224 CHAPTER 9. BASIC FUNCTION SPACES

the process must stop since otherwise, you would have an infinite sequence of points withnot limit point because they are all ε apart. This contradicts the compactness of K. ■

Recall Lemma 9.1.2. If you consider C (K,Rn) it is automatically equal to BC (K,Rn)because of the extreme value theorem applied to x→ |f(x)| for x ∈ K. Therefore, the spaceC (K,Rn) is complete with respect to the norm defined there.

Definition 9.2.2 Let A be a set of functions in C (K,Rn) . It is called equicontin-uous if for every ε > 0 there exists δ > 0 such that if |x−y| < δ , then |f(x)− f(y)| < ε

for all f ∈ A . In words, the functions in A are uniformly continuous for all f at once. Aset A ⊆C (K,Rn) is uniformly bounded if there is a large enough positive number M suchthat max{|f(x)| : x ∈ K, f ∈A }< M.

The significant property of an equicontinuous set of functions is the following.

Lemma 9.2.3 If {gk}∞

k=1 is equicontinuous and converges pointwise to g on a compactset K, then the sequence converges uniformly on K.

Proof of claim: Let ε > 0 be given and let δ go with ε/4 in the definition of equicon-tinuous. By compactness and Proposition 9.2.1, there are finitely many points of K, de-noted as {x1, · · · ,xs} such that K ⊆∪s

i=1B(xi,δ ). There exists Ni such that if k, l ≥Ni, then|gl (xi)−gk (xi)|< ε

4 . Thus if N ≥max{Ni, i = 1, · · · ,s} , then for all xi, |gl (xi)−gk (xi)|<ε

4 if k ≥ N. Then for k, l ≥ N, and x arbitrary, let x ∈ B(xi,δ ) . Then

|gl (x)−gk (x)| ≤ |gl (x)−gl (xi)|+ |gl (xi)−gk (xi)|+ |gk (xi)−gk (x)|

4+

ε

4+

ε

4

Thus for k, l ≥N, ∥gl−gk∥∞< 3ε

4 < ε. This shows {gk} is a Cauchy sequence in C (K,Rn)which is complete. Thus this sequence converges uniformly to some g ∈C (K,Rn). ■

The following is the Arzela Ascoli theorem . Actually, the converse is also true but Iwill only give the direction of most use in applications.

Theorem 9.2.4 Let A ⊆C (K,Rn) be both equicontinuous and uniformly bounded.Then every sequence in A has a convergent subsequence converging to some g∈C (K,Rn),the convergence taking place with respect to ∥·∥

∞, the uniform norm.

Proof: Let{

f j}∞

j=1 be a sequence of functions in A . Let D be a countable densesubset of K. Say D≡ {dk}∞

k=1 . Then{

f j (d1)}∞

j=1 is a bounded set of points in Rn. By theHeine Borel theorem, there is a subsequence, denoted by

{f( j,1) (d1)

}∞

j=1 which converges.

Now apply what was just done with{

f j}

to{

f( j,1)}

and feature d2 instead of d1. Thus{f( j,2)

}∞

j=1 is a subsequence of{

f( j,1)}

which converges at d2. This new subsequence stillconverges at d1 thanks to Theorem 2.2.10. Continue this way. Thus we get the following

f(1,1) f(2,1) f(3,1) · · · converges at d1f(1,2) f(2,2) f(3,2) · · · converges at d1,d2f(1,3) f(2,3) f(3,3) · · · converges at d1,d2,d3

......

......

f(1,l) f(2,l) f(3,l) · · · converges at d j, j ≤ l...

......

...

224 CHAPTER 9. BASIC FUNCTION SPACESthe process must stop since otherwise, you would have an infinite sequence of points withnot limit point because they are all € apart. This contradicts the compactness of K.Recall Lemma 9.1.2. If you consider C (K,R") it is automatically equal to BC (K,R")because of the extreme value theorem applied to x — |f(x)| for x € K. Therefore, the spaceC (K,R") is complete with respect to the norm defined there.Definition 9.2.2 Let @ be a set of functions in C(K,IR"). It is called equicontin-uous if for every € > 0 there exists 6 > 0 such that if |x —y| < 6, then |f(x) -f(y)| <€for all f € &. In words, the functions in & are uniformly continuous for all f at once. Aset & CC(K,R") is uniformly bounded if there is a large enough positive number M suchthat max {|f(x)|:x ¢ K,fe A} <M.The significant property of an equicontinuous set of functions is the following.Lemma 9.2.3 /f {gx}; is equicontinuous and converges pointwise to g on a compactset K, then the sequence converges uniformly on K.Proof of claim: Let ¢ > 0 be given and let 6 go with €/4 in the definition of equicon-tinuous. By compactness and Proposition 9.2.1, there are finitely many points of K, de-noted as {x;,--- ,X;} such that K C Uj_, B(x;,6). There exists N; such that if k,/ > Nj, then|g (xi) — gx (xi)| < 4. Thus if N > max {N;,i= 1,--- ,s}, then for all x;, |g (x;) — gx (xi)| <£ if k > N. Then for k,/ > N, and x arbitrary, let x € B(x;,6). ThenIgi (x) —ge(x)| < |g (x) — 7 (x)| +180 (x) — gx (%:)| + |Bx (Xi) — Bk (X)|- £,8,€Thus for k,/ > N, ||g: — g|l.. < *£ < €. This shows {g;} is a Cauchy sequence in C (K,R")which is complete. Thus this sequence converges uniformly to some g € C(K,R").The following is the Arzela Ascoli theorem . Actually, the converse is also true but Iwill only give the direction of most use in applications.Theorem 9.2.4 Let 7 C C(K,R") be both equicontinuous and uniformly bounded.Then every sequence in & has a convergent subsequence converging to some g € C (K,R"),the convergence taking place with respect to ||-\|,,, the uniform norm.Proof: Let {f ijt be a sequence of functions in <. Let D be a countable densesubset of K. Say D = {dy }7_, . Then {f; (di) } i is a bounded set of points in R”. By theHeine Borel theorem, there is a subsequence, denoted by {f (4.1) (d,) eo which converges.Now apply what was just done with {f;} to {f(;,1)} and feature dy instead of d;. Thus{f(;,2) van is a subsequence of {f;;.1) } which converges at dy. This new subsequence stillconverges at d; thanks to Theorem 2.2.10. Continue this way. Thus we get the followingfra) flo1) fan converges at dyfii2) f2) faz) +--+ converges at d1,dofii 3) f(23) f3) --- converges at d),d,d3fir) fl2n fay +++ converges at dj, j </