5 Compact subsets of C and Lp

We want to find out which subsets of the spaces C(K; K) (K compact) and Lp (Ω; K) (Ω
open) are compact. We already know that the closed unit balls in these spaces are not
compact, as this is true only for finite-dimensional spaces.
Two notions of compactness.

Proposition 5.1
Let (X, d) be a metric space. Then there are equivalent:

(i) X is sequentially compact, that is, every sequence in X has a subsequence which
converges to some element of X.

(ii) Every open covering of X has a finite subcovering, that is: If (Ui )i∈I is a family of
open subset of X such that [
X= Ui ,
i∈I

then there exists n ∈ N and indices i1 , . . . , in ∈ I such that

n
[
X= U ik .
k=1

If (ii) holds, we say that X is compact.

Proof: Omitted. 2

A side remark: In general topological spaces, (i) and (ii) are not equivalent.

Corollary 5.2 Let X be a normed space, let F be a compact subset of X. Then F is

closed and bounded.

Proof: Let (xn )n∈N be a sequence in F with xn → a ∈ X. By Proposition 5.1(i), for

some subsequence and some b ∈ F we have xnk → b. As xnk → a, we must have a = b.
This proves that F is closed. If F were unbounded, then there would exist a sequence
(xn )n∈N in F with kxn k ≥ n for all n ∈ N. Such a sequence does not have a convergent
subsequence. Consequently, F cannot be compact. 2

Compact sets in the space of continuous functions. For a subset of C(Ω; K) to

be compact, in addition to being closed and bounded, the following property must be
satisfied.

Definition 5.3 (Equicontinuity)

Let Ω ⊂ Rn . A subset F of C(Ω; K) is called equicontinuous, if for all x ∈ Ω and all
ε > 0 there exists a δ > 0 such that for all y ∈ Ω

kx − yk < δ ⇒ |f (x) − f (y)| < ε for all f ∈ F .

(“It is possible to find a common δ for all f ∈ F ”.) 2

46
Lemma 5.4 Let Ω ⊂ Rn , F ⊂ C(Ω; K), assume that there exists an L such that

|f (x) − f (y)| ≤ Lkx − yk , for all x, y ∈ Ω, f ∈ F . (5.1)

Then F is equicontinuous.

Proof: Setting δ = ε/L we see that F has the property required in Definition 5.3. 2

When F ⊂ C 1 (Ω), it follows from the mean value theorem that (5.1) holds, if there exists
a C > 0 such that

|∂i f (x)| ≤ C , for all x ∈ Ω, f ∈ F , i = 1, . . . , n. (5.2)

This criterion is often used to prove equicontinuity of a given set F of continuous functions.

Proposition 5.5 (Arzela-Ascoli)

Let K ⊂ Rn be compact, let F ⊂ (C(K; K), k · k∞ ). Then there are equivalent

(i) F is relatively compact in C(K), that is, F is compact.

(ii) F is bounded and equicontinuous.

Proof: Omitted. 2

Smoothing by convolution. We recall the definition of the convolution of two functions

f and g, Z
(f ∗ g)(y) = f (x)g(y − x) dx . (5.3)
Rn

It is defined for f, g ∈ L1 (Rn ) and yields a function f ∗ g ∈ L1 (Rn ) with the properties

f ∗g =g∗f, kf ∗ gk1 ≤ kf k1 · kgk1 . (5.4)

One can approximate a given function by smooth functions if one convolves it with suitable
functions. We define
(
exp − 1t , t > 0 ,


ψ : R → R , ψ(t) = (5.5)
0, t ≤ 0,
ψ̃ : R → R , ψ̃(r) = ψ(1 − r2 ) , (5.6)
η1 : R n → R , η1 (x) = αψ̃(kxk) . (5.7)

Here, α > 0 is chosen such that

Z
η1 (x) dx = 1 . (5.8)
Rn

For given ε > 0 we define the “standard mollifier”

1 x
ηε : Rn → R , ηε (x) = η1 . (5.9)
εn ε

47
The functions ηε are radially symmetric (that is, they only depend upon kxk), and have
the properties (as developed in integration theory)
Z
∞ n
ηε ∈ C0 (R ) , supp (ηε ) = K(0; ε) , ηε ≥ 0 , ηε (x) dx = 1 . (5.10)
Rn

For functions which are not defined on the whole of Rn we too want to consider convolution
with the standard mollifier.

Given f ∈ L1 (Ω), we define f˜ by setting f˜ = f in Ω and f˜ = 0 outside of Ω, and define

f ε by
Z Z
ε ˜
f (y) = (f ∗ηε )(y) = ˜
f (x)ηε (y −x) dx = f (x)ηε (y −x) dx , for all y ∈ Rn . (5.11)
Rn Ω

Since f ε = f˜ ∗ ηε = ηε ∗ f˜, we may represent f ε as

Z Z
ε ˜ −n x
˜
f (y) = ηε (x)f (y − x) dx = ε η1 f (y − x) dx
Rn Rn ε
Z Z (5.12)
= ˜
η1 (z)f (y − εz) dz = η1 (z)f˜(y − εz) dz .
Rn K(0;1)

In the following, we no longer distinguish between f and f˜, and just write f for both
functions.

Lemma 5.6 Let Ω ⊂ Rn be open and bounded, f ∈ Lp (Ω), 1 ≤ p < ∞. Then f ε = f ∗ ηε

satisfies

f ε ∈ C ∞ (Rn ) , supp (f ε ) ⊂ Ω + B(0; ε) , kf ε kLp (Ω) ≤ kf kLp (Ω) . (5.13)

Proof: We have f ∈ L1 (Ω) since Ω is bounded, and

Z
ε
f (y) = f (x)ηε (y − x) dx , for all y ∈ Rn . (5.14)
Ω

As ηε (y − x) = 0 holds for ky − xk ≥ ε, we get supp (f ε ) ⊂ Ω + B(0; ε). For all multi-

indices α, the functions x 7→ f (x)∂ α ηε (y − x) are bounded (uniformly with respect to y)
by the integrable functions k∂ α ηε k∞ |f |. By a result from integration theory, all partial
derivatives ∂ α f ε exist in Rn and are continuous; thus we have f ε ∈ C ∞ (Rn ). Let now
y ∈ Ω. In the case p > 1, we choose q such that p1 + 1q = 1. It follows from Hölder’s
inequality that
Z Z
1 1
ε
|f (y)| = f (x)ηε (y − x) dx ≤ |f (x)|(ηε (y − x)) p (ηε (y − x)) q dx
Ω Ω
Z  p1  Z  1q
p
≤ |f (x)| ηε (y − x) dx ηε (y − x) dx .
Ω
| Ω {z }
≤1

48
Next, we see that for p ≥ 1 (for p > 1 we use the estimate from the previous line)
Z Z Z Z Z
ε p p p
|f (y)| dy ≤ |f (x)| ηε (y − x) dx dy = |f (x)| ηε (y − x) dy dx
Ω ZΩ Ω Ω Ω

≤ |f (x)|p dx .
Ω
2

By C0 (Rn ) we denote the space of functions which are continuous on Rn and have compact
support.

Lemma 5.7 Let f ∈ C0 (Rn ). Then f ε → f uniformly on Rn when ε → 0.

Proof: For all y ∈ Rn we have

Z Z
ε
|f (y) − f (y)| = f (x)ηε (y − x) dx − f (y) = (f (x) − f (y))ηε (y − x) dx
Rn R n
Z
≤ ηε (y − x) dx · sup |f (y) − f (x)| ≤ sup |f (z) − f (x)|
Rn x∈Rn x,z∈Rn
kx−yk≤ε kx−zk≤ε

Since f has compact support and therefore is uniformly continuous on Rn , the assertion
follows. 2

Lemma 5.8 Let Ω ⊂ Rn be open and bounded, let f ∈ Lp (Ω), 1 ≤ p < ∞. Then f ε → f
in Lp (Ω) for ε → 0.

Proof: Let δ > 0. We choose g ∈ C0 (Rn ) such that

kf − gkLp (Ω) ≤ δ .
This is possible since C0 (Rn ) is dense in Lp (Ω) (see the exercises). It follows that
kf − f ε kLp (Ω) ≤ kf − gkLp (Ω) + kg − g ε kLp (Ω) + kg ε − f ε kLp (Ω)
(5.15)
≤ 2δ + kg − g ε kLp (Ω) ,
since
g ε − f ε = g ∗ ηε − f ∗ ηε = (g − f )ε
and by Lemma 5.6
k(g − f )ε kLp (Ω) ≤ kg − f kLp (Ω) ≤ δ .
By Lemma 5.7 we have g ε → g uniformly. Consequently,
lim sup kf − f ε kLp (Ω) ≤ 2δ .
ε→0

As δ > 0 was arbitrary, the assertion follows. 2

Compactness in Lp . We investigate the behaviour of

Z
|f (x + h) − f (x)|p dx , h → 0 .
Ω

49
Proposition 5.9 Let Ω ⊂ Rn be open and bounded, let f ∈ Lp (Ω), 1 ≤ p < ∞. Then
Z
lim |f (x + h) − f (x)|p dx = 0 . (5.16)
h→0 Ω

(Again we extend f by 0 outside of Ω.)

Proof: We define (τh f )(x) = f (x + h). Then

Z
|f (x + h) − f (x)|p dx = kτh f − f kpLp (Ω) .
Ω

We estimate from above

kτh f − f kLp (Ω) ≤ kτh f − τh f ε kLp (Ω) + kτh f ε − f ε kLp (Ω) + kf ε − f kLp (Ω) .

Let γ > 0. For ε > 0 sufficiently small, the third and also the first term on the right side
are smaller than γ/3 for all h with khk ≤ ε because of

kτh f − τh f ε kLp (Ω) = kτh (f − f ε )kLp (Ω) ≤ kf ε − f kLp (Ω+B(0;ε)) .

As f ε is uniformly continuous on Ω, for h sufficiently small the second term, too, is smaller
than γ/3. This proves the assertion. 2

Proposition 5.10 (Fréchet-Riesz-Kolmogorov)

Let Ω ⊂ Rn be open and bounded, 1 ≤ p < ∞, let F ⊂ (Lp (Ω; K), k · kp ). Then there are
equivalent:

(i) F is relatively compact in Lp (Ω), that is, F is compact.

(ii) F is bounded, and

Z
lim sup |f (x + h) − f (x)|p dx = 0 . (5.17)
h→0 f ∈F Ω

Condition (5.17) means that the passage to the limit h → 0 is uniform with respect to F .

Proof: Omitted. 2

50