1.

σ-algebras
Definition 1. Let X be any set and let F be a collection of subsets of
X. We say that F is a σ-algebra (on X), if it satisfies the following.
(1) X ∈ F.
(2) If A ∈ F, then Ac ∈ F.
(3) If A1 , A2 , · · · ∈ F, a countable collection, then ∪∞
n=1 An ∈ F.

In the above situation, the elements of F are called measurable sets

and X (or more precisely (X, F)) is called a measurable space.
Example 1. For any set X, P(X), the set of all subsets of X is a
σ-algebra and so is F = {∅, X}.
Theorem 1. Let X be any set and let G be any collection of subsets
of X. Then there exists a σ-algebra, containing G and smallest with
respect to inclusion.
Proof. Let S be the collection of all σ-algebras containing G. This set
is non-empty, since P(X) ∈ S. Then F = ∩A∈S A can easily be checked
to have all the properties asserted in the theorem. 
Remark 1. In the above situation, F is called the σ-algebra generated
by G.
Definition 2. Let X be a topological space (for example, a metric space)
and let B be the σ-algebra generated by the set of all open sets of X.
The elements of B are called Borel sets and (X, B), a Borel measurable
space.
Definition 3. Let (X, F) be a measurable space and let Y be any topo-
logical space and let f : X → Y be any function. We say that f is
measurable, if for any open set U ⊂ Y , f −1 (U ) ∈ F.
Example 2. (1) If (X, F) is a measurable space with X a topologi-
cal space and B ⊂ F, then any continuous function from X to
any topological space is measurable.
(2) If f : X → Y is measurable and g : Y → Z is continuous, g ◦ f
is measurable.
Theorem 2. Let (X, F) be a measurable space and let f : X → Y be
a function, where Y is any set.
(1) If G is defined as the collection of all subsets A of Y such that
f −1 (A) ∈ F, then G is a σ-algebra on Y .
(2) If f : X → [−∞, ∞] is any function, f is measurable if and
only if f −1 ((a, ∞]) ∈ F for any a ∈ [−∞, ∞].
1
2

Proof. The first part is mere checking. For the second part, notice that
since (a, ∞] is open in [−∞, ∞], the condition is clearly necessary. To
check sufficiency, one notes that [−∞, b) = ∪n [−∞, b− n1 ] for any b and
[−∞, c] = (c, ∞]c . So, one gets that f −1 ([−∞, b)) ∈ F too for any b and
thus we get f −1 ((a, b)) ∈ F for any a, b, since (a, b) = [−∞, b) ∩ (a, ∞]
and these generate all open sets. 
Theorem 3. Let fn : X → [−∞, ∞] be a sequence of measurable
functions. Then, supfn , inffn , lim supfn , lim inffn are all measurable.
Proof. Let g = supfn . Then g −1 ((a, ∞]) = ∪n fn−1 ((a, ∞]) and so we
are done by the previous theorem. A similar proof applies to infimum.
Let gn = supk≥n fk . Then all the gn ’s are measurable by the previous
part. Then lim supfn = infgn and hence it too is measurable. Similar
argument applies to lim inf. 
We get as a corollary the following results.
Corollary 1. Let f, g : X → [−∞, ∞] be measurable. Then so are
max{f, g} and min{f, g}. In particular, writing f + = max{f, 0} and
f − = − min{f, 0}, we see that these are measurable.
Remark 2. In all the above, we have dealt with [−∞, ∞] which has the
advantage that supremum, lim sup etc. make sense. But, notice that
if a, b ∈ [−∞, ∞], then a + b may not make sense. On the other hand,
if we worked with R, then the latter obviously makes sense, but not
necessarily the former. In most cases, it should be clear which space
we are working in and why.
Definition 4. A function f : X → [0, ∞) is called simple, if it takes
only finitely many values.
Example 3. (1) If A ⊂ X, the characteristic function χA defined as
χA (x) = 1 if x ∈ A an zero otherwise, is a simple function.
(2) More generally, if s is a simple function, let a1 , . .P
. , an be its
finitely many values and let Ai = s (ai ). Then s = ni=1 ai χAi .
−1

Notice that ∪i Ai = X and Ai ∩ Aj = ∅ if i 6= j (pairwise

disjoint).
(3) If s is a simple function as above from a measurable space X,
s is measurable
P if and only
P if all the Ai ’s are measurable.
(4) If s = ai χAi and t = bj χBj are simple (measurable) func-
tions, soPis s + t. Letting Cij = Ai ∩ Bj , one can see that
s + t = (ai + bj )χCij .
Theorem 4. Let f : X → [0, ∞] be a measurable function. Then
there exists simple measurable functions s1 ≤ s2 ≤ · · · ≤ f such that
lim sn = f .
 3

Proof. For any n and an i such that 1 ≤ i ≤ n2n , let En,i = f −1 ([ i−1 , i ))
2n 2n
and let Fn = f −1 ([n, ∞]). Then these are all measurable sets, pairwise
Pn2n i−1
disjoint for a fixed n and cover all of X. So, sn = i=1 χ
2n En,i
+ nχFn
is a simple measurable function. Easy to check that sn ≤ sn+1 ≤ f for
all n and lim sn = f . 

2. Measures
Definition 5. Let (X, F) be a measurable space. A (positive) measure
on X is a function µ : F → [0, ∞] which is countably additive. That
is, if A1 , A2 , . . . ∈ FPis a countable collection of pairwise disjoint sets,
∞
then µ(∪∞ n=1 An ) = n=1 µ(An ).

Since µ(A) = ∞ for all A ∈ F is trivially a measure by the above,

to avoid this case, we will always assume that there exists some A ∈ F
with µ(A) < ∞ for our definition.
Example 4. (1) Let X be any set and consider the function µ :
P(X) → [0, ∞] given by µ(A) is the cardinality of A if A is
finite and ∞ otherwise. Then µ is a positive measure. This is
called the counting measure.
(2) Again consider (X, P(X)) and let a ∈ X, fixed. Define µ(A) =
1 if a ∈ A and zero otherwise. This is a positive measure.
Theorem 5. Let µ be a positive measure on (X, F).
(1) µ(∅) = 0.
n
PnA1 , A2 , . . . , An ∈ F, pairwise disjoint, then µ(∪i=1 Ai ) =
(2) If
i=1 µ(Ai ). (Finite additivity)
(3) If A ⊂ B are measurable, µ(A) ≤ µ(B). (Monotonicity)
(4) If A1 ⊂ A2 ⊂ · · · are measurable sets and A = ∪∞ n=1 An , then
limn→∞ µ(An ) = µ(A).
(5) If A1 ⊃ A2 ⊃ · · · and A = ∩∞ n=1 An with µ(A1 ) < ∞, then
lim µ(An ) = µ(A).
Proof. (1) Let A be such that µ(A) < ∞. Let A1 = A and An = ∅
for n > 1. Then these are pairwise
P disjoint and its union is A.
So, we get µ(A) = µ(A) + µ(∅). Rest is clear, since we can
cancel µ(A) < ∞ from bot sides.
(2) This is clear by taking Am = ∅ for m > n and using the first.
(3) We can write B = A ∪ (B − A), pairwise disjoint and thus
µ(B) = µ(A) + µ(B − A) ≥ µ(A).
(4) Let Bn = An − An−1 for n > 1 and B1 = A1 . Then {Bn } are
∞ n
Pn and ∪n=1 Bn = A. Also,
pairwise disjoint P∞ ∪i=1 Bi = An . So, we
get µ(An ) = i=1 µ(Bi ) and µ(A) = n=1 µ(Bn ).
4

(5) For this, let Cn = A1 − An . Then Cn ⊂ Cn+1 and ∪∞ n=1 Cn =

A1 −A. So, from the previous part, we have lim µ(Cn ) = µ(A1 −
A) = µ(A1 ) − µ(A), since A ⊂ A1 . We also have by the same
reason, µ(Cn ) = µ(A1 ) − µ(An ). So, taking limits, we get,
lim µ(Cn ) = µ(A1 ) − lim(An ) = µ(A1 ) − µ(A) and canceling
µ(A1 ) < ∞, we are done.


3. Lebesgue Integral
Definition 6. Let (X, F, µ) be a measure space and let P s : X → [0, ∞)
be
R a simplePpositive measurable function. Writing s = ai χAi , define
X
sdµ = ai µ(Ai ), where we use the convention, 0 · µ(A) = 0 even
if
R µ(A) = ∞. If 0 ≤ f is any positive
R measurable function, we define
X
f dµ to be the supremum of X sdµ for simple measurable functions
0 ≤ s ≤ f.

Remark 3. The above gives two possible definitions for the integral of a
simple measurable function, but it is clear that the definitions coincide.

We collect a few easy facts which follow from the definitions.

Theorem 6. Let (X, F, µ) be a measure space. All functions men-

tioned below are measurable and so are all the subsets of X.
R R
(1) If 0 ≤ f ≤ g, then X f dµ ≤ X gdµ. R R
(2) If A ⊂ B in F, for 0 ≤ f on X, R we have, RA f dµ ≤ B
f dµ.
(3) If 0 ≤ f and 0 ≤ c < ∞, thenR X
cf dµ = c X
f dµ.
(4) If f (x) = 0 for all x ∈ A, A f dµ = 0, even if R µ(A) = ∞.
(5) If µ(A) = 0, for any measurable positive f , A f dµ = 0 even if
Rf (x) = ∞ forR some or all x ∈ A.
(6) X f χA dµ = A f dµ. R R R
(7) If s, t are simple functions, then X (s + g)dµ = X sdµ + X tdµ

Proof. The
P only part whichPrequires a proof isR the last one.
P
R If s = P a i χ A i
and t = b j χ Bj
, we have X
sdµ = Pai µ(Ai ) and
X
tdµ = bj µ(Bj ). On the other hand, we have s+t = (ai +bj )χCij ,
 5

where Cij = Ai ∩ Bj . Thus

Z X
(s + t)dµ = (ai + bj )µ(Cij )
X i,j
X X X X
= ai µ(Cij ) + bj µ(Cij )
i j j i
X X
= ai µ(Ai ) + bj µ(Bj )
Z Z
= sdµ + tdµ
X X

As a corollary, we get,
Corollary
R 2. Let s be a simple measurable function on X and define
φ(A) = A sdµ for A measurable. Then φ is a measure.
Proof. Let s = ni=1 ai χAi . If B1 , B2 , . . . is a countable collection of
P
pairwise disjoint measurable sets and B is their union, we have,
Xn X n X∞
φ(B) = ai µ(Ai ∩ B) = ai µ(Ai ∩ Bj )
i=1 i=1 j=1
X∞ Xn ∞
X
= ai µ(Ai ∩ Bj ) = φ(Bj )
j=1 i=1 j=1

Since φ(∅) = 0 < ∞, we see that φ is indeed a measure in our sense. 

Next we come to an important theorem.
Theorem 7 (Lebesgue Monotone Convergence Theorem). Let 0 ≤
f1 ≤ f2 ≤ · · · ≤ f be a sequenceR of positive
R measurable functions
converging to f . Then limn→∞ X fn dµ = X f dµ.
R R R
Proof. Let α = R limn→∞ X
f n dµ. Since X
f n dµ ≤ X
f dµ for all n, we
get that α ≤ X f dµ.
Next we prove the reverse inequality. For this, let 0 ≤ Rs ≤ f be a
simple measurable function. It suffices to prove that α ≥ X sdµ. Let
0 < c < 1 and consider the sets An = {x ∈ X|fn (x) ≥ cs(x)}.R Then An
R measurable,R A1 ⊂ A2 ⊂ · · · and ∪An = X. Also,
is R we have XRfn dµ ≥
f dµ ≥ c An sdµ for all n. Thus, α = lim X fn dµ ≥ lim c An sdµ.
An n
Noting that the latter is a measure from the previousRcorollary and
using an earlier theorem, we get that it is equalR to c X sdµ. Since
this is true for any 0 < c < 1, we see that α ≥ X sdµ, finishing the
proof. 
6
R
Corollary
R R 3. If f, g : X → [0, ∞] are measurable, X
(f + g)dµ =
X
f dµ + X gdµ.
Proof. By arguments as in the homework, it is immediate that f + g
is measurable. Let 0 ≤ s1 ≤ s2 ≤ · · · ≤ f be simple function such that
lim sn = f (see Theorem 4). Similarly, 0 ≤ t1 ≤ t2 ≤ · · · ≤ g. Then
sn +tn is simple and 0 ≤ s1 +t2 ≤ s2 +t2 ≤
R · · · ≤ f +g with lim(s
R n +tn ) =
f + g. So, by R Theorem 7, weR get lim RX (sn + tn )dµ = X (f + g)dµ.
But we have X (sn + tn )dµR = X sn dµ R + X tn dµ, sinceR these are Rsimple
functions. We have lim X sn dµ = X f dµ and lim X tn dµ = X gdµ
by Theorem 7 and so we are done. 
Corollary 4. Let fnP: X → [0, ∞] be a sequence of measurable func-
∞ R
tions
P∞ R and let f = n=1 nf . Then f is measurable and X
f dµ =
n=1 X fn dµ.
Pn
Proof. Let gn = k=1 fk . So, we have lim R gn = f Rwith 0 ≤ g1 ≤
g2 ≤ · · · ≤ f and thus by Theorem 7, lim X gn dµ = X f dµ. On the
other hand,P by a simple induction fromRthe previous P∞corollary, we have,
R n R R
g
X n
dµ = f
k=1 X k dµ and thus lim X n
g dµ = k=1 X kf dµ. 
P∞ P∞ P∞ P∞
Corollary 5. If aij ≥ 0 are real numbers, i=1 j=1 aij = j=1 i=1 aij .
Proof. Consider X = N with its power set as a σ-algebra and µ, the
counting measure. Let fn : X → [0, ∞) be defined as fn (m) = amn .
Then fn is measurable (if the σ-algebra is the power set, any function
is measurable) and so we have, from the previous corollary,
Z X ∞ X∞ Z
fn dµ = fn dµ.
X n=1 n=1 X

For any g : X → [0, ∞], we also have X gdµ = ∞

R P
n=1 g(n). For this,
consider simple functions sn : X → [0, ∞] defined as sn (k) = g(k) if
k ≤ Rn and zero other
R wise. Then P0 ≤ s1 ≤ s2 ≤ · · · ≤ g and lim sn = g.
So, X
= lim X sn dµ. sn =P nk=1 g(k)χ{k} + 0χ{k > n} and thus
gdµ P
s dµ = nk=1 g(k)µ({k}) = nk=1 g(k). Rest is clear.
R
X n

Theorem 8 (Fatou’s lemma). Let fn : X → [0, ∞] be a sequence of
measurable functions. Then,
Z Z
lim inf fn dµ ≤ lim inf fn dµ.
X X
Proof. Notice first, that we have proved lim inf fn is measurable, so the
left hand side above makes sense.
Consider gk = inf n≥k fn . Then gk is measurable and we have 0 ≤
g1 ≤ g2 ≤ · · · ≤ lim inf fn and lim gk = lim inf fn . Thus, by Theorem
 7
R
7, we get the left hand side in Rthe theorem R to be lim X gk dµ. But
gRk ≤ fn for all n R≥ k and thus X gk dµ ≤ X fn dµ for all n ≥ k. So,
g dµ ≤ inf n≥k X fn dµ. So,
X k
Z Z Z
lim gk dµ ≤ lim inf fn dµ = lim inf fn dµ.
X k n≥k X X


Theorem 9. Let f : X → [0, ∞] R be measurable. The the function
φ : F → [0, ∞] given by φ(A) = A f dµ isRa positiveR measure. If g is
any positive measurable function, we have X gdφ = X gf dµ
Proof. We have φ(∅) = 0. If A1 , A2 , . . . is a countablePcollection of
measurable
R disjoint R sets with AP=R ∪An , we have χA = χAi . Thus,
φ(A) = A f dµ = X χRA f dµ = X
χ
R Ai f dµ, the last by the corollary
to Theorem 7. Since X χAi f dµ = Ai f dµ = φ(Ai ), we get countable
additivity for φ proving that it is indeed a measure.
For the last part, we consider g = χA for some A ∈ F. Then,
Z Z Z
gdφ = φ(A) = f dµ = gf dµ.
X A X

Thus, the same holds for any simple measurable function and then
Theorem 7 finishes the proof for any g, since any positive measurable
function is the limit of an increasing sequence of simple functions. 

4. L1 spaces
In this section, we will consider functions f : X → R, not necessarily
positive, but only taking finite values. If f is measurable, we have seen
that both f + , f − are measurable and thus so is |f | = f + + f − .
DefinitionR 7. A measurable function f : X → R belongs to the set
L1 (µ) if X |f |dµ < ∞.
If f ∈ L1 (µ), then clearly |f | ∈ L1 (µ). Also, both X f + dµ, X f − dµ
R R
are finite and hence it makes sense to define,
Z Z Z
f dµ = +
f dµ − f − dµ.
X X X
R
For a measurable function f , not necessarily positive, X f dµ may not
make sense, since the only reasonable
R way we could do this
R was to write
f = f + − f − and attempt to define X f dµ = X f + dµ − X f − dµ. The
R
subtraction above will make sense only if at least on those integrals
were finite.
8

Theorem 10. The set L1 (µ) is a vector space over R. That is, given
f, g ∈ L1 (µ), a, b ∈ R, af + bg ∈ L1 (µ).
Proof. We already know that af + bg is measurable (since they take
finite values). So,
Z Z Z Z
|af +bg|dµ ≤ (|a||f |+|b||g|)dµ ≤ |a| |f |dµ+|b| |g|dµ < ∞.
X X X X

R R
Theorem 11. If f ∈ L1 (µ), | X f dµ| ≤ X |f |dµ.
R R R
Proof.
R If a = X f dµ, then | X f dµ| = θa where θ = ±1. So, | X f dµ| =
X
θf dµ. But, θf ≤ |θf | = |f | and so we are done. 
Theorem 12 (Lebesgue Dominated Convergence Theorem). Let fn :
X → R be a sequence of measurable functions converging pointwise to
1
a function f : X → R. Assume R exists a g ∈ L (µ) such that
R that there
|fn | ≤ g for all n. Then, lim X fn dµ = X f dµ.
Proof. Clearly, we have |f | ≤ g and thus |fn − f | ≤ 2g for all n. So,
hn = 2g − |fn − f | are positive measurable
R functions. ThusR we can
apply Fatou’s lemma, to conclude, X lim inf hn dµ ≤ lim inf R X hn dµ.
RClearly limRinf hn = lim hn = 2g. On the other hand, X hn dµ =
X
2gdµ − X |fn − f |dµ and thus,
Z Z Z
lim inf hn dµ = 2gdµ + lim inf(− |fn − f |dµ)
X X X
Z Z
= 2gdµ − lim sup |fn − f |dµ
X X
So, we get,
Z Z Z
2gdµ ≤ 2gdµ − lim sup |fn − f |dµ.
X X X
R R
Since X 2gdµ < ∞, we can cancel theseR to get, lim sup X
|fn −Rf |dµ ≤
0. EasyRto see that this implies lim X |fn − f |dµ = 0. Since | RX fn −
f dµ| ≤ X |fn − f |dµ from
R the previous
R result, we get that lim | X fn −
f dµ| = 0 and thus lim X fn dµ = X f dµ. 

5. Sets of measure zero

If (X, F, µ) is a measure space, any A ∈ F with µ(A) = 0 is naturally
called a set of measure zero. If B ⊂ A, we would like to say that
µ(B) = 0, but B may not be in F. So, we want to enlarge F to G
so that, if A ∈ F has measure zero, and B ⊂ A, then B ∈ G and of
 9

course F ⊂ G. This may not be a σ-algebra, but of course we can

replace G by the smallest σ-algebra containing G. Then it is not clear
that we can extend µ to this bigger σ-algebra, since we do not have
an explicit enough description of it. All this can be achieved by the
following construction, which can be easily checked to be the same as
above.
Consider G to be the collection of all subsets E ⊂ X such that there
exists A, B ∈ F with A ⊂ E ⊂ B and µ(B − A) = 0. Then using
simple set theory, one can easily check that G is a σ-algebra containing
F and if we define µ(E) = µ(A) = µ(B), it is well defined. We usually
call such a G (or more precisely (X, G, µ)) a complete measure space.
We will often say ‘something happens’ almost everywhere, abbrevi-
ated to a. e. to mean that it happens outside a set of measure zero.
For example, we may write f = g a. e. for two functions f, g on X
when they coincide outside some unspecified set of measure zero. It is
also common to use the expression for almost all x ∈ X to mean the
same.
Remark 4. Here is an important remark. Let (X, F, µ) be a complete
measure space. Let f be a measurable function defined a. e. This
means, f is defined on X − E where E is a set of measure zero and
measurable on X − E (with respect to the restricted σ-algebra). Then
extending f to all of X by assigning arbitrary values at points of E,
one easily checks that this function is measurable on all of X. So, one
can be loose about the definition of f on a set of measure zero.
Many of the earlier results we proved for functions on all of X can
thus be extended to functions defined a. e.
To illustrate, here is a corollary to Theorems 7 and 12
Theorem 13. Suppose fn : X → R (orP C) beRa sequence of measurable
∞
functions defined
P∞ a. e. on X such that n=1 X |fn |dµ < ∞. Then the
series f (x)P= Rn=1 fn (x) converges a. e. to a function f ∈ L1 (µ) and
f dµ = ∞
R
X n=1 X fn dµ.
Proof. Let En be the set where fnPis defined. Then letting E = ∩En , we
see that µ(E c ) = 0. Let φ(x) =R ∞ n=1 |fn (x)|.
P∞Then R φ(x) : X → [0, ∞]
is a measurable function and RX φdµ = n=1 X |fn |dµ < ∞. Let
−1
R
A = φ (∞). Then X φdµ ≥ A φdµ = ∞ · µ(A). This implies that
c
µ(A) = 0. So, outsideP the measure zero set E ∪ A, φ is a function to
R and thus f (x) = fn (x) converges (absolutely) on E ∩ Ac . Further
|f (x)| ≤ φ(x) on this set and thus by Theorem 12, we see that f ∈
L1 (µ) and the rest is clear too by Theorem 12. 
Below are some similar results, which are easy to prove.
10

Theorem
R 14. (1) If f : X → [0, ∞] is measurable, A ∈ F and
A
f dµ = 0, then Rf = 0 a. e. on A.
1
(2) If f ∈ L (µ) and A f dµ = 0 for every A ∈ F, then f = 0 a. e.
on X. R R
(3) If f ∈ L1 (µ) and | X f dµ| = X |f |dµ, then there is a constant
a such that af = |f | a. e. on X.
Theorem 15. Suppose µ(X) < ∞ and f ∈ L1 (µ) Rand let T ⊂ R (or
1
C) be a closed subset. If the averages AE (f ) = µ(E) E
f dµ lie in T for
every E ∈ F with µ(E) > 0, then f (x) ∈ T for almost all x ∈ X.
Proof. We can cover R − T by countably many open intervals of the
form I = (a − r, a + r) ⊂ R − T where a 6∈ T and r > 0. So, suffices to
prove that E = f −1 (I) has measure zero.
If µ(E) > 0, then,
Z  Z
1   1
|AE (f ) − a| = (f − a)dµ ≤ |f − a|dµ < r.
µ(E)  E µ(E) E
But AE (f ) ∈ T and thus |AE (f ) − a| ≥ r. This is a contradiction.

Theorem 16. Let An ∈ F such that ∞
P
n=1 µ(An ) < ∞. Then almost
all x lie in at most finitely many An s.
P∞
Proof.
R Consider Pφ(x)
R = n=1 χAn . Then φ : X → [0, ∞] is P
measurable
and X φdµ = χ dµ by Theorem 7. The latter is just µ(An ) <
X An
∞, by assumption. Thus as before, φ−1 (∞) has measure zero. If an
x belonged to infinitely many An ’s, then φ(x) = ∞. This finishes the
proof. 

6. Riesz Representation Theorem

We start with some notation. We will always use K for a compact
set in a topological space X and V for an open set. We also write
Cc (X) to denote continuous functions on X with compact support-
that is, functions vanishing outside a compact set. If K ⊂ V ⊂ X, and
if f ∈ Cc (X) with f = 1 on K and f = 0 outside V , with 0 ≤ f ≤ 1,
we write K ≺ f ≺ V . The notation K ≺ f would mean f = 1 on
K, where f ∈ Cc (X) and similarly f ≺ V would mean f = 0 outside
V with f ∈ Cc (X). The theorems in this section are valid for any
locally compact Hausdorff topological space, but we will work with Rk
(or subsets of it) , which is one such and we will denote it by X.
We start with a slight variation of Tietze extension theorem.
 11

Theorem 17. Let K ⊂ V ⊂ X. Then there exists an f ∈ Cc (X) such

that K ≺ f ≺ V .
Proof. Notice that we may replace V by any open set containing K and
contained in V . Since K is compact, it is bounded and hence contained
in a ball of radius R with center the origin and R > 0. Thus, we may
replace V by V ∩ B(0, R) and assume V is bounded. In particular, if
K ≺ f ≺ V , with f continuous, it is in Cc (X), since f = 0 outside the
closed and bounded (and hence compact) set, the closure of V .
If we take the function which is 1 on K and zero on X − V , then it
is continuous on the closed set K ∪ (X − V ) and then by Tietze, we
can extend this to a continuous function on X with |f | ≤ 1. Clearly,
we may replace f with |f | and so we are done. 
Remark 5. Some of you might have guessed that this could be done
much more easily, since we are working on a metric space with the
usual metric, which let me call d. Letting K 0 = X − V above, we can
0)
explicitly describe such a function by taking f (x) = d(x,Kd(x,K
0 )+d(x,K) .

Theorem 18 (Partition of Unity). Let K be a compact set contained

in ∪ni=1 Vi where
P Vi s are open. Then there exists fi ∈ Cc (X) such that
fi ≺ Vi and fi (x) = 1 for all x ∈ K.
Proof. For any point x ∈ K, we can find an open neighborhood Wx
such that the closure of Wx is compact and this compact neighborhood
is contained in Vi for some i. Since these cover K, by compactness, we
can find Wx1 , . . . , Wxm which cover K. Collecting the ones contained
in Vi , thus we get Hi ⊂ Vi , Hi compact and K ⊂ ∪Hi .
By the previous result, we can find gi such that Hi ≺ gi ≺ Vi . Let
hi = 1 − gi . Notice that hi ∈ Cc (X) and 0 ≤ hi ≤ 1. Now, consider fi
defined as,
f1 = 1 − h1
f2 = h1 − h1 h2
f3 = h1 h2 − h1 h2 h3
......
fn = h1 · · · hn−1 − h1 · · · hn
Clearly, all the fi ∈ Cc (X) and 0 ≤ fi ≤ 1, since each hi is. If x 6∈ Vi ,
then hi (x) = 1 and thus,
fi (x) = h1 · · · hi−1 (x) − h1 · · · hi (x) = 0.
So, fi ≺ Vi . If x ∈ K, then xP∈ Hi for some i and then hi (x) = 0. But,
P
fi = 1 − h1 · · · hn and so fi (x) = 1. 
12

Theorem 19 (Riesz Representation Theorem). Let L : Cc (X) → R

be linear map such that L(f ) ≥ 0 if f ≥ 0. (Usually referred to as a
positive linear functional.) Then there exists a σ-algebra F, containing
all open sets and a unique positive measure µ such that for any f ∈
Cc (X),
Z
L(f ) = f dµ.
X
Further, we have,
(1) For any compact set K, µ(K) < ∞.
(2) If V is any open set, then µ(V ) = sup{µ(K)|K ⊂ V }, K com-
pact.
(3) If E ∈ F, then µ(E) = inf{µ(V )|E ⊂ V }, V open.
(4) (X, F, µ) is a complete measure space.
We will not a give a complete proof, though it is not difficult, it is a
bit tedious.
Proof. First, we show that µ is unique. If (F1 , µ1 ), (F2 , µ2 ) are two
such, we will show that they agree on F1 ∩ F2 . So, the measures are
essentially the same.
It is clear that we only need to show that µ1 (K) = µ2 (K) for any
compact set. Given  > 0, by property 3) above, there exists an open
set V containing K such that µ1 (K) ≤ µ1 (V ) ≤ µ1 (K) + . Choose f
such that K ≺ f ≺ V . Then we have,
Z Z
µ2 (K) = χK dµ2 ≤ f dµ2 = L(f )
X X
Z Z
= f dµ1 ≤ χV dµ1 = µ1 (V ) ≤ µ1 (K) + 
X X

Since  was arbitrary, we get µ2 (K) ≤ µ1 (K) and reversing the roles of
µ1 , µ2 , we get equality.
Notice that since µ(K) = µ2 (K) ≤ L(f ) < ∞, we have Property 1)
of the theorem, if we have the equation connecting L and the integral.
Next, we define µ : P(X) → [0, ∞] as follows. For any open set V ,
define
µ(V ) = sup{L(f )|f ≺ V }.
It is clear that if U ⊂ V are open, then µ(U ) ≤ µ(V ). Next define for
any set E,
µ(E) = inf{µ(V )|E ⊂ V, V open}.
Notice that this definition agrees with the previous definition for open
sets. It looks as though we have defined µ on all subsets of X. But,
 13

we will construct a suitable σ-algebra, only on which all the properties

will be checked-mainly the countable additivity.
For this, we first look at F 0 consisting of all subsets of E ⊂ X such
that µ(E) < ∞ and
(1) µ(E) = sup{µ(K)|K ⊂ E, Kcompact}.
(Those of you who have seen Lebesgue measure using outer and inner
measure would see where one is going.)
Finally, define F to be the set of all E such that E ∩ K ∈ F 0 for any
compact set K.
First some remarks. Clearly µ(A) ≤ µ(B) if A ⊂ B. If µ(E) = 0,
then E ∈ F 0 , E ∈ F. So, we see that property 2) and property 4) both
hold for our µ.
Next we write down steps which will finish the proof.
(1) µ is countably sub-additive.P That is, if E1 , E2 , . . . are arbitrary
subsets of X, µ(∪En ) ≤ µ(En ).
(2) F 0 contains every compact set.
(3) F 0 contains every open set with finite measure. Further, they
satisfy equation 1 above.
(4) µ is countably additive on F 0 .
(5) If E ∈ F 0 and  > 0, there exists a compact K and open V such
that K ⊂ E ⊂ V with µ(V − K) < .
(6) If E1 , E2 ∈ F 0 , then so do E1 − E2 , E1 ∪ E2 , E1 ∩ E2 .
(7) F is a σ-algebra containing all open sets.
(8) F 0 consists of precisely those elements of F with finite measure.
(9) µ is a measure on F. R
(10) For every f ∈ Cc (X), L(f ) = X f dµ.


7. Lp -spaces
Definition 8. A function φ : (a, b) → R is called convex if for any
a < x, y < b and any t ∈ [0, 1], φ((1 − t)x + ty) ≤ (1 − t)φ(x) + tφ(y).
Lemma 1. A convex function is continuous.
Lemma 2. If φ is differentiable and φ0 is monotonically increasing,
then φ is convex.
Example 5. The above lemma implies φ(x) = ex is convex on R. Simi-
larly, φ(x) = xp is convex on (0, ∞) if p > 0.
Theorem 20 (Jensen’s inequality). Let (X, F, µ) be a measure space
with µ(X) = 1. Assume f ∈ L1 (X) with f (X) ⊂ (a, b). If φ is convex
14

on (a, b), one has,

Z  Z
φ f dµ ≤ (φ ◦ f )dµ.
X X

As an immediate corollary, we get that geometric mean is less than

the arithmetic mean.
Corollary 6. Let x1 , x2 , . . . , xn be n positive real numbers. Then,
1 1
(x1 xx · · · xn ) n ≤ (x1 + x2 + · · · + xn ).
n
Proof. Consider a space X = {p1 , . . . , pn } consisting of n elements
with measure µ({pi }) = n1 for all i. Consider the function f : X → R
given by f (pi ) = log xi and apply Jensen’s inequality with the convex
function exp. 
1 1
Definition 9. If p, q are positive real numbers with p
+ q
= 1, we say
that p, q are conjugate.
Notice that the above condition forces 1 ≤ p, q ≤ ∞ and p = 1 if
and only if q = ∞.
Theorem 21. Let p, q be conjugate with 1 < p, q < ∞ and let (X, µ)
be a measure space. Let f, g be positive measurable functions on X.
Then,
(1) Hölder’s inequality:
Z Z  p1 Z  1q
p q
f dµ ≤ f dµ g dµ
X X X
(2) Minkowski’s inequality:
Z  p1 Z  p1 Z  p1
p p p
(f + g) dµ ≤ f dµ + g dµ
X X X

Definition 10.R Let 0 < p < ∞ and let f be measurable on (X, µ). De-
fine ||f ||p = X f dµ and let Lp (X) consists of all measurable functions
p p

f with ||f ||p < ∞.

We define ||f ||∞ somewhat differently. Let f be any positive mea-
surable function and let S = {a ∈ R|f −1 ((a, ∞])has measure zero}. If
S = ∅, define ||f ||∞ = ∞ and otherwise define ||f ||∞ = inf S. Notice
that S is bounded below by zero (unless µ(X) = 0, which is a trivial
case and we will not usually consider it). Define L∞ (X) to be the set
of all measurable functions f such that g = |f | has ||g||∞ < ∞. The
previous theorem (Hölder’s and Minkowski’s inequalities) immediately
give the following.
 15

Theorem 22. Let p, q be conjugates with 1 ≤ p, q ≤ ∞.

(1) If f ∈ Lp (X), g ∈ Lq (X) then f g ∈ L1 (X) and,
||f g||1 ≤ ||f ||p ||g||q .
p
(2) If f, g ∈ L (X), then,
||f + g||p ≤ ||f ||p + ||g||p .
One immediately sees that Lp (X) is a vector space and || · || is a
norm. In particular, for f, g ∈ Lp (X), we can define a metric by
d(f, g) = ||f − g||p , the only difficulty is that it is possible d(f, g) = 0
even if f 6= g. But it is clear that d(f, g) = 0 if and only if f = g a. e.
and thus identifying this we still get a new vector space with actually
a metric on it. One usually sweeps this under the carpet, since it is
easier to think of elements of Lp (X) as functions and not equivalence
class of functions, and the context will make clear which of these we
are dealing with.
The fundamental theorem about these spaces is,
Theorem 23. Lp (X) is a complete metric space for 1 ≤ p ≤ ∞.
We finally state some density properties. Notice that if s is a simple
measurable functions with µ{x|s(x) 6= 0} < ∞, then s ∈ Lp for all
1 ≤ p < ∞.
Theorem 24. Let S be the set of all measurable functions satisfying
the above property. Then S is dense in Lp (X) for 1 ≤ p < ∞.
Next, we consider the case X to be an open or closed subset of Rn
(or more generally a locally compact Hausdorff space, so that Riesz
representation theorem is applicable) and the Lebesgue measure on it.
Then, Cc (X) ⊂ Lp (X) for 1 ≤ p ≤ ∞.
Theorem 25. Cc (X) is dense in Lp (X) for 1 ≤ p < ∞.