42 The Elements of Integration

In fact, since f + - f f = f; - f2, it follows that f + + f2 = f2 +

f If we apply Corollary 4.7(b), we infer that

CHAPTER 5
Since all these terms are finite, we obtain
Integrable Functions
f belongs on to R by

(5.2) A(E) = f d',

SE
In Definition 4.4 we defined the integral of each function in
M + = M + (X, X) with respect to a measure jt and permitted this then A is a charge.
integral to be +oo. In this chapter we shall discuss the integration PROOF. Since f + and f - belong to M +, Corollary 4.9 implies that
of measurable functions which may take on both positive and negative
the functions A+ and A-, defined by
real values. Here it is more convenient to require the values of the
functions and the integral to be finite real numbers.
A+(E) = f+ d [z, A - (.E) = f-- dd,
E E
5.1 DEFINITION. The collection L = L(X, X, p) of integrable (or
summable) functions consists of all real-valued X-measurable functions are measures on X; they are finite because f e L. Since A = A + - A -,
f defined on X, such that both the positive and negative parts f +, f -, it follows that A is a charge. Q.E.D.
off have finite integrals with respect to jc. In this case, we define the
The function A defined in (5.2) is frequently called the indefinite
integral off with respect to p to be
integral off (with respect to Since A is a charge, if (En) is a disjoint
p).

(5.1) ffdd= 5 f+d - 5 f- dd. sequence in X with union E, then

00

If E belongs to X, we define f dp = f dd.

fE n= 2 lEn
.f fd[t = f + d1i -- .f - dp We refer to this relation by saying that the indefinite integral of a function
E E E
in L is co un tabl y additive,
Although the integral of f is defined to be the difference of the The next result is sometimes referred to as the property of absolute
integrals off +, f -, it is easy to see that if f = f, - f2 where f2 , f2 are integrability of the Lebesgue integral. The reader will recall that,
any nonnegative measurable functions with finite integrals, then although the absolute value of a (proper) Riemann integrable function
is Riemann integrable, this may no longer be the case for a function
5fdp = fii dFi - f .fi dFi which has an improper Riemann integral (for example, consider
41 f (x) = x -' sin x on the infinite interval 1 c x < +ao) .
 Integrable Functions 43 44 The Elements of Integration

5.3 THEOREM. A measurable function f belongs to L if and only if If f and g belong to L, then IfI and IgI belong to L. Since
If belongs to L. In this case
I
If + SI < If I + I SI it follows from Corollaries 4.7 and 5.4 that f + g
belongs to L. To establish the desired relation, we observe that
(5.3) I ffdd fIfId. f +9 = (f+ + g+) - (f- +9-).
PROOF. By definition f belongs to L if and only if f + and f' belong Since f + + g+ and f - + g- are nonnegative integrable functions, it
to M + and have finite integrals. Since If I + = If I =.f + + f - and follows from the observation made after Definition 5.1 that
If I = 0, the assertion follows from Lemma 4.5(a) and Corollary
4.7(b). Moreover,
5(1+ S) dµ = 5(1 + g+) dµ - f(f- + g-) dµ.
ffdpl Iff+ dp - Jfdal If we apply Corollary 4.7(b) and rearrange the terms, we obtain
f f + dp + f f- dp = f III CIµ f(f+g)d/ = f f + dw - f f- dp + fg + du - fg - dw
5.4 COROLLARY. If f is measurable, g is integrable, and If I < IgI,
then f is integrable, and
= f f dµ + fg dµ. Q.E.D.

5IfId5IgId. We shall now establish the most important convergence theorem

for integrable functions.
PROOF. This follows from Lemma 4.5(a) and Theorem 5.3. Q.E.D.

We shall now show that the integral is linear on the space L in the 5.6 LEBESGUE DOMINATED CONVERGENCE THEOREM. Let (fn) be a
following sense. sequence of integrable functions which converges almost everywhere to a
real valued measurable function f. If there exists an integrable function
5.5 THEOREM. A constant multiple of and a sum f + g of functions
g such that I fn I c g for all n, then f is integrable and
in L belongs to L and

5afd_-affd,, 5(f+g)d=5fd,+5gd,. (5.4) J f dµ = lim fIn dµ.

PROOF. If a = 0, then of = 0 everywhere so that By redefining the functions fn, f on a set of measure 0 we
PROOF.
may assume that the convergence takes place on all of X. It follows
5afdi ,c = 0 = a f dp . from Corollary 5.4 that f is integrable. Since g + fn >, 0, we can
apply Fatou's Lemma 4.8 and Theorem 5.5 to obtain
If a > 0, then (af ) + = a f + and (af ) - = a f -, whence

5afdl,c = faf+ dp - 5af- dp fg dµ + f f dµ = f(g +J) dµ < lim inf f (g + fn) dµ

= lim inf (fg dµ + f fn dµ)
=a f+ d - 5f- dp. = a fdfL.
= 5gd,t + lim inf f fn dµ .
The case a < 0 is handled similarly.
 integrable Functions 45 46 The Elements of Integration

Therefore, it follows that PROOF. Let (t7) be a sequence in [a, b] which converges to to, and
apply the Dominated Convergence Theorem to the sequence (ft)
(5.5) ffdit c lim inf fn d/L . defined by f7t(x) = f (x , t,) for x e X. Q.E.D.

Since g - fn >, 0, another application of Fatou's Lemma and Theorem 5.8 COROLLARY. If the function t -*f (x, t) is continuous on [a, b]
5.5 yields for each x c- X, and if there is an integrable function g on X such that
f (x , t) c g(x), then the./unction F defined by
fgdp - f f dµ = f (g - f) dµ < lim inf f (g - fn) dµ
(5.8) F(t) = 5 f (x, t) dd(x)
= fg dµ - lim sup J J n dµ,
is continuous for t in [a, b].
from which it follows that PROOF. This is an immediate consequence of Corollary 5.7. Q.E.D.

(5.6) lim sup fin < f f dµ. 5.9 COROLLARY. Suppose that for some to c- [a, b], the function
x -*f (x, to) is integrable on X, that afl at exists on X x [a, b], and that
Combine (5.5) and (5.6) to infer that there exists an integrable function g on X such that

f f dµ = 1i1T1 f fn CIµ. Q.E.D.

Then the function F defined in Corollary 5.8 is differentiable on [a, b] and

DEPENDENCE ON A PARAMETER
dr
(t) =
dr J
l(x, t) 5(x. t)
Frequently one needs to consider integrals where the integrand
depends on a real parameter. We shall show how the Lebesgue PROOF. Let t be any point of [a, b] . If (tn} is a sequence in [a, b]
Dominated Convergence Theorem can be used in this connection. converging to t with to t, then
For the remainder of this chapter we shall let f denote a function
defined on X x [a, b] to R and shall assume that the function x -* of tn) - f(x, t}
at(
x t) = limf(x,

tn- t x c- X.
f(x, t) is X-measurable for each t e [a, b] . Additional hypotheses will
be stated explicitly. Therefore, the function x -* (aflat)(x, t) is measurable.
5.7 COROLLARY. Suppose that for some to in [a, b] If x E X and t c- [a, b], we can apply the Mean Value Theorem (see
Reference [1], page 210) to infer the existence of a s, between to and t
(5.7) f (x , to) = lim f (x , t } such that
t to
a
for each x e X, and that there exists an integrable function g on X such .f(x, t} - .f(x, to) = (t - to) (x, s2).
at
that I t) I c g(x). Then
Therefore we have
5 f (x, to) dd(x) = lim 5f(x, t) dp(x)
t "to I.f(x, 0 I < Il(x, ro)I + I t - tol 9W9
 Integrable Functions 47 48 The Elements of Integration

which shows that the function x - f (x, t) is integrable for each t in Therefore we have
[a, b]. Hence, if t,, t, then b
'(t) dt = H(b) - H(a)
F(t)
_ f(x,tn)-f(x,t)d
tn-- t
F(tn) a
t-t n `(x.
} [h(x, b) - h(x, a)] dd(x)
Since this integrand is dominated by g(x), we may apply the Dominated
Convergence Theorem to obtain the stated conclusion. Q.E.D. _f Ifa

Ifa f (X I
t) dt] dp(x) .
Q.E.D.
5.10 COROLLARY. Under the hypotheses of Corollary 5.8,
The interchange of the order of (Lebesgue) integrals will be considered
fF(t) dt = [ff(xt)d(x)] dt in Chapter 10.
a fa

f(x, t) dt dp(x), EXERCISES

[fa

5.A. If f E L(X, X, p) and a > 0, show that the set {x E X : I I

a} has finite measure. In addition, the set {x E X : f(x) O} has

PROOF. Recall that if 9) is continuous on [a, b] then
cr-finite measure (i.e., it is the union of a sequence of measurable sets
d with finite measure).
dt
ftcsds=ct),
a
a<t<b.
5.B. If f is an X-measurable real-valued function and iff(x) = 0 for
pt-almost all x in X, then f E L(X, X, p) and
Let h be defined on X x [a, b] by
5fdtz = 0.
h(x,t) = Jtf(x,s)ds.
a
5.C. If f E L(X, X, l,c) and g is an X-measurable real-valued function
It follows that (ahlat)(x, t) = f(x, t). Since this Riemann integral such thatf(x) = g(x) almost everywhere on X, then g E L(X, X, p) and
exists, it is the limit of a sequence of Riemann sums; hence the map
x -* h(x, t) is measurable for each t. Moreover, since l t) l c g(x), 5fdd = fgd/L.
we infer that I h(x, t) I c g(x)(b - a), so that the function x - h(x, t)
is integrable for each t c [a, b]. Let H be defined on [a, b] by 5.D. If f c L(X, X, p) and s > 0, then there exists a measurable
simple function 9) such that
H(t) = fh(x,t)dit(x);
ff_d<s.
it follows from Corollary 5.9 that
5.E. If f EL and g is a bounded measurable function, then the
dH productfg also belongs to L.
(t) = f ah (x, t) dµ(x) = ff(x , t) dµ(x) = F(t). 5.F. If f belongs to L, then it does not follow that f2 belongs to L .
 Integrable Functions 49 50 The Elements of Integration

5.G. Suppose that f is in L(X, X, p) and that its indefinite integral is 5.M. Show that the conclusion in the Exercise 5.L may fail if the
hypothesis t(X) < +ao is dropped.
A(E) = f dp, E E X. 5.N. Let f, = n x[o.1/n], where X = R, X = B, and is Lebesgue
E
measure. Show that the condition IfI c g cannot be dropped in the
Show that A(E) >, 0 for all E E X if and only if f (x) > 0 for almost all Lebesgue Dominated Convergence Theorem.
x E X. Moreover, A(E) = Q for all E if and only if f(x) = Q for almost
5.0. If fn EL(X, X, p), and if
allxEX. Go

5.H. Suppose that f1 and f2 are in L(X, X, p) and let Al and A2 be I fn I dd < +co,
their indefinite integrals. Show that A1(E) = A2(E) for all E E X if n=1

and only if f1(x) = f2(x) for almost all x in X. then the series > fn(x) converges almost everywhere to a function f in
5.1. If f is a complex-valued function on X such that Ref and Imf L(X, X, p). Moreover,
belong to L(X, X, p), we say that f is integrable and define ffdii _ j'fnd.
n=1
ffdd= 5Refdi, + i Imfdd. S.P. Let fn c- L(X, X, µ), and suppose that (fn) converges to a
function f Show that if
Let f be a complex-valued measurable function. Show that f is
integrable if and only if If I is integrable, in which case lim fin - .f I dµ = 0, then f If I dw = lim fin I dµ
f dµ f If I dw- S.Q. If t > 0, then
J +COe

[Hint: If Jfdjt = r e1° with r, 0 real, consider g(x) = er1° f(x).1

txdx= _.
t
a

5.J. Let (f7) be a sequence of complex-valued measurable functions Moreover, if t > a > 0, then e r tx c e_. Use this and Exercise
which converges to f. If there exists an integrable function g such 4.M to justify differentiating under the integral sign and to obtain
that f, c g, show that the formula
+cc
xne-x dx = n!
5 f dj = lim
5 f n dp . JO

5.R. Suppose that f is defined on X x [a, b] to R and that the

5.K. Let X = N, let X be all subsets of N, and let jt be the counting function x - f (t, x) is X-measurable for each t E [a, b]. Suppose
measure on X. Show that f belongs to L(X, X, p) if and only if the that for some to and t1 in [a, b] the function x - f(x, to) is integrable
series J f (n) is absolutely convergent, in which case
on X, that (aflat)(x, t1) exists, and that there exists an integrable
Go
function g on X such that
f dd = f (n)
J n=1
f(x, t} - f(x, t1)
t g(x)
5.L. If (fn) is a sequence in L(X, X, p) which converges uniformly t1
on X to a function f, and if p(X) < +ao, then for x E X, and t E [q, b] , t t1. Then

5fdµ = lim f fn dµ.

[jj f( x,tdx)=
) (trt at(x,t1)dp(}
x.
 Integrable Functions 51

5.S. Suppose the function x - f(x, t) is X-measurable for each

t E R, and the function t - f (x, t) is continuous on R for each X E X.
In addition, suppose that there are integrable functions g, h on X such
that l t) l c g(x) and such that the improper Riemann integral
CHAPTER 6
I t) dt < h(x).
f +- CO
co
The Lebesgue spaces Lp
f + co [f f(x, t) dd(x) I dt = f [f + c* f(x, t) dt] dp(x),
CO

where the integrals with respect to t are improper Riemann integrals.

5.T. Let f be an X-measurable function on X to R. For n E N, let
It is often useful to impose the structure of a Banach space on the
(f) be the sequence of truncates of f (see Exercise 2.K). If f is
set of all integrable functions on a measure space (X, X, p) . In
integrable with respect to jt, then
addition, we shall introduce the Lp, 1 c p c oo, spaces which occur
frequently in analysis. Aside from the intrinsic importance of these
ff dµ = lim f fn dµ. spaces, we examine them here partly to indicate applications of some
Conversely, if of the results in the earlier sections.
SUP f I .fnj dp < +ao, 6.1 DEFINITION. If V is a real linear (= vector) space, then a real-
then f is integrable. valued function N on V is said to be a norm for V in case it satisfies
(i) N(v) 0 for all V E V;
(ii) N(v) = 0 if and only if v = 0;
(iii) N(av) = I al N(v) for all V E V and real a;
(iv) N(u+v) c N(u)+N(v)forallu,vEV.
If condition (ii) is dropped, the function N is said to be a semi-norm or a
pseudo-norm for V. A normed linear space is a linear space V together
with a norm for V.
6.2 EXAMPLES. (a) The absolute value function yields a norm for
the real numbers.
(b) The linear space Rn of n-tuples of real numbers can be normed
by defining
NI(u1, . .. , un) = lull + + lunl ,
Np(u1, . , u ) = {1u11' + . . + IunIp 1Ip p
.
1,
1ump}1/p,

N,,(u1, .. . , un) = sup tlu11 , .. . , lung}.

52
 The Lebesgue Spaces Lp 53 54 The Elements of Integration

It is easy to check that N1 and Nc are norms and that Np satisfies (i), 6.5 LEMMA. The space L(X, X, p) is a linear space under the
(ii), (iii). It is a consequence of Minkowski's Inequality, which will operations defined by
be proved subsequently, that N. satisfies (iv).
(c) The linear space 11 of all real-valued sequences u = (un) such lJ + 9)(X) = f (X) + 9(X), (af)(x) = of (x), x E X,
that N1(u) = I l < +ao is a normed linear space under N1 . and NN is a semi-norm on L(X, X, p). Moreover, N ,,,(f) = 0 if and
Similarly, if 1 c p < co, the collection 1p of all sequences such that only if f (x) = Q for jc-almost all x in X.
NN(u) = 11 lunlP}1rp < +oo is normed by Np.
(d) The collection B(X) of all bounded real valued functions on X PROOF. It was seen in Theorem 5.5 that L = L(X, X, p) is a linear
is normed by space under the indicated operations. It is clear that Nu(f) > 0 for
Non -sup {Ir(x) I : x E X}. f EL, and that
In particular, the linear space of continuous functions on X = [a, b] is Nu(at) = f l all dp = l al f I.fI dp = I at Nu0')
normed.
All the preceding examples have been proper norms on a linear Moreover, it follows from the Triangle Inequality that
space. There are also semi-norms on a linear space that are of interest.
The following are some examples.
Nu(f + S) = 5f+gI dp _< 5(fI + I dp
6.3 EXAMPLES. (a) On the space Rn, consider the semi-norm
No(u1, ... , un) = sup {lull , ... , l= lunl}-
= f If I dw+ f I SI dw=Nub+Nu(S)
HereNa(u1,...,un) = 0 if and only if u2 = un = 0. Hence Nu is a semi-norm on L, and it follows from Corollary 4.10 that
(b) On the linear space C[0, 1] of continuous functions on [0, 1] to Nu(f) = 0 if and only iff(x) = 0 for almost all x. Q.E.D.

R, define the semi-norm In order to make L(X, X, µ) into a nonmed linear space, we shall
N0(1) = sup { If (x) l :0cx identify two functions that are equal almost everywhere; that is, we use
equivalence classes of functions instead of functions.
Here N0(f) = 0 if and only if f (x) vanishes for 0 c x c I.

(c) On the linear space of functions on [a, b] to R which have con- 6.6 DEFINITION. Two functions in L = L(X, X, p) are said to be
tinuous derivatives, consider the semi-norm p.-equivalent if they are equal jc-almost everywhere. The equivalence
N0(f)=sup{lf'(x) l :a <x <b}. class determined by f in L is sometimes denoted by [f] and consists of
the set of all functions in L which are ji-equivalent to f . The Lebesgue
Here N0(f) = 0 if and only if f is constant on [a, b].
space L1 = L1(X, X, fit) consists of all jt-equivalence classes in L. If
6.4 DEFINITION. Let (X, X, p) be a measure space. If f belongs to [ f ] belongs to L1, we define its norm by
L(X, X, p), we define
Ng(f) =ff l dd. Mill = f If I dp.
It will be shown that N. is a semi-norm on the space L(X, X, p) 6.7 THEOREM. The Lebesgue space L1(X, X, fit) is a normed linear
space.
 The Lebesgue Spaces L,p 55 56 The Elements of Integration

It is understood, of course, that the vector operations in L1

PROOF. space. It is understood that the vector operations between the
are defined by equivalence classes in Lp are defined pointwise: the sum of the equiv-
a[f] = [af], [f] + [g] _ [f+g], alence classes containing f and g is the equivalence class containing
f + g and similarly for the product cf.
and that the zero element of L1 is [0] . We shall check only that
In the special case where jt is the counting measure on all subsets of
equation (6.1) gives a norm on L1. Certainly II U] II 1 % 0 and II [0] II 1 =
N, the Lp-spaces can be identified with the sequence spaces 4, of
Moreover, if II[All 1 = 0 then
Example 6.2(c). In this case, each equivalence class contains one
0.

element. It is frequently enlightening to interpret assertions about

5 fI dd = 0,

general L.-spaces by considering the somewhat simpler /.-spaces.

so f (x) = 0 for µ-almost all x. Hence [f} = [0]. Finally, it is easily In order to establish that (6.3) yields a norm on Lp, we shall need
seen that properties (iii) and (iv) of Definition 6.1 are satisfied. There- the following basic inequality.
fore 1111 1 yields a norm on L1. Q.E.D.
6.9 HoLDER's INEQUALITY. Let f c Lp and g c Lq where p > 1 and
It should always be remembered that the elements of L1 are actually
(I1p) + (l/q) = 1. Then fg E L1 and Ilfg111 c 11f 11p11 g11q
equivalence classes of functions in L L. However, it is both convenient
PROOF. Let a be a real number satisfying 0 < a < 1, and consider
and customary to regard these elements as being functions, and we
the function 9) defined for t > 0 by
shall subsequently do so. Thus we shall make reference to the
equivalence class [f] by referring to "the element f of L1," and we shall 9p(t) = at - ta.
write Ilf II 1 in place of II [f] 11 1.
It is easy to check that 9)'(t) < 0 for 0 < t < 1 and p'(t) > 0 for t > 1.
It follows from the Mean Value Theorem of calculus that p(t) >, 9)(1)
THE SPACES Lp, I c p< +oo and that p(t) = 9)(1), if and only if t = I. Therefore we have

We now wish to consider a family of related normed linear spaces of to C at + (1 - a) , t'0.

equivalence classes of measurable functions. If a, b are nonnegative, and if we let t = alb and multiply by b, we
6.8 DEFINITION. If 1 c p < oo, the space Lp = Lp(X, X, p) con- obtain the inequality
sists of all jt-equivalence classes of X-measurable real-valued functions aabl ra C as + (I - a)b,
f for which If IP has finite integral with respect top over X. Two
functions are p-equivalent if they are equal p-almost everywhere. We where equality holds if and only if a = b.
set Now let p and q satisfy 1 < p < oo and (lip) + (I fq) = 1 and take
a = l/p. It follows that if A, B are any nonnegative real numbers,
lif lip = f f I dpll'p.
P then
AP
If p = 1, this reduces to the norm introduced previously on the AB + $q .
p q
space L1 of equivalence classes of integrable functions. We shall show
subsequently that if 1 c p < oo, then Lp is a normed linear space and that the equality holds if and only if AP = B.
under (6.3), and is complete under this norm ; thus L. is a Banach Suppose that fE Lp and geLQ, and that IIfiIn 0 and So 0.
 The Lebesgue Spaces L,p 57 58 The Elements of Integration

The product fg is measurable and (6.4) with A = I(x)/Iv and f + hI ° ELq. Hence we can apply Holder's Inequality to infer that
B = S(x)Igq implies that dll`ljQ

f I.fI If + h Ip-1 dp _< MfM{5f+hl (p-1)4

J
Ilf Ilp 11 g9 pJ IIPp j9ll gll9Q Ilf Ilp Ilf + h II PP19

Since both of the terms on the right are integrable, it follows from If we treat the second term on the right in (6.7) similarly, we obtain
Corollary 5.4 and Theorem 5.5 that fg is integrable. Moreover, on Ilf + h II.J _< MfM IIv Ilf + hIIPP/9 + IIhIIP Ilf + h PP/9
integrating we obtain
lJ P + h D} Mf+ h IIDD/9'
hIIDD/9'

MfgM1 ! I

MfMMMa p+v- If A = IIf + hII v = 0, then equation (6.6) is trivial. If A =A 0, we

can divide the above inequality by AD/9; since p - p/q = 1, we obtain
which is Holder's Inequality. Q.E.D.
Minkowski's Inequality. Q.E.D.
Holder's Inequality implies that the product of a function in Lp and It is readily seen that the space L. is a linear space and that formula
a function in Lq is integrable when p > 1 and q satisfies the relation (6.3) defines a norm on L. The only nontrivial thing to be checked
(1 /p) + (1 /q) = 1 or, equivalently, when p + q = pq. Two numbers here is the inequality 6.1(iv) and this is Minkowski's Inequality. We
satisfying this relation are said to be conjugate indices. It will be noted shall now show that L. is complete under this norm in the following
that p = 2 is the only self-conjugate index. Thus the product of two sense.
functions in L2 is integrable.
6.12 DEFINITION. A sequence (f7) in Lp is a Cauchy sequence in
6.10 CAUCHY---BUNYAKOVSKII---SCHWARZ INEQUALITY. If f and g LP if for every positive number s there exists an M(e) such that if
belong to L2, thenfg is integrable and m , n >, M(e) , then I- f7t 11 p < e. A sequence (f7) in L. is con-
vergent to f in L. if for every positive number s there exists an N(s)
(6.5) I ffgdµl 5 f I .15I dw < IMINKOWSKI's
I I f112 1196- such that if n >, N(s), then Ilf - f7t Il p < e. A normed linear space is
complete if every Cauchy sequence converges to some element of the
6.11 MINKOWSKI'S INEQUALITY. If f and h belong to L, p > 1, space.
then f + h belongs to Lp and
6.13 LEMMA. If the sequence (f7t) converges to f in L, then it is a
(6.6) Ilf + hlln s If Ilp + Ilh Ilp Cauchy sequence.
PROOF. The case p = 1 has already been treated, so we suppose PROOF. If m, n >, N(s/2), then
p > 1. The sum f + h is evidently measurable. Since

If + hID 5 [2 sup {I.fI ,IhI}lD 5 2D{IfID + IhIp} MI -JmP 2s MffMD

Hence we have
it follows from Corollary 5.4 and Theorem 5.5 that f + h E Lp . More-
over, Mim -I IIn -< Mim - flla + MI -f llo < E Q.E.D.

We shall now show that every Cauchy sequence in Lp converges in

(6.7) If + hID = If + h If + hip-1
_< IfI If + hIP-1 + h If + hIP-1.
LP to an element. This result is sometimes called the Riesz-Fischer
Since f + h e Lp, then If + hip EL1; since p = (p - 1)q it follows that Theorem.
 The Lebesgue Spaces L,p 59 60 The Elements of Integration

6.14 COMPLETENESS THEOREM. If 1 4, p < oo, then the space L. is a implies that f E Lp . Since If - gk I p c 2p gp , we infer from the
complete normed linear space under the norm Dominated Convergence Theorem that 0 = lim Ilf - gk II P, so that
1Tp (gk) converges in Lp to f.
MI Mp
If I p p In view of (6.8), if m > M(s) and k is sufficiently large, then

PROOF. It has been stated that Lp is a normed linear space. To JIfm gklpd <sp.
establish the completeness of Lp, let (f,,) be a Cauchy sequence relative
to the norm II lip . Hence, if s > 0 there exists an M (e) such that if Apply Fatou's Lemma to conclude that
m, n > M(s), then

(6.8) Ifm - fnI p dd = Vm - fnIIpp < e . f Ifm - .f' IP dj'G C lim in f

k" co
I.fm - gk I p dtZ C ep

whenever m > M(e). This proves that the sequence (f7) converges to
There exists a subsequence (gk) of (fn) such that II9k11 - gk II p < 2 -k f in the norm of Lp. Q.E.D.
for k E N. Define g by
A complete normed linear space is usually called a Banach space
(6.9) g(x) = I + co I gk + 1(x) gk(x) I , Thus the preceding theorem could be formulated : the space Lp is a
k
Banach space under the norm given in (6.3).
so that g is in M + (X, X). By Fatou's Lemma, we have
p
J I gI p dd c lim inf Ig1l + Igk+1- Al dd. THE SPACE LOD
n " Co
k=
We shall now introduce a space which is related to the Lp-spaces.
Take the pth root of both sides and apply Minkowski's Inequality to
obtain 6.15 DEFINITION. The space Lc = Lc (X, X, p) consists of all the
1/p n
equivalence classes of X-measurable real-valued functions which are
I gl P dp c liminf llglllp + llgk+l - gkllp
n"co k=1 almost everywhere bounded, two functions being equivalent when they
Mg1Mp + 1. are equal pc-almost everywhere. If f c L. and N E X with p(N) = 0,
we define
Hence, if E = {x E X : g(x) < +ao}, then E E X and p(X 1 E) = 0.
S(N) = sup {I f(x) I : x 0 N}
Therefore, the series in (6.9) converges almost everywhere and g XE
and
belongs to Lp.
We now define f on X by (6.10) lif 11 Go = inf IS (N) : N E X, µ(N) = 0}.

An element of Lo is called an essentially bounded function.

f (X) = g1(x) + 2, {g k + 1(x)
k=1
- gk(x)} , x c- E,
It follows (see Exercise 6.T) that if f E Lam, then If(x)I < jfor
_0, x0E. almost all x. Moreover, if A < Ilf 11 co , then there exists a set E with
Since I9 I C 191 I +
Ek-1
j= Ig+' - 9, I g and since (90 converges
almost everywhere to f, the Dominated Convergence Theorem 5.6
, positive measure such that If(x)I , A for X EE. It is also clear that
the norm in (6.10) is well-defined on L,,.
 The Lebesgue Spaces Lp 61 62 The Elements of Integration

6.16 THEOREM. The space Lc is a complete normed linear space C.C. Let N be a norm on a linear space V and let d be defined for
under the norm given by formula (6.10). u, v c- V by d(u, v) = N(u - 6). Show that d is a metric on V; that
is, (i) d(u, v) > 0 for all u, v E V; (ii) d(u, v) = 0 if and only if u = v;
PROOF. It is clear that Lc is a linear space and that lif IIco > 0,
(iii) d(u, v) = d(v, u); (iv) d(u, v) C d(u, w) + d(w, v).
II0IIC* = 0, and IIUf- IIC* = I Ilf IIc* . If Ilf I= 0, then there exists a
set Nk E X with p(Nk) = 0 such that f (x) I c Ilk for x 0 Nk. If we put C.D. If f E L1(X, X, p) and s > 0, then there exists a simple X
N = Uk Nk, then N e X, jc(N) = 0, and =0 for x 0 N. measurable function 9) such that Ilf - 9) 111 < s. Extend this to L,
1 I

Therefore, f (x) = 0 for almost all x . 1 c p < oo. Is this true for Lc ?
If f, g e LCO , there exist sets N1, N2 in X with 4(N1) = p(N2) = 0 6. E. If f c- Lp, 1 c p < oo, and if E = {x E X : I f(x) I O}, then
such that E is Q-finite.
IIlf 11co for x O N1, 6.F. If f c- Lp and if E7t = {x E X : If(x) I > n}, then p(E) --* 4 as
Ig(x)I cc for x 0 N2.
11 S 11
n -* 00.
Therefore I
+ g(x)I c IIfIIc* + IIgIIco for x0 (N1 u N2), from C.G. Let X = N, and let t be the counting measure on N. If f is
which it follows that Ilf + 911 co c Ilf If co + 11911 Co. defined on N by f (n) = 1 In, then f does not belong to L1, but it does
It remains to prove that Lo is complete. Let (f7) be a Cauchy belong to Lp for 1 < p c cc. [Alternatively, let X = R, X = B, and
sequence in LO, and let M be a set in X with p(M) = 0, such that let p be Lebesgue measure and define g(x) = 0 for x < 1 and g(x) _
I
I c II I c. for x 0 M, n = 1 , 2, ... , and also such that I fn(X) - 1 /x for x > 1 .]
fm(x) I c IIfn - fm II. for all x 0 M, n, m = 1, 2, .... Then the C.H. Let X = N, and let A be the measure on N which has measure
sequence (fn) is uniformly convergent on X 1 M, and we let 1 In' at the point n. (More precisely A(E) = I I1In2 : n c- E}.) Show
f(x) = l im fo(x), x 0 M, that A(X) < +ao . Let f be defined on X by f(n) = Vn . Show that
=0, xEM. f c- Lp if and only if 1 c p < 2. [For a similar example, let X = (0, 1)
with Lebesgue measure, and consider g(x) = 1 /V x .]
It follows that f is measurable, and it is easily seen that IIfn - f IIC0 -* 0.
C.I. Modify the Exercise 6.H to obtain a function on a finite measure
Hence Lc is complete. Q.E.D.
space which belongs to Lp if and only if 1 p < po.
C.J. Let (X, X, l,c) be a finite measure space. If f is X-measurable,
EXERCISES
let E7t = {x E X : (n -- 1) c If (x) I < n}. Show that f e L1 if and
G.A. Let CEO, 1] be the linear space of continuous functions on only if
Go

[0, 11 to R. Define No for f in C[O, 1] by N0(f) = I Show that n 4(En) < +co .
No is a semi-norm on C[O, 1]. n=1

G.B. Let CEO, 1] be as before and define N1 for f in C[O, 1] to be the More generally, f c- Lp for 1 c p < cc, if and only if
Riemann integral of If I over [0, 1] . Show that Nl defines a norm on cc

C[O, 1 ] . If fn is defined for n > I to be equal to 0 for 0 c x c np p (En) < +00-

n=
(1 - l In)/2, to be equal to I for I c x c 1, and to be linear for
(1 - lln)/2 c x c 1, show that (f) is a Cauchy sequence, but that it 6.K. If (X, X, p) is a finite measure space and f c- L, then f E Lr
does not converge relative to N1 to an element of C[O, 1]. for 1 c r c p. (Hint: Use Exercise 6.J or the inequality If r 1+
 The Lebesgue Spaces L,p 63 64 The Elements of Integration

If I P.)Apply Holder's Inequality to If I r in Lpjr and g = 1 to obtain G.S. Let f7z, /37, be as in the Exercise 6.R, and suppose that (f7) is a
the inequality Cauchy sequence. If s > 0, then there exists a S(e) > 0 such that if
MIMT -< Ilf 11 p tt(X) , E E X and p (E) < S(e), then /37z(E) < s for all n c- N. (Hint: Use
Corollary 4.11.)
where s = (11r) - (11p). Therefore, if /c(X) = 1, then IIfIIr c llfllp-
G.T. If f E L (X, X, p), then l c 11f 11. for almost all x. More-
6.L. Suppose that X = N and j is the counting measure on N. If over, if A < 11f ff co , then there exists a set E E X with p(E) > 0 such
f e Lp, then f e Ls with 1 < p c s < co, and MI Its c Mf. that I f(x)f > A for all x c- E.
. 6.M. Let X = (0, oo), let jt be Lebesgue measure on X, and let 6.U. If f E LP, 1 c p c ao, and g c- L., then the product fg c- L. and
.f(x) = x -1/2(1 + I log x 1) r 1. Then f e Lp if and only if p = 2.
MOP C 11AP11911COO
6.N. Let (X, X, p) be any measure space and let f belong to both G.V. The space Lro(X, X, p) is contained in L1(X, X, p) if and only
Lp1 and L,,,, with 1 c p1 < p2 < CC . Prove that f e L. for any value if p(X) < co. If p(X) = 1 and f c- L , then
of p such that p1 c p c p,2.
11ACO = lim 11APO
6.0. Let 1 < p < oo, and let (lip) + (11q) = 1. It follows from
Holder's Inequality that if f e Lp, then

ffgdw I < Ilf 11 p

for all g E Lq such that 11 S 11 9 1< 1 . If f =A 0, define go on X by go(x) _

c[signum f (x)] If (x) I p-1, where c = (II f II D) -p,9' Show that go c- Lq,
that 11 So 11 q = 1, and that

ffgod =IIfIIp
G.P. Let f e Lp(X, X, p) , 1 c p c oo , and let s > 0. Show that 1

there exists a set EE e X with p(EE) < +oo such that if' F e X and
FEE=0,then If XFP < .

6. Q. Let fn e Lp(X, X, 0, 1 c p < oo, and let /37, be defined for

Ee X by
1/p
fln(E) = I fn IP d
{fE
Show that I fln(E) - Pm(E) I Ilfn - fm II P . Hence, if (fn) is a Cauchy
sequence in Lp, then lim /3,,,(E) exists for each E e X.
6. R. Let fn, /3n be as in Exercise G.Q. If (fn) is a Cauchy sequence
and s > 0, then there exists a set EE e X with jt(EE} < +oo such that if
F e X and F n EE = 0, then /3n(F) < s for all n e N.