IE 581 Introduction to Stochastic Simulation

Name ___________________

One page of notes, front and back. Closed book. 50 minutes.

Score ___________________________

Schmeiser

Page 1 of 4

Test #2, Spring 2001

IE 581 Introduction to Stochastic Simulation

Name ___________________

1. True or false. (If you wish, write an explanation of your thinking.) (a) (b) (c) (d) (e) (g) (h) (i) (j) (k) (l) T F Acceptance/rejection can be used to generate random variates only from continuous distributions in one dimension. T F Composition can be used to generate random variates only from continuous distributions in one dimension. T F Special properties can be used to generate random variates only from continuous distributions in one dimension. T F The inverse transformation can be used to generate random variates only from continuous distributions in one dimension. T F The purpose of analyzing simulation output data using micro/macro . replications is to reduce the variance of the alternate point estimator, T F The purpose of antithetic variates is to reduce the variance of the point estimator. is a T F Micro/macro replications is applicable only when the point estimator sample mean. T F The instructor suggested using between 10 and 30 microreplications within each macroreplication, regardless of the total sample size n . T F Let Y = in=1 Yi / n . If Y 1, Y 2, . . . , Yn are identically distributed, then var(Y ) = var(Y 1) / n . T F Let Y = in=1 Yi / n . If Y 1, Y 2, . . . , Yn are identically distributed and independent, then var(Y ) = var(Y 1) / n . T F If arrivals to a system are generated from a Poisson process with rate (t ), then the times between the arrivals are independent and exponentially distributed.

2. In next-event simulation, an event can perform the following the actions: (a) Update statistics, (b) Schedule future events, (c) Update the state of the system. (a) Can one event do all three? yes no

(b) At any time t , what is the maximum number of states that the system can be in? one innite (c) For any statistic that is a mean, "updating statistics" involves adding an observation to an accumulator and incrementing the number of observations by one. True False (d) Each event is executed at a point in time. True False

Schmeiser

Page 1 of 4

Test #2, Spring 2001

IE 581 Introduction to Stochastic Simulation

Name ___________________

3. Statistics collected in a dynamic simulation fall into two categories. "Time" statistics involve an output variable that is dened for every time t . "Observational" statistics involve an output variable that is dened only occasionally by dynamic behavior of the model. Classify the following: (a) Estimate E("number in system") (b) Estimate V("number in system") (c) Estimate E("time in system") (d) Estimate P("time in system" > 10 minutes) 4. Consider the sentences In the proof that the acceptance/rejection algorithm is valid for generating a random variates from the random variable X with density function f , we used the notation that A = "Step 2 results in acceptance". Suppose that within the proof, we have P( A | X = x ) = f (x ) / [ c r (x ) ] . Using the above information for context, circle "constant", "random variable", "event", and "undened" for each part below. (a) x (b) A (c) X = x (d) c (e) P( A | X = x ) constant constant constant constant constant random variable random variable random variable random variable random variable event event event event event undened undened undened undened undened time time time time observational observational observational observational both both both both

Schmeiser

Page 2 of 4

Test #2, Spring 2001

IE 581 Introduction to Stochastic Simulation

Name ___________________

5. Suppose that n independent and identically distributed observations Y 1, Y 2, . . . , Yn are obtained in one replication of a simulation experiment. In performing micro/macroreplications, we use k macroreplications, each composed of m . These are microreplications, where n = mk . The j th macro replication results in j averaged to obtain . Suppose that we wish to estimate the mean Y and standard deviation Y using = and
n i =1

Yi / n
1/2 2

i =1

Yi

2 n .

n 1

in terms of Y , Y , . . . , Y . (a) Suppose that k = 3 and m = 100. Write 1 1 2 n

) in terms of std( ). (b) Write std( j

(c) The following items depend upon none, one or more than one of the experiment parameters n , m , and k . Specify which. (There can be more than one answer.) (i) (ii) (iii) j (iv) Y 1 (v) (vi) n n n n n n m m m m m m k k k k k k none none none none none none

Schmeiser

Page 3 of 4

Test #2, Spring 2001

IE 581 Introduction to Stochastic Simulation

Name ___________________

6. Consider a Poisson process with rate function (t ) = 2t for all non-negative times t . The cumulative rate function is then (t ) = t 2. At time ti 1, the cdf of the next time Ti is F (t ) = 1 exp[(t ) + (ti 1)]. (That is, the area under (t ) between ti 1 and Ti is exponentially distributed with mean one.) What is the logic to transform a random number u into the next arrival time t ?

7. We discussed the "three-step" method for generating a random vector (X 1, X 2, . . . , Xk ). Multivariate normal random variates are transformed to a multivariate uniform random vector, which is then transformed into the target random vector. (a) The method does not guarantee a particular joint distribution for the random vector. What model information does the method require?

(b) Why not assume that the components of the target random vector are independent? (Check all that are true.) (i) Modeling error would increase. (ii) Computation time would increase. (iii) Sampling error would increase. (iv) Verication error would increase.

Schmeiser

Page 4 of 4

Test #2, Spring 2001