Real Analysis MAA 6616

Lecture 24
p
L Spaces: Norm
Lp Space

Let E ⊂ Rn be a measurable set and consider the space of measurable functions on E that are
finite a.e.: 
M0 (E) = f : E −→ R : f measurable and finite a.e. on onE .
Consider the relation = on M0 (E) given by f ∼
∼ = g ⇐⇒ f = g a.e. on E. Then ∼ = is an
equivalence relation (exercise). Let M0 (E)/∼= be the quotient space. Hence M0 (E)/∼ = is the set
of all equivalence classes x = [f ] = {h ∈ M0 (E) : f ∼ = h}. It can be verified that M0 (E)/∼ = is
a vector space over R. For instance, if x = [f ] and y = [g] are in M0 (E)/∼ = and if a, b ∈ R,
then ax + by = z = [af + bg] is well defined. Indeed, if f ∼ = f 0 and g ∼
= g0 , then
(af + bg) ∼ = (af 0 + bg0 ).
When there is no ambiguity, we will identify x = [f ] ∈ M0 (E)/∼ = with its representative
f ∈ M0 (E) or with any other h ∈ M0 (E) such that h ∼ = f . For example if we say that
x = [f ] ∈ M0 (E)/∼ = is continuous, it means that the there exists a continuous function
h : E −→ R such that f ∼ = h.
Let p ∈ R such that 1 ≤ p < ∞. Note that if f = ∼ g, then |f |p ∈ L(E) if and only if
p p
|g| ∈ L(E). Denote by L (E) the space of functions Z (more precisely the space of equivalence
classes) in M0 (E)/∼
= that are p-integrable over E i.e. |f |p dx < ∞.
 ZE 
Lp (E) = [f ] ∈ M0 (E)/∼ = : |f |p dx < ∞ .
E
A function f ∈ M0 (E) is said to be essentially bounded if there exists M > 0 such that
|f (x)| ≤ M for almost all x ∈ E. Again if f ∼
= g, and f is essentially bounded then so is g.
Define L∞ (E) as the subspace of essentially bounded functions in M0 (E)/∼ =.
Lemma (1)
Let a, b ≥ 0 and 1 ≤ p < ∞. Then (a + b)p ≤ 2p−1 (ap + bp ).

Proof.
The inequality is trivial if either a = 0 or if p = 1. So assume that a > 0 and p > 1 and let b = ta with t ≥ 0. The
p p−1 p
inequality of the lemma is equivalent to (1 + t) ≤ 2 (1 + t ) for all t ≥ 0. Consider the function
p−1 p p
f (t) = 2 (1 + t ) − (1 + t) .

We have f (0) = 2p−1 − 1 > 0, f (1) = 0, and f (t) → ∞ as t → ∞. For t > 0, we have
f 0 (t) = p(2p−1 tp−1 − (1 + t)p−1 ). The equation f 0 (t) = 0 has a unique solution at t = 1. Therefore f (1) = 0 is a
global minimum of f and the lemma follows.

This lemma implies that for 1 ≤ p ≤ ∞, Lp (E) is a linear space. For instance, for
p p p−1 (|f |p + |g|p ) implies that
1Z ≤ p < ∞, and f , g ∈
Z L (E), then the Zinequality |f + g| ≤ 2
|f + g|p dx ≤ 2p−1 |f |p dx + 2p−1 |g|p dx < ∞.
E E E
For p = ∞, there exist Zf , Zg ⊂ E with measure 0 and Mf , Mg > 0 such that |f | < Mf on E\Zf
and |g| < Mg on E\Zg . Therefore |f + g| < Mf + Mg on E\(Zf ∪ Zg ) and so f + g ∈ L∞ (E).
Normed Spaces

Let X be a vector space over R (or over C). A function k·k : X −→ [0, ∞) is a norm on X if
I kαf k = |α| kf k for all f ∈ X and α ∈ R (or α ∈ C)
I kf k = 0 if and only if f = 0
I kf + gk ≤ kf k + kgk for all f , g ∈ X.
(X, k·k) is called a normed space.

Examples
Z
1. For f ∈ L1 (E), let kf k1 = |f | dx. Then k·k1 is a norm on L1 (E). Indeed, if kf k1 = 0, then |f | = 0 a.e. on E
E
1
and so f = 0 in L (E) (more precisely f ∼ = 0 and so [f ] = [0]). Now we verify the triangle inequality. Let
f , g ∈ L1 (E), then f and g areZfinite a.e. and so Zis f + g. FurthermoreZ|f + g| ≤ Z |f | + |g| a.e. We have therefore
kf + gk1 = |f + g| dx ≤ (|f | + |g|) dx ≤ |f | dx + |g| dx = kf k1 + kgk1
E E E E
Similarly, one can verify kαf k1 = |α| kf k1 .
n
2. Let B(0, R) be the closed ball with center 0 and raduis R > 0 in Rn : B(0, R) = {x ∈ R : |x| ≤ R}. Let
C0 (B(0, R)) be the space continuous function on B(0, R). For f ∈ C0 (B(0, R)), let kf kmax = max |f (x)|.
x∈B(0,R)
 
0
Then C (B(0, R)), k·kmax is a normed space. The verification is left as an exercise.
The space L∞ (E)

L∞ (E) is the space of essentially bounded functions: f ∈ L∞ (E) (or more precisely [f ] for the
equivalence relation ∼
=) if there exists a positive constant M such that |f (x)| ≤ M for almost all
x ∈ E. Define k·k∞ on L∞ (E) by
kf k∞ = inf{M : |f | ≤ M a.e. on E}.
Now we verify that k·k∞ is a norm.
We first prove that kf k∞ is an essential upper bound of f , i.e. |f | ≤ kf k∞ a.e. on E. To see
1
why, let n ∈ N and Mn = kf k∞ + , then Mn is an upper bound of f and so there exists a set
n
Zn ⊂ E with m(Zn ) = 0 and such that |f (x)| ≤ Mn for every x ∈ E\Zn . Let Z = ∞
S
n=1 Zn .
Then m(Z) = 0 and for every x ∈ E\Z, we have |f (x)| ≤ Mn for all n. Consequently
f (x) ≤ kf k∞ on E\Z. It follows that if kf k∞ = 0, then f = 0 a.e. on E.
Let α ∈ R∗ . Then ( )
M̂
kαf k∞ = inf{M̂ : |αf | < M̂ a.e.} = inf M̂ : |f | < a.e.
|α|

= inf{|α|M : |f | < M a.e.} = |α| inf{M : |f | < M a.e.} = |α| kf k∞

If f , g ∈ L∞ (E), then there exist Zf ⊂ E and Zg ⊂ E with measure 0 such that |f | ≤ kf k∞ on
E\Zf and |g| ≤ kgk∞ on E\Zg . Then for x ∈ E\(Zf ∪ Zg ) we have
|f (x) + g(x)| ≤ |f (x)| + |g(x)| ≤ kf k∞ + kgk∞
This means kf + gk∞ ≤ kf k∞ + kgk∞
Let E ⊂ Rn be measurable. For 1 < p < ∞ and f ∈ Lp (E) defines kf kp by
Z 1
p
kf kp = |f |p dx .
E
We will show that k·kp is norm on Lp (E). The first two conditions for a norm as easy to verify:
If α ∈ R, then kαf kp = |α| kf kp follows from the linearity of the integral; and if kf kp = 0,
Z
then |f |p dx = 0 and so f = 0 a.e. on E. To verify the triangle inequality, we need to use
E
Young’s and Hölder inequalities.
p
Define the conjugate of a number 1 < p < ∞ as the number q = so that 1 < q < ∞
p−1
and
1 1
+ =1
p q
The conjugate of 1 is ∞ and the conjugate of ∞ is 1

Proposition (1. Young’s Inequality)

Let a, b ≥ 0 and let p, q > 1 be a conjugate pair. Then
ap bq
ab ≤ +
p q
Proof.
First note that the inequality is trivial if a = 0 or b = 0. We assume ab > 0. We give two simple proofs:
1. The function et is convex on R: Thus for ever λ ∈ [0, 1] and for every t, s ∈ R, we have
λt+(1−λ)s t s
e ≤ λe + (1 − λ)e .
1
Young’s inequality follows by taking λ = so that 1 − λ = 1
p
and let t = log(ap ), s = log(bq ).
p
2. Young’s inequality is equivalent to
a 1 ap 1
≤ + .
bq−1 p bq q
p 1 p 1 a
Since q = = p(q − 1), then this inequality is equivalent to x ≤ x + where x = . Consider
p−1 p q bq−1
the function γ defined for x ≥ 0 by
1 p 1
γ(x) = x + − x
p q
1 0 p−1
We have γ(0) = , lim γ(x) = ∞ (p > 1) and γ (x) = x − 1 vanishes only at x = 1 where
p x→∞
1 1
γ(1) = + − 1 = 0. Therefore γ(x) ≥ 0 for all x ≥ 0.
p q
Theorem (1. Hölder’s Inequality )
Let E ⊂ Rn be measurable, 1 ≤ p < ∞ and let q be the conjugate of p. If f ∈ Lp (E) and
g ∈ Lq (E), then Z
|fg| dx ≤ kf kp kgkq
E

Proof. Z
If p = 1 and q = ∞, we have |g| ≤ kgk∞ a.e. on E and therefore |fg| dx ≤ kf k1 kgk∞ .
E
Suppose p > 1. If kf kp = 0 or if kgkq = 0, then |fg| = 0 a.e. on E and Hölder’s inequality follows.
f (x) g(x)
Now assume kf kp > 0 and kgkq > 0. Let F(x) = and G(x) = . Hence F ∈ Lp (E) with kFkp = 1 and
kf kp kgkq
Z
G ∈ Lq (E) with kGkq = 1. To prove Hölders inequality, it is enough to verify that |FG| dx ≤ 1.
E
Let Z = {x ∈ E : |F(x)| = ∞ or |G(x)| = ∞}. Then m(Z) = 0. For x ∈ E\Z, we can apply Young’s inequality to get
1 p 1 q
|F(x)G(x)| ≤ |F(x)| + |G(x)| .
p q
Therefore
1 1
Z Z Z
p q
|FG| dx ≤ |F| dx + |G| dx
E p E q E

1 p 1 q 1 1
≤ kFkp + kGkq = + =1
p q p q

A special case of Hölder’s inequality is the Cauchy-Schwarz

sZ inequality:
sZ If f , g ∈ L2 (E), then
Z
|fg| dx ≤ kf k2 kgk2 = |f |2 dx |g|2 dx
E E E
Theorem (2. Minkowski’s Inequality)
Let E ⊂ Rn be measurable, 1 ≤ p ≤ ∞, and f , g ∈ Lp (E), then f + g ∈ Lp and
kf + gkp ≤ kf kp + kgkp

Proof.
The case p = 1 and the case p = ∞ were considered earlier. Assume 1 < p < ∞. We know that Lp (E) is a vector space,
hence f + g ∈ Lp (E) if f , g ∈ Lp (E). If kf + gkp = 0 Minkowski is trivial. Assume kf + gkp > 0. We can write
p p−1 p−1 p−1
|f + g| = |f + g| |f + g| ≤ |f | |f + g| + |g| |f + g|
Now we apply Hölder’s inequality to |f | |f + g|p−1 and to |g| |f + g|p−1 to obtain
Z Z 1 Z Z 1
p−1 q(p−1) q p−1 q(p−1) q
|f | |f + g| dx ≤ kf kp |f + g| dx and |g| |f + g| dx ≤ kgkp |f + g| dx
E E E E
1 p−1
where q is the conjugate of p. Since (p − 1)q = p and = , it follows that
q p
Z h i Z  p−1
p p p
|f + g| dx ≤ kf kp + kgkp |f + g| dx .
E E
Minkowski’s inequality follows by rearranging this inequality.

Corollary (1)
Let E ⊂ Rn be measurable, 1 < p < ∞, and let F ⊂ Lp (E). If the family F is bounded, then
it is uniformly integrable over E ( F bounded in Lp (E) means that there exists M > 0, such that
kf kp ≤ M for all f ∈ F ).
Proof.
We need to prove that for any given  > 0, there exists δ > 0 such that for any measurable set A ⊂ E, with m(A)
 <δ,q we

Z
have |f | dx <  for all f ∈ F . Let M > 0 such that kf kp ≤ M for all f ∈ F . For a given  > 0, let δ =
A M
q
where q is the conjugate of p. Let A ⊂ E, with m(A) < δ. Since A has finite measure, then χA is in L (A) and the restriction
to A of any element in Lp (E) is in Lp (A). We apply Hölder inequality in the set A to f ∈ Lp (E) and χA to get
Z Z
|f | dx = |f | χA dx ≤ kf kA,p χA
A A A,q
where k·kA,p denotes the norm in Lp (A) to distinguish it from the norm k·kp = k·kE,p in Lp (E). We have
1
χA = (m(A)) q . Now for any f ∈ F we have kf kA,p ≤ kf kp ≤ M. It follows that
A,q
Z 1 1
|f | dx ≤ kf kA,p (m(A)) q ≤ M (m(A)) q < 
A

Corollary (2)
Let E ⊂ Rn be measurable with finite measure and let 1 ≤ p1 < p2 ≤ ∞. Then
Lp2 (E) ⊂ Lp1 (E). Furthermore, for every f ∈ L
p2 (E) we have
p2 −p1
 m(E) p1 p2 if p2 < ∞


kf kp1 ≤ C kf kp2 , with C =

 1
m(E) p1 if p2 = ∞

Proof.
First, consider the case p2 = ∞. Let f ∈ L∞ (E). Then there exists a set Z ⊂ E with m(Z) = 0 such that |f | ≤ kf k∞ on
Z
p p1
E\Z. Since m(E) < ∞, then |f | 1 dz ≤ kf k∞ m(E) and so kf kp ≤ C kf k∞ .
E 1
p2
Next if p2 < ∞, let p = > 1 and let q be the p-conjugate. If f ∈ Lp2 (E), then |f |p1 ∈ Lp (E). The function χE is in
p1
Lq (E) (since E has finite measure). We can therefore apply Hölder inequality to the pair |f |p1 and χE to get
Z 1 p2 −p1
p p
|f | 1 χE dx ≤ |f | 1 p χE q . By using χE q = m(E) q = m(E) p2 , we get from the above inequality
E
p2 −p1 Z 1 p2 −p1
p p p2 p p p p2
kf kp1 ≤ |f | 1 p m(E) = |f | 1 m(E)
1 E

Z  p1 p2 −p1
p p2
≤ |f | 2 m(E) p2 .
E
The estimate of the lemma follows by taking the p1 -root.

Remark (1)
In general, when m(E) = ∞ there is no inclusion between the different Lp (E) spaces (see
exercises)