Professional Documents
Culture Documents
Unit 9
Unit 9
9.1 Introduction
Objectives
9.2 Introduction to Lebesgue Integral
9.3 Convergence Theorems
9.4 Lebesgue Integral vs. Riemann Integral
9.5 Spaces of Integrable Functions
The Space L1
The Space L~
9.6 Summary
9.7 Hints/Solutions
9.1 INTRODUCTION
In the last unit, we introduced to you the concept of Lebesgue measure as a
countably additive real-valued function m defined on a a-algebra M of subsets
defined on R. Now, in this unit we shall see how we can use this measure to
define the integral which is called Lebesgue integral of certain functions defined
on R. This integral is called Lebesue integral. You are already familiar with the
notion of Rieman integral. Here you will see that when X = R the class of
functions which are Lebesgue integrable on R also includes the functions
which are Riemann integrable.
We shall first define integral for a simple function which are function of the
n
form s = a i z ~. ,Then we extend the definition to non-negative functions and
i=l
later to any measurable function. We shall cover this in Sec. 9.2. In Sec. 9.2, we
shall discuss two important theorems known as Montone convergence theorems
(also called Fatou's lemma) and dominated convergence theorem. In Sec. 9.3,
we recall the Riemann integration theory in brief and discuss its relationship
with the Lebesgue integral. Sec. 9.4 deals with two function spaces L1(E) and
L~(E) where (E) is a measurable subset of R. These spaces are called Lebesgue
spaces. L1(E) is the space of real-valued measurable functions defined on R
which are integrable over a measurable set E and L2(E)is the space of
real-valued measurable functions on R , the square of which are integrable over
the measurable space E. Both these spaces have lot of applications in
Engineering, Physics and other areas.
Objectives
After studying this unit, you should be able to
compute the Lebesgue integr~lfor simple function;
state Montone convergence theorem and prove the theorem using Fatou's lemma;
.
i use the converge theorem to compute Lebesgue integral;
Measure and Integral state and prove dominated convergence and apply it to compute Lebesgue
integral;
$?
9.2 INTRODUCTION TO LEBESGUE
INTEGRAL
are measurable subsets of R, i.e. Ai E M . Note that the sets A; are pairwise
disjoint and their union is R.
Lebesgue Integral
Clearly we can write
I=1
You can give many examples like this. Now we observe some simple properties
of simple hctions.
Proposition 1: Let (R, M , m) be a measure space. Let f : R -+ R and
g : R -+ R be simple functions. Then f + g is also a simple function. Also for
any X E R, then function Xf is a also a simple function.
The verification of these properties is left as an exercise for you (see E3).
Proposition 2: Let f : R -+ R be a non-negative measurable function. Then
there exists a sequence fi, fi . . of simple function fn : R -+ R such that, for
each x E R
O I f i ( x ) <_fi(x)F - .
We shall not prove this. But we state that if we consider f, such that
Next we shall define integral for non-negative funct~ons.For that we shall make
use of Proposition 2.
f dm = sup Y(E. f)
where
y(E, f) = (1 pdm : 0 5
1
d 5 f. 4 is simple .
Note that the integral can be + m , and is always non-negative. We observe that
the set Y(E, f) is always of the form [O, x] or [0, x), where the value x = +m is
allowed.
or even as jab
ff(x)dx,when no confision is possible (and we set
ffdm = - :j fdrn if a > b). The notation J fdm means JR f dm.
Let us now consider a measurable set A E M and a non-negative measurable
function g such that g = 0 on A", then any non-negative simple function that is
less than g must be zero on A". Applying this to g = f.XA we obtain the
important identity
JA JR
Next we show that the above properties of the integrals of simple functions
extent to the integrals of non-negative measurable functions.
Theorem 2: Suppose f and g are non-negative measurable functions.
i) If A E M, and f < g on A, then
ii) If B 2 A, A, B E M, then
/ fdrn 2
iii) For a > 0,
l afdm = a
S, fdm.
Measure and Integral Here we give a sketch of the proof
Proof: i) Notice that Y(A, f ) 2 Y(A, g) (we can squeeze more simple
functions under g than under f and the supremum of a bigger set is either
equal or larger).
ii) If y is a simple function such that p < f on B, then extending it by zero
outside B we obtain a simple function 4'such that 4' <_ f on A. The
integrals of these sjmple functions 6, and p' are the same. So
Y(B, f) Y(A, f3 and we conclude as in (i).
iii) The elements of the set Y(A, af) are of the form ax where x E Y(A, f), so
the same relation holds between their suprema.
iv) For any simple function 9,S, qdm = 0. To see this, let us take
9 = C C,Z)~E,,
say, then m(A n Ei) = 0 for each i, so Y(A, f ) = (0).
v) The elements of Y(A U B, f) are of the form JAu, qdm. Therefore by
+
Theorem 2(v), they are of the form JA cpdm J, ?dm. So
Y(A U B, f ) = Y(A, f ) + Y(B, f) and taking suprema this gives
For the opposite inequality suppose that the simple function 9 and
satisfy: 6. < f on A and ip = 0 on A", while $, 5 f on B and = 0 on Bc.
$1
On the right hand side of ( I ) we have an upper bound which remains valid
for all simple functions that are less than f on A U B. Thus taking suprema
over p and $! separately on the left hand side of () gives
E 1) Prove the Mean Value Theorem for the integral which states : if
a 5 f(x) < b for x E A, then a m(A) 2 S, f dm 5 b m(A).
We now confirm that null sets are precisely the 'negligible sets' for integration
theory.
Theorem 3: Suppose f is a non-negative function. Then f = 0 a.e. if and only
if J, fdm = 0
Proof: First, note that i f f = 0 a.e. and 0 < p 5 f is a simple function, then Lebesgue Integral
;? = 0 ax., since neither f nor 9 take negative values. Thus SRpdm = 0 for all
such 9 and so Jk fdm = 0 also.
Conversely, let J' fdm = 0, and E = {x : f(x) > 0). Our goal is to show that
m(E) = 0. Put
n=l
To show that m(E) = U it is sufficient to prove that m(En) = 0 for all n. We note
1
that the function ,- = -YE" is simple and g 5 f by the definition of En. So
This shows that m(E,) - 0 for all n. Thus m(E) = 0. Hence the result.
Using the results proved so far the following 'a.e.' version of the monotonicity
of the integral is not difficult to prove.
Proposition 4: I f f and g are measurable then f < g a.e. implies
J fdm 5 S gdm.
This can be verified by noting that let A = {x : f(x) < g(x)), then B = A' is
null and 5gy~.
The crux of Lebesgue integration is its convergence theory. In the next section,
we shall discuss this. We can make a start on that by giving a famous result.
f
I
-
Suppose that {x,) is a sequence of real numbers. Let E be the set of numbers x
such that x,, x for some subsequence x,, of {x,). x is called a limit point of
x,. So, E is the set of all limit points of {x,). Then supremum and infimum of E
, are denoted by the following
lim inf
n-+w
1 fndm > (hiIinf fn) dm.
Proof: We write
f(x) = lim inf f,(x)
n-m
We recall that for any x lim inf fn(x)= inf E, where Ex is the set of all limit
points of {fn(x)).
Let g,(x) = inf fk(x)'
k>n
and we have
00
Next,
5/ AnnE
r
fk dm for k 2 n( by the definition of g,)
<1
fkdm(as E is the larger set)
for k >_ n. Hence Lebesgue Integral
(Note that the symbol "4"indicates that the sum on the L.H.S. converges to
that on the R.H.S.)
The inequality in (2) remains true in the limiting case also and therefore we have
Now we substitute the expression for 7given in (3). Then by making the
assumption that
we get
The case m((x : y(x) > 0)) = oc has to be treated separately. Here
JE ydm = oo and so SEfdm = oc.We have to show that
lim inf
k-cc
[fxdm = m.
Let ci, i = 1. . . n since (F is simple it assumes finite value, say cl, . . . , c, for
1
some n E N and let a = Amin{cl, c2, . . . ,c,) we put
2
and
03
lirn inf
n+oo
1E
fndm 5 lim sup
n+w
1
E
fndm
gives
x
g(x) = { :(x)' when
when x
EA
E Ac,
I Then using E = [E fl Ac] U [E fl A] we get Lebesgue Integral
I gdm =
I fdm.
You already know from Proposition 2 that for any non-negative measurable f,
there is a sequence fn of non-negative simple functions such that fn /' f.
1
t Try this exercise now.
,
E3) Consider the sequence f, := ; ~ [ ~ , ~ +=~ 1l ,. 2( ,n. . .). Does this sequence
I satisfy all the conditions of the Monotone convergence theorem? Does the
t conclusion of the Monotone Convergence Theorem hold good for this
i sequence? Justify your answers.
L
1 We can now extend the integral verv easilv to general real-valued functions.
using the positive part f + = legal:ive part f - max
of any measurable function not use the non .egati
-
measurable function Ifl alonge: as we already saw, (fl can be measurable
without f being measurable!
! Then we have
f = f + -f-
and
Here we note that the function f : R -t R is measurable iff both f + and f - are
I measurable. lwoucan verify this by yourself.
Note thx f is int2gmble iff (flis integrable, and that
Proof: Step 1: Suppose first that f and g are non-negative simple functions. Let
z
f = a i x ~, g z
, = bjxB,. The sum f + g is also a simple function which can be
written in the form
Therefore
where we have used additivity of m and the facts that Ai cover R and the same
is true for Bj. -
Step 2: Now suppose that f, g are non-negative measurable (not necessarily
simple) functions. By Proposition 2 we can find sequences f,, g, of simple
functions such that for each x, f, (x) /" f(x) and gn(x) /" g(x). Clearly
f, -tg,(x) /" f + g hence using the monotone convergence theorem and the
additivity property for simple functions we obtain
(f + gjdm = lim
n-oc
= Iinl
n-+m
/
E
s.dm Iim J t d m
+ n-m
E
This, in particular, implies that theintegral o f f + g is finite if the integrals o f f Lebesgue Integral
and g are finite.
hence
By definition of the integral the last relation implies the claim of the theorem.
Proposition 6: I f f is integrable and c E R, then
JE(cf)dm = c l f d m .
The proof is left as an exercise for' you to verify. The proof is based on the fact
that f can be approximated by simple functions.
We can now answer an important question on the extent to which integral
determines the integrand.
Theorem 7: If JA f dm 5 JA g dm for all A E M, then f 5 g a.e. particular, if
JAfdm= JAgdmforallA€M,thenf=ga.e.
Proofi Here we give sketch of the proof.
By additivity of the integral and Proposition 6, it is sufficient to show that
JA h dm 2 0 for all A E M implies h >
0 (and then take h = g - f ) . Let us
1
write A = {x : h(x) < 0); then A = UA, where A, = {x : h(x) 5 --). By
n
monotonicity of the integra1,'we have
Measure and Integral which is non-negative but this can happen only if m(A,) = 0. The sequence of
>
sets An increases with n, hence m(A) = 0 and so h(x) 0 a.e.
A similar argument shows that if JA h dm 5 0 for all A, then h 5 0. This
implies the second claim of the theorem: put h = g - f and JA h drp is both
>
non-negative and non-positive, hence h 0 and h 5 0 a.e. thus h = 0 a.e. O
The limit theorem which turns out to be the most useful in practice states that
convergence holds for an a.e. convergent sequence which is dominated by an
integrable function. Again Fatou's lemma holds the key to the proof.
Theorem 9: (Dominated Convergence Theorem): Suppose E E M. Let
{f,) be a sequence of measurable functions such that Ifn(5 g a.e. on E for all
n 2 1, where g is integrable over E. If f(x) = lim,,, fn(x)a.e. then f is
integrable over E and
Lebesgue Integral
Proof: Suppose for the moment that fn 2 0. Fatou's lemma gives
/ L
f dm < ,lim irlf J IE f. dm
-+ x
I I;
Finally, subtract g dm (which is finite) and multiply by -1 and get the
desired result given in (5).
I
we have
and we can apply the result proved for non-negative hnctions to the sequence
f,,(x) + g(x) (the function 2g is ofcourse integrable) and get the desired result.
-
for x E ( 0 , l ) . Clearly fn(x) 0. To conclude that limn f fn dm = 0 we need to
find an integrable dominating function. To find such a function, we proceed as
follows:
I nsinx I , n , n 1 _ 1
I t 15 not easy to show thls directly. But, we observe that the hnction x -t fi 1s
R~emann~ntegrable.We shall show shortly that Lebesgue and Riemann
~ntegralsof a bounded function coincide ~fthe latter exists, and since
d 3 -
-( &) = k f x we can apply the Fundamental Theorem of the Calculus to get
dx
the desired result. The result, then follows from the dominated convergence
theorem. 1
where
. sequence gk
The proof is left as an exercise for you to v e r i ~The = xE=,fn is
increasing and converges to C z ,f,.
Note: We cannot yet conclude that the sum of the series on the right-hand side
of (8 )is a.e. finite. Therefore C z , fn need not be integrable. However, we have
the following result. I
Theorem I I : (Beppo-Levi): Suppose that
30
1 /fkjdm i! finite.
Then the series CE, fk(x)converges for almost all x, its sum is integrable, and Lebesgue Integral
as required.
+
Suppose f is a complex-valued function, and let f = u iv, where u and v are
the corresponding real functions. We say that f is measurable if and only if both
u and v are measurable.
Tt is easy to verify that sums and products of complex measurable functions are
again measurable. Since
1 Ifdm = udm + i
1 vdm, E E M
Measure and integral Then we can state that all the theorems, we have considered in Sec.9.3, hold for
complex functions also.
In the next section we shall consider the relationship between Lebesgue integral
and hemann integral.
Mi = sup (fix)).
X,-~<X<X,
Now, let
and
Note that xi - X,-I is the length of the sub-interval [ y i P l , x,]. The sums L(P, f)
and U(P, f) are called lower sum and upper sum o f f corresponding to the
partition P, respectively. These sums are also ref;:-red to as lower and upper
Riemann sums. Then, we have
m(b - a) < L(P, f j IU(P, f j < M(b - aj,
where m and M are the infiinunl and supremcin, respectively 01 f i n [a, b] (refer
Theorem 1, Unit 10, Block 5 of the IGNOU course material MTE-09).
Let P be the set of all partitions of [a, b]. Now conside, Lire set
If l L = Ill(= 1~say)
irk such cases 1 ns also called ihe definite integral o f f over [a, b].
We haw seen that the following class of functions are Riemann integrable.
: the <!ass of real valued cont~nxousf~nctlonsdefined on a closed interval
[a, bl.
) the ~qlassof buur~dednon-negative funr:tions with countably many
di\contlnuitlga.
We can also prove that a bounded real-valued function on [a, b] is Riemann
Integrable on [a: b] G f is continuous "almost everywhere" on [a, b].
;, not Kiemann integrable. Note that this function is bounded but everywhere
This function is known as Dirichlet's function. Here the question arises. How
to extend this definition to functions having discontinuity.
li~n
n-no
lb [
f(xjdx = f(x)dx?
The answer to this question, in general, is in the negative. For example consider
fn(x)= ne-nx,x E [O, 11,n = 1.2, . . . lirn fn(x)= f(x) = Ox E (O,l].
n--m
n + mfn(0) + oo as n + oo.Thus each f,, is Riemann Integrable on [0, 11, but
f is not. There are other examples that one can construct.
Yet another problem is that the class of functions.R([a,b]) Riemann Integrable
functions on [a, b] with a metric constructed from the Pseudo-metric
d ( f g) = (f(x)- g(x)ldx is not complete in the sense that Cauchy sequences
of Riemann Integrable functions need not converge to a R i e m a ~
Integrable
A Pseudo-metric is a function. The incompleteness of R[a,b] is a problem of serious concern and
function d : R x R -, R makes it difficult to do analysis on it.
which satisfies all the
conditions of a metric All these problems indicate that the concept of Riemann Integration is
except d(x,y) = 0 iff restructive and non inclusive of a desirable class of functions. This prompts a
x = y. worthwhile generalisation of Riemann Integration. The new integration theory
should include all Riemann Integrable functions for which the New Integral
should coincide with the Riemann Integral and should be applicable to a larger
class of functions. We would also like to have the larger class of functions with
new integrability, to be free from the drawbacks of Riemann Integrable
functions.
So the problem is how to generalise Riemam Integration. We have to reinvent
the concept of integration.
Let us get back to the basics.
where yo = 0
Now, consider the sets f ( [ Y , - ~ ,y,)). This need not be an interval. All that one
can say is that it is a subset of [a,b]. Consider the sums
where !( f - l [ ~ ~yi))
- ~ ,is the "length" of the set f - ' ( [ ~ , _ .yi)),
~ . if at all such a
concept makes sense! Then these two sums provides with a lower
approximation and an upper approximation to the area A(f) to be computed.
Note that for bounded functions these sums are finite sums. We encounter the
following problems.
Problem 1: ( f l [yiPl,yi)) may not be an interval! It may not be even a union of
intervals. Then what do we mean by E(f-l lyiPl,y,))? Concept "length" has to
be introduced for arbitrary sets!
Problem 2: Unlike in the Riemann Integration case we may have to work with Lebesgue Integral
infinite sums.
Assuming we can handle these problems successfully, we can recover the
integral, proceeding further as in the Riemann Integral Case. For,
Refinement of the partition of the range will move the lower sums upward and
the upper sums downward. '
Again, we ask whether these sums have limits and whether they are equal! If
they are equal then we say f is integrable in the 'new' sense. So our concerns is
the following:
The concept of mctsure which we have discussed in Unit 8 helps to extend the
notion of length to more complicated sets than intervals such that for intervals
the new concept co~ncideswith the notion of length in the conventional sense.
In the next section, we shall discuss the Integration theory based on the concept
of Lebesgue measure.
Next we shall show that the Lebesgue and Riemann integrals coincide whenever
the latter exists. In the process we shall find necessary and sufficient conditions
for the existence of the Riemann integral.
The next theorem gives a relationship between Riemann and Lebesgue integral.
In this section, we shall first study spaces of functions which are integrable.
The properties (i) and (ii) stated in Theorem 13 can be verified directly (see E5).
A vector (linear) space, say X, is said to be normed linear space if we can assign
a non-negative real number which we denote by to each x E X satisfying
the following conditions:
i) lixll 2 0 Vx E X
ii) llxll = 0 if and only if x = O
iii) llaxll = IQIIIxII,V Q E R and x E X Lebesgue Integral
-
For f, g t L1(E), we say that f is equivalent to g, denoted by f g, if the'set
{x E E : f(x) # g(x)) has measure 0, i.e. f and g are equal a.e. Then the
relation '--'is an equivalence relation.
You can verify that - is an equivalence relation on L1(E).
Write [fl for the equivalence class containing the function f E L1(E). Thus,
h E [fl ~ fh(x)
f = f(x) a.e.
Let C1(E) denotes the set of all equivalence classes.
We can show that L1(E) is a vector space, since L' (E) is a vector space. To do
this, we need to understand what we mean by a linear combination of
+ + +
equivalencc classes. We d e h e [q [g] as the class [f g] o f f g, i.e.
+ +
h E [fj [g] iff h(x) = f(x) g(x) a.e. Similarly for multiplication by constants:
~ [ f =j [afj for Q E R. Hence C1(E) is a vector space with these operations.
Strictly speaking we should continue to distinguish between the equivalence
class [fj f L1(E) and the function f f L1(E) which is a representative of this
class. But there is no serious loss of clarity by treating f interchangeably as a
member of L1(E) and of C1(E) ,depending on the context. In other words, by
treating the equivalence class [fj as if it were the function f, we implicitly
identify two functions as soon as they are a.e. equal. Thus, we do not
distinguish between L1(E) and L1(E). With this convention it will be clear that
the function 1) . 1) is a norm on L1(E) and is called L2L1-nom.
We now define a metric on L1(E).
For f, g t L1(E), we define
Definition 7: Let f, g E L' (R). Let S denote the set of x for which the
Lebesgue integral given by
and
= fdt
and observe that the integral on the R.H.S. does not exist. That means, the Lebesgue Integral
convolution is not defined for x = 1.
Now we shall state a theorem which gives a condition for which the con:/olution
integral exists for all x E R.
Before we state the theorem we shall state another theorem which w111 be used
as a lemma for proving the theorem. The theorem deals with interchanging the
order of integration. You rnight be already familiar with similar results when
you studied multiple integrals in your undergraduate Calculus courses.
Theorem 15: I.et X and Y be two intervals of R and let k be a function which
is defined, con~liiuousand bounded on X x Y, say
is continuous on Y.
b) For cach x in E, the lebesgue integral
tainctrcm G defined on X by the equation
S, g(y)k(x, y)dy exists, and the
is continuous on X.
C) The two lebesgue integral 1 g(y)f(y)dy and /X
f(x)G(x)dxexists and are
equal. That is
Thus, f * g is bounded. n
Try these exercises now.
The space we now introduce plays a special role. It provides the closest
analogue of the Euclidean space Rnamong the space of functions, and its
geometry is closely modelled on that of Rn. We introduce a function called
)I2
norm, which we denote by (1 . . . on L1(E). This function provides a concept
of orthogonality (and hence 'angles') between functions. This gives L2 many
pleasant such as a 'Pythagoras theorem' and the concept of
orthogonal projections, which plays vital role in many applications.
We make the following definition:
1
I
As in the case of L1(E),we do not distinguish between L1(E) and C1(E)
Henceforth we will be using L ~ ( E )With
. this convention L~(E) is a vector
space. We now define a metric on L ~ ( Egiven
) by
1 Then d is a metric on
The inequality in (14) is called Schwarz Inequality. Note that you are already
familiar with Cauchy-schwarz inequality for real sequences.
i
The inequality will be used in Functional Analysis course and other courses
very frequently.
In the next unit, you will see more properties of L1(E) and L2(E).
With this we come to an end of this unit.
I 9.6 SUMMARY
1 *
In this unit, we have covered the following points.
1) We have defined "simple" functions and the Lebesgue integral of a simple
function. We have also explained the computation of Lebesgue integral for
certain simple functions.
r1 r1
1 lirn fn(x)dm #
n-m
fn(x)dx
Hint: Show that each fn is measurable but the sequence ifn) is not
monotone. Also, show that, f,dm = 1,'d n, lim f, = 0, and
J(lim f,)dm = 0.
1 Ul L l l V l 011""I
1 nx3
bllUL 11111
< ca since f, g E X.
This shows that f + g E X.
ii) For f E X, 0 E R, we have