Brownian Motion and Stochastic Calculus. Brownian Motion and Stochastic Calculus

Similar documents
Asymptotic Behavior of Solutions of Nonlinear Delay Differential Equations With Impulse

8. Queueing systems lect08.ppt S Introduction to Teletraffic Theory - Fall

Upper Bound For Matrix Operators On Some Sequence Spaces

Key words: Fractional difference equation, oscillatory solutions,

(1) Cov(, ) E[( E( ))( E( ))]

QR factorization. Let P 1, P 2, P n-1, be matrices such that Pn 1Pn 2... PPA

14. Poisson Processes

The Poisson Process Properties of the Poisson Process

BSDEs with polynomial growth generators Philippe Briand IRMAR, Universite Rennes 1, Rennes Cedex, FRANCE Rene Carmona Statistics & Operations R

Continuous Time Markov Chains

Least Squares Fitting (LSQF) with a complicated function Theexampleswehavelookedatsofarhavebeenlinearintheparameters

Solution of Impulsive Differential Equations with Boundary Conditions in Terms of Integral Equations

Cyclone. Anti-cyclone

Learning of Graphical Models Parameter Estimation and Structure Learning

Continuous Indexed Variable Systems

The Mean Residual Lifetime of (n k + 1)-out-of-n Systems in Discrete Setting

On g Evaluations with L p Domains under Jump Filtration

Some Probability Inequalities for Quadratic Forms of Negatively Dependent Subgaussian Random Variables

MATH 247/Winter Notes on the adjoint and on normal operators.

THE PUBLISHING HOUSE PROCEEDINGS OF THE ROMANIAN ACADEMY, Series A, OF THE ROMANIAN ACADEMY Volume 10, Number 2/2009, pp

Density estimation III.

Comparison of the Bayesian and Maximum Likelihood Estimation for Weibull Distribution

Real-Time Systems. Example: scheduling using EDF. Feasibility analysis for EDF. Example: scheduling using EDF

Available online Journal of Scientific and Engineering Research, 2014, 1(1): Research Article

The Linear Regression Of Weighted Segments

VARIATIONAL ITERATION METHOD FOR DELAY DIFFERENTIAL-ALGEBRAIC EQUATIONS. Hunan , China,

Solution set Stat 471/Spring 06. Homework 2

FALL HOMEWORK NO. 6 - SOLUTION Problem 1.: Use the Storage-Indication Method to route the Input hydrograph tabulated below.

Fully Fuzzy Linear Systems Solving Using MOLP

Supplement Material for Inverse Probability Weighted Estimation of Local Average Treatment Effects: A Higher Order MSE Expansion

Midterm Exam. Tuesday, September hour, 15 minutes

Cyclically Interval Total Colorings of Cycles and Middle Graphs of Cycles

Chapter 3: Maximum-Likelihood & Bayesian Parameter Estimation (part 1)

UNIVERSITY OF OSLO DEPARTMENT OF ECONOMICS

Partial Molar Properties of solutions

AML710 CAD LECTURE 12 CUBIC SPLINE CURVES. Cubic Splines Matrix formulation Normalised cubic splines Alternate end conditions Parabolic blending

Moments of Order Statistics from Nonidentically Distributed Three Parameters Beta typei and Erlang Truncated Exponential Variables

Quantum Mechanics II Lecture 11 Time-dependent perturbation theory. Time-dependent perturbation theory (degenerate or non-degenerate starting state)

As evident from the full-sample-model, we continue to assume that individual errors are identically and

Density estimation. Density estimations. CS 2750 Machine Learning. Lecture 5. Milos Hauskrecht 5329 Sennott Square

The Signal, Variable System, and Transformation: A Personal Perspective

K3 p K2 p Kp 0 p 2 p 3 p

F. Inequalities. HKAL Pure Mathematics. 進佳數學團隊 Dr. Herbert Lam 林康榮博士. [Solution] Example Basic properties

-distributed random variables consisting of n samples each. Determine the asymptotic confidence intervals for

Comparison of Differences between Power Means 1

Interval Estimation. Consider a random variable X with a mean of X. Let X be distributed as X X

P a g e 5 1 of R e p o r t P B 4 / 0 9

The algebraic immunity of a class of correlation immune H Boolean functions

A New Generalized Gronwall-Bellman Type Inequality

Fundamentals of Speech Recognition Suggested Project The Hidden Markov Model

. The set of these sums. be a partition of [ ab, ]. Consider the sum f( x) f( x 1)

4. THE DENSITY MATRIX

STOCHASTIC CALCULUS I STOCHASTIC DIFFERENTIAL EQUATION

Complementary Tree Paired Domination in Graphs

The MacWilliams Identity of the Linear Codes over the Ring F p +uf p +vf p +uvf p

1 Solution to Problem 6.40

Lecture 02: Bounding tail distributions of a random variable

( ) :. : - 1. : - 2. : - 3. : - 4., - :. - :1 -.,M /M/1. :2 -. :* - 3.,M /M/ M, /M/c/, K M/M/c :4 - :. : - 1. : - 2.

EP2200 Queuing theory and teletraffic systems. 3rd lecture Markov chains Birth-death process - Poisson process. Viktoria Fodor KTH EES

The Properties of Probability of Normal Chain

Determination of Antoine Equation Parameters. December 4, 2012 PreFEED Corporation Yoshio Kumagae. Introduction

The Arithmetic-Geometric mean inequality in an external formula. Yuki Seo. October 23, 2012

The ray paths and travel times for multiple layers can be computed using ray-tracing, as demonstrated in Lab 3.

Fault Tolerant Computing. Fault Tolerant Computing CS 530 Probabilistic methods: overview

Unit 10. The Lie Algebra of Vector Fields

March 14, Title: Change of Measures for Frequency and Severity. Farrokh Guiahi, Ph.D., FCAS, ASA

An interesting result about subset sums. Nitu Kitchloo. Lior Pachter. November 27, Abstract

4. Runge-Kutta Formula For Differential Equations

Mu Sequences/Series Solutions National Convention 2014

SUPPLEMENT TO STATISTICAL PROPERTIES OF MICROSTRUCTURE NOISE (Econometrica, Vol. 85, No. 4, July 2017, ) for j = (0 j) j = 0 1 (B.

FORCED VIBRATION of MDOF SYSTEMS

Brownian Motion and Stochastic Calculus

Suppose we have observed values t 1, t 2, t n of a random variable T.

PTAS for Bin-Packing

2 f(x) dx = 1, 0. 2f(x 1) dx d) 1 4t t6 t. t 2 dt i)

Mathematical Formulation

Least squares and motion. Nuno Vasconcelos ECE Department, UCSD

P-Convexity Property in Musielak-Orlicz Function Space of Bohner Type

Chapter 8. Simple Linear Regression

( ) ( ) Weibull Distribution: k ti. u u. Suppose t 1, t 2, t n are times to failure of a group of n mechanisms. The likelihood function is

In the complete model, these slopes are ANALYSIS OF VARIANCE FOR THE COMPLETE TWO-WAY MODEL. (! i+1 -! i ) + [(!") i+1,q - [(!

Entropy, Relative Entropy and Mutual Information

OPTIMALITY AND SECOND ORDER DUALITY FOR A CLASS OF QUASI-DIFFERENTIABLE MULTIOBJECTIVE OPTIMIZATION PROBLEM

Sensors and Regional Gradient Observability of Hyperbolic Systems

Fresnel Equations cont.

Basic Results in Functional Analysis

D KL (P Q) := p i ln p i q i

A note on Turán number Tk ( 1, kn, )

The textbook expresses the stock price as the present discounted value of the dividend paid and the price of the stock next period.

Iterated Bernstein polynomial approximations

The textbook expresses the stock price as the present discounted value of the dividend paid and the price of the stock next period.

Lecture two. January 17, 2019

JORIND 9(2) December, ISSN

Mixed Integral Equation of Contact Problem in Position and Time

θ = θ Π Π Parametric counting process models θ θ θ Log-likelihood: Consider counting processes: Score functions:

= lim. (x 1 x 2... x n ) 1 n. = log. x i. = M, n

A Comparison of AdomiansDecomposition Method and Picard Iterations Method in Solving Nonlinear Differential Equations

Probability Bracket Notation, Probability Vectors, Markov Chains and Stochastic Processes. Xing M. Wang Sherman Visual Lab, Sunnyvale, CA, USA

A Remark on Generalized Free Subgroups. of Generalized HNN Groups

Solution. The straightforward approach is surprisingly difficult because one has to be careful about the limits.

Transcription:

Browa Moo Sochasc Calculus Xogzh Che Uversy of Hawa a Maoa earme of Mahemacs Seember, 8 Absrac Ths oe s abou oob decomoso he bascs of Suare egrable margales Coes oob-meyer ecomoso Suare Iegrable Margales 4 Browa Moo Sochasc Calculus Couou Tme Submargales Usually s su ce o oly dscuss submargales by symmery de o echues are he same. oob-meyer ecomoso oob-meyer decomoso clears he obsable for de g sochasc egral ( he somery sraegy) wr suare egrable margales hece s of foudameal morace e o A creasg rocess A s called aural f for every bouded, rgh couous margale fm ; F : < g we have E M s da s E M s da s () (;] (;] for every < < Problem Suose X fx ; F : < g s a rgh couous submargale. Show ha uder ay oe of he followg codos, X s of class L a) X ;a.s for every b) X has he secal form suggesed by oob decomoso. X M + A ; Show also ha f X s a uformly egrable margale, he s of class.

Proof. By ooal samlg heorem for bouded sog mes ( oe 4), we have X T fx T >g Also we have P (X T > ) E (X T ) for all a > ; > ; T I a : Therefore X a fx T >g E (X a) su X T dp T I a fx T >g For he secod ar. I su ces o show ha fm g he decomoso s uformly egrable (sce fa g s uformly egrable for all T I a ). Aga from ooal samlg heorem X T E (X a jf T ) for all T I a whch esablshed he eeded uformly egrably. If X s uformly egrable, he X closes X hece X s Problem 3 Le X fx ; F : < g be a couous, o-egave rocess wh X a.s. A fa ; F : < g ay couous, creasg rocess for whch E (X T ) E (A T ) holds for every bouded sog me T of ff g. Iroduce he rocess V max s X s cosder a couous, creasg fuco F o [; ) wh F () de e for < x < : Esablsh he euales for ay sog me T of ff g Proof. e e he sog mes G (x) F (x) + x x u df (u) P [V T ] E (A T ) ; 8 > () P [V T ; A T < ] E ( ^ A T ) ; 8; > (3) E (F (V T )) E (G (A T )) (4) H f f : X g ; S f f : A g (5) T T ^ ^ H

(Noce ha H T f V T < s o rue for arbrary e T;.e., f fv T g for some e T he H T ; oherwse, he lef sde s zero sce fv T g s emy for e T he ualy holds aurally) we have P (V T ) E X T fvt g E (XT ) E (A T ) E (A T ) Now by he couy of robably measure we ake a o boh sdes P (V T ) P (V T ^H ) E (A T ) (sce (V T ^H ) (V T ) sce V max s X s moooc o-decreasg o : ) O he oher h, we have P (V T ; A T < ) P (V T ^S ) E (A T ^S) E ( ^ A T ) (6) (sce A T < mles T S ) The we have P (V T ) P (V T ; A T < ) + P (V T ; A T ) E ( ^ A T ) + P (A T ) whch s he rs eualy he corollary ha follows eoe he cdf of V T by F VT (V T ) : By he assumo o F we have E (F (V T )) F (x) F (V T ) df VT (V T ) fxug df (u) fvt ugdf (u) df VT (V T ) (7) df (u) fvt ugdf VT (V T ) P (V T u) df (u) E (u ^ AT ) + P (A T u) df (u) u ( ) E AT fat <ug + P (A T u) df (u) u E F (A T ) + A T df (u) E (G (A T )) A T u Corollary 4 from he above roblem, we have P [V T ] E ( ^ A T ) + P [A T ] E (V T ) E (A T ) ; < < 3

Proof. Take F (x) x for < < x : The g (x) x + x u u du x + x u du x x x + x u x x + x x x + x x from he above eualy as desred. E (V T ) E (A T ) Suare Iegrable Margales Oe hg o oce s ha, whe suarg sums of margale cremes akg he execao, o ca eglece h cross-roduc erms. More recsely, f M fm g M s < u < v; he E ((M v M u ) (M M s )) E fe ((M v M u ) (M M s ) jf u )g (8) E f(m M s ) E (M v M u jf u )g Ths fac aled o boh M; M hm we ge E (M v M u ) jf E Mv M u M v + Mu jf E Mv jf E (M u E (M v jf u ) jf ) + E MujF E Mv jf E MujF + E M u jf E M v M ujf E (hmv hm u jf ) (9) sce E (M u M v jf ) E (E ((M u M v ) jf u ) jf ) E ([M u E (M v jf u )] jf ) E M ujf for margale M Thus M v hm v M u hm u (Mv M u ) (hm v hm u ) have zero execaos codoed o F Problem 5 Le fx ; F g be a couous rocess wh he roery ha for each xed > for some > () L V () kk! robably, where L s a rom varable akg values [; ) a.s.. Show ha for > ; kk! V () () robably kk! V () () for < < o fl > g Proof. If L a.s., he we have ohg o rve. So s reasoable o suose P (L > ) > () Le X (; ) su js sj<;s;s [;] fjx s X s j g 4

The by he uform couy of X s o he comac se [; ] ; we have! js for ay > : For ay aro su sj<;s;s [;] fjx s X s j g o () [; ] : () () () + () m le k k () max m + () o The s clear ha V () ( ) Xm X () + m X X () + X () X () Xm! X () + su m X X () () + X () + X () X () Coseuely, f > ; he V () ( ) k k! robably. For he secod half, le s assume he corary. The here are cosas ; K > a seuece of aros o () [; ] : () () () + () m such ha for he ses holds ha Thus from s fered ha whch coradcs V () ( ) Xm X () + m X A X () + K su m L > ; V () o ( ) K P (A ) P (L > ) X () X X () () + Xm! X () + su m X () X X () () + X () + X () P V () ( ) P (L > ) > k k! P V () ( ) L > < for all whose kk < () for some () for ay gve ; > : X () Problem 6 Le X be M c ; T be a sog me of ff g. If hx T a.s. he P (X T ^ ; 8 < ) 5

Proof. Sce hx s couous odecreasg ^ T we have P (X ^T ; < ) Sce M X hx s a couous margale, by he ooal samlg heorem E (M T ^ ) E (M ^T ) E X^T E (hx ^T ) E X^T whch mles robably for all < : Coseuely For ay [; ); by couy we have Thus as desred. X ^T P X ^T : Q [;) P @ \ fx r^t ga r Q [;) [ P (X r^t 6 ) r Q [;) r ;k #;r ;k Q [;) X ;rk^t X ^T P (X ^T : [; )) P X ;rk^t X ^T : [; ) P @ \ fx r^t ga r ;k #;r ;k Q [;) r Q [;) Problem 7 Show ha for X; Y M c f ; ; ; m g a aro of [; ] robably. kk! k (X k X k ) (Y k Y k ) hx; Y Proof. Le s ake a smarer sraegy wh he followg roadma: by he de o hx; Y (hx + Y hx Y ) 4 he fac ha m X+Y (; ) X [(X k X k ) + (Y k Y k )]! P hx + Y ; V () V () k m X Y (; ) X [(X k X k ) (Y k Y k )]! P hx Y k 6

as kk! ; we have oly o show m V () X;Y (; ) X (X k X k ) (Y k Y k )! P 4 V () k X+Y (; k! ) k Noce ha he above scheme he aros used may be very d ere. The Obvously, for he same aro ; V () X;Y (; ) 4 V () X;Y (; )!P hx; Y V () X+Y () (; ) V X Y (; ) V () k X k! Y ; whch yelds he resul. To make hs more formal, we showed ha, sce for ay ; > ; here exss some > ; such ha o V () max P X+Y (; ) hx + Y V () ; P X Y (; ) hx Y < for every wh kk < ; ha, for all such ; holds ha V () X+Y V () X;Y (; ) 4 () (; ) V X Y (; ) he as desred. P V () X;Y (; ) hx; Y < Remark 8 We do have o coe wh he mess by de g k X X k X k ; k Y Y k Y k ; k hx; Y hx; Y k hx; Y k show comue E! (X k X k ) (Y k Y k ) hx; Y k E E! (X k X k ) (Y k Y k ) hx; Y E k k +E @ su km ll we re bored [( k X k Y ) k hx; Y ] k o! ( k X k Y ) + ( k hx; Y ) + k X k Y k hx; Y j+ fj k X jg (( X Y ) hx; Y ) (( j X j Y ) j hx; Y ) A su km fj k Y jg E k! k X k Y + su fj k hx; Y jg E km!! ( k hx; Y ) Problem 9 Le X; Y be M c;loc : The here s a sochascally uue adaed, couous rocess of bouded varao hx; Y sasfyg hx; Y a.s., such ha XY hx; Y M c;loc. If X Y we wre hx hx; X hs rocess odecreasg. k 7

Proof. Ths roblem d ers from he orgal heorem he exbook ha here he rocessed are oly local margales. Thus s more d cul bu su ces o show X M c;loc ) X hx M c;loc The we ca ake XY hx; Y (X + Y ) hx + Y 4 o (X Y ) hx Y whch s a lear combao of local margales coseuely a local margale. So, le s de e seueces of sog mes fs g ; ft g such ha S ; T. The o o X () X ^S ; Y () Y ^T are ff g-margales (by he ooal samlg heorem). e e O f f : (jx j _ jy j) g se Noe ha R a.s.. Sce ~X () R S ^ T ^ O X ^R ; ~ Y () Y ^R ~X () X ^R X () ^R ; ~ Y () Y ^R Y () ^R hese rocesses are also ff g-margales are M c because hey re bouded. For m > ; ~ X () X (m) ^R (key observao) so ~X () E ~X (m) ^R X (m) ^R ~X (m) E ^R s a margale. Ths mles ~X () E by oob ecomoso. Thus we ca de e E hx ~X () ~X (m) E ^R () whever R be assured ha hx s well-de ed. The rocess hx s adaed, couous, odecreasg sa es hx a.s.. Furhermore E (sce hx ^R ~X () ^R X ^R hx ^R ~X (m)e By Theorem 5.3 he ex, we ca ake ^R ~X ()E ~X () ~X () E hx; Y (hx + Y hx Y ) 4 () ) s a margale for each ; so X hx M c;loc : 8

The XY hx; Y M c;loc For he uueess, suose boh A; B sasfy he codos reures of hx; Y. The M XY A B XY B are M c;loc ; so jus as before we ca cosruc a seuece fr g of sog mes wh R such ha M () M ^R N () N ^R are M c : Coseuely M () N () B ^R A ^R M c beg of bouded varao hs rocess mus be decally zero. (see he roof of Theorem 5.3) I follows ha A B Problem Esablsh he followg ) A local margale of class L s a margale ) A oegave lcoal margale s a suermargale ) If M M c;loc S s a sog me of ff g ; he E M S E (hms ) ; where M! M Proof. ) By hyohess, here exss a seuece of sog mes T T + a.s. such ha for each T he soed rocess fx ^T ; F g s a margale. Sce he seuece f ^ T : g s bouded by ; by L roery E (X ^T jf s ) X s^t mles a.s. whch jus ed he margale roery. ) Sce E (X ^T jf s ) E (X jf s ) X s^t X s E (X ^T jf s ) X s^t a.s. for all s for each T ; he (because he s exss a.s.) X s^t X s E (X ^T jf s ) E (X ^T jf s ) by Faou s lemma Lev s lemma, for ay A F s E (X ^T jf s ) dp E (X ^T jf s ) dp Sce A F s s arbrary, he Coseuely (Acually fx g s a margale) ) For A A A X ^T dp E (X ^T jf s ) E (X jf s ) X s E (X jf s ) E (X ^T jf s ) X s^t a.s. for all s for each such T : S s a sog me Esseally, esablsh (locally) uformly egrably E M S E (hms ) A X dp A X ^T dp 9

Remark Local margale: for ay s T ; we have ^ T ; s ^ T s E (X jf s ) X s For > T we have So deed s a margale locally. E (X T jf s ) X s Problem Le M fm ; F g M [ M c;loc assume ha s uadrac varao rocess hm s egrable: E (hm ) < : The ) M s a margale, M M are boh uformly egrable; arcular, M M exsa a.s. E M E (hm ) ) we may ake a rgh-couous mod cao of E MjF M ; ; whch s a oeal. Proof. Esseally, esablsh (locally) uformly egrably use mod cao. If M M ; he E M E (hm ) E (hm ) : If M M c;loc ; he E MS E (hms ) E (hm ) : Coseuely fm S g SI uformly egrable (by Holder s Ieualy). Thus M M a.s E (M jf ) M a.s. by Faou s lemma E M E M E M E (hm ) (3) Jesse s eualy yelds M E M jf If follows ha M has a las eleme, e., ha M ; F : egrable (3) holds wh eualy. Leg E MjF s a submargale, whch s uformly gves he reured rocess M Problem 3 Le M M c;loc show ha for ay sog me T of ff g ; P max jm j E ( ^ hm T ) T + P [hm T ] for ay ; > : I arcular, for a seuece M () submargale M ()E! P ) max T T M c;loc we have M () P! P Proof. Take X M A hm use he roblem oe 4. P max jm j P max M T T The secod follows readly by relacg. E ( ^ hm T ) + P [hm T ]