Simulation studies are important in investigating various characteristics of a system or process. They are generally employed when the mathematical analysis necessary to answer important questions is too complicated to yield closed form solutions. For example, in a system where the time between successive customer arrivals has a particular pdf and the service time of any particular customer has another particular pdf, simulation can provide information about the probability that the system is empty when a customer arrives, the expected number of customers in the system, and the expected waiting time in queue. Such studies depend on being able to generate observations from a speci ed probability distribution. The rejection method gives a way of generating an observation from a pdf f(# ) when we have a way of generating an observation from g(#) and the ratio f(x)/g(x) is bounded, that is, c for some nite c. The steps are as follows:
1. Use a software package s random number generator to obtain a value u from a uniform distribution on the interval from 0 to 1. 2. Generate a value y from the distribution with pdf g(y). 3. If u f(y)/cg(y), set x y ( accept x); otherwise return to step 1. That is, the procedure is repeated until at some stage u f(y)/cg(y).
a. Argue that c 1. Hint: If c 1, then f(y) g(y) for all y; why is this bad? b. Show that this procedure does result in an observation from the pdf f(); that is, P(accepted value x) F(x). Hint: This probability is P({U f(Y)/cg(Y)} {Y x}); to calculate, rst integrate with respect to u for xed y and then integrate with respect to y. c. Show that the probability of accepting at any particular stage is 1/c. What does this imply about the expected number of stages necessary to obtain an acceptable value? What kind of value of c is desirable? d. Let f(x) 20x(1
x) 3 for 0 x 1, a particular beta distribution. Show that taking g(y) to be a uniform pdf on (0, 1) works. What is the best value of c in this situation?
