Mathematics Department Stanford University

Math 171 Lecture Supplement

The Riemann and Lebesgue Integrals
Leon Simon
Autumn Quarter, 2012
These notes are meant as a quick introduction to the Riemann integral using step function terminology, followed by an almost-as-quick introduction to the Lebesgue integral, also via step functions.
We include a proof of Lebesgues theorem which precisely characterizes those functions which are
Riemann integrable.

Contents
1
2
3
4
5

Preliminaries: Step Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Riemann Integrable Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Lebesgue measure zero . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Definition and Properties of the Lebesgue Integral . . . . . . . . . . 7
The spaces L1 (R) and L2 (R) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

Preliminaries: Step Functions

Here we work in Rn , where n 1 is given. By an interval I in Rn we mean the cartesian

product I1 I2 In , where each Ij is an interval in R, thus each Ij is one of the following: [aj , bj ], (aj , bj ), [aj , bj ), (aj , bj ], where aj bj are real. The volume |I| of an interval
I = I1 I2 In is of course defined to be the product of the edge lengths. Thus |I| =
(b1 a1 ) (bn an ) in case Ij has end-points aj , bj with aj bj , and the volume is zero if I
is a degenerate interval (i.e. if aj = bj for some j).
We say the interval I is closed if each factor Ij is a closed interval; thus a closed interval in Rn is
one that can be written [a1 , b1 ] [an , bn ]. Similarly, an open interval is one that can be written
I = (a1 , b1 ) (an , bn ) (= if aj = bj for some j).
From now on R = [a1 , b1 ] [an , bn ] will be a fixed closed interval in Rn with aj < bj for each
j = 1, . . . , n.
Definition: A partition P of R is the collection of closed intervals I R obtained by partitioning
each of the edges of R; thus for each j = 1, . . . , n we select points aj = tj,0 < tj,1 < < tj,Nj = bj
and then P = {[t1,i1 1 , t1,i1 ] [t2,i2 1 , t2,i2 ] [tn,in 1 , tn,in ] : ij {1, . . . , Nj } for each j =
1, . . . , n}. The points tj,0 , . . . , tj,Nj are called the j-th edge points of the partition P. For any
I = [t1,i1 1 , t1,i1 ] [t2,i2 1 , t2,i2 ] [tn,in 1 , tn,in ] P we let I denote the corresponding open

interval (t1,i1 1 , t1,i1 ) (t2,i2 1 , t2,i2 ) (tn,in 1 , tn,in ), and I = I \ I.

Definition: is a step function on R if it is bounded and there is a partition P of R such that
thus
for each I P there is a real constant aI such that aI on I;
P

1.1
= IP aI I on IP I,
where we use the notation that, for any set A Rn , A denotes the indicator function of A; thus
A (x) = 1 if x A and A (x) = 0 if x Rn \ A.
1

Math. 171 Lecture Supplement 2012

Notice that in 1.1 we make no particular restriction on the values taken by on IP I beyond the
fact that (x) should be a real number for each x IP I and |(IP I) should be bounded.
P
P
If and are two such step functions on R, say with = IP aI I on IP I and = JQ bJ J
then the sum and difference is also a step function, as are max{, }, min{, }. One
on JQ J,
checks this by taking a common refinement R of the partitions P, Q (thus R is a partition of R
with j-th edge points including all the j-th edge points of P and all the j-th edge points of Q,
j = 1, . . . , n); then
P
P
1.2
= IR e
aI I and = IRebI I on IR I
for suitable e
aI , ebI (in fact e
aJ = aI whenever J R with J I and I P and ebJ = bI whenever
J R with J I and I Q; this makes sense because every J in the common refinement R is
contained in a unique I P and a unique Ie Q). Evidently it is also true that any scalar multiple
of a step function is also a step function (indeed any product of step functions is a step function
by 1.2) so in particular we have shown that the step functions form a subspace of the vector space
of functions f : R R. The vector space of step functions on R will be denoted
S(R).
Naturally the integral of a step function as in 1.1 is defined by
R

1.3

IP aI |I|,

where |I| is the volume of I (i.e. the product of the edge lengths of I). Of course this definition
is independent of the particular representation of step function used; thus if instead of 1.1 we use
P
then of course it turns out that
a different partition Q of R and if = JQ bJ J on JQ J,
P
P
IP aI |I| =
JQ bJ |J|; one can check this directly by using a common refinement R of P, Q as
where
in 1.2 above, as follows: Since each K R is contained in a unique I P and cK on K,
cK = aI , we have
P

KR cK |K|

IP

aI

KK with KI |K|

IP aI

|I|,

and similarly
P

KR cK |K|

JQ bJ |J|,

so as claimed
P

IP aI |I|

JQ bJ |J|.

Using the formulae 1.2 it is evident that the integral so defined is linear on the step functions; that
is, if , are both step functions on R and if , are real constants then
1.4

R (

+ ) =

R
+ R ,

, S(R).

It is also clear from the representation 1.2 that

1.5

, S(R) with

R .

Riemann & Lebesgue Integrals

Riemann Integrable Functions

Given a bounded function f : R R we can define the upper and lower Riemann integral of f ,
R
R
denoted R f and R f respectively, by

Rf

f
R

inf

sup

S(R),f R
S(R),f

and f is said to be Riemann integrable if it is bounded and

the above definitions we have
R

f
R

Rf

R
R

f . Of course using 1.5 and

Rf

in any case, so to prove f is Riemann integrable it suffices to prove the reverse inequality.
We begin with a convenient criterion for checking Riemann integrability:
2.1 Lemma: f is Riemann integrable for each given > 0 there are step functions , with
R
R
R
f on R and R < R + (i.e. R ( ) < ).
Proof : We assume that for each > 0 there are step functions , with f on R
R
R
R
R
R
R
R
R
R
and R < R + . Then R f R f R R + R f + , whence 0 R f R f < ,

R
R
and hence (since this holds for each > 0) R f = R f , and so f is Riemann integrable as claimed.

Proof : For any > 0, by definition of upper and lower integral we can find step functions
R
R
R
R
, with f and with R > R f /2 and R < R f + /2. Thus if f is Riemann
R
integrable (i.e. if the upper and lower integrals are equal), then we have R ( ) < /2 + /2 = .
We now check a few general facts about the set of Riemann integrable functions.
2.2 Theorem: Every continuous function f : R R is Riemann integrable.
Proof: Let > 0. R is compact and a continuous function on a compact set is uniformly continuous,
so there is > 0 such that |f (x) f (y)| < whenever x, y R with |x y| < . So take a partition
P of R such that each I P has diameter < , and observe that then maxI f minI f + for each
I P. Define step functions , on R as follows:
|I = min f, |I = max f,
I

IP

and take = minR f and = maxR f on the remaining points of R (i.e. on IP I). Then
f and

IP (max f
I

min f )|I|
I

IP |I|

= |R|,

and hence f is Riemann integrable by Lemma 2.1.

2.3 Theorem: The set of Riemann integrable functions on R is a vector space and the Riemann
integral is a linear operator on that vector space (i.e. f, g Riemann integrable on R and , R
R
R
R
f + g is also Riemann integrable, and R (f + g) = R f + R g).

Math. 171 Lecture Supplement 2012

Proof: Let R(R) be the set of Riemann integrable functions on R. It suffices to prove
(1)

f, g R(R) f + g R(R) and

R (f

+ g) =

Rf

Rg

and
(2)

f R(R) and R f R(R) and

R f

R
= R f.

To prove (1), let > 0 and, using Lemma 2.1, pick step functions , , ,
e e such that f
R
R
e and ( ) < /2, (e )
,
e g ,
e < /2, and so by virtue of the linearity 1.4 we have
R
R
R

R (

e
+ )

R (

+ )
e < /2 + /2 = ,

and
e
+
e f + g + ,
R
R
so we deduce that f + g is Riemann integrable by virtue of Lemma 2.1, and R f + R g
R
R
R
R
R
R
R
e
e ( + )
e R (f + g) R ( + )
e + R f + R g + , so
R ( + ) R ( + )
R
R

Rf

for each > 0, whence

R (f

Rg

+ g) =

Rf

R (f

+ g)

Rf

Rg

R g.

The proof of (2) is left as an easy exercise, again based on Lemma 2.1 and the linearity 1.4 (one
has to break consideration into the cases > 0, = 0, = 1).
The Riemann integral also preserves ordering of functions as follows:
2.4 Theorem: Suppose f, g are Riemann integrable on R and f g (meaning f (x) g(x) x R).
R
R
Then R f R g.
R
R
R
R
Proof: Let > 0. Since R f = R f we can find a step function with f and R f R + ,

R
R
R
R
and since R g = R g we can find a step function with g and R R g + . Then
R
R
R
R
R
R
f g and by 1.5 R f R + R + R g + + ; that is R f R g + 2, and
R
R
since is arbitrary this implies R f R g.

Lebesgue measure zero

A key difference between Lebesgue theory and the theory of volume related to the Riemann integral
is in the notion of measure zero. In the Riemann theory a set A is said to have volume zero (or
Jordan content zero) if for each > 0 there are open intervals I1 , . . . , IN with A N
j=1 Ij and
PN
j=1 |Ij | < . In the Lebesgue theory, we allow infinitely many intervals:
3.1 Definition: A set A Rn is said to have Lebesgue measure zero (or simply measure zero) if
P
for each > 0 there are open intervals I1 , I2 , . . . such that A
j=1 Ij and
j=1 |Ij | < .
This makes a profound difference in the related theory; for example we have the following lemma:

Riemann & Lebesgue Integrals

3.2 Lemma. Suppose A1 , A2 , . . . is a sequence of subsets of Rn such that each Aj has Lebesgue
measure zero. Then
j=1 Aj also has Lebesgue measure zero.
Proof: Let > 0. For each j we can select open intervals Ij,1 , Ij,2 , . . . with Aj
k=1 Ij,k and
P
P
P j
j

k=1 |Ij,k | < 2 . Then j=1 Aj j,k Ij,k and

j=1 2 =
j,k |Ij,k | <
As a first indication of the importance of the notion of sets of measure zero, we have the following very elegant theorem of Lebesgue, which completely characterizes those functions which are
Riemann integrable.
3.3 Theorem (Lebesgues Theorem on the Riemann Integral.) Let f : R R be a bounded
function. f is Riemann integrable on R there is a set A R of Lebesgue measure zero such
that f : R R is continuous at x for each point x R \ A.
Caution: f : R R is continuous at each x R \ A is a much stronger condition than f |R \ A
is a continuous function, and indeed f |R \ A continuous is in general not sufficient to ensure that
f is Riemann integrable even if A has measure zero. For example if we take R = [0, 1], A = the set
of rationals in [0, 1], then A has measure zero but the function f which is 1 on A and 0 on R \ A
R
R
is not Riemann integrable because evidently R f = 0 and R f = 1.

Proof of : Observe, by the definition of continuity, that f discontinuous at y R
0 > 0 such that supI f inf I f > 0 open interval I with y I R, which is the same as
saying there is a positive integer j such that supI f inf I f > 1/j open interval I with y I R.
can be written Sj , where
Thus the set of discontinuities of f |R
j=1
: supI f inf I f > 1/j for every open interval I with y I R}.
Sj = {y R
By Lemma 3.2 we know that
j=1 Sj has Lebesgue measure zero if each Sj has Lebesgue measure
zero, so it is thus enough to prove that Sj has Lebesgue measure zero for each j.
Let > 0, j {1, 2, . . .}. By Lemma 2.1 (with /j in place of ), we can pick a partition P and
corresponding , S(R) with f and
R
P
IP (supIf inf If )|I| R ( ) < /j.
Since supI f inf I f 1/j whenever Sj I 6= (by definition of Sj ), the above evidently implies
P
{IP:Sj I6=} (1/j)|I| < /j; that is,
P
()
j 6=} |I| < .
{IP:IS
But the intervals I P with I Sj 6= cover all of Sj except for the set E = IP I of measure
Thus () proves that Sj can be covered by a finite union
zero, and hence Sj \ E {IP:Sj I6=} I.
of intervals of total volume < and hence Sj has Lebesgue measure zero as required.
Proof of : Let > 0 and cover the set of discontinuities of f by a union
j=1 Ij of open
P
I is a compact set and f (as a function on
intervals such that
|I
|
<
.
Then
K

R
\

j=1 j
j=1 j
R) is continuous at each point of this compact set. We can therefore assert that there is > 0 such
that
()

|f (x) f (y)| < whenever x K, y R, and |x y| < .

Math. 171 Lecture Supplement 2012

Notice that the statement is stronger than the standard fact that a continuous function on a
compact set is uniformly continuous, because only the point x, and not necessarily the point y, is
required to be in the compact set Kon the other hand, the proof using the Bolzano-Weierstrass
theorem is almost identical to the usual Bolzano-Weierstrass proof of this standard fact, as follows:
If there is > 0 such that () fails for each > 0 then it fails with = k1 , k = 1, 2, . . ., and
hence there are points xk K, yk R such that |xk yk | < k1 but |f (xk ) f (yk )| . Then
by the Bolzano-Weierstrass theorem we can find a convergent subsequence xkj with x = lim xkj
K. Since |xkj ykj | < k1j 1j we also have lim ykj = x, and so by continuity of f at x we have
f (xkj ) f (ykj ) f (x) f (x) = 0, contradicting the fact that |f (xkj ) f (ykj )| for each j.
Now, with as in (), we select any partition P of R with each edge of each I P having length

< / n (so diameter of each I P is < ). For any I P such that I K 6= we have by () that
supI f inf I f = supz1 ,z2 I (f (z1 )f (z2 )) = supz1 ,z2 I ((f (z1 )f (yI ))(f (z2 )f (yI ))) + = 2,
where yI is any point in I K, while of course the sum of the volumes |I| over the remaining I P
is (because the remaining I have the property I K = and hence all such I R \ K j Ij ,
P
and j |Ij | < ). Thus we have
P

IP (sup f
I

inf f )|I| 2|R| + (sup f inf f ) 2(|R| + M ),

I

M = sup |f |.
R

and |I =
Thus if we define , S(R) by |I = inf I f, I P, and = infR f on R \ IP I,
R

supI f, I P, and = supR f on R \ IP I then R ( ) 2(|R| + M ), so f is Riemann

integrable by Lemma 2.1.
We conclude this section with two important lemmas about monotone sequences of step functions.
Again the concept of sets of measure zero plays a key role. In the statement of these lemmas, and
subsequently, we use the following terminology:
Terminology: We say that a property holds almost everywhere or for almost every point
(abbreviated a.e.) in a subset Rn if there is a set S of Lebesgue measure zero such that
the property in question holds at each point of \ S.
Thus for example fk (x) 0 a.e. x means that there is a set S with Lebesgue measure
zero such that lim fk (x) = 0 for every x \ S.
The first lemma relates to decreasing sequences of non-negative step functions:
3.4 Lemma. Suppose k S(R) with 0 k+1 k k = 1, 2, . . .. Then
k (x) 0 a.e. x R.

R k

Proof of : Let > 0 and S = {x R : lim k (x) }. Then S {x R : k (x) }

for each k, so, with P k any partition of R such that k |I is constant for each I P k , we have
R
P
( (IP I))
IF k |I| R k , where F k = {I P k : k |I }, so S (IF k I)
`
`=1
R
P
1

and IF k |I| R k 0. Since `=1 (IP ` I) has measure zero by Lemma 3.2, this shows
that S has measure zero for each > 0. Then, by another application of Lemma 3.2, {x R :
lim k (x) > 0} =
j=1 S1/j has measure zero as claimed.
Proof of : Let > 0. For each k = 1, 2, . . . let P k be a partition of R such that k is constant
on I for each I P k , and let E be a set of Lebesgue measure zero such that k (x) 0 for each

Riemann & Lebesgue Integrals

zero (by 3.2), we can select open

x R \ E. Since E (
k=1 IP k I) has Lebesgue measure P

intervals I1 , I2 , . . . such that E (k=1 IP k I) j=1 Ij and

j=1 |Ij | < .
Now K = R \ (
j=1 Ij ) is a compact set and by construction k (x) 0 for each x K and hence
for each given x K we can find kx such that kx (x) < and also (since x
/ IP kx I) there is

Jx P kx with x Jx and kx |Jx const. < . Trivially K xK Jx , so by compactness there is a

finite set of points x1 , . . . , xN K such that K N

j=1 Jxj . Then if k max{kx1 , . . . , kxN } we have
N
k < on j=1 Jxj because k () kxj () < for each Jxj and each k kxj . So let Q be a
partition of R with the property that, for i = 1, . . . , n, the i-th edge points of the partition Q include
all the i-th edge points of the all the intervals in the partitions P kx1 , . . . , P kxN . Then each J Q is

either such that J Jxj for some j or else J (N

j=1 Jxj ) = . Let be a step function which is on
the former intervals and M = supR 1 on the latter (and keep in mind that each J of the latter kind
is contained in R \ (N
J )
I ). So we have k on R for all k max{kx1 , . . . , kxN },
j=1
R
R j=1 xj
P j
P
and hence R k R |R| + M JQ:J(N Jx )= |J| |R| + M
j=1 |Ij | < (|R| + M ) for all
j=1
j
R
k max{kx1 , . . . , kxN }. Thus we have shown limk R k = 0 as required.
The second lemma concerns increasing sequences of step functions and is proved in a manner similar
to the proof of the easy direction () of Lemma 3.4:
3.5 Lemma. Suppose k S(R) with k k+1 for each k = 1, 2, . . . and suppose also that
R
{ R k }k=1,2,... is bounded. Then {k (x)}k=1,2,... is bounded for a.e. x R.
Proof: We have to prove that S has measure zero, where
S = {x R : {k }k=1,2,... is unbounded}.
R
R
We are given { R k }k=1,2,... is bounded, and it is increasing by 1.5, hence lim R k R and we can
R
select a subsequence {kj }j=1,2,... such that R (kj+1 kj ) < 22j for each j = 1, 2, . . .. For any
> 1 observe that
j
S
j=1 Sj , where Sj = {x R : kj+1 (x) kj (x) 2 },
PN j
P 1
<
because x
/ Sj for each j implies kN (x) k1 (x) = N
j=1 (kj+1 (x) kj (x)) < j=1 2
for each N 2.
On the other hand for each j, with P j a partition of R such that (kj+1 kj )|I = constant for each
R
P
I P j , we have 2j IF j |I| R (kj+1 kj ) < 22j , where F j = {I P j : (kj+1 kj )|I
P P
P j
2j }, hence
= 1 and by construction {I : I F j } covers all of
j=1
IF j |I| <
j=1 2
Sj \ (IP j I). Since
j=1 IP j I has measure zero by Lemma 3.2, we can select open intervals
P

I1 , I2 , . . . such that j=1 IP j I Ij and

j=1 |Ij | < 1/. Then (j=1 {I : I F j }){I1 , I2 , . . .}
is a collection of open intervals covering
j=1 Sj ( S) and the sum of their volumes 1/ + 1/ =
2/. Since is arbitrarily large this shows that S has measure zero as claimed.

Definition and Properties of the Lebesgue Integral

We let S + (R) denote the set of non-negative step functions (i.e. S + (R) = { S(R) : 0}).
L+ (R) denotes the set of functions f : R [0, ) with the property that there exists an increasing

Math. 171 Lecture Supplement 2012

R
sequence of {k }k=1,2,... S + (R) with { R k }k=1,2,... bounded and f (x) = lim k (x) a.e. x R.
L+ (R) is closed under the operations of addition and multiplication by non-negative scalars and
also under the operation of taking max and min of two functions in L+ (R); thus
4.1

, 0, g, h L+ (R) g + h L+ and also max{g, h}, min{g, h} L+ (R),

R
because if k , k are increasing sequences in S + (R) converging a.e. to g, h respectively with R k
R
and R k bounded, then k + k , max{k , k }( k + k ) and min{k , k } are increasing
sequences in S + (R) with bounded integrals converging a.e. to g + h, max{g, h} and min{g, h}
respectively.
However L+ (R) is not a vector space because it is not closed under the operation of multiplication
by 1. On the other hand
L1 (R) = L+ (R) L+ (R) (= {f : f = g h with g, h L+ (R)})
is a vector space since it is a non-empty subset of the vector space of real-valued functions on R
and it is closed under addition and multiplication by scalarsit is closed under multiplication by
non-negative scalars because L+ (R) is (by 4.1), and it is also trivially closed under multiplication
by -1.
We are going to define the Lebesgue integral on the vector space L1 (R). In order to do this we first
need to define it on L+ (R):
R
4.2 Definition: For f L+ (R) we define the Lebesgue integral R f of f by
R
R
R f = lim R k ,
R
where {k }k=1,2,... is an increasing sequence in S + (R) such that { R k }k=1,2,... is bounded and
f (x) = lim k (x) a.e. x R (as in the definition of L+ (R)).
R
R
Notice that of course lim R k exists because { R k } is bounded by assumption, and it is monotone
R
increasing by 1.5. However to be sure the definition makes sense we have to show that lim R k =
R
R
lim R k for any other increasing sequence {k }k=1,2,... S + (R) with { R k }k=1,2,... bounded and
f (x) = lim k (x) a.e. x R. This is in fact an easy consequence of Lemma 3.4, as follows: For each
fixed k = 1, 2, . . . we have that {(k ` )+ }`=1,2,... is a decreasing sequence of non-negative step
functions which converges to zero for a.e. x R, and hence by Lemma 3.4 (the direction of
R
R
R
R
that lemma) we have lim` R (k ` )+ = 0 for each k, and hence R k R ` = R (k ` )
R
R
R
R
R
R (k ` )+ 0 as ` . Thus R k lim R ` for each k and hence lim R k lim R k .
The reverse inequality is proved by interchanging k and k .
Thus the definition of the Lebesgue integral makes sense on L+ (R) and evidently has the additivity
property that


R
R
R
4.3
g, h L+ (R) and , 0 g + h L+ (R) and R f + g = R g + R h,
by 4.1, Definition 4.2, and the linearity 1.4 of the integral on step functions.
We also observe that (again by Definition 4.2)
4.4

f : R [0, ), f = 0 a.e. in R f L+ (R) and

because the zero sequence 0, 0, . . . of step functions converges a.e. to f .

Rf

= 0,

Riemann & Lebesgue Integrals

We can now immediately extend the definition of the Lebesgue integral to the vector space L1 (R)
as follows:
R
R
R
4.5 Definition: If f L1 (R) then we define R f = R g R h, where g, h L+ (R) are such that
f = g h.
We of course have to check that this definition is independent of which g, h L+ (R) we choose to
represent f . So suppose g, h, ge, e
h L+ (R) and g h = ge e
h. Then g + e
h = ge + h and by the
R
R
R
R
e
additivity 4.3 we then have R g R h = R ge R h.
Notice that by 4.3 and the Definition 4.5 we have the linearity
4.6

f, g L1 (R), , R f + g L1 (R) and

R
R

R
R
f + g = R f + R g.

We now check some of the basic further properties of the Lebesgue integral on L1 (R):
First we show that changing an L1 (R) function on a set of measure zero neither changes the fact
that the function is in L1 (R) nor does it change the value of the integral:
4.7

f L1 (R), fe : R R, and fe = f a.e. fe L1 (R) and

Rf

Rf .

This is clear because fe = f + (fe f ) = f + (fe f )+ (fe f ) and both (fe f ) are in L+ (R)
and have integral zero by 4.4.
Next observe that if f L1 (R) with f 0 then by definition we have g, h L+ (R) with f = g h
R
R
and g h (because f 0), and hence, by Definition 4.2 and by 1.5, R g R h, and Definition 4.5
R
R
R
R
then implies R f = R g R h 0. Thus f L1 (R), f 0 R f 0. Since f, g L1 (R) with
f g f g 0 we can use the linearity 4.6 to conclude
4.8

f, g L1 (R) with f g

Rf

R g.

Next we observe that L1 (R) is closed under the operation of taking absolute values:
R  R


4.9
f L1 (R) |f | L1 (R) and  R f  R |f |.
To check this we choose g, h L+ (R) with f = g h and observe that then |f | = max{g, h}
min{g, h} and also max{g, h}, min{g, h} L+ (R) (by 4.1), and so |f | L1 (R) and, by the DefiniR
R
R
R
R
R
tion 4.5 and by 4.8, R f = R g R h R max{g, h} R min{g, h} = R |f |, and then the proof
of 4.9 is completed by applying the same argument to f . In view of 4.9 we also have
4.10

f, g L1 (R) max{f, g}, min{f, g} L1 (R)

because max{f, g} = 12 (f + g + |f g|) and min{f, g} = 12 (f + g |f g|).

The next property seems intuitively obvious but the proof requires Lemma 3.4:
4.11

f L1 (R), f 0,

Rf

= 0 f = 0 a.e.

To check 4.11 choose g, h L+ (R) with f = g h, and then observe that by definition of L+ (R)
we have increasing sequences {` }`=1,2,... and {k }k=1,2,... S + (R) with ` g a.e. and k h

10

Math. 171 Lecture Supplement 2012

R
R
a.e., and hence, with = sup ` , limk R ( (` k )+ ) = R ( (` h)+ ) by Definition 4.2.
R
R
R
Hence limk R (` k )+ = R (` h)+ R f , where at the last step we used the hypothesis
R
f 0. Since R f = 0 we can then apply Lemma 3.4 (the direction of that lemma with
(` k )+ in place of k ) to give (` h)+ = lim(` k )+ = 0 a.e., and hence ` h 0 a.e.,
so by taking the pointwise limit with respect to ` we have f = g h 0 a.e., and hence f = 0 a.e.
as claimed.
Well now check that the Lebesgue integral exists and agrees with the Riemann integral whenever
the Riemann integral existsthus the Lebesgue integral entirely subsumes the Riemann integral.
At the same time well give an alternate characterization (supplementary to Theorem 3.3) of exactly
which of the Lebesgue integrable functions are Riemann integrable.
Using Lemma 2.1 with = 1/k we see that if f : R R is Riemann integrable then there are step
R
R
R
R
functions k , k with k f k and R k 1/k R k RI(f ) R k R k + 1/k for
R
R
k = 1, 2, . . ., where RI(f ) is the Riemann integral of f , and hence R k and R k both converge
to RI(f ), where k = min{1 , . . . , k } and k = max{1 , . . . , k }. Thus we have k f k ,
k is a bounded increasing sequence of step functions, and k is a bounded decreasing sequence
R
R
of step functions. Also R (k k ) R (k k ) 0, so by Lemma 3.4 (the direction of
that lemma) we have k (x) k (x) 0 a.e. x R. Since k f k for all x we thus have
e+ (R) L
e (R) where L
e+ (R) denotes
lim k (x) = lim k (x) = f (x) a.e. x R. But then f L
the set of functions f : R R such that there is an increasing sequence {k }k=1,2,... S(R) with
e (R) = {f : f L
e + (R)}. Also, observe that,
k f on R and f (x) = lim k (x) a.e. x R and L
with = supR |1 |, k + S + (R), so by definition of the Lebesgue integral on L+ (R) we have
R
R
R
R
R ( + f ) = limk R ( + k ), hence R f = limk R k which is also the Riemann integral of
f by construction. Thus we have proved:
4.12 Lemma. The set R(R) of Riemann integrable functions f : R R is a linear subspace of
e (R) and the Riemann integral and the Lebesgue integral coincide on R(R).
e+ (R) L
L
e+ (R) L
e (R) then we can find an increasing
Actually there is also a partial converse: If f L
sequence k and a decreasing sequence k of step functions with f (x) = lim k (x) = lim k (x) a.e.
x R. For each k, let P k be a partition of R such that both k and k are constant on each I
with I P k . Evidently we must then have k |I k |I for each I P k . Let M > 0 be such
that M < 1 and 1 < M everywhere on R. Then we can define
ek (x) = k (x) for x IP k I
and
ek (x) = M on IP k I, and similarly ek (x) = k (x) for x IP k I and ek (x) = M on
IP k I. Then
ek (x) ek (x) at every point and they converge to f (x) a.e. Indeed if E is a set
of measure zero such that both k (x) and k (x) converge to f (x) at every point of R \ E, then
by construction
ek (x) and ek (x) both converge to f (x) for every x R \ (E (
k=1 IP k I)).
However
ek , ek are not monotone sequences. To remedy this we define k = max{
e1 , . . . ,
ek }
e
e
and k = min{1 , . . . , k }. These are respectively increasing and decreasing sequences of step
functions with M k k M everywhere on R and k (x), k (x) converge to f (x) at every
e
point x R \ (E (
k (x) for every x R, then we
k=1 IP k I)). Thus if weRlet f (x) = limk
R
have k fe k for each k, fe = f a.e., and R (k k ) R (k k ) 0, so fe is Riemann
integrable by Lemma 2.1. Thus we have proved:
e+ (R) L
e (R) then there is a Riemann integrable fe with fe = f a.e.
4.13 Lemma. If f L

Riemann & Lebesgue Integrals

11

e+ (R)
Remark: Note that by combining Lemma 4.12 and Lemma 4.13 we have proved that f L
e (R) f is a.e. equal to a Riemann integrable function fe, and in that case the Riemann
L
integral of fe is equal to the Lebesgue integral of f (also equal to the Lebesgue integral of fe by 4.7).
We conclude this section with a discussion of some important convergence theorems. First:
4.14 Theorem (Monotone Convergence Theorem.) If {fk } is an increasing sequence in
R
L1 (R) with { R fk }k=1,2,... bounded, then f L1 (R) such that fk (x) f (x) a.e. x R and
R
R
R fk R f . Furthermore we can take f L+ (R) if fk L+ (R) for each k.
Proof: We consider first Case 1: fk L+ (R) for each k = 1, 2, . . .. In this case, by definition
of L+ (R), for each k = 1, 2, . . . we can find an increasing sequence {k,j }j=1,2,... S + (R) with
R
R
k,j (x) fk (x) a.e. x R and R (fk k ) 2k , where k = k,1 . Then R (k,j k )
2k , k, j = 1, 2, . . ., and hence
R Pj
k=1 (k,j k ) 1 j = 1, 2, . . . ,
R
P
and also { jk=1 (k,j k )}j=1,2,... is an increasing sequence in S + (R). So by Lemma 3.5 we see that
P
{ jk=1 (k,j (x) k (x))}j=1,2,... is bounded a.e. x R. Thus a.e. x R there is a fixed constant
P
Cx such that jk=1 (k,j (x) k (x)) Cx for all j, and hence, since {k,i }i=1,2,... is increasing, we
have, a.e. x R,
P
k=1 (k,i (x) k (x)) Cx i = 1, 2, . . . .
Since limi k,i (x) = fk (x) a.e. x R, by taking limi of the partial sums of the above series
we conclude
P
k=1 (fk (x) k (x)) Cx a.e. x R.
In particular fk (x) k (x) 0 a.e. x R, and hence k = max{1 , . . . , k } is an increasing
sequence in S + (R) with fk (x) k (x) 0 a.e. x R, because k k fk for each k. The latter
R
R
R
inequality implies R k R fk for each k, and so { R k }k=1,2,... is bounded and, by Lemma 3.5,
{k (x)} is bounded a.e. x R and we can define f L+ (R) by taking f (x) = lim k (x) for x such
that {k (x)}k=1,2,... is bounded and f (x) = 0 elsewhere. Then lim fk (x) = lim k (x) = f (x) a.e.
R
R
x R, and, by Definition 4.2, R f = lim R k . Since k fk on R and fk f a.e. on R we then
R
R
also have R fk R f . This completes the proof in Case 1.
R
Case 2: The general case when {fk } is an increasing sequence in L1 (R) with { R fk }k=1,2,... bounded.
In this case set f0 f1 and observe that, for k = 1, 2, . . ., fk fk1 L1 (R) and hence there are
gk , hk L+ (R) with fk fk1 = gk hk , and by Definition 4.2 we can select k S + (R) with
R
k hk and R (hk k ) < 2k . Then gk k = (hk k ) + (fk fk1 ) 0 and also
R
R
k
+ R (fk fk1 ).
R (gk k ) 2
Thus with
Pk

j=1 (gj

j ),

Pk

j )
R
R
we see that {Gk }, {Hk } are increasing sequences in L+ (R) with { R Gk }k=1,2,... , { R Hk }k=1,2,...
bounded, and Gk Hk = fk f1 . By Case 1 above we have G, H L+ (R) such that Gk G
R
R
R
R
and Hk H a.e. in R and R Gk R G, and R Hk R H. So with f = G H + f1 we have
R
R
f L1 (R), fk f a.e. and R fk R f . This completes the proof of the monotone convergence
theorem.
Gk =

Hk =

j=1 (hj

12

Math. 171 Lecture Supplement 2012

R
We can now prove the important corollary that for any sequence {fk } L1 (R) with { R |fk |}k=1,2,...
bounded, pointwise convergence a.e. to some f implies f L1 (R):
R
4.15 Corollary. Suppose {fk }k=1,2,... L1 (R) with { R |fk |}k=1,2,... bounded, and suppose there is
f : R R with lim fk (x) = f (x) a.e. x R. Then f L1 (R).
Proof: Since we can write fk = fk+ fk , where fk+ = max{fk , 0} and fk = max{fk , 0},
so that fk L1 (R) by 4.10 and fk+ (x) f+ (x) and fk (x) f (x) a.e. x R, we see
that it suffices to prove the corollary under the additional assumption that fk 0 for each k.
So for each k, ` = 1, 2, . . . we assume fk 0 and we let Fk,` = min{fk , fk+1 , . . . , fk+` } and
observe that the sequence {Fk,` }`=1,2,... is monotone increasing and bounded above by 0, and of
course in L1 (R) by 4.10. Thus by the Monotone Convergence Theorem inf{fk , fk+1 , . . .} L1 (R)
because inf{fk (x), fk+1 (x), . . .} = lim` Fk,` (x) for each x R. But clearly inf{fk , fk+1 , . . .}
is increasing and it is fk on R for each k = 1, 2, . . .. It also evidently has limit f (x) a.e.
R
x R, so we can apply the Monotone Convergence Theorem again to infer f L1 (R) and R f =
R
lim R inf{fk , fk+1 , . . .}.
R
R
4.16 Remarks (1): The above does not show that R f = limk R fk and indeed this is not true in
R
generalobviously, because e.g. it may not even be true that the bounded real sequence R fk has
a limit. The above proof in fact only shows that, for non-negative fk L1 (R) with fk f a.e.
R
and R fk bounded,
()

inf f` L1 (R) and

`k

R f = lim


R 

lim
inf
inf
f
`
R f` .
R

`k

k `k

For example in the case n = 1 it is easy to construct a sequence fk of non-negative step functions
R
such that R fk = 1 for each k and fk (x) 0 x R, thus showing that equality does not hold in
R
R
general in the inequality R f limk inf `k R f` .
(2): The above corollary makes it elementary to check that various algebraic operations leave
L1 (R) invariant. For example
(
f, g L1 (R) and g bounded f g L1 (R)
f, g L1 (R) and g 1 f /g L1 (R)
which are easily checked by using the definition L1 (R) = L+ (R) L+ (R) in combination with the
corollary.
R
R
As Remark 4.16(1) above makes clear, we cannot in general conclude R f = lim R fk under the
hypotheses of 4.15. However if a sequence {fk }k=1,2,... L1 (R) has the additional property that
there is F L1 (R) such that |fk | F k then, still assuming fk f a.e., () can be applied to
both F + fk and F fk in order to give inf `k f` , sup`k f` L1 (R) and
R
R
R
R
R
R
lim R inf f` R F + lim inf R f` ,
R F + R f = R F + k
`k
k `k
R
R
hence R f limk inf `k R f` , and
R
R
R
R
R
R
lim R sup f` R F lim sup R f` ,
R F R f = R F k
k `k
`k
R
R
R
R
hence limk sup`k R f` R f limk inf `k R f` , and so limk R fk exists and is equal to
R
R f . That is, we have proved:

13

Riemann & Lebesgue Integrals

4.17 Theorem (Dominated Convergence Theorem). Suppose fk L1 (R), f : R R with

fk (x) f (x) a.e. x R, and suppose there is a function F L1 (R) such that |fk | F k. Then
R
R
f L1 (R) and limk R fk = R f .

The Spaces L1 (R) and L2 (R)

We conclude these notes with a discussion of some important further properties of the space L1 (R),
and we also introduce the linear subspace L2 (R).
First, for f L1 (R) we define
kf k1 =

R |f |,

and observe that k k1 has the first two properties of a norm: kf k1 = ||kf k1 and kf + gk1
kf k1 + kgk1 , but in place of the usual positivity we have only kf k1 0 for each f L1 (R), because
kf k1 = 0 evidently does not imply f = 0 but rather that f = 0 a.e. by 4.11. We therefore classify
k k1 as a seminorm rather than a norm, but we keep in mind that the positivity of k k1 fails in
very mild way, namely (by 4.11)
kf k1 = 0 f (x) = 0 a.e. x R.
Here we establish the following completeness property of L1 (R):
5.1 Theorem. Suppose {fk }k=1,2,... is a Cauchy sequence with respect to the seminorm k k1 above;
R
that is, for each > 0 there is N such that R |fk f` | < whenever both k, ` N . Then there is
R
f L1 (R) such that R |fk f | 0 and such that, for some subsequence {fkj }j=1,2,... of {fk }k=1,2,... ,
fkj (x) f (x) a.e. x R.
Proof: We use the definition of Cauchy sequence with = 2j , j = 1, 2, . . ., in order to select kj
R
such that R |f` fkj | < 2j for all ` kj , and by selecting these kj successively we can at the
same time arrange that kj+1 > kj for all j 1, so that we have in particular that {fkj }j=1,2,... is a
subsequence of {fk }k=1,2,... and
kfkj+1 fkj k1 < 2j
Pi
for each j. In view of (1) we thus have that
j=1 (fkj+1 fkj )+ , i = 1, 2, . . . is an increasing
sequence in L1 (R) with integrals bounded above by the fixed constant 1, hence by the Monotone
P
1
Convergence Theorem (4.14) we have g =
j=1 (fkj+1 fkj )+ L (R) and has integral equal to
R Pi
P
limi R j=1 (fkj+1 fkj )+ . Likewise h = j=1 (fkj+1 fkj ) L1 (R) and has integral equal
R P
to limi R ij=1 (fkj+1 fkj ) .
(1)

Thus with f = g h + fk1 we have, a.e. in R,

|fki+1 f | = |

i
X
j=1

(fkj+1 fkj ) g + h| (g

i
X

(fkj+1 fkj )+ ) + (h

j=1

i
X

(fkj+1 fkj ) ),

j=1

so fki (x) f (x) a.e. x R and kfki+1 f k1 0. Of course then limk kf fk k1 = 0 because
{fk } is Cauchy with respect to the L1 seminorm.
We conclude this discussion by introducing the space L2 (R):
5.2 Definition: L2 (R) = {f L1 (R) : f 2 L1 (R)}.

14

Math. 171 Lecture Supplement 2012

At first sight it is not obvious that L2 (R) is a linear subspace of L1 (R), but one can very easily
check this using the Dominated Convergence Theorem and Remark 4.16(2) as follows: By 4.16(2),
f, g L2 (R) f 1+kg1 |g| L1 (R) for each k = 1, 2, . . .. Also, |f 1+kg1 |g| | |f g| 21 (f 2 + g 2 )
L1 (R) and f
That is
()

g
1+k1 |g|

f g pointwise, so by the Dominated Convergence Theorem f g L1 (R).

f, g L2 (R) f g L1 (R).

Since (f + g)2 = f 2 + g 2 + 2f g we thus have f, g L2 (R) f + g L2 (R). Since it is also clearly

true that f L2 (R) f L2 (R), this completes the proof that L2 (R) is a linear subspace of
L1 (R).
In view of (), we can define an inner product (or more correctly a semi inner product) hf, gi
on L2 (R) by
R
5.3
hf, gi = R f g, f, g L2 (R).
Notice this has properties analogous to the dot product of vectors in Rn :
(i)
(ii)
(iii)

hf, gi = hg, f i

(symmetry)

hf + g, hi = hf, hi + hg, hi

(linearity)

hf, f i 0 with equality f = 0 a.e.,

(positivity)

f2

where in checking (iii) we used 4.11 (with

in place of f ). Using the usual argument (that,
2
by (i)(iii), 0 hf + tg, f + tgi = t hg, gi + 2thf, gi + hf, f i, which, for hg, gi =
6 0, takes the non2
negative minimum value of (hf, f ihg, gi hf, gi )/hg, gi when t = hf, gi/hg, gi) we then have the
Cauchy-Schwarz inequality:


hf, gi kf k2 kgk2 ,
5.4
p
where kf k2 = hf, f i. kf k2 is called the inner product semi-norm of f L2 (R). Since kf +gk22 =
kf k22 + kgk22 + 2hf, gi kf k22 + kgk22 + 2kf k2 kgk2 by 5.4, we thus have the triangle inequality
5.5

kf + gk2 kf k2 + kgk2 ,

f, g L2 (R).

We claim finally that L2 (R) is complete with respect to the semi-norm k k2 :

5.6 Theorem. The space L2 (R) is complete relative to the inner product semi-norm; that is if
{fk }k=1,2,... L2 (R) is Cauchy (i.e. for each > 0 there is N such that k ` N kfk f` k2 < )
then there is f L2 (R) such that lim kfk f k2 = 0.
Proof: Without loss of generality, because fk = fk+ fk and {fk } Cauchy for k k2 implies both
{fk+ } and {fk } are Cauchy for k k2 , we can assume fk 0 on R for each k. Then observe
R
R
kfk2 f`2 k1 = R |fk2 f`2 | = R |fk f` |(fk + f` ) kfk f` k2 kfk + f` k2 kfk f` k2 (kfk k2 + kf` k2 ),
where we used 5.4,5.5. Since kfk k2 , kf` k2 1 + kfN k for k, ` N , with N as in the definition of
Cauchy sequence with = 1, we have thus shown that fk2 is a Cauchy sequence with respect to the
semi-norm k k1 of L1 (R) and so by Theorem 5.1 there is h L1 (R) with kfk2 hk1 0 and, also
by 5.1, there is a subsequence fk2j with fk2j (x) h(x) a.e. x R. Since kfk f` k1 |R|1/2 kfk f` k2
(by Cauchy-Schwarz again), we also have that fk is Cauchy with respect to the semi-norm k k1 of
L1 (R), and hence there is an f L1 (R) such that kfk f k1 0. Theorem 5.1 also guarantees
that there is a subsequence {fkj` }`=1,2,... of {fkj }j=1,2,... which converges a.e. to f . But then fk2j
`

converges a.e. to f 2 and so h = f 2 a.e., and we have proved that f L2 (R) and kfk2 f 2 k1 0,
R
R
R
which implies kfk f k22 = R |fk f |2 R |fk f |(fk + f ) = R |fk2 f 2 | = kfk2 f 2 k1 0.