13
POISSON DISTRIBUTION
Examples 1. You have observed that the number of hits to your web site occur at a rate of 2 a day. Let X be be the number of hits in a day
2. You observe that the number of telephone calls that arrive each day on your mobile phone over a period of a year, and note that the average is 3. Let X be the number of calls that arrive in any one day.
3. Records show that the average rate of job submissions in a busy computer centre is 4 per minute. Let X be the number of jobs arriving in any one minute. 4. Records indicate that messages arrive to a computer server at the rate of 6 per hour. Let X be the number of messages arriving in any one hour.
Generally X = number of events, distributed independently in time, occurring in a xed time interval. X is a Poisson variable with pdf: P (X = x) = e where is the average. Example: Consider a computer system with Poisson job-arrival stream at an average of 2 per minute. Determine the probability that in any one-minute interval there will be (i) 0 jobs; (ii) exactly 2 jobs; (iii) at most 3 arrivals. (iv) What is the maximum jobs that should arrive one minute with 90 % certainty?
x
x!
, x = 0, 1, . . . ,
Solution: Job Arrivals with = 2 (i) No job arrivals: P (X = 0) = e2 = .135 In R dpois(0, 2) [1] 0.1353353 (ii) Exactly 3 job arrivals: P (X = 3) = e In R dpois(3, 2) [1] 0.1804470 (iii) At most 3 arrivals P (X 3) = P (0) + P (1) + P (2) + P (3) 2 3 2 2 2 2 2 2 2 +e +e = e +e 1 2! 3! = 0.1353 + 0.2707 + 0.2707 + 0.1805 = 0.8571 In R ppois(3,2) [1] 0.8571235 more than 3 arrivals: P (X > 3) = 1 P (X 3) = 1 0.8571 = 0.1429
2 2 3
3!
= .18
(iv) Maximum arrivals with at least 90% certainty: i.e. 90% quantile Choose k so that P (X k) .9 In R qpois(.9, 2) [1] 4 at least a 90% chance that the number of job submissions in any minute does not exceed 4. equivalently less than a 10% chance that there will be more than 4 job submissions in any one minute.
Poisson Probability Density Functions
Web Site Hits: Poisson(2)
0.20 P(X = x) 0 2 4 6 8 10 12 0.00 0.10
Calls to Mobile: Poisson(3)
P(X = x)
0.00
0.10
0.20
10
12
Number of Hits
Number of Calls
Job Submissions: Poisson(4)
0.20
Messages to Server: Poisson(6)
0.15 P(X = x)
P(X = x)
0.10
0.00
10
12
0.00 0
0.05
0.10
10
12
Number of Submissions
Number of Messages
Poisson Probability Density Functions par(mfrow = c(2,2)) # multiframe x<-0:12 #look at the first 12 probabilities plot (x, dpois(x, 2), xlab = "Number of Hits", ylab = "P(X = x)", type = "h", main= "Web Site Hits: Poisson(2)") plot (x, dpois(x, 3), xlab = "Number of Calls", ylab = "P(X = x)", type = "h", main= "Calls to Mobile: Poisson(3)") plot (x, dpois(x, 4), xlab = "Number of Submissions", ylab = "P(X = x)", type = "h", main= "Job Submissions: Poisson(4)") plot (x, dpois(x, 6), xlab = "Number of Messages", ylab = "P(X = x)", type = "h", main= "Messages to Server: Poisson(6)")
Poisson Cumulative Distribution Functions
Web Site Hits:lambda=2
P(X = x) 1.0 P(X = x) 0 2 4 6 8 Number of Hits 12
Calls to Mobile:lambda=3
0.2 0.6 1.0 0 2 4 6 8 Number of Calls
0.2
0.6
12
Submissions:lambda=4
0.0 0.4 0.8 P(X = x) P(X = x) 0 2 4 6 8 12
Server Messages:lambda=6
0.0 0.4 0.8 0 2 4 6 8
12
Number of Submissions
Number of Messages
Poisson Cumulative Distribution Functions
par(mfrow = c(2,2)) # multiframe x<-0:12 plot (x, ppois(x, 2), xlab = "Number of Hits", ylab = "P(X = x)", type = "s", main= "Web Site Hits:lambda=2") plot (x, ppois(x, 3), xlab = "Number of Calls", ylab = "P(X = x)", type = "s", main= "Calls to Mobile:lambda=3") plot (x, ppois(x, 4), xlab = "Number of Submissions", ylab = "P(X = x)", type = "s", main= "Submissions:lambda=4") plot (x, ppois(x, 6), xlab = "Number of Messages", ylab = "P(X = x)", type = "s", main= "Server Messages:lambda=6", )
Derivations of Some Properties of Poisson 1. Clearly Also since i.e. e
x
x!
> 0 since > 0 e
x
x=0
x!
=1
2 3 e =1++ + + 2! 3!
x = e x=0 x!
2. E(X) =
E(X) =
x=0
xe
x x!
= e = e
x x x=1 x! x x=1 (x 1)!
x1 = e x=1 (x 1)!
2 1 + + + . . . = e 1! 2! = e e =
APPLICATIONS OF THE POISSON The Poisson distribution arises in two ways: 1. Events distributed independently of one another in time: X = the number of events occurring in a xed time interval has a Poisson distribution. P DF : Example: hour. p(x) = e x , x = 0, 1, 2, ; > 0 x! X = the number of telephone calls in an
2. As an approximation to the binomial when p is small and n is large, When examining the number of defectives in a large batch where p, the defective rate, is usually small.
Example: The manufacturer of the disk drives in one of the well-known brands of microcomputers expects 2% of the disk drives to malfunction during the microccomputers warranty period. Calculate the probability that in a sample of 100 disk drives, that not more than three will malfunction.
No. of disk drives malfunctioning k 0 1 2 3 Total
Binomial
100 k
.02k .98100k 0.13262 0.27065 0.27341 0.18228 0.85890
Poisson Approximation e2 2k /k! 0.13534 0.27067 0.27067 0.18045 0.85713
Poisson as an approximation to the binomial when n is large p is small Recall: mean of binomial = np mean of Poisson = PDF of Binomial P (x) = =
px (1
n x
p)nx; p =
x
n x
1 n
nx
n! x (1 n )n x = e x x!(n x)! nx 1 n x!
Now (1 )n e as n n (1 )x 1 as n n n(n 1) . . . (n (x 1)) n! = (n x)!nx nx 1 2 x1 = 1(1 )(1 ) . . . (1 )1 n n n
Some Examples 1. Messages arrive to a server at the rate of 6 per hour. What is the maximum number k so that the probability that the number of messages to the server in an hour more than this value is .75 Solution qpois(.75, 6) [1] 8 At least a 75% chance of 8 messages 25% change of > 8; i.e. 3rd quartile. For 1st quartile, qpois(.25, 6) [1] 4 Interquartile range: [4, 8], i.e. 50% of the time, arrivals will be between 4 and 8.
Example: A web server receives an average of 1000 queries per day. How much excess capacity should it build in to be 95% sure that it can meet the demand? Poisson with = 1000 Choose k so that P (X k) .95 qpois(.95, 1000) [1] 1052 and to double check ppois(1052, 1000) [1] 0.9506515 The extra capacity needed is 52.
Example: The mean number of errors due to a particular bug occurring in a minute is 0.0001 1. What is the probability that no error will occur in 20 minutes? 2. How long would the program need to run to ensure that there will be a 99.95% chance that an error will show up to highlight this bug? Solution Poisson = .0001 per minute = 20(.0001) = .002 per 20 minute interval. probability that no error will occur in 20 min: P (0) = e.002 = 0.998002 99.88% chance OF 0 in 20 min. Equivalently 1.12% that an error will show up in the rst 20 minutes. For 99.95% sure of catching bug: P (X 1) .9995, or equivalently P (X = 0) = .0005 P (No occurrence in k minutes) = e(.0001)k
To be 99.95% sure, choose k so that e(.0001)k .0005 or equivalently 1 e(.0001)k .9995 k > 75,000 mins k 75000/60 = 1250 hours k . 1250/24 = 52 days In R we can examine the probabilities of at least one error in diering timespans: k<-seq(50000,100000, 5000) #Running time in 5,000 int. y<-1-exp(-(.0001)*k) #At least one in k minutes plot(k,y, xlab = "Running time of the package in minutes", ylab = "Probability of at least one bug") abline(h = .9995) Bug Detection.
0.993 0.994 0.995 0.996 0.997 0.998 0.999 1.000
5 e+04
Probability of at least one bug in minutes
e+04
e+04
e+04
e+04
e+05
Running
time of the package
Example: The average number of defects per wafer (defect density) is 3. The redundancy built into the design allows for up to 4 defects per wafer. What is the probability that the redundancy will not be sucient if the defects follow a Poisson distribution? Poisson with = 4. For k defects:
3 3 k
k where X be the number of defects per wafer. The redundancy will not be sucient when X > 4. P (X > 4) = 1 (P (X = 0) + P (X = 1) + P (X = 2) + P (X = 3) + P (X = 4)) 31 32 33 34 = 1 e3 (1 + + + + ) 1 2! 3! 4! In R we obtain P (X > 4) using ppois 1-ppois(4, 3) [1] 0.1847368 over an 18% chance that the defects will exceed redundancy.
P (X = k) = e
R Functions for the Poisson Distribution
dpois dpois(x = 4, or dpois(4, 3) P (X = 4) = 3 ppois ppois(x = 4, or ppois(4, 3) P (X 4) with = 3 qpois qpois(.95, lambda = 3) or qpois(.95, 3) [1] 7 Choose k so that P (X k) .95. > qpois(.95, 3) [1] 6 \lambda = 3) lambda = 3)
