),f(x′))≤1/nd(f(x), f(x')) \leq 1/n. Every function f∈C(X;Y)f \in \mathscr{C}(X; Y) is uniformly continuous (Chapter II, § 4, no. 1, Theorem 2) and therefore, for each n>0n > 0, C(X;Y)\mathscr{C}(X; Y) is the union of the sets GmnG_{mn} (m>0m > 0). Let {a1,…,ap(m)}\{a_1, \ldots, a_{p(m)}\} be a finite subset of XX such that the open balls with centres aia_i and radii 1/m1/m cover XX (1≤i≤p(m)1 \leq i \leq p(m)); and let (br)r∈N(b_r)_{r \in \mathbf{N}} be a countable sequence which is dense in YY. For each mapping φ:[1,p(m)]→N\varphi : [1, p(m)] \to \mathbf{N}, let HφH_\varphi be the set of those f∈Gmnf \in G_{mn} such that d′(f(ak),bφ(k))≤1/nd'(f(a_k), b_{\varphi(k)}) \leq 1/n for 1≤k≤p(m)1 \leq k \leq p(m). By the definition of the brb_r, GmnG_{mn} is the union of the sets HφH_\varphi for φ∈Np(m)\varphi \in \mathbf{N}^{p(m)}; let CmnC_{mn} be the set of mappings φ∈Np(m)\varphi \in \mathbf{N}^{p(m)} such that Hφ≠∅H_\varphi \neq \emptyset, and for each φ∈Cmn\varphi \in C_{mn} let gφg_\varphi be an element of HφH_\varphi; finally, let LmnL_{mn} denote the countable set of gφg_\varphi for φ∈Cmn\varphi \in C_{mn}. Let f∈Gmnf \in G_{mn}, and let φ\varphi be an element of CmnC_{mn} such that f∈Hφf \in H_\varphi; then it follows immediately from the definitions that we have d′(f(x),gφ(x))≤4/nd'(f(x), g_\varphi(x)) \leq 4/n for all x∈Xx \in X, i.e. δ(f,gφ)≤4/n\delta(f, g_\varphi) \leq 4/n. Hence the union of the sets LmnL_{mn} is dense in Cu(X;Y)\mathscr{C}_u(X; Y), because for every integer n>0n > 0 and every f∈C(X;Y)f \in \mathscr{C}(X; Y) there exists mm such that f∈Gmnf \in G_{mn}, and we have just seen that the distance from ff to LmnL_{mn} is ≤4/n\leq 4/n.

b) Let I=[0,1]I = [0, 1]. Since Cu(X;I)\mathscr{C}_u(X; I) is a uniform subspace of Cu(X;R)\mathscr{C}_u(X; \mathbf{R}), it is of countable type. Let (fn)(f_n) be a sequence which is dense in Cu(X;I)\mathscr{C}_u(X; I). Consider the product space K=NK = \mathbf{N} and the mapping ψ:x→(fn(x))\psi : x \to (f_n(x)) of XX into KK, which is obviously continuous. The mapping ψ\psi is injective; for, by definition of the sequence (fn)(f_n), the relation fn(x)=fn(x′)f_n(x) = f_n(x') for all nn implies, on passing to the limit, f(x)=f(x′)f(x) = f(x') for every function f∈C(X;I)f \in \mathscr{C}(X; I); but this is impossible if x≠x′x \neq x' by virtue of Axiom (OIV) applied to the point xx and to a neighbourhood VV of xx which does not contain x′x' (Chapter IX, § 1, no. 5, Theorem 2). It follows that the compact space XX is homeomorphic to the subspace ψ(X)\psi(X) of KK (Chapter I, § 9, no. 4, Theorem 2, Corollary 2); since KK is metrizable and of countable type, so is ψ(X)\psi(X) and therefore so is XX.

Q.E.D.

Corollary

Let XX be a locally compact space whose topology admits a countable base, and let YY be a metrizable uniform space of countable type.

a) The space L\mathscr{L} of continuous mappings of XX into YY which have a limit at infinity, endowed with the topology of uniform convergence in XX, is a metrizable space of countable type.

b) The space Cc(X;Y)\mathscr{C}_c(X; Y) of continuous mappings of XX into YY, endowed with the topology of compact convergence, is a metrizable space of countable type.

a) Let X′X' be the compact space obtained by adjoining a point at infinity to XX (Chapter I, § 9, no. 8, Theorem 4); by definition, every function f∈Lf \in \mathscr{L} can be uniquely extended to a continuous function f‾:X′→Y\overline{f} : X' \to Y, and f→f‾f \to \overline{f} is therefore a bijection of L onto C(X′;Y)C(X'; Y); and this bijection is a homeomorphism of the space L onto Cu(X′;Y)C_u(X'; Y) by Proposition 6 of § 1, no. 6. Since X′X' is metrizable (Chapter IX, § 2, no. 9, Proposition 16, Corollary) the result follows from Theorem 1, applied to X′X' and Y.

b) Let (Un)(U_n) be a covering of X by relatively compact open sets, such that every compact subset of X is contained in some UnU_n (Chapter I, § 9, no. 9, Proposition 15, Corollary 1). If SS is the set of the U‾n\overline{U}_n, the topology of compact convergence on C(X;Y)C(X; Y) is the same as the topology of SS-convergence. Consequently (§1,§ 1, no. 2, Remark 3) the space Cc(X;Y)C_c(X; Y) is homeomorphic to a subspace of the product ∏nCu(U‾n;Y)\prod_n C_u(\overline{U}_n; Y); since each of the compact spaces U‾n\overline{U}_n has a countable base, it is metrizable (Chapter IX, § 2, no. 9, Proposition 16); each of the Cu(U‾n;Y)C_u(\overline{U}_n; Y) is therefore metrizable and of countable type by Theorem 1, and hence so is Cc(X;Y)C_c(X; Y).

Note that the space of all bounded continuous real-valued functions on R\mathbf{R}, endowed with the topology of uniform convergence, is not of countable type (Exercise 4).

4. The compact-open topology

Theorem 2

Let X be a topological space, Y a uniform space. For each compact subset K of X and each open subset U of Y, let T(K, U) denote the set of all continuous mappings u:X→Yu : X \to Y such that u(K)⊂Uu(K) \subset U. Then the sets T(K,U)T(K, U) generate the topology of compact convergence on C(X;Y)C(X; Y).

Let Y′Y' be the Hausdorff uniform space associated with Y (Chapter II, § 3, no. 8) and let i:Y→Y′i : Y \to Y' be the canonical mapping of Y onto Y′Y'. The topology of compact convergence is the coarsest topology for which the mappings u→(i∘u)∣Ku \to (i \circ u)|K of C(X;Y)C(X; Y) into Cu(K;Y′)C_u(K; Y') are continuous, as K runs through the set of all compact subsets of X (§1,§ 1, no. 4, Proposition 4). Hence we obtain a subbase of the topology of Cc(X;Y)C_c(X; Y) by taking a subbase of the topology of Cc(K;Y′)C_c(K; Y') for each compact subset K of X and then taking the union [in B(C(X;Y))\mathfrak{B}(C(X; Y))] of the inverse images of these subbases in C(X,Y)C(X, Y). On the other hand, every open subset of Y is of the form i‾−1(U′)\overline{i}^{-1}(U'), where U′U' is open in Y′Y' (Chapter II, § 3, no. 7, Proposition 12); hence, for each compact subset K′⊃KK' \supset K, T(K,i‾−1(U′))T(K, \overline{i}^{-1}(U')) is the inverse image of T(K,U′)T(K, U') under the mapping

C(X;Y)→Ca(K′;Y′). C(X; Y) \to C_a(K'; Y').

It thus remains for us to prove the theorem when X is compact and Y is Hausdorff; we shall make these assumptions from now on.

Let us first show that T(K,U)T(K, U) is open in Cc(X;Y)C_c(X; Y). Let u0u_0 be a point of this set; since u0(K)u_0(K) is compact (Chapter I, § 9, no. 4, Theorem 2, Corollary 1) and contained in the open set UU, there exists a symmetric entourage VV of YY such that V(u0(K))⊂UV(u_0(K)) \subset U (Chapter II, § 4, no. 3, Proposition 4, Corollary). Let WW be the neighbourhood of u0u_0 in Cc(X;Y)C_c(X; Y) consisting of all continuous mappings u:X→Yu : X \to Y such that (u(x),u0(x))∈V(u(x), u_0(x)) \in V for all x∈Kx \in K. For such mappings we clearly have u(K)⊂V(u0(K))⊂Uu(K) \subset V(u_0(K)) \subset U; hence u∈T(K,U)u \in T(K, U) and therefore W⊂T(K,U)W \subset T(K, U), which proves our assertion.

Conversely, if WW is a neighbourhood of a point u0∈Cc(X;Y)u_0 \in C_c(X; Y), let us show that WW contains the intersection of a finite number of neighbourhoods of the form T(K,U)T(K, U). We may suppose that WW is the set of all u∈C(X;Y)u \in C(X; Y) such that (u(x),u0(x))∈V(u(x), u_0(x)) \in V for all x∈Xx \in X, VV being a given entourage of YY. Since u0u_0 is continuous on XX, it is uniformly continuous (Chapter II, § 4, no. 1, Theorem 2). Let V1V_1 be a symmetric entourage of YY, open in Y×YY \times Y and such that V12⊂VV_1^2 \subset V. XX can be covered by a finite number of compact sets KiK_i (1≤i≤n1 \leq i \leq n) such that each u0(Ki)u_0(K_i) is V1V_1-small (1≤i≤n1 \leq i \leq n). Let UiU_i be the open set V1(u0(Ki))V_1(u_0(K_i)), and let u:X→Yu : X \to Y be a continuous mapping contained in the intersection of the nn sets T(Ki,Ui)T(K_i, U_i) (which are neighbourhoods of u0u_0). Then, for every x∈Kix \in K_i, u(x)u(x) belongs to UiU_i and therefore u0(x)u_0(x) and u(x)u(x) are V12V_1^2-close, hence VV-close. Since each x∈Xx \in X belongs to some KiK_i, we have u∈Wu \in W and the proof is complete.

This result leads us to make the following definition:

Definition 1

Let X,YX, Y be two topological spaces, not necessarily uniformizable. For each compact subset KK of XX and each open subset UU of YY, let T(K,U)T(K, U) be the set of all u∈C(X;Y)u \in C(X; Y) such that u(K)⊂Uu(K) \subset U. The topology on C(X;Y)C(X; Y) generated by the sets T(K,U)T(K, U) is called the topology of compact convergence or the compact-open topology; and we denote by Cc(X;Y)C_c(X; Y) the topological space obtained by endowing C(X;Y)C(X; Y) with this topology.

If YY is a uniform space it follows from Theorem 2 that this definition agrees with that given in § 1, no. 3.

If HH is a subset of C(X;Y)C(X; Y) we shall say that the topology induced on HH by that of Cc(X;Y)C_c(X; Y) is the compact-open topology on HH.

Example

Let II be the interval [0,1][0, 1] in R\mathbf{R}. If YY is any topological space, the space Cc(I;Y)C_c(I; Y) is called the space of paths in YY. For each y∈Yy \in Y, the subspace Ωy(Y)\Omega_y(Y) of Cc(I;Y)C_c(I; Y) consisting of paths uu such that u(0)=u(1)=yu(0) = u(1) = y is called the space of loops (in YY) at the point yy.

Remark 1

Likewise, the topology induced on C(X;Y)C(X; Y) by the product topology on YX=F(X;Y)Y^X = \mathscr{F}(X; Y) is called the topology of pointwise convergence (YY being not necessarily uniformizable); it is generated by sets of the form T({x},U)T(\{x\}, U) as xx runs through XX and UU runs through the set of all open subsets of YY, and it is therefore coarser than the compact-open topology. We deduce that, if YY is Hausdorff, the space Cc(X;Y)C_c(X; Y) is Hausdorff (Chapter I, § 8, no. 1, Proposition 5, Corollary).

Remark 2

Let S\mathscr{S} be a subbase of the topology of YY, and let K\mathscr{K} be a set of compact subsets of XX with the following property:

(R) If LL is any compact subset of XX and VV is any neighbourhood of LL, there exists a finite number of sets Ki∈KK_i \in \mathscr{K} such that L⊂⋃iKi⊂VL \subset \bigcup_i K_i \subset V.

Then the sets T(K,U)T(K, U), where K∈KK \in \mathscr{K} and U∈SU \in \mathscr{S}, form a subbase for the compact-open topology on C(X;Y)C(X; Y). To prove this, we have to show that if LL is any compact subset of XX and VV any open subset of YY, and if u∈T(L,V)u \in T(L, V), then there exists a finite number of pairs (Ki,Ui)(K_i, U_i) such that Ki∈KK_i \in \mathscr{K}, Ui∈SU_i \in \mathscr{S} and u∈⋂iT(Ki,Ui)⊂T(L,V)u \in \bigcap_i T(K_i, U_i) \subset T(L, V). Note first that for every finite sequence (sk)(s_k) of sets of S\mathscr{S} and every compact subset MM of XX, we have T(M,⋂kSk)=⋂kT(M,Sk)T\left(M, \bigcap_k S_k\right) = \bigcap_k T(M, S_k) by definition. We may therefore first of all replace S\mathscr{S} by the set of finite intersections of sets of S\mathscr{S}, i.e. we may suppose that S\mathscr{S} is a base of the topology of YY. By hypothesis, u(L)u(L) is quasi-compact and contained in VV, hence there exists a finite number of sets Ui∈SU_i \in \mathscr{S} contained in VV which cover u(L)u(L).

The sets u‾−1(Ui)\overline{u}^{-1}(U_i) are open in XX and cover LL. For each x∈Lx \in L there is therefore a compact neighbourhood NxN_x of xx in LL, contained in some one of the u‾−1(Ui)\overline{u}^{-1}(U_i). We can cover LL with a finite number of these sets Nxj=LjN_{x_j} = L_j; for each jj, let us denote by i(j)i(j) one of the indices ii such that Lj⊂u‾−1(Ui)L_j \subset \overline{u}^{-1}(U_i). This being so, for each index jj there exists [by (R)] a finite number of sets Kjk⊂u‾−1(Ui(j))K_{jk} \subset \overline{u}^{-1}(U_{i(j)}), belonging to K\mathscr{K}, which cover LjL_j. For each v∈⋂j,kT(Kjk,Ui(j))v \in \bigcap_{j,k} T(K_{jk}, U_{i(j)}) we have ⋃kv(Kjk)⊂Ui(j)\bigcup_k v(K_{jk}) \subset U_{i(j)} and therefore v(Lj)⊂Ui(j)v(L_j) \subset U_{i(j)}, and v(L)=⋃jv(Lj)⊂⋃jUi(j)⊂Vv(L) = \bigcup_j v(L_j) \subset \bigcup_j U_{i(j)} \subset V; thus our assertion is proved.

Theorem 3

Let X,Y,ZX, Y, Z be three topological spaces and let ff be a mapping of X×YX \times Y into ZZ. If ff is continuous then f~:x→f(x,.)\tilde{f}: x \to f(x, .) is a continuous mapping of XX into Cc(Y;Z)C_c(Y; Z). The converse is true if YY is locally compact.

Suppose that ff is continuous. To show that f~\tilde{f} is continuous we have to prove that, for each compact subset KK of YY and each open subset UU of ZZ, the inverse image VV of T(K,U)T(K, U) under f~\tilde{f} is open in XX. Let x0∈Vx_0 \in V; for each y∈Ky \in K, we have f(x0,y)∈Uf(x_0, y) \in U, and since ff is continuous there is a neighbourhood VyV_y of x0x_0 in XX and a neighbourhood WyW_y of yy in YY such that f(Vy×Wy)⊂Uf(V_y \times W_y) \subset U. Since KK is compact, there exists a finite number of points yi∈Ky_i \in K such that the sets Wyi(1≤i≤n)W_{y_i} (1 \leq i \leq n) cover KK. Let V′V' be the intersection of the neighbourhoods VyiV_{y_i} of x0x_0, which is a neighbourhood of x0x_0; if x∈V′x \in V' and y∈Ky \in K, we have f(x,y)∈Uf(x, y) \in U, since yy is contained in some one of the WyiW_{y_i} and xx is contained in each VyiV_{y_i}; hence V′⊂VV' \subset V, and therefore VV is a neighbourhood of each of its points and consequently is open in XX.

Conversely, suppose that f~\tilde{f} is continuous and that YY is locally compact, and let us show that ff is continuous. Let x0∈Xx_0 \in X, let y0∈Yy_0 \in Y and let UU be an open neighbourhood of f(x0,y0)f(x_0, y_0) in ZZ; we shall show that there is a neighbourhood VV of x0x_0 in XX and a neighbourhood WW of y0y_0 in YY such that f(V×W)⊂Uf(V \times W) \subset U. Since y→f(x0,y)y \to f(x_0, y) is continuous, there is a compact neighbourhood WW of y0y_0 such that f({x0}×W)⊂Uf(\{x_0\} \times W) \subset U. On the other hand, since f~\tilde{f} is continuous, the set VV of x∈Xx \in X such that f(x,.)∈T(W,U)f(x, .) \in T(W, U) [i.e. such that f(x,y)∈Uf(x, y) \in U for all y∈Wy \in W] is an open subset of XX and therefore a neighbourhood of x0x_0; and we have f(V×W)⊂Uf(V \times W) \subset U.

Q.E.D.

Corollary 1

Let XX be a locally compact space, YY a topological space, HH a subset of C(X;Y)C(X; Y). Then the compact-open topology on HH is the coarsest for which the mapping (u,x)→u(x)(u, x) \to u(x) of H×XH \times X into YY is continuous.

For, by Theorem 3, this mapping is continuous if and only if the canonical injection H→Cc(X;Y)H \to C_c(X; Y) is continuous.

Remark 3

Let XX be a locally compact space and YY a Hausdorff topological space. If T\mathscr{T} is a topology on a subset HH of C(X;Y)C(X; Y) such that the mapping (u,x)→u(x)(u, x) \to u(x) is continuous on H×XH \times X and if also HH is compact with respect to T\mathscr{T}, then T\mathscr{T} is the compact-open topology. For it is finer than the latter by Corollary 1, and since the compact-open topology is Hausdorff, the two topologies are identical. Note that if in addition YY is completely regular, then HH is equicontinuous with respect to every uniformity compatible with the topology of YY (§ 2, no. 5, Theorem 2, Corollary 3), and for every compact subset KK of XX the set

H(K)=⋃x∈KH(x) H(K) = \bigcup_{x \in K} H(x)

is compact, since it is the image of H×KH \times K under the continuous mapping (u,x)→u(x)(u, x) \to u(x).

Corollary 2

Let X,Y,ZX, Y, Z be three topological spaces such that XX is Hausdorff and YY is locally compact. Then the restriction to C(X×Y;Z)C(X \times Y; Z) of the canonical bijection F(X×Y;Z)→F(X;F(Y;Z))\mathscr{F}(X \times Y; Z) \to \mathscr{F}(X; \mathscr{F}(Y; Z)) (Set Theory, R, § 4, no. 14) is a homeomorphism of Cc(X×Y;Z)C_c(X \times Y; Z) onto Cc(X;Cc(Y;Z))C_c(X; C_c(Y; Z)).

This restriction is certainly a bijection

ρ:C(X×Y;Z)→C(X;Cc(Y;Z)) \rho : \mathscr{C}(X \times Y; Z) \to \mathscr{C}(X; \mathscr{C}_c(Y; Z))

by Theorem 3; it remains therefore to be shown that the compact-open topology on C(X×Y;Z)\mathscr{C}(X \times Y; Z) is the inverse image under ρ\rho of the compact-open topology on C(X;Cc(Y;Z))\mathscr{C}(X; \mathscr{C}_c(Y; Z)). Since the sets T(K,U)T(K, U), where KK is a compact subset of YY and UU is an open subset of ZZ, form a subbase of the topology of Cc(Y;Z)\mathscr{C}_c(Y; Z), it follows from Remark 2 that the topology of Cc(X;Cc(Y;Z))\mathscr{C}_c(X; \mathscr{C}_c(Y; Z)) is generated by the sets of the form T(J,T(K,U))T(J, T(K, U)), where KK and UU are as above and JJ is a compact subset of XX. Now the image of T(J,T(K,U))T(J, T(K, U)) under ρ\rho is precisely T(J×K,U)T(J \times K, U), and is therefore an open set; so we have shown that ρ\rho is continuous. To prove that ρ\rho is a homeomorphism, we note first that the sets of the form J×KJ \times K in X×YX \times Y (where JJ is a compact subset of XX, and KK is a compact subset of YY) satisfy the condition (R) of Remark 2: for if LL is a compact subset of X×YX \times Y and VV is a neighbourhood of LL in X×YX \times Y, the projections M=pr⁡1(L)M = \operatorname{pr}_1(L), N=pr⁡2(L)N = \operatorname{pr}_2(L) are compact, since XX and YY are Hausdorff and V∩(M×N)V \cap (M \times N) is a neighbourhood of LL in the compact space M×NM \times N, so that every point of LL has a neighbourhood in M×NM \times N of the form J×K⊂VJ \times K \subset V, where J⊂MJ \subset M and K⊂NK \subset N are compact; since LL can be covered by a finite number of these neighbourhoods, the assertion is proved. Therefore sets of the form T(J×K;U)T(J \times K; U), where JJ is a compact subset of XX, KK a compact subset of YY and UU an open subset of ZZ, generate the topology of Cc(X×Y;Z)\mathscr{C}_c(X \times Y; Z). But we have already seen that the image of T(J×K,U)T(J \times K, U) under ρ\rho is the open set T(J,T(K,U))T(J, T(K, U)) in Cc(X;Cc(Y;Z))\mathscr{C}_c(X; \mathscr{C}_c(Y; Z)); hence ρ\rho is a homeomorphism.

Note that if in addition ZZ is assumed to be uniformizable, Corollary 2 is a trivial consequence of § 1, no. 4, Proposition 2.

Proposition 9

Let X,Y,ZX, Y, Z be three topological spaces, YY being locally compact. Then the mapping (u,v)→v∘u(u, v) \to v \circ u of Cc(X;Y)×Cc(Y;Z)\mathscr{C}_c(X; Y) \times \mathscr{C}_c(Y; Z) into Cc(X;Z)\mathscr{C}_c(X; Z) is continuous.

We have to show that, for every compact subset KK of XX and every open subset UU of ZZ, the set RR of pairs (u,v)(u, v) such that v(u(K))⊂Uv(u(K)) \subset U is open in Cc(X;Y)×Cc(Y;Z)\mathscr{C}_c(X; Y) \times \mathscr{C}_c(Y; Z). Let (u0,v0)∈R(u_0, v_0) \in R; then u0(K)u_0(K) is a compact subset of the locally compact space YY, contained in the open set v0−1‾(U)\overline{v_0^{-1}}(U), and hence there is a compact neighbourhood LL of u0(K)u_0(K) contained in v0−1‾(U)\overline{v_0^{-1}}(U) (Chapter I, § 9, no. 7, Proposition 10). The set VV of all u∈Cc(X;Y)u \in \mathscr{C}_c(X; Y) such that u(K)⊂Lu(K) \subset L is a neighbourhood of u0u_0, and the set WW of all v∈Cc(Y;Z)v \in \mathscr{C}_c(Y; Z) such that v(L)⊂Uv(L) \subset U is a neighbourhood of v0v_0; furthermore, the relation (u,v)∈V×W(u, v) \in V \times W implies v(u(K))⊂Uv(u(K)) \subset U. Hence the result.

5. Topologies on groups of homeomorphisms

Proposition 10

Let XX be a uniform space and let HH be an equicontinuous set of homeomorphisms of XX onto itself. If HH and H−1H^{-1} are endowed with the topology of pointwise convergence in XX, then the mapping u→u−1u \to u^{-1} of H−1H^{-1} onto HH is continuous.

It is enough to show that, for each x0∈Xx_0 \in X, the mapping u→u−1(x0)u \to u^{-1}(x_0) of H−1H^{-1} into XX is continuous at every point u0∈H−1u_0 \in H^{-1}. Let VV be a symmetric entourage of XX, and let y0=u0−1(x0)y_0 = u_0^{-1}(x_0). By hypothesis there is a symmetric entourage UU of XX such that the relation (x,x0)∈U(x, x_0) \in U implies (u−1(x),u−1(x0))∈V(u^{-1}(x), u^{-1}(x_0)) \in V for all u∈H−1u \in H^{-1}. Take an element u∈H−1u \in H^{-1} which is W({y0},U)W(\{y_0\}, U)-close to u0u_0; then we have (u(y0),u0(y0))∈U(u(y_0), u_0(y_0)) \in U, i.e. (u(y0),x0)∈U(u(y_0), x_0) \in U. It follows that (y0,u−1(x0))∈V(y_0, u^{-1}(x_0)) \in V, i.e. (u0−1(x0),u−1(x0))∈V(u_0^{-1}(x_0), u^{-1}(x_0)) \in V; this completes the proof.

Corollary

Let XX be a uniform space and let HH be an equicontinuous group of homeomorphisms of XX onto itself. Then the topology of pointwise convergence in XX is compatible with the group structure of HH.

This is a consequence of Proposition 10, together with § 2, no. 1, Proposition 1, Corollary 5.

Proposition 11

Let XX be a compact space and let Γ\Gamma be the group of all homeomorphisms of XX onto itself. Then the topology of uniform convergence in XX is compatible with the group structure of Γ\Gamma.

We know already (no. 4, Proposition 9) that the mapping (u,v)→v∘u(u, v) \to v \circ u of Γ×Γ\Gamma \times \Gamma into Γ\Gamma is continuous with respect to this topology; thus we have to show that u→u−1u \to u^{-1} is continuous at every point u0u_0 of Γ\Gamma. Since u0−1u_0^{-1} is uniformly continuous on XX, given any symmetric entourage VV of XX there exists an entourage WW of XX such that the relation (x,x′)∈W(x, x') \in W implies (u0−1(x),u0−1(x′))∈V(u_0^{-1}(x), u_0^{-1}(x')) \in V. Hence, if u∈Γu \in \Gamma is such that (u0(x),u(x))∈W(u_0(x), u(x)) \in W for all x∈Xx \in X, it follows that (x,u0−1(u(x)))∈V(x, u_0^{-1}(u(x))) \in V for all x∈Xx \in X, and therefore (as uu is bijective) (u−1(x),u0−1(x))∈V(u^{-1}(x), u_0^{-1}(x)) \in V for all x∈Xx \in X. This completes the proof.

Now let XX be a locally compact space and let Γ\Gamma be the group of all homeomorphisms of XX onto itself. The topology of compact convergence in XX is not necessarily compatible with the group structure of Γ\Gamma (Exercise 17). Let X′X' denote the compact space obtained by adjoining a point at infinity ω\omega to XX. Every homeomorphism uu of XX onto itself extends uniquely to a homeomorphism u′u' of X′X' onto itself such that u′(ω)=ωu'(\omega) = \omega (Chapter I, § 10, no. 3, Corollary to Proposition 7), so that Γ\Gamma can be identified with the subgroup of the group Γ′\Gamma' of all homeomorphisms of X′X' onto itself, consisting of all homeomorphisms which leave ω\omega fixed. The topology induced on Γ\Gamma by that of Cu(X′;X′)C_u(X'; X') is therefore compatible with the group structure of Γ′\Gamma' (Proposition 11), and Γ\Gamma is closed in Γ′\Gamma' [with respect to the topology induced by that of Cu(X′;X′)C_u(X'; X')] because it is defined by the equation u(ω)=ωu(\omega) = \omega (§ 1, no. 2, Remark 6). We denote by Tβ\mathscr{T}_\beta the group topology thus defined on Γ\Gamma; it is finer than the topology of compact convergence and can also (by virtue of § 1, no. 6, Proposition 6) be defined as the topology of uniform convergence on XX, when XX is endowed with the uniformity induced by the unique uniformity of X′X'.

The topology Tβ\mathscr{T}_\beta can be characterized as follows:

Proposition 12

On the group Γ\Gamma of all homeomorphisms of a locally compact space XX, the topology Tβ\mathscr{T}_\beta is the coarsest for which the mappings u→uu \to u and u→u−1u \to u^{-1} of Γ\Gamma into Cc(X;X)C_c(X; X) are continuous.

Let us denote the latter topology for the moment by C′\mathscr{C}'. Since u→u−1u \to u^{-1} is continuous with respect to Tβ\mathscr{T}_\beta and since Tβ\mathscr{T}_\beta is finer than the topology of compact convergence, it is clear that Tβ\mathscr{T}_\beta is finer than C′\mathscr{C}'. To prove the converse, endow X′X' with its unique uniformity; let u0∈Γu_0 \in \Gamma and let VV be an entourage of X′X'; then we have to prove that there is a compact subset KK of XX and a symmetric entourage WW of X′X' such that the relations

u∈Γ, (u0(x),u(x))∈W and (u0−1(x),u−1(x))∈W for all x∈K u \in \Gamma, \ (u_0(x), u(x)) \in W \text{ and } (u_0^{-1}(x), u^{-1}(x)) \in W \text{ for all } x \in K

imply

(u0(x),u(x))∈V for all x∈X. (u_0(x), u(x)) \in V \text{ for all } x \in X.

Let V1V_1 be a symmetric open entourage of X′X' such that V21⊂V\overset{2}{V}_1 \subset V; then K1=X′−V1(ω)K_1 = X' - V_1(\omega) is a compact subset of XX. Choose a symmetric open entourage WW of X′X' such that W⊂VW \subset V and W(ω)∩W(u0−1(K1))=∅W(\omega) \cap W(u_0^{-1}(K_1)) = \emptyset; this is possible by Proposition 4 of Chapter II, § 4, no. 3. Let K2=X′−W(ω)K_2 = X' - W(\omega), which is a compact subset of XX. We shall see that WW and the compact set K=K1∪K2K = K_1 \cup K_2 do what is required. Since W⊂VW \subset V, it is enough to show that the relation

(u0−1(x),u−1(x))∈W for all x∈K1(u∈Γ) (u_0^{-1}(x), u^{-1}(x)) \in W \text{ for all } x \in K_1 \quad (u \in \Gamma)

implies that

(u(y),ω)∈V1 for all y∈W(ω); (u(y), \omega) \in V_1 \text{ for all } y \in W(\omega);

for we shall then also have (u0(y),ω)∈V1(u_0(y), \omega) \in V_1 and thence

(u0(y),u(y))∈V21⊂V (u_0(y), u(y)) \in \overset{2}{V}_1 \subset V

for all y∈W(ω)=X′−K2y \in W(\omega) = X' - K_2. Now if we had y∈W(ω)y \in W(\omega) and

u(y)∈X′−V1(ω)=K1, u(y) \in X' - V_1(\omega) = K_1,

it would follow that y∈u−1(K1)⊂W(u0−1(K1))y \in u^{-1}(K_1) \subset W(u_0^{-1}(K_1)), contrary to the choice of WW; the proof is therefore complete.

In general the group Γ\Gamma, endowed with Tβ\mathscr{T}_\beta, is not locally compact; but we have the following criterion:

Theorem 4

Let GG be a subgroup of the group Γ\Gamma of all homeomorphisms of a locally compact space XX. Suppose that, in the space Cc(X;X)C_c(X; X), there is a neighbourhood VV of the identity mapping ee such that V∩G=HV \cap G = H is symmetric in GG and relatively compact in Cc(X;X)C_c(X; X). Then the closure G‾\overline{G} of GG in Γ\Gamma with respect to the topology Tβ\mathscr{T}_\beta is a locally compact group with respect to the topology induced by Tβ\mathscr{T}_\beta; this topology induced on G‾\overline{G} is the same as the topology of compact convergence, and the closure H‾\overline{H} of HH in Cc(X;X)C_c(X; X) is a neighbourhood of ee in G‾\overline{G} with respect to this topology.

Let us show first that H‾\overline{H} is contained in Γ\Gamma and that the topology induced on H‾\overline{H} by Tβ\mathscr{T}_\beta is the same as the topology of compact convergence. Let u0∈H‾u_0 \in \overline{H}; u0u_0 is therefore the limit, in Cc(X;X)C_c(X; X), of an ultrafilter Φ\Phi on HH. Since Φ−1\Phi^{-1} (the image of Φ\Phi under u→u−1u \to u^{-1}) is an ultrafilter base on H⊂H‾H \subset \overline{H}, it converges in the compact subspace H‾\overline{H} of Cc(X;X)C_c(X; X) to an element v0v_0. The mapping (u,v)→uv(u, v) \to uv converges to u0v0u_0 v_0 with respect to Φ×Φ−1\Phi \times \Phi^{-1} (no. 4, Proposition 9); a fortiori, u→uu−1=eu \to uu^{-1} = e converges to u0v0u_0 v_0 with respect to Φ\Phi, hence u0v0=eu_0 v_0 = e since Cc(X;X)C_c(X; X) is Hausdorff. Similarly v0u0=ev_0 u_0 = e; hence u0u_0 is a homeomorphism of XX, i.e., u0∈Γu_0 \in \Gamma. Thus H‾\overline{H} is contained in Γ\Gamma. Moreover, this argument shows that H‾−1=H‾\overline{H}^{-1} = \overline{H} and that, for every ultrafilter Φ\Phi on H‾\overline{H} which converges to u0u_0, Φ−1\Phi^{-1} converges in Cc(X;X)C_c(X; X) to u0−1u_0^{-1}; hence the mapping u→u−1u \to u^{-1} of H‾\overline{H} into Cc(X;X)C_c(X; X) is continuous when H‾\overline{H} carries the topology of compact convergence (Chapter I, § 7, no. 4, Proposition 9, Corollary 1). Proposition 12 then shows that, on H‾\overline{H}, the topology of compact convergence is the same as the topology induced by Tβ\mathscr{T}_\beta.

Furthermore, since the topology Tβ\mathscr{T}_\beta on Γ\Gamma is finer than the topology of compact convergence, H‾\overline{H} is also the closure of HH with respect to Tβ\mathscr{T}_\beta. But HH is a neighbourhood of ee in GG with respect to the topology of compact convergence, and a fortiori with respect to the topology induced by Tβ\mathscr{T}_\beta; it follows (Chapter I, § 3, no. 1, Proposition 2) that H‾\overline{H} is a neighbourhood of ee in G‾\overline{G} with respect to the topology induced by Tβ\mathscr{T}_\beta, and hence G‾\overline{G} is locally compact in this topology. If WW is the interior of VV with respect to the topology of compact convergence, then W∩ΓW \cap \Gamma is open in Tβ\mathscr{T}_\beta, hence W∩G‾W \cap \overline{G} is contained in the closure of H=V∩GH = V \cap G with respect to Tβ\mathscr{T}_\beta (Chapter I, § 1, no. 6, Proposition 5); this shows that H‾\overline{H} is also a neighbourhood of ee in G‾\overline{G} with respect to the topology of compact convergence. Finally, for each u0∈Γu_0 \in \Gamma, the bijections v→u0∘vv \to u_0 \circ v and v→u0−1∘vv \to u_0^{-1} \circ v of Cc(X;X)C_c(X; X) onto itself are continuous (no. 4, Proposition 9), and hence, if u0∈G‾u_0 \in \overline{G}, u0H‾u_0 \overline{H} is a neighbourhood of u0u_0 in G‾\overline{G} with respect to the topology of compact convergence. This completes the proof.

Corollary

Let GG be a group of homeomorphisms of a locally compact space XX. If the closure G‾\overline{G} of GG in Cc(X;X)C_c(X; X) is compact, then G‾\overline{G} is a group of homeomorphisms of XX, and the topology of compact convergence is compatible with the group structure of G‾\overline{G}, which is therefore a compact topological group.

A group of homeomorphisms of a locally compact space XX which is locally compact but not compact with respect to the topology of compact convergence is locally closed in Cc(X;X)C_c(X; X) by virtue of Chapter I, § 9, no. 7, Proposition 12, but is not necessarily closed.

For example, in the ring L(Rn)\mathscr{L}(\mathbf{R}^n) of endomorphisms of Rn\mathbf{R}^n, identified with the ring Mn(R)M_n(\mathbf{R}) of square n×nn \times n matrices over R\mathbf{R} and endowed with the topology of compact convergence, the group GL(n,R)GL(n, \mathbf{R}), identified with the group of non-singular matrices, is locally compact but dense (Chapter VI, § 1, no. 6, Proposition 6).

Exercises

See the exercises for § 3.

Transcribed from the printed volume. Not translated.