You are on page 1of 3

Measure theory class notes - 1 September 2010, class 7 1

The monotone class theorem


Recall that a monotone class over Ω is a collection of subsets of Ω closed under countable increasing
unions and countable decreasing intersections. M(F ) denotes the smallest monotone class which
includes F .

Theorem (Monotone class theorem). Let F be a field of subsets of Ω. Then M(F ) = σ(F ).

Proof. Clearly M(F ) ⊆ σ(F ), since σ(F ) is a monotone class. To show σ(F ) ⊆ M(F ), we
need to show that M(F ) is a σ-field.
We first show that M(F ) is a field. Since F ⊆ M(F ), ∅ ∈ M(F ). To show that M(F ) is
closed under complementation, let

M0 = {A ∈ M(F ) : Ω \ A ∈ M(F )}

F ⊆ M0 because F ⊆ M(F ) and F is a field. Let {An }∞ n=1 be an increasing sequence of sets in
M0 with union A. Since M(F ) is a monotone class, A ∈ M(F ). Since each An belongs to M0 ,
Ω \ An ∈ M(F ). \
Ω\A= (Ω \ An )
n∈N

and {Ω\An }∞n=1 is a decreasing sequence, so Ω\A ∈ M(F ). Since A ∈ M(F ) and Ω\A ∈ M(F ),
we have A ∈ M0 . M0 is closed under increasing unions, and a similar argument shows that it
is closed under decreasing intersections. M0 is a monotone class and F ⊆ M0 ⊆ M(F ), so
M0 = M(F ). So M(F ) is closed under complementation.
We show that M(F ) is closed under finite intersections in steps. First, fix A ∈ F (only in F , for
now!), and we will show that for all B ∈ M(F ), A ∩ B ∈ M(F ). As usual, let

M0 = {B ∈ M(F ) : A ∩ B ∈ M(F )}

Clearly F ⊆ M0 , since F ⊆ M(F ) and F is a field. Let {Bn }∞


n=1 be an increasing sequence of
sets in M0 whose union is B. Clearly B ∈ M(F ), since M(F ) is a monotone class. We have
that A ∩ Bn ∈ M(F ) for all n, and since
[
A∩B = (A ∩ Bn )
n∈N

and {A ∩ Bn }∞n=1 is an increasing sequence, we have A ∩ B ∈ M(F ). B ∈ M(F ) and A ∩ B ∈


M(F ), so B ∈ M0 . M0 is closed under increasing unions and a similar argument shows that it
is closed under decreasing intersections. So M0 is a monotone class. Since F ⊆ M0 ⊆ M(F ),
we have M0 = M(F ).
We have that for all A ∈ F and B ∈ M(F ), A ∩ B ∈ M(F ). To complete the argument, now
fix C ∈ M(F ). Let
M0 = {D ∈ M(F ) : C ∩ D ∈ M(F )}
By the previous step we know that F ⊆ M0 . By the same argument as before we have that M0
is a monotone class. So M0 = M(F ). Since C was an arbitrary set in M(F ), we have shown
that if C, D ∈ M(F ), then C ∩ D ∈ M(F ).
Measure theory class notes - 1 September 2010, class 7 2

So M(F ) is a field. To show that it is a σ-field, we need to show it is closed under countable
unions. Let {An }∞ ∞
n=1 be a sequence of sets in M(F ). Define {Bn }n=1 ,

n
[
Bn = Am
m=1

{Bn }∞
n=1 is an increasing sequence of sets and each of them is in M(F ) since M(F ) is a field.
S
Since M(F ) is a monotone class, their union, which is n∈N An belongs to M(F ).
M(F ) is a σ-field, and M(F ) = σ(F ).

An application of the monotone class theorem


Theorem. Let Ω be a set and A a σ-field on it. Suppose µ and ν are measures on (Ω, A ). Let
F be a field on Ω such that σ(F ) = A . Suppose µ and ν are σ-finite on F and agree on F .
Then µ = ν.1

Proof. Let {Ωn }n∈N be a disjoint collection of sets in F whose union is Ω and such that each µ(Ωn )
is finite. Since ν(Ωn ) = µ(Ωn ), ν(Ωn ) is also finite. We will show that for all A ∈ A and n ∈ N,

µ(A ∩ Ωn ) = ν(A ∩ Ωn )

Fix n and consider the set


{A ∈ A : µ(A ∩ Ωn ) = ν(A ∩ Ωn )}
This includes F (since A ∩ Ωn ∈ F ) and is a monotone class (since finite measures respect
increasing unions and decreasing intersections), and so includes M(F ) which equals σ(F ) = A .
For any A ∈ A , X X
µ(A) = µ(A ∩ Ωn ) = ν(A ∩ Ωn ) = ν(A)
n∈N n∈N

Directly dealing with sets in a σ-field is often hard - we may not have a nice description of all sets
in the σ-field. The above theorem tells us that to show that two measures are equal on the σ-field,
it suffices to check equality only for sets in a field which generates it (provided the measures are
σ-finite on the field), which is likely to be easier since we may have an explicit description of the
sets in the field.
Note that the measure being σ-finite on the field is important. It is possible to have two σ-finite
measures on (Ω, A ) which agree on F but are not equal: For example, consider (R, B), and let
D1 and D2 be countable dense subsets of R such that D1 ∩ D2 = ∅. Consider µ1 , µ2 : B → [0, ∞]R̄ ,

µ1 (A) = |A ∩ D1 |
µ2 (A) = |A ∩ D2 |

It is easy to see that µ1 and µ2 are measures. µ1 is σ-finite; we can take the cover by finite-measure
sets to be
{{x} : x ∈ D1 } ∪ {R \ D1 }
1
This follows from the uniqueness part of the Carathodory extension theorem, but we give a proof anyway.
Measure theory class notes - 1 September 2010, class 7 3

Similarly, µ2 is also σ-finite. We can take our field F to be all finite disjoint unions of intervals
of the type (a, b]. σ(F ) = B. For A ∈ F , we have µ1 (A) = µ2 (A) = ∞ if A is nonempty and
µ1 (∅) = µ2 (∅) = 0. So µ1 and µ2 agree on F but are not equal (since, for example, µ1 (D2 ) = 0
and µ2 (D2 ) = ∞).

Integration
To define the Riemann integral of a function f : [a, b] → [c, d], we divided [a, b] into small intervals,
and approximated the region under the graph of [a, b] on each of these small intervals by a rectangle.
This way we can approximate the area under the graph of f by the sum of the areas of these thin
rectangles.
Rb If this quantity tends to a number z as the partition of [a, b] becomes finer, we say that
a
f (x)dx = z. To be able to approximate the value of f on a small interval by a single value, we
would like that f does not vary too much on that interval. Riemann integration works well for
continuous functions (and also some other functions like piecewise continuous ones).
Lebesgue integration takes a different approach. Instead of dividing the doman, we divide the
range (say [c, d] as above) into intervals A1 , A2 , . . . An , where n is large, so that each interval is
small. We can pick a point ai in each Ai , and approximate the integral of f by
n
X
ai (length of f −1 (Ai ))
i=1

Note that unlike in the case of Riemann integration, here taking ai as an approximation for any
value in Ai is reasonable, since Ai is a small interval, regardless of what f is! So this approach is
likely to work better, provided we can assign a length to f −1 (Ai ). This is one of the motivations
for developing measure theory. We now define the class of functions for which the f −1 (Ai ) will
have length defined, in a more general setting.

Definition. Let Ω be a nonempty set and A a σ-field on it. f : Ω → R is said to be measurable


if for any interval I ⊆ R, f −1 (I) ∈ A .

It follows that for a measurable function f , for any Borel set B ⊆ R, f −1 (B) ∈ A (as usual, since
the collection of Borel sets for which the inverse image belongs to A has all intervals and is closed
under complements and countable unions).
When (Ω, A ) = (R, B), then every continuous function is measurable, since the collection of all
open intervals in R generates B too.

You might also like