Department of Biostatistics and Computational Biology
University of Rochester
September 14, 2009
Qiu, Lee BST 401
Outline
1 Measures
Qiu, Lee BST 401
Countable additivity of a set function Let µ(·) be an extended real-valued set function defined on a σ-field F , i.e., It takes sets A in F as input variables. It maps these sets into extended real numbers, e.g., real numbers plus ∞. It is said to be finitely additive if: ForSfinitecollection of disjoint sets N A1 , . . . , AN in F , µ P n An = n µ(An ). Countably additive (σ-additive) if: For countably infinite collection of disjoint sets A1 , . . . , An , . . . in F , ∞ ! [ X µ An = µ(An ). (1) n n
Qiu, Lee BST 401
Countable additivity of a set function Let µ(·) be an extended real-valued set function defined on a σ-field F , i.e., It takes sets A in F as input variables. It maps these sets into extended real numbers, e.g., real numbers plus ∞. It is said to be finitely additive if: ForSfinitecollection of disjoint sets N A1 , . . . , AN in F , µ P n An = n µ(An ). Countably additive (σ-additive) if: For countably infinite collection of disjoint sets A1 , . . . , An , . . . in F , ∞ ! [ X µ An = µ(An ). (1) n n
Qiu, Lee BST 401
Countable additivity of a set function Let µ(·) be an extended real-valued set function defined on a σ-field F , i.e., It takes sets A in F as input variables. It maps these sets into extended real numbers, e.g., real numbers plus ∞. It is said to be finitely additive if: ForSfinitecollection of disjoint sets N A1 , . . . , AN in F , µ P n An = n µ(An ). Countably additive (σ-additive) if: For countably infinite collection of disjoint sets A1 , . . . , An , . . . in F , ∞ ! [ X µ An = µ(An ). (1) n n
Qiu, Lee BST 401
Countable additivity of a set function Let µ(·) be an extended real-valued set function defined on a σ-field F , i.e., It takes sets A in F as input variables. It maps these sets into extended real numbers, e.g., real numbers plus ∞. It is said to be finitely additive if: ForSfinitecollection of disjoint sets N A1 , . . . , AN in F , µ P n An = n µ(An ). Countably additive (σ-additive) if: For countably infinite collection of disjoint sets A1 , . . . , An , . . . in F , ∞ ! [ X µ An = µ(An ). (1) n n
Qiu, Lee BST 401
Basic definition
A measure on a σ-field F is a function µ(A) such that
It is non-negative: µ(A) ≥ 0. Measure of an empty set is always zero: µ(φ) = 0. (Remark: The reverse is not true.) It is countably additive. Exercise: finite/countable additivity are about commuting union and addition of disjoint sets. What happens if A1 , A2 , . . . are not disjoint?
Qiu, Lee BST 401
Basic definition
A measure on a σ-field F is a function µ(A) such that
It is non-negative: µ(A) ≥ 0. Measure of an empty set is always zero: µ(φ) = 0. (Remark: The reverse is not true.) It is countably additive. Exercise: finite/countable additivity are about commuting union and addition of disjoint sets. What happens if A1 , A2 , . . . are not disjoint?
Qiu, Lee BST 401
Basic definition
A measure on a σ-field F is a function µ(A) such that
It is non-negative: µ(A) ≥ 0. Measure of an empty set is always zero: µ(φ) = 0. (Remark: The reverse is not true.) It is countably additive. Exercise: finite/countable additivity are about commuting union and addition of disjoint sets. What happens if A1 , A2 , . . . are not disjoint?
Qiu, Lee BST 401
Finite/Probability Measure
If µ(Ω) is finite, µ is called a finite measure.
Let µ2 (A) = Cµ1 (A). µ1 , µ2 share a lot of mathematical properties. As a special case, for every finite measure µ, we can 1 definite µ∗ (A) = µ(Ω) µ(A) so that µ∗ (Ω) = 1. If µ(Ω) = 1, µ is called a probability measure. In other words, essentially every finite measure is equivalent to a probability measure. By convention, we define a measure/probability space to be a triple (Ω, F , µ). Where Ω is the whole space, F is the σ-field with Ω as its whole space, and µ is a measure defined on F .
Qiu, Lee BST 401
Finite/Probability Measure
If µ(Ω) is finite, µ is called a finite measure.
Let µ2 (A) = Cµ1 (A). µ1 , µ2 share a lot of mathematical properties. As a special case, for every finite measure µ, we can 1 definite µ∗ (A) = µ(Ω) µ(A) so that µ∗ (Ω) = 1. If µ(Ω) = 1, µ is called a probability measure. In other words, essentially every finite measure is equivalent to a probability measure. By convention, we define a measure/probability space to be a triple (Ω, F , µ). Where Ω is the whole space, F is the σ-field with Ω as its whole space, and µ is a measure defined on F .
Qiu, Lee BST 401
Finite/Probability Measure
If µ(Ω) is finite, µ is called a finite measure.
Let µ2 (A) = Cµ1 (A). µ1 , µ2 share a lot of mathematical properties. As a special case, for every finite measure µ, we can 1 definite µ∗ (A) = µ(Ω) µ(A) so that µ∗ (Ω) = 1. If µ(Ω) = 1, µ is called a probability measure. In other words, essentially every finite measure is equivalent to a probability measure. By convention, we define a measure/probability space to be a triple (Ω, F , µ). Where Ω is the whole space, F is the σ-field with Ω as its whole space, and µ is a measure defined on F .
Qiu, Lee BST 401
Finite/Probability Measure
If µ(Ω) is finite, µ is called a finite measure.
Let µ2 (A) = Cµ1 (A). µ1 , µ2 share a lot of mathematical properties. As a special case, for every finite measure µ, we can 1 definite µ∗ (A) = µ(Ω) µ(A) so that µ∗ (Ω) = 1. If µ(Ω) = 1, µ is called a probability measure. In other words, essentially every finite measure is equivalent to a probability measure. By convention, we define a measure/probability space to be a triple (Ω, F , µ). Where Ω is the whole space, F is the σ-field with Ω as its whole space, and µ is a measure defined on F .
Qiu, Lee BST 401
Finite/Probability Measure
If µ(Ω) is finite, µ is called a finite measure.
Let µ2 (A) = Cµ1 (A). µ1 , µ2 share a lot of mathematical properties. As a special case, for every finite measure µ, we can 1 definite µ∗ (A) = µ(Ω) µ(A) so that µ∗ (Ω) = 1. If µ(Ω) = 1, µ is called a probability measure. In other words, essentially every finite measure is equivalent to a probability measure. By convention, we define a measure/probability space to be a triple (Ω, F , µ). Where Ω is the whole space, F is the σ-field with Ω as its whole space, and µ is a measure defined on F .
Qiu, Lee BST 401
About Ω
The Ω term of a probability space is a set of all outcomes,
and are not necessarily numbers. For example, we sometimes denote the set of two possible outcomes of a Bernoulli distribution as Ω = {H, T }, represents Head and Tail. In fact you can have Ω defined as the set of humans such as“all students who are taking BST401”, a σ-algebra containing all subsets of you, and define a probability to quantify the probability that one of you guys in such a subset will one day win the Nobel prize.
Qiu, Lee BST 401
About Ω
The Ω term of a probability space is a set of all outcomes,
and are not necessarily numbers. For example, we sometimes denote the set of two possible outcomes of a Bernoulli distribution as Ω = {H, T }, represents Head and Tail. In fact you can have Ω defined as the set of humans such as“all students who are taking BST401”, a σ-algebra containing all subsets of you, and define a probability to quantify the probability that one of you guys in such a subset will one day win the Nobel prize.
Qiu, Lee BST 401
About Ω
The Ω term of a probability space is a set of all outcomes,
and are not necessarily numbers. For example, we sometimes denote the set of two possible outcomes of a Bernoulli distribution as Ω = {H, T }, represents Head and Tail. In fact you can have Ω defined as the set of humans such as“all students who are taking BST401”, a σ-algebra containing all subsets of you, and define a probability to quantify the probability that one of you guys in such a subset will one day win the Nobel prize.
Qiu, Lee BST 401
Examples
General definition of a counting measure: for any Ω, µ is
well defined on the largest σ-field 2Ω . Counting measures on a finite set, Z, and Q. Formally prove that every counting measure is a valid measure. Discrete probabilities: K unique outcomes: , . . . , ωK }, p1 , . . . , pK , Kk=1 pk = 1, P Ω = {ω1P µ(A) = ωk ∈A pk Here K can be countably infinite. Bernoulli distribution. (fair coin/non-fair coin). Binomial distribution. Ω = {0, 1, . . . , N}, F = 2Ω pk = Nk pk (1 − p)N−k . + Poisson distribution. Ω = Z+ = {0, 1, 2, . . .}, F = 2Z , k pk = e−λ λk ! .
Qiu, Lee BST 401
Examples
General definition of a counting measure: for any Ω, µ is
well defined on the largest σ-field 2Ω . Counting measures on a finite set, Z, and Q. Formally prove that every counting measure is a valid measure. Discrete probabilities: K unique outcomes: , . . . , ωK }, p1 , . . . , pK , Kk=1 pk = 1, P Ω = {ω1P µ(A) = ωk ∈A pk Here K can be countably infinite. Bernoulli distribution. (fair coin/non-fair coin). Binomial distribution. Ω = {0, 1, . . . , N}, F = 2Ω pk = Nk pk (1 − p)N−k . + Poisson distribution. Ω = Z+ = {0, 1, 2, . . .}, F = 2Z , k pk = e−λ λk ! .
Qiu, Lee BST 401
Examples
General definition of a counting measure: for any Ω, µ is
well defined on the largest σ-field 2Ω . Counting measures on a finite set, Z, and Q. Formally prove that every counting measure is a valid measure. Discrete probabilities: K unique outcomes: , . . . , ωK }, p1 , . . . , pK , Kk=1 pk = 1, P Ω = {ω1P µ(A) = ωk ∈A pk Here K can be countably infinite. Bernoulli distribution. (fair coin/non-fair coin). Binomial distribution. Ω = {0, 1, . . . , N}, F = 2Ω pk = Nk pk (1 − p)N−k . + Poisson distribution. Ω = Z+ = {0, 1, 2, . . .}, F = 2Z , k pk = e−λ λk ! .
Qiu, Lee BST 401
Examples
General definition of a counting measure: for any Ω, µ is
well defined on the largest σ-field 2Ω . Counting measures on a finite set, Z, and Q. Formally prove that every counting measure is a valid measure. Discrete probabilities: K unique outcomes: , . . . , ωK }, p1 , . . . , pK , Kk=1 pk = 1, P Ω = {ω1P µ(A) = ωk ∈A pk Here K can be countably infinite. Bernoulli distribution. (fair coin/non-fair coin). Binomial distribution. Ω = {0, 1, . . . , N}, F = 2Ω pk = Nk pk (1 − p)N−k . + Poisson distribution. Ω = Z+ = {0, 1, 2, . . .}, F = 2Z , k pk = e−λ λk ! .
Qiu, Lee BST 401
Examples
General definition of a counting measure: for any Ω, µ is
well defined on the largest σ-field 2Ω . Counting measures on a finite set, Z, and Q. Formally prove that every counting measure is a valid measure. Discrete probabilities: K unique outcomes: , . . . , ωK }, p1 , . . . , pK , Kk=1 pk = 1, P Ω = {ω1P µ(A) = ωk ∈A pk Here K can be countably infinite. Bernoulli distribution. (fair coin/non-fair coin). Binomial distribution. Ω = {0, 1, . . . , N}, F = 2Ω pk = Nk pk (1 − p)N−k . + Poisson distribution. Ω = Z+ = {0, 1, 2, . . .}, F = 2Z , k pk = e−λ λk ! .
Qiu, Lee BST 401
Examples
General definition of a counting measure: for any Ω, µ is
well defined on the largest σ-field 2Ω . Counting measures on a finite set, Z, and Q. Formally prove that every counting measure is a valid measure. Discrete probabilities: K unique outcomes: , . . . , ωK }, p1 , . . . , pK , Kk=1 pk = 1, P Ω = {ω1P µ(A) = ωk ∈A pk Here K can be countably infinite. Bernoulli distribution. (fair coin/non-fair coin). Binomial distribution. Ω = {0, 1, . . . , N}, F = 2Ω pk = Nk pk (1 − p)N−k . + Poisson distribution. Ω = Z+ = {0, 1, 2, . . .}, F = 2Z , k pk = e−λ λk ! .
Qiu, Lee BST 401
Examples
General definition of a counting measure: for any Ω, µ is
well defined on the largest σ-field 2Ω . Counting measures on a finite set, Z, and Q. Formally prove that every counting measure is a valid measure. Discrete probabilities: K unique outcomes: , . . . , ωK }, p1 , . . . , pK , Kk=1 pk = 1, P Ω = {ω1P µ(A) = ωk ∈A pk Here K can be countably infinite. Bernoulli distribution. (fair coin/non-fair coin). Binomial distribution. Ω = {0, 1, . . . , N}, F = 2Ω pk = Nk pk (1 − p)N−k . + Poisson distribution. Ω = Z+ = {0, 1, 2, . . .}, F = 2Z , k pk = e−λ λk ! .
Qiu, Lee BST 401
Basic Properties of a Measure
Monotonicity. If A ⊆ B, then µ(A) 6 µ(B).
P Subadditivity. If A ⊆ ∪i A Pi , then µ(A) 6 i µ(Ai ). As a special case, µ(∪i Ai ) 6 i µ(Ai ). Continuity from below. Ai ↑ A =⇒ µ(Ai ) ↑ µ(A). Continuity from above. Ai ↓ A =⇒ µ(Ai ) ↓ µ(A).
Qiu, Lee BST 401
Basic Properties of a Measure
Monotonicity. If A ⊆ B, then µ(A) 6 µ(B).
P Subadditivity. If A ⊆ ∪i A Pi , then µ(A) 6 i µ(Ai ). As a special case, µ(∪i Ai ) 6 i µ(Ai ). Continuity from below. Ai ↑ A =⇒ µ(Ai ) ↑ µ(A). Continuity from above. Ai ↓ A =⇒ µ(Ai ) ↓ µ(A).
Qiu, Lee BST 401
Basic Properties of a Measure
Monotonicity. If A ⊆ B, then µ(A) 6 µ(B).
P Subadditivity. If A ⊆ ∪i A Pi , then µ(A) 6 i µ(Ai ). As a special case, µ(∪i Ai ) 6 i µ(Ai ). Continuity from below. Ai ↑ A =⇒ µ(Ai ) ↑ µ(A). Continuity from above. Ai ↓ A =⇒ µ(Ai ) ↓ µ(A).
Qiu, Lee BST 401
Basic Properties of a Measure
Monotonicity. If A ⊆ B, then µ(A) 6 µ(B).
P Subadditivity. If A ⊆ ∪i A Pi , then µ(A) 6 i µ(Ai ). As a special case, µ(∪i Ai ) 6 i µ(Ai ). Continuity from below. Ai ↑ A =⇒ µ(Ai ) ↑ µ(A). Continuity from above. Ai ↓ A =⇒ µ(Ai ) ↓ µ(A).
Qiu, Lee BST 401
The Borel σ-fields
Borel measure on R: Ω = R, B, µ(A) = the length of A.
Where the Borel σ-field is defined to be the smallest σ-field containing all open intervals (a, b). (or all closed intervals [a, b], or all intervals of form (a, b].) First step: open sets. More details will be discussed later. Terminology: the smallest σ-field containing a collection of sets S is called a) the minimal σ-field over S ; b) the σ-field generated by S . B is a σ-field generated by the collection of open/closed intervals/sets. Not every subset of R is a Borel set! Checkout the Vitali set from Wikipedia.
Qiu, Lee BST 401
The Borel σ-fields
Borel measure on R: Ω = R, B, µ(A) = the length of A.
Where the Borel σ-field is defined to be the smallest σ-field containing all open intervals (a, b). (or all closed intervals [a, b], or all intervals of form (a, b].) First step: open sets. More details will be discussed later. Terminology: the smallest σ-field containing a collection of sets S is called a) the minimal σ-field over S ; b) the σ-field generated by S . B is a σ-field generated by the collection of open/closed intervals/sets. Not every subset of R is a Borel set! Checkout the Vitali set from Wikipedia.
Qiu, Lee BST 401
The Borel σ-fields
Borel measure on R: Ω = R, B, µ(A) = the length of A.
Where the Borel σ-field is defined to be the smallest σ-field containing all open intervals (a, b). (or all closed intervals [a, b], or all intervals of form (a, b].) First step: open sets. More details will be discussed later. Terminology: the smallest σ-field containing a collection of sets S is called a) the minimal σ-field over S ; b) the σ-field generated by S . B is a σ-field generated by the collection of open/closed intervals/sets. Not every subset of R is a Borel set! Checkout the Vitali set from Wikipedia.
Qiu, Lee BST 401
The Borel σ-fields
Borel measure on R: Ω = R, B, µ(A) = the length of A.
Where the Borel σ-field is defined to be the smallest σ-field containing all open intervals (a, b). (or all closed intervals [a, b], or all intervals of form (a, b].) First step: open sets. More details will be discussed later. Terminology: the smallest σ-field containing a collection of sets S is called a) the minimal σ-field over S ; b) the σ-field generated by S . B is a σ-field generated by the collection of open/closed intervals/sets. Not every subset of R is a Borel set! Checkout the Vitali set from Wikipedia.
Qiu, Lee BST 401
The Borel σ-fields
Borel measure on R: Ω = R, B, µ(A) = the length of A.
Where the Borel σ-field is defined to be the smallest σ-field containing all open intervals (a, b). (or all closed intervals [a, b], or all intervals of form (a, b].) First step: open sets. More details will be discussed later. Terminology: the smallest σ-field containing a collection of sets S is called a) the minimal σ-field over S ; b) the σ-field generated by S . B is a σ-field generated by the collection of open/closed intervals/sets. Not every subset of R is a Borel set! Checkout the Vitali set from Wikipedia.
Qiu, Lee BST 401
The Borel σ-fields
Borel measure on R: Ω = R, B, µ(A) = the length of A.
Where the Borel σ-field is defined to be the smallest σ-field containing all open intervals (a, b). (or all closed intervals [a, b], or all intervals of form (a, b].) First step: open sets. More details will be discussed later. Terminology: the smallest σ-field containing a collection of sets S is called a) the minimal σ-field over S ; b) the σ-field generated by S . B is a σ-field generated by the collection of open/closed intervals/sets. Not every subset of R is a Borel set! Checkout the Vitali set from Wikipedia.