Probabilistic Meanings of Numerical Characteristics for Single Birth Processes

Similar documents
Complete Convergence for Weighted Sums of Arrays of Rowwise Asymptotically Almost Negative Associated Random Variables

PROJECTION PROBLEM FOR REGULAR POLYGONS

STRONG CONSISTENCY FOR SIMPLE LINEAR EV MODEL WITH v/ -MIXING

Complete Convergence and Some Maximal Inequalities for Weighted Sums of Random Variables

Research Article Multidimensional Hilbert-Type Inequalities with a Homogeneous Kernel

Q-analogue of a Linear Transformation Preserving Log-concavity

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

Journal of Mathematical Analysis and Applications

The Mathematical Appendix

Research Article Some Strong Limit Theorems for Weighted Product Sums of ρ-mixing Sequences of Random Variables

Functions of Random Variables

Eulerian numbers revisited : Slices of hypercube

Almost Sure Convergence of Pair-wise NQD Random Sequence

LINEAR RECURRENT SEQUENCES AND POWERS OF A SQUARE MATRIX

Fibonacci Identities as Binomial Sums

Extend the Borel-Cantelli Lemma to Sequences of. Non-Independent Random Variables

Assignment 7/MATH 247/Winter, 2010 Due: Friday, March 19. Powers of a square matrix

Analysis of a Repairable (n-1)-out-of-n: G System with Failure and Repair Times Arbitrarily Distributed

A New Measure of Probabilistic Entropy. and its Properties

A Markov Chain Competition Model

arxiv: v4 [math.nt] 14 Aug 2015

Strong Convergence of Weighted Averaged Approximants of Asymptotically Nonexpansive Mappings in Banach Spaces without Uniform Convexity

Marcinkiewicz strong laws for linear statistics of ρ -mixing sequences of random variables

Arithmetic Mean and Geometric Mean

STRONG CONSISTENCY OF LEAST SQUARES ESTIMATE IN MULTIPLE REGRESSION WHEN THE ERROR VARIANCE IS INFINITE

Research Article A New Iterative Method for Common Fixed Points of a Finite Family of Nonexpansive Mappings

Cubic Nonpolynomial Spline Approach to the Solution of a Second Order Two-Point Boundary Value Problem

ON THE LOGARITHMIC INTEGRAL

THE PROBABILISTIC STABILITY FOR THE GAMMA FUNCTIONAL EQUATION

Non-uniform Turán-type problems

Estimation of Stress- Strength Reliability model using finite mixture of exponential distributions

Application of Generating Functions to the Theory of Success Runs

Summary of the lecture in Biostatistics

CS286.2 Lecture 4: Dinur s Proof of the PCP Theorem

Chapter 3 Sampling For Proportions and Percentages

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

A NEW LOG-NORMAL DISTRIBUTION

Extreme Value Theory: An Introduction

Introduction to local (nonparametric) density estimation. methods

Bounds on the expected entropy and KL-divergence of sampled multinomial distributions. Brandon C. Roy

Analysis of System Performance IN2072 Chapter 5 Analysis of Non Markov Systems

Entropy ISSN by MDPI

IS 709/809: Computational Methods in IS Research. Simple Markovian Queueing Model

CHAPTER 4 RADICAL EXPRESSIONS

Part 4b Asymptotic Results for MRR2 using PRESS. Recall that the PRESS statistic is a special type of cross validation procedure (see Allen (1971))

Third handout: On the Gini Index

MARKOV CHAINS. 7. Convergence to equilibrium. Long-run proportions. Part IB Michaelmas 2009 YMS. Proof. (a) state j we have π (i) P ) = π

Bounds for the Connective Eccentric Index

AN EULER-MC LAURIN FORMULA FOR INFINITE DIMENSIONAL SPACES

About k-perfect numbers

STK4011 and STK9011 Autumn 2016

Point Estimation: definition of estimators

Chapter 5 Properties of a Random Sample

The Primitive Idempotents in

ANALYSIS ON THE NATURE OF THE BASIC EQUATIONS IN SYNERGETIC INTER-REPRESENTATION NETWORK

TESTS BASED ON MAXIMUM LIKELIHOOD

A New Method for Decision Making Based on Soft Matrix Theory

Assignment 5/MATH 247/Winter Due: Friday, February 19 in class (!) (answers will be posted right after class)

X X X E[ ] E X E X. is the ()m n where the ( i,)th. j element is the mean of the ( i,)th., then

Comparing Different Estimators of three Parameters for Transmuted Weibull Distribution

Lower Bounds of the Kirchhoff and Degree Kirchhoff Indices

arxiv:math/ v1 [math.gm] 8 Dec 2005

Class 13,14 June 17, 19, 2015

Solving Constrained Flow-Shop Scheduling. Problems with Three Machines

ρ < 1 be five real numbers. The

{ }{ ( )} (, ) = ( ) ( ) ( ) Chapter 14 Exercises in Sampling Theory. Exercise 1 (Simple random sampling): Solution:

X ε ) = 0, or equivalently, lim

DIFFERENTIAL GEOMETRIC APPROACH TO HAMILTONIAN MECHANICS

Investigation of Partially Conditional RP Model with Response Error. Ed Stanek

FORECASTING USING MARKOV CHAIN

EVALUATION OF FUNCTIONAL INTEGRALS BY MEANS OF A SERIES AND THE METHOD OF BOREL TRANSFORM

Generalized Convex Functions on Fractal Sets and Two Related Inequalities

Lecture 2 - What are component and system reliability and how it can be improved?

A Study on Generalized Generalized Quasi hyperbolic Kac Moody algebra QHGGH of rank 10

Hájek-Rényi Type Inequalities and Strong Law of Large Numbers for NOD Sequences

Chapter 14 Logistic Regression Models

arxiv: v1 [math.st] 24 Oct 2016

Mu Sequences/Series Solutions National Convention 2014

Bayes Estimator for Exponential Distribution with Extension of Jeffery Prior Information

Lecture 3 Probability review (cont d)

The internal structure of natural numbers, one method for the definition of large prime numbers, and a factorization test

UNIVERSITY OF OSLO DEPARTMENT OF ECONOMICS

On the Primitive Classes of K * KHALED S. FELALI Department of Mathematical Sciences, Umm Al-Qura University, Makkah Al-Mukarramah, Saudi Arabia

Double Dominating Energy of Some Graphs

On the construction of symmetric nonnegative matrix with prescribed Ritz values

Unimodality Tests for Global Optimization of Single Variable Functions Using Statistical Methods

#A27 INTEGERS 13 (2013) SOME WEIGHTED SUMS OF PRODUCTS OF LUCAS SEQUENCES

PERRON FROBENIUS THEOREM FOR NONNEGATIVE TENSORS K.C. CHANG, KELLY PEARSON, AND TAN ZHANG

Chapter 8. Inferences about More Than Two Population Central Values

Approximation for Collective Epidemic Model

On Submanifolds of an Almost r-paracontact Riemannian Manifold Endowed with a Quarter Symmetric Metric Connection

hp calculators HP 30S Statistics Averages and Standard Deviations Average and Standard Deviation Practice Finding Averages and Standard Deviations

Analysis of Variance with Weibull Data

02/15/04 INTERESTING FINITE AND INFINITE PRODUCTS FROM SIMPLE ALGEBRAIC IDENTITIES

CHAPTER VI Statistical Analysis of Experimental Data

CIS 800/002 The Algorithmic Foundations of Data Privacy October 13, Lecture 9. Database Update Algorithms: Multiplicative Weights

Lecture 8: Linear Regression

Chapter 4 Multiple Random Variables

18.413: Error Correcting Codes Lab March 2, Lecture 8

On Face Bimagic Labeling of Graphs

Transcription:

A^VÇÚO 32 ò 5 Ï 206 c 0 Chese Joural of Appled Probablty ad Statstcs Oct 206 Vol 32 No 5 pp 452-462 do: 03969/jss00-426820605002 Probablstc Meags of Numercal Characterstcs for Sgle Brth Processes LIAO Zhogwe (School of Mathematcs Su Yat-se Uversty Guagzhou 50275 Cha WANG Lgd (School of Mathematcs ad Statstcs Hea Uversty Kafeg 475004 Cha ZHANG Yuhu (School of Mathematcal Sceces Bejg Normal Uversty Bejg 00875 Cha Abstract: We cosder probablstc meags for some umercal characterstcs of sgle brth processes Some probabltes of evets such as extcto probablty returg probablty are represeted terms of these umercal characterstcs Two examples are also preseted to llustrate the value of the results Keywords: sgle brth (upwardly skp-free processes; extcto probablty; brth-death processes 200 Mathematcs Subject Classfcato: 60J60 Itroducto ad Ma Results Wag ad Yag [] preset probablstc meags for a lot of umercal characterstcs of brth-death processes such as returg probablty extcto probablty Ths paper s devoted to cosderg the correspodg problems for the sgle brth processes descrbed as follows O a probablty space (Ω F P cosder a cotuous-tme homogeeous ad rreducble Markov cha {X(t : t 0} wth trasto probablty matrx P (t = (p j (t ad state space Z + = {0 2 } We call {X(t : t 0} a sgle brth process f ts The project was supported by the Natoal Natural Scece Foudato of Cha (3003 526075 57043 Specalzed Research Fud for the Doctoral Program of Hgher Educato (20000030005 ad 985 Project from the Mstry of Educato Cha ad Scetfc Research foudato of Hea Uversty (203Y- BZR045 Correspodg author E-mal: waglgd@malbueduc Receved May 4 204 Revsed July 22 205

No 5 LIAO Z W et al: Probablstc Meags of Numercal Characterstcs for Sgle Brth Processes 453 desty matrx Q = (q j : j Z + has the followg form q 0 q 0 0 0 0 q 0 q q 2 0 0 Q = q 20 q 2 q 2 q 23 0 ( where q := q j q j q + > 0 q +j = 0 for Z + ad j 2 The matrx ( s called a sgle brth Q-matrx deduced by q j t + o(t f j < or j = + ; p j (t = q t + o(t f j = as t 0 Throughout the rest of the paper we cosder oly totally stable ad coservatve sgle brth Q-matrx: q = q = j q j < for Z + Especally f q j = 0 for 0 j 2 ad j 2 the ( s just a brth death Q-matrx Some otatos are ecessary before movg o Defe (k = k j for 0 k < (k Z + ad m 0 = q 0 m = + ( d 0 = 0 d = + ( F ( = F ( = + + k=0 + k= k=0 q (k j=0 q (k m k q (k d k F ( k 0 < The the umercal characterstcs defed below play mportat roles studyg sgle brth processes: R = m Z m = =0 =m d = sup >0 [ / d F (0 =0 =0 ] S = sup k 0 k (F (0 =0 d d To expla what the umercal characters mght mea probablty we troduce some stopg tmes respectvely e Deote the frst leapg tme ad the -th jumpg tme by η ad η η = f{t > η : X(t X(η } ; η = lm η where η 0 0 The frst httg tme ad the frst returg tme of the state are defed respectvely as follows τ = f{t > 0 : X(t = } σ = f{t η : X(t = }

454 Chese Joural of Appled Probablty ad Statstcs Vol 32 Though these umercal characterstcs may seem complex they do have explct probablstc meags ad make a postve cotrbuto towards uderstadg the process clearly Let P (A = P(A X 0 = e the codto probablty gve {X 0 = } ad E A = P (A for some measurable set A The Zhag [2] proved that m = E τ + R = E 0 η ad poted out that P 0 (σ 0 < η = Z 0 So R s the mea tme of the frst httg of the sgle brth process wth startg from 0 ad P 0 (σ 0 < η = oce /Z 0 = I [3] we see that d = E τ 0 E 0 σ 0 = /q 0 + d ad E τ 0 = (F (0 =0 d d It s easy to see that S = sup E τ 0 0 Based o the above results the followg explct crtera for several classcal problems ca be uderstood clearly (cf [2 4 7] The process s uque f ad oly f R = Assume that the Q-matrx s rreducble ad regular The the process s recurret f ad oly f Z 0 = For the regular case the process s ergodc f ad oly f d < ad the process s strogly ergodc f ad oly f S < Now we stll eed to study the probablstc meags of Z m ad defed as = wth the coveto that = =m =m > m 0 = 0 f m It wll be see later that these quattes are related to P k (τ m < τ whch s the probablty of arrvg at m alog the trajectory before reachg wth startg from k Before presetg our ma results we meto that f the sgle brth process s ergodc the the statoary dstrbuto (π ca be descrbed as (cf [8] Moreover π k = q kk+ c k c k =k [ / ] c k = sup m F (k k 0 (2 >k =k =k F (k = E k τ + E τ k 0 k < (3 It s easy to see that c k s the mea commute tme betwee k ad k + Now we preset our ma results as follows Theorem Suppose that m < The P k (τ < τ m + P k (τ m < τ = ad

No 5 LIAO Z W et al: Probablstc Meags of Numercal Characterstcs for Sgle Brth Processes 455 ( for 0 k P k (τ < τ m = Z mk P k (τ m < τ = Z mk ; ( for k > P k (τ m < τ = Z k P + (τ m < τ + Z kf (m Z mk + Moreover f the process s ergodc the P + (τ m < τ = ( c c m + + j=m+ j Z mj q( + It s easy to see that P k (τ m < τ = ad P k (τ < τ m = 0 for 0 k m P (τ m < τ = 0 ad P (τ < τ m = As for P k (σ m < τ t s obvous that P k (σ m < τ = P k (τ m < τ for k m ad P k (σ < τ m = P k (τ < τ m for k Moreover we have the followg theorem Theorem 2 Suppose that m < ( Suppose the sgle brth process s ergodc The ( P (τ m < σ = +c c m P m (σ m < τ = m+ ad P m (τ m < σ m = P m (σ m < τ m = 0 P (σ < τ m = +c c m P m (τ < σ m = m+ P k (σ m < η s the probablty of reachg m alog the trajectory through ftely may jumps wth startg from k I partcular P m (σ m < η s the probablty startg from m of returg to m alog the trajectory through ftely may jumps after leavg m whch s called a returg probablty Corollary 3 For P k (σ m < η we have f k < m; P k (σ m < η = m+ f k = m; Z m where we use the coveto that / = 0 Z mk f k > m Z m

456 Chese Joural of Appled Probablty ad Statstcs Vol 32 I practcal applcatos P k (σ 0 < η s called a extcto probablty e the probablty that there exst k dvduals tally but (through ftely may steps of trasto they fally de out (amely reach the state 0 About extcto probablty oe may also refer to [9; Chapter 9] for the case m = 0 Corollary 3 2 Proofs of the Ma Results Proof of Theorem It s easy to see that P k (τ m < τ = ad P k (τ < τ m = 0 for 0 k m To prove the remaders deote P k (τ m < τ by p k By the strog Markov property of the process for m < k we have p k = q kk+ q k p k+ + k j=0 q kj q k p j The by the coservatve property of Q-matrx ad p k = for 0 k m t follows from the above equalty that q kk+ (p k p k+ = k =m q ( k (p p + m < k (4 Deote p p + by v for 0 So we have the dfferece equato v k = k q ( q k v kk+ =m m < k < wth the boudary codtos p m = ad p = 0 By the ducto t s see that By deftos of v ad t s derved that v = v m m < (5 = p m p = =m v = v m =m = v m So v m = / ad v = / (m < Therefore t follows from p = 0 that p k = p k p = =k v = ( =k /Zm ( = =m k =m /Zm = Z mk for m < k < By the smlar argumet oe ca prove the secod part of the asserto ( Of course t s followed mmedately from the property of P k (τ m < τ + P k (τ < τ m = too

No 5 LIAO Z W et al: Probablstc Meags of Numercal Characterstcs for Sgle Brth Processes 457 To prove the asserto ( we wll dscuss frstly the relato betwee p + ad p k wth k + By (4 ad (5 t s see that The v k = ( q kk+ q ( k =m k v m v k = q kk+ k = v m + k = Defe u = v m v ( Thus oe obtas that u k = q kk+ k = q ( k v k > q ( (m k (F v m v k > q ( k u k > By the equaltes above ad the ducto t follows that u = F ( u ( Hece oe deduces that v = = ( v m u = v m F ( u = v m F ( F ( F (m v m + F ( v ( v m v Note that v = p p + = p + Furthermore t s obtaed that for k > p k = p k p = k = ( Z k = k = v = k = Z k p + + Z kf (m = ( ( (F F (m /Zm + Z k p + Z mk + v m + F ( p + By the smlar argumet or the property P k (τ m < τ + P k (τ < τ m = oe ca prove the secod part of the asserto ( Now t remas to show the asserto o the expresso of p + Usg the strog Markov property wth Theorem we have P (τ m < σ = + = + = + p + + j=0 j p j p + + q(m + j=m+ p + + q( j=m+ q ( j Z mj j Zmj Combg the above equalty wth the asserto ( Theorem 2 whch eeds oly some smple calculatos the requred asserto holds mmedately Before provg Theorem 2 we troduce the followg result (refer to [0]

458 Chese Joural of Appled Probablty ad Statstcs Vol 32 Proposto 4 dstrbuto (π The for j we have Proof of Theorem 2 Gve a ergodc Markov cha {X(t : t 0} wth the statoary P (τ j < σ = q π (E τ j + E j τ The asserto ( follows drectly from (2 (3 ad Proposto 4 by some smple calculatos By the strog Markov property ad the asserto below Theorem t turs out that P m (τ < σ m = m j=0 j P j (τ < τ m + m+ P m+ (τ < τ m = m+ P m+ (τ < τ m m+ = m+ f = m + Z mm+ = m+ Z mm+ = m+ f > m + = m+ The remaders of the asserto ( are easly obtaed Remark 5 By ducto t s ot dffcult to obta that F ( F (m m Further we get the followg equalty: Z mk Z k m < < k Hece t follows that Z Z m m < I partcular we see that Z F (0 Z 0 Z 0 for all > 0 Proof of Corollary 3 Note that τ η as almost surely wth respect to P k Hece P k (τ m < τ P k (σ m < η as for k m Combg these facts wth the assertos proved above oe gets easly the frst ad the thrd parts of the asserto By the strog Markov property ad argumet above t s see that P m (σ m < η = m j=0 = q(m m j P j (σ m < η + m+ P m+ (σ m < η = m+ Z m + q ( mm+ Z mm+ Z m I the last equalty we use the fact that Z mm+ = m =

No 5 LIAO Z W et al: Probablstc Meags of Numercal Characterstcs for Sgle Brth Processes 459 3 Examples The frst example s about the brth-death process whch s a specal class of sgle brth processes Example 6 For brth-death processes wth brth rate a ad death rate b at deoted by (a b We have these mportat quattes wth smple forms as follows m = µ[0 ]ν d = µ[ ]ν = ν ν m m 0 where µ[ k] = k µ j wth {µ } s the varat measure havg the followg form j= µ 0 = µ = b 0b b a a 2 a ( ; ad ν s aother measure related to the recurrece of the process wth ν = /µ b ( 0 I the followg we always let ν[ k] deote the term k ν j ad ν[ := ν j for some measure ( For the process we have the followg results whch ca also refer to [] Corollary 7 j= Suppose that m < For brth-death processes we have P k (τ m < τ = ad P k (τ < τ m = 0 for all 0 k m; P k (τ m < τ = 0 ad P k (τ < τ m = for all k ; ( For m < k < P k (τ < τ m = ν [m k ] ν [m ] P k(τ m < τ = ν [k ] ν [m ] ; ( P m (τ m < σ m = P m (σ m < τ m = 0 ad (v (v P m (τ < σ m = P (τ m < σ = j= (a m + b m µ m ν[m ] = P m(σ m < τ ; (a + b µ ν[m ] = P (σ < τ m ; f k < m P k (σ m < η = f k = m µ m (a m + b m ν[m ν[m k ] f k > m ν[m coveto that / = 0

460 Chese Joural of Appled Probablty ad Statstcs Vol 32 Proof By Theorems 2 ad Corollary 3 all the assertos are derved drectly except the asserto (v whch s prove as follows By the strog Markov property ad the asserto ( as well as ( we have P (τ m < σ = The proof s fshed b P + (τ m < τ + a P (τ m < τ a + b a + b = a P (τ m < τ a + b a f m = a + b = / [ a ν (a + b ] ν f m < = /[ (a + b µ Especally the extcto probablty =m =m ν ] P k (σ 0 < η = k / ν ν k =0 =0 The followg example s a exteso of the oe [8] or [] Example 8 Let + = for all 0 q 0 = b = b a 2 = a for all 2 ad q j = 0 for other j where a ad b are costats satsfyg b a > 0 k By computg we kow that {F (k } k are geeralzed Fboacc umbers for every F (k k+ = p+ + 0 k 0 p q where p = (b + b 2 + 4a/2 ad q = (b b 2 + 4a/2 Note that p > b ad < q < 0 Now ad ( p m+ p q m+ q f p ; p q p q = m q + q m+ q ( q 2 f p = ( p +2 p q+2 q f p ; p q p q m = + q + q+2 q ( q 2 f p = ( p + p q+ q f p ; p q p q d = q + q+ q ( q 2 f p =

No 5 LIAO Z W et al: Probablstc Meags of Numercal Characterstcs for Sgle Brth Processes 46 Hece t turs out that R = m =0 = e the process s always uque for all b a > 0 Moreover we get that f p ; Z m = ( p p q p + q f p < q Thus whe p (equvaletly a + b we have Z 0 = the process s recurret ad P k (σ m < η = for all k 0 Whe p < (equvaletly a + b < we have Z 0 < ad the process s traset f k < m; a + b f k = m = 0; P k (σ m < η = ( p( q f k = m > 0; + b Moreover for p we get that p( q( pk m ( pq( q k m p q f k > m P k (τ < τ m = pq(pk m q k m p k m+ + q k m+ + p q pq(p m m p m+ + m+ + p q m < k < ; (p q(p (q P 0 (τ < σ 0 = pq(p p + + + + p q 0 < ; (p q(p (q P m (τ < σ m = ( + b(pq(p m m p m+ + m+ m < + p q for p = oe obtas that k m (k m + q + qk m+ P k (τ < τ m = m ( m + q + q m+ m < k < ; ( q 2 P 0 (τ < σ 0 = m ( m + q + q m+ 0 < ; ( q 2 P m (τ < σ m = ( + b( m ( m + q + m+ m < By the way whe p = the process s ull recurret because d = Refereces [] Wag Z K Yag X Q Brth ad Death Processes ad Markov Chas [M] Bejg: Scece Press 992

462 Chese Joural of Appled Probablty ad Statstcs Vol 32 [2] Zhag J K O the geeralzed brth ad death processes (I the umeral troducto the fuctoal of tegral type ad the dstrbutos of rus ad passage tmes [J] Acta Math Sc 984 4(2: 24 259 [3] Zhag Y H Momets of the httg tme for sgle brth processes [J] J Bejg Normal Uv (Natur Sc 2003 39(4: 430 434 ( Chese [4] Che M F From Markov Chas to No-Equlbrum Partcle Systems [M] 2d ed Sgapore: World Scetfc 2004 [5] Che M F Sgle brth processes [J] Chese A Math Ser B 999 20(: 77 82 [6] Che M F Explct crtera for several types of ergodcty [J] Chese J Appl Probab Statst 200 7(2: 3 20 [7] Zhag Y H Strog ergodcty for sgle-brth processes [J] J Appl Probab 200 38(: 270 277 [8] Zhag Y H The httg tme ad statoary dstrbuto for sgle brth processes [J] J Bejg Normal Uv (Natur Sc 2004 40(2: 57 6 ( Chese [9] Aderso W J Cotuous-Tme Markov Chas: A Applcatos-Oreted Approach [M] New York: Sprger-Verlag 99 [0] Aldous D Fll J A Reversble Markov Chas ad Radom Walks o Graphs [M/OL] Preprt 204 [204-06-02] http://wwwstatberkeleyedu/ aldous/rwg/book Ralph/bookhtml [] Mao Y H Zhag Y H Expoetal ergodcty for sgle-brth processes [J] J Appl Probab 2004 4(4: 022 032 ül êavç¹â % l ( ìœæêææ 2² 50275 (àhœæêæ ÚOÆ mµ 475004 Ü{Ÿ ( ŒÆêÆ ÆÆ 00875 Á : ÄüL êa /ÏùêA x (X«ýžm! ˆž uvç ^ (JOŽü ~fƒ'êa ' c: ül ; «ývç; «L ã aò: O262