Probability and Random Processes

Similar documents
ST5215: Advanced Statistical Theory

Introduction to Probability. Ariel Yadin

Convergence of random variables. (telegram style notes) P.J.C. Spreij

Probability for mathematicians INDEPENDENCE TAU

Distribution of Random Samples & Limit theorems

7.1 Convergence of sequences of random variables

1 Convergence in Probability and the Weak Law of Large Numbers

Notes 5 : More on the a.s. convergence of sums

7.1 Convergence of sequences of random variables

Solutions of Homework 2.

Introduction to Probability. Ariel Yadin. Lecture 7

Probability Theory. Muhammad Waliji. August 11, 2006

Lecture 8: Convergence of transformations and law of large numbers

This section is optional.

Lecture 3 : Random variables and their distributions

Advanced Stochastic Processes.

Lecture 20: Multivariate convergence and the Central Limit Theorem

Probability 2 - Notes 10. Lemma. If X is a random variable and g(x) 0 for all x in the support of f X, then P(g(X) 1) E[g(X)].

Lecture 19: Convergence

Lecture Chapter 6: Convergence of Random Sequences

Chapter 3. Strong convergence. 3.1 Definition of almost sure convergence

Learning Theory: Lecture Notes

Solutions to HW Assignment 1

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 19 11/17/2008 LAWS OF LARGE NUMBERS II THE STRONG LAW OF LARGE NUMBERS

Lecture 6: Coupon Collector s problem

Glivenko-Cantelli Classes

Math 525: Lecture 5. January 18, 2018

6 Infinite random sequences

ECE 330:541, Stochastic Signals and Systems Lecture Notes on Limit Theorems from Probability Fall 2002

for all x ; ;x R. A ifiite sequece fx ; g is said to be ND if every fiite subset X ; ;X is ND. The coditios (.) ad (.3) are equivalet for =, but these

Lecture 2: Concentration Bounds

1 = δ2 (0, ), Y Y n nδ. , T n = Y Y n n. ( U n,k + X ) ( f U n,k + Y ) n 2n f U n,k + θ Y ) 2 E X1 2 X1

MAXIMAL INEQUALITIES AND STRONG LAW OF LARGE NUMBERS FOR AANA SEQUENCES

An Introduction to Randomized Algorithms

ACO Comprehensive Exam 9 October 2007 Student code A. 1. Graph Theory

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 21 11/27/2013

Large Sample Theory. Convergence. Central Limit Theorems Asymptotic Distribution Delta Method. Convergence in Probability Convergence in Distribution

2.1. Convergence in distribution and characteristic functions.

STA Object Data Analysis - A List of Projects. January 18, 2018

Chapter 5. Inequalities. 5.1 The Markov and Chebyshev inequalities

5 Birkhoff s Ergodic Theorem

Entropy Rates and Asymptotic Equipartition

1 The Haar functions and the Brownian motion

Lecture 01: the Central Limit Theorem. 1 Central Limit Theorem for i.i.d. random variables

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 6 9/23/2013. Brownian motion. Introduction

Lecture 12: November 13, 2018

Agnostic Learning and Concentration Inequalities

Chapter 1. Probability

Solution to Chapter 2 Analytical Exercises

Lecture 2: April 3, 2013

Parameter, Statistic and Random Samples

Generalized Semi- Markov Processes (GSMP)

2.2. Central limit theorem.

Notes 19 : Martingale CLT

Some Basic Probability Concepts. 2.1 Experiments, Outcomes and Random Variables

The Central Limit Theorem

Probability and Statistics

The Borel-Cantelli Lemma and its Applications

Econ 325/327 Notes on Sample Mean, Sample Proportion, Central Limit Theorem, Chi-square Distribution, Student s t distribution 1.

Partial match queries: a limit process

ECE534, Spring 2018: Final Exam

THE STRONG LAW OF LARGE NUMBERS FOR STATIONARY SEQUENCES

EE 4TM4: Digital Communications II Probability Theory

SDS 321: Introduction to Probability and Statistics

BIRKHOFF ERGODIC THEOREM

4. Partial Sums and the Central Limit Theorem

Asymptotic distribution of products of sums of independent random variables

On Random Line Segments in the Unit Square

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 6 9/24/2008 DISCRETE RANDOM VARIABLES AND THEIR EXPECTATIONS

Math 341 Lecture #31 6.5: Power Series

Solution. 1 Solutions of Homework 1. Sangchul Lee. October 27, Problem 1.1

Integrable Functions. { f n } is called a determining sequence for f. If f is integrable with respect to, then f d does exist as a finite real number

Probability, Random Variables and Random Processes

Notes 27 : Brownian motion: path properties

Basics of Probability Theory (for Theory of Computation courses)

Elements of Statistical Methods Lots of Data or Large Samples (Ch 8)

Mi-Hwa Ko and Tae-Sung Kim

Introduction to Probability. Ariel Yadin. Lecture 2

This exam contains 19 pages (including this cover page) and 10 questions. A Formulae sheet is provided with the exam.

It is always the case that unions, intersections, complements, and set differences are preserved by the inverse image of a function.

AMS570 Lecture Notes #2

ECE 6980 An Algorithmic and Information-Theoretic Toolbox for Massive Data

Sequences and Series of Functions

Sieve Estimators: Consistency and Rates of Convergence

HOMEWORK #4 - MA 504

An Introduction to Asymptotic Theory

Singular Continuous Measures by Michael Pejic 5/14/10

LECTURE NOTES ON PROBABILITY

ECE534, Spring 2018: Solutions for Problem Set #2

Introductory Ergodic Theory and the Birkhoff Ergodic Theorem

Monkeys and Walks. Muhammad Waliji. August 12, 2006

Definition 4.2. (a) A sequence {x n } in a Banach space X is a basis for X if. unique scalars a n (x) such that x = n. a n (x) x n. (4.

Real Numbers R ) - LUB(B) may or may not belong to B. (Ex; B= { y: y = 1 x, - Note that A B LUB( A) LUB( B)

Lecture 2 February 8, 2016

January 25, 2017 INTRODUCTION TO MATHEMATICAL STATISTICS

32 estimating the cumulative distribution function

LECTURE 8: ASYMPTOTICS I

Introduction to Extreme Value Theory Laurens de Haan, ISM Japan, Erasmus University Rotterdam, NL University of Lisbon, PT

CS166 Handout 02 Spring 2018 April 3, 2018 Mathematical Terms and Identities

Statistical Theory; Why is the Gaussian Distribution so popular?

Transcription:

Probability ad Radom Processes Lecture 5 Probability ad radom variables The law of large umbers Mikael Skoglud, Probability ad radom processes 1/21 Why Measure Theoretic Probability? Stroger limit theorems Coditioal probability/expectatio Proper theory for cotiuous ad mixed radom variables Mikael Skoglud, Probability ad radom processes 2/21

Probability Space A probability space is a measure space Ω, A, P the sample space Ω is the uiverse, i.e. the set of all possible outcomes the evet class A is a σ-algebra of measurable sets called evets the probability measure is a measure o evets i A with the property P Ω = 1 Mikael Skoglud, Probability ad radom processes 3/21 Iterpretatio A radom experimet geerates a outcome ω Ω For each A A either ω A or ω / A A evet A i A occurs if ω A with probability P A sice A is the σ-algebra of measurable sets, we are esured that all reasoable combiatios of evets ad sequeces of evets are measurable, i.e., have probabilities Mikael Skoglud, Probability ad radom processes 4/21

With Probability Oe A evet E A occurs with probability oe if P E = 1 almost everywhere, almost certaily, almost surely,... Mikael Skoglud, Probability ad radom processes 5/21 Idepedece E ad F i A are idepedet if P E F = P EP F The evets i a collectio A 1,..., A are pairwise idepedet if A i ad A j are idepedet for i j mutually idepedet if for ay {i 1, i 2,..., i k } {1, 2,..., } P A i1 A i2 A ik = P A i1 P A i2 P A ik A ifiite collectio is mutually idepedet if ay fiite subset of evets is mutually idepedet mutually pairwise but ot vice versa Mikael Skoglud, Probability ad radom processes 6/21

Evetually ad Ifiitely Ofte A probability space Ω, A, P ad a ifiite sequece of evets {A }, defie lim if A = A k, lim sup A = A k =1 k= =1 k= ω lim if A iff there is a N such that ω A for all > N, that is, the evet lim if A occurs evetually, {A evetually} ω lim sup A iff for ay N there is a > N such that ω A, that is, the evet lim sup A occurs ifiitely ofte {A i.o.} Mikael Skoglud, Probability ad radom processes 7/21 Borel Catelli The Borel Catelli lemma: A probability space Ω, A, P ad a ifiite sequece of evets {A } 1 if P A <, the P {A i.o} = 0 2 if the evets {A } are mutually idepedet ad P A =, the P {A i.o} = 1 Mikael Skoglud, Probability ad radom processes 8/21

Radom Variables A probability space Ω, A, P. A real-valued fuctio Xω o Ω is called a radom variable if it s measurable w.r.t. Ω, A Recall: measurable X 1 O A for ay ope O R X 1 A A for ay A B the Borel sets Notatio: the evet {ω : Xω B} X B P {X A} {X B} P X A, X B, etc. Mikael Skoglud, Probability ad radom processes 9/21 Distributios X is measurable P X B is well-defied for ay B B The distributio of X is the fuctio µ X B = P X B, for B B µ X is a probability measure o R, B The probability distributio fuctio of X is the real-valued fuctio F X x = P {ω : Xω x} = otatio = P X x F X is obviously the distributio fuctio of the fiite measure µ X o R, B, i.e. F X x = µ X, x] Mikael Skoglud, Probability ad radom processes 10/21

Idepedece Two radom variables X ad Y are pairwise idepedet if the evets {X A} ad {Y B} are idepedet for ay A ad B i B A collectio of radom variables X 1,..., X is mutually idepedet if the evets {X i B i } are mutually idepedet for all B i B Mikael Skoglud, Probability ad radom processes 11/21 Expectatio For a radom variable o Ω, A, P, the expectatio of X is defied as E[X] = XωdP ω For ay Borel-measurable real-valued fuctio g E[gX] = gxdf X x = gxdµ X x Ω i particular E[X] = xdµ X x Mikael Skoglud, Probability ad radom processes 12/21

Variace The variace of X, VarX = E[X E[X] 2 ] Chebyshev s iequality: For ay ε > 0, P X E[X] ε VarX ε 2 Kolmogorov s iequality: For mutually idepedet radom variables {X k } with VarX k <, set S j = j X k, 1 j, the for ay ε > 0 P max S j E[S j ] ε VarS j ε 2 = 1 Chebyshev Mikael Skoglud, Probability ad radom processes 13/21 The Law of Large Numbers A sequece {X } is iid if the radom variables X all have the same distributio ad are mutually idepedet For ay iid sequece {X } with µ = E[X ] <, the evet lim occurs with probability oe 1 X k = µ Toward the ed of the course, we will geeralize this result to statioary ad ergodic radom processes... Mikael Skoglud, Probability ad radom processes 14/21

S = 1 X µ with probability oe S µ i probability, i.e., for each ε > 0 lim P { S µ ε} = 0 i geeral i probability does ot imply with probability oe covergece i measure does ot imply covergece a.e. Mikael Skoglud, Probability ad radom processes 15/21 The Law of Large Numbers: Proof Lemma 1: For a oegative radom variable X P X E[X] =1 P X =0 Lemma 2: For mutually idepedet radom variables {X } with VarX < it holds that X E[X ] coverges with probability oe Lemma 3 Kroecker s Lemma: Give a sequece {a } with 0 a 1 a 2 ad lim a =, ad aother sequece {x k } such that lim k x k exists, the lim 1 a a k x k = 0 Mikael Skoglud, Probability ad radom processes 16/21

Assume without loss of geerality why? that µ = 0 Lemma 1 =1 P X = =1 P X 1 < Let E = { X k k i.o.}, Borel Catelli P E = 0 we ca cocetrate o ω E c Let Y = X χ { X <}; if ω E c the there is a N such that Y ω = X ω for N, thus for ω E c lim 1 X k = 0 lim 1 Y k = 0 Note that E[Y ] µ = 0 as Mikael Skoglud, Probability ad radom processes 17/21 Lettig Z = 1 Y, it ca be show that =1 VarZ < requires some work. Hece, accordig to Lemma 2 the limit Z = lim Z k E[Z k ] exists with probability oe. Furthermore, by Lemma 3 1 Y k E[Y k ] = 1 kz k E[Z k ] 0 where also 1 E[Y k ] 0 sice E[Y k ] E[X k ] = E[X 1 ] = 0 Mikael Skoglud, Probability ad radom processes 18/21

Proof of Lemma 2 Assume w.o. loss of geerality that E[X ] = 0, set S = X k For E A with E 1 E 2 it holds that P E = lim P E Therefore, for ay m 0 P { S m+k S m ε} = lim P = lim P { S m+k S m ε} max S m+k S m ε 1 k Mikael Skoglud, Probability ad radom processes 19/21 Let Y k = X m+k ad T k = k Y j = S m+k S m, j=1 the Kolmogorov s iequality implies P max S m+k S m ε 1 k VarS m+ S m ε 2 = 1 ε 2 m+ k=m+1 VarX k Hece P { S m+k S m ε} 1 ε 2 k=m+1 VarX k Mikael Skoglud, Probability ad radom processes 20/21

Sice VarX <, we get lim P m { S m+k S m ε} = 0 Now, let E = {ω : {S ω} does ot coverge}. The ω E iff {S ω} is ot a Cauchy sequece for ay there is a k ad a r such that S +k S r 1. Hece, equivaletly, { E = S +k S 1 } r k r=1 k For F 1 F 2 F 3, P k F k = lim P F k, hece for ay r > 0 { P S +k S 1 } { = lim r P S +k S 1 } r That is, P E = 0 k Mikael Skoglud, Probability ad radom processes 21/21