Lecture 4. We also define the set of possible values for the random walk as the set of all x R d such that P(S n = x) > 0 for some n.

Similar documents
Sequences and Series of Functions

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

6a Time change b Quadratic variation c Planar Brownian motion d Conformal local martingales e Hints to exercises...

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

Notes 27 : Brownian motion: path properties

7.1 Convergence of sequences of random variables

7.1 Convergence of sequences of random variables

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

Lecture Chapter 6: Convergence of Random Sequences

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 3 9/11/2013. Large deviations Theory. Cramér s Theorem

Random Models. Tusheng Zhang. February 14, 2013

1 Convergence in Probability and the Weak Law of Large Numbers

Lecture 19: Convergence

M17 MAT25-21 HOMEWORK 5 SOLUTIONS

An Introduction to Randomized Algorithms

Lecture 8: Convergence of transformations and law of large numbers

LECTURE 8: ASYMPTOTICS I

Probability for mathematicians INDEPENDENCE TAU

Math 525: Lecture 5. January 18, 2018

Entropy Rates and Asymptotic Equipartition

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

It is often useful to approximate complicated functions using simpler ones. We consider the task of approximating a function by a polynomial.

Infinite Sequences and Series

5 Birkhoff s Ergodic Theorem

n=1 a n is the sequence (s n ) n 1 n=1 a n converges to s. We write a n = s, n=1 n=1 a n

(A sequence also can be thought of as the list of function values attained for a function f :ℵ X, where f (n) = x n for n 1.) x 1 x N +k x N +4 x 3

Lecture 3 : Random variables and their distributions

This section is optional.

K. Grill Institut für Statistik und Wahrscheinlichkeitstheorie, TU Wien, Austria

6.3 Testing Series With Positive Terms

Ma 530 Introduction to Power Series

Notes #3 Sequences Limit Theorems Monotone and Subsequences Bolzano-WeierstraßTheorem Limsup & Liminf of Sequences Cauchy Sequences and Completeness

Chapter 0. Review of set theory. 0.1 Sets

Introduction to Probability. Ariel Yadin

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

Section 11.8: Power Series

Chapter 6 Infinite Series

Random Matrices with Blocks of Intermediate Scale Strongly Correlated Band Matrices

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

A Proof of Birkhoff s Ergodic Theorem

6. Uniform distribution mod 1

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 2 9/9/2013. Large Deviations for i.i.d. Random Variables

1 Lecture 2: Sequence, Series and power series (8/14/2012)

Axioms of Measure Theory

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

Complex Analysis Spring 2001 Homework I Solution

MS&E 321 Spring Stochastic Systems June 1, 2013 Prof. Peter W. Glynn Page 1 of 5

lim za n n = z lim a n n.

Achieving Stationary Distributions in Markov Chains. Monday, November 17, 2008 Rice University

Math Solutions to homework 6

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

Chapter 6 Principles of Data Reduction

Lecture 12: November 13, 2018

Distribution of Random Samples & Limit theorems

ECONOMETRIC THEORY. MODULE XIII Lecture - 34 Asymptotic Theory and Stochastic Regressors

6 Infinite random sequences

Lecture 2: April 3, 2013

Discrete Mathematics for CS Spring 2007 Luca Trevisan Lecture 22

5 Many points of continuity

EECS564 Estimation, Filtering, and Detection Hwk 2 Solns. Winter p θ (z) = (2θz + 1 θ), 0 z 1

Lecture 3: August 31

Math 113 Exam 3 Practice

Recurrence Relations

STAT Homework 1 - Solutions

62. Power series Definition 16. (Power series) Given a sequence {c n }, the series. c n x n = c 0 + c 1 x + c 2 x 2 + c 3 x 3 +

Define a Markov chain on {1,..., 6} with transition probability matrix P =

January 25, 2017 INTRODUCTION TO MATHEMATICAL STATISTICS

MATH 312 Midterm I(Spring 2015)

Central limit theorem and almost sure central limit theorem for the product of some partial sums

BIRKHOFF ERGODIC THEOREM

Math 113 Exam 3 Practice

6 Integers Modulo n. integer k can be written as k = qn + r, with q,r, 0 r b. So any integer.

Econ 325 Notes on Point Estimator and Confidence Interval 1 By Hiro Kasahara

ST5215: Advanced Statistical Theory

How to Maximize a Function without Really Trying

Ma 530 Infinite Series I

Math 210A Homework 1

Lecture 12: Subadditive Ergodic Theorem

MAT1026 Calculus II Basic Convergence Tests for Series

Lecture Notes for Analysis Class

Metric Space Properties

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.

Asymptotic distribution of products of sums of independent random variables

Advanced Stochastic Processes.

A sequence of numbers is a function whose domain is the positive integers. We can see that the sequence

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

The natural exponential function

ECE 901 Lecture 12: Complexity Regularization and the Squared Loss

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

Differentiable Convex Functions

Let us give one more example of MLE. Example 3. The uniform distribution U[0, θ] on the interval [0, θ] has p.d.f.

Final Review for MATH 3510

Sequence A sequence is a function whose domain of definition is the set of natural numbers.

Lecture 7: Properties of Random Samples

Lecture 12: September 27

Simple Random Walk. Timo Leenman, June 3, Bachelor scription Supervisor: Prof. dr. F den Hollander

CHAPTER 10 INFINITE SEQUENCES AND SERIES

Analytic Continuation

Chapter 2. Periodic points of toral. automorphisms. 2.1 General introduction

Direction: This test is worth 250 points. You are required to complete this test within 50 minutes.

Transcription:

Radom Walks ad Browia Motio Tel Aviv Uiversity Sprig 20 Lecture date: Mar 2, 20 Lecture 4 Istructor: Ro Peled Scribe: Lira Rotem This lecture deals primarily with recurrece for geeral radom walks. We preset several criteria for a radom walk to be recurret, ad prove Polya s theorem o recurrece ad trasiece for the simple radom walk o Z d Recurrece for geeral radom walks Remember the followig result from the previous lecture: Every -dimesioal radom walk S satisfies exactly oe of the followig almost surely: (i) S = 0 for every. (ii) lims =. (iii) lims =. (iv) limsups = ad limifs =. We would like to cotrast this with the followig defiitio: Defiitio A radom walk takig values i R d is called poit-recurret if P(S = 0 ifiitely ofte) =. (remember this probability is either 0 or, by the Hewitt-Savage zero-oe law). We also defie the set of possible values for the radom walk as the set of all x R d such that P(S = x) > 0 for some. Exercise If a radom walk is poit-recurret, the for every possible value x. P(S = x ifiitely ofte) = Remark 2 If the radom walk is ot discrete (havig a atomic distributio for each step) the these defiitios are ot very useful. Istead we say that the radom walk is eighborhood-recurret if for some (ad the for ay, as ca be proved) ǫ > 0 P( S < ǫ ifiitely ofte) =. 4-

Similary, possible values are chaged to those x for which for every ǫ > 0 there exists a such that P( S x < ǫ) > 0. Propositio 3 A radom walk i R d is poit-reucrret if ad oly if oe of the followig holds: (i) P(, S = 0) =. (ii) P(S = 0 ifiitely ofte) = (this is the defiitio of poit-recurrece). (iii) P(S = 0) =. Proof The equivalece betwee (i) ad (ii) is clear. For the equivalece betwee (i) ad (iii), ote that by the strog markov property, the umber of visits to 0 is distributed geometrically Geo(p), where p is the probability of o retur to 0. Therefore P(S = 0) = E(Geo(p)) = p, ad ow the equivalece is obvious. Theorem 4 (Pólya) A simple radom walk i Z d is recurret if ad oly if d = or d = 2. Proof We cosider each dimesio separately: d = : We already saw this result i dimesio, but we ow give a differet proof. Notice that by Stirlig s formula ( ) 2 P(S 2 = 0) = 2 2. π Therefore P(S 2 = 0) = ad we are doe. d = 2: By rotatig Z 2 i 45 we get that each step is like movig oe step i each of two idepedet, oe dimesioal, simple radom walks. Hece ad agai P(S 2 = 0) =. P(S 2 = 0) ( π ) 2 = π, 4-2

d = 3: We calculate explicitly: P(S 2 = 0) = 6 2 j,k 0 j+k = 2 2 ( 2 (2)! (j! k! ( j k)!) 2 ) j,k 0 j+k ( ) 3! 2 j! k! ( j k)! (here j is the umber of steps i directio (,0,0), ad k is the umber of steps i directio (0,,0) ). Sice 3! j! k! ( j k)! =, it follows that j,k 0 j+k P(S 2 = 0) 2 2 ( 2 ) max j,k 0 j+k ( ) 3!. j! k! ( j k)! The maximum is achieved whe j ad k are as close as possible to 3, ad this maximum is smaller tha c for some costat c. Hece P(S 2 = 0) c so P(S 2 = 0) < ad the radom walk is trasiet. d 4: The radom walk is still trasiet, sice the first 3 coordiates are trasiet. Exercise Fid a poit-recurret radom walk o R whose set of possible values is a coutable dese set i R. 3 2, 2 The Chug-Fuchs theorem Theorem 5 (Chug-Fuchs, 95) Let S be a radom walk i R d. The: (i) If d = ad S 0 i probability, the S is eighborhood-recurret. I particular, this happes if EX = 0. 4-3

(ii) If d = 2 ad S coverges i distributio to a cetered ormal distributio, the S is eighborhood-recurret. I particular, this happes if EX = 0 ad EX 2 <. (iii) If d = 3 ad the radom walk is ot cotaied i a plae the it is eighborhoodtrasiet (the coditio just meas that the set of eighborhood-possible values is ot cotaied i a plae). Remark 6 If the walk is o Z d the eighborhood-recurrece is the same as poitrecurrece ad the theorem still applies. Remark 7 If EX = µ the by the strog law of large umbers S µ almost surely. Therefore if EX exists ad is ozero, it s obvious that the radom walk is trasiet. We wo t prove the Chug-Fuchs theorem i its full geerality, but we will show some partial results, together with other criteria for recurrece. We will eed the followig theorem: Theorem 8 (Birkhoff ergodic theorem for fuctios of IIDs) Let X,X 2,... be IID radom variables i a state space S ad let g : S R be ay measureable fuctio. Defie Y = g(x,x +,...). If E Y < the almost surely ad i L. lim Y k = EY k= Remark 9 I fact, it is sufficiet to take the Y s to be a statioary ergodic sequece with E Y < We wo t prove this theorem (a proof ca be foud, for example, i []). However, we will use this result to prove the followig: Theorem 0 (Keste-Spitzer-Whitma rage theorem, 964) For a radom walk i R d let R = {x : k,s k = x} be the umber of distict poits visited, the almost surely. R P(o retur to 0) = lim 4-4

Remark This theorem also holds for ergodic markov chais ad eve statioary ergodic sequeces. Proof Note that Thus R R = [S j S k for k+ j ]. [S k ever revisited] = g(x k+,x k+2,...) for a appropriate measureable fuctio g. Now we ca use Birkhoff ergodic theorem ad get that R lim if Eg(X,X 2,...) = P(o retur to 0) almost surely. O the other had, fix M ad ote that R M Agai by Birkhoff we get that ( R lim sup limsup almost surely. But as M so it follows that ad we are doe. [S k ot visited agai by time k +M]+M. M + M = P(0 ot revisited by time M) [S k ot visited agai by time k+m] P(0 ot revisited by time M) P(0 is ever revisited), lim sup R P(0 is ever revisited) We are ow ready to prove part (i) of the Chug-Fuchs theorem, uder the additioal assumptios that the radom walk is o Z ad that EX exists (ad therefore EX = 0): Proof By the strog law of large umbers we kow that S 0 almost surely. Therefore for every ǫ > 0 we ca fid 0 (which is a radom variable by itself) such that S < ǫ for > 0. Hece for much larger tha 0 we get that R 3ǫ, so limsup R 3ǫ almost surely. Sice ǫ was arbitrary it follows that R 0, so by the Keste-Spitzer-Whitma theorem P(o retur to 0) = 0, ad the walk is recurret. ) = 4-5

3 Fourier aalytic criterio for recurrece For a radom walk o Z d, let ϕ(θ) = Ee iθ X be the characteristic fuctio of X. Note that ϕ is essetially from T d = [ π,π] d to C. Remider (i) Ee iθ S = ϕ (θ) (ii) We have the Fourier iversio formula: P(S = y) = (2π) d T d e iy θ ϕ (θ)dθ Theorem 2 A radom walk S o Z d is reucrret if ad oly if ( ) lim dθ = rր rϕ(θ) T d R Remark 3 (i) It is also true, but more difficult to prove, that S is recurret if ad oly if ( ) dθ = ϕ(θ) T d R (ii) Similarly, for geeral radom walks i R d, S is eighbourhood-recurret if ad oly if for some δ > 0 (ad the for every δ > 0) ( ) dθ = ϕ(θ) (this time ϕ is defied o R d ) ( δ,δ) d R (iii) If X has first ad secod momets, the (i dimesio): as θ 0. ϕ(θ) = +iθex θ2 2 EX2 +o(θ2 ) (iv) The weak law of large umbers holds if ad oly if ϕ (0) exists, ad the S µ i probability, where ϕ (0) = iµ (By a result of E. Pitma) Proof P(S = 0) = (2π) d T d ϕ (θ)dθ, 4-6

so for 0 < r < we have r P(S = 0) = (2π) d r ϕ (θ)dθ = (2π) d T d Here the fial step was Fubii s theorem, which applies because T d r ϕ (θ) dθ Sice the left had side is real it follows that ( r P(S = 0) = (2π) T d R d T d r dθ <. rϕ(θ) r ϕ (θ)dθ. T d ) dθ. Whe r ր the left had side coverges to P(S = 0), so we are doe. We would like to use this result for a few remarks about the relatio betwee recurrece ad momets Defiitio 4 The symmetric stable radom variable of idex α is the radom variable X o R such that Ee iθ X = e θ α Such a radom variable exists for every α 2. If α = 2 this is just the gaussia. If α < 2 this is a symmetric radom variable such that E X p < for every p < α, but E X α =. Note that as θ 0. Therefore ( δ,δ) d ϕ(θ) = e θ α θ α { dθ : α ϕ(θ) = < : α <, so these radom walks are recurret if ad oly if α. For α > this is a cosequece of the Chug-Fuchs theorem, sice EX = 0. For α = X has the Cauchy distributio, with desity π(+x 2 ). For such X, S is agai Cauchy distributed (as ca be see by calculatig its characteristic fuctio), so the weak law of large umbers does t hold eve though the walk is recurret. The recurrece coditio is ot oly about the tail of the radom varaible. I 964, Shepp gave examples of recurret walks with arbitrary heavy tails. I fact he proved the followig: 4-7

Theorem 5 For every fuctio ǫ(x) such that lim x ǫ(x) = 0, there exists a symmetric radom variable X such that P( X > x) ǫ(x) for all large x, ad the radom walk S is eighborhood (ad eve poit) recurret. Refereces [] Lalley S. Oe-dimesioal radom walks. http://galto.uchicago.edu/ lalley/courses/32/rw.pdf. 4-8