Вы находитесь на странице: 1из 53

BST 401 Probability Theory

Xing Qiu Ha Youn Lee

Department of Biostatistics and Computational Biology


University of Rochester

September, 16

Qiu, Lee BST 401


Outline

1 Extension of Measures

Qiu, Lee BST 401


Motivations

To show that a measure defined on an algebra F0 (not


necessarily a σ-algebra) can be extended to a measure on
F = σ(F0 ), the σ-algebra generated by F0 .
Remark: F is generated by F0 is just another way of
saying F is the minimal σ-algebra containing F0 .
It serves as a bridge between finite step set operations
(pertain to an algebra) and infinite step set operations
(e.g., taking set limit) that are essential to σ-algebras.
As a consequence, we can use finite steps (valid within the
algebra) to approximate measures of members of the
σ-algebra.

Qiu, Lee BST 401


Motivations

To show that a measure defined on an algebra F0 (not


necessarily a σ-algebra) can be extended to a measure on
F = σ(F0 ), the σ-algebra generated by F0 .
Remark: F is generated by F0 is just another way of
saying F is the minimal σ-algebra containing F0 .
It serves as a bridge between finite step set operations
(pertain to an algebra) and infinite step set operations
(e.g., taking set limit) that are essential to σ-algebras.
As a consequence, we can use finite steps (valid within the
algebra) to approximate measures of members of the
σ-algebra.

Qiu, Lee BST 401


Motivations

To show that a measure defined on an algebra F0 (not


necessarily a σ-algebra) can be extended to a measure on
F = σ(F0 ), the σ-algebra generated by F0 .
Remark: F is generated by F0 is just another way of
saying F is the minimal σ-algebra containing F0 .
It serves as a bridge between finite step set operations
(pertain to an algebra) and infinite step set operations
(e.g., taking set limit) that are essential to σ-algebras.
As a consequence, we can use finite steps (valid within the
algebra) to approximate measures of members of the
σ-algebra.

Qiu, Lee BST 401


Motivations

To show that a measure defined on an algebra F0 (not


necessarily a σ-algebra) can be extended to a measure on
F = σ(F0 ), the σ-algebra generated by F0 .
Remark: F is generated by F0 is just another way of
saying F is the minimal σ-algebra containing F0 .
It serves as a bridge between finite step set operations
(pertain to an algebra) and infinite step set operations
(e.g., taking set limit) that are essential to σ-algebras.
As a consequence, we can use finite steps (valid within the
algebra) to approximate measures of members of the
σ-algebra.

Qiu, Lee BST 401


Other Similar Ideas

Classical example of rational number approximation to real


numbers.
Tolerance of small errors is of crucial importance.
Essentially we are replacing the “real thing” with an
imperfect finitely constructed replacement.
For rational/real number approximation, it means that we
replace an irrational number a by a rational number an ≈ a,
such that the “discrepancy” between the two measured by
f (x), dist(f (an ), f (a)) ,is less than some pre-specified .
For measure extension: we replace an arbitrary Borel set A
by An ≈ A, An ∈ F0 , which is constructed from intervals by
finitely many set operations, such that the “discrepancy”
between the two measured by µ(·), µ(An ∆A) ,is less than
some pre-specified .
Qiu, Lee BST 401
Other Similar Ideas

Classical example of rational number approximation to real


numbers.
Tolerance of small errors is of crucial importance.
Essentially we are replacing the “real thing” with an
imperfect finitely constructed replacement.
For rational/real number approximation, it means that we
replace an irrational number a by a rational number an ≈ a,
such that the “discrepancy” between the two measured by
f (x), dist(f (an ), f (a)) ,is less than some pre-specified .
For measure extension: we replace an arbitrary Borel set A
by An ≈ A, An ∈ F0 , which is constructed from intervals by
finitely many set operations, such that the “discrepancy”
between the two measured by µ(·), µ(An ∆A) ,is less than
some pre-specified .
Qiu, Lee BST 401
Other Similar Ideas

Classical example of rational number approximation to real


numbers.
Tolerance of small errors is of crucial importance.
Essentially we are replacing the “real thing” with an
imperfect finitely constructed replacement.
For rational/real number approximation, it means that we
replace an irrational number a by a rational number an ≈ a,
such that the “discrepancy” between the two measured by
f (x), dist(f (an ), f (a)) ,is less than some pre-specified .
For measure extension: we replace an arbitrary Borel set A
by An ≈ A, An ∈ F0 , which is constructed from intervals by
finitely many set operations, such that the “discrepancy”
between the two measured by µ(·), µ(An ∆A) ,is less than
some pre-specified .
Qiu, Lee BST 401
Other Similar Ideas

Classical example of rational number approximation to real


numbers.
Tolerance of small errors is of crucial importance.
Essentially we are replacing the “real thing” with an
imperfect finitely constructed replacement.
For rational/real number approximation, it means that we
replace an irrational number a by a rational number an ≈ a,
such that the “discrepancy” between the two measured by
f (x), dist(f (an ), f (a)) ,is less than some pre-specified .
For measure extension: we replace an arbitrary Borel set A
by An ≈ A, An ∈ F0 , which is constructed from intervals by
finitely many set operations, such that the “discrepancy”
between the two measured by µ(·), µ(An ∆A) ,is less than
some pre-specified .
Qiu, Lee BST 401
Discountinuous Examples

Without continuity there is no prediction. The Dirichlet


function 1Q (x).
(Exercise 1.4, Pg 438:) µ does not satisfy countable
additivity, so it is “discontinuous”, therefore the extension to
σ(A ) does not exist.
A technical term, σ-finiteness, is also important. It basically
says µ(Ω) can be infinite, but it can not be “too infinite”. We
must be able to break Ω into countably many subsets An ,
each with finite measure.
Most measures we will study in this course will be σ-finite.
Some infinite dimensional spaces have natural measures
that are not σ-finite, but they are not useful for probability
theory.

Qiu, Lee BST 401


Discountinuous Examples

Without continuity there is no prediction. The Dirichlet


function 1Q (x).
(Exercise 1.4, Pg 438:) µ does not satisfy countable
additivity, so it is “discontinuous”, therefore the extension to
σ(A ) does not exist.
A technical term, σ-finiteness, is also important. It basically
says µ(Ω) can be infinite, but it can not be “too infinite”. We
must be able to break Ω into countably many subsets An ,
each with finite measure.
Most measures we will study in this course will be σ-finite.
Some infinite dimensional spaces have natural measures
that are not σ-finite, but they are not useful for probability
theory.

Qiu, Lee BST 401


Discountinuous Examples

Without continuity there is no prediction. The Dirichlet


function 1Q (x).
(Exercise 1.4, Pg 438:) µ does not satisfy countable
additivity, so it is “discontinuous”, therefore the extension to
σ(A ) does not exist.
A technical term, σ-finiteness, is also important. It basically
says µ(Ω) can be infinite, but it can not be “too infinite”. We
must be able to break Ω into countably many subsets An ,
each with finite measure.
Most measures we will study in this course will be σ-finite.
Some infinite dimensional spaces have natural measures
that are not σ-finite, but they are not useful for probability
theory.

Qiu, Lee BST 401


Discountinuous Examples

Without continuity there is no prediction. The Dirichlet


function 1Q (x).
(Exercise 1.4, Pg 438:) µ does not satisfy countable
additivity, so it is “discontinuous”, therefore the extension to
σ(A ) does not exist.
A technical term, σ-finiteness, is also important. It basically
says µ(Ω) can be infinite, but it can not be “too infinite”. We
must be able to break Ω into countably many subsets An ,
each with finite measure.
Most measures we will study in this course will be σ-finite.
Some infinite dimensional spaces have natural measures
that are not σ-finite, but they are not useful for probability
theory.

Qiu, Lee BST 401


Approximations in the real world

Finite step approximation is more than “approximation”. It


is pretty much the only way we, as animals equipped with
finite step logic calculation ability, can deal with the real
world, which is infinitely complex.
Philosophical implications. Almost all engineering solutions
assumes continuity of the real world. Think: why you even
dare to drive a car? Predictability.
No perfect predictability in this world. In fact there is no
perfect measurement of any sort: time, length, force,
thickness of your car, smoothness of the road, etc.

Qiu, Lee BST 401


Approximations in the real world

Finite step approximation is more than “approximation”. It


is pretty much the only way we, as animals equipped with
finite step logic calculation ability, can deal with the real
world, which is infinitely complex.
Philosophical implications. Almost all engineering solutions
assumes continuity of the real world. Think: why you even
dare to drive a car? Predictability.
No perfect predictability in this world. In fact there is no
perfect measurement of any sort: time, length, force,
thickness of your car, smoothness of the road, etc.

Qiu, Lee BST 401


Approximations in the real world

Finite step approximation is more than “approximation”. It


is pretty much the only way we, as animals equipped with
finite step logic calculation ability, can deal with the real
world, which is infinitely complex.
Philosophical implications. Almost all engineering solutions
assumes continuity of the real world. Think: why you even
dare to drive a car? Predictability.
No perfect predictability in this world. In fact there is no
perfect measurement of any sort: time, length, force,
thickness of your car, smoothness of the road, etc.

Qiu, Lee BST 401


Measure extension (I)

Extending a probability measures µ0 on F0 , the algebra


generated by intervals, to σ(F0 ), the Borel σ-algebra.
First, extend F0 to G = F0 ∪ {limits of increasing sequence
of sets}
It’s pretty easy to check there exist a set function defined
on this larger collection:
For those A ∈ F0 , define µ1 (A) = µ0 (A).
For those B that can only be expressed as a limit of
increasing sequence of A1 , A2 , . . . in F0 , define
µ1 (B) = limn→∞ µ0 (An ). Why this is a good definition?
It satisfies:
Non-negativity,
µ1 (φ) = 0., µ1 (Ω) = 1,
Finite additivity, and a prototype of countable additivity.

Qiu, Lee BST 401


Measure extension (I)

Extending a probability measures µ0 on F0 , the algebra


generated by intervals, to σ(F0 ), the Borel σ-algebra.
First, extend F0 to G = F0 ∪ {limits of increasing sequence
of sets}
It’s pretty easy to check there exist a set function defined
on this larger collection:
For those A ∈ F0 , define µ1 (A) = µ0 (A).
For those B that can only be expressed as a limit of
increasing sequence of A1 , A2 , . . . in F0 , define
µ1 (B) = limn→∞ µ0 (An ). Why this is a good definition?
It satisfies:
Non-negativity,
µ1 (φ) = 0., µ1 (Ω) = 1,
Finite additivity, and a prototype of countable additivity.

Qiu, Lee BST 401


Measure extension (I)

Extending a probability measures µ0 on F0 , the algebra


generated by intervals, to σ(F0 ), the Borel σ-algebra.
First, extend F0 to G = F0 ∪ {limits of increasing sequence
of sets}
It’s pretty easy to check there exist a set function defined
on this larger collection:
For those A ∈ F0 , define µ1 (A) = µ0 (A).
For those B that can only be expressed as a limit of
increasing sequence of A1 , A2 , . . . in F0 , define
µ1 (B) = limn→∞ µ0 (An ). Why this is a good definition?
It satisfies:
Non-negativity,
µ1 (φ) = 0., µ1 (Ω) = 1,
Finite additivity, and a prototype of countable additivity.

Qiu, Lee BST 401


Measure extension (I)

Extending a probability measures µ0 on F0 , the algebra


generated by intervals, to σ(F0 ), the Borel σ-algebra.
First, extend F0 to G = F0 ∪ {limits of increasing sequence
of sets}
It’s pretty easy to check there exist a set function defined
on this larger collection:
For those A ∈ F0 , define µ1 (A) = µ0 (A).
For those B that can only be expressed as a limit of
increasing sequence of A1 , A2 , . . . in F0 , define
µ1 (B) = limn→∞ µ0 (An ). Why this is a good definition?
It satisfies:
Non-negativity,
µ1 (φ) = 0., µ1 (Ω) = 1,
Finite additivity, and a prototype of countable additivity.

Qiu, Lee BST 401


Measure extension (I)

Extending a probability measures µ0 on F0 , the algebra


generated by intervals, to σ(F0 ), the Borel σ-algebra.
First, extend F0 to G = F0 ∪ {limits of increasing sequence
of sets}
It’s pretty easy to check there exist a set function defined
on this larger collection:
For those A ∈ F0 , define µ1 (A) = µ0 (A).
For those B that can only be expressed as a limit of
increasing sequence of A1 , A2 , . . . in F0 , define
µ1 (B) = limn→∞ µ0 (An ). Why this is a good definition?
It satisfies:
Non-negativity,
µ1 (φ) = 0., µ1 (Ω) = 1,
Finite additivity, and a prototype of countable additivity.

Qiu, Lee BST 401


Measure extension (I)

Extending a probability measures µ0 on F0 , the algebra


generated by intervals, to σ(F0 ), the Borel σ-algebra.
First, extend F0 to G = F0 ∪ {limits of increasing sequence
of sets}
It’s pretty easy to check there exist a set function defined
on this larger collection:
For those A ∈ F0 , define µ1 (A) = µ0 (A).
For those B that can only be expressed as a limit of
increasing sequence of A1 , A2 , . . . in F0 , define
µ1 (B) = limn→∞ µ0 (An ). Why this is a good definition?
It satisfies:
Non-negativity,
µ1 (φ) = 0., µ1 (Ω) = 1,
Finite additivity, and a prototype of countable additivity.

Qiu, Lee BST 401


Measure extension (I)

Extending a probability measures µ0 on F0 , the algebra


generated by intervals, to σ(F0 ), the Borel σ-algebra.
First, extend F0 to G = F0 ∪ {limits of increasing sequence
of sets}
It’s pretty easy to check there exist a set function defined
on this larger collection:
For those A ∈ F0 , define µ1 (A) = µ0 (A).
For those B that can only be expressed as a limit of
increasing sequence of A1 , A2 , . . . in F0 , define
µ1 (B) = limn→∞ µ0 (An ). Why this is a good definition?
It satisfies:
Non-negativity,
µ1 (φ) = 0., µ1 (Ω) = 1,
Finite additivity, and a prototype of countable additivity.

Qiu, Lee BST 401


Measure extension (I)

Extending a probability measures µ0 on F0 , the algebra


generated by intervals, to σ(F0 ), the Borel σ-algebra.
First, extend F0 to G = F0 ∪ {limits of increasing sequence
of sets}
It’s pretty easy to check there exist a set function defined
on this larger collection:
For those A ∈ F0 , define µ1 (A) = µ0 (A).
For those B that can only be expressed as a limit of
increasing sequence of A1 , A2 , . . . in F0 , define
µ1 (B) = limn→∞ µ0 (An ). Why this is a good definition?
It satisfies:
Non-negativity,
µ1 (φ) = 0., µ1 (Ω) = 1,
Finite additivity, and a prototype of countable additivity.

Qiu, Lee BST 401


Measure extension(II)

You may think if we could just extend G to G¯, which include


all limits of interval-sequences then it would be a σ-algebra
and we are done. After all, that’s precisely how we extend
Q to R, right?
Unfortunately, this is not the case. G¯ is indeed closed
under σ-union/intersection and complement for F0 , but it is
not close for such operations acting on those non-F0
members!
The analogy for real numbers would be: is the limit of a
sequence of real numbers (which are the limits of a
sequence of rational numbers) again a real number?
 
lim lim qnm ∈ R? (1)
n→∞ m→∞

Qiu, Lee BST 401


Measure extension(II)

You may think if we could just extend G to G¯, which include


all limits of interval-sequences then it would be a σ-algebra
and we are done. After all, that’s precisely how we extend
Q to R, right?
Unfortunately, this is not the case. G¯ is indeed closed
under σ-union/intersection and complement for F0 , but it is
not close for such operations acting on those non-F0
members!
The analogy for real numbers would be: is the limit of a
sequence of real numbers (which are the limits of a
sequence of rational numbers) again a real number?
 
lim lim qnm ∈ R? (1)
n→∞ m→∞

Qiu, Lee BST 401


Measure extension(II)

You may think if we could just extend G to G¯, which include


all limits of interval-sequences then it would be a σ-algebra
and we are done. After all, that’s precisely how we extend
Q to R, right?
Unfortunately, this is not the case. G¯ is indeed closed
under σ-union/intersection and complement for F0 , but it is
not close for such operations acting on those non-F0
members!
The analogy for real numbers would be: is the limit of a
sequence of real numbers (which are the limits of a
sequence of rational numbers) again a real number?
 
lim lim qnm ∈ R? (1)
n→∞ m→∞

Qiu, Lee BST 401


Measure extension (III)

As for the Borel sets, we want to know whether


 
lim lim Anm ∈ G¯, Anm ∈ G¯. (2)
n→∞ m→∞

The answer for the real number analogy is (fortunately)


yes.
The answer for the Borel sets is no.
You can checkout the wikipedia entry “Borel set”, and pay
attention to the first section, “Generating the Borel
algebra”.

Qiu, Lee BST 401


Measure extension (III)

As for the Borel sets, we want to know whether


 
lim lim Anm ∈ G¯, Anm ∈ G¯. (2)
n→∞ m→∞

The answer for the real number analogy is (fortunately)


yes.
The answer for the Borel sets is no.
You can checkout the wikipedia entry “Borel set”, and pay
attention to the first section, “Generating the Borel
algebra”.

Qiu, Lee BST 401


Measure extension (III)

As for the Borel sets, we want to know whether


 
lim lim Anm ∈ G¯, Anm ∈ G¯. (2)
n→∞ m→∞

The answer for the real number analogy is (fortunately)


yes.
The answer for the Borel sets is no.
You can checkout the wikipedia entry “Borel set”, and pay
attention to the first section, “Generating the Borel
algebra”.

Qiu, Lee BST 401


Measure extension (III)

As for the Borel sets, we want to know whether


 
lim lim Anm ∈ G¯, Anm ∈ G¯. (2)
n→∞ m→∞

The answer for the real number analogy is (fortunately)


yes.
The answer for the Borel sets is no.
You can checkout the wikipedia entry “Borel set”, and pay
attention to the first section, “Generating the Borel
algebra”.

Qiu, Lee BST 401


Measure extension (IV)

Next, we extend µ1 to µ∗ , a set function that is almost a


measure defined on 2Ω , the maximum σ-algebra of Ω:
For G ∈ G , µ∗ (G) = µ1 (G).
For every A ∈ Ω,

µ∗ (A) = inf {µ1 (G) : G ∈ G , G ⊂ A} . (3)

Essentially, it tries to define measure of an arbitrary set A


by surrounding it with many larger sets, and let these sets
shrink to their infimum. So µ∗ is called an outer measure,
which defines upper bounds of an arbitrary set.

Qiu, Lee BST 401


Measure extension (IV)

Next, we extend µ1 to µ∗ , a set function that is almost a


measure defined on 2Ω , the maximum σ-algebra of Ω:
For G ∈ G , µ∗ (G) = µ1 (G).
For every A ∈ Ω,

µ∗ (A) = inf {µ1 (G) : G ∈ G , G ⊂ A} . (3)

Essentially, it tries to define measure of an arbitrary set A


by surrounding it with many larger sets, and let these sets
shrink to their infimum. So µ∗ is called an outer measure,
which defines upper bounds of an arbitrary set.

Qiu, Lee BST 401


Measure extension (IV)

Next, we extend µ1 to µ∗ , a set function that is almost a


measure defined on 2Ω , the maximum σ-algebra of Ω:
For G ∈ G , µ∗ (G) = µ1 (G).
For every A ∈ Ω,

µ∗ (A) = inf {µ1 (G) : G ∈ G , G ⊂ A} . (3)

Essentially, it tries to define measure of an arbitrary set A


by surrounding it with many larger sets, and let these sets
shrink to their infimum. So µ∗ is called an outer measure,
which defines upper bounds of an arbitrary set.

Qiu, Lee BST 401


Measure extension (IV)

Next, we extend µ1 to µ∗ , a set function that is almost a


measure defined on 2Ω , the maximum σ-algebra of Ω:
For G ∈ G , µ∗ (G) = µ1 (G).
For every A ∈ Ω,

µ∗ (A) = inf {µ1 (G) : G ∈ G , G ⊂ A} . (3)

Essentially, it tries to define measure of an arbitrary set A


by surrounding it with many larger sets, and let these sets
shrink to their infimum. So µ∗ is called an outer measure,
which defines upper bounds of an arbitrary set.

Qiu, Lee BST 401


Properties of an outer measure
Properties of µ∗ , also serves as general definitions of outer
measure:
Non-negativity, µ∗ (φ) = 0, µ∗ (Ω) = 1. (All inherited from
µ1 , which in turn inherited from µ0 .
A ⊂ B implies µ∗ (A) ≤ µ∗ (B). (monotonicity)
Countable sub-additivity

X
µ∗ (∪∞
n=1 An ) ≤ µ∗ (An ). (4)
n=1

And yes, there are indeed some F ∈ 2Ω , such that:

µ∗ (F ) < µ∗ (F1 ) + µ∗ (F2 ), F = F1 + F2 , F1 ∩ F2 = φ. (5)

Qiu, Lee BST 401


Properties of an outer measure
Properties of µ∗ , also serves as general definitions of outer
measure:
Non-negativity, µ∗ (φ) = 0, µ∗ (Ω) = 1. (All inherited from
µ1 , which in turn inherited from µ0 .
A ⊂ B implies µ∗ (A) ≤ µ∗ (B). (monotonicity)
Countable sub-additivity

X
µ∗ (∪∞
n=1 An ) ≤ µ∗ (An ). (4)
n=1

And yes, there are indeed some F ∈ 2Ω , such that:

µ∗ (F ) < µ∗ (F1 ) + µ∗ (F2 ), F = F1 + F2 , F1 ∩ F2 = φ. (5)

Qiu, Lee BST 401


Properties of an outer measure
Properties of µ∗ , also serves as general definitions of outer
measure:
Non-negativity, µ∗ (φ) = 0, µ∗ (Ω) = 1. (All inherited from
µ1 , which in turn inherited from µ0 .
A ⊂ B implies µ∗ (A) ≤ µ∗ (B). (monotonicity)
Countable sub-additivity

X
µ∗ (∪∞
n=1 An ) ≤ µ∗ (An ). (4)
n=1

And yes, there are indeed some F ∈ 2Ω , such that:

µ∗ (F ) < µ∗ (F1 ) + µ∗ (F2 ), F = F1 + F2 , F1 ∩ F2 = φ. (5)

Qiu, Lee BST 401


Properties of an outer measure
Properties of µ∗ , also serves as general definitions of outer
measure:
Non-negativity, µ∗ (φ) = 0, µ∗ (Ω) = 1. (All inherited from
µ1 , which in turn inherited from µ0 .
A ⊂ B implies µ∗ (A) ≤ µ∗ (B). (monotonicity)
Countable sub-additivity

X
µ∗ (∪∞
n=1 An ) ≤ µ∗ (An ). (4)
n=1

And yes, there are indeed some F ∈ 2Ω , such that:

µ∗ (F ) < µ∗ (F1 ) + µ∗ (F2 ), F = F1 + F2 , F1 ∩ F2 = φ. (5)

Qiu, Lee BST 401


The existence part of the extension theorem

The outer measure µ∗ is defined on the maximum


σ-algebra of Ω (which is good), but in general it does not
have countable additivity (which is bad).
A workaround: restrict µ∗ to a collection F ∗ , so a set
E ∈ F ∗ satisfy:

µ∗ (F ) = µ∗ (F ∩ E}) + µ∗ (F
| {z | ∩
c
{zE }), for all F ⊂ Ω. (6)
F1 F2

Equation (6) sometimes serve as the definition of


measurable sets. It implies countable additivity of µ∗ on
F ∗.

Qiu, Lee BST 401


The existence part of the extension theorem

The outer measure µ∗ is defined on the maximum


σ-algebra of Ω (which is good), but in general it does not
have countable additivity (which is bad).
A workaround: restrict µ∗ to a collection F ∗ , so a set
E ∈ F ∗ satisfy:

µ∗ (F ) = µ∗ (F ∩ E}) + µ∗ (F
| {z | ∩
c
{zE }), for all F ⊂ Ω. (6)
F1 F2

Equation (6) sometimes serve as the definition of


measurable sets. It implies countable additivity of µ∗ on
F ∗.

Qiu, Lee BST 401


The existence part of the extension theorem

The outer measure µ∗ is defined on the maximum


σ-algebra of Ω (which is good), but in general it does not
have countable additivity (which is bad).
A workaround: restrict µ∗ to a collection F ∗ , so a set
E ∈ F ∗ satisfy:

µ∗ (F ) = µ∗ (F ∩ E}) + µ∗ (F
| {z | ∩
c
{zE }), for all F ⊂ Ω. (6)
F1 F2

Equation (6) sometimes serve as the definition of


measurable sets. It implies countable additivity of µ∗ on
F ∗.

Qiu, Lee BST 401


(Cont’)

Lemma (2.4), pg 447 proves that F ∗ is indeed a σ-algebra.


So µ∗ restricted on F ∗ is a true probability measure!
If you prefer, you can further restrict µ∗ to F = σ(F0 ).
In fact, F ∗ and F are very similar: sets in F ∗ can always
be expressed as A ∪ N, where A ∈ F is the main part, N is
a null set (µ∗ (N) = 0).

Qiu, Lee BST 401


(Cont’)

Lemma (2.4), pg 447 proves that F ∗ is indeed a σ-algebra.


So µ∗ restricted on F ∗ is a true probability measure!
If you prefer, you can further restrict µ∗ to F = σ(F0 ).
In fact, F ∗ and F are very similar: sets in F ∗ can always
be expressed as A ∪ N, where A ∈ F is the main part, N is
a null set (µ∗ (N) = 0).

Qiu, Lee BST 401


(Cont’)

Lemma (2.4), pg 447 proves that F ∗ is indeed a σ-algebra.


So µ∗ restricted on F ∗ is a true probability measure!
If you prefer, you can further restrict µ∗ to F = σ(F0 ).
In fact, F ∗ and F are very similar: sets in F ∗ can always
be expressed as A ∪ N, where A ∈ F is the main part, N is
a null set (µ∗ (N) = 0).

Qiu, Lee BST 401


Carathéodory extension theorem

Theorem (1.1), pg. 438.


It generalized the above extension theorem to the case of
σ-finite measures.
It claims this extended measure is unique. In other words,
you can’t have two measures as extensions of µ0 , and they
disagree on some set A ∈ F .

Qiu, Lee BST 401


Carathéodory extension theorem

Theorem (1.1), pg. 438.


It generalized the above extension theorem to the case of
σ-finite measures.
It claims this extended measure is unique. In other words,
you can’t have two measures as extensions of µ0 , and they
disagree on some set A ∈ F .

Qiu, Lee BST 401


Carathéodory extension theorem

Theorem (1.1), pg. 438.


It generalized the above extension theorem to the case of
σ-finite measures.
It claims this extended measure is unique. In other words,
you can’t have two measures as extensions of µ0 , and they
disagree on some set A ∈ F .

Qiu, Lee BST 401


Measure approximation

Definition: set difference operation:


A∆B = (A ∩ B c ) ∪ (Ac ∩ B).
Assume µ is σ-finite on F0 .
For any A ∈ F and a given  > 0, there exists a set B ∈ F0
such that µ(A∆B) < .
This theorem is almost the -δ technique in probability
theory.

Qiu, Lee BST 401


Measure approximation

Definition: set difference operation:


A∆B = (A ∩ B c ) ∪ (Ac ∩ B).
Assume µ is σ-finite on F0 .
For any A ∈ F and a given  > 0, there exists a set B ∈ F0
such that µ(A∆B) < .
This theorem is almost the -δ technique in probability
theory.

Qiu, Lee BST 401


Measure approximation

Definition: set difference operation:


A∆B = (A ∩ B c ) ∪ (Ac ∩ B).
Assume µ is σ-finite on F0 .
For any A ∈ F and a given  > 0, there exists a set B ∈ F0
such that µ(A∆B) < .
This theorem is almost the -δ technique in probability
theory.

Qiu, Lee BST 401


Measure approximation

Definition: set difference operation:


A∆B = (A ∩ B c ) ∪ (Ac ∩ B).
Assume µ is σ-finite on F0 .
For any A ∈ F and a given  > 0, there exists a set B ∈ F0
such that µ(A∆B) < .
This theorem is almost the -δ technique in probability
theory.

Qiu, Lee BST 401

Вам также может понравиться