O QP P. Limit Theorems. p and to see if it will be less than a pre-assigned number,. p n

Similar documents
Limit Theorems. Convergence in Probability. Let X be the number of heads observed in n tosses. Then, E[X] = np and Var[X] = np(1-p).

Continuous-Time Fourier Transform. Transform. Transform. Transform. Transform. Transform. Definition The CTFT of a continuoustime

Bohr type models of the atom give a totally incorrect picture of the atom and are of only historical significance.

Topic 5:Discrete-Time Fourier Transform (DTFT)

Probability & Statistics,

H2 Mathematics Arithmetic & Geometric Series ( )

Chapter Taylor Theorem Revisited

International Journal of Mathematical Archive-7(5), 2016, Available online through ISSN

Some Families of Higher Order Three-Step Iterative Techniques. where is a real number and y (5)

Chapter (8) Estimation and Confedence Intervals Examples

Discrete Fourier Transform. Discrete Fourier Transform. Discrete Fourier Transform. Discrete Fourier Transform. Discrete Fourier Transform

DTFT Properties. Example - Determine the DTFT Y ( e ) of n. Let. We can therefore write. From Table 3.1, the DTFT of x[n] is given by 1

6. Negative Feedback in Single- Transistor Circuits

1985 AP Calculus BC: Section I

A Simple Proof that e is Irrational

APPENDIX: STATISTICAL TOOLS

PREPARATORY MATHEMATICS FOR ENGINEERS

MONTGOMERY COLLEGE Department of Mathematics Rockville Campus. 6x dx a. b. cos 2x dx ( ) 7. arctan x dx e. cos 2x dx. 2 cos3x dx

Problem Value Score Earned No/Wrong Rec -3 Total

ENGI 4421 Central Limit Theorem Page Central Limit Theorem [Navidi, section 4.11; Devore sections ]

Option 3. b) xe dx = and therefore the series is convergent. 12 a) Divergent b) Convergent Proof 15 For. p = 1 1so the series diverges.

Chapter 3.1: Polynomial Functions

10. Joint Moments and Joint Characteristic Functions

8(4 m0) ( θ ) ( ) Solutions for HW 8. Chapter 25. Conceptual Questions

Worksheet: Taylor Series, Lagrange Error Bound ilearnmath.net

SAFE OPERATION OF TUBULAR (PFR) ADIABATIC REACTORS. FIGURE 1: Temperature as a function of space time in an adiabatic PFR with exothermic reaction.

Lectur 22. RF and Microwave Circuit Design Γ-Plane and Smith Chart Analysis. ECE 303 Fall 2005 Farhan Rana Cornell University

Law of large numbers

Quantum Mechanics for Scientists and Engineers. David Miller

Chapter 2 Infinite Series Page 1 of 11. Chapter 2 : Infinite Series

ln x = n e = 20 (nearest integer)

Review Exercises. 1. Evaluate using the definition of the definite integral as a Riemann Sum. Does the answer represent an area? 2

Acid Base Reactions. Acid Base Reactions. Acid Base Reactions. Chemical Reactions and Equations. Chemical Reactions and Equations

ALOHA Product no.: 03007

Partition Functions and Ideal Gases

Outline. Heat Exchangers. Heat Exchangers. Compact Heat Exchangers. Compact Heat Exchangers II. Heat Exchangers April 18, ME 375 Heat Transfer 1

Markov processes and the Kolmogorov equations

MATH Midterm Examination Victor Matveev October 26, 2016

. This is made to keep the kinetic energy at outlet a minimum.

PURE MATHEMATICS A-LEVEL PAPER 1

Modern Physics. Unit 5: Schrödinger s Equation and the Hydrogen Atom Lecture 5.6: Energy Eigenvalues of Schrödinger s Equation for the Hydrogen Atom

07 - SEQUENCES AND SERIES Page 1 ( Answers at he end of all questions ) b, z = n

LECTURE 5 Guassian Wave Packet

Solutions. Definitions pertaining to solutions

ENGI 4421 Central Limit Theorem Page Central Limit Theorem [Navidi, section 4.11; Devore sections ]

Chapter 33 Gauss s Law

GUC (Dr. Hany Hammad) 4/20/2016

Discrete Fourier Transform (DFT)

The Matrix Exponential

Math 34A. Final Review

Session : Plasmas in Equilibrium

European Business Confidence Survey December 2012 Positive expectations for 2013

Another Explanation of the Cosmological Redshift. April 6, 2010.

Digital Signal Processing, Fall 2006

A Study on Estimation of Lifetime Distribution with Covariates Under Misspecification

z 1+ 3 z = Π n =1 z f() z = n e - z = ( 1-z) e z e n z z 1- n = ( 1-z/2) 1+ 2n z e 2n e n -1 ( 1-z )/2 e 2n-1 1-2n -1 1 () z

Derivation of a Predictor of Combination #1 and the MSE for a Predictor of a Position in Two Stage Sampling with Response Error.

Even/Odd Mode Analysis of the Wilkinson Divider

, the random variable. and a sample size over the y-values 0:1:10.

Topic 5: Discrete-Time Fourier Transform (DTFT)

The Matrix Exponential

They must have different numbers of electrons orbiting their nuclei. They must have the same number of neutrons in their nuclei.

The Language of SOCIAL MEDIA. Christine Dugan

Statistics 3858 : Likelihood Ratio for Exponential Distribution

Part B: Transform Methods. Professor E. Ambikairajah UNSW, Australia

Ch. 1 Introduction to Estimation 1/15

Chapter Five. More Dimensions. is simply the set of all ordered n-tuples of real numbers x = ( x 1

High Energy Physics. Lecture 5 The Passage of Particles through Matter

Thomas J. Osler. 1. INTRODUCTION. This paper gives another proof for the remarkable simple

Electrical field generated by a charged harmonic oscillator at thermodynamic equilibrium

DFT: Discrete Fourier Transform

(Reference: sections in Silberberg 5 th ed.)

Mount Vernon Charles Station Potential Visibility Assessment. December 4, 2017 Chesapeake Conservancy Annapolis, MD 21401

9.5 Complex variables

Lecture 26: Quadrature (90º) Hybrid.

120~~60 o D 12~0 1500~30O, 15~30 150~30. ..,u 270,,,, ~"~"-4-~qno 240 2~o 300 v 240 ~70O 300

- Prefixes 'mono', 'uni', 'bi' and 'du' - Why are there no asprins in the jungle? Because the parrots ate them all.

A L A BA M A L A W R E V IE W

Physics 43 HW #9 Chapter 40 Key

EE 119 Homework 6 Solution

SIMPLE METHOD FOR THE SOLUTION OF INCOMPRESSIBLE FLOWS ON NON-STAGGERED GRIDS

are given in the table below. t (hours)

CDS 101: Lecture 5.1 Reachability and State Space Feedback

Time : 1 hr. Test Paper 08 Date 04/01/15 Batch - R Marks : 120

Review for cumulative test

INTRODUCTION TO SAMPLING DISTRIBUTIONS

ELEC9721: Digital Signal Processing Theory and Applications

Search sequence databases 3 10/25/2016

Helping you learn to save. Pigby s tips and tricks

MATHEMATICS 9740/01 Paper 1 14 Sep hours

The Phase Probability for Some Excited Binomial States

Exponential Functions

ANOVA- Analyisis of Variance

y = 2xe x + x 2 e x at (0, 3). solution: Since y is implicitly related to x we have to use implicit differentiation: 3 6y = 0 y = 1 2 x ln(b) ln(b)

Chemical Physics II. More Stat. Thermo Kinetics Protein Folding...

Lectures 9 IIR Systems: First Order System

BACKFILLED 6" MIN TRENCH BOTT OF CONT FTG PROVIDE SLEEVE 1" CLR AROUND PIPE - TYP BOTTOM OF TRENCH PIPE SHALL NOT EXTEND BELOW THIS LINE

On the approximation of the constant of Napier

Engineering 323 Beautiful HW #13 Page 1 of 6 Brown Problem 5-12

AP Statistics Notes Unit Eight: Introduction to Inference

Transcription:

Limit Trms W ft av t d fr apprximatis w gts vry larg. Fr xampl, smtims fidig t prbability distributis f radm variabls lad t utractabl matmatical prblms. W ca s tat t distributi fr crtai fuctis f a radm variabl ca asily b apprximatd fr larg valus f, v tug t xact distributi s difficult t btai. Cvrgc i Prbability Supps a ci as prbability f cmig up ads a sigl tss, ad if w tss it tims w wdr abut t fracti f ads tat sws up. Ituitivly, w wuld xpct tat t largr is, t cls t p t fracti bcms. Tis is t always s. Tis lads t t qusti "wat apps t t radm diffrc btw a stimat ad its targt valu?" Lt X b t umbr f ads bsrvd i tsss. T, E[X] = p ad Var[X] = p(-p). O way t masur t clsss f x/ t p is t xami x p ad t s if it will b lss ta a pr-assigd umbr,. L NM O QP P Tis P x x p p suld b cls t uity fr larg if ur ituiti is crrct. Dfiiti: T squc f radm variabls, x, x, x3,..x, is said t "Cvrg i Prbability" t t cstat c if fr vry psitiv umbr, lim P[ x c ] = Trm Lt X, X,. X b idpdt ad idtically distributd radm variabl wit E[X I ] = µ ad Var[X i ] = s < i. i i= Furtr, lt X = X lim P[ X µ ] = 0 lim P[ X µ < ] =. T fr ay psitiv ral umbr Tat is, X cvrgs i prbability t µ.. Tis is calld t "Wak law f larg umbrs" Atr trm usig t abv ccpts is Trm Supps X cvrgs i prbability tµ ad Y cvrgs t µ. T: () X ad Y cvrg i prbability t µ ad µ. () X Y cvrgs i prbability t µ µ (3) X /Y cvrgs i prbability t µ /µ ; µ 0 (4) X cvrgs i prbability t µ prvidd tat P[X 0]=

Exampl; Exampl: Supps X, X,. X ar idpdt ad idtically distributd radm variabls wit E[X i ] = µ, E[X i ] = µ ', E[X 3 i ] = µ ' 3, E[X 4 i ] = µ ' 4, wit all fiit Lt S ' b t sampl variac, c, ' S = ( Xi X ) i= Sw S ' cvrgs i prbability t Var(X i ) Sluti: ' S = ( Xi X ) i= X = ( Xi X Xi + X ) = Xi Xi + X Wr i= i= i= i= = Xi i= X = i= X X i ' Lkig at trms i S, X i is t avrag f idpdt ad idtically i= distributd radm variabls f t frm X wit E[ X ] = µ ', ad Var[ X ] = µ ' ( µ ' ) X. i i i X i i= Sic Var( X i ) is fiit, t 4 cvrgs i prbability t µ (Try tlls us tat X cvrgs i prbability t µ). Sic X i ad X cvrg i i= prbability t ' ' ' ' µ, ad µ it fllws tat S = Xi X cvrgs i prbability t µ µ = Var( Xi ) i= Tis sws tat fr larg sampls. T sampl variac suld b cls t ppulati variac wit ig prbability. If w ar ccrd wit wat apps t t prbability distributi f crtai typs f radm variabls as w d t fllwig dfiiti. Dfiiti: If w lt Y b a radm variabl wit a distributi fucti, F (x) ad w lt Y b a radm variabl wit distributi fucti F(y), t if lim F ( y) F( y) at vry pit y fr wic F(y) is ctiuus t Y is said t "cvrg i distributi t Y. F(y) is calld t limitig distributi fucti f y.

Lt X, X, X b idpdt radm variabls, uifrmly distributd vr t itrval (0, θ ) fr a psitiv θ. Als. Lt y = max(x,x, X ) Fid t limitig distributi f Y. Sluti: F ( y) = P( X y) = X i R S T 0 y θ Frm rdrd statistics w kw y 0 0 < y < θ y θ G( y) = P[ Y y] = F ( y), wr F X (y) is t distributi fucti fr ac X. R T 0 y T lim G( y) = Slim( ) θ X y 0 0 < y > θ y θ Tus, Y cvrgs i distributi t a radm variabl tat as a prbability f at t pit θ ad zr lswr. It is ft asir t fid limitig distributis wrkig wit mmt gratig fuctis. Trm: Lt Y ad Y b radm variabls wit mmt gratig fucti M (t) ad M(t), rspctivly. If lim M = M fr a ll ral t, t Y cvrgs i distributi t Y. Exampl; ` A xprimtr wats t cut t umbr f bactria/small vlum f watr. T sampl siz is t vlum f watr i wr cut is mad. Fr purpss f apprximati, t prbability distributi f cuts w tik f t vlum, ad c t avrag cut pr vlum, as quatitis gttig larg. Lt X dt t bactria cut pr cc f watr, ad assum tat X as a Piss prbability dsity fucti, wit ma valu f. W wat t apprximat t prbability distributi f X fr larg valus f.. W X d tis by swig ;Y = cvrgs i distributi t a stadard rmal radm variabl as. Spcifically, if t allwabl plluti i a watr supply is a cut f 0/cc apprximat t prbability tat X will b at mst 0, assumig = 00. Sluti: W tak t limit f t MGF f Y as ad us ur trm. t ( ) M = ad c, t MGF fr Y is M X X t t t t / = M X( ) = xp[ ( )]

3 / t t T trm ( ) =... 3/ + + 6 + Ad up addig xpts, 3 t t t M Y = xp[ t + ( + + +...] / 3 6 3 t t = xp[ + +...] / 6 W s tat t first trm is fr f ad t trs av raisd t sm pwr i t dmiatr.. Trfr, lim M Y = t Tis is t MGF fr a stadard rmal fucti. W wat F HG P( X ) = P X 0 0. X W av sw tat Y = is apprximatly a stadard rmal radm variabl fr larg valus f. S, fr. = 00 F 0 00I P Y P Y HG K J = ( ) =. 843 0 T rmal apprximati t Piss prbability wrks rasably wll fr gratr ta 5 I K J Piss Prcss Additial Tpics i Prbability Lt us csidr t applicati f Prbability try t mdlig radm pm tat cag wit tim. Tis is t s-calld STOCHASTIC PROCESS. Tus, a tim paramtr is itrducd (wr "tim" is usd i a vry gral ss) ad t radm variabl, Y(t), is rgardd as a fucti f tim. Exampls ar dlss, but Y(t) culd rprst t siz f a bilgical ppulati at tim t; t cst f pratig a cmplx idustrial systm fr t tim uits; t umbr f custmrs waitig t b srvd at a cck-ut cutr t tim uits aftr pig.

A stcastic prcss, Y(t), is said t av "idpdt" icrmts if, fr ay st f tim pits, t 0 <t < <t, t radm variabl [Y(t i )-Y( ti- )] ad [Y(t j )-Y(t j- )] ar idpdt fr i j. T prcss as "statiary" idpdt icrmts if, i additi, t radm variabls [Y(t + )-Y(t + )] ad [Y(t )-Y(t )] av idtical distributis fr ay >0. I makig a matmatical mdl fr a ral wrld pm it is always cssary t mak crtai simplifyig assumptis s as t rdr t matmatics tractabl. O t tr ad, w cat mak t may simplifyig assumptis fr t ur cclusis wuld t b applicabl. O assumpti usually mad is t assum tat crtai radm variabls ar xptially distributd. T ras fr tis ids tat t xptial distributi is bt rlativly asy t wrk wit ad is ft gd apprximati t t actual distributi. T prprty f t xptial distributi tat maks it asy t aalyz is tat it ds t dtrirat wit tim. I.. if liftim f a itm is xptially distributd, t a itm wic as b usd fr t (ay umbr) urs is as gd as a w itm is, rgardlss t t amut f tim rmaiig util failur. Tis is als rfrrd t as t xptial distributi NOT avig a mmry, r mmrylss. T xptial distributi is t ly distributi tat psssss tis prprty. A ctiuus radm variabl is said t av a xptial distributi wit paramtr, >0, if x R x 0 f ( x) = S 0 x < 0 r T zx x F(x) = dx = R S T 0 x x 0 x < 0 T; tx E[ X ] = MGF = E[ ] = t < t E[ X ] = Var[ X ] = A radm variabl is mmrylss if P[X>s+t X>t] = P[X>s} s,t 0. Tis ca als b writt as P[X > s + t, X > t] = p[ X > s] P[ X > t] P[X > s + t] = P[X > s]p[x > t]

Ts quatis stat tat if t lctric cmpt (sic) is pratig at tim t, t t distributi f t rmaiig tim it survivs is t sam as rigial liftim distributi i.. T cmpt dis t rmmbr tat it as alrady b i us fr a tim t.\ Exampl: Supps tat t amut f tim spds i a bak is xptially distributd, wit a ma f t miuts, ( = /0). (a) Wat is t prbability tat a custmr will spd mr ta 5 miuts i t bak? (b) Wat is t prbability tat a custmr will spd mr ta 5 miuts giv tat still is i t bak aftr 0 miuts? Sluti: (a) Lt X = amut f tim i t bak. P[X>5]=-F(5) = -+ t/0. > = 5 / P[ X 5] 3 = = 0. 0 (b) Sic t prprty f mmrylss applis, giv tat spds 0 miuts i t bak, w d ly 5 miuts mr. > = 5 / P[ X 5] = =. 604 Furtr prprtis f t xptial distributi ar: () If X, X,. X ar idpdt ad idtically distributd radm variabls wit a ma valu f /, t t ( t) f X X X t.. ( ) = (Gamma distributi) ( )! () P( X < X ) = + A Stcastic prcss {N(t), t 0} is a cutig prcss if N(t) rprsts t ttal umbr f vts tat av ccurrd up t tim t. Fr xampl, a) Numbr f ppl trig a str prir t tim t. b) Ttal umbr f ppl br by tim t. c) Numbr f gals sccr playrs scr by tim t. S, t b a cutig prcss, i) N(t) 0 ii) N(t) is itgr valud iii) Fr s < t, t N(s) < N(t) iv) Fr s<t, N(t) - N(s) quals umbr f vts tat av ccurrd i t itrval (s,t)

A cutig prcss as idpdt icrmt if t umbr f vts tat ccur i disjit itrvals ar idpdt. I.. Evts ccurrd by tim 0 (tat is, N(0)) must b idpdt f vts ccurs btw 0 ad 5 [N(5) - N(0)]. A cutig prcss as statiary icrmts if t distributi f t umbr f vts wic ccur i ay itrval f tim dpds ly t lgt f t tim itrval, ad t t spcific tim. i.. [N(t +s) - N(t +s)] as t sam distributi as [N(t ) - N(t )], fr all t < t, s >0. O f t mst imprtat cutig Prcsss is t Piss Prcss. Dfiiti: T cutig prcss {N(t),t 0} is said t b a Piss prcss avig t rat, >0, if ) N(0) = 0 ) Prcss as idpdt icrmts 3) T umbr f vts i ay itrval, t, is Piss distributd, wit ma t. Tat is, fr all s,t 0 - t ( t) P[N(t + s) - N(s) = } = = 0,,,3,4..! Nt: Frm 3), a Piss prcss as statiary icrmts ad E[N(t)]=t -- wic xplais wy is calld t rat f t prcss I rdr t dtrmi if a arbitrary cutig prcss is actually. a Piss prcss w must sw tat ), ) ad 3) ar satisfid. Cditi ) is simply tat t cutig bgis at t=0. Cditi ) ca b dirctly vrifid by kwldg f t prcss. Hwvr, 3) is t clar as t w w culd dtrmi t cditi as satisfid. S. T lp dtrmi tis, a quivalt dfiiti is usful. As a prlud, lt us dfi t ccpt f a fucti, f( ) big (). Dfiiti: T fucti is said t b () if f lim ( ) = 0 Fr xampl, f() () f = x is () sic lim = lim = 0 0 0 () f = x is t () sic lim = lim = 0 0 0 (3) If f( ) is () ad g( ) is () t f( ) + g( ) is (), sic f() + g() f() g() lim = lim + lim = 0+ 0 = 0 0 0

cf() (4) If f( ) is () t s is g( ) = cf( ), sic lim = c lim f() = c 0 = 0 0 Frm 3) ad 4) it fllws tat ay fiit liar cmbiati f fuctis, ac f wic is ()m is als (). I rdr fr t fucti f( ) t b () it is cssary tat f()/ g t zr as gs t zr. But if gs t zr, t ly way fr f() t g t zr is fr f() t g t zr fastr ta ds. Tat is, fr small, f() must b small cmpard t. w ar w i a psiti t giv a altrativ dfiiti f a Piss prcss. Dfiiti: T cutig prcss {N(t),t 0} is said t b a Piss prcss avig rat, > 0, if i.) N(0) =0 ii.) Prcss as statiary ad idpdt icrmts iii.) P{N()=}= + () iv.) P{N() =() Examiig ts cditi, iii) stats t prbability f ccurrc i small tim itrval,, is prprtial t. Cditi iv) stats, prbability fmr ta ccurrc i gs t zr fastr ta itslf. Ts tw axims imply P{N(t+) - N(t)]=0}=--()= P[N()=0]= - - () T tw dfiitis ar ttally quivalt. Lt us dfi P ( t ) = P ( N ( t ) = ), ad w will driv a diffrtial quati fr P (t) as fllws: P ( t + ) = P{ N ( t + ) = 0} = P{ N = 0, N ( t + ) N = 0} = P{ N = 0} P{ N ( t + ) N = 0} = P [ + ( )] ( t + ) t) ( ) Hc, P P( = P + Lttig 0, i t limit w gt P' (t)=-p (t), r P ' =, wic, by itgrati, implis P l P = t + c r P = k Sic P ( 0) = P{ N ( 0) = 0} = (crtai vt frm i)) w arriv at P = t t

Similarly, fr >0 P ( t + ) = P[ N( t + ) = ] = P{ N =,[ N ( t + ) N = 0]} = P{ N =,[ N ( t + ) N = 0]} + P{ N =,[ ( t + ) N = ]} + ki= P{ N = k,[ N ( t + ) N = k]} Hwvr, frm iv) abv, t last trm is (); P ( t + ) = P P + P P ( ) + ( ) But, S, P ( ) = P{ N ( ) = 0} = + ( ) P ( ) = P{ N( ) = } = + ( ) P ( t + ) = ( ) P + P + ( ) Tus, P ( t + ) P = P + P + Lttig 0 yilds ' P = P + P r quivaltly, t ' t [ P + P ] = P t Usig P =, ( ) d t t t ( P ) = = dt r P ( t ) = ( t + c ) t Sic w av frm I), N(0)=0, t N=0 givs P =0, ad c=0 P ( t ) = t t Usig matmatical iducti, assumig t sluti fr =, t d t t ( P ) = P dt t t = ( )( t ) P = M t ( t) P =! Sic P ( 0) = 0 = t t t

Exampl: Supps tat i a crtai ara plats ar radmly disprsd wit a ma dsity f 0/squar yard. If a bilgist radmly lcats 00 -squar yard samplig quadrats t ara, w may f tm ca b xpctd t ctai plats? Sluti: Assumig t plat cuts/uit ara t av a Piss distributi wit a ma f 0/sq. yd., t prbability f plats i a squar yard ara is: =0 ad P[N(t=)=0]= t ( t) = = = 0! 0 0 40 If t 00 quadrats ar radmly lcatd, t xpctd umbr f quadras ctaiig zr plats is 40 00P[ Y( ) = 0] = 00 = 4. 5x0-6 Exampl: Custmrs arriv at a bak at a Piss rat. Supps tw custmrs arrivd durig t first ur. Wat is t prbability tat (a) bt arrivd durig t first 0 miuts (b) at last arrivd durig t first 0 miuts. Sluti: Lt s= fracti f r. (i.. s=/ mas 30 miuts) Giv = /r -,a rat f ccurrc Piss wit =, t=s Piss wit =, t=-s (a) P[ arrivalsi itrval (0, s) arrivalsi itrval (0,)] = P[ i (0, s),0 i (s,) P[ i (0,) Piss wit =, t= L = ( ) N M = s = 9! OL QP NM s ( s) ( s) ( [ s])! 0! ad w s = 3 0 O QP (0 miuts)

Exampl: (b) At last i first 0 miuts = at last i first s= /3 P[At last i first /3 ] = - P[Bt i last s=/3 (40 miuts)] Sic mmry, P[arrivals i last 40 miuts] is t sam as P[arrivals i first 40 miuts]. Hc, P[at last arrival i first 0 miuts]= -s = (-(/3) ) = 5/9 Cars crss a crtai pit i t igway i accrdac wit a Piss prcss wit rat = 3 pr miut. If Al blidly rus acrss t igway, t wat is t prbability tat will b ijurd if t amut f tim tat it taks im t crss t rad is s scds (assum tat if is t igway w t car passs, t will b ijurd)? D it fr s=,5,0,0. Sluti; = 3, ad s= scds t igway. t P t! ( ) = s = 0! 3 0 3 60 Hc, s/ 0 s/ 0 s/ 0 s/ 0 = s/ 0 ( / ) (fr s = ) =.90 (fr s = 5) =.779 (fr s = 0) =.607 (fr s = 0) =.368 Summary f Piss Prcsss Ctral t t dvlpmt is t ccpt f (). Rcall tat a fucti is said t b () if f ( ) / 0. Tat is, f is )) if, fr small valus f, f() is small v i rlati lim 0 = t. Supps w tat "vts" ar ccurrig at radm tim pits ad lt N)t) dt t umbr f vts tat ccur i t tim itrval [0,t]. T stcastic prcss {N(t),t 0} is a=said t b a Piss prcss avig t=rat, >0 if (i) N(0)=0.

(ii) T umbr f vts tat ccur i disjit tim itrvals ar idpdt (iii) T distributi f t umbr f vts tat ccur i a giv itrval dpds ly t lgt f tat itrval ad t its lcati. (iv) P { N( ) = } = t = ( ) (v) P { N( ) } = ( ). Tus, cditi (I) stats tat t prcss bgis at tim 0. Cditi (ii), t idpdt icrmt assumpti, stats fr istac, tat t umbr f vts by tim t ]tat is, N(T(] is idpdt f r umbr f vys tat ccur btw t ad t+s [tat is, N(t+s) - N(t)]. Cditi (iii), t statiary icrmt assumpti, stats tat, t prbabiltiy distributi f N(t+s) - N(t) is t sam fr all valus f t. W av dvlpd t Piss distributi by a limitig vrsi f t bimial distributi. W w ca btai t=is sam rsult by a diffrt mtd. Lmma. Fr a Piss prcss wit rat t P{ N( t) = 0}= Prf: Lt P = P{ N( t) = 0}. W driv a diffrtial quati fr P (t) i t fllwig mar. P ( t + ) R S T = P{ N ( t + ) = 0} = P{ N = 0, N ( t + ) N = 0} = P{ N = 0} P{ N ( t + ) N = 0} = P [ + ( ) Wr t fial tw quatis fllw frm Assumpti (ii) plus t fact tat Assumptis (iii) ad (iv) imply tat P{N()=0}=-+(). Hc P ( t + ) P ( ) = P + ' ' PO 9t0 Nw, lttig 0, w btai P = P, r quivaltly =, wic P implis by itgrati, tat lg P (t) = -t +c, r P (t)=k -t. Sic P (0)=P{N(0)=0}=, w arriv at P{ N( t) = 0}= Fr a Piss prcss, lt us dt by T t tim f t first vt. Furtr, fr >, lt T dt t lapsd tim btw t (-)st ad t t vt. T squc f itrarrival tims. Fr istac, if T = 5 ad t

T =0, t t first vt f t Piss prcss wuld av ccurrd at tim 5 ad t scd at tim 5. W w dtrmi t distributi f t T. T d s, w first t tat t vbt{t>t} taks plac if ad ly if vts f t Piss P{ T t P N t t > } = { ( ) = 0} prcss ccur i t itrval [0,t]. ad tus, Hc T as a xptial distributi wit a ma f. Nw Hwvr P{ T > t} = P{ T > t T } P{ T > t T = s} = P{ vts i (s,s + t) T = s} 0 = P{ 0 vts i(s,s + t)} = t wr t last tw quatis fllwd frm idpdt ad statiary icrmts. Trfr, frm t abv w cclud tat T is als a xptial radm variabl wit ma /, ad furtrmr, tat T is idpdt f T. Usig tis argumt rpatdly, w ca cclud: " T, T, ar idpdt xptial radm variabls ac wit ma /" Atr quatity f itrst is S, t arrival tim f t t vt, als calld t waitig tim util t t vt. It is asily s tat S = T ad c frm prpsiti statd abv, ad t rsults btaid arlir, it fllws tat S as a gamma distributi wit paramtrs ad. Tat is, t prbability dsity f S is giv by x f ( x S x ) ( ) = ( )! W ar w rady t prv tat N(t) is a Piss radm variabl wit ma t. Trm: Fr a Piss prcss wit rat, i= i P{ N = } =! t Prf: Nt tat t t vt f t Piss prcss will ccur bfr r at tim t if, ad ly if t umbr f vts tat ccur by tim t is a last. tat is

N S t ad s P{ N = } = P{ N } P{ N + } = P{ S t} P{ S t} z t x t t ( ) t ( x) = dx dx 0 ( )! 0 ( )! z z But itrgrati by parts frmula udv = uv vdu yilds, wit u x =, dv = [( γ x) / ( )!] dx, z + zt t 0 t t ( x) t ( t) dx = + 0 ( )!! Wic cmplts t prf. z ( x) d ( )!