122 8.

INTRODUCTION TO BANACH SPACES

3. The Lp Space
In this section we consider a space Lp (E) which resembles `p on many aspects. After general
concepts of measure and integral were introduced, we will see that these two spaces can be viewed
as special cases of a more general Lp space.
Definition 3.1. Given a measurable set E ⊂ Rn . For 0 < p < ∞, define the space Lp (E) and
the real-valued function k · kp on Lp (E) by
Z Z 1
p
p p p
L (E) = {f : f is measurable on E and |f | < ∞}, kf kp = |f | .
E E
The essential supremum of a measurable function f on E is defined by
ess sup f = inf{α ∈ (−∞, ∞] : m({f > α}) = 0}.
E

The space L∞ (E) and the real-valued function k · k∞ on L∞ (E) are given by
L∞ (E) = {f : f is measurable on E and ess sup |f | < ∞}, kf k∞ = ess sup |f |.
E

Functions in L∞ (E) are said to be essentially bounded.

The measurable function f in the definition of Lp (E) for 0 < p < ∞ can be complex-valued,
but functions in L∞ (E) are assumed to be real-valued. We leave it to the readers to check that
m(f > ess supE f ) = 0 for any f ∈ L∞ (E) (Exercise 3.1). In other words, f ≤ ess supE f and
|f | ≤ kf k∞ almost everywhere.
For any 0 < p ≤ ∞, two functions f1 , f2 in Lp (E) are considered equivalent if f1 = f2 almost
everywhere on E. The space of equivalence classes, still denoted by Lp (E), are called Lp (E) classes
or Lp (E) spaces.
Similar to `p , the space Lp (E) is a vector space for any 0 < p ≤ ∞. Indeed, kαf kp = |α|kf kp
for any scalar α, kαf kp = |α|kf kp and
f, g ∈ Lp (E) ⇒ |f + g|p ≤ (2 max{|f |, |g|})p ≤ 2p (|f |p + |g|p ),
f, g ∈ L∞ (E) ⇒ kf + gk∞ ≤ kf k∞ + kgk∞ .
The second line follows by observing that

|f | ≤ kf k∞ almost everywhere
⇒ |f + g| ≤ kf k∞ + kgk∞ almost everywhere.
|g| ≤ kgk∞ almost everywhere
When 1 ≤ p ≤ ∞, the function k · kp is a norm on Lp (E). This follows from the theorem below,
the proof for which is similar to that of `p .
1 1
Theorem 3.1. Given 1 ≤ p, q ≤ ∞ with p+q = 1. Let f, g be measurable functions on E ⊂ Rn .
(a) (Hölder’s Inequality for Lp ) If f ∈ Lp (E), g ∈ Lq (E), then f g ∈ L1 (E) and
kf gk1 ≤ kf kp kgkq .
(b) (Minkowski’s Inequality for Lp )
kf + gkp ≤ kf kp + kgkp .
 3. THE Lp SPACE 123

Proof. (a) The cases p = 1, q = ∞ and p = ∞, q = 1 are obvious. Consider 1 < p, q < ∞. If
kf kp = 0 or kgkq = 0, then f g = 0 almost everywhere on E, and the asserted inequality is obvious.
We may now assume 0 < kf kp , kgkq < ∞.
Let F = kffkp , G = kgkg
q
. By Young’s inequality,

|F |p |G|q kF kpp kGkqq

Z Z
1 1
|F G| ≤ + = + = + = 1,
E p q p q p q
ZE Z
kf gk1 = |f g| = kf kp kgkq |F G| ≤ kf kp kgkq .
E E
(b) The case p = 1 is obvious, and the case p = ∞ has been proved. Now we consider 1 < p < ∞.
p
Note that q = p−1 . Minkowski’s inequality follows easily from (a):
Z
kf + gkpp = |f + g|p
ZE Z
p−1
≤ |f + g| |f | + |f + g|p−1 |g|
E E
Z  p−1 Z 1 Z  p−1 Z 1
p p p p
p p p p
= |f + g| |f | + |f + g| |g|
E E E E
= kf + gkp−1
p (kf kp + kgkp ).


The special case p = q = 2 of the Hölder inequality is also known as the Cauchy-Schwarz
inequality. The assumption 1 ≤ p ≤ ∞ is necessary. For example, let E = [0, 1], f = χ[0, 1 ] ,
2
g = χ[ 1 ,1] . Then
2
 1  1
1 p 1 p 1− 1
kf kp + kgkp = + = 2 p < 1 = kf + gkp .
2 2
Corollary 3.2. Suppose 0 < p < q < ∞, m(E) < ∞. Then
 Z 1  Z 1
1 p
p 1 q
q
|f | ≤ |f | .
m(E) E m(E) E
In particular, Lq (E) ⊂ Lp (E).
q 1 1
Proof. Let r = q−p , then q/p + r = 1. Therefore,
Z Z  p Z 1 Z p
q q r q q−p
p p r q
|f | ≤ (|f | ) p 1 = |f | m(E) q .
E E E E
Then the corollary follows from
Z 1
p q−p 1
p −1
kf kp = |f | ≤ kf kq m(E) qp = kf kq m(E) p q .
E

124 8. INTRODUCTION TO BANACH SPACES

Example 3.1. Consider f (x) = xr , r 6= 0, defined on [0, ∞).

1 1
When r < 0, f ∈ Lp [1, ∞) if and only if p > − , f ∈ Lp [0, 1) if and only if 0 < p < − .
r r
/ Lp [1, ∞) for any p > 0, f ∈ Lp [0, 1) for any p > 0.
When r > 0, f ∈
This example shows that the assumption m(E) < ∞ is necessary in the above corollary, and
Lq (E) ( Lp (E) if 0 < p < q < ∞ and E = [1, ∞).
Example 3.2. The function log x belongs to Lp (0, 1) for any 0 < p < ∞ but it is not in L∞ (0, 1).
Theorem 3.3. (Riesz-Fisher) For any 1 ≤ p ≤ ∞, the space (Lp (E), k · kp ) is a Banach space.
Proof. Consider p = ∞ first. Note that convergence in L∞ (E) means uniform convergence
outside a set of measure zero.
Let {fn } be a Cauchy sequence in L∞ (E).
S For each n, m ∈ N, |fn − fm | ≤ kfn − fm k∞ except
on a set Zn,m of measure zero. Let Z = Zn,m , then Z has measure zero and
n,m∈N

|fn − fm | ≤ kfn − fm k∞ on E \ Z
In particular, for any x ∈ E \ Z, {fn (x)} converges. Let f (x) = limn→∞ fn (x) for x ∈ E \ Z and
set f (x) = 0 on Z. Then
fn → f uniformly on E \ Z.
This implies that fn converges to f in L∞ (E), and so L∞ (E) is complete.
Now we consider 1 ≤ p < ∞. By Theorem 1.3, we only have to show that every absolutely
convergent p
P∞series converges to some element in L (E).
Let k=1 fk be an absolutely convergent series. Then ∞
P
k=1 kfk kp = M is finite. Let
n
X n
X
gn = |fk |, sn = fk .
k=1 k=1
n
p
≤ M p for any n. For any x ∈ E,
P R
By Minkowski’s inequality, kgn kp ≤ kfk kp ≤ M . Thus E gn
k=1
the function gn (x) is increasing in n, and so gn converges pointwise to some function g : E → [0, ∞].
The function g is measurable and, by Fatou’s lemma,
Z Z
p
g ≤ lim inf gnp ≤ M p .
E n→∞ E
Therefore g is finite almost everywhere and g ∈ Lp (E). When g(x) is finite, ∞
P
k=1 fk (x) is absolutely
convergent. Let s(x) be its value, and set s(x) = 0 elsewhere. Then the function s is defined
everywhere, measurable on E, and
Xn
fk = sn → s almost everywhere on E.
k=1
Since |sn (x)| ≤ g(x) for all n, we have |s(x)| ≤ g(x), where hence s ∈ Lp (E) and |sn (x) − s(x)| ≤
2g(x) ∈ Lp (E). By the Lebesgue dominated convergence theorem,
Z
|sn − s|p → 0 as n → ∞.
E
This proves that ∞ p p
P
k=1 fk converges to s ∈ L (E), and thus proves completeness of L (E). 
 3. THE Lp SPACE 125

Theorem 3.4. If 1 ≤ p < ∞, then Lp (E) is separable.

Proof. Consider E = Rn . Consider the collection of cubes of the form [k1 , k1 + 1] × . . . ×
[kn , kn + 1], k1 . . . , kn ∈ Z. Bisect each of these cubes into 2n congrument subcubes, and repeat this
process. The collection of all these cubes are called dyadic cubes. Let D be the set of finite linear
combinations of characteristic functions on these dyadic cubes with rational coefficients. Clearly D
is countable. All we need to prove is that D is dense in Lp (Rn ). That is, given f ∈ Lp (Rn ), there
exists a sequence fk ∈ D such that kfk − f kp → 0 as k → ∞.
It suffices to consider the case f ≥ 0 since
f = f + − f −, kfk − f kp ≤ kfk+ − f + kp + kfk− − f − kp (by Minkowski’s inequality).
In fact, it suffices to consider the case f ≥ 0 with compact support since
Z Z
|fk − f |p = lim |fk − f |p
Rn m→∞ [−m,m]n

Let {gk } be an increasing sequence of nonnegative simple functions such that gk % f , f ≥ 0

has compact support. Then, by the monotone convergence theorem,
Z
p n
gk ∈ L (R ), |gk − f |p → 0 as k → ∞.
Rn
Therefore, it suffices to consider the case f ≥ 0, f is a simple function with compact support.
N
P
For a simple function f = ak χEk ,
k=1
Z N Z
X
p
|f − g| = |ak − g|p for any g ∈ Lp (Rn ).
Rn k=1 Ek

From this observation, it suffices to consider the case when f is the characteristic function of some
bounded measurable set E. There exists a Gδ set G containing E with m(G \ E) = 0, so that we
may consider only the case E being a Gδ set.
∞
T
Let E = Ok , O1 ⊃ O2 ⊃ · · · is a nested sequence of bounded open sets. Then, by the
k=1
monotone convergence theorem,
Z
|χOk − χE |p → 0 as k → ∞.
Rn
Therefore, it suffices to consider f = χO , where O is a bounded open set. But in this case,
∞
χck for some dyadic cubes ck . This proves D is dense in Lp (Rn ).
P
f=
k=1
Now consider arbitrary measurable set E ⊂ Rn . Let D0 = {g · χE : g ∈ D}. Then D0 is a
countable set consisting of finite linear combinations of characteristic functions on dyadic cubes
which intersect with E and with rational coefficients.
Given f ∈ Lp (E). Let f˜ = f on E, f˜ = 0 on Rn \E. Choose {fk } ⊂ D such that Rn |fk − f˜|p → 0
R
as k → ∞. Then Z Z
˜ p
|fk · χE − f | = |fk · χE − f˜|p → 0 as k → ∞.
E Rn
This proves that D0 is dense in Lp (E). 
126 8. INTRODUCTION TO BANACH SPACES

Given h ∈ Rn . Let τh f (x) = f (x + h) be the translation operator. Similar to the case L1 (Rn ),
we have continuity of variable translations with respect to k · kp :
Theorem 3.5. If 1 ≤ p < ∞ and f ∈ Lp (Rn ), then
lim kτh f − f kp = 0.
h→∞
Proof. Let Cp be the collection of Lp (Rn ) functions
satisfying this property. It follows easily
from the Minkowski inequality that it is a subspace of Lp (Rn ).
Given E ⊂ Rn with m(E) < ∞. By the Lebesgue dominated convergence theorem,
Z
|χE (x + h) − χE (x)|p dx = m(E \ Eh ) + m(Eh \ E) → 0 as h → 0,
Rn
where Eh = E − h = {e − h : e ∈ E}. This shows that χE ∈ Cp , and as well as simple functions
in Lp (Rn ). Suppose f ∈ Lp (Rn ) is nonnegative. Choose simple functions fk ≥ 0, fk % f . Then
fk ∈ Lp (Rn ) and, by the monotone convergence theorem, kfk → f kp → 0 as k → ∞. Therefore,
kτh f − f kp ≤ kτh f − τh fk kp + kτh fk − fk kp + kfk − f kp
≤ kτh fk − fk kp + 2kf − fk kp
Let h → 0, then let k → ∞, we find that lim suph→∞ kτh f − f kp = 0. This proves that f ∈ Cp . This
actually implies Cp = Lp (Rn ) since any f ∈ Lp (Rn ) is the difference of two nonnegative measurable
functions in Lp (Rn ). 
Exercises.
3.1. Given any f ∈ L∞ (E). Show that m({f > ess supE f }) = 0.
3.2. Use the generalized Young’s inequality in Exercise 2.2 to formulate a generalization of
Hölder’s inequality for Lp (E).
3.3. Suppose m(E) < ∞. Show that kf k∞ = limp→∞ kf kp . How about if m(E) = ∞?
3.4. Let f be a real-valued measurable function on E. Define the essential infimum on E by
ess inf f = sup{α ∈ (−∞, ∞] : m({f < α}) = 0}.
E
Show that, if f ≥ 0, then ess inf E f = 1/ ess supE (1/f ).
3.5. Consider Lp (E) with 0 < p < 1. Verify that ρp (f, g) = |f − g|p is a metric on Lp (E).
R
E
Prove that (Lp (E), ρp ) is a complete separable metric space.
3.6. Given 0 < p < ∞, fn ∈ Lp (E). Suppose fn converges to f almost everywhere. Prove that
each of the following conditions implies kfn − f kp → 0 as n → ∞.
(a) There exists some g ∈ Lp (E) such that |fn | ≤ g for any n.
(b) kfn kp → kf kp as n → ∞.
3.7. For what kind of f ∈ Lp (E) and g ∈ Lq (E), p1 + 1q = 1, do we have equality for the Hölder
inequality? For what kind of f, g ∈ Lp (E) do we have equality for the Minkowski inequality?
3.8. Consider 1 < p < ∞. Give a proof for the Minkowski inequality using convexity of xp , and
without using Hölder’s inequality.
3.9. Show that L∞ (E) is not separable whenever m(E) > 0.