0% found this document useful (0 votes)
119 views19 pagesSimulation Models ORIntroduction Hamdy Taha
Uploaded by
yonasCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
119 views19 pagesSimulation Models ORIntroduction Hamdy Taha
Uploaded by
yonasCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
/ 19
Chapter 1 8
Simulation Modeling
18.1. WHAT Ss SIMULATION?
‘Simulation isthe next best thin to observing areal system in operation. It allows us
to collect pertinent information about the behavior of the system by executing
‘computerized model The collected data are then used to design the system. Simla
tion is not an optimization technique. Rather, itis a technique for estimating the
‘measures of performance of the modeled system,
‘Simulation has been used in all aspects of science and technology asthe fol
Towing partial ist demonstrates:
1. Basie science
(@) Estimation of the area under a curve or, more generally, evaluation of
‘multiple integrals
(b) Estimation ofthe constant (= 3.14159)
(©) Matrix inversion
(@) Study of particle diffusion
2, Practical situations
(a) Industrial problems including the design of queueing ystems, commu
nication networks, inventory coutuul wt chemical processes
() Business and economic problems including consumer behavior, price
determination, economic forecasting, and total firm operation
(©) Behavioral and social problems, including population dynamics, envi
ronmental heath effects epidemiological studies, and group behavior
(@) Biomedical systems, including uid balance, electroyte distribution in
the human body, blood cell proliferation, and brain activites
(©) Warstrategies and tats
on om (Chap. 18 Simulation Modeling
Estimation of simulation output i based on random sampling, much the same
way we do when observing areal situation. his mears that the output of simulation
is subject to random variations, and thus,as in any statistical experiment, must be ex-
famined using formal statistical inference tests. This important point is stressed
throughout the chapter
18,2 MONTE CARLO SIMULATION
Monte Carlo simulation refers tothe use of random sampling to estimate the output
‘of an experiment. It is regarded as the forerunner to present-day simulation
‘This section uses an example to demonstrate the procedure. The objective of
the example is to stress the statistical nature ofthe simulation experiment.
Example 2-1
"We wll we Monte Caro sampling to estimate the aes of cce whose equation is
(r= 0F + y= 2F #25
The radius ofthe circle isr = Sm, and its ceteris ix») = (1.2)
The procedure fr estimating the area requires enlosng the cele “snugly in a
square whose side equals the diameter ofthe clea shown in Figure 18-1, The corer
pins are determined from the geometry ofthe quae,
|
a 9 rire tet
‘The estimation ofthe area of the czle based om the assumption that all the
points in the square ae equally key to oecr, Ths if out ofa random sample of
points inthe square, are found t fall within the cece then
(Areaot) my AreRot Ym 9 x 1)
\thecicle) ~ (ine square) *
The ideas implemented by representing the coordinates x and y ofa pint inthe
quate by the following uniform distributions
Te)
d. wsers6
SSec.182 Monte Carlo Simulation ons
£0) 3eye7
rT}
Both functions equil zero ouside the indicated ranges.
‘Let and be two ditnct random numbers between 0 and 1.Then an equally
‘ely random point (9) nthe square i determined ax
B= -8+ [6 (-8)]R, = -44 108,
yo -34 (T= (COIR, = 3 410%
Table 18-1 give sual list of 01) random numbers These numbers are gener-
ed by using specal arithmetic operations ar wil be explined in Section 185.
ee ee ee
613336461281 A871 76982465
«m9 76162867 S28 4220
is 8981 2168912954
6139 3019526129113 s
S833 7816 3866 zn 0s 408
gui 519971255954 16056037
17826358 21085403 35672560
Mn 4 ASS 42083000546
Sou 9542706 607555190305
For each pir of distinct random numbers (R, .), we can generate a (uniform)
random point (9) in the quate by wing the ven formulas A sample pola 2)"
falls within the cil if
(= DF + Gt = 2F 525
For example, given R, = 0889 and Ry = 6733, then
B= =4-6 108, = 4 + 10¢ 0880) = —3.411
y= 34 10R, = 3 + 10(6733) = 3.733
Because (~3411 ~ 1)? + (733 ~ 2) = 22.46 ess than 25, the point (fli oe
chap. 18 Simulation Modeling
We investigate now the effect of random sampling onthe aecuray of estimating
the area ofthe ele Figure 18-2 summaries the results for various sample ies rang
ing fromm = 100 tom = 10000. The experiment replicated 10 times foreach, with
plcaton sing a ifferent sequence of (0,1) eandom numbers The igre graphs
the estimates for replication | and 2 only, together wi the mean and standard devia
tion forall 10 replications
&
=e
Figure 18-2
The following conclusions can be dawn from the results of the experiment
1. The estimate ofthe area ofthe cc improves asthe sample size increases, a
demonstrated in Figure 1-2 by replications 1 and 2.
‘Sec. 182 Monte Carlo Simulation or
2 The average ofall 10 replications for each supe size n provides a etter es
timate ofthe ara than any one rephction, In Figure 18-2, the mean ofthe 10
replications approaches the exact area (= 7884 en) more raply thn the
éstimates provided by the individu repctins
5 The “accuray" ofthe mean value fall 10 replications increases withthe sm:
ple sass m as evident by the dectne inthe valve of the standard deviation
Because ofthe random variation in the output ofthe experiment itis essential to
express the results 8a confidence interval, Letting A snd. be the mean and variance
Of replications, the 100(1 ~ a) % conidence interval forthe tue area A ie
wheter the 8 percentage point ofthe tistbuton with N — 1 degrees of
freedom (ee the etable in AppenditD). (Note that equals the number of repli
‘ons which tint from the sample size) la terms of he preset experiment, we
‘would be interested in establishing the confidence intrel bared onthe largest sample
Sze (Le, = 10000) Given N= 10, A= 7857 cn and = 47 cr the resting
‘3% eondencewnteral 7423 = A'S 74S
The example pores two questions tegarding the simulation experiment
1. How large should the sample ize, be?
2. How many replications,
re needed?
The answers depend on the nature ofthe simulation experiment, Asin any statistical
experiment, higher values of and N provide more relible simulation result. The
drawback s that higher valves of and W imply higher cots for onduting the simul
tion experiment
Problem set 18.2a
1. In Example 182-1, do the followin:
() Estimate the area ofthe circle using the fist two columns of the (0,1)
random numbers in Table 18-1. (Fer convenience, o down each col
umn, selecting R, fist and then Rs)
(©) Suppose that the equation ofthe circle is
(x-3F + (y +2 = 16
Define the corresponding distributions f(x) and f(y), and then show
how a sample point (x y) ean be determined using the (0,1) random
pair (R,, Ry).
2, Use Monte Carlo sampling to estimate the area of the lake shown ia Figure
18-3. Base the estimate on the fist two columns of (0,1) random numbers
in Table 18-1 chap. 18 Simulation Modsting
ie 18-3
3. Consider the game in w
fair coin. Ifthe outcome:
$10 from i
(@) How isthe game simulated as a Monte Carlo experiment?
(@) Ran the experiment for 5 replications of 10 tosses each. Use the fist
five columns of the (0,1) random numbers in Table 18-1, with each cok
uum corresponding ‘ one replication.
(©) Establish a 95% conidence interval on Jan’ winnings.
(@) Compare the confidence interval in (c) with Jans expected theoretical
winnings
4. Consider the following éefinite integral:
two players, Jan and Jim, take turns in tossing a
heads Jan gets $10 from Jim. Otherwise, Jim gets
[eae
(@) Develop the Monte Carlo experiment to estimate the integral
(@) Use the frst four columns in Table 18-1 o evaluate the integral based
‘on 4 replications of sie 10 each. Comnpute a 95% confidence interval,
and compare it with the exact value of the integra
§5, Simulate five wins or lowes ofthe following game of craps:The player rolls
‘wo fair dice. If the outcome sum is 7 or 11, the player wins $10. Otherwise,
the player records the resulting sum (called poi) and keeps on rolling the
tice until the outcome sum matches the recorded point, in which case the
player wins $10. Ifa 7 is obtained, the player loses $10.
6, The led time for receiving an order can be or 2 days, with equal probabil
ties The demand per day asumes the values 0,1, and 2 with the respective
probabilities of 2, 5,and.3. Use the random numbers in Table 18-1 (stating
‘with column 1) to estima the joint distribution ofthe demand and lead time
From the join distribution, estimate the pf of demand during lead time
(Hint The demand duritg lead time assumes discrete values from 004)
7. Consider the Buffon needle experiment. A horizontal plane is ruled with
parallel lines spaced D em apart. A needle of length d em (d < D) is
‘Sec. 183 Types of Simulation ora
«dropped randomly on the plane. The objective ofthe experiment is to de~
{ermine the probability that either end ofthe needle tou
ofthe lines Define
‘n= Perpendicular distance from the nesdle center toa (parallel ine
8 = Inclination angle of the needle witha line
(2) Show thatthe needle will touch or crossa line only if
D
2
() Design the Monte Catlo experiment, and provide an estimate ofthe de
sired probability.
(©) Use the first four columns of Table 18-1 to obtain 4 replications of size
Oech ofthe desired probability. Determine a 95% confidence interval
for the estimate.
(@ Prove that the theoretical probability is given by the formula
n=Ssin, oshs2.0s0s0
2
PD
Use the result in (¢) together with this formula to estimate
18.3 TYPES OF SIMULATION
“The execution of present-day simulation is based geterally on the sampling idea ot
the Monte Carlo method. It differs in that itis concerned withthe study of the be
vir o real ystems a function of ine. Two dint peso inlaton md
1. Continuous models deal with systems whose behavior changes continuously
with time typical example of a continuous simulation isthe study of world popu
Tation dynamic. Continuous simulation models usually are represented in terms of
ditference-ditferental equations that deseribe the interactions among the different
elements of the system,
2. Discrete models deal with systems whose behavior changes only at given in
stants A typical example occurs in waiting lines where we are intersted in estimat-
ing such measures asthe average waiting time or the length of the waiting line. Such
measures change only when a customer enters or leaves the system, At all other in-
stants nothing from the standpoint of collecting statistics occurs inthe system,
‘The instants at which changes inthe system occur identify the mode'’s events
(g.,arrval and departure of customers) The fact that these events occur at disrete
points gives rise tothe name diserete event simulation, 600 chap. 18 Simulation Modeling
Although both continuous and dissrete simulations are important tools in
practice, the discrete type is the one that is usualy presented in conjunction with
Cperations research tapi The reason i hat disrete simulation is more closely al
Hed with queueing models (see Chapter 17). Indeed, tis fair to state that practically
all diserete simulation situations can be described in some form or another as
‘queueing models.
“This chapter concentrates on discussing the basics of discrete event simula
tion. We start with a description of events and how they can be generated in asim-
tlation model, Next, we present the procedures for collecting simulation statistics
‘and discuss the statistical aspect ofthe simulation experiment. The chapter also em-
Dhasizes the role ofthe computer and simolation languages inthe execution of sim-
‘lation models
Problem set 18.3
1. Categorize the following situations as either discrete or continuous (or a
‘combination of both). In each cae, specify the objective of developing the
simulation model,
{@) Orders for an item arrive randomly ata warehouse. An order that
not be filed immediately fom available stock must await the arrival of
ew shipments
‘World population is affected by the availability of natural resources,
food production, environmental conditions, educational level, health
cate, and capita investments
(©) Goods arrive on pallets at a receiving bay of an automated warehouse
The pallets ae loaded on a lower conveyor belt and lifted through an,
tupelevator to an upper coaveyor that moves the pallets to corridors
“The corridors are served by cranes that pick up the pallets from the con
‘veyor and place them in storage bins.
2. Explain why you would agree or disagree with the following statement
“Most discrete event simulation models can be viewed in some form or an
other as queueing systems consisting of sources from which customers are
generated, quewes where customers may wait, and facies where cus
tomers are served”
©
118.4 ELEMENTS OF DISCRETE EVENT SIMULATION
“This section shows how the concept of events is implemented and how the statistics
ofthe simulated system are collected
10.4.1 Generic Definition of Events
All discrete-event simulations describe, directly or indirectly, queueing (or waiting
{ine) situations in which customers arrive, wat in a queue if necessary, and then re=
Sec. 18.4 Elements of Discrete Event Simulation 6st
ceive serie before they depart he system. In general any dscrete-event model i
composed ofa network of merrelated queues
Given that dscet-event models in reality a composite of queues we noice
that forthe purpose of eallecting sats (mets of performance) changin the
‘omem (aioe gt anda he evs fli) ca only our when 8c
tomer ave athe gute and when acstomer eaves he lai ater bing seve
‘his mean that the two principal evens in ony Get simulation model are an
tral an dare Tes se the oly twoustans st which we need toexamine he
Sotem Atal other ines no changer atloctng the statis othe syste cour
Example 18,
MetalcoJobshop receives wo types of jobs: regular and rush, lobe are procesed on
two consecutive machines with ample buffer areas. Rush obs aways assume nonpre
emptive priority over regular obs Identify the evens of hesitation
"Ti situation consis of two tandem quevescortesponding tthe two machines
"pects At nt thoughon maybe nines tent the even of he stan
ALIA regular job arives at machine 1
‘A2I:A tush ob arrives at machine 1
DII:A regular job departs machine 1
D21:A rush ob departs machine 1
‘A12:A separ ob arrives at machine 2.
‘A22:A rush job arrives at machine 2.
DI2:A regular job departs machine 2
1D22:A rushjob departs machine 2
In realy, we only have exactly two events an arrival of a (new) jb atthe shop
and. departure of (completed) job from a machine, is note that events DIL ang
[Alzcoinide exactly and hence are not distinguishable. The same observation apis
to D2l and AZZ Next, in ascete simulation we ean use one (aval or departure)
‘vent for both ype offs and ply “tag” the event wth am alate hat enties
the job type as either regular o rush. (We en think ofthe atribute inthis cse a per-
sonal ideation number, and, indeed, itis) Given this teasoing the evens ofthe
‘model reduce to (1) an aeivalA (atthe shop) and (2) a departure D (rom each mi
chine) Te ations associated with the departure event wll depend onthe machine at
‘which hey oecu
Having defined the basic evens of a simulation model, we now show how the
models executed. Figure 18-4 gives a schematic representation of typical ocarrences
Figure 18-4 en Chap. 18 Simulation Modeling
‘of events onthe simulation time sale. After all the actions associated witha current
treat have been performed, te simulaton advances by “jumping” tothe next chrono
Tosi event In exene, the eneetion of the simulation occurs a the instant at which
the events ocr.
tow does the simulation determine the occurrence time of the events? Theat
rival events are separated by te interacial time (the interval between successive
rivals), andthe departure events ar specie by the service time inthe fei
rte times may be deterministic (eg a tain arivng ats station every 5 minutes)
Or probeblistie (ete random anv ofestomers as bank). Ie the ume between
vents i deterministic, the determination of their occurrence times i staightfor.
Ward. I itis probabilistic, we use a special procedare to sample the interevent ime
From the corresponding probability Sistribution. This points discussed inthe next
Problem set 18.44
1. Identify the discrete everts needed to simulate the following situation: Two
types of jobs arrive from two different sources Both types are processed on
‘single machine, with priority given tothe jobs artiving from te firs source.
2, Jobs arive at a constant rate at a carousel conveyor system. Three service
“ations are spaced equaly around the carousel. Ifthe server is idle when @
job arrives at the station, the job is removed from the conveyor for process=
ing. Otherwise, the job continues to rotate about the carousel until a server
becomes availabe, A processed job is stored in an adjacent shipping area.
Identify the disrete events needed to simulate this situation
3. Cars arrive at a two-lane driven bank, where each lane can house a maxi
num of four ears. If the two lanes are full, arriving cars seek service ese
‘where If at anytime one lane isa least two cars longer than the other, the
last cari the longer lane wil jockey to the lst position inthe shorter lane
‘The bank operates the drive-in facility from 800 A.M. to 3:00 PM. each
‘work day Define the discrete events forthe situation.
4. The cafeteria at Elmdale Elementary provides a single-ray, fixed-menu
tunch toa its pupils Kids arrive atthe dispensing window every 30 seconds.
It takes 18 seconds to receive the lunch tray. Map the arrival-departure
events on the time scale forthe first five pupils.
18.4.2 Sampling from Probability Distributions
‘Randomness in simulation arises when the interval, , Between successive events is
probabilistic, This section presents methods for generating successive random
Samples (F = ffs) from a probsbilty distribution (Q). All the methods are
footed inthe use of independent and identically distributed uniform (0,1) random
numbers.
See. 184 Elements of Diserete Event Simulation ee
Inverse Method. Suppose that its desired to obtain random sample x from
the (continuous or discrete) probability density function f(s). The inverse method
first determines a closed form expression of the cumulative density function
Fi) = Ply = x), where 0 = F(x) = 1, forall defined values of y.Given R isa ran-
‘dom value obtained from a uniform (0, 1) distribution, and assuming that F's the
inverse of F the steps ofthe method areas follows:
Step 1. Generate the (0,1) random number R.
Step 2. Compute the desired sample x = F(R),
Figure 18-S illustrates the procedure for both a continuous and a discrete ran-
dom distribution. The uniform (0, 1) random value R, is projected from the vertical
F(x)-scaleto yield the desired sample value x, on the horizonal scale
Figure 18-5
“The validity of the proposed procedure rests on showing thatthe random vat
able z = F(x) is uniformly distributed in the interval 0 = 2 1, as the following
theorem proves.
‘Theorem 184-1. Given the cumulative density function F(x) ofthe random
variable x, = x 5 % the random variable 2 = F(x), z I, has the following.
probability density function:
1@
Which isa uniform (01) distribution.
Proof The random variable is uniformly distributed if, and only if
PesZ)=Z, 0=Z51
‘This result follows directly from the following equalities:
Ple SZ) = PIF) = 2) = Pes F(Z) = FP) = Z
Additionally,0 = Z = 1 because 0 = Ple = Z} = 1 Chap. 18 Simulation Modeling
‘Example 1842 (EXPONENTIAL DISTRIBUTION).
tial distribution with mean le =| time units that
ted by an exponen
f=", > 0
Determine random sample trom
The cumulative density function i determined as
Roy [rewar= ie e>0
Setting = FQ), we can solve fort which yields
1
r=-E)ina-m
Because 1 ~ Ris the complement of R, we may replace I ~ R with R
"ters of simulation, the result means thatthe occurrences ofthe events are
‘Stnple ofthe period unt the rival of the next customers computed as
t= -({)ime = 9) srr nour = 345 mites
“The values of R used to obtain successive samples mist be selected randomly
fom a uniform (0,2) distribution, We wil show Iter in Section 18.5 how these (0,1)
andom values are generated during the exceution ofthe simulation,
Problem set 18.4
1. In Example 184-2, suppose that the fist customer arrives at time 0, Use the
frst three random numbers in columa 1 of Table 18-1 to generate the ar-
rival times of the next tee customers and plot the resulting events on the
time scale.
2 Uniform distribution. Suppose thatthe time needed to manufacture a part,
‘oma machine is uniformly disteibuted in the interval fa, b],a< b—that is,
1
f= sha asrsh
ar ors
Determine an expression for the sample given the random number &
3, Jobs are received randomly at a one-machine shop. The time between a
rival is exponential with mean 2 hours. The time needed to manufacture a
job is uniform between 1.1 and 2 hours. Assume thatthe first job arrives at
time 0. Determine the arrival and departure time forthe fist iv jobs using
the (0,1) random numbers in column 1 of Table 18-1,
‘4. The demand for an expensive spare part of a passenger jet is 0, 1,2, or 3
units per day with probabilities 2, 3, 4, and 1, respectively. The aisle
Sec, 184 Elements of Diserete Event Simulation 08
maintenance shop starts operation with a stock ofS wnits, and will bring the
stock level back fo 5 units immediately after it drops below 2 unit.
(2) Devise the procedure for determining demand samples.
(b) How many days wil elapse until the first replenishment occurs? Use
successive valves of R from the fst column in Table 18
Ina simulation situation, TV units are inspected for possible defects. There
isan 80% chance that a unit will pas inspection, in which itis sent to pack
aging. Otherwise, the unit i repaired, We can represent the situation sym.
bolicaly in one of two ways
goto REPAIR/2, PACKAGES
goto PACKAGE/S8, REPAIR/2
equivalent. Yt, when a given sequence of
(0,1) random numbers is applied to the wo representations, different dec:
sions (REPAIR or PACKAGE) may result. Explain why
[A player tosses a fair coin repeatedly until a read occurs The associated
payott is 2, where mis the number of tosses unl a head is realized
(a) Devise the sampling procedure ofthe game
() Use the random numbers in column 1 of Table 18-1 to det
smulative payoff after two heads ae realized,
1. Triangular distribution. Ia simulation, the lack of data may make it impos
bie to determine the probability distribution essociated with a simulation
activity. In most ofthese situations it may be easy to describe the desired
vatiable by estimating its smallest, most likely, and largest values, These
three values are sufficient to define a triangular distribution, which can then
be used as “rough cut” estimation ofthe “true” distribution
(@) Develop the formula for sampling from the following triangular dst
bution whose parameters area, 5 and ¢, where a = B < c
2-4)
@- 96-0
2e-x)
© de-a)
©) Generate three samples from & triangular distribution with parame:
ters (1,3, 7) using the first three random numbers in column 1 of
Table 18-1
fe)
. (a) Develop the sampling procedure fora probability distribution that con:
sists of a rectangle flanked onthe left and rght sides by two symmetri
cal right triangles. The respective ranges ‘or the triangle on left, the
rectangle, and the triangle on right are fa, 61, (5 cl, and fe dl
0
where « > Oi the shape parameter, and 8 > Oi the scale parameter.
“The inverse method works well when the cumulative density function F(x) is
analytically tractable. Distributions, such asthe gamma, norma}, and Poisson, do not
fallin this category. The methods listed inthe remainder of this section show how
these distributions are sampled,
Convolution Method. The basic idea ofthe convolution method is to express
the desired sample asthe statistical sum of other easy-to-sample random variables.
Typical among these distributions are the Erlang andthe Poisson whose sample can
belobtained from the exponential distribution samples,
Example 184-3 (ERLANG DISTRIBUTION.
The m-Erlang random variable is defined asthe statistical sum (convoltions) of min
‘pendent and idenscally distibuted exponential random variables Let y represent
the m-Eriang random valble, then
ndent and identically distributed exponential ran
From Example 184-2, the th exponential distribution canbe sampled as
1)
In(R), f= 1am
Ths the m-Evang sample is computed as
y= —(2) amc + tn) += +R
= -(2ncitte
Sec. 18.4 Elements of Disorete Event Simulation or
illustrate the use ofthe formula, suppose that m = 3,and A = 4 events per hour
The first 3 random numbers in column 1” of Table. 18-1" yield Risky
(0589),6733)(4799) = 0190, which yields y ~ ~({)in 019) ~ 991 hour.
"Example 184-4 (POISSON DISTRIBUTIONS,
Ifthe distibution between the oocurrence of sucesive event is exponential then the
sistrbuton ofthe number of events per uit ime mut be Piston and vie versa, We
ts this relationship to sample the Polson ditrbution
Assume thatthe Poisson distribution hata mean value of events per unit time,
‘Then the time between event exponential with mean! time unit Thismeans that ¢
Poison sample, wll occur during tine nis and oly
‘This condition tansates to
Ostsy, mao
where ia sample from the exponential distribution with mean |. From the esl in
Example 1843, we hive
(Him RoR) =< (Ein RonRasds m>0
o=r<-()nary
whieh reduces to
n=0
oilustrate the implementation of the sampling process, suppote that A= 4
‘evens per hour and tat we wish to obtain a sample fora period = 3 hour Thi gives
"= 383. Using the random numbers in calm 1 of Table 16-1, we note that
1, = £0889 islessthan -* = 1383 Hence, the corresponding simple sn 0.
Example 184-5 (NORMAL DISTRIBUTION.
"The Central Limit Theorem states thatthe sum (convolution) of» independent and
idenscllydstrbuted random variables tend o be normaly dsributed as n becomes
suid large We use this result to generate sample from normal distbuton vith
‘mean and standard deviation
SER TRH OER,
The random vail i asymptotically normal bythe Central Limit Theorem, Given
thatthe uniform (0, random number Rhas a mean of} snd variance off fllows aa Chap. 18 Simulation Modeling Sec. 18.4 Elements of Diserete Event Simulation 609
Lee aa eins Te Sa 2. tn Example 184-4 generate hee Poison sample rng our period
rattan na uaa Sots sn ohcanbecomped tons fiven ta the mea of he Potton eens per hour
3. In Example 185-5, generate ewe same rom M1 by wing bh the
Conveton method tn he Bor Maer method
4. Jobs ave t Melo jbshop scrding toa Poison dsbaon wih 3
Iman of eco er da) The shop neues ve macing centers wh
In practice, we tke n = 1 for convenience, which reduces the formula to ‘one sample of the interval between the arrival of jobs atthe fist machine
lini 5. The ACT scores forthe 1994 senior class at Springdale High is normal, with
Toiiestrate the use of ths method, suppose that we wish to generate a sample
from 2 N(0, 2) (rmeun = 10 and standard deviation o ~ 2)-Taking the sum of the
first 12 random aumbers in columns 1 and 2 of Table 18-1, we get x ~ 61094. Thus,
y= 10+ 2661098 ~ 6) = 102188,
"The disadvantage ofthis procedure is that it requires generating 12 (0,1) 6. Psychology professor Yatahais conducting a learning experiment in which
random numbers for each normal sample whichis computationally inefficient. more mice are rained to find their way around a restricted maze. The base ofthe
efficient procedure cals for sing the transformation ‘maze is square. The mouse enters the maze at one of the four corners and
‘must find its way through the maze to exit atthe same point where it en
‘mean of 27 points and a standard deviation of 3 points Suppose that we
draw a random sample of six seniors from that class. Use the Box Miller
method to determine the mean and standard deviation of the sample
V=TINTR) «08 2e R)
a tered. Te design ofthe maze is such that the mouse must pass by each of
Box and Muller (1958) prove that xia standard N(,1)- Thus y= y+ ax wil ro the remaining three comer points exacly once before it exits. The mit
doce a sample from NG 0). Tae new procedure is more ecient because it requires paths ofthe maze connect the four corners ina stit clockwise order. Pro
‘two (0,1) random numbers only. fessor Yataha estimates that the time the mouse takes to reach one commer
a a ie ae a prods anater MC.) semple sept by Te point from another is uniformly distributed between 10 and 20 seconds, de-
eR) This means tat the two random numbers , and Ry pending on the path it takes. Develop a sampling procedure for the time a
rultaneous (0,1) samples. mouse spends inthe maze.
prove that the
Placing oe(2R) wit i
nay be used to generate wo simul
"po llustrate the implementation of the Box-Mller procedure tothe normal dis 7. In Problem 6, suppose that once a mouse makes an exit from the maze, en
tsbution N(LO.2)the fist two random aumbers inclu 1 of able 1-1 yield the fl other mouse instantly enters, Develop a sampling procedure forthe number
owing N(0, 1) samples of mice that exit the maze in 5 minutes.
= VERTIS coucae x 733) = —1.108 8, Negative Binomial. Show how a random sample can be determined from
the negative binomial whose distribution is given as
ay = V~Tin OD) singe x 6738) ~ —2.108,
$0) = CPP ~ py, x= 0.1.2
Thus the corresponding N(0,2) samples are where xis the numberof failures until the yth success occurs ina sequence
yo = 10+ 2¢-1.108) = 2798 of m independent Bernouli trials and pis the probability of success
0< p <1. (Hint'The negative binomial isthe convolution of independent
eometrie samples. See Problem 18.46-9,)
na
2.108) = 5.7823
Acceptance-Rejection Method. ‘The acceptance-rejection method is de
ee re signed to handle complex pdfs for which the preceding methods cannot be applied
The general idea of the method isto replace the complex pf fx) with a more ane
Iyticaly manageable “proxy” pat h(x)-Samples from h(x) can then be used to deter:
mine samples from the original pdf f(x).
oral he pobiens ofthis setae the random umber n Table 16-1 starting wih hose of Define the majoring funetion g(x) such tha it dominates fl) in its entire
range—that i,
and A=
1. In Example 184-3, compute an Erlang sample, given m
3 per hour ‘600 Chap. 18 Simulation Modeling
a@)= fl), -#