The Hypergeometric Coupon Collection Problem and its Dual

Similar documents
0, otherwise. EX = E(X 1 + X n ) = EX j = np and. Var(X j ) = np(1 p). Var(X) = Var(X X n ) =

Binomial transform of products

distinct distinct n k n k n! n n k k n 1 if k n, identical identical p j (k) p 0 if k > n n (k)

An Introduction to Randomized Algorithms

ECE 901 Lecture 4: Estimation of Lipschitz smooth functions

CS 330 Discussion - Probability

Lecture 6: Coupon Collector s problem

On Random Line Segments in the Unit Square

Discrete Mathematics: Lectures 8 and 9 Principle of Inclusion and Exclusion Instructor: Arijit Bishnu Date: August 11 and 13, 2009

Lecture 10: Bounded Linear Operators and Orthogonality in Hilbert Spaces

The Coupon Collector Problem in Statistical Quality Control

AN EFFICIENT ESTIMATION METHOD FOR THE PARETO DISTRIBUTION

1 Convergence in Probability and the Weak Law of Large Numbers

Contents Two Sample t Tests Two Sample t Tests

Bertrand s postulate Chapter 2

AVERAGE MARKS SCALING

The random version of Dvoretzky s theorem in l n

Lecture Outline. 2 Separating Hyperplanes. 3 Banach Mazur Distance An Algorithmist s Toolkit October 22, 2009

q-fibonacci polynomials and q-catalan numbers Johann Cigler [ ] (4) I don t know who has observed this well-known fact for the first time.

Statistics for Applications Fall Problem Set 7

The Differential Transform Method for Solving Volterra s Population Model

Randomized Algorithms I, Spring 2018, Department of Computer Science, University of Helsinki Homework 1: Solutions (Discussed January 25, 2018)

Supplementary Material

A New Type of q-szász-mirakjan Operators

7.1 Convergence of sequences of random variables

Automated Proofs for Some Stirling Number Identities

A new sequence convergent to Euler Mascheroni constant

Chapter 2. Asymptotic Notation

Axioms of Measure Theory

Double Derangement Permutations

CS 70 Second Midterm 7 April NAME (1 pt): SID (1 pt): TA (1 pt): Name of Neighbor to your left (1 pt): Name of Neighbor to your right (1 pt):

Sequences, Mathematical Induction, and Recursion. CSE 2353 Discrete Computational Structures Spring 2018

Q-BINOMIALS AND THE GREATEST COMMON DIVISOR. Keith R. Slavin 8474 SW Chevy Place, Beaverton, Oregon 97008, USA.

A PROBABILITY PROBLEM

Massachusetts Institute of Technology

Some remarks on the paper Some elementary inequalities of G. Bennett

FUZZY RELIABILITY ANALYSIS OF COMPOUND SYSTEM BASED ON WEIBULL DISTRIBUTION

Discrete Mathematics for CS Spring 2007 Luca Trevisan Lecture 22

HOMEWORK 2 SOLUTIONS

arxiv: v1 [math.nt] 26 Feb 2014

COMP 2804 Solutions Assignment 1

f(1), and so, if f is continuous, f(x) = f(1)x.

Birth-Death Processes. Outline. EEC 686/785 Modeling & Performance Evaluation of Computer Systems. Relationship Among Stochastic Processes.

Reliability Equivalence Analysis of a Parallel-Series System Subject to Degradation Facility

Jacobi symbols. p 1. Note: The Jacobi symbol does not necessarily distinguish between quadratic residues and nonresidues. That is, we could have ( a

CSCI-6971 Lecture Notes: Stochastic processes

1.2 AXIOMATIC APPROACH TO PROBABILITY AND PROPERTIES OF PROBABILITY MEASURE 1.2 AXIOMATIC APPROACH TO PROBABILITY AND

DISTANCE BETWEEN UNCERTAIN RANDOM VARIABLES

ECE Spring Prof. David R. Jackson ECE Dept. Notes 20

Application to Random Graphs

1 (12 points) Red-Black trees and Red-Purple trees

Singular Continuous Measures by Michael Pejic 5/14/10

ABOUT CHAOS AND SENSITIVITY IN TOPOLOGICAL DYNAMICS

Element sampling: Part 2

The standard deviation of the mean

7.1 Convergence of sequences of random variables

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

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

On twin primes associated with the Hawkins random sieve

w (1) ˆx w (1) x (1) /ρ and w (2) ˆx w (2) x (2) /ρ.

(s)h(s) = K( s + 8 ) = 5 and one finite zero is located at z 1

J. Stat. Appl. Pro. Lett. 2, No. 1, (2015) 15

Orthogonal Functions

Chapter 6 Infinite Series

Math 4707 Spring 2018 (Darij Grinberg): midterm 2 page 1. Math 4707 Spring 2018 (Darij Grinberg): midterm 2 with solutions [preliminary version]

THE ASYMPTOTIC COMPLEXITY OF MATRIX REDUCTION OVER FINITE FIELDS

A GENERALIZED BERNSTEIN APPROXIMATION THEOREM

Optimal Estimator for a Sample Set with Response Error. Ed Stanek

Product measures, Tonelli s and Fubini s theorems For use in MAT3400/4400, autumn 2014 Nadia S. Larsen. Version of 13 October 2014.

The Binomial Multi- Section Transformer

Lecture 2: April 3, 2013

#A18 INTEGERS 11 (2011) THE (EXPONENTIAL) BIPARTITIONAL POLYNOMIALS AND POLYNOMIAL SEQUENCES OF TRINOMIAL TYPE: PART I

BIRKHOFF ERGODIC THEOREM

Some results on the Apostol-Bernoulli and Apostol-Euler polynomials

Lecture 2. The Lovász Local Lemma

Lecture 7: Properties of Random Samples

This is an introductory course in Analysis of Variance and Design of Experiments.

Queueing Theory II. Summary. M/M/1 Output process Networks of Queue Method of Stages. General Distributions

42 Dependence and Bases

International Journal of Multidisciplinary Research and Modern Education (IJMRME) ISSN (Online): (

Exercise 8 CRITICAL SPEEDS OF THE ROTATING SHAFT

Topic 9: Sampling Distributions of Estimators

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

5.6 Binomial Multi-section Matching Transformer

1 Introduction to reducing variance in Monte Carlo simulations

Discrete Mathematics for CS Spring 2005 Clancy/Wagner Notes 21. Some Important Distributions

X. Perturbation Theory

Introduction to Optimization, DIKU Monday 19 November David Pisinger. Duality, motivation

1 The Primal and Dual of an Optimization Problem

Surveying the Variance Reduction Methods

Optimization Results for a Generalized Coupon Collector Problem

MATH 320: Probability and Statistics 9. Estimation and Testing of Parameters. Readings: Pruim, Chapter 4

THE N-POINT FUNCTIONS FOR INTERSECTION NUMBERS ON MODULI SPACES OF CURVES

THE KALMAN FILTER RAUL ROJAS

PROBLEM SET 5 SOLUTIONS 126 = , 37 = , 15 = , 7 = 7 1.

Distribution of Random Samples & Limit theorems

1 Duality revisited. AM 221: Advanced Optimization Spring 2016

A Hadamard-type lower bound for symmetric diagonally dominant positive matrices

CSE 1400 Applied Discrete Mathematics Number Theory and Proofs

Topic 9: Sampling Distributions of Estimators

Transcription:

Joural of Idustrial ad Systes Egieerig Vol., o., pp -7 Sprig 7 The Hypergeoetric Coupo Collectio Proble ad its Dual Sheldo M. Ross Epstei Departet of Idustrial ad Systes Egieerig, Uiversity of Souther Califoria, Los Ageles, CA, USA (soss@usc.edu) ABSTRACT Suppose a ur cotais M balls, of differet types, which are reoved fro the ur i a uifor rado aer. I the hypergeoetric coupo collectio proble, we are iterested i the set of balls that have bee reoved at the oet whe at least oe ball of each type has bee reoved. I its dual, we are iterested i the set of reoved balls at the first oet that this set cotais all of the balls of at least oe type.. ITRODUCTIO AD SUMMARY Cosider a ur that cotais M balls, of which i are of type i, for i,...,, M i i. The hypergeoetric coupo collectig proble arises whe balls are reoved fro the ur i uifor rado aer util at least oe ball of each type has bee reoved. Let i deote the uber of type i balls that are reoved i the coupo collectig proble, ad let i deote the total uber of reoved balls. I sectio we derive the oit probability ass fuctio of,..., as well as the argial ass fuctios ad the eas ad variaces of the i. We also give forulas for the factorial oets of. The tail distributio P (>r) is explicitly obtaied whe all i are equal, ad upper ad lower bouds are provided i the geeral case. We also show that P(>r) is a Schur covex fuctio of the paraeters,...,. The dual of the hypergeoetric coupo collectig proble, which stops reovig balls at the oet whe all the balls of at least oe type have bee reoved, is cosidered i Sectio 3.. RESULTS FOR THE HYPERGEOMETRIC COUPO COLLECTIO PROBLEM We start with the oit ass fuctio of,...,. Propositio With r i, P ( i, i,..., i) i : i M M r + ( r )

Ross Proof: With L beig the last of the types to have oe of its balls reoved, we have P ( i, i,, i ) P( i, i,, i, L ) : i i : i M M r + r To obtai the argial ass fuctio of we will fid it useful to cosider a cotiuous tie odel i which each of the balls is reoved at rado tie that is uiforly distributed o (, ), with these M ties beig idepedet. Clearly, the order i which the balls are reoved i this cotiuous tie odel is probabilistically the sae as i the origial odel. Propositio r r r, P( ) ( t) [ ( t) ] dt + ( t) t( t) [ ( t) ] dt i i i r, P( i) ( t) t ( t) [ ( t) ] dt, i > Proof: Let T deote the tie of the first reoval of type ball. The, with L desigatig the last type ball reoved P ( i) P ( il, ) + P ( il, ) P ( il, ) + P ( il, T t ) ( t) dt i i + i r, P ( il, ) ( t) t( t) [ ( t) ] dt where ( t) [ ( t) ] dt, if r, if i > P ( il, ) ad the proof is coplete. Rear: The precedig result ca be used to obtai a expressio for P ( i ), the expected uber of types to have exactly i balls reoved. Expressios for these quatities i the classical coupo collectig proble - where each ew coupo is, idepedetly of the past, of type i with probability p i, i,...,, - were give i Adler et al. (3). We ow obtai the ea ad variace of.

The Hypergeoetric Coupo Collectio 3 Propositio 3 With r, p ( t) ( ( t) ) dt+ ( t) t ( ( t) ) dt, E [ ] p ( ) + r, Var( ) p ( p ) [ ( t) t ( ( t) ) dt p ] Proof: uber the type balls, ad let R i deote the evet that type ball uber i is i the fial set, i,...,. The, PR ( ) PR (, L ) + PR (, L ) i i i r, ( t) ( ( t) ) dt+ ( t) t ( ( t) ) dt Also, if, the for i s PRR ( ) PRR (, L ) i s i s r, ( t) t ( ( t) ) dt the result follows sice I. Ri The ext propositio yields a forula for factorial oets of balls oe eeds to reove to obtai a coplete set., the total uber of Propositio 4 ( M + r)! i r E[ ( + )...( + r )] r [ ( p )]( p) dp M! Proof: Cosider a syste with M copoets i which each copoet is either worig or failed, ad suppose there is a ootoe structure fuctio that defies whe the syste wors as a fuctio of which copoets are worig. Suppose that the M iitially worig copoets fail i a rado, uiforly distributed order, ad let equal to the uber of copoets that eed fail to cause the syste to fail. It was show by Ross et al. (98) that ( M + r)! r E[ ( + )...( + r )] r r( p)( p) dp M! where r(p) is the probability that the syste wors whe each copoet idepedetly wors with probability p.

4 Ross For a syste of M copoets, partitioed ito disoit subsets of sizes,...,, that is said to wor if all of the copoets of at least oe subset wor r( p) ( p i ) dp ad the result follows. Our ext result shows that the distributio fuctio of is a Schur fuctio of the paraeters (,..., ). Propositio 5 P (>r) is a Schur covex fuctio of (,..., ). Proof: Let (,..., ) be the uber of balls reoved to obtai at least oe of each type. It suffices to show, for >, that P ( (,,,..., ) > r) P ( (, +,,..., ) > r) 3 3 Let I(,..., ) be the idicator variable of the evet that the set of the first balls reoved cotais at least oe of each of the types 3,...,; ad let R(,..., ) deote the total uber of balls of types 3,...,, that are aog the first r balls reoved. The P ( (,..., ) > r) P ( (,..., ) > ri (,..., ) ir, (,..., ) ) P( I(,..., ) i, R(,..., ) ) Because (I,R ) (I(,, 3,, ),R(,, 3,, )), ad (I,R ) (I( -, +, 3,, ),R( -, +, 3,, )) have the sae oit distributio, it suffices to show that for > ad > P( (,..., ) > ri, R r ) P( (, +,..., ) > ri, R r ) which is equivalet to showig that P ( (, ) > ) P ( (, + ) > ) or, equivaletly, that or, equivaletly, that + + + + +

The Hypergeoetric Coupo Collectio 5 or, equivaletly, that ( )! ( )!( )! which is iediate. The followig propositio gives bouds for P( >r), as well as a closed for expressio whe i,,,. Propositio 6 M M r r M M P ( > r) M M r r + r r If all, i ( r) ( ) P i+ > i M i r M r Proof: Let A i deote the evet that there are o type i balls aog the first r selected. The P> r PA A A ( ) (... ) The right had iequality follows fro Boole s iequality, ad the left had oe fro the coditioal expectatio iequality (Ross, ). The result whe all, follows fro the iclusio-exclusio idetity. Rears: (a) The coditioal expectatio iequality, stroger tha the secod oet iequality, states that P( A ) PA ( ) + PAA ( ) i (b) Although we could also use the iclusio-exclusio idetity to derive a expressio for P( >r) i the geeral case, it would require suig over ters.

6 Ross (c) Whe all i,, we ca also efficietly copute P( >r) by a recursio. For, i <, let A(i,) deote a geeric rado variable havig the distributio of the uber of additioal balls oe ust reove to have coplete set if balls reai i the ur ad there are still i types that have yet to be collected. The, with we have the recursio P (, i ) P( A(, i ) ) i i P(, i ) P ( i, ) + P (, i ), i> with the boudary coditio P (, ) ; uerical coputatio yields the result P(>r)-P r (,). 3. THE DUAL PROBLEM The dual proble to the oe so far cosidered is iterested i the uber of balls that have bee reoved at the first oet that all the balls of ay of the types have bee reoved. ow, iagiig that we cotiue reovig balls util they are all tae fro the ur, the outcoe of the experiet for both the origial ad dual proble is a vector ( i,..., i M ), where i is the type of the th ball reoved. ow, as a fuctio of the outcoe, let ( i,..., i M ) be the uber of type balls that have bee reoved, ad let L ( i,..., i M ) be the last type to have bee reoved, at the first oet whe at least oe ball of each type has bee reoved. Siilarly, let ( i,..., im) be the uber of type balls that have bee reoved, ad let L ( i,..., im ) be the last type to have bee reoved, at the first oet that all the balls of ay of the types have bee reoved. It is easy to see that + ( im,..., i), if L( i,..., im) ( i,..., im) ( im,..., i), if L( i,..., im) () (,..., ) (,..., ) () L i im L im i Let ad L be the ubers of type balls reoved ad the last type reoved, respectively, i the dual proble, ad let ad L be the correspodig quatities for the origial proble. Because all outcoes are equally liely, it follows fro () ad () that (, ), P + rl if P ( rl, ) P ( rl, ), if (3)

The Hypergeoetric Coupo Collectio 7 I additio, it follows fro () that i im + M im i (,..., ) (,..., ) iplyig, because all outcoes are equally liely, that P ( ) P ( M+ ) (4) where ad are the ubers of balls reoved i the origial proble ad i the dual proble, respectively. Rears (a) The dual proble was previously cosidered by El-eweihi et al.(978), where their ai cocer was deteriig the probability ass fuctio of the type of the last ball reoved. I additio, they coectured that was a icreasig failure rate rado variable. This coecture was the prove by Ross et al. (98). (b) The duality result give by Equatio (4) was previously give i Ross et al. (98). REFERECES [] Adler I., Ore S., Ross S. M. (3), The coupo collector s proble revisited; Joural of Applied Probability, 4; 53-58. [] El-eweihi E., Proscha F., Sethuraa J. (978), A siple odel with applicatios i structural reliability, extictio of species, ivetory depletio ad ur saplig; Advaces i Applied Probability, (); 3-54. [3] Ross S. M. (), Probability odels for coputer sciece, Acadeic Press. [4] Ross S. M., Shahshahai, M., Weiss G. (98), O the uber of copoet failures i systes whose copoet lives are exchageable, Matheatics of Operatios Research, 5(3); 358-365.