Chapter 2: Probability Spaces And Random Variables Flashcards
Sample space
Ω
Set containing every possible outcome
Event
Collection of possible outcomes
Subset of Ω
Subset of sample space
Definition 2.1.1
F, set of subsets
Sigma field
Let F be a set of subsets of Ω. We say F is a sigma-field if:
1) empty and sample space are elements
2) complements are elements
3) unions are elements
P(Ω)
The power set F= P( Ω )
Sub-sigma-field
g is subset of F
g is a sub-sigma-field of F
Ie a more restrictive set of information than F has
Measurable subset
A subset A of Ω is measurable/ a measurable event if A is element of F
Ie A is F-measureable
Ie A is an event
Probability space
(Ω, F, P)
Sample space
Sigma field
Probability measure
Empty set
Test nothing happens
Examples of sigma fields for chosen experiments
Coin toss
H or T
Ω = {H,T}
F= { ∅, {H}, {T} , Ω}
Set of all possible subsets
In order
It’s a test we get nothing ( this has chance 0)
A test we get a Head
A test we get a tail
A test we get a head or tail
Examples of sigma fields for chosen experiments
Coin toss with 2 coins
Ω = { HH, HT, TH,TT}
How we F (set of subset) only contains events we are interested in..
Suppose that all we care about is whether we get two heads
So define F as set of subsets we care about
For example if interested in whether both coins are heads we have
F = { ∅, {H,H}, Ω{H,H}, Ω}
Ie complement of HH is Ω\HH “not HH”
And empty and sample space for sigma field
Test nothing (chance0)
Test HH
Test not HH
Test two coins
Probability measure
P is a function P:F to [0,1] st
1) P[Ω] =1
2) if A_1, A_2,… in F are pair-wise disjoint
(ie A_i intersection A_j = empty set for all I,j st I not equal to j)
Then
P[ Union from I=1 to infinity ] = sigma from I=1 to Infinity of P[A_i] *
*the probability of the Union is the probability that any one of the events happens
- uses the probability of A1 Union A_2 that are disjoint is the sum of the individual probabilities
P[A_1 Union A_2 ] = P[A_1] + P[A_2] - this last condition is sigma additivity
(We have sigma-additivity and * by def 2.1.1 of sigma field)
Probability space
Triple ( Ω, F, P)
Ω sample space
Where F is a sigma-field
And P is a probability measure
Examples of F:
* F= {∅, Ω} this is a no information sigma field as we have Ω the event that anything happens (probability 1) and ∅ the event that nothing happens ( always probability 0)
- we care about one event eg if outcome is in a subset A so we use sigma field:
F= {∅, A, Ω\A, Ω}
These F are sigma fields- we can check the conditions
Single coin toss probability measure
Ω ={H,T}
F= { ∅, {H}, {T}, Ω}
And define P[{H}] | = P [{T}] = 0.5 if it’s a fair coin toss ie probability of a set containing H = probability of the set containing T
And define P(Ω) =1 and P(∅) =0
We want to choose F to be smaller than P(Ω)
Lemma 2.1.5 intersections of sigma fields
Let I be any set. For any i in I let F_i be a sigma-field. Then F = intersections of i in I of F_i
Is a σ-field
Ie we contain all the information that is common to all
Proof: conditions
1) F_i is a sigma field so empty set is an element of F_i and so empty set is an element of intersections F_i
2) if A is an element of F = intersections of F_i then A is an element of F_i for each i. Since each F_i is a sigma-field then Ω\A is an element of F_i for each i. Hence Ω\A is an element of the intersections of F_i
3)
If A_j in F for all j then A_j is in F_i for all i and j. Since each F_i is a sigma-field, unión of A_j is an element of F_i for all i. Hence union of A_j is an element of the intersections of F_i
Corollary 2.1.6
Intersections of sigma-fields σ-fields
If F_1 and F_2 are sigma-fields then so is F_1 intersection F_2
Simple case of lemma
Def 2.1.7
Events in sigma fields
Let E_1, E_2,.. be subsets of Ω.
σ(E_1, E_2,..) is the smallest σ -fields containing E_1, E_2,…
(Any event in any of F_i)
Defined Big curly F
How do we construct a sigma field without writing it down
For a given Ω (sample space), finite or countable E_1, E_2,.. subset of Ω (events). ( we are interested in)
Let big_curly_F be the set of all sigma-fields that contain E_1,E_2,..
Enumerate
Big_curly_F ={ F_i: i is in I}
And by applying lemme 2.1.5 obtain a sigma-field F which contains all events we want
(Each of these F_i contains events we want And maybe some others)
This means F is the smallest σ-field that has E_1,E_2,.. as events
(Applying lemma 2.1.5)
(F is contained inside any σ-field F which has those events we want
And
It’s the smallest σ -field which contains all the E_i’s
As it’s the intersection of all that contain and thus big_curly_F is smaller than all of F_i ‘s
By the lemma
Ie it EXISTS
With Ω as any set and A a subset of Ω
{ empty set, A, Ω\A, Ω} is σ(A)
If F_1, F_2 ,.. are sigma fields then
Write sigma( F_1,F_2,..) for the smallest sigma-algebra wrt which all events are measurable
Properties of events in F
If A is an element of F then Ω\A element of F and since Ω =A ∪(Ω\A) we have 1=P[A] + P[Ω\A]
=P(Ω)=P(A ∪ (Ω\A))
If A,B in F and A subset of B then B=A Union (B\A)
Which gives
P[B] = P[B\A] + P[A]
as P[B\A] is bigger than or equal to 0
Implying
P[A] is less than or equal to P[B]
Lemma 2.1.8
Let A_1, A_2,.. in F where F is a sigma-field. Then intersection from i=1 to infinity of A_i in F
(Countable)
Proof/ we can write
*in general uncountable unions and intersections of measurable sets need not be measurable but lemma may not hold so that F isn’t too large for a Probability measure as harder to define probabilities the bigger F is
The empty set
Is an element of F the sigma-field and is the test that nothing happens
Set of all subsets of Ω
The set of all subsets of Ω is a sigma field
what is the smallest sigma field of unions?
σ(F_1,F_2,…)
Random variables
pre-image
X:Ω to R. For each outcome ω ∈ Ω
X(ω) is a property .
X^-1 (A) = {ω ∈ Ω : X(ω) ∈ A}
is a PRE-IMAGE ( not inverse) of A under X.
and used to find set of outcomes ω ∈ Ω mapping to some set A under X.
X-1(a,b) for interval (a,b)
EXAMPLE
Ω={1,2,3,4,5,6}
and F=P(Ω).
Property X(ω)= {0 if ω is odd {1 if ω is even
ω ∈ Ω
X-1({0}) = {1,3,5} X-1({1}) = {2,4,6} X-1({0,1}) = {1,2,3,4,5,6}
DEF 2.2.1:
Definition of a random variable/measurable function
Let G be a σ-field. A function X:Ω to R is G-measurable if for all subintervals I⊆R we have X-1(I) ∈ G
For a probability space (Ω , F,P) we say that X:Ω to R is a RANDOM VARIABLE if X if F-measurable
measurable notes:
- X is measurable (using σ-field G) I.e. x∈ mG
- P[X∈ A] = P[X-1(A)] (X-1(A) ∈ F)
*P[a less than X less than b]
= P[X-1(a,b)] = P[ω ∈ Ω ; X(ω) ∈ (a,b) ]
*P[X=a]= P[X-1(a)] = P[ω ∈ Ω ; X(ω)=a ]
EXAMPLE: toss a coin twice
Ω= {HH, HT, TH, TT} F= P(Ω) then every funct X:Ω to R is F-measurable
G= {∅, {H,H}, {HT,TT, TH}, Ω }
(this is info of “did we get 2 heads” with sigma-field that gives this)
if we are interested in #tails:
X: Ω to R given by X(ω) = #total tails occurred
Then X is NOT G-measurable
ie if all we knew was whether or not we had HH, we can’t work out exact #tails
We need to be able to deduce the info from the given info
We can take the preimage of an interval eg [0,1] we can find set is not an element of G ( X-1([0,1]) = {HH,HT,TH} is not an element of G) so X is not G measurable
information and sigma-fields:
σ-field: G chooses which info we care about
X is G-measurable ie X∈mG: X depends only on the information in G / information we care about
given an outcome ω of our experiment, but not knowing which ω ∈ Ω it was, as each event “G” in G represents a piece of info this info is whether or not ω ∈ G ( ie whether or not event “G” has occurred), If this info allows us to deduce the exact value of X(ω) and if this is the case for any ω ∈ Ω then X is G-measurable
G and “G”
in notes G is written curly eg similar to g (as is F)
writing G as “G” in flashcards
σ-field generated by random variables
X is RV
A σ-field is σ(X), containing information given by X
LEMMA 2.2.5 σ(X)
X is σ(X)-measurable
σ(X) is a σ-field, containing information given by X
σ(X)
to construct:
include ∅ and Ω
look at preimages of X
look at complements
look at unions
remember
operations on random variables produce random variables
Lemma 2.2.2:
If X is a discrete RV:
suppose we have (Ω, F, P) and a RV X:Ω to R and we want to check that X is measurable wrt some smaller σ-field
Let G be a σ-field on Ω. Let X:Ω to R and suppose X takes values {x_1,x_2,…} ( a countable set).
Then
X is measurable wrt G/ X∈ mG
⇔
for all j, {X=x_j} ∈ G
PROOF lemma 2.2.2:
countable set measurable wrt G
⇔
each element in G
..
example: take Ω={1,2,3,4,5,6}m rolling a dice
F= P(Ω)
consider
G_1 ={∅, {1,3,5}, {2,4,6}, Ω}
G_1 ={∅, {1,2,3}, {4,5,6}, Ω}
G_1 ={∅, {1,3,5}, {2,4,6}, Ω}
“test is Ω even or odd”
G_1 ={∅, {1,2,3}, {4,5,6}, Ω}
“test is Ω less than or equal to 3 or bigger than 2”
Here, G_1 contains the info of whether roll even or odd etc. We can check both are σ-fields
DEFINE
X_1(ω) =
{0 if ω is odd
{1 if ω is even
X_2(ω) =
{0 if ω is less than or equal to 3
{1 if ω is bigger than 3.
X_1 tests if X is even
X_2 tests is X is odd
we expect that X_1 is measurable wrt G_1 but not wrt G_2 etc
- X_1^-1(0) ={1,3,5}
- X_1^-1(1) ={2,4,6}
both are in G_1 but not in G_2 so
X_1 is measurable wrt G_1 but not wrt G_2
similarly * X_2^-1(0) ={1,2,3} * X_2^-1(1) ={4,5,6} both are in G_2 but not in G_1 so X_2 is measurable wrt G_2 but not wrt G_1
Example: consider generated σ-field and smallest contained events
G_3 = σ( {1,3}. {2}, {4}, {5}, {6}) has 32 elements but
the information given cant tell 1 from 3
X-1(0) = {1,3,5} = { {1}, {1,3}, {3}} ∪ {5}
as {1,3} is an element of G_3 and {5} is we also have
{ {1}, {1,3}, {3}} ∪ {5} is an element of G_3
X-1(1) = {2,4,6} = {2}∪ {4} ∪ {6}
each is an elemnt of G_3 and so union is an elemnt of G_3
by def 2.1.7
we thus have
X_1 is measurable wrt G_3 and X_2 is also
σ-field are associated to
σ-field are associated to each function X:Ω to R
DEF 2.2.4 σ-field generated by X
The σ-field generated by X, denoted σ(X) is
σ(X)= σ( X-1(I) : I is a subinterval of R)
the smallest σ-field that contains events
F=P(Ω)
Ω=(1,2,3,4,5,6}
X(ω) =
{1 if ω is odd
{2 if ω is even
X-1(1) = {1,3,5} X-1(2) = {2,4,6}
thus the smallest σ-field that contains events is
σ(X) = {∅, {1,3,5}, {2,4,6}, Ω}
ie we don’t need any more elements as all unions and complements are contained
the smallest σ-field that contains events
F=P(Ω)
Ω=(1,2,3,4,5,6}
Y(ω) =
{1 if ω=1
{2 if ω=2
{3 if ω=3
Y-1(1) = {1} Y-1(2) = {2} Y-1(3) = {3}
thus the smallest σ-field that contains events is
σ(Y) = {∅,
{1}, {2},{3}, {2,3,4,5,6}, {1,3,4,5,6}, {1,2,4,5,6},
{1,3}, {1,2}, {2,3} , {2,4,5,6}, {3,4,5,6}, {1,4,5,6},
{1,2,3},{4,5,6}
Ω}
taking unions and complements etc to include the 3 events, empty set, sample space and all their complements and unions
LEMMA 2.2.5: σ(X) measurable
Let X:Ω to R. Then X is σ(X) measurable
Let X:Ω to R. Then X is σ(X) measurable
PROOF:
…
Define σ(X_1, X_2,…)
σ(X_1, X_2,…) = σ(X_i(I)) : X_i is element of sequence and I is a subinterval of R
this is the σ-field containing the info given by X_1, X_2,…
2.2.4 gives us:
Considering a collection of RVs its better to consider a sub-σ-field G ⊆F
Let α∈R and let X, Y, and X_1,_X_2,… be G-measurable functions Ω to R
Then
α, αX, X+Y, XY, 1/X
are all G-measurable.
Further, if X_∞ given by
X_∞ (ω) = limit as n tends to infinity of X_n(ω)
exists for all ω,
then X_∞ is G-measurable
EXAMPLE: if X∈mG then
Then (X^2 + X)/ 2 ∈mG
Y= e^X then Y ∈mG
because Y(ω)= sum from n=0 to infinity X(ω)^n/(n!)
we know this limit exists as e^x is defined for all x in the reals consider the partial sums
Y_N (ω) = sum from n=0 to N of X(ω)^n /n! ∈mG by (1) and Y(ω) = limit as N to infinity Y_N(ω) exists so Y ∈mG
IF X ∈mG and g: R to R is any SENSIBLE FUNCTION then Y=g(X) ∈mG
ie all powers, trig functs, e^x limit, polynomial, monotone functions, all piecewise linear functs
RECALL FOR INDEPENDENT EVENTS
Events E_1, E_2
are independent if P( E_1 ∩ E_2) = P(E_1) P(E_2)
ie change that E_1 and E_2 happens is
chance that E_1 * chance that E_2
2.2.7 INDEPENDENCE
we define independence of events using sigma fields
Sub- σ-fields G_1, G_2 of F are indep if
(note notation F is curly F, G is curly G , “G” is G)
P(“G”_1 ∩ “G”_2) = P(“G”_1)P(“G”_2) for all “G_1” in G_1 and “G_2” in G_2
Events E_1 and E_2 are independent if
σ(E_1) and σ(E_2) are independent
Random vars X_1 and X_2 are indep if σ(X_1) and σ(X_2) are independent
2.3 INFINITE SAMPLE SPACE
for probability space
(Ω, F, P)
Recall if Ω= {x_1, x_2,…,x_n} is finite. We normally take F=P(Ω).
Since F contains every subset of Ω,
any σ-field on Ω is a sub-σ-field of F. We have seen how it is possible to construct other σ-fields
on Ω too.
In this case we can define a probability measure on Ω
We can define P by choosing some sequence (a_1, a_2,..,a_n) st a_i ∈ [0,1] and sum from i=1 to n of a_i = 1
and set P({x_i}) = a_i
This naturally extends to defining
P[A] for any subset A ⊆ Ω, by setting
more generally P(A) = sum from i=1 to n of P({x_i}) * (indicator function of x_i ∈ A)
If Ω is countable we have Ω = {x_1,x_2,..} can replace n with infinity. Infinite sequence transformed into infinite series which is bounded above by 1.
EXAMPLE:
toss a coin countably many times. Outcome is ω= ω_1ω_2…. = (ω_1, ω_2,…)
where each ω_i is in {H,T}
the set of ω is uncountable :
Ω= {H,T}^N
Define X_n(ω) = ω_n = result of the nth toss
σ(X_1)
σ(X_1, X_2)
Define X_n(ω) = ω_n = result of the nth toss X_n: Ω to {H,T} this isn't a subset of R but we can imagine as 0 or 1 etc
We want F = σ(X_1, X_2,..)
= σ(X^{-1}_i (I): i in N, I is a subinterval of R)
contains all info generated by coin tosses
NOTE
σ(X_1) = ( { ∅, {H…}, {T…} , Ω}
ie if first toss is H or T followed by anything else
σ(X_1) = ( { ∅, {H…}, {T…} , Ω}
σ(X_1, X_2) = σ( {HH..}, {TH…}, {HT..}, {TT…})
= {∅, Ω,
{HH…}, {TH..}, {HT..}, {TT…},
{H..}, {T…}, {H..}, {T..},
{HH..,
TT..},
{HT..,
TH..},
{HH..}^c, {TH…}^c, {HT..}^c, {TT..}^c }
ie considering first two tosses unions and complements
if 2 ω have same 1st and 2nd outcomes they fall into same subset of σ(X_1, X_2)
if a RV dep on anything more than 1st or 2nd outcomes will not be σ(X_1, X_2) measurable
HOW TO DEFINE P?
when using infinite Ω will be given probability measure P and properties
EXAMPLE HERE GIVEN: P: F to [0,1]
1) X_n are indep RVs
2) P(X_n = H) = P(X_n = T) = 0.5 for all n in N
from this we work with P without being constructed. Don’t need to know which subsets of Ω in F as 2.2.6 allows us
if we try to take F= P(Ω) there is no probability measure such that 1 and 2 are satisfied
2.3.2 Ω={H,T} sequence if fair independent coin tosses
P(X_i = T) = P(X_2 =H) = 0.5
P(X_1 = ω_1, X_2 = ω_2, X_3 = ω_3,….)
P[X_1 =H]
P[ eventually throw a head]
P[never throw a head ]
P(X_1 = ω_1, X_2 = ω_2, X_3 = ω_3,…) =
P(X_1 =ω_1)P(X_2= ω_2)P(X_3 = ω_3)……
= 0.5 0.5 0.5 *… = 0
e.g the probability that we never throw a head
So
P[X_1 =H] = P[ {ω_i} in Ω : ω_1 =H] = 0.5
P[ eventually throw a head] = P[for some n, X_n =H] = 1- P[for all n X_n =T]=1-0=1 event has probability 1 but not equal to whole sample space
P[never throw a head ] = P[for all n X_n =T]= 0.50.5…=0
DEF 2.3.2
ALMOST SURELY
if the event E has P(E) =1 we say E occurs ALMOST SURELY
ie coin will eventually throw a headie Y less than or equal to 1 almost surely
IFF P[Y less than or equal to 1] =1
DEFINE
proportions of heads and tails in the first n tossesX_1,… X_n are q_n ^H and q_n ^T
q_n ^H = (1/n) sum from i=1 to n of (indicator funct of X_i =H)
q_n ^T = (1/n) sum from i=1 to n of (indicator funct of X_i =T)
q_n ^T + q_n ^H = 1
the random vars indicator of {X_i=H} are iid with
E[1_{X_i=H}] =0.5
hence by thm 1.1.1 we have P[q_n ^T tends to 0.5 as n tends to infinity] = 1
by strong law of large #
& P[q_n ^H tends to 0.5 as n tends to infinity] = 1
event that half tosses H and half are T
is event E = {limit as n tends to infinity of q_n ^T = 0.5 and limit as n tends to infinity of q_n ^H = 0.5}
occurs almost surely
as P[q_n ^H tends to 0.5 , q_n ^T tends to 0.5 ] =1
EXPECTATION
if X is continuous RV E[X] = integral from -infinity to infinity of [xf_x(alpha).dx] (pdf)
if X is discrete RV E[X] = sum of x in R_x of [xP[X=x]]
(R_x is range of x)
WHEN X IS AN F-MEASURABLE FUNCTION FROM Ω to R, RVS MIGHT NOT BE DISCRETE OR CONTINUOUS .
we use Lebesgue integration to define E(X) in this case
E[X] defined:
E[X} is defined for all X se either:
1) X is bigger than or equal to 0, in which case its possible that E[X] = infinity
2) General X for which E[ |X|] less than infinity
* we haven’t got E[X] = - infinity
* we use the modulus to avoid “infinity take away infinity”
***if X is bigger than or equal to 0 and P[X= infinity] bigger than 0, this implies E[X] = infinity
(the chance of X being infinite will outweigh all of the finite possibilities and make E[X] Infinite)
PROPOSITION 2.4.1
Expectations for RVs X and Y
For random variables X, Y such that E[X] and E[Y] exist:
LINEARITY
if a,b in R then E[ aX +bY] = a E[X] + bE[y]
INDEPENDENCE if X, Y are indep then E[XY] = E[X]E[Y] ABSOLUTE VALUES |E[X]| is less thn or equal to E[|X|] MONOTONICITY if X is less than or equal to Y then E[X] is less than or equal to E[Y]
indicator function for events
for event A in F,
1_A: Ω to R
1_A(ω) =
{1 if ω is in A
{0 if ω isn’t in A
E(1_A) = 1P(A) + 0(P(A^c))
=P(A)
- sometimes write 1_{event} = 1{event}
LEMMA 2.4.2
indicator function for events
Let A∈G
(remember G is curly G similar to g)
Then function 1_A is G-measurable
proof: Range of 1_A is {0,1}. Pre-images are: 1_A -1 (0) =Ω\A ∈G 1_A -1 (1) = A ∈G by2.2.2, Y is G-measurable
CONDITIONING
breaking RV into cases
for example, given any random var X we write:
X= X 1{X≥1} + X1{X less than 1}
( as only one of the RHS terms is non-zero)
for example, given any random var X we write:
X= X 1{X≥1} + X1{X less than 1}
( as only one of the RHS terms is non-zero)
we use this to prove an inequality:
putting |X| in place of X then taking the expectation to obtain:
E[|X|] = E[|X| 1{|X|≥1} ] + E[|X| 1{|X|less than 1} ]
≤
E[X^2 1_{|X|≥1} ] + 1
≤
E[X^2] +1
(to reduce second key point we can only use |x| is less than or equal to x^2 if x is bigger than or equal to 1)
ie we want to prove
E[|X|] ≤E[X^2] +1
|X| = |X| 1{|X|≥1} + |X| 1{|X| less thn 1}
BY MONOTONICITY:
FIRST TERM
|X| 1{|X|≥1} ≤ X^2 1{|X|≥1}
assume |X| ≥ 1 and for reals X ≤ X^2
SECONT TERM
|X| 1{|X|less thn 1}≤ 1{|X| Less thn 1}
assume |X| less than 1 and for reals X ≤ 1
|X| = |X| 1{|X|≥1} + |X| 1{|X| less thn 1}
≤ X^2* 1{|X|≥1} + 1 *1{|X|less thn 1}
≤ X^2 + 1
by monotonicity of expectation:
E(|X|) ≤ E [1 +X^2] -1 + E[X^2]
DEF 2.4.3 L^p
Let p∈ [1, infinity)
we say that X ∈ L^p
if E[|X|^p] is less than infinity
we usually care about cases p = 1 or p = 2
L^1 and L^2
*by definition L^1 is the set of random variables such that E[|X|] is less than infinity
ie that its finite
- L^2 is the set of random variables such that Var(X) is less than infinity
Var(X) = E(X^2) - E(X)^2 ie finite var
- from 2.7, if X ∈L^2 then X∈L^1
DEF 2.4.4
bounded
We say that a random variable X is bounded if there exist a deterministic constant c ∈R such that |X| ≤ c
if X is bounded then using monotonicity:
E[|X|^p] ≤ E[C^p] =c^p less than infinity
meaning that X ∈ L^p for all p