Let D be a dense subset of X and A a subset of the space Cs(X). The following conditions are equivalent : (i) A is relatively compact in Cs(X). (ii) From every infinite sequence of elements of A, we can extract a sequence converging in Cs(X). (iii) Every infinite sequence of elements of A has a limit point in Cs(X). (iv) Let (fn)n∈N be a sequence of functions belonging to A and (xm)m∈N a sequence of points of D. If the iterated limits
γ=m→∞limn→∞limfn(xm),δ=n→∞limm→∞limfn(xm)
exist, then they are equal. In addition, we have supf∈A∣f(x)∣<+∞ for all x∈X.
(i) ⇒ (ii) : let A be the closure of A in Cs(X). Assume that A is compact, and consider a sequence of functions fn∈A (for n∈N). Let ϕ be the continuous mapping x↦(fn(x))n∈N from X into the metrizable space KN. The image X′ of X under ϕ is a compact metrizable space, since X is compact. Let E be the closed subspace of Cs(X) consisting of continuous functions f on X such that the relation ϕ(x)=ϕ(y) implies f(x)=f(y) for every pair of points x,y in X. By cor. 2 of GT, I, § 9, No. 4 and prop. 3 of GT, I, § 5, No. 2, the mapping f′↦f′∘ϕ is a homeomorphism ϕ∗ from Cs(X′) onto E. Hence the set A′=(ϕ∗)−1(A) is compact in Cs(X′), and it is clear that there exist elements fn′ in A′ such that ϕ∗(fn′)=fn′∘ϕ is equal to fn.
Since X′ is a compact metrizable space, there exists a countable dense subset D' in X′ (GT, IX, § 2, No. 8, prop. 12 and § 2, No. 9, prop. 16). Let T1 (resp. T2) be the topology on A′ induced by the topology of simple convergence on D' (resp. X′). Then T1 is metrizable, T2 is compact and finer than T1, hence T1 and T2 coincide; in other words, A′ is a compact metrizable subspace of Cs(X′). Therefore, there exists a sequence (fnk′) extracted from (fn′) and converging to an element f′ of Cs(X′). Therefore, the sequence (fnk) converges to f=f′∘ϕ in Cs(X).
(ii) ⇒ (iii) : this is clear.
(iii) ⇒ (iv) : suppose that every infinite sequence of elements of A has a limit point in Cs(X). Let x∈X. The mapping ϕx:f↦f(x) from A into K is continuous. Consequently, every infinite sequence in ϕx(A) has a limit point; since the field K (equal to R or C) is metrizable, the set ϕx(A) is relatively compact in K, hence bounded. In other words, we have supf∈A∣f(x)∣<∞.
Let fn,xm,γ and δ be as in (iv). Let f be a limit point of the sequence (fn) in Cs(X), and let x be a limit point of the sequence (xm) in the compact space X. For every m, the mapping h↦h(xm) from Cs(X) into K is continuous. In view of the hypotheses, we have f(xm)=limn→∞fn(xm), and hence γ=limm→∞f(xm); since f:X→K is continuous and x is a limit point of the sequence (xm), we get γ=f(x). In an analogous way, we prove the equality δ=f(x), whence γ=δ.
(iv) ⇒ (i) : suppose that the set of numbers f(x), as f ranges over A, is bounded in K for all x∈X. This is equivalent to assuming that the closure A of A in the product space KX is compact (GT, I, § 9, No. 5). Suppose that A is not relatively compact in Cs(X). This means that there exists a function u∈A and a point a∈X such that u is not continuous at a. Hence there exists a real number ε>0 such that in every neighbourhood U of a, there exists a point x with ∣u(x)−u(a)∣≥ε.
We shall construct by induction a sequence (xn)n∈N of points in D and a sequence (fn)n∈N of elements of A, satisfying the following relations :
(1)m
∣u(xm)−u(a)∣≥εform≥1;
(2)m
∣u(xi)−fm(xi)∣≤m+11for0≤i≤m−1;
(3)m,i
∣fm(xi)−fm(a)∣≤i+11for0≤m≤i.
We take x0=a with f0 arbitrary in A (the set A is not empty, otherwise it will be relatively compact in Cs(X)). Let n≥1 and x0,x1,...,xn−1,f0,f1,...,fn−1 satisfy relations (1)m, (2)m for 1≤m<n and (3)m,i for 0≤m≤i<n. Since u belongs to A, there exists fn∈A satisfying (2)n. Let Vn be the set of all x∈X such that ∣fm(x)−fm(a)∣≤n+11 for 0≤m≤n. Since fn is continuous, Vn is a neighbourhood of a; choose a point xn in D∩Vn such that ∣u(xn)−u(a)∣≥ε, hence (1)n and (3)m,n are satisfied. Therefore, the construction can be continued.
Since u(X) is a compact subset of K, there exists a sequence (yk) extracted from (xm) and such that the limit γ=limk→∞u(yk) exists. By (2)m, we have u(xi)=limn→∞fn(xi) for all i∈N, hence
(4)
γ=k→∞limn→∞limfn(yk).
On the other hand we have fn(a)=limi→∞fn(xi) by (3)m,i hence fn(a)=limk→∞fn(yk). Since x0=a, we deduce from (2)m that limn→∞fn(a)=u(a). Consequently,
u(a)=n→∞limk→∞limfn(yk).
Finally, from (1)m, we get ∣γ−u(a)∣≥ε, and so γ=u(a). This contradicts assertion (iv); we have thus proved that (iv) implies (i).
Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.
Nothing.